Scheme intelligent reasoning generation method based on process knowledge graph

Abstract

The invention relates to the technical field of process scheme generation, and discloses an intelligent scheme reasoning generation method based on a process knowledge graph, and the method comprises the steps: firstly collecting multi-source process data, constructing a knowledge element extraction system, and outputting standardized knowledge entries through semantic analysis; establishing a process knowledge graph, and when a scheme reasoning demand is detected, positioning a target knowledge node through a logic association algorithm and generating an association relationship label; meanwhile, constructing a historical scheme case information base, and dynamically recording reasoning to generate a record; establishing a multi-dimensional weight configuration model, calculating a scheme matching degree score, and generating a recommendation priority sequence; solving an optimal scheme reasoning result by adopting an improved ant colony algorithm; and finally, outputting a result through the interactive verification platform, tracking and feeding back. The method improves the efficiency and quality of process scheme generation, and is suitable for process scheme formulation in industrial production.

Description

Technical Field

[0001] This invention relates to the field of process scheme generation technology, specifically to a method for intelligent reasoning and generation of schemes based on process knowledge graphs. Background Art

[0002] In modern industrial production, the design and generation of process solutions is a complex and crucial task. With the continuous development of industrial manufacturing, process data is characterized by being multi-source, heterogeneous, and massive, posing numerous challenges to traditional methods of generating process solutions.

[0003] Processing multi-source process data presents challenges. Industrial production generates data from various sources, including process documents, equipment logs, and expert experience. This data is often inconsistent in format and quality, containing a large amount of redundant information and noise. Traditional methods struggle to efficiently collect, preprocess, and extract knowledge from this data, resulting in an inability to accurately obtain process knowledge elements and impacting the accuracy and reliability of subsequent solution generation.

[0004] There is a lack of effective means for representing and managing process knowledge. Process knowledge involves multiple aspects such as industry fields, process types, and technical standards, and contains rich information such as process names, parameter ranges, operating procedures, and entity attributes. Traditional knowledge representation methods are unable to clearly express the logical relationships between processes and cannot build a complete and systematic process knowledge system. This makes it difficult to quickly and accurately locate target knowledge nodes during the solution reasoning process, affecting reasoning efficiency and effectiveness.

[0005] The process of generating solutions lacks scientific decision-making models and optimization algorithms. Traditional methods, when matching and recommending solutions, often fail to comprehensively consider multi-dimensional factors such as process complexity, technological maturity, cost constraints, and environmental adaptability. This makes it difficult to reasonably assign values ​​to various indicators and calculate solution matching scores, resulting in an unscientific and unreasonable priority sequence for generated solutions. Furthermore, when solving for the optimal solution reasoning result, traditional algorithms suffer from low matching coverage, high conflict probability, and slow reasoning process, failing to meet the demands of industrial production for efficient and accurate solution generation. The utilization of historical solution cases is insufficient. Traditional methods lack systematic management and dynamic updating of historical solution case information, failing to effectively utilize historical reasoning generation records to optimize the current solution generation process. This leads to low knowledge reusability and difficulty in continuously improving the quality and efficiency of solution generation.

[0006] The interaction and feedback mechanism between solution generation and practical application is imperfect. Traditional methods lack continuous tracking and feedback on the actual application effects after outputting solution reasoning results, and cannot optimize and improve the solution generation method based on actual application conditions, making it difficult for the solution generation method to adapt to the ever-changing industrial production needs. Summary of the Invention

[0007] The purpose of this invention is to provide a method for intelligent reasoning and generating solutions based on process knowledge graphs, so as to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, this invention provides a method for intelligent reasoning and generating solutions based on process knowledge graphs, the method comprising the following steps: Collect multi-source process data, construct a process knowledge element extraction system, and output standardized knowledge items through a semantic parsing module; A process knowledge graph is established based on knowledge representation technology. When a solution reasoning requirement is detected, the target knowledge node is located and a relationship label is generated through a logical association algorithm. Build a historical solution case information database to dynamically record the reasoning and generation records of different process types, application scenarios, and technical parameters; A multi-dimensional weight configuration model is established. By combining factors such as process complexity, technology maturity, cost constraints, and environmental adaptability, the Delphi method is used to assign values ​​to each indicator. A weighted fusion method is used to construct an inference model to calculate the matching degree score of the scheme, thereby generating a priority sequence of recommended schemes. An improved ant colony algorithm is used to solve the inference results of the optimal solution, with the optimization goal of improving matching coverage, reducing the probability of conflict and accelerating the inference process. The final solution reasoning results are output through the interactive verification platform, and feedback on the actual application effect is continuously tracked.

[0009] Preferably, the process of collecting multi-source process data, constructing a process knowledge element extraction system, and outputting standardized knowledge items through a semantic parsing module specifically includes: Based on industrial-grade data interfaces, multi-source data such as process documents, equipment logs, and expert experience are acquired. The data acquisition terminal receives the raw data from the interface and performs deduplication, noise filtering, and semantic cleaning preprocessing to remove redundant and interfering information. Based on the semantic features and logical relationships of process knowledge, factor analysis was used to screen parameters closely related to knowledge elements as candidate elements. Candidate elements are divided into process parameters, technical indicators, and operating procedures, and a multi-level element system is constructed. The overall knowledge correlation is taken as the top-level element, and the top-level element is decomposed into several first-level elements, and each first-level element is further subdivided into several second-level elements. The semantic parsing module is trained using a recurrent neural network model based on a multi-level element system. By associating process information of technological steps with knowledge elements of preprocessed data, a process-knowledge element mapping relationship is established. Knowledge management software was selected as the semantic analysis platform, and the process steps were marked on the interface according to the actual process flow. Different labels were used for visualization based on the differences in knowledge elements. Steps with large differences in elements were marked with prominent labels, while steps with small differences in elements were marked with conventional labels.

[0010] Preferably, the establishment of the process knowledge graph based on knowledge representation technology specifically includes: Collect basic process data for common industry sectors, process types, and technical standards, including process names, parameter ranges, and standard descriptions of operating procedures; Obtain the entity attribute information of the process, including process category, applicable materials, and equipment requirement parameters; Abstract the process into data nodes, and the logical relationships between processes into data links, and construct a node-link data structure. Establish a graph architecture, including defining the graph database, setting attributes, and establishing links, and storing process information, attribute information, and links; The preprocessed data is imported into the graph database of the knowledge management software to establish a process knowledge graph.

[0011] Preferably, the step of locating the target knowledge node and generating association tags through a logical association algorithm specifically involves: Based on the process knowledge graph, each process node is regarded as a sample point in the dataset, and the logical links between processes are regarded as the correlation between samples. Using the knowledge requirement of the solution to be reasoned as the query point, the nearest neighbor algorithm is used to calculate the association distance to each sample point. When the sample point with the highest association degree is traversed, the feature information of the sample point is recorded. By analyzing and calculating the correlation distance, and combining it with process complexity and technical standard information, the matching range of the target knowledge node is obtained; Extract the process name, parameter range, and operation step description of the target node from the process knowledge graph to generate the association relationship; The relationships between target nodes, along with process types and application scenarios, are combined into a tag, which is then visualized on the knowledge management software interface.

[0012] Preferably, the construction of the historical solution case information database dynamically records the reasoning generation records for different process types, application scenarios, and technical parameters, specifically as follows: The main dimensions for constructing the information database are obtained, including inference and generation records for different process types, application scenarios, and technical parameters. At the same time, specific information fields are planