Cigarette primary processing workshop simulation model auxiliary system, operation method, electronic equipment and storage medium

Abstract

The invention discloses a simulation model auxiliary system for a cigarette cut tobacco manufacturing workshop, an operation method, electronic equipment and a storage medium. Comprising a workshop simulation model mark management module, a workshop simulation basic knowledge extraction module, a workshop simulation multi-mode knowledge graph construction module, a workshop simulation knowledge interaction and problem diagnosis module and a workshop system quality evaluation module. According to the invention, the workshop simulation model and the multi-modal knowledge graph are associated through the simulation model marking and management module, and model recommendation of a dialogue system is supported; a basic knowledge extraction module is utilized to extract data from MES, ERP, PLM and process documents, a multi-modal knowledge graph is constructed, and the construction efficiency is improved; the knowledge interaction module analyzes user requirements and provides structured support through knowledge graph retrieval; the optimization suggestion and problem diagnosis unit is combined with various reasoning methods to analyze requirements and provide diagnosis and optimization suggestions; and the system quality evaluation module ensures the reasoning accuracy, integrity and practicability, and provides guarantee for simulation model optimization and decision making.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing of manufacturing production lines, in particular to a simulation model auxiliary system and operation method, electronic equipment and storage medium for a cigarette thread making workshop. Background Art

[0002] At present, there are three main auxiliary solutions for workshop simulation model construction, namely, auxiliary methods for workshop simulation model construction based on secondary review by professionals, auxiliary decision-making methods for equipment support based on simulation modeling, and simulation modeling methods for auxiliary cooling systems of large passenger aircraft based on graph theory. These existing auxiliary solutions for workshop simulation model construction cannot provide reasonable optimization suggestions and problem location; and most of the existing technical solutions are mainly based on manual simple verification and problem finding, and a small number of them have modeling suggestion capabilities.

[0003] The auxiliary method of workshop simulation model construction based on secondary review by professionals has limitations in the current technical environment.

[0004] In the early workshop simulation model construction auxiliary scheme, Oser, RL (1999) mentioned in A Structured Approach for Scenario-Based Training. Human Factors and Ergonomics Society Annual Meeting Proceedings that in the scenario-based training method, the construction and review of the simulation model requires a lot of manual intervention; the scenario-based training method relies on the in-depth knowledge and experience of professionals to identify the various variables and dynamic elements in the training scenario. The article emphasizes that since the model needs to be manually verified and adjusted, it is often very time-consuming; in the case of multiple scenarios or complex training requirements, the lengthy review process not only affects the overall production efficiency, but also significantly increases the cost. The high cost of manual verification stems from the need to invest a lot of human resources to perform these tasks, and the cost of personnel training and knowledge updating also needs to be considered. It can be seen that the manual operation of model construction and verification requires the in-depth intervention of professionals, consumes a lot of human resources, and leads to high costs. This method must trigger a restart of the review process every time a new scenario changes or requirements, which is too long and inefficient; and it is highly dependent on the subjective judgment of professionals, resulting in poor adaptability and inconsistent results.

[0005] In a recent workshop simulation model construction assistance program, Ji, Z., Pons, D., & Pearse, J. (2020) mentioned in the article Plant system simulation for engineering training workshops. Computer Applications in Engineering Education that in engineering training workshops, the manual review method of simulation models is inherently complex, requiring professionals to conduct detailed review and verification of each step and operation in the model to ensure that it meets the training objectives and the actual operating environment. This includes analyzing the movement of operators and workpieces between multiple workstations, and considering waiting times and various decision-making stages in the model. In situations involving multi-level and multi-stage decisions, since each new work scenario requires adjusting and verifying the simulation model from scratch, the cycle of the overall review and optimization process becomes very long. This method relies on the experience and judgment of professionals, lacks standardized processes and tools for rapid review and adjustment, and is difficult to expand to a wider range of applications. In addition, manual review relies on the in-depth knowledge and judgment of professionals, which limits the application of the model in practice. Each step relies on the judgment of professionals, which not only limits the wide application of the model, but also makes the final result inconsistent and unpredictable.

[0006] Beijing Anmosi Technology Co., Ltd. has disclosed an equipment support decision-making assistance method based on simulation modeling. The implementation principle is based on modular structure, data-driven decision support, dynamic simulation and real-time response capability, fault management optimization strategy, and performance evaluation and feedback mechanism. Through this modular structure, each module can be executed and optimized independently, thereby improving the flexibility, maintainability and scalability of the system. In the equipment support decision-making assistance method based on simulation modeling, the core of the simulation model is to use a data-driven method to evaluate the equipment support performance. The data collection module is responsible for recording various key data (such as intermediate data, result data, equipment status data, and simulation time data), which are then used for statistical analysis and evaluation of the task evaluation module, thereby ensuring the scientificity and transparency of equipment support decisions, and being able to optimize the decision-making process through continuously updated and accumulated data. When a fault event occurs, the system can respond in real time according to the output of the fault sampling module, and advance the simulation time to the corresponding time point in order to quickly simulate and handle the fault. The fault sampling module randomly tests and simulates the occurrence of various fault events, and determines the driving events that should be handled; the simulation calculation module then optimizes the equipment support strategy based on these driving events, including calculating the start time, duration and delay time of each event, thereby optimizing the response strategy to various emergencies. Finally, by constructing a task evaluation module, the performance of the simulation system is statistically analyzed and evaluated to ensure that the feedback mechanism can continuously improve the simulation capability and decision support capability of the system, forming a closed-loop optimization process. Although the equipment support decision-making support method based on simulation modeling adopts a modular design, its shortcomings mainly lie in the complexity of the system, lack of real-time performance, reliance on the accuracy of initial data, limited adaptability and scalability, and lack of automated optimization capabilities. When the number of fault events increases or the scale of the system expands, the mutual dependence between modules is likely to lead to a decline in system performance, increase the difficulty of debugging and maintenance, and the complex simulation system requires a large amount of computing resources, which limits the application of workshop environments with limited resources. In the operation of the actual workshop simulation model, the real-time performance and response speed of the system will be limited. For example, when dealing with multiple concurrent fault events or facing high-frequency or complex fault events, the simulation calculation module and the task generation module need to process a large amount of data, which directly leads to a slower system response. System delays will affect the timeliness and effectiveness of the entire workshop system decision-making.

[0007] Professor Hu Haitao of Shanghai Jiao Tong University disclosed a simulation modeling method for auxiliary cooling systems of large passenger aircraft based on graph theory. The implementation principle is to use graph theory technology, encoding, directed graphs, adjacency matrices, decoding and state calculation steps to provide a fast, automated and general modeling method to help users efficiently build simulation models of auxiliary cooling systems of large passenger aircraft. This technical solution sets coding rules to encode the subsystems and subunits (such as compressors, condensers, evaporators, etc.) in the auxiliary cooling system as independent nodes, and the connections between nodes (such as pipes and flow paths) as edges, abstracting the physical system structure into a graphical model; after the coding is completed, a directed graph is generated according to the coding, and the edge set and point set represent the connection relationship and flow direction of the system; then, the directed graph is converted into an adjacency matrix, and the matrix elements represent the connections between nodes ("1" means connection, "0" means no connection), which is efficiently stored and processed in the computer; then, by decoding the adjacency matrix, the topological information and state information of the system (such as subsystem type, connection mode and coolant flow direction) are automatically extracted, and the inlet and outlet states (such as temperature, pressure, flow) of each subunit are further calculated to support accurate simulation analysis and optimization decision-making. This technical solution relies on pre-set coding rules and fixed graph structures, which have limitations when dealing with real-time changes that often occur in workshops (such as equipment failures, order changes, production plan adjustments, etc.). The reason is that workshop simulation requires rapid response to environmental changes and instant decision-making, while graph theory methods require regeneration or adjustment of graph models, which makes it difficult to quickly adapt to dynamic changes, resulting in reduced efficiency. Although graph theory models can represent the physical connections and paths in the system, they are not expressive enough to describe these diverse workshop dynamic behaviors (such as production line scheduling, inventory management, etc.), and it is difficult to fully adapt to different types of workshop scenarios. Workshop simulation models usually contain a large number of discrete events (such as process start / end, equipment start / stop, etc.), and the timing and interrelationships of these events are complex. Graph theory methods are more suitable for state analysis of continuous systems, and have limited description and simulation capabilities for the time-driven characteristics of discrete event systems, making it difficult to fully support the needs of discrete event simulation.

[0008] Therefore, there are obvious deficiencies in the construction and verification of current workshop simulation models, including the lack of automated detection and verification methods, over-reliance on the experience of engineering personnel, lack of comparative analysis methods, and lack of effective quality assessment support. These technical problems limit the accuracy, optimization and reuse capabilities of workshop simulation models.

[0009] The existing simulation model of the cigarette silk workshop is constructed by modeling software, which greatly relies on the modeling experience of engineers and relevant knowledge of the workshop. The modeling process is complicated and time-consuming. If errors occur, it will cause great losses to the subsequent construction of the cigarette silk workshop flow. The third party manually evaluates the simulation model, and performs manual problem diagnosis and optimization suggestions after the simulation model is fully developed. It is extremely expensive and time-consuming. Therefore, there is an urgent need for an auxiliary system and working method in the modeling of the cigarette silk workshop to provide timely and accurate problem diagnosis and optimization suggestions for the cigarette silk workshop for the simulation information platform. Summary of the invention

[0010] The first purpose of the present invention is to provide a simulation model auxiliary system for a cigarette silk-making workshop. By utilizing the intention recognition algorithm and the slot filling mechanism in combination with the extraction workshop modeling method and the construction of a multimodal knowledge graph, the simulation model recommendation of the cigarette silk-making workshop, the problem diagnosis and optimization suggestion functions of the cigarette silk-making workshop operations are enhanced, and the high-quality performance of the simulation model system in practical applications is ensured, decision support for comprehensive quality evaluation is provided, and the efficiency and quality of cigarette silk-making workshop modeling are improved.

[0011] The second purpose of the present invention is to provide an operating method of a cigarette silk-making workshop simulation model. According to the illustrated content and the actual working conditions of the cigarette silk-making workshop, the operating process of the cigarette silk-making workshop is divided into five stages: knowledge extraction and construction, demand analysis and reasoning of the cigarette silk-making workshop, recommendation and optimization of cigarette silk-making workshop simulation models, matching and recommendation of cigarette silk-making workshop simulation models, and system quality evaluation. This ensures that a cigarette silk-making workshop simulation model auxiliary system can efficiently extract data, understand the actual needs of cigarette silk-making workshops, infer knowledge, recommend suitable cigarette silk-making workshop simulation models and evaluate their quality.

[0012] The third object of the present invention is to provide an electronic device for a simulation model of a cigarette silk-making workshop, the purpose of which is to provide powerful computing and storage support for the construction and optimization of the simulation model of the cigarette silk-making workshop. Through the combination of high-performance computing and large-capacity storage, the device can process a large amount of multimodal data, perform complex reasoning tasks, and provide reliable data management and disaster recovery capabilities, thereby achieving efficient operation of the system and optimizing decision support.

[0013] The fourth object of the present invention is to provide a non-temporary storage medium for a cigarette silk-making workshop simulation model, which can provide computer-readable and executable instructions, support the implementation of a cigarette silk-making workshop simulation model auxiliary system and operating method, and provide the system with efficient, reliable and flexible data storage and management capabilities, thereby ensuring the effectiveness and reliability of the system in different scenarios.

[0014] To achieve the above-mentioned purpose, the present invention adopts the following technical scheme: a simulation model auxiliary system for cigarette making workshop

[0015] The first object of the present invention is to provide a cigarette silk-making workshop simulation model auxiliary system, which includes five independent modules, namely, a cigarette silk-making workshop simulation model tag management module, a cigarette silk-making workshop simulation basic knowledge extraction module, a cigarette silk-making workshop simulation multimodal knowledge graph construction module, a cigarette silk-making workshop simulation knowledge interaction and problem diagnosis module, and a cigarette silk-making workshop system quality assessment module;

[0016] The cigarette silk workshop simulation model tagging management module includes a simulation model feature extraction unit, a simulation model annotation unit, and a simulation model feature data management unit, which are used to accurately define and describe the cigarette silk workshop simulation model, and combine the marked information with the knowledge graph, so as to provide a cigarette silk workshop model recommendation function based on the knowledge graph for the dialogue simulation platform system;

[0017] The basic knowledge extraction module for cigarette silk workshop simulation includes a cigarette silk workshop data extraction unit and a cigarette silk workshop data processing unit, which are used to extract key rules and data of the cigarette silk workshop, and obtain basic knowledge data related to the cigarette silk workshop simulation platform by constructing entity relationships, ternary relationships and knowledge inheritance hierarchical structures of knowledge graphs;

[0018] The multimodal knowledge graph construction module for simulating the cigarette silk-making workshop includes a basic graph construction unit and a feature fusion unit, which are used to integrate and analyze data from various information systems of the cigarette silk-making workshop (such as the manufacturing execution system MES, the enterprise resource planning system ERP, the product lifecycle management system PLM, and data in process documents related to cigarette silk-making), and to construct a comprehensive knowledge graph that supports optimization analysis of the operation of the cigarette silk-making workshop, fault diagnosis of cigarette silk-making operations, and decision support;

[0019] The cigarette silk workshop simulation knowledge interaction and problem diagnosis module includes a dialogue management and knowledge query unit, a problem diagnosis and optimization suggestion unit, a model recommendation and interactive learning unit, which is used to accurately parse the natural language input of the cigarette silk workshop, extract key knowledge using the intent recognition algorithm and slot filling mechanism, and use knowledge graph reasoning to capture the complex relationship between related entities in the cigarette silk operation, perform multi-step relationship reasoning, and convert the actual demand description of the cigarette silk workshop operation into a graph query expression. Then, using subgraph matching and query parsing methods, it extracts improvement suggestions for related production scheduling, quality management, process formula management, and tobacco leaf or tobacco material supply chain management from the graph, and comprehensively obtains the final knowledge interaction and problem diagnosis;

[0020] The cigarette silk-making workshop system quality assessment module includes a storage device and processing equipment accuracy verification unit, a tobacco processing rule integrity check unit, a tobacco production practicality assessment unit, a tobacco finished product quality index calculation unit, and a system dynamic update support unit. After the system is built, it is used to comprehensively evaluate the overall quality of the cigarette silk-making workshop logistics simulation multimodal system to ensure that the actual application of the system in the cigarette silk-making operation shows accuracy, completeness, consistency and practicality, thereby effectively supporting the optimization and decision-making process of the cigarette silk-making workshop simulation model.

[0021] Furthermore, in the cigarette thread-making workshop simulation model auxiliary system described in the present invention, the composition and function of each module and unit are as follows:

[0022] The simulation model feature extraction unit automatically identifies and extracts key feature data from simulation models of different workshops;

[0023] The technical means adopted by the simulation model feature extraction unit are structured data analysis technology, computer vision and image processing technology, natural language processing technology and data fusion and integration technology;

[0024] The simulation model annotation unit is used to perform hierarchical annotation of explicit information and implicit information on the simulation model; the model annotation unit adopts a variety of technical means to achieve comprehensive annotation of the model, including automatic annotation and manual annotation;

[0025] The simulation model feature data management unit is responsible for storing simulation model files and feature data to ensure integrity and traceability. By establishing a multidimensional index structure, it supports rapid retrieval and associated query of model feature data. For each model, the feature data management unit will classify the acquired data and perform structured storage. The optional data format includes JSON. The technical means adopted by the simulation model annotation unit are feature data storage and multidimensional indexing, high-dimensional feature mapping and dynamic embedding mechanism, dynamic feature mapping tensor adjustment and similarity calculation and clustering algorithm.

[0026] The data extraction unit of the cigarette silk-making workshop obtains relevant data from different information systems (such as manufacturing execution system MES, enterprise resource planning system ERP, product lifecycle management system PLM) and process documents to support the construction of workshop process simulation model; the technical means adopted by the data extraction unit of the cigarette silk-making workshop are the workshop's manufacturing execution system, enterprise resource planning system, product lifecycle management system and related process documents to obtain relevant data of workshop equipment and processes; the technical means adopted by the data extraction unit of the cigarette silk-making workshop are data integration tools (ETL), API (application programming interface) calls, database queries, document parsing tools and data cleaning and conversion technologies;

[0027] The data processing unit of the cigarette silk workshop performs data splitting and noise reduction after successfully acquiring relevant data; for structured data, concepts and relationships are automatically generated through predefined templates, and knowledge is extracted in combination with ETL tools; for unstructured data, part of the data is annotated through concepts and relationships, a domain knowledge training library is constructed, and knowledge is extracted using relevant models; and multiple training algorithms are used in the extraction process to improve the recognition and extraction capabilities of data types, sequence annotation models such as BiLSTM are used for text data, and feature extraction is completed for image data through the SURF feature extraction algorithm; the technical means adopted by the data processing unit of the cigarette silk workshop are ETL tools, natural language processing (NLP) technology, image processing algorithms, noise reduction algorithms and multiple machine learning training algorithms;

[0028] The basic graph construction unit constructs multiple basic knowledge graphs through preliminary structured processing and data integration to realize the conceptual links between different systems such as workshop equipment, process flow, simulation platform, etc.; each basic knowledge graph uses the workshop equipment, process flow, workstation in the simulation platform, etc. in their respective systems as node concepts, and the association between equipment and process, the association between workstation and process, etc. as edges, reflecting the equipment, process flow and their association in the workshop, and ultimately supporting a more accurate and intelligent simulation process; the technical means adopted by the basic graph construction unit are knowledge graph construction tools, data integration and ETL processing, relationship extraction and graph generation and graph optimization;

[0029] The feature fusion unit generates and disambiguates candidate concepts by integrating the features of multimodal data (such as text, images, etc.), matches and links them with concepts in the basic knowledge graph, and combines multimodal information from different data sources to eliminate ambiguity and ensure that the generated candidate concepts can be accurately linked to the corresponding concept nodes in the knowledge graph; the technical means adopted by the feature fusion unit are weighted multimodal feature fusion model, attention mechanism disambiguation and matching and linking of candidate concepts;

[0030] The dialogue management and knowledge query unit is used to parse workshop input, query knowledge graph, generate response and manage dialogue status, and finally complete natural language parsing; the technical means adopted by the dialogue management and knowledge query unit are natural language processing (NLP), intent recognition and slot filling, dialogue management, knowledge graph retrieval and natural language generation (NLG);

[0031] The problem diagnosis and optimization suggestion unit is to perform problem identification, optimization suggestion, problem diagnosis and comprehensive reasoning; the problem identification is to automatically detect potential problems or abnormal conditions in the simulation model; the technical means adopted by the problem diagnosis and optimization suggestion unit combine multiple reasoning and analysis methods to help systematically analyze the simulation model of the cigarette silk workshop and provide optimization suggestions; these technical means include rule-based, case-based, representation learning-based reasoning and fault tree analysis (FTA);

[0032] The recommendation and interactive learning unit is to perform model recommendation, interactive optimization and recommendation explanation; analyze the needs input by the workshop, match them with the classified simulation model feature library, generate a model recommendation list most relevant to the needs, and then recommend the simulation model that best meets the needs to the cigarette silk workshop based on similarity evaluation, accelerate the deployment and optimization of the model, and then provide a detailed explanation of the recommendation results, display the matching basis of each recommended model with the needs of the cigarette silk workshop, help the workshop understand the reasons for the recommendation, and finally support the workshop to dynamically adjust the recommended model, through the interactive feedback mechanism, continuously optimize the model according to the actual application scenario, to ensure that the model is closer to the actual needs of the workshop; the technical means adopted by the recommendation and interactive learning unit are model feature library construction and classification, demand analysis and matching, similarity evaluation and recommendation, recommendation explanation function and interactive learning and feedback mechanism;

[0033] The storage device and processing equipment accuracy verification unit verifies the accuracy of the data, parameters and their relationships in the simulation model to ensure that the equipment, processes and concepts and relationships used in the system and the simulation model are consistent with the equipment and process flow in the actual production environment, and then compares the equipment parameters (such as capacity, operating speed, energy consumption, etc.) in the simulation model with the actual equipment parameters to ensure parameter matching, and then checks whether the relationship between the equipment and the process flow in the system is consistent with the actual situation, such as the upstream and downstream relationship of the equipment and the operation sequence; if inconsistencies or errors are found, a feedback report is generated and correction suggestions are made; the technical means adopted by the storage device and processing equipment accuracy verification unit are data comparison and verification, knowledge graph consistency verification, rule engine verification and data mapping and feedback mechanism;

[0034] The main task of the integrity check unit is to detect whether there are any missing process links, equipment parameters, and whether there are any internal contradictions or inconsistent relationships through systematic evaluation, so as to ensure that the simulation system contains all key concepts, equipment, process flows and their relationships involved in the workshop production process without missing or contradictory concepts and information; the technical means adopted by the integrity check unit are concept coverage check, logic consistency verification, rule engine and consistency check, and data mapping and feedback mechanism;

[0035] The finished tobacco product quality index calculation unit is used to calculate key performance indicators, generate quality reports and support system optimization and adjustment, so as to evaluate the performance of the system, measure the system's performance in processing cigarette shred workshop inquiries, data feedback, simulation model support, etc., and generate detailed quality reports, using the system's accuracy, recall rate, F1 value, and coverage rate as indicators to optimize and adjust the system; the technical means used by the finished tobacco product quality index calculation unit are accuracy calculation, recall rate calculation, F1 value calculation, coverage rate calculation, performance evaluation tools, and automated testing and feedback mechanisms;

[0036] The system dynamic update support unit performs version management, incremental updates, compatibility verification, and automatic updates and notifications to ensure that the system can be updated and adapted to new data and requirements in a timely manner when equipment and process flows change. Furthermore, through version management and incremental update mechanisms, the system changes are tracked and their real-time and accuracy are maintained to ensure that the simulation system can continuously optimize and reflect the latest production environment. The technical means adopted by the dynamic update support unit are version management tools, incremental update mechanisms, compatibility verification mechanisms, and automatic updates and notification mechanisms.

[0037] Preferably, the simulation model feature extraction unit automatically identifies and extracts key feature data, including structured data, image data, and text data; the structured data is data organized and stored in a specific format, data information stored in a database, table or other structured data storage system, which is easy to represent and process in the form of a table or database, including the attributes of equipment objects (such as type, specification, performance parameters), upstream and downstream relationships between equipment, equipment spatial distribution data, and process parameters (such as temperature, speed, and time); the image data is information captured visually, stored and processed in the form of an image, and is used to intuitively describe the appearance, layout, and spatial distribution of the equipment, including a three-dimensional shape cross-sectional view of the equipment, a workshop layout diagram, and a photograph of the equipment installation location; the text data is information presented in the form of natural language text, and comes from sensor data, equipment status information, working parameter information, operation log information, workshop transport parameter information, quality index information, and / or fault information, including the process description, operation manual, simulation log, workshop input instructions, or description text of a cigarette silk workshop.

[0038] Preferably, the explicit information of the simulation model annotation unit is information that can be clearly defined and identified in the simulation model, including intuitive and specific data information, which is easy to extract and annotate, and can directly describe the appearance, structure and basic properties of the model; the explicit information includes basic information, hierarchical information, and layout image of the simulation model; basic information includes model name, model type, and model update date; hierarchical information includes object library structure, object attributes, and object subordination; layout image information includes a two-dimensional image of the workshop and a three-dimensional shape cross-sectional view of a single-machine equipment; the implicit information of the simulation model annotation unit is a feature that is not clearly defined in the simulation model but describes the model. The implicit information needs to be further analyzed and refined from the basic marking content of the model, and can supplement and deepen the understanding of the simulation model.

[0039] Preferably, the data acquired in the data extraction unit of the cigarette silk-making workshop include the basic material data of the workshop, the material processing rules and related process standards of the workshop, the production scheduling data of the workshop, the equipment modeling data of the workshop and the process sections and process data of the workshop; the basic material data of the workshop include data information on tobacco material name, material type, material code and material grade; the material processing rules and related process standards of the workshop include information on group processing, blending weight, blending sequence, upper and lower limits of storage time, and conversion rate of tobacco materials in the processing stage; the production scheduling data of the workshop include data information on cigarette silk-making workshop shifts, cigarette silk-making workshop teams and factory calendar; the equipment modeling data of the cigarette silk-making workshop include data information on basic equipment parameters, control rules and status of storage equipment, processing equipment and auxiliary equipment; the process sections and process data of the workshop include data information on design flow, brand change time, process length and processing time of main equipment; the ownership relationship between the process sections, processes and equipment of the workshop includes data on processes included in the process sections and the number of equipment included in the processes.

[0040] Furthermore, the simulation model annotation unit allows manual input of text descriptions and uses a large language model (LLM) to split and convert the manually input text into standardized tags or data formats; the implicit information annotation uses simulation logs, user input, and NLP technologies (such as NER, LDA) in the cigarette silk-making workshop to extract implicit information, and combines multiple NLP technologies for the implicit information extraction process of the cigarette silk-making workshop simulation model to form new and specific simulation model combinations and application scenarios; the manual input and LLM conversion uses a large language model (LLM) to perform data conversion and annotation in a specific field to split the input text into multiple meaningful parts and convert it into structured data entries (such as workshop scene names, category data information), and the system automatically marks the converted data entries as part of the implicit information field, and finally saves and updates the annotation information for subsequent retrieval and optimization.

[0041] Furthermore, the high-dimensional feature mapping and dynamic embedding mechanism maps the feature data of the simulation model into a high-dimensional topological space and classifies it using the dynamic feature embedding mechanism; then, a nonlinear transformation (such as the Tanh function) and a weighting coefficient matrix are used to transform the model feature data into a multidimensional feature data vector y, y = [y1, y2, ..., y n ], each y i Represents different model features, multi-dimensional logistics feature mapping function f MLFM As shown below:

[0042] L = f MLFM (y)=[L1(y),L2(y),…,L m (y)]

[0043] Each mapping dimension L i The formula for (y) is as follows:

[0044]

[0045] In the formula, λ ij is the weight coefficient of the i-th dimension to the j-th feature, γ ij is the scaling factor of each dimension of the simulation model, δ ij is the normalization coefficient of the i-th dimension to the j-th feature; in the simulation model of the cigarette shred workshop, the physical properties and operating parameters (such as temperature, speed, and output rate) of the equipment objects (such as shredders, shredders, and flavoring machines) will be mapped into high-dimensional vectors and converted into embedded representations in the high-dimensional topological space through the above mapping formula;

[0046] The dynamic feature map tensor adjustment is performed using the dynamic feature map tensor D(k), and an adjustment factor is introduced to represent the weight change of the feature in different logistics operations; the adjustment factor is calculated using the Sigmoid function:

[0047] The dynamic feature map tensor D(k) is generated by introducing an adjustment factor based on the vector y, which represents the weighted value in the kth logistics operation:

[0048] D(k)=[λ1(k)y1,λ2(k)y2,…,λ n (k)y n ]

[0049] Among them, λ(k) uses the sigmoid function to calculate the adjustment factor of each feature in the kth logistics operation; the sigmoid function is often used to limit the value range of the adjustment factor to between (0,1) to avoid the influence of extreme values ​​and ensure that the adjusted feature weight is within a reasonable range;

[0050] In the simulation model of the cigarette making workshop, the operating characteristics of different equipment have different importance in different logistics operations (such as feeding, blending, cutting, etc.); and by dynamically adjusting the weights of these characteristics, the influence of characteristic data in specific operations can be more accurately reflected. For example, in the drying process, the rotation speed and temperature characteristics of the drying machine may have higher weights, while in the cutting process, the blade speed and feed speed characteristics of the cutting machine have higher weights.

[0051] The similarity calculation is to calculate the similarity score between the simulation models through multiple metrics (such as logistics topology space distance, dynamic feature mapping tensor similarity, cosine similarity); calculate the classification score between the simulation models and classify them according to the score using a clustering algorithm; the classification score S(L) between the i-th and j-th logistics simulation models in the k-th logistics operation i ,L j ,k)The calculation formula is as follows:

[0052] S(L i ,L j ,k)=α·d(L i ,L j )+β·sim(D i (k),D j (k))

[0053] In the formula, α represents the weight coefficient of the logistics topological space distance, β represents the weight coefficient of the dynamic feature map tensor similarity, sim(D i (k),D j (k)) indicates D i (k) and D j (k) The cosine similarity between them.

[0054] Furthermore, the data extraction unit of the cigarette silk-making workshop combines NLP and knowledge graph to convert the entities and relationships extracted by NLP into structured data forms, which are used to build a domain knowledge base of the cigarette silk-making workshop to support the simulation model's understanding and simulation of complex process flows; the image processing of the data extraction unit of the cigarette silk-making workshop uses the SURF algorithm, which is not only used for equipment layout detection, but also closely combines the spatial information of these feature points with the simulation model to reflect the equipment status and layout dynamics in real time, thereby improving the mapping accuracy of the simulation model to the physical workshop environment; the comprehensive processing and intelligent optimization of multimodal data of the data extraction unit of the cigarette silk-making workshop comprehensively uses machine learning models to process time series data, image data and structured data, to achieve real-time optimization and fault prediction of the workshop production process, especially through deep learning training based on the specific equipment and process data of the cigarette silk-making workshop to enhance the intelligence of the simulation model.

[0055] Furthermore, the feature fusion unit further constructs a multimodal knowledge graph after constructing the basic knowledge graph; candidate concept generation is based on a weighted multimodal feature fusion candidate concept generation model; the multimodal feature fusion model is shown in Formula 1:

[0056]

[0057] In the formula, C i represents the i-th candidate concept, such as workshop equipment, process flow, or a specific workstation in the simulation platform; D m Refers to the dataset of the mth modality, such as the text dataset D1 contains the text descriptions in the workshop process documents, and the image dataset D2 contains the images of factory equipment; M refers to the number of modalities, W m represents the weight parameter for the mth mode; f m (D m ,θ m ) refers to the feature extraction function used for this modality, D m is the modal data set, θ m is the model parameter set under each mode, such as the weight matrix of BiLSTM and the key point detection parameters of SURF;

[0058] Concept disambiguation performs weighted fusion of multimodal features based on the attention mechanism and eliminates ambiguity by maximizing the similarity between related concepts. The introduced weighted attention mechanism model is shown in Formula 2:

[0059]

[0060] In the formula, S(c i ) represents the candidate concept c i The disambiguation score is finally used to judge c i Is it the correct concept? Represents the candidate concept c i The multimodal feature representation at the Lth layer; σ represents the activation function, M(L-1) represents the weight matrix of the L-1th layer, and M(c i ) represents the candidate concept c i The associated set of neighbor concepts, is the attention weight, indicating the importance of neighbor concept x in generating c i Importance when scoring disambiguation; is the multimodal feature representation of the neighbor concept x at the L-1 layer, which is used to calculate the candidate concept c i The feature of , d(L-1) is the bias vector of the L-1 layer, which corrects the final feature representation. Attention weight The calculation is shown in formula 3:

[0061]

[0062] Where k is the vector parameter used to calculate the attention score, M key and M val is the weight matrix used to map the candidate concept and neighbor concept features into the same space;

[0063] After candidate concepts are generated and disambiguated, the final concept linking is completed by comparing the candidate concepts with the concepts in the basic knowledge graph. This process uses the aforementioned weighted multimodal feature fusion model to calculate the matching scores of candidate concepts and link the candidate concepts with the concept nodes in the knowledge graph. In the concept linking, concept confidence and distance metrics are additionally introduced to enhance the accuracy of concept linking. The matching and linking process of candidate concepts can be expressed as Formula 4:

[0064]

[0065] In the formula, c * It represents the candidate concept with the highest score after considering multiple factors. α is the weight of the candidate concept score, which is used to adjust the influence of the candidate concept score in the final decision. β is the weight parameter considering the similarity of neighbor concepts, which is used to adjust the influence of neighbor concepts on the candidate concept link. Z is the normalization factor to ensure that the influence of similarity is within a reasonable range. It provides a platform for multiple sources such as workshop equipment, process flow or simulation platform to communicate with c i the associated set of neighbor concepts; Represents the candidate concept c i The weight between the neighbor concept x and the neighbor concept x indicates its influence in the link decision; sim(c i ,x) represents the candidate concept c i The similarity between the candidate concept and the neighbor concept x measures their distance in the feature space; δ represents the weight parameter of the confidence and distance metric, which is used to balance the confidence of the candidate concept and its distance from the existing knowledge graph concept; conf(c i ) represents the candidate concept c i The confidence of , indicating the matching confidence with the target concept; d(c i ,E) represents the candidate concept c i The cosine similarity with the concept set E in the knowledge graph; the graph can be stored and managed using the graph database Neo4j or Amazon Neptune.

[0066] Furthermore, the deep integration of NLP and domain-specific knowledge graph in the dialogue management and knowledge query unit is to deeply integrate the Transformer-based NLP model with the multimodal knowledge graph of the specially constructed cigarette silk workshop, so as to directly retrieve, reason and generate responses that are highly relevant to the specific needs of the workshop from natural language; the dynamic dialogue management of the dialogue management and knowledge query unit and the multi-round interaction specific to cigarette silk making are dialogue systems for cigarette silk making workshops, which not only track the dialogue history through LSTM or GRU, but also perform dynamic context management in combination with the actual process and equipment status of the workshop, so as to be able to continuously improve the interaction quality in multiple rounds. The dialogue strategy is continuously adjusted and optimized to make the system response more in line with the actual production environment; the deep reasoning and multi-step query optimization based on the knowledge graph in the dialogue management and knowledge query unit performs multi-step reasoning and optimization on the complex association between SPARQL and graph query algorithms and equipment and processes to form an intelligent query and reasoning engine; the adaptive natural language generation model of the dialogue management and knowledge query unit in NLG, in addition to template generation, the system also uses deep learning models (such as GPT) for adaptive generation, which can dynamically adjust the generated content according to the real-time data and changes of the workshop, so that the response is more in line with the complex cigarette silk workshop scene.

[0067] Furthermore, the reasoning based on representation learning in the problem diagnosis and optimization suggestion unit is realized through knowledge graph embedding, multi-step reasoning and implicit relationship optimization, and process flow linkage optimization; the knowledge graph embedding and multi-step reasoning are to convert entities and relationships in the knowledge graph into low-dimensional vectors through deep learning technology, and perform multi-step reasoning through these vectors, which can mine the implicit connection between equipment and process flow and find optimization points that cannot be captured by traditional rules or case methods;

[0068] The implicit relationship optimization and process flow linkage optimization capture complex implicit relationships through representation learning, and perform process linkage optimization, which deeply combines deep learning with the workshop simulation model, can identify clear equipment and parameter problems in the workshop, and discover potential collaborative relationships between equipment; the reasoning based on representation learning is a comprehensive integration of multiple reasoning methods, which integrates rule reasoning, case reasoning, reasoning based on representation learning and fault tree analysis (FTA) in the same system, allowing the system to flexibly use different reasoning methods at different levels and in different scenarios to form more comprehensive and accurate diagnosis and optimization suggestions; and then through knowledge graph embedding and multi-step reasoning, the system can combine the operating status of the equipment in real time to make dynamic optimization suggestions.

[0069] The recommendation and interactive learning unit has deep characterization and accurate classification in specific areas. For the simulation model specific to the cigarette thread making workshop, the equipment parameters and process stage information are deeply extracted and accurately classified, making the simulation model management and retrieval more accurate.

[0070] The multi-dimensional demand analysis and precise matching of the recommendation and interactive learning unit is to identify basic needs, extract more detailed slot data (such as production efficiency, material processing time), and perform precise matching to enhance the pertinence of simulation model recommendations;

[0071] The combination of similarity evaluation and feature space mapping of the recommendation and interactive learning unit is to use a traditional similarity algorithm combined with spatial mapping and multi-dimensional analysis of model features when making recommendations;

[0072] The in-depth combination of the recommendation explanation and user interaction of the recommendation and interactive learning unit is to explain the recommendation reasons in detail through visualization tools, and help the workshop understand the matching relationship between the characteristics and requirements of each model, and provide in-depth decision support;

[0073] The real-time interactive learning and adaptive recommendation optimization of the recommendation and interactive learning unit is achieved by recording feedback from the workshop and instantly recalculating the similarity and optimization results of the recommendation model after each adjustment.

[0074] Furthermore, the real-time bidirectional data verification and dynamic correction mechanism of the storage device and processing equipment accuracy verification unit can compare the simulation model with the knowledge graph in real time for consistency, dynamically verify whether the upstream and downstream relationships of the equipment and process flow are correct, and provide immediate feedback; the advanced data mapping and bidirectional feedback mechanism of the storage device and processing equipment accuracy verification unit adopts more sophisticated mapping technology to achieve one-to-one precise matching of actual equipment and equipment in the simulation model, and maintain dynamic synchronization of data. At the same time, it not only provides static feedback reports, but also supports dynamic adjustment suggestions, and directly converts feedback into automatic adjustment instructions for the simulation model; the intelligent knowledge graph consistency verification of the storage device and processing equipment accuracy verification unit automatically optimizes the node and relationship matching rules in the knowledge graph through intelligent algorithms (such as machine learning) to enhance the accuracy of the verification.

[0075] Furthermore, the deeply integrated version management and simulation model update of the dynamic update support unit uses a deeply integrated version control tool, so that each time the process parameters or equipment configuration is updated, a version record can be automatically generated, and rollback can be allowed at the model level; an incremental update mechanism is adopted to update only part of the data of the relevant equipment or process, reducing the system downtime and update burden; when new equipment is introduced into the workshop or the process is changed, the system automatically performs a compatibility check and generates a detailed conflict report; the system not only automatically detects and updates data, but also can immediately feedback the update content and impact to the workshop through the notification system, and recommends verification.

[0076] (2) An operation method of a cigarette making workshop simulation model

[0077] The second purpose of the present invention is to provide an operating method of a cigarette silk-making workshop simulation model described in the present invention. According to the illustrated content and the actual working conditions of the cigarette silk-making workshop, the operating process of the cigarette silk-making workshop is divided into five stages: knowledge extraction and construction, demand analysis and reasoning of the cigarette silk-making workshop, recommendation and optimization of cigarette silk-making workshop simulation models, matching and recommendation of cigarette silk-making workshop simulation models, and system quality evaluation. This ensures that the cigarette silk-making workshop simulation model auxiliary construction system described in the present invention can efficiently extract data, understand the actual needs of the cigarette silk-making workshop, infer knowledge, recommend suitable cigarette silk-making workshop simulation models and evaluate their quality.

[0078] The operation method of the cigarette thread making workshop simulation model described in the present invention needs to go through the following five stages:

[0079] Phase 1: Data extraction and knowledge graph construction - extract basic data of the cigarette silk workshop, generate knowledge entities, and build a multimodal knowledge graph;

[0080] Phase 2: Demand analysis and reasoning of the cigarette silk workshop - Analyze the actual demand of the cigarette silk workshop and determine the optimization direction through multi-step reasoning;

[0081] Phase 3: Cigarette silk workshop simulation model recommendation and optimization - extract the characteristics of the cigarette silk workshop simulation model, perform index management, and put forward suggestions for optimizing the operation of the cigarette silk workshop;

[0082] Phase 4: Cigarette silk workshop simulation model matching and recommendation - matching the most suitable simulation model category according to the needs of the cigarette silk workshop, generating a recommended simulation model and giving reasons for the recommendation;

[0083] Phase 5: System Quality Assessment - Conduct a comprehensive quality assessment of the cigarette making workshop simulation model to ensure reliability and practicality.

[0084] Furthermore, the operation method of the cigarette thread making workshop simulation model of the present invention comprises the following steps:

[0085] Phase 1: Data extraction and knowledge graph construction

[0086] S1. Extraction of basic knowledge of cigarette silk-making workshop simulation: Extract basic data from the manufacturing execution system (MES), enterprise resource planning system (ERP), product lifecycle management system (PLM) and related process documents of the cigarette silk-making workshop; these data include information on commonly used equipment for cigarette silk-making, process flow of cigarette silk-making, material data of tobacco leaves or modules, production schedule of cigarette silk-making workshop, etc.;

[0087] S2. Basic knowledge entity concepts and relationship generation: Based on the basic data extracted from the basic knowledge of the cigarette silk workshop simulation, the entity concepts of the knowledge graph are generated. The entity concepts include information such as the types of common equipment for cigarette silk making, the process flow of cigarette silk making, and the relationship between entity concepts (such as the upstream and downstream relationship of common equipment for cigarette silk making, the association of the process steps of cigarette silk making, etc.);

[0088] S3. Construct a multimodal knowledge graph for the simulation of a cigarette silk workshop: Integrate multimodal data (such as text, images, process data, etc.) obtained from different data sources to construct a multimodal knowledge graph, covering information such as equipment in the cigarette silk workshop, cigarette silk process flow, materials of tobacco leaves or modules, and the simulation platform of the cigarette silk workshop; this multimodal knowledge graph will serve as the basis for subsequent reasoning and model management;

[0089] Phase 2: Demand analysis and reasoning for the cigarette making workshop

[0090] S4. Cigarette silk workshop demand input (natural language): The cigarette silk workshop inputs its needs and problems through natural language (voice or text), such as optimizing the workshop production process or identifying equipment failures;

[0091] S5. Knowledge acquisition through intent recognition algorithm and slot filling mechanism: Using natural language processing (NLP) technology, through intent recognition algorithm and slot filling mechanism, the core requirements and key information are extracted from the demand input of the cigarette silk workshop, and then relevant data is obtained from the knowledge graph;

[0092] S6. Knowledge graph multi-step reasoning: Based on the constructed knowledge graph, multi-step reasoning is performed to analyze the needs of the cigarette silk workshop, identify the complex relationships between related entities, and determine the actual production links of the cigarette silk workshop that need to be optimized or the potential sources of failure;

[0093] Phase 3: Model recommendation and optimization

[0094] S7. Cigarette silk workshop simulation model feature extraction and data management module: automatically extract feature data from the existing cigarette silk workshop simulation model, and perform structured storage and management to ensure data integrity and traceability;

[0095] S8. Multidimensional indexing of workshop simulation model feature data: multidimensional indexing of the extracted cigarette silk workshop simulation model feature data is performed to construct an efficient retrieval structure to quickly locate and retrieve the cigarette silk workshop simulation model feature data;

[0096] S9. Identify the causes of failures in cigarette silk making operations and propose optimization measures for cigarette silk making workshops: Based on the results of knowledge reasoning, analyze and identify potential causes of failures in cigarette silk making operations, propose specific optimization suggestions or improvement measures for cigarette silk making workshops, and assist in optimizing the simulation model of cigarette silk making workshops;

[0097] Phase 4: Simulation model matching and recommendation

[0098] S10. Model recommendation and interactive learning: By matching the demand description of the cigarette silk-making workshop with the characteristics of the simulation model of the cigarette silk-making workshop, the most suitable simulation model is recommended, and the simulation model is dynamically adjusted and fed back according to the needs of the cigarette silk-making workshop, thereby realizing interactive learning and optimization of the simulation model of the cigarette silk-making workshop;

[0099] S11. Retrieve the simulation model category that best matches the needs of the cigarette silk-making workshop: Use multi-dimensional indexing and feature data to retrieve and match the simulation model category that best matches the needs of the cigarette silk-making workshop, and ensure that the recommended simulation model of the cigarette silk-making workshop can meet the specific operational needs of the cigarette silk-making workshop;

[0100] S12. Generate similarity recommendation and recommendation reasons: Based on the matching results of the cigarette silk workshop simulation model, generate a list of recommended cigarette silk workshop simulation models, and provide specific recommendation reasons, explain the reasons for selecting and recommending these cigarette silk workshop simulation models and the applicable scenarios of different working conditions of specific cigarette silk workshops;

[0101] Phase 5: System Quality Assessment

[0102] S13. Comprehensive quality assessment: Conduct a comprehensive quality assessment on the recommended cigarette silk-making workshop simulation model and system, check the performance of the cigarette silk-making workshop simulation model in terms of accuracy, completeness, consistency and practicality, and ensure the efficiency and reliability of the final cigarette silk-making workshop simulation model.

[0103] Furthermore, the operating method of the cigarette silk making workshop simulation model described in the present invention includes four sub-methods, namely, a method for extracting workshop modeling-related information based on the combination of voice interaction and knowledge graph retrieval, a reasoning method based on a multimodal workshop modeling knowledge graph, a method for constructing a multimodal workshop simulation basic knowledge graph, and a quality assessment method for a workshop simulation multimodal system.

[0104] (3) A method for extracting workshop modeling related information based on voice interaction and knowledge graph retrieval

[0105] A method for extracting workshop modeling related information based on voice interaction and knowledge graph retrieval, which includes the following steps: S1. Workshop demand input

[0106] S11. Input form: The cigarette making workshop inputs system requirements through voice or text to express its problems or requirements related to workshop modeling;

[0107] S12. Speech-to-text: Convert speech input to text using automatic speech recognition (ASR) technology;

[0108] S13. Text preprocessing: standardize the text, remove noise words, and perform word segmentation and lemmatization;

[0109] S2. Intent Recognition Algorithm

[0110] S21. Intent Recognition: Use NLP and intent recognition models (such as BERT or GPT) to parse text and identify core intent and key devices and parameters;

[0111] S22. Model training: Train the model based on historical data and equipment parameters to ensure accurate identification of the specific needs of the cigarette making workshop;

[0112] S3. Slot filling mechanism

[0113] S31. Slot identification: Identify detailed information (equipment type, parameters, and target) in the requirements of the cigarette making workshop;

[0114] S32. Slot filling: Based on the identified slots, the system fills the detailed content of the cigarette silk workshop requirements into the corresponding slots to form a structured query;

[0115] S4. Knowledge extraction and query

[0116] S41. Knowledge graph retrieval: Use structured queries (SPARQL, Cypher) to retrieve relevant information from the knowledge graph;

[0117] S42. Reasoning based on representation learning: Convert the search results of the knowledge graph into vectors, use graph neural networks for reasoning, and find the most relevant improvement suggestions;

[0118] S43. Result generation: Generate improvement suggestions based on the reasoning results and feed them back to the cigarette making workshop in natural language;

[0119] S5. Workshop feedback and result optimization

[0120] S51. Workshop feedback collection: Collect user feedback on suggestions to support confirmation, rejection or further optimization;

[0121] S52. Result optimization and learning: Use interactive learning mechanisms to continuously adjust the recommendation model based on feedback, update the knowledge graph and model parameters, and improve the accuracy and practicality of the system's recommendations.

[0122] (4) A method for constructing a basic knowledge graph for multimodal workshop simulation

[0123] A method for constructing a basic knowledge graph of multimodal workshop simulation includes the following steps:

[0124] S1. Extraction of basic knowledge of workshop simulation

[0125] S11. Data source selection: extract equipment, process flow and tobacco material data from the workshop's manufacturing execution system (MES), enterprise resource planning system (ERP), product lifecycle management system (PLM), etc.;

[0126] S12. Data extraction and collection: Use ETL tools to extract data, collect raw data through API, database query and file reading, and convert it into a standard format (such as CSV, JSON);

[0127] S13. Data cleaning and preprocessing: clean up missing values, outliers and duplicate data in the data, segment the text and remove stop words, standardize the image and numerical data to ensure data consistency and accuracy;

[0128] S2. Basic knowledge entity concepts and relationship generation

[0129] S21. Entity recognition and entity classification: Use NLP technology to identify entities such as equipment, process flow, tobacco materials, etc. in the text, and classify and annotate attributes according to categories;

[0130] S22. Relationship definition and relationship generation: Extract the relationship between entities from the data (such as the association between equipment and process), define the relationship type (such as "belongs to", "connects"), and describe the interaction between entities;

[0131] S23. Structured storage: Store entities and relationships in a relational database or graph database to form a preliminary knowledge graph structure;

[0132] S3. Constructing a multimodal knowledge graph for workshop simulation

[0133] S31. Multimodal data integration: Integrate multiple data types (such as text, images, sensor data) and extract features (such as text features, image features);

[0134] S32. Multimodal feature fusion: Use feature fusion techniques (such as weighted fusion and attention mechanism) to integrate features of different modalities and assign weights to retain key information;

[0135] S33. Graph construction and optimization: Use graph databases to map multimodal data into knowledge graphs, and improve the structure and application value of graphs through redundancy removal and node optimization.

[0136] S34. Multimodal knowledge reasoning: Use graph neural networks and other technologies to represent graphs, capture implicit relationships between entities, and support workshop process optimization, equipment fault diagnosis, and intelligent decision-making.

[0137] (5) A Reasoning Method Based on Multimodal Shop Floor Modeling Knowledge Graph

[0138] A reasoning method based on multimodal workshop modeling knowledge graph should include the following steps:

[0139] S1. Knowledge graph multi-step reasoning

[0140] S11. Input workshop requirements: parse the natural language input of the cigarette silk workshop, use the intent recognition algorithm to convert the requirements into a structured query format, and locate the relevant nodes and relationships;

[0141] S12. Preliminary query: Use graph query languages ​​(such as SPARQL and Cypher) to query in the multimodal knowledge graph to obtain nodes and relationships related to user needs;

[0142] S13. Multi-step reasoning: Perform recursive or iterative reasoning on preliminary query results, use graph neural networks (GNNs) and deep learning models (such as RGCN) to process embedding vectors, identify complex relationships, and generate reasoning paths and lists of related nodes;

[0143] S14. Result aggregation and optimization suggestions: Use rule engines or machine learning models to aggregate reasoning results, extract key related information, find commonalities and differences, and generate optimization suggestions;

[0144] S2. Causes of failure and optimization suggestions

[0145] S21. Fault cause identification: Use rule sets and historical case libraries to analyze reasoning results and identify potential fault causes;

[0146] S22. Generation of optimization suggestions: Retrieve successful solutions through case library and decision tree algorithm, generate optimization suggestions for current problems, and improve equipment operation or process flow;

[0147] S23. Feedback and verification: Apply the optimization suggestions to the simulation model or actual test, verify the effect and update the knowledge graph, and feed back the verification results to the cigarette making workshop;

[0148] S24. Cyclic improvement: Collect feedback from the cigarette silk workshop, adjust the inference model through reinforcement learning, continuously update the knowledge graph and optimization suggestions, and enhance the accuracy and practicality of the system.

[0149] (6) A quality assessment method for a workshop simulation multimodal system

[0150] A quality assessment method for a workshop simulation multimodal system includes the following steps:

[0151] S1. Accuracy Assessment

[0152] S11. Model prediction accuracy: Evaluate the prediction accuracy of the model, using indicators such as accuracy, mean square error, etc., test the prediction ability of the model through the validation set to ensure that the error is within an acceptable range;

[0153] S12. Recommendation accuracy: Compare the recommendation model with the actual usage according to the needs of the cigarette silk workshop, and use indicators such as precision, recall, and F1 value to evaluate the relevance of the recommendation to ensure that the recommendation meets the needs of the cigarette silk workshop;

[0154] S2. Completeness Assessment

[0155] S21. Data integrity: Use data audit tools to check the integrity of system data, calculate the coverage of the knowledge graph for all equipment and processes in the workshop, and ensure that there are no missing data;

[0156] S22. Model completeness: Check the parameters and settings of the recommended model and compare it with the standard model to ensure that all necessary parameters and components are included and the simulation process is fully described;

[0157] S3: Consistency Assessment

[0158] S31. Data consistency: Compare similar data from different data sources, use consistency detection tools to check conflicts and duplications, and ensure data consistency within the system;

[0159] S32. Model output consistency: By running the recommended model multiple times, the variance of the output results is calculated to check the consistency and stability of the output;

[0160] S4. Practicality evaluation

[0161] S41. Workshop satisfaction: Collect feedback and satisfaction scores through surveys of cigarette-making workshops to ensure that user satisfaction reaches the set target;

[0162] S42. Actual application effect: Test the recommended model in the actual workshop, measure the changes in key performance indicators, and verify the improvement effect of the model in actual application;

[0163] S5. Quality assessment report generation: Generate a comprehensive quality assessment report based on the assessment results, including accuracy, completeness, consistency and practicality indicators, and put forward specific suggestions for system optimization.

[0164] (7) An electronic device for simulating a cigarette making workshop model

[0165] The third object of the present invention is to provide an electronic device of a cigarette silk-making workshop simulation model, comprising a memory, a processor and a computer program stored in the memory and runnable on the processor; the memory is used to non-temporarily store computer-readable instructions, and the processor is used to run the computer-readable and writable instructions; when the computer-readable and writable instructions are run by the processor, the system for auxiliary construction and optimization of a cigarette silk-making workshop simulation model described in the first object is executed, and the operating method for running the cigarette silk-making workshop simulation model described in the second object is executed.

[0166] The memory is used to store data in the system, including multimodal data, knowledge graphs, sensor data and other data. It has high capacity and high speed and can support data storage, reading and writing functions; the storage devices include large-capacity storage servers (NVMe SSD, SATA SSD), distributed databases, distributed storage systems (Amazon S3, etc.), backup storage and disaster recovery equipment (tapes, RAID, cloud backup).

[0167] The processor is responsible for executing computing tasks and processing data, including functions such as speech recognition, natural language processing, deep learning reasoning, etc.; the processor includes a high-performance server or cloud computing platform and edge computing devices, specifically including a server cluster's CPU (Intel Xeon, AMD EPYC), GPU (NVIDIA Tesla, A100), and edge computing devices (NVIDIA Jetson, Intel Movidius).

[0168] (8) A non-transitory storage medium for a simulation model of a cigarette making workshop

[0169] The fourth object of the present invention is to provide a non-transitory storage medium for a simulation model of a cigarette silk-making workshop, which is used to store computer-readable and writable instructions, readable instructions and mixed-purpose storage. The computer-readable and writable instructions, readable instructions and mixed-purpose storage include instructions for executing a cigarette silk-making workshop simulation model auxiliary system described in the first object. When the computer-readable and writable instructions in the non-transitory storage medium are executed by a computer, the first object of a cigarette silk-making workshop simulation model auxiliary system is realized, which can realize the operation of any system, module and unit in embodiments 1-21; the computer-readable and writable instructions, readable instructions and mixed-purpose storage also include the operation method of running a cigarette silk-making workshop simulation model described in the second object, which can realize the operation of the operation method and sub-method of the cigarette silk-making workshop simulation model described in any one of embodiments 22-26.

[0170] The present invention provides a non-temporary storage medium for a simulation model of a cigarette silk-making workshop, which includes both a readable medium and a readable and writable medium, and also includes mixed-purpose storage, depending on the specific function and purpose. In the auxiliary system of a simulation model of a cigarette silk-making workshop described in the present invention, the non-temporary storage medium can store data and support data carriers accessed by the system. The readable and writable medium is a device used for real-time processing and storage of data by the system, and all belong to readable and writable media, and the system can read and write data on these media. The readable and writable media include solid-state drives (SSDs), mechanical hard disks (HDDs), distributed file systems, graph databases, cloud storage services, cache storage, database storage, and object storage; the readable medium supports writing, but the function focuses more on data reading, and the writing operation is less, and is used to store data that needs to be stored for a long time and is less modified. The readable medium includes tape storage and read-only optical disks; the mixed-purpose storage can be used for main reading or reading and writing according to different system requirements; such as HDD or cloud storage in backup and recovery systems; although these media are readable and writable, in the backup system, mixed-purpose storage is more used to read backup data, and the writing operation is only performed when the backup occurs.

[0171] (9) Technical Effects

[0172] First, compared with the prior art, the cigarette silk workshop simulation model auxiliary system of the present invention has the following beneficial and significant technical effects:

[0173] A simulation model auxiliary system for a cigarette-making workshop achieves precise support and optimization of workshop production through comprehensive simulation model management, data extraction and knowledge construction, intelligent interaction and problem diagnosis, system quality assessment, dynamic update and continuous optimization. The system can automatically extract explicit and implicit features of the simulation model, and combine the knowledge graph to achieve accurate model recommendation and optimization. By integrating data from multiple information systems, the system constructs a multimodal knowledge graph, which provides powerful decision support for workshop optimization analysis and fault diagnosis.

[0174] In addition, the intelligent dialogue management module uses natural language processing (NLP), intent recognition and knowledge graph reasoning technology to accurately parse user input, diagnose workshop problems and provide optimization suggestions. The system's model recommendation and interactive learning mechanism ensures that the recommended simulation model is constantly close to actual needs through feature library construction and classification, similarity evaluation and dynamic optimization. The quality assessment module comprehensively verifies the accuracy, completeness and practicality of the system to ensure a high degree of consistency between the simulation system and actual production.

[0175] In addition, the system integrates version management and incremental update mechanisms, supports real-time updates and compatibility verification, and keeps the simulation system up to date. By integrating multiple technical means (such as NLP, image processing, machine learning, etc.), the system greatly improves the simulation model's understanding of complex process flows and automated decision-making capabilities, providing comprehensive intelligent support for workshop production.

[0176] Secondly, compared with the prior art, the operation method of the cigarette silk workshop simulation model of the present invention has the following beneficial and significant technical effects:

[0177] The operation method of a simulation model of a cigarette silk workshop described in the present invention significantly improves the intelligent management and optimization level of the workshop simulation model through systematic data extraction, knowledge graph construction, demand analysis, model recommendation and optimization, and quality evaluation. Specifically, the method constructs a comprehensive knowledge graph through the extraction and integration of multimodal data, provides an accurate data basis for the reasoning and management of the simulation model, and ensures the high consistency between the model and the actual production environment; then, by using NLP and knowledge graph reasoning technology, the system can accurately analyze the workshop needs and perform multi-step reasoning, automatically identify optimization opportunities and potential failure points in the production process, and improve the accuracy of problem diagnosis and optimization suggestions; then, through model feature extraction, indexing and interactive learning mechanism, the system can recommend the most suitable simulation model according to the actual needs of the workshop, and support dynamic adjustment and continuous optimization to ensure that the recommended model is always close to the actual needs; finally, by evaluating the accuracy, completeness, consistency and practicality of the model, the efficiency and reliability of the simulation model are guaranteed, providing solid support for the efficient operation of the workshop.

[0178] Third, compared with the prior art, the method for extracting workshop modeling related information based on the combination of voice interaction and knowledge graph retrieval described in the present invention has the following beneficial and significant technical effects:

[0179] The method described in the present invention is a method for extracting workshop modeling related information based on the combination of voice interaction and knowledge graph retrieval, which realizes the accurate extraction and intelligent optimization of the modeling related information of the cigarette silk workshop through the steps of voice recognition, intent recognition, slot filling, knowledge graph retrieval and reasoning. Specifically, the method uses ASR and NLP technology to accurately convert the voice requirements of the workshop into text and parse the core intent, identify key equipment and parameters, and ensure that the system can understand and respond to complex workshop modeling requirements; then through the structured query of the knowledge graph and the reasoning of the graph neural network, the system can deeply analyze the workshop problems and provide the most relevant optimization suggestions from multiple dimensions, improving the intelligent level of problem diagnosis and decision support; finally, the feedback from the workshop is collected and the recommendation model is continuously adjusted through interactive learning, so as to realize the dynamic update of the knowledge graph and model parameters, improve the applicability of the system in different scenarios and the accuracy of the suggestions, and provide more refined and personalized support for the modeling of the cigarette silk workshop.

[0180] Fourthly, compared with the prior art, the method for constructing a basic knowledge graph for multimodal workshop simulation according to the present invention has the following beneficial and significant technical effects:

[0181] The method for constructing a multimodal workshop simulation basic knowledge map described in the present invention integrates various data from systems such as MES, ERP, PLM, etc., systematically extracts and processes basic knowledge such as equipment, process flow and tobacco materials, and realizes comprehensive cognition and management of workshop simulation models. Specifically, the method ensures the accuracy and consistency of multi-source data through ETL tools and data cleaning technology, laying a solid foundation for the construction of knowledge maps; using NLP technology and graph databases, entities such as equipment, process flow and their relationships are systematically stored and classified to form a structured knowledge map, improving the accessibility and relevance of information; through feature fusion and graph neural network technology, multimodal data such as text and images are effectively integrated, enhancing the expression ability of knowledge maps, and supporting the accurate simulation of complex processes in cigarette silk workshops; finally, the implicit relationship between entities in the map is captured and reasoned, providing intelligent solutions for equipment fault diagnosis, process optimization and decision support, and significantly improving the management and optimization capabilities of workshop simulation systems.

[0182] Fifth, compared with the prior art, the reasoning method based on the multimodal workshop modeling knowledge graph described in the present invention has the following beneficial and significant technical effects:

[0183] The reasoning method based on the knowledge graph of multimodal workshop modeling described in the present invention realizes intelligent diagnosis and optimization of the process flow of the cigarette silk workshop through multi-step reasoning, fault identification, optimization suggestion generation, feedback and loop improvement. Specifically, the method uses intent recognition and graph query to convert natural language requirements into structured queries, and uses graph neural networks and deep learning models to perform multi-step reasoning to accurately identify implicit faults and complex relationships and improve diagnostic accuracy; based on rule engines, case libraries and decision tree algorithms, targeted optimization suggestions are generated to optimize equipment parameters and process flows, and improve production efficiency and quality; through simulation verification and actual testing, the knowledge graph is dynamically updated to achieve closed-loop improvement of optimization suggestions, and reinforcement learning further enhances the adaptability and practicality of the model; finally, multimodal data and deep learning technology are integrated to achieve a comprehensive understanding and continuous optimization of complex production environments, and improve the management and optimization level of simulation models.

[0184] Sixth, compared with the prior art, the operation method of the cigarette silk workshop simulation model of the present invention has the following beneficial and significant technical effects:

[0185] The operating method of a simulation model of a cigarette silk workshop described in the present invention comprehensively evaluates the system through four dimensions: accuracy, completeness, consistency and practicality. Specifically, the method evaluates the prediction ability of the model through indicators such as accuracy and mean square error, ensures that the error is within an acceptable range, and verifies the accuracy and relevance of the recommended model; uses data auditing and model parameter verification to check the complete coverage of data and models, ensures that there is no data missing in the system and that the simulation model contains all necessary components; compares data sources and multiple model outputs, checks the consistency and stability of data and model outputs, ensures that there is no conflict in the internal data of the system, and the output results are consistent and reliable; finally, through workshop surveys and actual application tests, evaluates the actual improvement effect of the model and workshop satisfaction, ensures the effectiveness of the simulation model in a real environment and the recognition of the cigarette silk workshop, generates a quality report based on the evaluation results, puts forward improvement suggestions, and continuously optimizes system performance and application effects.

[0186] Seventh, compared with the prior art, the electronic device of the cigarette silk workshop simulation model of the present invention has the following beneficial and significant technical effects:

[0187] The electronic device of a simulation model of a cigarette silk-making workshop described in the present invention integrates high-performance memory and processor to improve the construction, optimization and operation efficiency of the simulation model. Among them, the memory supports high-speed storage and reading of multimodal data and knowledge graphs, ensuring the stable and efficient operation of the system in a big data environment; distributed databases and storage systems such as Amazon S3 provide high availability and security of data. The processor includes high-performance servers, cloud computing platforms and edge computing devices, which can efficiently perform tasks such as speech recognition, natural language processing and deep learning reasoning; GPU accelerators such as NVIDIA Tesla and A100 accelerate the training and reasoning of deep learning models, and edge computing devices realize real-time processing and response on site. The electronic device of the simulation model of the cigarette silk-making workshop as a whole, through the combination of hardware and algorithms, realizes the intelligent and automated management of the workshop simulation model, and helps the efficient operation of the cigarette silk-making workshop.

[0188] Eighth, compared with the prior art, the non-temporary storage medium of the cigarette silk workshop simulation model of the present invention has the following beneficial and significant technical effects:

[0189] The non-temporary storage medium of the cigarette silk workshop simulation model of the present invention provides a flexible data storage and access solution by integrating readable and writable media (such as SSD, HDD, distributed file system), readable media (such as tape, optical disk) and mixed-purpose storage (such as HDD in backup system, cloud storage). Among them, the readable and writable media supports real-time data processing and efficient reading and writing, the readable media ensures long-term data preservation and security, and the mixed-purpose storage meets the backup and recovery requirements; and the combination of multiple storage media realizes efficient data management, fast access and reliable backup, and improves the performance and data processing capability of the simulation model. BRIEF DESCRIPTION OF THE DRAWINGS

[0190] Figure 1 It is a flow chart of the overall system of a simulation model auxiliary system and operation method, electronic equipment and storage medium of a cigarette silk workshop;

[0191] Figure 2 It is a schematic diagram of the operation of a workshop simulation model marking management module of a cigarette silk workshop simulation model auxiliary system and operation method, electronic equipment and storage medium;

[0192] Figure 3 It is a schematic diagram of the operation of a workshop simulation basic knowledge extraction module of a cigarette silk workshop simulation model auxiliary system and operation method, electronic equipment and storage medium;

[0193] Figure 4 It is a schematic diagram of the operation of a workshop simulation multimodal knowledge graph construction module of a cigarette silk workshop simulation model auxiliary system and operation method, electronic equipment and storage medium;

[0194] Figure 5 It is a schematic diagram of the operation of a workshop simulation knowledge interaction and problem diagnosis module of a cigarette silk workshop simulation model auxiliary system and operation method, electronic equipment and storage medium;

[0195] Figure 6 The present invention is a schematic diagram of the operation of a workshop system quality assessment module of a cigarette silk workshop simulation model auxiliary system and operation method, electronic equipment and storage medium. DETAILED DESCRIPTION

[0196] In order to enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention is clearly and completely described below in conjunction with the accompanying drawings of the present invention. Based on the embodiments in this application, other similar embodiments obtained by ordinary technicians in this field without making any creative work should all fall within the scope of protection of this application.

[0197] It should be noted that the terms used herein are only for describing specific embodiments, and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should also be understood that the terms "include" and "have" and any of their variations are intended to cover non-exclusive inclusions, for example, the process, method, system, module, medium or equipment comprising a series of steps or units need not be limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, systems, modules, medium or equipment.

[0198] The cigarette silk-making workshop simulation model auxiliary system of the present invention comprises five independent modules, namely, a cigarette silk-making workshop simulation model tag management module, a cigarette silk-making workshop simulation basic knowledge extraction module, a cigarette silk-making workshop simulation multimodal knowledge graph construction module, a cigarette silk-making workshop simulation knowledge interaction and problem diagnosis module, and a cigarette silk-making workshop system quality assessment module;

[0199] The cigarette silk workshop simulation model tagging management module includes a simulation model feature extraction unit, a simulation model annotation unit, and a simulation model feature data management unit, which are used to accurately define and describe the cigarette silk workshop simulation model, and combine the marked information with the knowledge graph, so as to provide a cigarette silk workshop model recommendation function based on the knowledge graph for the dialogue simulation platform system;

[0200] The basic knowledge extraction module for cigarette silk workshop simulation includes a cigarette silk workshop data extraction unit and a cigarette silk workshop data processing unit, which are used to extract key rules and data of the cigarette silk workshop, and obtain basic knowledge data related to the cigarette silk workshop simulation platform by constructing entity relationships, ternary relationships and knowledge inheritance hierarchical structures of knowledge graphs;

[0201] The multimodal knowledge graph construction module for simulating the cigarette silk-making workshop includes a basic graph construction unit and a feature fusion unit, which are used to integrate and analyze data from various information systems of the cigarette silk-making workshop (such as the manufacturing execution system MES, the enterprise resource planning system ERP, the product lifecycle management system PLM, and data in process documents related to cigarette silk-making), and to construct a comprehensive knowledge graph that supports optimization analysis of the operation of the cigarette silk-making workshop, fault diagnosis of cigarette silk-making operations, and decision support;

[0202] The cigarette silk workshop simulation knowledge interaction and problem diagnosis module includes a dialogue management and knowledge query unit, a problem diagnosis and optimization suggestion unit, a model recommendation and interactive learning unit, which is used to accurately parse the natural language input of the cigarette silk workshop, extract key knowledge using the intent recognition algorithm and slot filling mechanism, and use knowledge graph reasoning to capture the complex relationship between related entities in the cigarette silk operation, perform multi-step relationship reasoning, and convert the actual demand description of the cigarette silk workshop operation into an expression form of graph query. Then, using subgraph matching and query parsing methods, it extracts improvement suggestions for related production scheduling, quality management, process formula management, and tobacco leaf or tobacco material supply chain management from the graph, and comprehensively obtains the final knowledge interaction and problem diagnosis;

[0203] The cigarette silk-making workshop system quality assessment module includes a storage device and processing equipment accuracy verification unit, a tobacco processing rule integrity check unit, a tobacco production practicality assessment unit, a tobacco finished product quality index calculation unit, and a system dynamic update support unit. After the system is built, it is used to comprehensively evaluate the overall quality of the tobacco silk-making workshop logistics simulation multimodal system to ensure that the actual application of the system in the cigarette silk-making operation shows accuracy, completeness, consistency and practicality, thereby effectively supporting the optimization and decision-making process of the cigarette silk-making workshop simulation model.

[0204] Example 1 A specific implementation method of a simulation model auxiliary system for a cigarette making workshop

[0205] In a specific embodiment, a simulation model auxiliary system for a cigarette silk-making workshop aims to provide a complete solution for the construction, management, optimization and quality assessment of simulation models in industrial workshops. The simulation model auxiliary system for a cigarette silk-making workshop integrates functional modules such as simulation model tagging and management, basic knowledge extraction, multimodal knowledge graph construction, knowledge interaction and problem diagnosis, and system quality assessment to support efficient simulation model management and optimization decisions. Through this system, engineers can accurately define and describe the workshop simulation model, automatically extract key workshop knowledge, build an efficient multimodal knowledge graph, use a dialogue system to achieve intelligent knowledge interaction and problem diagnosis, and comprehensively evaluate and optimize the quality of the entire system.

[0206] The cigarette cutting workshop simulation model auxiliary system comprises a simulation model marking and management module, a workshop simulation basic knowledge extraction module operation, a workshop simulation multimodal knowledge graph construction module, a workshop simulation knowledge interaction and problem diagnosis module and a workshop system quality assessment module.

[0207] The operation mode of the simulation model marking and management module is as follows: the system first starts the simulation model marking and management module, automatically identifies and extracts key feature data from various simulation models in the workshop, including equipment objects, physical properties, equipment upstream and downstream relationships, spatial distribution data, process parameters and other data information; then uses the model annotation unit to annotate the extracted data with explicit information (such as model name, type, update date, layout image, etc.) and implicit information (such as layout description method, whether the model is a local simulation, etc.); allows the cigarette silk workshop to input supplementary text descriptions to further enrich the model features; and then stores the extracted and annotated feature data in the model feature data management unit, and realizes rapid retrieval and query by establishing a multidimensional index structure. Therefore, the simulation model marking and management module ensures that all simulation models can be systematically defined and stored through precise marking and management, providing basic data support for the operation of subsequent modules.

[0208] The operation mode of the workshop simulation basic knowledge extraction module is as follows: the system calls the data extraction unit to extract relevant data of workshop equipment and processes from MES, ERP, PLM systems and related process documents, including basic material data, processing rules, production scheduling data, equipment parameters and process section information; then, the data processing unit splits, reduces noise and structures the extracted data. Concepts and relationships are automatically generated for structured data using predefined templates, and unstructured data is processed using the BiLSTM sequence labeling model and SURF feature extraction algorithm. Therefore, the workshop simulation basic knowledge extraction module can build a comprehensive workshop basic knowledge base, providing rich and accurate basic data for the subsequent construction of a multimodal knowledge graph.

[0209] The operation mode of the workshop simulation multimodal knowledge graph construction module is as follows: in the multimodal knowledge graph construction module, the system first starts the basic graph construction unit, performs preliminary structural processing on the workshop, logistics simulation platform and process data, and generates multiple basic knowledge graphs; then, the feature fusion unit performs weighted fusion on multimodal data such as text and images through a weighted multimodal feature fusion model to generate candidate concepts; then the attention mechanism is used to disambiguate the candidate concepts to ensure that the concepts are unique and accurate, and then these candidate concepts are linked to the multimodal knowledge graph to form a unified and comprehensive knowledge graph. Therefore, the workshop simulation multimodal knowledge graph construction module directly constructs a multimodal knowledge graph that integrates workshop equipment, process flow and simulation platform, supports complex knowledge query and reasoning, and provides a strong data foundation for the system's knowledge interaction and problem diagnosis.

[0210] The operation mode of the workshop simulation knowledge interaction and problem diagnosis module is as follows: the dialogue management and knowledge query unit parses the natural language input of the user (cigarette silk workshop) through NLP technology, and identifies the intention and key needs of the user (cigarette silk workshop); then the system retrieves data related to the user (cigarette silk workshop) query from the multimodal knowledge graph, and extracts key information such as equipment type and process flow through the intention recognition algorithm and slot filling mechanism; furthermore, the problem diagnosis and optimization suggestion unit uses rule reasoning, case comparison and representation learning methods to analyze the user (cigarette silk workshop) input, identify potential problems in the simulation model, and provide optimization suggestions; finally, the system combines the model recommendation and interactive learning unit to recommend the most suitable simulation model according to the needs of the user (cigarette silk workshop), and supports the user (cigarette silk workshop) to dynamically adjust and optimize the model. Therefore, the workshop simulation knowledge interaction and problem diagnosis module realizes intelligent knowledge interaction and problem diagnosis, helping the cigarette silk workshop to make more accurate decisions in the creation, optimization and application of simulation models, and improve overall efficiency.

[0211] The operation mode of the workshop system quality assessment module is as follows: the system starts the quality assessment module, verifies the accuracy of the concepts and their relationships in the system through the accuracy verification unit, and ensures that the simulation model description is consistent with the actual workshop situation; then the integrity check unit is used to evaluate the system to detect whether all key concepts and relationships are covered, and whether there are internal contradictions or inconsistencies; then the practicality evaluation unit is used to simulate the actual application scenario, test the system's responsiveness to user queries and support for simulation models; then the quality indicator calculation unit is used to generate a quality report of the system, including indicators such as accuracy, recall rate, F1 value and coverage, and the support unit is dynamically updated to ensure that the system can adapt to new data and changes at any time. Therefore, the workshop system quality assessment module ensures that the system provides high-quality support and optimization functions in actual applications through comprehensive quality assessment; the evaluation results provide a basis for subsequent system adjustments and optimizations, and improve the work efficiency and decision-making ability of simulation engineers.

[0212] The simulation model auxiliary system of a cigarette silk-making workshop is comprehensively operated in an actual workshop, and through the seamless integration of various modules, it realizes the precise management and optimization support of the workshop simulation model. The construction and use of multimodal knowledge graphs enable the system to perform complex knowledge reasoning and query, meeting the needs of the cigarette silk-making workshop in simulation model optimization and decision-making. The cigarette silk-making workshop can easily query the simulation model information, obtain problem diagnosis and optimization suggestions through the interactive dialogue system; the multimodal knowledge graph generated by the system provides comprehensive and accurate knowledge support for the cigarette silk-making workshop, and improves the application effect of the simulation model. The introduction of the system quality evaluation module ensures the overall performance and application quality of the system, so that the workshop simulation multimodal system can maintain efficient operation in a dynamically changing production environment. Therefore, a simulation model auxiliary system of a cigarette silk-making workshop realizes the all-round management, optimization and evaluation of the workshop simulation model, supports the various needs of the cigarette silk-making workshop simulation model in the modeling and decision-making process, and greatly improves the application effect and production efficiency of the workshop simulation model.

[0213] Furthermore, the present invention discloses a cigarette silk-making workshop simulation model auxiliary system, wherein the five different modules are a cigarette silk-making workshop simulation model tag management module, a cigarette silk-making workshop simulation basic knowledge extraction module, a cigarette silk-making workshop simulation multimodal knowledge graph construction module, a cigarette silk-making workshop simulation knowledge interaction and problem diagnosis module, and a cigarette silk-making workshop system quality assessment module; specifically, the main tasks, connection relationships and transfer relationships between the five modules, and the coordination relationships between the modules are as follows:

[0214] Module 1: Cigarette Making Workshop Simulation Model Marking Management Module

[0215] The main task of the cigarette cutting workshop simulation model tag management module is to accurately define, describe and tag the simulation model, and associate these tags with the knowledge graph to form the basic structure data of the model;

[0216] The connection and transfer relationship between the cigarette silk-making workshop simulation model marking management module and the cigarette silk-making workshop simulation basic knowledge extraction module is that the marking management module provides the basic knowledge extraction module with a preliminary definition and marking of the model to form a basic framework of the data structure; and then guides subsequent data extraction through the annotation of equipment, process flow, procedures, etc.

[0217] The connection and transfer relationship between the cigarette silk-making workshop simulation model marking management module and the cigarette silk-making workshop simulation multimodal knowledge graph construction module is that the annotation data of the marking management module will be passed to the multimodal knowledge graph construction module as the basic concept information of nodes and relationships in the knowledge graph.

[0218] The connection and transmission relationship between the cigarette silk-making workshop simulation model marking management module and the cigarette silk-making workshop simulation knowledge interaction and problem diagnosis module is that the marking management module provides marked model information to the interaction and problem diagnosis module, so that the system can understand the cigarette silk-making workshop input and provide relevant model query services.

[0219] The coordination relationship between the cigarette silk workshop simulation model marking management module and the other four modules is the marking management module

[0220] It is the foundation of the entire system and ensures that the labeling information of each simulation model can be effectively integrated into the knowledge graph, so that subsequent data extraction, query, and diagnosis processes are based on accurate model definition and labeling.

[0221] Example 2 Specific implementation method of the marking management module of the cigarette silk workshop simulation model

[0222] In a specific embodiment, the production process of a cigarette shred workshop is complex, involving multiple equipment and process links. The simulation model is used to model and analyze the operation of the entire workshop. In order to effectively connect the simulation model with subsequent knowledge extraction, graph construction, and problem diagnosis, each model must be accurately labeled and described, and the corresponding relationship between the model and the knowledge graph must be established.

[0223] The tag management module requires the steps of model definition and description, tagging and classification, data structure formation, association with the knowledge graph, and integration with the knowledge interaction and problem diagnosis module to be implemented.

[0224] (1) Implementation examples of model definition and description

[0225] In the marking management module of the cigarette making workshop simulation model, it is first necessary to accurately define each cigarette making equipment and its process flow.

[0226] The operation mode of the model definition and description is as follows: first, the metadata of all equipment (such as equipment number, model, location, etc.) is obtained from the equipment database of the cigarette shredding workshop; then, according to the production process, the process links are sorted out, and the processes are divided in sequence (such as shredding, shredding, blending, flavoring, etc.); then, the process parameters corresponding to each process are defined (such as temperature, humidity, time, etc.). For example, the simulation model object of the "shredder" is defined as "equipment 001", and its function is described as "cutting the tobacco leaves into shreds of uniform width according to the set requirements", and the corresponding process parameters are "cutting width, cutting speed, etc.".

[0227] (2) Examples of Labeling and Classification

[0228] In the cigarette silk workshop simulation model tagging management module, the defined equipment and process models need to be tagged for subsequent data extraction and knowledge graph construction.

[0229] The marking and classification process is as follows: first, a unique identifier (ID) is assigned to each equipment and process model, such as equipment ID, process flow ID, etc.; then each model is classified into categories, such as "equipment type", "process flow type", "process type", etc.; and then each equipment and process model is labeled with attributes, such as the attributes of "equipment 001" can be labeled as "type: shredder; function: cutting tobacco leaves; process parameters: equipment size, design flow, cutting width"; finally, the system will automatically generate these tags and save them in the database.

[0230] (3) Example of Formation of Data Structure

[0231] According to the definition and labeling of equipment and process flow, the basic data structure of the model is formed. The data structure includes the following contents:

[0232] The data structure needs to include the following contents: a structure tree of the equipment, which represents the relationship and hierarchy between the equipment, such as the connection relationship between the shredder and other equipment; then, the process flow chain is set up to describe the sequence of process steps and their logical relationship; the association between the equipment and the process flow is through model marking to clarify the correspondence between the equipment and the process, such as defining a structure tree to show the sequential relationship between "equipment 001" and "equipment 002" (shredded wire drying machine).

[0233] (4) Implementation Examples Associated with Knowledge Graphs

[0234] The tag management module transfers the generated equipment and process tag data to the cigarette making workshop simulation multimodal knowledge graph construction module.

[0235] The knowledge graph construction module uses these labeled data as the basic information of nodes and relationships to construct a knowledge graph containing multi-level information such as equipment, processes, and parameters; for example, in the knowledge graph, "Equipment 001 (shredding machine)" is used as a node, and "cutting width" is used as the attribute of the node. There is a "sequential execution" relationship between "Equipment 002 (shredded wire drying machine)" and "Equipment 001".

[0236] (5) Implementation Example of Integration with Knowledge Interaction and Problem Diagnosis Module

[0237] The tag management module also provides the tagged simulation model information to the knowledge interaction and problem diagnosis module. When the cigarette shredding workshop makes a query or diagnosis request, the system can quickly locate the relevant equipment or process model based on the tag information and query the relevant information through the knowledge graph. For example, if the workshop inputs "How many types of faults does the shredding machine have?", the system can provide relevant fault types based on the tagged model information of "shredding machine" to assist in problem diagnosis.

[0238] Module 2: Cigarette Making Workshop Simulation Basic Knowledge Extraction Module

[0239] The main task of the basic knowledge extraction module for the simulation of the cigarette silk workshop is to extract relevant data from the workshop's MES, ERP, PLM and other systems, integrate information such as equipment and process flow, and provide basic data for knowledge graphs and simulation models.

[0240] The connection and transmission relationship between the basic knowledge extraction module for cigarette silk-making workshop simulation and the model tag management module for cigarette silk-making workshop simulation is to receive preliminary model tags such as equipment and process from the tag management module, and improve them by docking with actual production data.

[0241] The connection and transfer relationship between the basic knowledge extraction module for cigarette silk workshop simulation and the multimodal knowledge graph construction module for cigarette silk workshop simulation is that the extracted data is transferred to the knowledge graph construction module to generate specific information of nodes and edges in the graph. This module is responsible for integrating the extracted multi-source data and transferring it to the graph construction module.

[0242] The connection and transfer relationship between the basic knowledge extraction module of cigarette silk workshop simulation and the knowledge interaction and problem diagnosis module of cigarette silk workshop simulation is that the basic knowledge extraction module provides specific process and equipment data to the interaction module to ensure that relevant information can be extracted when users query.

[0243] The coordination relationship between the basic knowledge extraction module for simulation of cigarette silk workshop and the other four modules is that the basic knowledge extraction module provides core data support for the entire system, ensuring that the entity concepts in the labels and graphs of the simulation model have a real production data basis, so that subsequent analysis and optimization are based on accurate workshop data.

[0244] Example 3 Specific implementation of the basic knowledge extraction module for cigarette silk workshop simulation

[0245] In a specific embodiment, the production of a cigarette silk workshop involves multiple information systems, such as MES (manufacturing execution system), ERP (enterprise resource planning system), PLM (product lifecycle management system), etc. These systems store important data such as equipment, process flow, production parameters, etc. In order to build simulation models and knowledge graphs, data must be extracted and integrated from these systems. The task of the basic knowledge extraction module for the simulation of cigarette silk workshops is to extract basic knowledge from these information systems and integrate them into the simulation model and knowledge graph.

[0246] The basic knowledge extraction module for simulation of the cigarette silk-making workshop requires multi-source data connection and extraction, data cleaning and conversion, data integration, data annotation and transmission, transmission and docking with the multimodal knowledge graph construction module, and collaboration with the knowledge interaction and problem diagnosis module to be realized.

[0247] The multi-source data connection and extraction is to first establish a connection with the MES, ERP, PLM and other systems through the basic knowledge extraction module to obtain the real-time and historical data of the workshop production equipment and process flow. The operation mode of the multi-source data connection and extraction is: extracting equipment operation status data (such as equipment startup time, downtime, maintenance records, etc.) from the MES system; obtaining material management information (such as incoming materials, inventory, consumption, etc.) from the ERP system; extracting the design and technical parameters of equipment and process flow (such as equipment model, process flow specification, key process parameters, etc.) from the PLM system. For example, the basic knowledge extraction module extracts the average startup time, cutting width, cutting speed and other production data of "Slicer 001" from the MES system; obtains the current tobacco leaf inventory information from the ERP system; and extracts the technical parameters of "Slicer 001" from the PLM system.

[0248] The data cleaning and conversion is done by extracting data from various systems, which are often in inconsistent formats and contain redundant or missing data, and need to be cleaned and standardized. The operation mode of the data cleaning and conversion is: removing duplicate and invalid data, for example, deleting duplicate equipment maintenance records; filling missing data, for missing values ​​of some non-critical parameters, using interpolation or default values ​​to fill in; converting timestamps and units of different systems to make the data consistent in time and space. For example, converting the UNIX epoch time in the PLM system to a standard time format to ensure data consistency.

[0249] The data integration is to integrate the data extracted from different systems to form a unified data view of the equipment and process flow of the cigarette shredding workshop. The operation mode of the data integration is: compare the real-time data of the MES system with the design parameters of the PLM system, check whether the equipment is running according to the design parameters, and identify abnormal situations; then match the inventory data of the ERP system with the material consumption data of the MES system to form a material consumption chain; through equipment identification and process identification, the data of different systems are linked to form a complete production process data set. For example, the basic knowledge extraction module compares the startup time and actual cutting width of "shredding machine 001" with the designed cutting width in the PLM system, and finds that the actual cutting width has a large deviation, which is marked as a potential process abnormality.

[0250] The data labeling and transmission is to receive the preliminary equipment and process model labels transmitted from the simulation model labeling management module through the basic knowledge extraction module, and further improve these labels through the extracted production data. The improved labeling data will be transmitted to other modules in a standard format. The operation mode of the data labeling and transmission is as follows: first, the equipment is labeled, and the equipment operation status is labeled according to the extracted data, such as "normal operation" and "shutdown maintenance" and other data information; then the process flow is labeled, and the status of each link in the process flow is labeled according to the extracted data, such as "normal temperature" and "cutting width exceeds the standard" and other data information. For example, "cutter 001" is labeled as "normal operation" in the basic knowledge extraction module, but there is a slight deviation in the cutting width, which is labeled as "abnormal width".

[0251] The transmission and docking with the multimodal knowledge graph construction module is to transmit the integrated and annotated data to the multimodal knowledge graph construction module through the basic knowledge extraction module. The knowledge graph construction module uses these data to generate the information of nodes and edges in the graph. The transmission and docking with the multimodal knowledge graph construction module is as follows: first, the equipment node is transmitted and docked, and the basic knowledge extraction module provides the graph construction module with the operating status, technical parameters and other information of each device as the node attributes in the graph; then the process flow node is transmitted and docked, and the operating parameters and status of each link in the process flow are provided for the graph to construct the process flow chain. For example, the knowledge graph construction module generates a node in the graph containing attributes such as the equipment operating status and parameter deviation based on the data of "Slicer 001" provided by the basic knowledge extraction module.

[0252] The collaboration with the knowledge interaction and problem diagnosis module is to provide the integrated equipment and process data to the knowledge interaction and problem diagnosis module through the basic knowledge extraction module to support user queries and problem diagnosis. The operation mode of the collaboration with the knowledge interaction and problem diagnosis module is: when the cigarette silk workshop queries through the system, the interaction module can return relevant information through the data provided by the basic knowledge extraction module. For example, when the cigarette silk workshop queries "cutter 001", the system can quickly find the basic information of the simulation model involving the equipment, obtain the historical parameters and process anomalies of the equipment in the simulation calendar from the basic knowledge extraction module, and help the workshop to evaluate the equipment status and diagnose problems.

[0253] Module 3: Cigarette Making Workshop Simulation Multimodal Knowledge Graph Construction Module

[0254] The main task of the multimodal knowledge graph construction module for the simulation of the cigarette silk workshop is to construct a multimodal knowledge graph, and to structure the information such as equipment, processes, and procedures in the form of nodes and edges to support subsequent query, reasoning, and optimization.

[0255] The connection and transmission relationship between the cigarette silk-making workshop simulation multimodal knowledge graph construction module and the cigarette silk-making workshop simulation model marking management module is to receive the marking information from the marking management module, convert these markings into nodes and relationships in the knowledge graph, and form the knowledge structure of the workshop simulation model.

[0256] The connection and transfer relationship between the multimodal knowledge graph construction module for cigarette silk-making workshop simulation and the basic knowledge extraction module for cigarette silk-making workshop simulation is to integrate the equipment parameters, process flow and other data obtained from the basic knowledge extraction module to form a multimodal knowledge graph covering multiple data types.

[0257] The connection and transfer relationship between the cigarette silk-making workshop simulation multimodal knowledge graph construction module and the cigarette silk-making workshop simulation knowledge interaction and problem diagnosis module is that the constructed knowledge graph provides query services for the interaction module, enabling it to retrieve information based on the knowledge graph, and perform reasoning and optimization.

[0258] The coordination relationship between the multimodal knowledge graph construction module of the cigarette silk workshop simulation and the other four modules is that the knowledge graph construction module forms a complete simulation model graph structure by integrating the annotated data and basic knowledge. As the core support, the knowledge graph provides the data foundation and reasoning ability for interaction, query and reasoning.

[0259] Example 4 Specific implementation method of the multimodal knowledge graph construction module for simulating a cigarette making workshop

[0260] In a specific example, the production process of a cigarette silk workshop involves multiple equipment, process flows, and production links, and this information exists in different data sources. In order to improve the efficiency of production management, optimize process flows, and fault diagnosis, it is necessary to structure the equipment, process, and process data through a multimodal knowledge graph to achieve a comprehensive understanding of the operating status of a complex workshop.

[0261] The task of the multimodal knowledge graph construction module for simulating a cigarette making workshop is to convert the data from the tag management module and the basic knowledge extraction module into nodes and edges in the knowledge graph, forming a graph covering multidimensional information to support query, reasoning and optimization.

[0262] The multimodal knowledge graph construction module for simulating a cigarette silk-making workshop requires data integration and preprocessing, creation of nodes and edges of the knowledge graph, integration of multimodal data, construction of reasoning rules of the knowledge graph, collaborative work between the knowledge graph and the interactive module, and dynamic updating of the graph to be effectively implemented.

[0263] The data integration and preprocessing is to first receive the annotation information from the simulation model tag management module and the equipment and process flow data from the basic knowledge extraction module through the knowledge graph construction module; these data may come from different systems, so they need to be integrated and preprocessed before entering the knowledge graph. The operation method of the data integration and preprocessing is: unify the tag information of the equipment (such as equipment name, number, operating status) and its operating parameters (such as temperature, humidity, cutting speed); and then integrate the tag information of the process flow (such as process name, process number, dependency) and process parameters (such as processing time, temperature, humidity, etc.). For example, "cutter 001" from the tag management module is used as the equipment node, and the parameters such as cutting speed and cutting width extracted from the basic knowledge extraction module are used as the attributes of the node.

[0264] The creation of nodes and edges of the knowledge graph is through the knowledge graph construction module according to the diversity and relevance of data, taking different equipment, process flow and parameters as nodes, and defining the association relationship (edge) between them. The operation mode of creating nodes and edges of the knowledge graph is as follows: first create equipment nodes, each equipment as a node in the graph, and the attributes of the node include equipment number, equipment type, status information, technical parameters, etc.; then create process flow nodes, each step of the process flow as a node in the graph, and the order and dependency between process flows are represented by edges; finally create parameter nodes; the key parameters of each process and equipment (such as temperature, humidity, cutting width, etc.) are independent nodes, and are connected to the corresponding equipment or process flow nodes through edges; finally define the relationship between equipment and process, and connect the equipment node and process flow node by creating edges to represent the execution relationship between equipment and process flow. For example, the equipment node of "cutter 001" is created in the knowledge graph, and is connected to the process node of "cutter process" through edges. At the same time, the operating parameters of the cutter (cutting speed, cutting width, etc.) are attached to the equipment node as attributes, or independent parameter nodes are created to connect with the equipment node.

[0265] The integration of multimodal data means that the knowledge graph construction module not only includes structured equipment and process information, but also integrates data from different modes, such as text descriptions, image data, etc. The operation mode of the integration of multimodal data is as follows: first, unstructured data such as equipment maintenance records and technical documents are obtained in the basic knowledge extraction module, and they are used as text nodes in the knowledge graph, and associated with the corresponding equipment or process nodes; if there is image data of equipment in the workshop (such as equipment drawings and monitoring screens), these image information can be used as multimodal data nodes in the graph and associated with the equipment nodes. For example, the knowledge graph not only contains the structured data of "Slicer 001", but also connects the maintenance history of the equipment through text nodes, or connects the process drawings of the equipment through image nodes.

[0266] The reasoning rules of the knowledge graph are constructed to support reasoning and optimization. The knowledge graph construction module also defines some reasoning rules that can derive new knowledge or discover potential anomalies based on existing graph information. The operation mode of the reasoning rules of the knowledge graph is: first, based on the reasoning of the equipment status, according to the historical operation data of the equipment, it is inferred when the equipment needs maintenance or replacement; then based on the optimization suggestions of the process parameters, according to the historical data of the process flow, the optimal process parameter settings are derived to reduce the defective rate in the production process; then, anomaly detection is performed to infer potential failure risks based on the status and operating parameters of the equipment. For example, the knowledge graph can find through reasoning rules that if the cutting width of "cutter 001" often exceeds the standard range in the recent production cycle, it is inferred that the equipment may be worn and needs maintenance.

[0267] The collaborative work of the knowledge graph and the interactive module is that the knowledge graph construction module provides support for the knowledge interaction and problem diagnosis module; the interactive module can query the relevant information of the equipment and process through the knowledge graph, and even provide problem diagnosis and optimization suggestions through inference rules. The collaborative work of the knowledge graph and the interactive module is as follows: when the workshop queries the operating status of a certain equipment, the interactive module quickly obtains the current status, historical data and related process parameters of the equipment through the knowledge graph; in the problem diagnosis process, the interactive module can find the operating anomalies of the equipment based on the nodes and edge information in the graph, and prompt the possible causes of the fault. For example, when the cigarette silk workshop queries the status of "Slicer 001", the knowledge graph returns the historical operating data, maintenance records and associated process parameters (such as cutting width, temperature) of the equipment to help users quickly understand the status of the equipment.

[0268] The dynamic update of the graph is a dynamic update system. As the production in the workshop proceeds, the equipment status, process flow and parameters will continue to change. The knowledge graph construction module will regularly obtain the latest data from the basic knowledge extraction module and update the nodes and edges in the graph. The dynamic update of the graph operates in the following way: first monitor the operating status and process parameters of the equipment, and update the properties of the equipment nodes in real time; when the process flow changes, update the process nodes and their corresponding dependencies; and then update the reasoning rules and node relationships in the graph based on the equipment's maintenance records, fault history and other data. For example, when the "Slicer 001" is maintained, the maintenance status of the equipment node in the graph will be updated, recording the new maintenance time and the information on the replacement of parts.

[0269] In actual applications, the tobacco silk workshop can query the equipment and process status of the entire workshop through the knowledge interaction module. By connecting with the multimodal knowledge graph, the system can display the operation status of each device, the process flow, and the relationship with other equipment. For example, when there is a problem with the product quality of a certain process, the tobacco silk workshop can track the operation status of related equipment through the knowledge graph, find out the possible cause of the failure, and provide optimization suggestions, ultimately improving production efficiency and reducing losses.

[0270] The simulation multimodal knowledge graph construction module of the cigarette silk workshop effectively integrates equipment, processes, procedures and other multimodal data to form a multi-level and multi-dimensional knowledge graph, providing strong support for workshop management, query, reasoning and optimization.

[0271] Module 4: Cigarette Making Workshop Simulation Knowledge Interaction and Problem Diagnosis Module

[0272] The main task of the cigarette silk workshop simulation knowledge interaction and problem diagnosis module is to parse user queries through natural language processing and knowledge graph retrieval, perform reasoning and problem diagnosis based on the knowledge graph, and provide model optimization suggestions.

[0273] The connection and transfer relationship between the cigarette silk-making workshop simulation knowledge interaction and problem diagnosis module and the cigarette silk-making workshop simulation multimodal knowledge graph construction module is that the knowledge interaction and problem diagnosis module directly relies on the knowledge graph to retrieve the data queried by the user; when the user enters a query, the system searches for relevant equipment and process flow information in the knowledge graph and returns the query results.

[0274] The connection and transfer relationship between the cigarette silk workshop simulation knowledge interaction and problem diagnosis module and the cigarette silk workshop simulation model marking management module is that the knowledge interaction and problem diagnosis module relies on the model information in the marking management module, and supports the user's query request by understanding and retrieving the labeled data.

[0275] The connection and transfer relationship between the cigarette silk workshop simulation knowledge interaction and problem diagnosis module and the cigarette silk workshop simulation basic knowledge extraction module is to ensure the accuracy of problem diagnosis. The knowledge interaction and problem diagnosis module requires the equipment data and process flow in the basic knowledge module to support reasoning and optimization suggestion generation.

[0276] The connection and transmission relationship between the cigarette silk workshop simulation knowledge interaction and problem diagnosis module and the cigarette silk workshop system quality assessment module is that after the knowledge interaction and problem diagnosis module gives problem diagnosis and optimization suggestions, it will pass feedback to the system quality assessment module for evaluation to verify the feasibility and effectiveness of the optimization suggestions.

[0277] The coordination relationship between the tobacco silk workshop simulation knowledge interaction and problem diagnosis module and the other four modules is that the knowledge interaction and problem diagnosis module is the interface between the user and the system. By integrating the knowledge graph and basic data, it realizes automated reasoning from user query to problem diagnosis, helping users to quickly obtain effective optimization suggestions.

[0278] Example 6 Specific implementation of the knowledge interaction and problem diagnosis module for the simulation of the cigarette silk workshop

[0279] In a specific embodiment, in the daily production of a cigarette silk workshop, it is necessary to query the status of a certain equipment, detailed information of the process flow, or diagnose and optimize certain production problems (such as equipment failure, process abnormality) at any time. In order to improve work efficiency, the knowledge interaction and problem diagnosis module realizes direct interaction between the cigarette silk workshop and the system through natural language processing technology, and uses multimodal knowledge graphs and basic data for automated reasoning to provide diagnosis and optimization suggestions.

[0280] The cigarette silk workshop simulation knowledge interaction and problem diagnosis module requires natural language query parsing, knowledge graph retrieval and data matching, problem diagnosis and reasoning, provision of optimization suggestions, feedback and collaboration with the system quality assessment module, and user interaction and feedback implementation.

[0281] The natural language query parsing is a query request input by the cigarette silk workshop through text or voice, such as "What are the process parameters of the 001 model shredder?" or "Are there any ready-made processes for using the 001 model shredder in the existing local simulation model?". The knowledge interaction and problem diagnosis module first uses natural language processing (NLP) technology to parse the query input by the user. The natural language query parsing is operated in the following way: first, semantic analysis is performed, and the sentence input by the cigarette silk workshop is segmented and parsed, and key entities (such as equipment name, process flow name) and query intent (such as status query, abnormality troubleshooting) are identified; then keyword extraction is performed to extract key query words from the input of the cigarette silk workshop, such as "model", "shredder", "process parameters" and other data information. For example, if the cigarette silk workshop inputs the query "What are the process parameters of the 001 model shredder?", the system will parse it as a "process parameter query" for the "shredder 001" equipment.

[0282] The knowledge graph retrieval and data matching is that after the analysis is completed, the knowledge interaction and problem diagnosis module retrieves the relevant equipment or process flow information through the multimodal knowledge graph. The operation mode of the knowledge graph retrieval and data matching is: first retrieve the equipment node, and locate the equipment node corresponding to "cutter 001" in the knowledge graph according to the keywords in the query of the cigarette silk workshop; then retrieve the associated data, and obtain the current operating status, historical data, maintenance records and other information of the equipment according to the association relationship in the knowledge graph; and then query the process flow.

[0283] The problem diagnosis and reasoning is that when the cigarette silk workshop inquires about the problem diagnosis (such as "Why is the shredding process abnormal?"), the knowledge interaction and problem diagnosis module will reason based on the data in the knowledge graph and the basic knowledge extraction module, and automatically provide the diagnosis results. The operation mode of the problem diagnosis and reasoning is: first analyze the equipment and process parameters, and the system identifies potential abnormal points according to the equipment operation parameters in the knowledge graph and the set values ​​of the process flow. If the "cutting width" parameter exceeds the normal range, the system will mark the equipment as abnormal; then the reasoning rules are applied, and the system infers the possible causes of the fault through predefined reasoning rules (such as the association between the equipment status and the fault mode, and the parameter optimization rules in the process flow); then generate the diagnosis results, and based on the reasoning results, the system generates a problem diagnosis report, pointing out the specific abnormal causes in the equipment or process flow. For example, when the cigarette silk workshop inquires "Why is the shredding process abnormal?", the system analyzes the relevant equipment operation parameters (such as "cutting width exceeds the standard"), and infers that the abnormality may be caused by blade wear based on the historical operation data of the equipment.

[0284] The optimization suggestions are provided based on the diagnosis results, and the knowledge interaction and problem diagnosis module will also generate optimization suggestions. The operation mode of providing optimization suggestions is as follows: First, based on historical data and best practices, the system generates optimization suggestions for equipment or processes by analyzing historical data and optimal process parameter settings in the knowledge graph. For example, the system can suggest adjusting the "cutting width" to a reasonable range; then seek a solution to the problem. If the problem involves equipment failure, the system will give corresponding maintenance suggestions, such as replacing equipment parts, adjusting process parameters and other data information. For example, after detecting that the cutting width of "cutter 001" is abnormal, the system recommends "checking and replacing the blade" and at the same time feeding back the recommended optimal cutting width parameter setting (such as "0.7 mm") to the user.

[0285] The feedback and collaboration with the system quality assessment module is that when the knowledge interaction and problem diagnosis module generates optimization suggestions, the suggestions will be passed to the system quality assessment module for verification. The feedback and collaboration with the system quality assessment module is operated as follows: first, the optimization plan is verified. The system quality assessment module evaluates whether the suggestion is feasible and has practical benefits by comparing the existing data with the optimization suggestions; then feedback on the implementation results is given. Once the optimization plan is implemented, the system quality assessment module will monitor the actual effect of the workshop production, and pass the feedback data back to the knowledge graph to update the information and reasoning rules in the graph. If the cigarette silk workshop implements the suggestion to replace the blade, the system quality assessment module monitors the production data after the replacement. After confirming that the cutting width has returned to normal, the system will record this feedback in the knowledge graph to improve future diagnosis and optimization rules.

[0286] The user interaction and feedback is to apply the knowledge interaction and problem diagnosis module to provide a friendly interactive interface for the cigarette silk workshop. Users can query the equipment status, process flow or diagnose problems at any time; the cigarette silk workshop can also provide feedback on the optimization suggestions provided by the system to help the system further learn and optimize reasoning capabilities. The operation mode of the user interaction and feedback is: first, real-time response is performed, and the system can respond to user queries in real time and provide the latest equipment status and process information; then, feedback from the cigarette silk workshop is collected, and the cigarette silk workshop can provide feedback on the diagnosis and optimization suggestions (such as "the suggestion is effective" or "the problem is not solved") to help the system continuously improve the accuracy of diagnosis. For example, when the cigarette silk workshop queries "What are the best parameters for the shredding process?", the system can provide the best setting value based on the knowledge graph (such as "cutting width 0.7 mm, cutting speed 550 rpm"), and the cigarette silk workshop can feedback the parameter adjustment effect after adjustment according to the actual production situation.

[0287] In actual production, the tobacco silk workshop inputs queries through the knowledge interaction and problem diagnosis module, such as "What problems did the shredder 001 encounter recently?" The system parses out the key equipment and problem types, retrieves the equipment's maintenance records and operating parameters through the knowledge graph, and generates diagnostic results. If the system finds that the cutting width exceeds the preset range, it infers that blade wear is the possible cause of the problem. Subsequently, the system provides optimization suggestions for replacing the blade, and generates a report to pass to the system quality assessment module for verification. Finally, the simulation model of the tobacco silk workshop implements the suggestions and optimizes the production process. The tobacco silk workshop simulation knowledge interaction and problem diagnosis module realizes seamless interaction between users and the system. Based on the reasoning function of the knowledge graph, it provides efficient query, diagnosis and optimization suggestion services for workshop management, improving workshop production efficiency and the accuracy of problem handling.

[0288] Module 5: Cigarette Making Workshop System Quality Assessment Module

[0289] The main task of the cigarette silk workshop system quality assessment module is to evaluate the performance of the entire system, calculate indicators such as accuracy, recall rate, coverage rate, and evaluate the actual application effect of the system.

[0290] The connection and transmission relationship between the cigarette silk-making workshop system quality assessment module and the cigarette silk-making workshop simulation knowledge interaction and problem diagnosis module is that the system quality assessment module evaluates the accuracy, effectiveness and practicality of its optimization suggestions based on the feedback data of the interaction and diagnosis module; and then provides the workshop with quality feedback on the optimization suggestions by calculating various performance indicators in the reasoning process.

[0291] The connection and transfer relationship between the cigarette silk-making workshop system quality assessment module and the cigarette silk-making workshop simulation multimodal knowledge graph construction module is that the system quality assessment module can evaluate the performance of the knowledge graph in practical applications and ensure the integrity of the graph structure and the accuracy of reasoning; and the graph data is an important basis for system quality assessment.

[0292] The connection and transmission relationship between the cigarette silk-making workshop system quality assessment module and the cigarette silk-making workshop simulation model marking management module is that the system quality assessment module can evaluate whether the marking of the simulation model is clear and accurate, and the quality of the marking information will directly affect the subsequent map construction and optimization process.

[0293] The connection and transmission relationship between the cigarette silk-making workshop system quality assessment module and the cigarette silk-making workshop simulation basic knowledge extraction module is that the system quality assessment module ensures the accuracy of basic knowledge extraction and the integrity of data, and compares it with the actual operation effect of the simulation model to find out the optimization space.

[0294] The coordination relationship between the cigarette silk workshop system quality assessment module and the other four modules is that the quality assessment module ensures the efficiency and accuracy of the data flow and reasoning process of each module by evaluating the actual application effect of the entire system, puts forward improvement suggestions, and continuously optimizes the performance of the system.

[0295] Example 6 Specific implementation of the system quality assessment module of the cigarette silk workshop

[0296] In a specific embodiment, the simulation system of the cigarette silk workshop includes multiple modules, from data extraction to knowledge graph construction, to problem diagnosis and optimization suggestions, each module has its own functions and data interactions. In order to ensure the effectiveness and accuracy of the entire system, the performance of each module and the overall effect of the system must be comprehensively evaluated through the system quality evaluation module. This includes evaluating the accuracy of optimization suggestions, the effectiveness of knowledge graph reasoning, the clarity of labeled data, and the integrity of basic data, so as to ensure the efficient operation of the entire system.

[0297] The cigarette silk-making workshop system quality assessment module can only be realized through data collection and feedback aggregation, system performance evaluation index calculation, knowledge graph application effect evaluation, optimization suggestion evaluation and verification, data labeling and basic knowledge extraction evaluation, and system overall performance analysis and improvement suggestions.

[0298] The data collection and feedback summary is that the system quality assessment module first collects feedback data from each module. The operation mode of the data collection and feedback summary is: first, the execution results of the optimization suggestions are fed back in the simulation knowledge interaction and problem diagnosis module, including information such as the accuracy of the suggestions and the effectiveness of troubleshooting; then in the multimodal knowledge graph construction module, the reasoning accuracy of the knowledge graph, the integrity of the nodes and edges, and the coverage of the graph in practical applications; then in the simulation model tag management module, the clarity and accuracy of the tag information, and whether the tag structure can support the reasoning and diagnosis requirements of the knowledge graph; finally, in the basic knowledge extraction module, the integrity and accuracy of the data extracted from the data source (such as MES, ERP, PLM). For example, the system collects maintenance feedback data from "cutter 001", reasoning results extracted from the knowledge graph, and equipment parameter data provided by the basic knowledge extraction module.

[0299] The calculation of the system performance evaluation index is based on the aggregated feedback data, and the system quality evaluation module calculates a series of performance indicators to comprehensively evaluate the overall performance of the system. The operation mode of the calculation of the system performance evaluation index is as follows: first calculate the accuracy rate: used to measure the correctness of the optimization suggestions or problem diagnosis, that is, how many suggestions generated by the system are accurate; then calculate the recall rate, that is, measure the system's ability to find problems, when the real problems that the system can detect account for the proportion of the total number of actual problems; then calculate the coverage rate, indicating how many equipment and process nodes in the workshop are covered by the knowledge graph to ensure the extensiveness of the knowledge graph; then calculate the F1 value to find the harmonic mean of the accuracy rate and the recall rate, which is used to comprehensively evaluate the system performance. If the system has generated 100 optimization suggestions in the past month, and 85 of them have effectively solved the problem, the accuracy of the optimization suggestion is 85%.

[0300] The evaluation of the application effect of the knowledge graph is an analysis of the actual application effect of the multimodal knowledge graph by the system quality evaluation module. The operation mode of the evaluation of the application effect of the knowledge graph is: first, the accuracy of reasoning is to determine whether the nodes and edges in the knowledge graph can support effective reasoning and whether the status of the equipment and process flow can be correctly associated; then the structural integrity of the workshop is evaluated, that is, whether the knowledge graph contains all the key equipment and process nodes in the workshop, and whether the number of nodes and edges is sufficient to support the analysis and diagnosis needs of the system; then the reasoning performance of different scenarios is evaluated, and the reasoning efficiency and performance of the knowledge graph in different application scenarios, whether it can provide optimization suggestions and fault diagnosis within a reasonable time. For example, whether the connection between the "cutter 001" node in the evaluation graph and its related process nodes, fault records, and operating parameters is complete, and whether the system can accurately reason through these nodes and edges.

[0301] The evaluation and verification of the optimization suggestion is that the system quality evaluation module evaluates its accuracy and practicality based on the feedback of the optimization suggestion execution results. The evaluation and verification of the optimization suggestion is operated as follows: first, compare the equipment operation status, process efficiency, and production quality changes before and after the optimization suggestion to confirm whether the optimization suggestion has played a practical role in improvement; then analyze the long-term effect of the optimization suggestion through historical data and quality feedback, such as improved production efficiency, reduced failure rate and other data information; finally verify the operability of the suggestion, that is, whether the suggestion is easy to implement, whether additional resources are required and other data information. If the system recommends replacing the blade of "cutter 001" and adjusting the cutting width, evaluate whether the cutting quality is improved, whether equipment failures are reduced, and whether production efficiency is improved after the operation.

[0302] The evaluation of data labeling and basic knowledge extraction is that the system quality evaluation module also evaluates the labeling clarity and accuracy of the simulation model labeling management module, as well as the data integrity of the basic knowledge extraction module. The evaluation of data labeling and basic knowledge extraction is operated as follows: first check whether the labeled data is clear enough and whether it can support the subsequent knowledge graph construction and reasoning operations, such as whether the status of the equipment and the sequence of the process flow are accurately marked; then evaluate whether the data extracted by the basic knowledge extraction module is complete and accurate, whether it can cover the key equipment and process parameters of the workshop, and whether there is any missing or abnormal data. For example, evaluate whether the basic knowledge extraction module has accurately extracted all the key operating parameters of "Slicer 001" from the MES system, and whether these parameters have been fully passed to the knowledge graph.

[0303] The overall performance analysis and improvement suggestions of the system are based on all evaluation indicators and feedback data. The system quality assessment module generates an overall performance report of the system, points out the advantages and disadvantages of each module, and puts forward specific improvement suggestions. The operation mode of the overall performance analysis and improvement suggestions of the system is: first, some nodes or edges in the knowledge graph are supplemented or adjusted to improve the reasoning performance; then the data source or extraction method of the basic knowledge extraction module is adjusted to ensure the real-time and accuracy of the data; finally, the simulation model labeling rules are optimized to ensure that the labeling information is clearer and reduce the errors caused by labeling errors. If the system finds that certain parameters of the "wire cutting process" are often missing, the evaluation module will recommend adjusting the extraction frequency of the basic knowledge extraction module to ensure data integrity.

[0304] In actual applications, the system quality assessment module conducts a comprehensive assessment of the simulation system of the cigarette silk workshop. For example, within one production cycle, the system generated 200 optimization suggestions with an accuracy of 90%, a recall rate of 85%, and an F1 value of 87.5%. The assessment also found that some equipment nodes in the knowledge graph lacked associated process parameters, which resulted in some equipment problems not being discovered in time during reasoning. Based on these results, the assessment module recommends strengthening the real-time performance of data extraction, supplementing the missing nodes and edges in the knowledge graph, and optimizing the marking rules of the equipment. The assessment report ultimately helped the cigarette silk workshop adjust and optimize the system, improving the efficiency of problem discovery and resolution in the production process. The system quality assessment module of the cigarette silk workshop can comprehensively evaluate the system performance, ensure the efficient and coordinated operation of each module, and continuously improve the accuracy and optimization capabilities of the simulation system.

[0305] Furthermore, in the cigarette thread-making workshop simulation model auxiliary construction and optimization system described in the present invention, each module has a different working principle, and the synergistic effect realizes efficient simulation model construction, optimization and diagnosis; specifically, the key technologies, main functions and working principles of the five different modules are as follows:

[0306] Module 1: Cigarette Making Workshop Simulation Model Marking Management Module

[0307] The key technologies of the cigarette cutting workshop simulation model tag management module are model feature extraction and annotation, knowledge graph association, and model feature data management;

[0308] The main function of the cigarette cutting workshop simulation model tagging management module is to ensure the definition of the simulation model is clear through tagging and management, and to support the recommendation function of the model in the dialogue system;

[0309] The working principle of the cigarette cutting workshop simulation model tagging management module is that the module is responsible for accurately defining, describing and managing the simulation model. By marking the models of equipment, process flow and procedures in the workshop, a simulation model consistent with the actual production process is generated, and these tagging information is associated with the knowledge graph to ensure that the model is consistent with the equipment and process data in the actual workshop, thereby providing an accurate basis for subsequent model recommendation and optimization.

[0310] Module 2: Cigarette Making Workshop Simulation Basic Knowledge Extraction Module

[0311] The key technologies of the basic knowledge extraction module for the simulation of the cigarette silk workshop are data extraction and integration, multimodal feature fusion, noise reduction processing, and basic knowledge graph construction;

[0312] The main function of the basic knowledge extraction module for cigarette silk workshop simulation is to extract key knowledge of workshop equipment and processes from multiple systems, build and integrate basic knowledge graphs, and provide data support for the simulation system;

[0313] The working principle of the basic knowledge extraction module for the simulation of the cigarette silk-making workshop is that the module extracts key data of equipment, processes and production processes from the workshop's manufacturing execution system (MES), enterprise resource planning system (ERP), product lifecycle management system (PLM) and related process documents; then, through data processing and noise reduction, the data is cleaned and standardized, and finally a basic knowledge map is generated.

[0314] Module 3: Cigarette Making Workshop Simulation Multimodal Knowledge Graph Construction Module

[0315] The key technologies of the multimodal knowledge graph construction module for the simulation of the cigarette making workshop are natural language processing, intent recognition, knowledge graph retrieval, and structured data extraction;

[0316] The main function of the multimodal knowledge graph building block of the cigarette silk workshop simulation is to support users' query and optimization of the simulation model through natural language interaction and provide detailed information on related equipment and processes;

[0317] The working principle of the multimodal knowledge graph construction module for the simulation of the cigarette silk workshop is that the module uses natural language processing (NLP) technology to parse and identify the user's language input and extract key information from it. The knowledge graph retrieval engine is used to extract relevant structured data from the constructed multimodal knowledge graph, locate the equipment and process flow involved in the user's query, and feed back the relevant information to the workshop; this module can understand the workshop's needs and provide targeted query results.

[0318] Module 4: Cigarette Making Workshop Simulation Knowledge Interaction and Problem Diagnosis Module

[0319] The cigarette silk workshop simulation knowledge interaction and problem diagnosis module is an intelligent auxiliary system, which aims to help engineers better identify and solve potential problems in the simulation model, so as to optimize the workshop production efficiency and product quality; the key technology of the cigarette silk workshop simulation knowledge interaction and problem diagnosis module is the use of a variety of advanced reasoning technologies, including rule-based reasoning, case-based reasoning, representation learning reasoning and knowledge reasoning, to provide powerful diagnosis and optimization suggestions for the workshop simulation model platform;

[0320] The main function of the cigarette cutting workshop simulation knowledge interaction and problem diagnosis module is to identify problems in the simulation model, give optimization suggestions based on knowledge reasoning, and improve equipment parameters and process flow;

[0321] The working principle of the cigarette silk workshop simulation knowledge interaction and problem diagnosis module is that the module automatically identifies problems in the simulation model and puts forward optimization suggestions through rule-based reasoning, case-based reasoning and representation learning-based reasoning methods combined with the user's demand description; then, by analyzing the parameters, processes and equipment status of the simulation model, the system can quickly identify potential problems and generate improvement suggestions, helping engineers optimize the model and improve the accuracy and efficiency of the simulation model.

[0322] Module 5: Cigarette Making Workshop System Quality Assessment Module

[0323] The key technologies of the quality assessment module of the cigarette silk workshop system are accuracy verification, integrity check, practicality assessment, dynamic update and optimization;

[0324] The main function of the cigarette silk workshop system quality assessment module is to ensure that the simulation model can be applied in real scenarios and support effective decision-making of engineers through systematic quality control and assessment methods;

[0325] The working principle of the cigarette silk workshop system quality assessment module is that the module ensures the accuracy, completeness, consistency and practicality of the system in practical applications through multi-dimensional evaluation methods. By performing quality analysis on the running results of the simulation model, key indicators such as accuracy, recall rate, F1 value and coverage are calculated to generate system quality reports; these reports help engineers adjust and optimize the simulation model to ensure its reliability in practical applications.

[0326] Furthermore, the specific units and functions of the five different modules of the cigarette thread making workshop simulation model auxiliary system described in the present invention are as follows:

[0327] Module 1: Cigarette Making Workshop Simulation Model Marking Management Module

[0328] The cigarette cutting workshop simulation model marking management module comprises a simulation model feature extraction unit, a simulation model annotation unit, and a simulation model feature data management unit.

[0329] Unit 1: Simulation Model Feature Extraction Unit

[0330] The simulation model feature extraction unit automatically identifies and extracts key feature data from different cigarette silk-making workshop simulation models, including different parameters such as known equipment objects and physical properties in the cigarette silk-making workshop simulation model, upstream and downstream relationships of equipment in cigarette silk-making operations, spatial distribution data of equipment in cigarette silk-making operations, and process parameters set by the workshop simulation model; the feature extraction unit can directly access the workshop simulation model structure, three-dimensional geometric data, input of the cigarette silk-making workshop and simulation log in the simulation model.

[0331] The main task of the simulation model feature extraction unit is to automatically identify and extract key feature data from the simulation models of different workshops;

[0332] The technical means adopted by the simulation model feature extraction unit are structured data analysis technology, computer vision and image processing technology, natural language processing technology and data fusion and integration technology;

[0333] Specifically, the structured data parsing technology needs to be completed through data parsing, pattern matching and template recognition; the data parsing is to automatically parse the equipment objects and their physical properties, equipment upstream and downstream relationships, equipment spatial distribution data and process parameters in the model by analyzing the structured data (such as XML, JSON, CAD file format, etc.) in the simulation model file of the cigarette silk workshop; the pattern matching and template recognition is to use predefined templates or pattern matching technology to identify the common equipment types and process parameters of the cigarette silk workshop in the simulation model; the template can be customized according to the industry characteristics of different workshops, so as to accurately identify the needs of the cigarette silk workshop for simulation model;

[0334] Specifically, the computer vision and image processing technology needs to be completed through image recognition and deep learning models; the image recognition is to use computer vision technology (such as SURF, ORB and other feature extraction algorithms) to extract spatial layout features and the three-dimensional shape of the equipment for the three-dimensional geometric data and two-dimensional layout diagram in the simulation model; the deep learning model is to use convolutional neural network (CNN) deep learning algorithm to perform object detection and classification of image data, and identify the type and location of equipment from complex three-dimensional data.

[0335] Specifically, the natural language processing technology needs to be completed through text data processing and named entity recognition (NER);

[0336] The text data processing is to use natural language processing (NLP) technology to analyze the text description and simulation log input by the cigarette silk-making workshop, and then use a sequence labeling model (such as BiLSTM-CRF) to extract the process parameters and equipment information of the key cigarette silk-making workshop; the named entity recognition (NER) is to extract key information such as equipment name, model, parameters, etc. from the text through named entity recognition technology.

[0337] Specifically, the data fusion and integration technology needs to be completed through multi-source data fusion; the multi-source data fusion is to combine the feature data types (structured data, image data, text data) from different sources, and use the data fusion algorithm to integrate multimodal information to improve the accuracy and comprehensiveness of feature extraction of cigarette silk workshops.

[0338] The operation process of the simulation model feature extraction unit is as follows:

[0339] S1. Data collection and preprocessing: Collect simulation model files, two-dimensional images, three-dimensional geometric data, and text data; perform preliminary cleaning and preprocessing on the data to ensure data integrity and consistency.

[0340] S2. Feature extraction and analysis: Use structured data analysis technology to parse simulation model files and extract structured data such as equipment objects and process parameters; use computer vision technology to extract equipment location, shape and spatial distribution information from image data; use NLP technology to extract semantic information related to equipment and processes from text data.

[0341] S3. Data fusion and feature vector construction: Use data fusion technology to integrate feature data from different sources into a multidimensional feature vector; then optimize the fusion result according to the weight and credibility of the data source to generate the final feature vector set.

[0342] The result generated by the simulation model feature extraction unit after the operation process is: a complete data set containing all equipment and process features in the cigarette cutting workshop; this data set can be used for optimization, classification and recommendation of the simulation model.

[0343] In the actual operation of the simulation model of the cigarette silk workshop, the complete data set obtained by the simulation model feature extraction unit contains the necessary feature data, which can be used to optimize the layout of the cigarette silk workshop, improve the process flow and support the interaction of the dialogue system. The complete data set can optimize the equipment layout of the workshop and reduce the material transmission time and energy consumption by analyzing the spatial distribution data and process sequence between the equipment. The complete data set makes optimization suggestions based on the extracted process parameters and equipment performance data to improve production efficiency. The complete data set provides rich background data support for the dialogue system, enabling it to more accurately answer user queries and provide optimization suggestions.

[0344] Example 7 Implementation of the Simulation Model Feature Extraction Unit in the Cigarette Making Workshop Simulation Model

[0345] In a specific embodiment, a cigarette-making workshop applies the simulation model of the present invention, wherein the operation of the model feature extraction unit undergoes the process of model data analysis, image processing and recognition, text data processing, multimodal data fusion and feature vector construction.

[0346] The process of simulation model data parsing is as follows: First, read and parse structured data from the simulation model file of the cigarette silk workshop (such as files in XML and JSON formats), and extract the equipment objects and their related physical properties (such as temperature, pressure, speed, etc.). Then, use the pattern matching algorithm to automatically identify the upstream and downstream relationships of the equipment; for example, identify the position of each device in the process flow, and determine which equipment is the upstream processing equipment and which is the downstream processing equipment. Furthermore, parse the spatial distribution data, such as the three-dimensional coordinate information of the equipment in the cigarette silk workshop, and analyze the specific location and layout of the equipment in the workshop through geometric data. Finally, extract the process parameters specified by the simulation model of the cigarette silk workshop, such as blending weight, blending order, processing time, etc.; these data usually come from the attribute fields or preset parameters in the simulation model of the cigarette silk workshop.

[0347] In the tobacco making workshop, equipment objects include cutters, dryers, flavoring machines, etc. By analyzing the physical properties of these devices, the operating parameters of each device (such as temperature setting value, working speed, etc.) can be obtained; the upstream and downstream relationship analysis of the equipment can determine the path of the tobacco material flow, for example, identifying the process sequence of the cutter before the dryer, and the dryer before the flavoring machine.

[0348] The image processing and recognition process requires the application of computer vision technology. The specific extraction process is as follows: First, use computer vision technology, such as convolutional neural network (CNN), to detect objects in the two-dimensional image in the simulation model (such as the layout diagram of the cigarette making workshop), so as to identify the position and shape of the cigarette making equipment (such as cutters, dryers, and flavoring machines) in the workshop layout. Next, extract the spatial characteristics of the single equipment from the three-dimensional geometric data to identify its position and shape in three-dimensional space; for example, extract the three-dimensional model of each equipment and identify its shape, size, position and other features. Finally, analyze the spatial distribution relationship between the equipment through computer vision algorithms (such as SURF, ORB, etc.) to obtain the relative position relationship and layout information of the equipment.

[0349] In the tobacco cutting workshop, image processing technology can automatically detect the installation location and spatial distribution of equipment from the two-dimensional plane layout map, and identify the specific location of the drying machine in the workshop and the relative position of its adjacent equipment (such as the cutting machine and the flavoring machine) through the deep learning model. The processing of the three-dimensional model can help analyze the optimization degree of the workshop layout, identify whether the equipment is in the right position, and whether there is a spatial bottleneck that may affect production efficiency.

[0350] Text data processing requires the application of natural language processing (NLP) technology. The specific extraction process is as follows: use NLP technology to analyze the text description of the simulation log and the input of the cigarette silk workshop, and extract key information related to the simulation model of the cigarette silk workshop, such as equipment name, model, process parameters and operating instructions; through named entity recognition (NER) technology, identify specific equipment names (such as "thin plate drying machine No. 1"), equipment models (such as "model SH625"), processing rules (such as "feed temperature 80℃, drying drum speed 10r / min") and other information from the text.

[0351] In the cigarette making workshop, simulation logs and user text input contain descriptions of equipment status (such as "sheet tobacco drying machine No. 1 is working normally") and operating instructions (such as "start the sheet tobacco drying machine and set the temperature to 65°C"); NLP technology can automatically extract key information from these descriptions and form structured data for subsequent analysis.

[0352] Multimodal data fusion requires the application of multimodal data fusion algorithms (such as Bayesian networks and Dempster-Shafer evidence theory). The specific fusion process is as follows: First, the structured data (equipment objects, physical attributes, upstream and downstream relationships, etc.), the spatial feature data obtained by image processing (equipment location, shape, spatial distribution information) and the semantic information obtained by text processing (equipment name, process parameters, operating instructions, etc.) are integrated. Then, the multimodal data fusion algorithm is used to map the feature data from different sources into a unified feature vector space to construct a complete feature vector set. Finally, a feature weighted fusion strategy is adopted to optimize the result of feature fusion according to the credibility and weight of different data sources to ensure the accuracy and consistency of data fusion.

[0353] In the cigarette making workshop, multimodal data fusion integrates the feature data from model file parsing, image recognition and text analysis into a complete data set. For example, the different physical properties (such as power, speed), positions (extracted from image data) and operation rules (extracted from text data) of the cutter, dryer and flavoring machine are fused together to form a feature vector containing multidimensional information.

[0354] Unit 2: Simulation Model Annotation Unit

[0355] The main task of the model annotation unit is to perform hierarchical annotation of explicit information and implicit information of the simulation model. Specifically, the model annotation unit uses a variety of technical means to achieve comprehensive annotation of the model, including automatic annotation and manual annotation.

[0356] The simulation model annotation unit adopts a hierarchical annotation strategy to annotate the explicit information and implicit information of the model respectively. The explicit information includes the basic information, hierarchical information, and layout image of the simulation model; the basic information includes the model name, model type, model update date, etc.; the hierarchical information includes the object library structure, object attributes, and object affiliation, etc.; the layout image information includes the two-dimensional image of the workshop, the three-dimensional shape cross-sectional view of the stand-alone equipment, etc. The implicit information is extracted from the basic marking content of the model to further describe the features that are not clearly defined in the model itself, such as whether the simulation model is a local simulation model or a describable way of its layout. The simulation model annotation unit allows manual input of text descriptions to further describe the features and functions of the scene in which the simulation model is located, and the description content is split and transformed through LLM, etc.

[0357] The technical means adopted by the simulation model annotation unit are automatic annotation technology and manual annotation technology.

[0358] The automatic annotation technology needs to go through three processes: rule matching and template recognition, image processing and pattern recognition, and natural language processing (NLP) and semantic analysis; the rule matching and template recognition is to automatically identify the explicit information (such as model name, type, update date, etc.) and hierarchical information (such as object library structure, object attributes and object affiliation, etc.) of the simulation model through predefined rule sets and templates. This technology relies on structured data parsing technology and preset rules; the image processing and pattern recognition is to use computer vision technology, such as edge detection, feature extraction algorithms (such as SURF, SIFT), to analyze the layout image in the model, and annotate explicit information such as two-dimensional layout diagrams and three-dimensional shape cross-section diagrams; the natural language processing (NLP) and semantic analysis are for the annotation of implicit information, using NLP technology to analyze text descriptions, and through methods such as named entity recognition (NER) and topic modeling, extract and annotate features that are not clearly defined in the model but mentioned in the text description, such as whether the simulation model is a local simulation model, and the describable way of its layout.

[0359] The manual annotation technology needs to go through two processes: user manual input and editing and large language model (LLM) splitting and conversion; the user manual input and editing is to provide a user interface to allow engineers to manually input text descriptions to supplement or refine scene features and functions not covered in automatic annotation. These manually input contents may include special model uses, specific operating environments and other feature data; the large language model (LLM) splitting and conversion is to use a large language model (LLM), such as GPT or BERT, to split, convert and classify the manually input text descriptions to ensure that they can be recognized and processed by the system, and convert the complex operation scene descriptions of the cigarette silk workshop into structured or semi-structured data for easy storage and query.

[0360] The operation process of the model annotation unit is as follows:

[0361] S1. Initial data collection and preprocessing: Automatically extract metadata from simulation model files to collect basic information and structural information of the model; use image processing technology to analyze the two-dimensional and three-dimensional images of the simulation model to obtain layout and shape information;

[0362] S2. Automatic annotation of explicit information: Using automated rule matching and template recognition technology, the basic information, hierarchical information, and layout image information of the model are annotated from the structured data;

[0363] S3. Implicit information labeling and processing: Use NLP and semantic analysis technology to process unstructured data, extract implicit information and label it; provide a user interface to allow cigarette silk workshops to manually input additional descriptions, and split and transform the descriptions through the LLM model to generate recognizable structured data.

[0364] S4. Storage and update of annotation results: The explicit and implicit information of annotations is stored in the model feature data management unit, and a multi-dimensional index structure is established to support fast retrieval and associated query.

[0365] The extraction method for automatic annotation of explicit information is as follows: the system loads metadata from a simulation model file or database; runs a rule matching algorithm to identify the model's name, type, update time and other fields according to a preset template; and then automatically fills the identified information into the corresponding annotation field; and stores the annotation information for subsequent retrieval and use.

[0366] The extraction method of hierarchical information annotation is as follows: parsing the object library in the simulation model to identify all device nodes and attributes; using a recursive parsing method to traverse each device node to extract device type and attribute information; then automatically establishing subordinate relationships between devices (such as upstream and downstream relationships) and annotating them in the hierarchical structure of the model; saving the annotation results and updating the object library structure to reflect the relationship between devices.

[0367] The extraction method of layout image information annotation is as follows: loading a two-dimensional layout diagram of a workshop and a three-dimensional cross-sectional diagram of equipment; using an edge detection algorithm to identify regional boundaries in the image; using pattern recognition technology and a deep learning model (such as CNN) to identify the type and spatial position of the equipment; automatically annotating the identified layout information in the layout image data of the model; and finally saving the processed image and annotation data.

[0368] The extraction method of the annotation of implicit information is as follows: collecting simulation logs and description texts input by users; preprocessing the texts (such as word segmentation, removal of stop words, stem extraction, etc.); then using named entity recognition (NER) and topic modeling techniques (such as LDA) to extract key information and features; then annotating the extracted information as implicit information, such as adding "local simulation model" or "straight line layout" to the implicit information field of the model; and finally saving the annotated information for subsequent query and optimization.

[0369] The method of manual input and LLM conversion is as follows: the user manually inputs the scene description through the interface; the system calls LLM to perform semantic analysis and understanding on the input natural language text; then LLM splits the input text into multiple meaningful parts and converts them into structured data entries (such as scene name, category, etc.); then the system automatically marks the converted data entry as part of the implicit information field; finally, the marked information is saved and updated for subsequent retrieval and optimization.

[0370] The result of the model annotation unit after running is: a complete annotation data set containing explicit and implicit information is generated; the explicit information includes the name, type, hierarchical structure, layout image, etc. of the model; the implicit information includes the application scenario and description method of the model, etc. The complete annotation data set provides detailed annotation information for the simulation model of the cigarette silk workshop, so that the model can be accurately understood and applied in the subsequent use process; these annotation information can be used for the recommendation, optimization, query and analysis of the simulation model, and improve the intelligent management and use efficiency of the simulation model.

[0371] In the cigarette-making workshop, in the simulation model annotation unit, the explicit and implicit information of the model is fully annotated through the combination of automatic annotation technology and user manual input to ensure the accuracy and completeness of the information; while manual input and LLM conversion technology allow users to flexibly add and update the scene description and functional characteristics of the model to adapt to different application scenarios and demand changes. Automatic annotation technology greatly reduces manual workload and improves annotation efficiency. At the same time, combined with multimodal data processing, it improves the comprehensiveness and accuracy of annotation.

[0372] Example 8 Implementation of the Simulation Model Annotation Unit in the Cigarette Making Workshop Simulation Model

[0373] In a specific embodiment, in a simulation model of a cigarette cutting workshop, the specific technical content and operation steps of the simulation model annotation unit are to comprehensively annotate the explicit information and implicit information of the model through a combination of automation and manual methods.

[0374] First, the automatic annotation is as follows:

[0375] The annotation of the explicit information needs to be done through rule matching and template recognition technology, using rule matching (RuleMatching) and template recognition (Template Recognition) technology to automatically extract the basic information of the simulation model from the model file or database; this basic information is structured and located in the metadata part of the model, such as tags or key-value pairs in XML or JSON file format.

[0376] The labeling process of the explicit information is as follows: the system first defines a set of rule sets and template libraries, including expected patterns of information such as model name format, type format and update date; then, data is loaded from the simulation model file, and fields that conform to the rule set are identified through regular expressions (Regex) or template matching algorithms, such as using regular expressions to match text formats similar to "Model Name: Silk Making Process Simulation Model", and extracting "Silk Making Process Simulation Model" as the model name; finally, the extracted information is automatically filled in, such as "Silk Making Process Simulation Model" as the model name, "Process Simulation" as the model type, and "2024-09-06" as the update time, and filled in the corresponding labeling fields.

[0377] Specifically, the hierarchical information annotation needs to parse the object library structure of the simulation model through structured data parsing technology, automatically identify device objects and their attributes, and the subordinate relationships between devices (such as upstream and downstream relationships).

[0378] The hierarchical information annotation process is as follows: first, the system uses an XML or JSON parser to extract object library structure data from the simulation model file; through a recursive parsing method, each device node in the object library is traversed to identify the device type (such as "shredded wire cutter", "shredded wire drying machine") and its parameters (such as "power" and "speed"); then based on the object tree structure, the subordinate relationship between devices is identified, for example, the upstream device is a "shredded wire cutter" and the downstream device is a "shredded wire drying machine"; finally, this information is annotated into the hierarchical information of the model to form a device object tree to display the logical connection relationship between devices.

[0379] Specifically, the layout image information annotation needs to be performed through image processing technology, using image processing technology (such as edge detection, pattern recognition, etc.) to extract and annotate layout information from the two-dimensional image of the workshop and the three-dimensional cross-sectional view of the equipment.

[0380] The process of labeling the layout image information is as follows: first, the two-dimensional layout diagram of the workshop is processed using an edge detection algorithm (such as Canny edge detection) to identify the edges in the image and determine the regional boundaries; then, a pattern recognition algorithm (such as Hough transform) is used to identify common geometric shapes and graphic patterns in the image, for example, a rectangular area is identified as a "tobacco recipe library" or a "semi-finished tobacco library"; further, the shape, size and position data of the equipment are extracted from the three-dimensional cross-sectional diagram of the equipment, and a deep learning model (such as a convolutional neural network CNN) is used to further classify and label the equipment type (such as "cutting machine", "drying machine") and its spatial position; finally, these labeling information is finally stored in the form of image annotations or vector graphics for subsequent use.

[0381] The annotation of the implicit information requires the use of natural language processing (NLP) technology, which extracts implicit information from simulation logs, descriptive texts input from the cigarette making workshop, and other unstructured data using NLP technology; this information is not clearly defined in the workshop simulation model file, but is very important for the application and understanding of the workshop simulation model.

[0382] The labeling process of the implicit information is as follows: first, the system preprocesses the unstructured text data, such as word segmentation, stop word removal, stem extraction, etc.; then, specific entities (such as "local simulation model", "U-shaped layout") and key phrases are identified through named entity recognition (NER) technology; then, topic modeling (such as LDA model) is used to analyze the subject and semantics of the text, and extract feature information hidden in the text, such as the application scope of the simulation of the cigarette silk workshop, layout description and other feature data; finally, according to the extracted feature information, it is labeled as implicit information, such as if it is extracted that "the model only performs local simulation on the cutting, baking and blending section in the cigarette silk making process", it is labeled as "local simulation model".

[0383] Second, the manual annotation is as follows:

[0384] The manual input and LLM conversion is the manual input and large language model (LLM) conversion of the cigarette silk workshop; the cigarette silk workshop is allowed to manually input specific scene descriptions, and use LLM (such as GPT, BERT, etc.) to split and convert the input text into processable structured data entries.

[0385] The process of manual input and LLM conversion is as follows: the cigarette silk-making workshop inputs a specific scenario description through the system interface, such as "the local simulation model of the cigarette silk-making workshop is used to perform process verification on the cutting, baking and blending section"; then the LLM model is used to receive the input natural language description, and perform syntactic analysis and semantic understanding on it; then, based on the input content, the LLM splits the text into multiple meaningful parts and converts them into structured or semi-structured data entries, such as splitting the title into "local simulation", "cutting, baking and blending section", "process verification", and marking the corresponding categories; finally, the converted structured data is stored in the model database for subsequent query and analysis.

[0386] Unit 3: Simulation model feature data management unit

[0387] The main task of the simulation model feature data management unit is to store simulation model files and feature data to ensure integrity and traceability. The simulation model feature data management unit supports fast retrieval and associated query of model feature data by establishing a multidimensional index structure; and for each model, the feature data management unit will classify the acquired data and perform structured storage, and the optional data formats include JSON, etc.

[0388] The technical means adopted by the simulation model annotation unit are feature data storage and multi-dimensional indexing, high-dimensional feature mapping and dynamic embedding mechanism, dynamic feature mapping tensor adjustment and similarity calculation and clustering algorithm.

[0389] The characteristic data storage is that the simulation model characteristic data management unit uses a structured storage format (such as JSON, XML) to save the simulation model file and its characteristic data. These formats have good data integrity and traceability, and can effectively store model metadata, equipment attributes, process parameters and other information; in the simulation model of the cigarette silk workshop, characteristic data such as equipment name, equipment type, physical parameters (such as power, temperature, pressure, etc.) and process flow data will be stored as JSON objects, each object represents a set of attributes of a model or equipment.

[0390] The multidimensional index structure supports fast retrieval and associated query of the simulation model feature data. The multidimensional index structure can be an inverted index, a multidimensional tree index (such as a KD tree, an R tree) or a hash index in a high-dimensional space; for example, in a cigarette silk workshop simulation model, a KD tree can be used to quickly find the attributes of similar equipment, and an R tree can be used to store the spatial distribution information of equipment. A high-dimensional space hash index can be used to quickly retrieve matching models in the feature space.

[0391] The high-dimensional feature mapping and dynamic embedding mechanism maps the feature data of the simulation model into a high-dimensional topological space and classifies it using the dynamic feature embedding mechanism; then uses nonlinear transformation (such as Tanh function) and weighted coefficient matrix to transform the model feature data into a multi-dimensional feature data vector y, y = [y1, y2, ..., y n ], each y i Represents different model features, multi-dimensional logistics feature mapping function f MLFM As shown below:

[0392] L = f MLFM (y)=[L1(y),L2(y),…,L m (y)]

[0393] Each mapping dimension L i The formula for (y) is as follows:

[0394]

[0395] In the formula, λ ij is the weight coefficient of the i-th dimension to the j-th feature, γ ij is the scaling factor of each dimension of the simulation model, δ ij is the normalization coefficient of the i-th dimension to the j-th feature;

[0396] In the simulation model of the cigarette making workshop, the physical properties and operating parameters (such as temperature, speed, and output rate) of equipment objects (such as cutters, dryers, and flavoring machines) will be mapped into high-dimensional vectors and converted into embedded representations in the high-dimensional topological space through the above mapping formula.

[0397] The dynamic feature map tensor adjustment is performed using the dynamic feature map tensor D(k), and an adjustment factor is introduced to represent the weight change of the feature in different logistics operations; the adjustment factor is calculated using the Sigmoid function:

[0398] The dynamic feature map tensor D(k) is generated by introducing an adjustment factor based on the vector y, which represents the weighted value in the kth logistics operation:

[0399] D(k)=[λ1(k)y1,λ2(k)y2,…,λ n (k)y n ]

[0400] Among them, λ i (k) using a sigmoid function to calculate the adjustment factor for each feature in the kth logistics operation;

[0401] The Sigmoid function is often used to limit the value range of the adjustment factor to (0,1) to avoid the influence of extreme values ​​and ensure that the adjusted feature weights are within a reasonable range.

[0402] In the simulation model of the cigarette making workshop, the operating characteristics of different equipment have different importance in different logistics operations (such as feeding, blending, cutting, etc.); then, by dynamically adjusting the weights of these characteristics, the influence of characteristic data in specific operations can be more accurately reflected. For example, in the drying process, the rotation speed and temperature characteristics of the drying machine may have higher weights, while in the cutting process, the blade speed and feed speed characteristics of the cutting machine have higher weights.

[0403] The similarity calculation is to calculate the similarity score between the simulation models through multiple metrics (such as logistics topology space distance, dynamic feature mapping tensor similarity, cosine similarity); calculate the classification score between the simulation models and classify them according to the score using a clustering algorithm; the classification score S(L) between the i-th and j-th logistics simulation models in the k-th logistics operation i ,L j ,k)The calculation formula is as follows:

[0404] S(L i ,L j ,k)=α·d(L i ,L j )+β·sim(D i (k),D j (k))

[0405] In the formula, α represents the weight coefficient of the logistics topological space distance, β represents the weight coefficient of the dynamic feature map tensor similarity, sim(D i (k),D j (k)) indicates D i (k) and D j (k) cosine similarity between;

[0406] The clustering algorithm is to classify the models using a clustering algorithm (such as K-means or hierarchical clustering) based on the similarity score results obtained by similarity calculation; the K-means algorithm classifies the models into K clusters by minimizing the total square error between the models, while the hierarchical clustering generates a clustering tree by continuously merging or splitting clusters;

[0407] In the simulation model of the cigarette cutting workshop, similarity calculation can be used to compare the similarity of different simulation models under the same process. For example, by comparing the similarity of two cutting process simulation models, it is possible to identify whether the processes have similar operating conditions and performance. Based on the results of these similarity calculations, a clustering algorithm is used to group similar simulation models for subsequent optimization and analysis.

[0408] The operation process of the simulation model feature data management unit is as follows:

[0409] S1. Data collection: Extract all relevant feature data from simulation model files (such as JSON, XML and other formats), including equipment objects (such as shredders, shredders, flavoring machines, etc.), physical properties (such as equipment temperature, power, speed, etc.), and process parameters (such as blending weight, cutting speed, drying time, etc.);

[0410] S2. Data preprocessing: preprocess the collected data and format it into a standard multidimensional vector form y = [y1, y2, ..., y n ], ensuring that all data are of uniform scale to facilitate subsequent feature mapping and analysis; data preprocessing includes data cleaning (removing invalid data), normalization (standardizing data values), and feature engineering (extracting useful features);

[0411] S3. Feature mapping: Use the feature mapping function f MLFM The extracted data is embedded into a high-dimensional space; this process maps the feature vector through nonlinear transformation (such as Tanh function) and weighted coefficient matrix to generate a high-dimensional feature representation L i (y),capturing the complex relationships and interdependencies among features;

[0412] S4. Dynamic embedding: Through the dynamic feature mapping tensor D(k), the feature weights are adjusted according to the needs of different logistics operations. The adjustment factor is calculated using the Sigmoid function to dynamically adjust the weight of each feature to reflect the importance of the feature in different logistics operations.

[0413] S5. Similarity calculation: Calculate the similarity score S(L) between each simulation model i ,L j ,k), use multiple similarity metrics (such as spatial distance, dynamic mapping similarity and cosine similarity) to evaluate the similarity between models;

[0414] S6. Cluster analysis: Based on the similarity score results, clustering algorithms (such as K-means or hierarchical clustering) are used to classify the simulation models. Clustering algorithms classify models with high similarity into the same category, making it easier to identify and optimize similar model groups.

[0415] S7. Data storage: Store the generated feature vectors and classification results in the data management unit, establish an index structure (such as inverted index, multidimensional tree index, etc.) to support fast query and retrieval;

[0416] S8. Data update: Dynamically update feature data and classification information based on new simulation model data or optimization results to ensure the timeliness and accuracy of the data.

[0417] The results generated by the simulation model feature data management unit after the operation process are as follows: first, a multidimensional feature vector set is generated, which accurately represents the various types of equipment, physical properties and process parameters in the simulation model of the cigarette silk workshop. These feature vectors capture the complex relationship and interdependence between the features, and provide data support for subsequent model analysis; then, through the dynamic feature mapping tensor, the weights of the features in different logistics operations are adjusted, and a weight matrix reflecting the changes in the importance of the features is generated to ensure the best performance of the features in specific operations; then the similarity scores between the simulation models are calculated, and the models are divided into different categories according to the scoring results. The clustering results can be used to identify similar model groups, which is convenient for subsequent model optimization and decision-making; finally, through data management and optimization suggestions, a database containing feature vectors and classification results is generated to support fast query and retrieval. Based on classification and similarity analysis, the system can put forward targeted optimization suggestions (such as adjusting equipment parameters, improving process flow, etc.).

[0418] Example 9: Implementation Method of Simulation Model Feature Data Management Unit in Cigarette Making Workshop Simulation Model

[0419] In a specific embodiment, in the simulation model of a cigarette shred workshop, the equipment and process parameters in the simulation model are diverse and complex. In order to perform effective model optimization and decision-making, it is necessary to perform high-dimensional mapping, dynamic feature adjustment, similarity calculation and classification on these data through a feature data management unit. The simulation model feature data management unit involves the extraction, analysis and storage of the physical properties and operating parameters of the equipment to facilitate subsequent rapid query and optimization suggestions.

[0420] The high-dimensional mapping of the characteristic data is represented by a high-dimensional vector, and various characteristic data are extracted from the simulation model of the cigarette shred workshop. These characteristic data include equipment objects (such as shredders, shredders, flavoring machines, etc.), physical properties (such as equipment temperature, power, speed, etc.), and process parameters (such as mixing weight, cutting speed, drying time, etc.); these characteristic data are represented as a multidimensional vector y, y = [y1, y2, ..., y n ], each y i Represents a specific characteristic attribute; then use the multidimensional logistics feature mapping function f MLFM Map these feature vectors into a high-dimensional space to generate multiple mapping dimension values ​​L i (y); This process can capture the complex relationships and interdependencies between features.

[0421] The extraction process of the high-dimensional mapping of the feature data requires feature extraction, high-dimensional mapping and feature analysis; first extract all relevant feature data from the file or database of the simulation model and convert it into a standardized multi-dimensional vector form; then use the feature mapping function f MLFM The extracted feature data is nonlinearly mapped to generate feature representation in high-dimensional space; finally, through high-dimensional mapping, the relationship and importance of various features between different processes and equipment can be analyzed.

[0422] In the cigarette making workshop, this high-dimensional mapping can more accurately analyze the interactions between different equipment and process parameters; for example, the relationship between the temperature of the tobacco drying machine and the speed of the tobacco cutter can be clearly identified, which helps to optimize the process parameters and improve production efficiency.

[0423] (1) Example of data collection and preprocessing

[0424] In the simulation model of the cigarette making workshop, equipment (such as cutters, dryers, flavoring machines, etc.) and process parameters (such as temperature, power, cutting speed, drying time, etc.) are collected by the system in real time and stored in a structured data format (such as JSON).

[0425] The operation mode of the data collection and preprocessing is as follows: first, the system extracts all equipment and process data from the workshop equipment management system (EMS) and the manufacturing execution system (MES); then the equipment data (such as the temperature of the drying machine and the speed of the cutting machine) and the process parameters (such as the mixing weight, cutting speed, and drying time) are uniformly formatted into a standard multi-dimensional vector form; finally, the data is preprocessed, and data cleaning (removing invalid data) and normalization (ensuring that all data are compared on the same scale) are performed, and useful feature data are extracted through feature engineering.

[0426] The data collection and preprocessing generates results in which equipment data and process parameters are successfully extracted and formatted into standardized multi-dimensional vectors, ready for high-dimensional mapping and further analysis.

[0427] (2) Implementation of high-dimensional feature mapping

[0428] In the simulation model of the cigarette making workshop, the system maps these feature data into a high-dimensional space to capture the complex relationships and dependencies between the features. For example, the system analyzes the correlation between the temperature of the drying machine and the speed of the cutting machine through high-dimensional mapping.

[0429] The operation mode of the high-dimensional feature mapping is as follows: first, the attributes and process parameters of each device (such as the temperature and power of the wire drying machine, the rotation speed and blade speed of the wire cutting machine, etc.) are expressed as multi-dimensional vectors; each dimension represents a device characteristic or process parameter. Then, these multi-dimensional vectors are nonlinearly transformed (such as using the Tanh function) using the feature mapping function, and a mapping in a high-dimensional space is generated; the value of each dimension is operated through a weighted coefficient matrix to form a new high-dimensional feature vector, which represents the complex dependencies between different devices and process parameters.

[0430] The result of the high-dimensional feature mapping is that different equipment (shredder, shredder, and flavoring machine) and their process parameters are mapped into a high-dimensional topological space; this high-dimensional space captures the potential relationships between the equipment, such as the synergy between the temperature of the shredder and the rotation speed of the shredder.

[0431] (3) Implementation of Dynamic Feature Mapping and Weight Adjustment

[0432] In the simulation model of the cigarette making workshop, different processes (such as cutting, drying, and flavoring) have different degrees of dependence on equipment operation characteristics. In order to more accurately reflect the importance of these features, the system uses dynamic feature mapping tensors to adjust the weights of the features.

[0433] The operation mode of the dynamic feature mapping and weight adjustment is as follows: first, the system dynamically adjusts the weight of each feature according to different logistics operations (such as feeding, blending, cutting, etc.). For example, in the drying process, the rotation speed and temperature of the drying machine have higher weights; in the cutting process, the blade speed and feed speed of the cutting machine have higher weights. Then, the Sigmoid function is used to calculate the weight adjustment factor of each feature, limiting the range of weight adjustment to (0,1) to ensure the rationality of weight adjustment; the dynamically adjusted feature weight reflects the importance of different equipment parameters in each process step, thereby optimizing the equipment operation performance of each process in the simulation model.

[0434] The results of the dynamic feature mapping and weight adjustment are as follows: in the wire drying process, the weights of the temperature and speed of the wire drying machine are amplified; in the wire cutting process, the weight of the blade speed of the wire cutting machine is increased; the system dynamically adjusts the feature weights of each device based on the characteristics of logistics operations, generates an adjusted weight matrix, and optimizes the performance of the simulation model.

[0435] (4) Implementation Examples of Similarity Calculation and Cluster Analysis

[0436] In the simulation model of the cigarette making workshop, through similarity calculation and clustering algorithm, the system can identify the similarities between different simulation models. For example, the system compares the simulation models of two drying processes to determine whether they have similar operating conditions and performance.

[0437] The operation mode of the similarity calculation and cluster analysis is as follows: first, the system uses multiple metrics such as cosine similarity and logistics topology space distance to calculate the similarity score between each two simulation models;

[0438] Then, through the formula: S = αd topo +βd tensor +λcos(θ)

[0439] Among them, αd topo represents the distance in logistics topological space, d tensor represents the dynamic feature map tensor similarity, cos(θ) represents the cosine similarity;

[0440] Finally, based on the similarity score, the models are classified using K-means or hierarchical clustering algorithms to classify similar simulation models into the same group.

[0441] The results generated by the similarity calculation and cluster analysis are as follows: the system identifies and classifies similar equipment and process conditions in different simulation models. For example, the system finds that the temperature and speed of the two tofu drying machines in the tofu drying process are similar, so the two tofu drying machines are classified into the same model; the clustering results can be used for subsequent optimization and model improvement.

[0442] (5) Implementation of Data Storage and Dynamic Update

[0443] In the simulation model of the cigarette making workshop, the system stores the calculation results and classification information, and provides support for future data query and retrieval; when new simulation models or equipment data are introduced, the system dynamically updates these feature data.

[0444] The operation mode of the data storage and dynamic update is as follows: the system stores high-dimensional feature vectors, similarity scores and clustering results in a feature data management unit, and establishes multi-dimensional indexes (such as KD trees, R trees, etc.) to support fast queries of equipment and process parameters; then, when new simulation models or equipment data are added, the system performs incremental updates to ensure that the latest data and classification results are always accurate.

[0445] The results of the data storage and dynamic update generation are: the system generates a database containing high-dimensional feature vectors and classification results, and supports fast retrieval and dynamic update; when a new simulation model is introduced, the system can automatically integrate the new data into the existing data.

[0446] Module 2: Cigarette Making Workshop Simulation Basic Knowledge Extraction Module

[0447] The cigarette silk-making workshop simulation basic knowledge extraction module comprises a cigarette silk-making workshop data extraction unit and a cigarette silk-making workshop data processing unit.

[0448] Unit 4: Cigarette Making Workshop Data Extraction Unit

[0449] The main task of the data extraction unit of the cigarette silk making workshop is to obtain relevant data from different information systems (such as manufacturing execution system MES, enterprise resource planning system ERP, product lifecycle management system PLM) and process documents to support the construction of the workshop process simulation model; the data extraction unit is responsible for collecting, organizing and transmitting these data to ensure that the data required for the simulation model is accurate and comprehensive.

[0450] The technical means adopted by the data extraction unit of the cigarette thread workshop are the workshop's manufacturing execution system, enterprise resource planning system, product life cycle management system and related process documents to obtain relevant data on workshop equipment and processes;

[0451] The data obtained in the data extraction unit of the cigarette silk-making workshop include: 1) the basic material data of the workshop, taking the cigarette silk-making workshop as an example, including material name, material type, material code, material grade, etc.; 2) the material processing rules and related process standards of the workshop, taking the cigarette silk-making workshop as an example, including whether it belongs to group processing, blending weight, blending order, upper and lower limits of storage time, material conversion rate in the processing stage, etc.; 3) the production scheduling data of the workshop, taking the cigarette silk-making workshop as an example, including workshop shifts, workshop teams, factory calendar and other related data; 4) the equipment modeling data of the workshop, taking the cigarette silk-making workshop as an example, including basic equipment parameters, control rules, status and other data of storage equipment, processing equipment and auxiliary equipment; 5) the process section and process data of the workshop, taking the cigarette silk-making workshop as an example, including design flow, brand change time, process length, main equipment processing time and other data; 6) the ownership relationship of the process section, process and equipment of the workshop, taking the cigarette silk-making workshop as an example, including the process section included, the number of equipment included in the process and other data.

[0452] The technical means adopted by the data extraction unit of the cigarette silk workshop are data integration tools (ETL), API calls, database queries, document parsing tools and data cleaning and conversion technologies;

[0453] The data integration tool (ETL) is used to extract data from multiple systems, convert and clean it, and load the cleaned data into the simulation platform; the data integration tool (ETL) implementation method requires extraction, conversion and loading; firstly, extract data from multiple data sources (MES, ERP, PLM and other systems) to obtain the original production data, material information, equipment status, etc. of cigarette silk making; then standardize and clean the original data, handle missing values ​​and duplicate values, and ensure that the format and unit of the data are consistent, such as converting the time units and material codes in different systems in a unified manner; finally, load the converted data into the database of the simulation platform for use by the cigarette silk making workshop simulation model;

[0454] The API call is to directly connect with the production management system (such as MES, ERP, PLM) by calling the API (application programming interface) to obtain the latest data in real time; the operation mode of the API call is that the simulation platform directly extracts real-time data such as production schedule, materials, equipment parameters, process standards, etc. from the system by calling the API interface provided by these systems, and then the API call can be executed regularly or in real time to ensure that the data in the simulation platform is the latest, such as obtaining the status data of the workshop equipment through the API every 30 minutes;

[0455] The database query is to use SQL (Structured Query Language) or NoSQL technology to directly query the underlying database to obtain the structured data required for simulation; the operation mode of the database query is to first directly query the underlying databases such as MES, ERP, equipment management system, etc., use SQL statements to extract specific equipment, process, and material information from the relational database, and then for non-relational databases (such as NoSQL databases), obtain semi-structured data or distributed storage data through corresponding query mechanisms;

[0456] The document parsing tool uses text processing and parsing tools to extract key information from unstructured process documents, instructions and other documents and convert them into structured data; the operation mode of the document parsing tool is to use regular expressions or natural language processing (NLP) tools to parse texts of process documents, standard operating procedure documents, etc., extract information such as process parameters, material rules, equipment operating standards, etc., and then structure the extracted information, such as converting it into a table or database format, for use in the cigarette silk workshop simulation model;

[0457] The data cleaning and conversion is to standardize the data extracted from multiple sources to ensure that the data is complete, consistent, and meets the input format requirements of the simulation platform; the operation mode of the data cleaning and conversion requires data cleaning, data conversion and data standardization, specifically first processing missing data, outliers, duplicate data, deleting or correcting erroneous data, such as removing duplicate production plan entries, correcting erroneous equipment status, and then standardizing the fields and formats of different systems, such as unifying the tobacco leaf or tobacco material coding in different systems, ensuring the consistency of time format and unit, and finally converting the heterogeneous data from various sources into a unified format so that the cigarette silk workshop simulation platform can correctly read and apply these data.

[0458] The results of the operation of the data extraction unit of the cigarette silk workshop are: forming a unified material and equipment data model, establishing workshop simulation model rules and generating simulation operation parameters. All extracted data will be standardized and converted into material and equipment objects in the simulation model; according to the extracted process rules, the simulation model can automatically define the material processing steps, process flow and equipment usage, including production scheduling, material conversion rate, equipment process parameters, etc., to ensure that the simulation can be connected with the actual production process.

[0459] Example 10 Implementation of the Cigarette Making Workshop Data Extraction Unit in the Cigarette Making Workshop Simulation Model

[0460] In a specific embodiment, in a simulation model of a cigarette making workshop, the operation of a data extraction unit involves multiple steps and technical means to acquire, clean and convert data from different systems through an automated or semi-automated process so that it can be used in the simulation model.

[0461] In the simulation model of the cigarette making workshop, the operation process of the data extraction unit is as follows:

[0462] S1. Extraction of tobacco material data: Tobacco material data is stored in the company's ERP or MES system, which records all raw material information used in the cigarette making process; extract basic material information from ERP or MES through API interface or database query, such as tobacco material name, type, code, and grade; obtain the processing rules of tobacco materials, including tobacco blending weight, processing sequence, storage requirements, etc., to ensure that tobacco materials are processed according to the actual process in the simulation model; and then the extracted data will be cleaned and converted to ensure that it conforms to the input format of the cigarette making simulation model, such as unified coding and removal of redundant data, to ensure that the material data in the simulation model is completely consistent with the materials used in the actual workshop, and the processing rules can accurately reflect the actual operation process.

[0463] S2. Production scheduling data extraction: The production scheduling data of the cigarette silk workshop mainly comes from the MES system, which records the workshop shifts, work groups, factory calendar and other information; through API calls or database queries, the production scheduling data is regularly extracted from the MES, including the workshop shift arrangements, the working hours of each work group, and the overall calendar arrangement of the factory (such as rest days, maintenance time, etc.); the scheduling data is formatted and converted into a format that can be recognized by the simulation platform to ensure that the production schedule in the simulation model is synchronized with the actual workshop; at the same time, the system will compare the extracted data with the historical data to ensure the accuracy of the data, so that the working hours and rhythm of each shift and work group can be reflected in the simulation process.

[0464] S3. Equipment modeling data extraction: The parameters, status and control rules of the equipment in the cigarette making workshop are stored in the equipment management system in the form of a database or system interface. By calling the API interface of the equipment management system or directly querying the equipment database, the basic parameters of various types of equipment (such as equipment model, power, production capacity, etc.), current status (such as idle, working, fault) and control rules (such as start and stop conditions, operation mode) are obtained; at the same time, the relevant data of storage equipment (such as storage cabinets) and processing equipment (such as shredders and shredders) are extracted; the equipment data will be converted into equipment objects in the simulation model, which can correctly reflect the performance and status of the equipment in the simulation model, ensure that the equipment operation in the simulation process matches the reality, and can simulate the working status and fault conditions of the equipment;

[0465] S4. Extraction of process segment and procedure data: The relevant data of the process segments and procedures in the cigarette silk-making workshop are usually stored in the PLM system and also included in the process documents. The design parameters of the process segments are extracted from the PLM system through the API interface or database query, such as the design flow rate, process length, main equipment processing time, process switching time and other information. The specific content of the process segments and procedures (such as processing steps, material flow sequence, etc.) is recorded in the process documents, which can be parsed by document parsing tools to convert unstructured process information into structured data. The data of these process segments and procedures will be loaded into the simulation model to ensure that the simulation process follows the actual process flow of the workshop and accurately simulates the production process of each process segment and procedure, including material flow, processing sequence, conversion rate, etc.

[0466] S5. Extraction of attribution relationships: The hierarchical relationship between the process segments, procedures and equipment in the cigarette silk workshop is maintained by the company's production management system (such as PLM or MES), or can be parsed from the process documents; call the PLM or MES system, obtain the process information contained in the process segment through API or database query, clarify the procedures involved in each process segment, and the number and type of equipment used in each process; if the data comes from the document, extract these hierarchical structures through the document parsing tool and convert them into structured data that the system can process; finally, the simulation model will establish a logical hierarchy between the process segments, procedures and equipment based on these attribution relationships, so that the model structure matches the actual workshop production structure, and then the simulation model can reflect the hierarchical relationship between the process segments, procedures and equipment, ensure that the operating logic of each part is consistent with the actual production process, and facilitate simulation operation and analysis.

[0467] The operation process of the data extraction unit in the simulation model of the cigarette making workshop is automated, and it can regularly update the data in the simulation model to ensure that the model is synchronized with the actual operation of the workshop, ultimately providing high-quality and accurate data support for the simulation.

[0468] Unit 5: Cigarette Making Workshop Data Processing Unit

[0469] The main task of the data processing unit of the cigarette silk workshop is to process the acquired data, including data splitting, noise reduction, structuring and knowledge extraction, to ensure that the data can be effectively used for the construction and operation of the simulation model, while improving the quality and structuring of the data, and providing high-quality input for subsequent analysis and simulation.

[0470] The data processing unit of the cigarette silk workshop performs data splitting and noise reduction after successfully acquiring relevant data; for structured data, concepts and relationships are automatically generated through predefined templates, and knowledge is extracted in combination with ETL tools; for unstructured data, part of the data is annotated through concepts and relationships, a domain knowledge training library is constructed, and relevant models are used for knowledge extraction; and a variety of training algorithms are used in the extraction process to improve the recognition and extraction capabilities of data types. Sequence annotation models such as BiLSTM are used for text data, and feature extraction is completed for image data through the SURF feature extraction algorithm.

[0471] The technical means adopted by the data processing unit of the cigarette silk workshop are ETL tools, natural language processing (NLP) technology, image processing algorithms, noise reduction algorithms and multiple machine learning training algorithms.

[0472] The ETL tool is used to process structured data, such as extracting production data from MES, ERP and other systems, and converting it into a format that can be used by the simulation model; the operation mode of the ETL tool needs to extract data, convert data and value data. First, the basic information of materials (such as tobacco material name, code, grade, etc.) is extracted from the ERP system, and the production scheduling data, equipment status data and production plan data are extracted from the MES system. The extracted data is then converted into a format, such as standardizing the tobacco material codes in different systems to ensure that the material properties in the simulation model are consistent with the actual ones. The processed data is finally loaded into the simulation model to form the basis for key data such as tobacco materials, equipment, and processes; structured data is extracted and processed by the ETL tool to ensure that tobacco materials, equipment and production scheduling data can be correctly mapped to the simulation model, providing basic data support for simulation operation.

[0473] The natural language processing technology is used to process unstructured text data, and a sequence annotation model (such as BiLSTM) is used to extract key entities and relationships in the text; the operation mode of the natural language processing technology requires process document processing, entity and relationship extraction, and knowledge base construction. First, the tobacco silk making process is often recorded in documents, such as the processing standards and process rules of tobacco materials. Natural language processing technology extracts key information such as processing steps, material blending ratios, storage time, etc. by processing these documents. Then, through the BiLSTM model, entities (such as equipment, process section names, material types, etc.) and the relationships between them (such as the process steps executed by a certain equipment, the processing methods of materials in different processes, etc.) are extracted from the process documents. Then, the extracted information is used to construct a domain knowledge base to enhance the understanding and application of the simulation model to the process flow; natural language processing technology extracts unstructured data from the process documents of the cigarette silk making workshop, generates the process parameters required by the simulation model, and ensures the accuracy of the process flow.

[0474] The image processing algorithm is used to process workshop image data and extract key features therein, such as using the SURF (Speeded-Up Robust Features) algorithm to extract important information from equipment layout diagrams or workshop monitoring images; the specific operation mode of the SURF algorithm is to process the equipment layout image of the workshop, and the SURF algorithm extracts the feature points of the equipment and determines their relative positions, so as to help the simulation model accurately reflect the spatial distribution and operation process of the equipment, and then extract the equipment status information from the real-time monitoring image of the workshop, and reflect the current operation status or fault conditions of these devices in the simulation model; by performing image processing with the SURF algorithm, the simulation model can better understand the equipment layout and operation conditions in the cigarette silk workshop, and ensure that the spatial configuration matches the actual production environment.

[0475] The noise reduction algorithm ensures the quality of data input and removes noise and abnormal data by cleaning and filtering the data; the noise reduction algorithm needs to clean the equipment data of the cigarette silk workshop and verify the production plan data. For the equipment status data extracted from the MES or equipment management system, the noise reduction algorithm is used to clean the abnormal equipment operation data, eliminate sensor errors or unreasonable data points, and ensure that the simulation model obtains accurate equipment status, and then cleans the production scheduling data, eliminates redundant data, and corrects abnormal scheduling arrangements to ensure that the scheduling data used in the simulation process is consistent with the actual situation; the acquired data is cleaned by the noise reduction algorithm to ensure the accuracy of the simulation model input data and improve the reliability of the simulation results.

[0476] The multiple machine learning training algorithms are used to improve the ability to identify data types and extract knowledge, especially to process complex data types, such as time series data and process parameter optimization; the multiple machine learning training algorithms require data classification and identification, optimization and prediction, and deep learning model processing of image data. First, by training classifiers, different types of process data are identified, such as equipment operation mode, conversion rate in material processing and other data, and then based on historical data training models, production process parameters are optimized (such as production rhythm, material usage) and possible situations in future production are predicted (such as equipment failure, production bottlenecks). Complex image data is then processed through a convolutional neural network (CNN), and time series data of equipment operation is processed through an LSTM (long short-term memory network) to identify equipment failure modes or optimize process parameters. Through a variety of machine learning training algorithms, the optimization and prediction capabilities of the cigarette silk making simulation model in the production process are improved, making the simulation model more intelligent.

[0477] The operation steps of the data processing unit are as follows:

[0478] S1. Data splitting and noise reduction: The system first splits the data extracted from different sources and classifies them by data type (such as materials, processes, equipment, etc.); then uses noise reduction algorithms to remove invalid or inaccurate data;

[0479] S2. Concept and relationship generation: For structured data, ETL tools automatically generate data concepts and relationships based on predefined templates, such as defining the transformation rules for a certain type of material in a specific process segment and generating a mapping relationship between the process flow and the equipment;

[0480] S3. Processing of unstructured data: For text data, the key entities and relationships in the process are extracted through sequence annotation models (such as BiLSTM). For example, process standards such as "blending weight" and "storage time" are extracted from process documents; for image data, the features of key equipment are extracted through the SURF algorithm to generate image features of the equipment in the simulation model;

[0481] S4. Knowledge extraction and training: Use machine learning algorithms to train unstructured data to improve the accuracy of data type recognition and knowledge extraction. The generated knowledge base will continuously optimize the process rules and equipment operation logic of the simulation model;

[0482] S5. Data loading: The processed data will be loaded into the simulation platform for the construction and operation of the simulation model. The processed data will cover various aspects such as materials, processes, equipment, production scheduling, etc. to ensure the accuracy and reliability of the simulation model operation.

[0483] The results generated by the data processing unit of the cigarette silk workshop after the operation process are: structured simulation data is generated, concepts and relationships are automatically generated, a domain knowledge base is constructed and can be converted into feature extraction values. The processed data is converted into structured data, including information such as materials, equipment, process flow, production scheduling, etc. These data can be directly used as input for the simulation model; the concepts and relationships generated in the structured data include the matching of materials and process sections, the use of equipment in different processes, the association between production scheduling and equipment running time, etc.; based on the results of processing and extraction, a domain knowledge base is constructed, covering the rules and standards of the cigarette silk process, to provide support for subsequent simulation analysis; the equipment layout and key process information in the image data are converted into numerical features, and used to optimize the equipment configuration and workshop layout of the simulation model.

[0484] Example 11 Implementation of the data processing unit of the cigarette making workshop in the cigarette making workshop simulation model

[0485] In a specific embodiment, in a simulation model of a cigarette thread-making workshop, a data processing unit processes structured and unstructured data through a variety of technical means to ensure that the simulation model accurately reflects the actual production situation.

[0486] In the simulation model of the cigarette silk-making workshop, the operation process of the data processing unit of the cigarette silk-making workshop is as follows:

[0487] S1. Processing of structured data: Use ETL tools to process structured data and automatically generate concepts and relationships such as materials, equipment, and process flows based on predefined templates.

[0488] S1.1 Material data processing: Extract tobacco material data (such as tobacco material name, code, grade, etc.) from MES and ERP systems, and generate relevant process and material relationships based on material usage rules.

[0489] S1.2 Production scheduling data processing: Obtain production scheduling data from the MES system, including workshop shifts, factory calendars, and production plans, to ensure that the simulation model is consistent with the actual production schedule.

[0490] S1.3 Equipment parameter processing: Extract equipment parameters (such as equipment name, status, power, etc.) from the equipment management system, and automatically generate the role of the equipment in different processes and its relationship with materials in the process flow.

[0491] After S1. structured data processing, a structured data model containing elements such as materials, equipment, and process flow is generated to ensure that the simulation model reflects the actual operation of the workshop.

[0492] S2. Processing of unstructured data: Use natural language processing (NLP) technology and image processing algorithms to process unstructured data and extract useful process information; unstructured data includes text processing and image processing;

[0493] S2.1 Text processing: Use the BiLSTM sequence labeling model to process the text data in the process documents, identify entity information such as process steps, material conversion rates, processing rules, and generate a structured knowledge base; extract entity relationships in the process documents, such as the requirements for material processing in a certain process, blending weight, upper and lower limits of storage time, and other information.

[0494] S2.2 Image processing: Use the SURF feature extraction algorithm to process the image data of the workshop layout or equipment, and extract key image features, such as the location of the equipment, and the correspondence between the equipment and the process section, to ensure that the workshop layout in the simulation model is consistent with the actual production environment.

[0495] After S2. unstructured data processing, structured process data extracted from text and equipment layout data extracted from images are formed, ensuring that the simulation model completely contains process and workshop layout information.

[0496] S3. Data denoising and data cleaning: Ensure data accuracy and consistency through data cleaning and denoising algorithms;

[0497] The data denoising is to use a denoising algorithm to filter abnormal data, such as the error status of production equipment or unreasonable process time arrangement, to ensure the accuracy and reliability of simulation input data;

[0498] The data cleaning is to clean the extracted data such as materials, production plans, equipment status, etc., to eliminate duplicate, redundant or erroneous data, such as removing duplicate equipment status records and correcting unreasonable production scheduling time.

[0499] After S3. data denoising and data cleaning, the data quality is improved, ensuring that the simulation model can be simulated and analyzed based on accurate and clean data.

[0500] S4. Knowledge extraction and domain training: construct a domain knowledge base through knowledge extraction and machine learning algorithms, and perform model training and optimization; the knowledge extraction and domain training require the construction of a domain knowledge training base and the training and optimization;

[0501] The domain knowledge training library constructed uses annotation tools to annotate the content extracted from the unstructured data to generate a domain-specific knowledge base for the cigarette shred process, which contains information such as process steps, material processing standards, and equipment operation procedures, to help the simulation model better understand the process flow;

[0502] The training and optimization is to use a variety of machine learning algorithms (such as random forests, support vector machines, etc.) to train simulation data, improve the ability to recognize and process data types, and optimize the parameters of the model so that the simulation model can be optimized and adjusted according to new production data, thereby improving the prediction ability and accuracy of the simulation model.

[0503] After S4. knowledge extraction and domain training, a knowledge base is generated, and through machine learning model training and optimization, the simulation model can simulate and optimize the process flow more intelligently and has adaptive learning capabilities.

[0504] In the simulation model of the cigarette making workshop, the technical means of the data processing unit include structured data processing (ETL tools), unstructured data processing (NLP and image processing), data cleaning and noise reduction, as well as knowledge extraction and machine learning training. Through these technical means, the simulation model can simulate based on accurate and clear data, improving the simulation accuracy and optimization capabilities of the production process.

[0505] Module 3: Cigarette Making Workshop Simulation Multimodal Knowledge Graph Construction Module

[0506] The cigarette silk workshop simulation multimodal knowledge graph construction module includes a basic graph construction unit and a feature fusion unit.

[0507] Unit 6: Basic graph construction unit

[0508] The main task of the basic graph construction unit is to construct multiple basic knowledge graphs through preliminary structured processing and data integration to realize the conceptual links between different systems such as workshop equipment, process flow, simulation platform, etc.; these knowledge graphs reflect the equipment, process flow and their relationships in the workshop through systematic node and edge descriptions, and ultimately support a more accurate and intelligent simulation process.

[0509] The basic graph construction unit needs to construct multiple basic knowledge graphs; these graphs are generated through preliminary structured processing and integration of workshops, logistics simulation platforms and related process data; each basic knowledge graph uses the workshop equipment, process flow, workstations in the simulation platform, etc. in their respective systems as node concepts, and the association between equipment and processes, the association between workstations and processes, etc. as edges.

[0510] The technical means adopted by the basic graph construction unit are knowledge graph construction tools, data integration and ETL processing, relationship extraction and graph generation and graph optimization.

[0511] The knowledge graph construction tool uses professional knowledge graph construction tools (such as Neo4j, GraphDB) to create and manage knowledge graphs. The operation mode of the knowledge graph construction tool is to use Neo4j / GraphDB, which are graph databases specifically used to store and query knowledge graphs, and efficiently process the relationship between nodes (such as equipment, processes, and workstations) and edges (such as the association between equipment and process segments, and the association between processes and materials); when constructing a knowledge graph, equipment, processes, workstations, etc. are all defined as "nodes" in the graph. The association between different nodes (for example, a certain equipment is used in a specific process segment, and the association between a certain workstation and a material flow path) is defined as an "edge"; through the query language of the graph database (such as Neo4j's Cypher), complex graph structures can be easily managed and queried, supporting real-time node relationship query and analysis, and improving the information retrieval capability in the simulation model. The knowledge graph construction tool is used to construct a knowledge graph of workshop equipment, process flow, material flow and production operations, providing efficient data query and relationship retrieval functions for the simulation model.

[0512] The data integration and ETL processing is to extract structured and unstructured data from different systems (such as MES, ERP, PLM, etc.) through ETL tools, and clean and standardize them to ensure the consistency of data formats. The operation mode of the data integration and ETL processing is to extract production equipment and status data from the MES system, extract materials and process flow data from the ERP system, and obtain production process documents from the PLM system. These data sources are integrated and extracted through the ETL (Extract, Transform, Load) tool; during the integration process, the data format is unified according to the differences in data formats of different systems, such as unified time format, material coding, equipment ID, etc., to ensure that the data can be consistently associated in the knowledge graph; data with redundant or missing values ​​is processed, such as missing equipment parameters can be supplemented by historical data or default values, and repeated material information is removed. The data integration and ETL processing is used to extract and transform data from different systems, and standardize these data to ensure that the data can be seamlessly connected and used in the knowledge graph.

[0513] The relationship extraction and graph generation uses the relationship extraction algorithm and natural language processing (NLP) technology to extract the relationship between nodes from unstructured data (such as process documents, equipment configuration tables), and generate the associated edges in the graph. The operation mode of the relationship extraction and graph generation is to use the relationship extraction algorithm to analyze the relationship between equipment and process flow, process section and process, such as automatically identifying the operation performed by a certain equipment in a specific process section and the flow rules of materials in different process stages by analyzing the document; and then extract the process information and equipment configuration from the process document by NLP technology. Using BiLSTM or similar sequence labeling models, the relationship between "equipment-process-(tobacco) material" can be extracted from the process description, and these relationships can be input into the knowledge graph; finally, the extracted equipment, process section, material and other information are used as "nodes", and "edges" are generated through the extracted relationships, thereby forming a complete graph structure of equipment, process, and material flow. The relationship extraction and graph generation are used to convert the equipment, process, material and other information in the unstructured data into nodes and edges in the knowledge graph, and establish a complete association relationship between equipment and process flow.

[0514] The graph optimization is to optimize the knowledge graph through automated tools and manual verification to ensure that the definitions of all nodes and edges are accurate, and to optimize the structure in the graph to make it easy to query and run efficiently.

[0515] The operation mode of the graph optimization is to first manually check the logic and boundary conditions between key nodes for complex production process flows or multi-node relationships to ensure that there are no unreasonable process links in the simulation model; then optimize the connection mode of the edges in the knowledge graph through algorithms to reduce unnecessary node redundancy and improve the efficiency of graph query, such as merging repeated equipment nodes, simplifying multiple repeated connections in the same process segment, and optimizing the hierarchical structure of complex processes. The graph optimization is to ensure that the logical relationship of the graph is correct and optimize the query performance of the graph after the knowledge graph is constructed, so as to improve the overall operation efficiency and data consistency of the simulation model.

[0516] The basic graph construction unit operates as follows:

[0517] S1. Data extraction and cleaning: Extract raw data from MES, ERP and other systems, clean and convert these data to ensure that their formats are unified; construct data from workshop equipment, process flow, material flow and other systems into a queryable knowledge graph to facilitate data management and rapid retrieval of simulation models; extract, clean and standardize data from multiple systems through ETL tools to ensure that structured and unstructured data can be consistently expressed in the graph;

[0518] S2. Node and relationship creation: Based on the extracted data, define the nodes (equipment, process section, workstation, etc.) in the knowledge graph, and generate the associations between equipment and process, workstation and process, etc. through relationship extraction technology; extract key nodes and relationships from unstructured data such as process documents and equipment tables through relationship extraction algorithms and NLP technology, and generate associations in the graph;

[0519] S3. Graph construction and optimization: Use graph construction tools to generate complete knowledge graphs from nodes and relationships, optimize graph structures through algorithms and manual verification to ensure their correctness and completeness; optimize node and edge structures in graphs through automated and manual verification to ensure accuracy of graph expression and query efficiency, and to ensure efficient operation of simulation models;

[0520] S4. Graph maintenance and update: The basic graph can be regularly updated and maintained based on new data and simulation results to ensure that it always reflects the latest workshop production status; the basic graph construction unit can construct a complete knowledge graph of the cigarette silk workshop, so that the simulation model can accurately map the production process, equipment relationships and process flow in the workshop, and support more intelligent and efficient simulation analysis and optimization.

[0521] The result of the basic graph construction unit after the operation process is: the knowledge graph of equipment and process flow is generated, and the association relationship graph is constructed to unify the concepts and support intelligence. The basic graph contains detailed information of all equipment, process flow and workstations in the workshop, which can accurately reflect the production process in the actual workshop; the association relationship between equipment, process, materials and workstations is constructed through the basic graph, which provides a complete association network for the simulation model to ensure the logical correctness of equipment, process flow and material flow during simulation; through the knowledge graph, the simulation model can quickly query the equipment status, process flow and material flow relationship. The construction of the knowledge graph enables the simulation model to have intelligent processing and optimization capabilities, supporting complex production analysis and optimization.

[0522] Example 12: Implementation of the basic graph construction unit in the cigarette thread workshop simulation model

[0523] In a specific embodiment, in the simulation model of a cigarette silk-making workshop, the technical means of the basic graph construction unit covers the steps of data extraction, node definition, relationship extraction, knowledge graph construction, and verification optimization.

[0524] In the simulation model of the cigarette making workshop, the operation steps of the basic map construction unit are as follows:

[0525] S1. Data extraction and integration: Use ETL tools to extract structured and unstructured data from MES, ERP, PLM and other systems, and clean and format them to ensure data consistency and standardization. Ensure that data extracted from different sources can be integrated and standardized so that the graph can correctly represent the actual situation of workshop production;

[0526] S1.1 Data extraction: extract the real-time status of workshop equipment and production scheduling information from the MES system; obtain basic material data from the ERP system; extract process design data from the PLM system;

[0527] S1.2 Data cleaning and formatting: Clean the extracted data, remove redundant information, correct erroneous values, process missing data, and unify the data format, such as standardizing time and equipment parameters, so that it can be used by subsequent map construction tools;

[0528] S1.3 Data integration: Integrate materials, equipment, process flow, and scheduling information from multiple systems to form a unified basic data set to provide reliable input for the construction of the knowledge graph.

[0529] S2. Node definition: Based on the results of data extraction, the concepts of equipment, process sections, materials, etc. in the workshop are defined as nodes in the knowledge graph, providing a basic node structure for the knowledge graph to ensure that core elements such as equipment, process sections, materials, etc. can be clearly expressed in the graph;

[0530] The device node definition is that devices such as shredders and shredders are defined as nodes of the graph, and their device parameters (such as power, processing capacity, status, etc.) are recorded;

[0531] The process section node definition is that each step in the process section (such as silk making, drying, and silk cutting) is also defined as a node, and these nodes can be associated with equipment and materials;

[0532] The material node definition is to define different types of materials (such as tobacco leaves, flavors) as nodes, and record the properties and processing requirements of the materials (such as grade, processing sequence, conversion rate, etc.);

[0533] The workstation node definition is to use the workstations in the simulation platform as nodes to represent their geographical locations and functions in the actual production process.

[0534] S3. Relationship extraction: Using relationship extraction algorithms and natural language processing (NLP) technology, the associations between nodes are extracted from unstructured data (such as process documents, production reports, etc.) to capture and express the interactions and physical connections between equipment, process sections, and workstations, and to add associations to the nodes in the knowledge graph;

[0535] The extraction of relations from the process documents is to use NLP technology to identify the process flow description in the document and extract the logical relations between equipment, materials and process sections, such as the flow sequence of tobacco materials from one process to another, the working association between equipment, etc.

[0536] The relationship between the equipment and the process is extracted by a relationship extraction algorithm, such as the relationship between the shredder and the shredder process, the relationship between the shredder and the shredder process, etc.

[0537] The association between the workstation and the process section is to extract the position and function of the workstation in the process flow from the production report, and to establish the corresponding relationship between the workstation and the tobacco material and the process section.

[0538] S4. Knowledge graph construction: Use professional knowledge graph construction tools (such as Neo4j, GraphDB) to integrate nodes and relationships into the knowledge graph, so that the equipment, process flow and material flow in the graph can be linked to each other, which is used to build a visual and structured knowledge graph to support data interconnection and dynamic query in the simulation model; the knowledge graph construction requires node integration, edge creation and graph display, which is used to build a visual and structured knowledge graph to support data interconnection and dynamic query in the simulation model;

[0539] The node integration is to import the defined nodes such as equipment, materials, process sections, etc. into the knowledge graph construction tool. Through the integration of nodes, the knowledge graph can represent the physical structure and production factors of the workshop;

[0540] The edge creation utilizes the relationship information extracted in the previous step to generate associated edges between nodes, such as the correspondence between equipment and process sections, and the association between workstations and material flow paths, to ensure that the graph can fully express the production process;

[0541] The graph display is to visualize different nodes and associations through a graph construction tool, and to intuitively display information such as equipment layout, material flow, and process flow in the workshop.

[0542] S5. Graph verification and optimization: Check the accuracy of nodes and edges in the knowledge graph through manual verification and automatic optimization algorithms to ensure that the logical relationship of the data is clear and reasonable, and optimize the graph structure to improve its calculation efficiency, ensure that the constructed knowledge graph can accurately reflect the actual situation of the workshop, and ensure that the simulation model can run efficiently;

[0543] The manual verification is to manually check the key nodes and edges in the graph to ensure the correctness of the equipment, process and tobacco material flow path, such as confirming whether the role of a certain equipment in a specific process section is correct and whether the order of tobacco materials in the process is reasonable;

[0544] The automated optimization uses an automated optimization algorithm to analyze the density of nodes and edges in the graph, optimize the structure in the graph, avoid node duplication or relationship redundancy, and thereby reduce unnecessary complexity through optimization, thereby improving the query and computation efficiency of the graph.

[0545] The graph update is to regularly update the knowledge graph based on the latest data and production changes to ensure that it always reflects the latest workshop status.

[0546] In the simulation model of the tobacco cutting workshop, the basic graph construction unit constructs a highly structured knowledge graph through technical means such as data extraction, node definition, relationship extraction, graph construction and verification optimization. The basic graph construction unit organically links nodes such as equipment, process, tobacco materials, and workstations to ensure that the simulation model can accurately simulate and analyze based on complete and standardized data.

[0547] Unit 7: Feature Fusion Unit

[0548] The main task of the feature fusion unit is to generate and disambiguate candidate concepts by integrating the features of multimodal data (such as text, images, etc.), and match and link them with the concepts in the basic knowledge graph; the feature fusion unit combines multimodal information from different data sources to eliminate ambiguity and ensure that the generated candidate concepts can be accurately linked to the corresponding concept nodes in the knowledge graph.

[0549] The technical means adopted by the feature fusion unit are weighted multimodal feature fusion model, attention mechanism disambiguation and matching and linking of candidate concepts;

[0550] Formula 1 is used to generate candidate concepts by weighted fusion of data features from different modalities. The generation of candidate concepts is based on multimodal feature fusion, and the weight W m The relative importance of different modalities is determined, so that the system can make reasonable weighted decisions based on the quality or importance of the data source. Each modality has different weights and feature extraction functions. For example, text data is processed using BiLSTM, and image data is processed using the SURF feature extraction algorithm. The features of multiple modalities are weighted and fused to generate candidate concepts.

[0551] Formula 2 is used to introduce the attention mechanism to perform weighted processing on candidate concepts, and to eliminate ambiguity by maximizing the similarity between related concepts. The disambiguation score of a candidate concept is calculated by weighted similarity of neighbor concepts, and this process uses attention weights to determine the importance of neighbor concepts for disambiguation;

[0552] Formula 3 is used to calculate the attention weight using the LeakyReLU activation function to map the features of the candidate concept and the neighbor concept into the same space, so as to better evaluate the similarity between concepts; by introducing the attention mechanism, different weights are given to different neighbor concepts to ensure that the candidate concept c i Neighbor concepts with high correlation receive more attention, thus helping to disambiguate concepts.

[0553] Formula 4 is used to generate the final link by matching the candidate concepts with the existing concepts in the knowledge graph; by introducing three factors, namely, disambiguation score, neighbor concept similarity, and confidence and distance measurement, it ensures that the candidate concepts can be accurately matched with the existing concept nodes in the knowledge graph; Formula 4 is used in the simulation model to finally determine the link relationship between the candidate concepts and the concepts in the graph.

[0554] Specifically, the feature fusion unit is after constructing the basic knowledge graph, and the present invention further constructs a multimodal knowledge graph; candidate concept generation is based on a weighted multimodal feature fusion candidate concept generation model; the multimodal feature fusion model is shown in Formula 1:

[0555]

[0556] In the formula, C i represents the i-th candidate concept, such as workshop equipment, process flow, or a specific workstation in the simulation platform; D m Refers to the dataset of the mth modality, such as the text dataset D1 contains the text descriptions in the workshop process documents, and the image dataset D2 contains the images of factory equipment; M refers to the number of modalities, W m represents the weight parameter for the mth mode; f m (D m ,θ m ) refers to the feature extraction function used for this modality, D m is the modal data set, θ m It is the set of model parameters under each mode, such as the weight matrix of BiLSTM and the key point detection parameters of SURF.

[0557] Concept disambiguation performs weighted fusion of multimodal features based on the attention mechanism and eliminates ambiguity by maximizing the similarity between related concepts. The introduced weighted attention mechanism model is shown in Formula 2:

[0558]

[0559] In the formula, S(c i ) represents the candidate concept c i The disambiguation score is finally used to judge c i Is it the correct concept? Represents the candidate concept ci The multimodal feature representation at the Lth layer; σ represents the activation function, M(L-1) represents the weight matrix of the L-1th layer, and M(c i ) represents the candidate concept c i The associated set of neighbor concepts, is the attention weight, indicating the importance of neighbor concept x in generating c i Importance when scoring disambiguation; is the multimodal feature representation of the neighbor concept x at the L-1 layer, which is used to calculate the candidate concept c i The feature of , d(L-1) is the bias vector of the L-1 layer, which corrects the final feature representation. Attention weight The calculation is shown in formula 3:

[0560]

[0561] Where k is the vector parameter used to calculate the attention score, M key and M val is the weight matrix used to map the candidate concept and neighbor concept features into the same space.

[0562] After candidate concepts are generated and disambiguated, the final concept link is completed by comparing the candidate concepts with the concepts in the basic knowledge graph. This process uses the aforementioned weighted multimodal feature fusion model to calculate the matching score of the candidate concepts and link the candidate concepts with the concept nodes in the knowledge graph. In the concept link, concept confidence and distance metrics are additionally introduced to enhance the accuracy of concept link. The matching and linking process of candidate concepts can be expressed as Formula 4:

[0563]

[0564] In the formula, c * It represents the candidate concept with the highest score after considering multiple factors. α is the weight of the candidate concept score, which is used to adjust the influence of the candidate concept score in the final decision. β is the weight parameter considering the similarity of neighbor concepts, which is used to adjust the influence of neighbor concepts on the candidate concept link. Z is the normalization factor to ensure that the influence of similarity is within a reasonable range. It provides a platform for multiple sources such as workshop equipment, process flow or simulation platform to communicate with c i The set of associated neighbor concepts. Represents the candidate concept c i The weight between the neighbor concept x and the neighbor concept x indicates its influence in the link decision; sim(c i ,x) represents the candidate concept c iThe similarity between the candidate concept x and its neighbor concept x measures their distance in the feature space; δ represents the weight parameter of the confidence and distance metric, which is used to balance the confidence of the candidate concept and its distance from the existing knowledge graph concept. i ) represents the candidate concept c i The confidence of , indicating the matching confidence with the target concept; d(c i ,E) represents the candidate concept c i The cosine similarity with the concept set E in the knowledge graph; the graph can be stored and managed using graph databases such as Neo4j or Amazon Neptune.

[0565] The operation process of the feature fusion unit is as follows:

[0566] S1. Data extraction: Extract text and image features from process documents and equipment images;

[0567] S2. Multimodal feature fusion: weighted fusion of extracted text and image features to generate candidate concepts;

[0568] S3. Disambiguation and linking: Disambiguation is done through the attention mechanism, and candidate concepts are eventually linked to concepts in the knowledge graph to ensure that the concepts in the simulation model are consistent with the actual situation.

[0569] The results of the feature fusion unit after the operation process are: generating candidate concepts, disambiguation and optimization, and linking knowledge graphs. First, based on the multimodal data of the cigarette silk workshop, multiple candidate concepts are generated; through the attention mechanism, ambiguity is eliminated to ensure that the candidate concepts accurately reflect the actual equipment or process flow in the workshop; finally, the system links the candidate concepts with the existing concept nodes in the knowledge graph to ensure that the concept expression in the simulation model is accurate.

[0570] Example 13: Implementation of the feature fusion unit in the cigarette making workshop simulation model

[0571] In a specific embodiment, in a simulation model of a cigarette silk-making workshop, a feature fusion unit realizes multimodal feature fusion and concept linking through data extraction and multimodal feature generation, multimodal feature fusion and weighted calculation, and concept disambiguation and linking.

[0572] The operation steps of the feature fusion unit are as follows:

[0573] S1. Data extraction and multimodal feature generation: The data sources of the cigarette silk workshop are wide-ranging, including text data (such as process documents) and image data (such as equipment pictures); these data are processed and feature extracted through different technical means, laying the foundation for subsequent fusion and linking.

[0574] For text processing, the BiLSTM model is needed. Text data mainly comes from process documents, equipment instructions and process descriptions; the text is processed by BiLSTM (bidirectional long short-term memory network) to extract key information, such as equipment name, material handling process, production steps, etc.; BiLSTM can efficiently extract key entities and relationships in the process by capturing sequence information, ensuring a comprehensive understanding of process information.

[0575] For image processing, the SURF algorithm needs to be used. The image data mainly comes from photos or layout drawings of workshop equipment. The SURF (Speeded Up Robust Features) algorithm is used to process the image, extract key points in the image, and generate corresponding equipment feature vectors; these features can be used to identify the shape, position and appearance characteristics of the equipment, thereby providing physical feature information for the simulation model.

[0576] S2. Multimodal feature fusion and weighted calculation: After data extraction, data from different modalities such as text and images need to be fused by weighted fusion to generate candidate concepts. The core of multimodal feature fusion and weighted calculation is to effectively combine information from multiple data sources so that the candidate concepts finally generated can accurately reflect the actual situation of the workshop.

[0577] The model weights are weighted features of different modalities to determine the importance of each modality in the fusion process; for example, for some specific workshop tasks, text information may be more important than image information (such as process flow), while in other tasks, the physical features of the equipment may account for a larger proportion (such as equipment status recognition). Therefore, during the fusion process, the weights are adjusted according to actual needs to ensure that the generated candidate concepts can accurately reflect the production situation of the workshop.

[0578] The weighted fusion model assigns different weights W according to the quality and actual contribution of different modalities. m , for example, the features extracted by BiLSTM for text data may be given higher weights, while the SURF features for device image data are second.

[0579] S3. Concept disambiguation and linking: After feature fusion is completed, candidate concepts need to be generated and disambiguated to ensure that the candidate concepts can accurately match the existing concepts in the knowledge graph. This requires three stages: candidate concept generation, attention mechanism disambiguation, and concept linking and matching. The candidate concept generation is based on fused features. The candidate concepts are generated based on data such as equipment and process flow in the workshop by fusing text and image features. For example, after extracting the image features of the equipment and the relevant descriptions in the process documents, the generated candidate concept alternatives are "shredder" or "shredder". These candidate concepts contain the physical and operational information of the equipment. The feature fusion of the generated candidate concepts: the multimodal features after weighted calculation can accurately describe the equipment and processes in the workshop, ensuring that the candidate concepts contain complete information.

[0580] S4. Attention mechanism disambiguation: The candidate concepts are disambiguated by introducing an attention mechanism. The system can calculate the disambiguation score of each candidate concept based on the similarity between the candidate concept and its neighbor concepts. The attention mechanism assigns different weights to neighbor concepts to ensure that the association between the candidate concept and the adjacent processes or equipment is more reasonable.

[0581] The similarity calculation of the neighbor concepts is that the attention mechanism calculates the similarity between the neighbor concepts and the candidate concepts to determine whether the candidate concepts conform to the context information. For example, if a candidate concept is highly related to the surrounding equipment and process flow, the concept is given a higher weight, thereby improving the accuracy of disambiguation.

[0582] S5. Concept linking and matching: Concept linking and matching requires confidence, similarity of neighbor concepts and final linking; the confidence and similarity of neighbor concepts are the calculation of confidence and similarity introduced in the matching process of candidate concepts with existing concepts in the knowledge graph, which further measures the matching degree between the candidate concept and the target concept in the knowledge graph. The higher the confidence, the higher the consistency between the candidate concept and the actual workshop concept.

[0583] The similarity between the candidate concept and the neighboring concepts is used to ensure that the candidate concept is not only accurate in an independent case, but also consistent with other concepts in the context. This is achieved by measuring the similarity between the candidate concept and the neighboring concepts (such as the continuity of the process flow, the physical layout between the equipment, etc.);

[0584] The final link is that once the calculation is completed, the system will link the candidate concept with the corresponding node in the knowledge graph according to the disambiguation score, similarity and confidence, to ensure that the concepts in the simulation model are consistent with the actual concepts in the knowledge graph. For example, the system will match the candidate concept "shredder" with the corresponding "shredder" node in the knowledge graph to ensure the integrity and accuracy of the simulation model.

[0585] Module 4: Cigarette Making Workshop Simulation Knowledge Interaction and Problem Diagnosis Module

[0586] The cigarette silk workshop simulation knowledge interaction and problem diagnosis module includes a dialogue management and knowledge query unit, a problem diagnosis and optimization suggestion unit, and a model recommendation and interactive learning unit.

[0587] Unit 8: Dialogue Management and Knowledge Query Unit

[0588] The main tasks of the dialogue management and knowledge query unit are to parse user input, query the knowledge graph, generate responses and manage the dialogue state, and finally complete the natural language parsing. Parsing user input is to parse the user's natural language input through natural language processing (NLP) technology, identify the user's query intention, and extract relevant key information (such as equipment type, process flow, etc.); querying the knowledge graph is the retrieval function of the knowledge graph query engine (such as SPARQL or graph query algorithm), extracting relevant structured data from the logistics simulation multimodal knowledge graph; generating responses is to convert the data in the knowledge graph into natural language output through natural language generation (NLG) technology, and generate text replies that meet user needs; managing the dialogue state is to maintain the coherence and logic of the dialogue through context-aware technology and dialogue history tracking.

[0589] The technical means adopted by the dialogue management and knowledge query unit are natural language processing (NLP), intent recognition and slot filling, dialogue management, knowledge graph retrieval and natural language generation (NLG).

[0590] The natural language processing (NLP) uses a Transformer-based model (such as BERT, GPT) to semantically understand the user's natural language input. The Transformer model is good at processing natural language, can capture complex language context relationships, and identify the intention behind the user's query. The natural language processing (NLP) is a query input by the cigarette silk workshop through text or voice, such as "check the status of the shredder". The Transformer model will embed the input into words and analyze the context through a multi-layer attention mechanism to extract the user's intention and key slot information. For example, the model will parse the input as "equipment status query" and identify "shredder" as the queried device.

[0591] The intention recognition and slot filling are to perform intent recognition on the query of the cigarette silk workshop through a deep learning model (such as a classification model based on BERT) and extract relevant slot data. First, intent recognition is required, that is, using a deep learning model to identify the user's query intention (such as querying equipment status, process flow, fault diagnosis, etc.). For example, when the cigarette silk workshop inputs "What are the process parameters of the shredder?", the simulation model recognizes that the user's intention is to query the equipment status. Next, slot filling is required, that is, the system extracts slot data related to the intent, such as equipment name, status type and other information; for the above query, slot filling will recognize that "shredder" is the equipment name and "process parameters" is the information to be queried.

[0592] The dialogue management is to track the dialogue history through LSTM or GRU to ensure the consistency and coherence of the dialogue; these recurrent neural networks (RNN) can remember the context of the dialogue, so that the system can effectively deal with multiple rounds of dialogue and long-term interaction. First, the dialogue history needs to be tracked, that is, when the workshop makes inquiries in multiple rounds, the system records and remembers the previous dialogue content through LSTM or GRU. For example, the workshop first asks "the process parameters of the wire cutter" and then asks "what is the conversion rate?" The system can remember that the last device was "wire cutter" and accurately respond to the conversion rate information; then context management is performed, that is, the system updates the state according to the context of the dialogue to ensure that consecutive questions get coherent answers.

[0593] The knowledge graph retrieval is to retrieve relevant nodes and relationships in the knowledge graph through SPARQL or graph query algorithms to extract the required structured data. First, a SPARQL query is performed, that is, the system generates a SPARQL query statement based on the extracted intent and slot data, and locates the relevant equipment or process nodes in the knowledge graph. For example, if the user queries the "status of the wire cutter", the system will generate a SPARQL query to query the wire cutter node and its related "status" information; then, a graph query algorithm is performed, that is, if the knowledge graph is relatively complex, the system can also use a graph query algorithm to locate all information related to a certain node (such as equipment or process flow), such as the equipment's process parameters, status, historical records, etc.

[0594] The natural language generation (NLG) is to convert structured data into natural language through template generation method or deep generative model to generate user-readable responses. First, template generation is performed: the system can predefine multiple templates, fill in the data extracted from the knowledge graph into these templates, and quickly generate responses. For example, when querying "the process manufacturing capacity of shredder 001", the response generated by the template is: "The process manufacturing capacity of shredder 001 is 6400kg / h"; then, a deep generative model is performed, that is, for more complex and diverse responses, the system can use GPT-type models to dynamically generate natural language responses based on data content.

[0595] The operation process of the dialogue management and knowledge query unit is as follows:

[0596] S1. The workshop inputs questions through text or voice, and the system first uses the Transformer model to parse the semantics and identify the workshop's intention;

[0597] S2. The system tracks the historical information of the conversation through LSTM or GRU to maintain context consistency;

[0598] S3. The system generates SPARQL queries based on the extracted intent and slot data to retrieve data such as equipment status and process flow in the knowledge graph;

[0599] S4. The retrieved data is converted into natural language output through the natural language generation (NLG) module and finally fed back to the workshop.

[0600] The results of the dialogue management and knowledge query unit after running are: accurate intent recognition, continuous dialogue management, fast knowledge retrieval and natural language response generation. The system can extract key intent and slot data from user input and accurately understand the problems needed by the workshop; through context-aware technology, the system can maintain consistency in multiple rounds of dialogue, avoid repetitive problems, and improve the experience of the workshop dialogue system; the system can quickly retrieve the structured information needed by the workshop in the knowledge graph through SPARQL or graph query algorithms; finally, the system converts the data in the knowledge graph into natural language and generates clear text responses that are useful to the workshop.

[0601] Example 14: Implementation of the Dialogue Management and Knowledge Query Unit in the Cigarette Making Workshop Simulation Model

[0602] In a specific embodiment, in the simulation model of a cigarette silk-making workshop, the dialogue management and knowledge query unit implements the complete process from natural language parsing, dialogue management, knowledge graph retrieval to natural language generation, helping the system to achieve intelligent human-computer interaction, accurately identifying user needs, retrieving relevant knowledge, and providing feedback in the form of natural language.

[0603] The operation steps of the dialogue management and knowledge query unit are as follows:

[0604] S1. Natural language parsing and intent recognition: The first step in input processing, that is, through deep learning and NLP technology, the system can accurately understand the user's natural language input, identify the core intent of the user's query, and extract relevant information; the natural language parsing and intent recognition require the use of Transformer model and slot filling.

[0605] The Transformer model (such as BERT, GPT) is used to process the natural language input of the cigarette making workshop, and the context of the input sentence is deeply analyzed through a multi-layer attention mechanism to extract the query intention of the cigarette making workshop. For example, when the cigarette making workshop asks "What is the working status of the shredder?", the system recognizes through the Transformer model that the user's intention is to query the equipment status and extracts the specific equipment name "shredder".

[0606] The slot filling is to extract key information (such as equipment name, process flow, etc.) from the input of the cigarette shredding workshop. This information will serve as the basis for the subsequent knowledge retrieval. For example, in the query "working status of the shredder", the slot filling extracts "equipment name: shredder" and "query content: working status".

[0607] S2. Dialogue state management: It needs to be able to handle multiple rounds of dialogue to ensure that the system remembers the previous interactions in the cigarette silk workshop and can give a coherent response based on the context; the dialogue state management requires context awareness, history tracking, and status updates.

[0608] The context perception and history tracking is that the system tracks the conversation history of the cigarette silk workshop through the LSTM or GRU model and maintains the context consistency of the conversation; then the context perception technology of the recursive neural network (RNN) enables the system to understand and remember the key information in multiple rounds of interaction in the cigarette silk workshop, so that when the cigarette silk workshop continues to ask other equipment or related questions, the system can still maintain the conversation coherence. For example, if the user first asks "the working status of the shredder", and then asks "what is its conversion rate?" the system can remember the "shredder" mentioned above as the query object, and can understand that "it" refers to the "shredder".

[0609] The status update is the system updating the conversation status based on the input from the cigarette shredding workshop, ensuring that subsequent queries can be made based on the latest information. The status update ensures that the system can always correctly parse the context of the current conversation and ensure that the response given is accurate and relevant. For example, after querying "shredder status", the system will dynamically update the conversation status to "Current query object: shredder", so that subsequent queries are automatically associated with the device.

[0610] S3. Knowledge graph retrieval: In order to give accurate answers, the system needs to search in the knowledge graph; through the SPARQL query language or graph query algorithm, the system can find the equipment, process flow or status information related to the cigarette silk workshop query, and return structured data, and parse it into specific information; the knowledge graph retrieval requires SPARQL query and knowledge graph structured data extraction.

[0611] The SPARQL query is based on the fact that SPARQL is a query language used for knowledge graph data retrieval. The system will generate SPARQL query statements based on user input and slot data to locate data on related equipment or process flows from the knowledge graph. For example, if a user queries "fault records of tobacco drying machine", the system will generate a corresponding SPARQL query to retrieve fault record data on "tissue drying machine" in the knowledge graph; if the system queries the tobacco silk workshop, it will use SPARQL to retrieve the process flow and status parameters of the shredder, and use these data as the basis for subsequent responses.

[0612] The knowledge graph structured data extraction is to return the retrieved data in a structured form, and the system will extract and parse these structured data, including equipment status, process flow, material conversion rate and other data. For example, when the system retrieves "process parameters of the shredder", the returned data includes information such as equipment status, current operating parameters, tobacco material processing speed, etc.

[0613] S4. Natural Language Generation (NLG): The system needs to convert structured data into natural language responses that can be understood by the cigarette making workshop. It can use templates to generate simple responses, or use deep generative models for complex natural language generation; Natural Language Generation (NLG) technology includes template generation and deep generative models.

[0614] The template generation means that the system can generate standardized answers through preset templates. For example, for equipment status inquiries, there can be a fixed template: "The origin of equipment XX is XX, and the process manufacturing capacity is XX."

[0615] The deep generative model is for the complex scenario of multi-operational operation in the cigarette silk workshop. The system can use deep generative models such as GPT to automatically generate fluent natural language text according to the specific content of the data. For example, if the cigarette silk workshop asks about the "status of the shredder", the system can generate the following reply after searching the knowledge graph: "The current working state of the shredder is normal, and the material conversion rate is 95%."; For example, when the cigarette silk workshop asks about the "status of the shredder", the system can generate a response: "The current working state of the shredder is normal, and the last maintenance was two weeks ago."

[0616] Unit 9: Problem diagnosis and optimization suggestions unit

[0617] The main task of the problem diagnosis and optimization suggestion unit is to assist in identifying problems in the simulation model, optimize model parameters, and provide improvement suggestions through systematic analysis and reasoning; the ultimate goal of the problem diagnosis and optimization suggestion unit is to improve the efficiency of the modeling process and the accuracy of the model, ensure that the simulation results can reflect the actual situation, and assist effective decision-making.

[0618] The main functions of the problem diagnosis and optimization suggestion unit are problem identification, optimization suggestion, problem diagnosis and comprehensive reasoning.

[0619] The problem identification is to automatically detect potential problems or abnormal situations in the simulation model; the optimization suggestions are to make suggestions for optimizing model parameters, improving model structure or operating procedures based on the identified problems; the problem diagnosis is to analyze the abnormalities that occur in the simulation model during operation, identify the root cause of the problem, and ensure that the cause of the fault can be accurately located; the comprehensive reasoning is based on three reasoning methods of representation learning: rules, cases and deep learning, which integrates different information sources to improve the accuracy and practicality of diagnosis and suggestions.

[0620] The technical means adopted by the problem diagnosis and optimization suggestion unit combine multiple reasoning and analysis methods to help systematically analyze the simulation model of the cigarette silk workshop and provide optimization suggestions; these technical means include rule-based, case-based, representation learning-based reasoning and fault tree analysis (FTA).

[0621] The rule-based reasoning requires a preset rule set. The system uses the predefined rule set for automatic checking according to the structure and parameters of the simulation model. These rules include standards for equipment operating parameters, process flow settings, equipment operation time, etc. The rule set defines the optimal parameter range for certain key operations, such as temperature, pressure, flow, etc. When the parameters in the simulation model exceed the preset standard range, the system will automatically trigger a warning and make adjustment suggestions. The system will check whether these parameters are within the expected optimization range. If the operating temperature of a certain device exceeds the optimal setting value, the system will give temperature adjustment suggestions; if the optimal speed of the shredder is 300-500r / min, if the shredder blade speed in the simulation model exceeds this range, the system will issue a pr...

Claims

1. A cigarette silk-making workshop simulation model auxiliary system, comprising five independent modules, namely, a cigarette silk-making workshop simulation model tag management module, a cigarette silk-making workshop simulation basic knowledge extraction module, a cigarette silk-making workshop simulation multimodal knowledge graph construction module, a cigarette silk-making workshop simulation knowledge interaction and problem diagnosis module, and a cigarette silk-making workshop system quality assessment module, characterized in that: The cigarette silk workshop simulation model tagging management module includes a simulation model feature extraction unit, a simulation model annotation unit, and a simulation model feature data management unit, which are used to accurately define and describe the cigarette silk workshop simulation model, and combine the marked information with the knowledge graph, so as to provide a cigarette silk workshop model recommendation function based on the knowledge graph for the dialogue simulation platform system; The basic knowledge extraction module for cigarette silk workshop simulation includes a cigarette silk workshop data extraction unit and a cigarette silk workshop data processing unit, which are used to extract key rules and data of the cigarette silk workshop, and obtain basic knowledge data related to the cigarette silk workshop simulation platform by constructing entity relationships, ternary relationships and knowledge inheritance hierarchical structures of knowledge graphs; The multimodal knowledge graph construction module for simulating the cigarette silk-making workshop includes a basic graph construction unit and a feature fusion unit, which are used to integrate and analyze data from various information systems of the cigarette silk-making workshop (such as the manufacturing execution system MES, the enterprise resource planning system ERP, the product lifecycle management system PLM, and data in process documents related to cigarette silk-making), and to construct a comprehensive knowledge graph that supports optimization analysis of the operation of the cigarette silk-making workshop, fault diagnosis of cigarette silk-making operations, and decision support; The cigarette silk workshop simulation knowledge interaction and problem diagnosis module includes a dialogue management and knowledge query unit, a problem diagnosis and optimization suggestion unit, a model recommendation and interactive learning unit, which is used to accurately parse the natural language input of the cigarette silk workshop, extract key knowledge using the intent recognition algorithm and slot filling mechanism, and use knowledge graph reasoning to capture the complex relationship between related entities in the cigarette silk operation, perform multi-step relationship reasoning, and convert the actual demand description of the cigarette silk workshop operation into a graph query expression. Then, using subgraph matching and query parsing methods, it extracts improvement suggestions for related production scheduling, quality management, process formula management, and tobacco leaf or tobacco material supply chain management from the graph, and comprehensively obtains the final knowledge interaction and problem diagnosis; The cigarette silk-making workshop system quality assessment module includes a storage device and processing equipment accuracy verification unit, a tobacco processing rule integrity check unit, a tobacco production practicality assessment unit, a tobacco finished product quality index calculation unit, and a system dynamic update support unit. After the system is built, it is used to comprehensively evaluate the overall quality of the cigarette silk-making workshop logistics simulation multimodal system to ensure that the actual application of the system in the cigarette silk-making operation shows accuracy, completeness, consistency and practicality, thereby effectively supporting the optimization and decision-making process of the cigarette silk-making workshop simulation model.

2. According to claim 1, a cigarette thread making workshop simulation model auxiliary system is characterized in that: The simulation model feature extraction unit automatically identifies and extracts key feature data from simulation models of different workshops; the technical means adopted by the simulation model feature extraction unit are structured data analysis technology, computer vision and image processing technology, natural language processing technology and data fusion and integration technology; The simulation model annotation unit is used to perform hierarchical annotation of explicit information and implicit information on the simulation model; the model annotation unit adopts a variety of technical means to achieve comprehensive annotation of the model, including automatic annotation and manual annotation; The simulation model feature data management unit is responsible for storing simulation model files and feature data to ensure integrity and traceability. By establishing a multidimensional index structure, it supports rapid retrieval and associated query of model feature data. For each model, the feature data management unit will classify the acquired data and perform structured storage. The optional data format includes JSON. The technical means adopted by the simulation model annotation unit are feature data storage and multidimensional indexing, high-dimensional feature mapping and dynamic embedding mechanism, dynamic feature mapping tensor adjustment and similarity calculation and clustering algorithm. The data extraction unit of the cigarette silk-making workshop obtains relevant data from different information systems (such as manufacturing execution system MES, enterprise resource planning system ERP, product lifecycle management system PLM) and process documents to support the construction of workshop process simulation model; the technical means adopted by the data extraction unit of the cigarette silk-making workshop are the workshop's manufacturing execution system, enterprise resource planning system, product lifecycle management system and related process documents to obtain relevant data of workshop equipment and processes; the technical means adopted by the data extraction unit of the cigarette silk-making workshop are data integration tools (ETL), API (application programming interface) calls, database queries, document parsing tools and data cleaning and conversion technologies; The data processing unit of the cigarette making workshop performs data splitting and noise reduction after successfully acquiring relevant data; For structured data, concepts and relationships are automatically generated through predefined templates, and knowledge is extracted in combination with ETL tools; for unstructured data, part of the data is labeled through concepts and relationships, a domain knowledge training library is constructed, and knowledge is extracted using relevant models; multiple training algorithms are used in the extraction process to improve the recognition and extraction capabilities of data types, sequence labeling models such as BiLSTM are used for text data, and feature extraction is completed for image data using the SURF feature extraction algorithm; the technical means used by the data processing unit of the cigarette silk workshop are ETL tools, natural language processing (NLP) technology, image processing algorithms, noise reduction algorithms, and multiple machine learning training algorithms; The basic graph construction unit constructs multiple basic knowledge graphs through preliminary structured processing and data integration to achieve conceptual links between different systems such as workshop equipment, process flow, simulation platform, etc. Each basic knowledge graph uses the workshop equipment, process flow, workstations in the simulation platform, etc. in their respective systems as node concepts, and the association between equipment and process, and the association between workstations and processes as edges, reflecting the equipment, process flow and their association in the workshop, ultimately supporting a more accurate and intelligent simulation process; The technical means adopted by the basic graph construction unit are knowledge graph construction tools, data integration and ETL processing, relationship extraction and graph generation and graph optimization; The feature fusion unit generates and disambiguates candidate concepts by integrating the features of multimodal data (text, image), matches and links them with concepts in the basic knowledge graph, and combines multimodal information from different data sources to eliminate ambiguity and ensure that the generated candidate concepts can be accurately linked to the corresponding concept nodes in the knowledge graph; the technical means adopted by the feature fusion unit are weighted multimodal feature fusion model, attention mechanism disambiguation and matching and linking of candidate concepts; The dialogue management and knowledge query unit is used to parse workshop input, query knowledge graph, generate response and manage dialogue status, and finally complete natural language parsing; the technical means adopted by the dialogue management and knowledge query unit are natural language processing (NLP), intent recognition and slot filling, dialogue management, knowledge graph retrieval and natural language generation (NLG); The problem diagnosis and optimization suggestion unit is to perform problem identification, optimization suggestion, problem diagnosis and comprehensive reasoning; the problem identification is to automatically detect potential problems or abnormal conditions in the simulation model; the technical means adopted by the problem diagnosis and optimization suggestion unit are combined with a variety of reasoning and analysis methods to help systematically analyze the simulation model of the cigarette silk workshop and provide optimization suggestions; These technical approaches include rule-based, case-based, representation learning-based reasoning, and fault tree analysis (FTA); The recommendation and interactive learning unit is to perform model recommendation, interactive optimization and recommendation explanation; analyze the needs input by the workshop, match them with the classified simulation model feature library, generate a model recommendation list most relevant to the needs, and then recommend the simulation model that best meets the needs to the cigarette silk workshop based on similarity evaluation, accelerate the deployment and optimization of the model, and then provide a detailed explanation of the recommendation results, display the matching basis of each recommended model with the needs of the cigarette silk workshop, help the workshop understand the reasons for the recommendation, and finally support the workshop to dynamically adjust the recommended model, through the interactive feedback mechanism, continuously optimize the model according to the actual application scenario, to ensure that the model is closer to the actual needs of the workshop; the technical means adopted by the recommendation and interactive learning unit are model feature library construction and classification, demand analysis and matching, similarity evaluation and recommendation, recommendation explanation function and interactive learning and feedback mechanism; The storage device and processing equipment accuracy verification unit verifies the accuracy of the data, parameters and their relationships in the simulation model to ensure that the equipment, processes and concepts and relationships in the simulation model used in the system are consistent with the equipment and process flow in the actual production environment, and then compares the equipment parameters (such as capacity, operating speed, energy consumption, etc.) in the simulation model with the actual equipment parameters to ensure parameter matching, and then checks whether the relationship between the equipment and the process flow in the system is consistent with the actual situation, such as the upstream and downstream relationship of the equipment and the operation order; If inconsistencies or errors are found, generate feedback reports and make correction suggestions; The technical means adopted by the storage device and processing equipment accuracy verification unit are data comparison and verification, knowledge graph consistency verification, rule engine verification and data mapping and feedback mechanism; The main task of the integrity check unit is to detect whether there are any missing process links, equipment parameters, and whether there are any internal contradictions or inconsistent relationships through systematic evaluation, so as to ensure that the simulation system contains all key concepts, equipment, process flows and their relationships involved in the workshop production process without missing or contradictory concepts and information; the technical means adopted by the integrity check unit are concept coverage check, logic consistency verification, rule engine and consistency check, and data mapping and feedback mechanism; The finished tobacco product quality index calculation unit is used to calculate key performance indicators, generate quality reports and support system optimization and adjustment, so as to evaluate the performance of the system, measure the system's performance in processing cigarette shred workshop inquiries, data feedback, simulation model support, etc., and generate detailed quality reports, using the system's accuracy, recall rate, F1 value, and coverage rate as indicators to optimize and adjust the system; the technical means used by the finished tobacco product quality index calculation unit are accuracy calculation, recall rate calculation, F1 value calculation, coverage rate calculation, performance evaluation tools, and automated testing and feedback mechanisms; The system dynamic update support unit performs version management, incremental update, compatibility verification, and automatic update and notification to ensure that the system can be updated and adapted to new data and requirements in a timely manner when equipment and process flow change. Then, through version management and incremental update mechanism, the system changes are tracked and kept real-time and accurate, ensuring that the simulation system can continuously optimize and reflect the latest production environment. The technical means adopted by the dynamic update support unit are version management tools, incremental update mechanism, compatibility verification mechanism and automatic update and notification mechanism.

3. According to claim 2, a cigarette thread making workshop simulation model auxiliary system is characterized in that: The simulation model feature extraction unit automatically identifies and extracts key feature data, including structured data, image data, and text data; the structured data is data organized and stored in a specific format, and is data information stored in a database, table, or other structured data storage system, which is easy to be represented and processed in the form of a table or database, including the attributes of the equipment object (such as type, specification, performance parameters), the upstream and downstream relationships between equipment, the spatial distribution data of the equipment, and process parameters (such as temperature, speed, and time); the image data is information captured by vision, stored and processed in the form of an image, and is used to intuitively describe the appearance, layout, and spatial distribution of the equipment, including a three-dimensional shape cross-sectional view of the equipment, a workshop layout diagram, and a photo of the equipment installation location; the text data is information presented in the form of natural language text, and comes from sensor data, equipment status information, working parameter information, operation log information, workshop transport parameter information, quality index information, and / or fault information, including the process description, operation manual, simulation log, workshop input instruction, or description text of the cigarette silk workshop; The explicit information of the simulation model annotation unit is information that can be clearly defined and identified in the simulation model, including intuitive and specific data information, which is easy to extract and annotate, and can directly describe the appearance, structure and basic properties of the model; the explicit information includes basic information, hierarchical information, and layout images of the simulation model; basic information includes model name, model type, and model update date; hierarchical information includes object library structure, object attributes, and object affiliation; layout image information includes a two-dimensional image of the workshop and a three-dimensional shape cross-sectional view of a single machine; the implicit information of the simulation model annotation unit is a feature that is not clearly defined in the simulation model but describes the model. The implicit information needs to be further analyzed and refined from the basic marking content of the model, and can supplement and deepen the understanding of the simulation model; The data acquired in the data extraction unit of the cigarette silk-making workshop include the workshop's basic material data, the workshop's material processing rules and related process standards, the workshop's production scheduling data, the workshop's equipment modeling data, and the workshop's process sections and process data; the workshop's basic material data include data information on tobacco material name, material type, material code, and material grade; the workshop's material processing rules and related process standards include information on group processing, mixing weight, mixing sequence, upper and lower limits of storage time, and conversion rate of tobacco materials in the processing stage; the workshop's production scheduling data include data information on cigarette silk-making workshop shifts, cigarette silk-making workshop teams, and factory calendars; the equipment modeling data of the cigarette silk-making workshop include data information on basic equipment parameters, control rules, and status of storage equipment, processing equipment, and auxiliary equipment; the workshop's process sections and process data include data information on design flow, brand change time, process length, and processing time of main equipment; the ownership relationship between the workshop's process sections, processes, and equipment, including data on processes included in a process section and the number of equipment included in a process section.

4. According to claim 2, a cigarette thread making workshop simulation model auxiliary system is characterized in that: The simulation model annotation unit allows manual input of text descriptions and utilizes a large language model (LLM) to split and convert the manually input text into standardized tags or data formats; The implicit information labeling is to extract implicit information using simulation logs, user input, and NLP technology (such as NER and LDA) in the cigarette silk workshop, and combine multiple NLP technologies in the implicit information extraction process of the cigarette silk workshop simulation model to form a new and specific simulation model combination and application scenario; The manual input and LLM conversion uses a large language model (LLM) to perform data conversion and annotation in a specific field to split the input text into multiple meaningful parts and convert them into structured data entries (such as workshop scene names, category data information). The system automatically annotates the converted data entries as part of the implicit information field, and finally saves and updates the annotation information for subsequent retrieval and optimization.

5. According to claim 2, a simulation model auxiliary system for a cigarette thread making workshop is characterized in that: The high-dimensional feature mapping and dynamic embedding mechanism maps the feature data of the simulation model into a high-dimensional topological space and classifies it using the dynamic feature embedding mechanism; then uses nonlinear transformation (such as Tanh function) and weighted coefficient matrix to transform the model feature data into a multidimensional feature data vector y, y = [y1, y2, ..., y n ], each y i Represents different model features, multi-dimensional logistics feature mapping function f MLFM As shown below: L=f MLFM (y)=[L1(y),L2(y),...,L m (and)] Each mapping dimension L i The formula for (y) is as follows: In the formula, λ ij is the weight coefficient of the i-th dimension to the j-th feature, γ ij is the scaling factor of each dimension of the simulation model, δ ij is the normalization coefficient of the i-th dimension to the j-th feature; in the simulation model of the cigarette shred workshop, the physical properties and operating parameters (such as temperature, speed, and output rate) of equipment objects (such as mixers, shredders, and dryers) will be mapped into high-dimensional vectors and converted into embedded representations in high-dimensional topological space through the above mapping formula; The dynamic feature map tensor adjustment is performed using the dynamic feature map tensor D(k), and an adjustment factor is introduced to represent the weight change of the feature in different logistics operations; the adjustment factor is calculated using the Sigmoid function: The dynamic feature map tensor D(k) is generated by introducing an adjustment factor based on the vector y, which represents the weighted value in the kth logistics operation: D(k)=[λ1(k)y1,λ2(k)y2,...,λ n (k)y n ] Among them, λ i (k) The sigmoid function is used to calculate the adjustment factor of each feature in the kth logistics operation; the sigmoid function is often used to limit the value range of the adjustment factor to between (0, 1) to avoid the influence of extreme values ​​and ensure that the adjusted feature weight is within a reasonable range; In the simulation model of the cigarette shredding workshop, the operating characteristics of different equipment have different importance in different logistics operations (such as raw material input, mixing, cutting, etc.); by dynamically adjusting the weights of these characteristics, the influence of characteristic data in specific operations can be more accurately reflected. For example, in the mixing process, the speed and temperature characteristics of the mixer may have higher weights, while in the shredding process, the blade speed and feed speed characteristics of the shredder have higher weights; The similarity calculation is to calculate the similarity score between the simulation models through multiple metrics (such as logistics topology space distance, dynamic feature mapping tensor similarity, cosine similarity); calculate the classification score between the simulation models and classify them according to the score using a clustering algorithm; the classification score S(L) between the i-th and j-th logistics simulation models in the k-th logistics operation i , L j , k) is calculated as follows: S(L i ,L j ,k)=α·d(L i ,L j )+β·sim(D i (k),D j (k)) In the formula, α represents the weight coefficient of the logistics topological space distance, β represents the weight coefficient of the dynamic feature map tensor similarity, sim(D i (k), D j (k)) indicates D i (k) and D j (k) The cosine similarity between them.

6. According to claim 2, a simulation model auxiliary system for a cigarette thread making workshop is characterized in that: The feature fusion unit further constructs a multimodal knowledge graph after constructing the basic knowledge graph; candidate concept generation is based on a weighted multimodal feature fusion candidate concept generation model; the multimodal feature fusion model is shown in Formula 1: In the formula, C i represents the i-th candidate concept, such as workshop equipment, process flow, or a specific workstation in the simulation platform; D m Refers to the dataset of the mth modality, such as the text dataset D1 contains the text descriptions in the workshop process documents, and the image dataset D2 contains the images of factory equipment; M refers to the number of modalities, W m represents the weight parameter for the mth mode; f m (D m ,θ m ) refers to the feature extraction function used for this modality, D m is the modal data set, θ m is the model parameter set under each mode, such as the weight matrix of BiLSTM and the key point detection parameters of SURF; Concept disambiguation performs weighted fusion of multimodal features based on the attention mechanism and eliminates ambiguity by maximizing the similarity between related concepts. The introduced weighted attention mechanism model is shown in Formula 2: In the formula, s(c i ) represents the candidate concept c i The disambiguation score is finally used to judge c i Is it the correct concept? Represents the candidate concept c i The multimodal feature representation at the Lth layer; σ represents the activation function, M(L-1) represents the weight matrix of the L-1th layer, and M(c i ) represents the candidate concept c i The associated set of neighbor concepts, is the attention weight, indicating the importance of neighbor concept x in generating c i Importance when scoring disambiguation; is the multimodal feature representation of the neighbor concept x at the L-1 layer, which is used to calculate the candidate concept c i The feature of , d(L-1) is the bias vector of the L-1 layer, which corrects the final feature representation. Attention weight The calculation is shown in formula 3: Where k is the vector parameter used to calculate the attention score, M key and M val is the weight matrix used to map the candidate concept and neighbor concept features into the same space; After candidate concepts are generated and disambiguated, the final concept linking is completed by comparing the candidate concepts with the concepts in the basic knowledge graph. This process uses the aforementioned weighted multimodal feature fusion model to calculate the matching scores of candidate concepts and link the candidate concepts with the concept nodes in the knowledge graph. In the concept linking, concept confidence and distance metrics are additionally introduced to enhance the accuracy of concept linking. The matching and linking process of candidate concepts can be expressed as Formula 4: In the formula, c * It represents the candidate concept with the highest score after considering multiple factors. α is the weight of the candidate concept score, which is used to adjust the influence of the candidate concept score in the final decision. β is the weight parameter considering the similarity of neighbor concepts, which is used to adjust the influence of neighbor concepts on the candidate concept link. Z is the normalization factor to ensure that the influence of similarity is within a reasonable range. It provides a platform for multiple sources such as workshop equipment, process flow or simulation platform to communicate with c i the associated set of neighbor concepts; Represents the candidate concept c i The weight between the neighbor concept x and the neighbor concept x indicates its influence in the link decision; sim(c i ,x) represents the candidate concept c i The similarity between the candidate concept and the neighbor concept x measures their distance in the feature space; δ represents the weight parameter of the confidence and distance metric, which is used to balance the confidence of the candidate concept and its distance from the existing knowledge graph concept; conf(c i ) represents the candidate concept c i The confidence of , indicating the matching confidence with the target concept; d(c i ,E) represents the candidate concept c i The cosine similarity with the concept set E in the knowledge graph; the graph can be stored and managed using the graph database Neo4j or Amazon Neptune.

7. The cigarette thread making workshop simulation model auxiliary system according to claim 2 is characterized in that: The data extraction unit of the cigarette silk-making workshop combines NLP and knowledge graph to convert the entities and relationships extracted by NLP into structured data forms, which are used to build a domain knowledge base of the cigarette silk-making workshop to support the simulation model's understanding and simulation of complex process flows; the image processing of the data extraction unit of the cigarette silk-making workshop uses the SURF algorithm, which is not only used for equipment layout detection, but also closely combines the spatial information of these feature points with the simulation model to reflect the equipment status and layout dynamics in real time, and improve the mapping accuracy of the simulation model to the physical workshop environment; the comprehensive processing and intelligent optimization of multimodal data of the data extraction unit of the cigarette silk-making workshop uses a machine learning model to process time series data, image data and structured data, to achieve real-time optimization and fault prediction of the workshop production process, especially through deep learning training of specific equipment and process data of the cigarette silk-making workshop to enhance the intelligence of the simulation model; The deep integration of NLP and domain-specific knowledge graph in the dialogue management and knowledge query unit is to deeply integrate the Transformer-based NLP model with the multimodal knowledge graph of the specially constructed tobacco silk workshop, so as to directly retrieve, reason and generate responses highly relevant to the specific needs of the workshop from natural language; the dynamic dialogue management of the dialogue management and knowledge query unit and the multi-round interaction specific to tobacco silk are dialogue systems for tobacco silk workshops, which not only track the dialogue history through LSTM or GRU, but also perform dynamic context management in combination with the actual process and equipment status of the workshop, and can continuously adjust and optimize the dialogue strategy in multi-round interactions, so that the system response is more in line with the actual production environment; The dialogue management and knowledge query unit uses deep reasoning and multi-step query optimization based on knowledge graphs to perform multi-step reasoning and optimization on the complex associations between SPARQL and graph query algorithms and equipment and processes, forming an intelligent query and reasoning engine. In the adaptive natural language generation model of the dialogue management and knowledge query unit, in addition to template generation, the system also uses deep learning models (such as GPT) for adaptive generation in NLG, which can dynamically adjust the generated content according to the real-time data and changes of the workshop, so that the response is more in line with the complex tobacco silk workshop scene. The deep characterization and precise classification of specific fields of the recommendation and interactive learning unit, for the simulation model specific to the cigarette silk-making workshop, deeply extracts equipment parameters and process stage information, and performs precise classification, making the simulation model management and retrieval more accurate; the multi-dimensional demand analysis and precise matching of the recommendation and interactive learning unit is to identify basic needs, extract more detailed slot data (such as production efficiency, material processing time), and perform precise matching, so as to enhance the pertinence of simulation model recommendations; the similarity evaluation and feature space mapping of the recommendation and interactive learning unit are combined by using traditional similarity algorithms when making recommendations, combined with spatial mapping and multi-dimensional analysis of model features; the in-depth combination of recommendation explanation and user interaction of the recommendation and interactive learning unit is to explain the reasons for the recommendation in detail through visualization tools, and help the workshop understand the matching relationship between the characteristics of each model and the needs, and provide in-depth decision support; the real-time interactive learning and adaptive recommendation optimization of the recommendation and interactive learning unit is to record the feedback from the workshop, and immediately recalculate the similarity and optimization results of the recommendation model after each adjustment. The real-time two-way data verification and dynamic correction mechanism of the storage device and processing equipment accuracy verification unit can compare the simulation model with the knowledge graph in real time, dynamically verify whether the upstream and downstream relationships of the equipment and process flow are correct, and provide immediate feedback; the advanced data mapping and two-way feedback mechanism of the storage device and processing equipment accuracy verification unit adopts more sophisticated mapping technology to accurately match the actual equipment with the equipment in the simulation model one-to-one, and maintain dynamic synchronization of data. At the same time, it not only provides static feedback reports, but also supports dynamic adjustment suggestions, and directly converts feedback into automatic adjustment instructions for the simulation model; the intelligent knowledge graph consistency verification of the storage device and processing equipment accuracy verification unit automatically optimizes the node and relationship matching rules in the knowledge graph through intelligent algorithms (such as machine learning) to enhance the accuracy of verification; The deeply integrated version management and simulation model update of the dynamic update support unit uses a deeply integrated version control tool, so that each time the process parameters or equipment configuration is updated, a version record can be automatically generated, and rollback is allowed at the model level; an incremental update mechanism is adopted to update only part of the data of related equipment or processes, reducing the system downtime and update burden; when new equipment is introduced into the workshop or the process is changed, the system automatically performs a compatibility check and generates a detailed conflict report; the system not only automatically detects and updates data, but also can immediately feedback the update content and impact to the workshop through the notification system, and recommends verification.

8. The cigarette thread making workshop simulation model auxiliary system according to claim 2, characterized in that: The reasoning based on representation learning in the problem diagnosis and optimization suggestion unit is achieved through knowledge graph embedding, multi-step reasoning and implicit relationship optimization, and process flow linkage optimization; The knowledge graph embedding and multi-step reasoning are to convert entities and relationships in the knowledge graph into low-dimensional vectors through deep learning technology, and perform multi-step reasoning through these vectors, so as to mine the implicit connection between equipment and process flow and find optimization points that cannot be captured by traditional rules or case methods; The implicit relationship optimization and process linkage optimization are to capture complex implicit relationships through representation learning and perform process linkage optimization. By deeply combining deep learning with the workshop simulation model, it is possible to identify specific equipment and parameter problems in the workshop and discover potential collaborative relationships between equipment. The reasoning based on representation learning is a comprehensive integration of multiple reasoning methods, integrating rule reasoning, case reasoning, reasoning based on representation learning and fault tree analysis (FTA) in the same system, allowing the system to flexibly use different reasoning methods at different levels and in different scenarios to form more comprehensive and accurate diagnosis and optimization suggestions; further, through knowledge graph embedding and multi-step reasoning, the system can combine the operating status of the equipment in real time and make dynamic optimization suggestions.

9. An operating method of a cigarette thread workshop simulation model, characterized in that: The following steps are involved: Phase 1: Data extraction and knowledge graph construction S1. Data extraction: Extract basic data from the systems (MES, ERP, PLM, etc.) and process documents of the cigarette making workshop, covering equipment information, process flow, tobacco material data, production scheduling and other data information; S2. Entity and relationship generation: Convert the extracted data into entity concepts in the knowledge graph and define the relationships between entities, such as the upstream and downstream relationships of equipment, the association of process steps, and other data information; S3. Multimodal knowledge graph construction: Integrate text, images, and process data from different data sources to build a multimodal knowledge graph that includes equipment, process flow, and tobacco materials, providing a basis for subsequent reasoning and model management; Phase 2: Requirements Analysis and Reasoning S4. Demand input: The cigarette making workshop inputs its needs through voice or text, such as optimizing production processes or diagnosing equipment failures; S5. Intent recognition and slot filling: Use NLP technology to identify the core information of the demand, extract key information through the slot filling mechanism, and obtain relevant data from the knowledge graph; S6. Knowledge graph multi-step reasoning: Perform multi-step reasoning based on the knowledge graph to analyze the complex relationships between entities and identify production links that need to be optimized or potential sources of failure; Phase 3: Model recommendation and optimization S7. Model feature extraction and management: Extract feature data from existing simulation models, store and manage them in a structured manner to ensure data integrity; S8. Multidimensional index of model feature data: construct a multidimensional index to facilitate rapid retrieval of simulation model feature data; S9. Failure cause analysis and optimization suggestions: Identify the cause of the operation failure based on the reasoning results and make optimization suggestions to improve the simulation model; Phase 4: Simulation model matching and recommendation S10. Model recommendation and interactive learning: Recommend the most suitable simulation model based on the matching of requirements and simulation model characteristics, support dynamic adjustment and feedback, and realize interactive learning and optimization; S11. Model category matching: Retrieve the simulation model category matching the requirements through indexing to ensure that the recommended model meets the actual operation requirements of the workshop; S12. Generate recommendations and explanations: Generate a list of recommended simulation models and provide reasons for the recommendations, explaining the applicability of the models and the reasons for the recommendations; Phase 5: System Quality Assessment S13. Comprehensive quality assessment: Conduct quality assessment on the recommended simulation models and systems to check their accuracy, completeness, consistency and practicality to ensure the efficiency and reliability of the models.

10. The method for operating the simulation model of a cigarette thread-making workshop according to claim 9, characterized in that: The execution of S4. demand input, S5. intent recognition and slot filling, and S6. knowledge graph multi-step reasoning includes sub-methods: a method for extracting workshop modeling related information based on the combination of voice interaction and knowledge graph retrieval; The S6. knowledge graph multi-step reasoning and S9. fault cause analysis and optimization suggestion execution include sub-methods: a reasoning method based on multimodal workshop modeling knowledge graph; The execution of said S1. data extraction, S2. entity and relationship generation and S3. multimodal knowledge graph construction includes sub-methods: a method for constructing a basic knowledge graph of multimodal workshop simulation; The execution of the comprehensive quality assessment of S13. includes a sub-method: a quality assessment method for a workshop simulation multi-modal system.

11. The method for operating a cigarette thread-making workshop simulation model according to claim 9 or 10, characterized in that: A method for extracting workshop modeling related information based on voice interaction and knowledge graph retrieval, comprising the following steps: S1. Workshop demand input S11. Input form: The cigarette making workshop inputs system requirements through voice or text to express its problems or requirements related to workshop modeling; S12. Speech-to-text: Convert speech input to text using automatic speech recognition (ASR) technology; S13. Text preprocessing: standardize the text, remove noise words, and perform word segmentation and lemmatization; S2. Intent Recognition Algorithm S21. Intent Recognition: Use NLP and intent recognition models (such as BERT or GPT) to parse text and identify core intent and key devices and parameters; S22. Model training: Train the model based on historical data and equipment parameters to ensure accurate identification of the specific needs of the cigarette making workshop; S3. Slot filling mechanism S31. Slot identification: Identify detailed information (equipment type, parameters, and target) in the requirements of the cigarette making workshop; S32. Slot filling: Based on the identified slots, the system fills the detailed content of the cigarette silk workshop requirements into the corresponding slots to form a structured query; S4. Knowledge extraction and query S41. Knowledge graph retrieval: Use structured queries (SPARQL, Cypher) to retrieve relevant information from the knowledge graph; S42. Reasoning based on representation learning: Convert the search results of the knowledge graph into vectors, use graph neural networks for reasoning, and find the most relevant improvement suggestions; S43. Result generation: Generate improvement suggestions based on the reasoning results and feed them back to the cigarette making workshop in natural language; S5. Workshop feedback and result optimization S51. Workshop feedback collection: Collect user feedback on suggestions to support confirmation, rejection or further optimization; S52. Result optimization and learning: Use interactive learning mechanisms to continuously adjust the recommendation model based on feedback, update the knowledge graph and model parameters, and improve the accuracy and practicality of the system's recommendations.

12. The method for operating a cigarette thread-making workshop simulation model according to claim 13 or 14, characterized in that: A method for constructing a basic knowledge graph of multimodal workshop simulation includes the following steps: S1. Extraction of basic knowledge of workshop simulation S11. Data source selection: extract equipment, process flow and tobacco material data from the workshop's manufacturing execution system (MES), enterprise resource planning system (ERP), product lifecycle management system (PLM), etc.; S12. Data extraction and collection: Use ETL tools to extract data, collect raw data through API, database query and file reading, and convert it into a standard format (such as CSV, JSON); S13. Data cleaning and preprocessing: clean up missing values, outliers and duplicate data in the data, segment the text and remove stop words, standardize the image and numerical data to ensure data consistency and accuracy; S2. Basic knowledge entity concepts and relationship generation S21. Entity recognition and entity classification: Use NLP technology to identify entities such as equipment, process flow, tobacco materials, etc. in the text, and classify and annotate attributes according to categories; S22. Relationship definition and relationship generation: Extract the relationship between entities from the data (such as the association between equipment and process), define the relationship type (such as "belongs to", "connects"), and describe the interaction between entities; S23. Structured storage: Store entities and relationships in a relational database or graph database to form a preliminary knowledge graph structure; S3. Constructing a multimodal knowledge graph for workshop simulation S31. Multimodal data integration: Integrate multiple data types (such as text, images, sensor data) and extract features (such as text features, image features); S32. Multimodal feature fusion: Use feature fusion techniques (such as weighted fusion and attention mechanism) to integrate features of different modalities and assign weights to retain key information; S33. Graph construction and optimization: Use graph databases to map multimodal data into knowledge graphs, and improve the structure and application value of graphs through redundancy removal and node optimization. S34. Multimodal knowledge reasoning: Use graph neural networks and other technologies to represent graphs, capture implicit relationships between entities, and support workshop process optimization, equipment fault diagnosis, and intelligent decision-making.

13. The method for operating a simulation model of a cigarette thread-making workshop according to claim 13 or 14, characterized in that: A reasoning method based on multimodal workshop modeling knowledge graph should include the following steps: S1. Knowledge graph multi-step reasoning S11. Input workshop requirements: parse the natural language input of the cigarette silk workshop, use the intent recognition algorithm to convert the requirements into a structured query format, and locate the relevant nodes and relationships; S12. Preliminary query: Use graph query languages ​​(such as SPARQL and Cypher) to query in the multimodal knowledge graph to obtain nodes and relationships related to user needs; S13. Multi-step reasoning: Perform recursive or iterative reasoning on preliminary query results, use graph neural networks (GNNs) and deep learning models (such as RGCN) to process embedding vectors, identify complex relationships, and generate reasoning paths and lists of related nodes; S14. Result aggregation and optimization suggestions: Use rule engines or machine learning models to aggregate reasoning results, extract key related information, find commonalities and differences, and generate optimization suggestions; S2. Causes of failure and optimization suggestions S21. Fault cause identification: Use rule sets and historical case libraries to analyze reasoning results and identify potential fault causes; S22. Generation of optimization suggestions: Retrieve successful solutions through case library and decision tree algorithm, generate optimization suggestions for current problems, and improve equipment operation or process flow; S23. Feedback and verification: Apply the optimization suggestions to the simulation model or actual test, verify the effect and update the knowledge graph, and feed back the verification results to the cigarette making workshop; S24. Cyclic improvement: Collect feedback from the cigarette silk workshop, adjust the inference model through reinforcement learning, continuously update the knowledge graph and optimization suggestions, and enhance the accuracy and practicality of the system.

14. The method for operating a simulation model of a cigarette thread-making workshop according to claim 13 or 14, characterized in that: A quality assessment method for a workshop simulation multimodal system includes the following steps: S1. Accuracy Assessment S11. Model prediction accuracy: Evaluate the prediction accuracy of the model, using indicators such as accuracy, mean square error, etc., test the prediction ability of the model through the validation set to ensure that the error is within an acceptable range; S12. Recommendation accuracy: Compare the recommendation model with the actual usage according to the needs of the cigarette silk workshop, and use indicators such as precision, recall, and F1 value to evaluate the relevance of the recommendation to ensure that the recommendation meets the needs of the cigarette silk workshop; S2. Completeness Assessment S21. Data integrity: Use data audit tools to check the integrity of system data, calculate the coverage of the knowledge graph for all equipment and processes in the workshop, and ensure that there are no missing data; S22. Model completeness: Check the parameters and settings of the recommended model and compare it with the standard model to ensure that all necessary parameters and components are included and the simulation process is fully described; S3: Consistency Assessment S31. Data consistency: Compare similar data from different data sources, use consistency detection tools to check conflicts and duplications, and ensure data consistency within the system; S32. Model output consistency: By running the recommended model multiple times, the variance of the output results is calculated to check the consistency and stability of the output; S4. Practicality evaluation S41. Workshop satisfaction: Collect feedback and satisfaction scores through surveys of cigarette-making workshops to ensure that user satisfaction reaches the set target; S42. Actual application effect: Test the recommended model in the actual workshop, measure the changes in key performance indicators, and verify the improvement effect of the model in actual application; S5. Quality assessment report generation: Generate a comprehensive quality assessment report based on the assessment results, including accuracy, completeness, consistency and practicality indicators, and put forward specific suggestions for system optimization.

15. An electronic device for simulating a cigarette-making workshop model, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: The memory is used for non-temporarily storing computer-readable instructions or readable and writable instructions, and the processor is used for executing the computer-readable instructions or readable and writable instructions; When the computer-readable instructions are executed by the processor, the operations of the cigarette-making workshop simulation model auxiliary system described in any one of claims 1 to 12 and the cigarette-making workshop simulation model operation method and its sub-methods described in any one of claims 13 to 18 are performed.

16. A non-transitory storage medium for a simulation model of a cigarette-making workshop, used for storing computer-readable and writable instructions, computer-readable instructions and mixed-purpose storage, characterized in that: The computer-readable and writable instructions, readable instructions and mixed-purpose storage include instructions for executing the cigarette-making workshop simulation model auxiliary system described in any one of claims 1 to 12 above. When the computer-readable and writable instructions or readable instructions in the non-transitory storage medium are executed by a computer, the operation of the cigarette-making workshop simulation model auxiliary system described in any one of claims 1 to 12 above is realized; The computer-readable and writable instructions, readable instructions and mixed-purpose storage also include an operating method for running a cigarette silk-making workshop simulation model as described in the second purpose, which can realize the operation of the operating method of the cigarette silk-making workshop simulation model and its sub-methods as described in any one of claims 13-18 above.

17. The non-transitory storage medium of the cigarette thread-making workshop simulation model according to claim 11, characterized in that: The non-transitory storage medium is a readable medium, a readable and writable medium, and a mixed-purpose storage; The readable media include magnetic tape storage and read-only optical disks; the readable and writable media include solid-state drives, mechanical hard disks (HDDs), distributed file systems, graph databases, cloud storage services, cache storage, database storage, and object storage; the mixed-use storage includes mechanical hard disks (HDDs) and cloud storage services; The non-temporary storage medium can non-temporarily retain computer-readable instructions so that they can be accessed and executed by the computer when needed; through the computer-readable and writable instructions, readable instructions and mixed-purpose storage stored in the non-temporary storage medium, the computer can perform corresponding operations in accordance with the steps of the operating method and sub-methods of a cigarette silk-making workshop simulation model as described in any one of claims 13-18 above, and support the implementation of the cigarette silk-making workshop simulation model auxiliary system and operating method.

Images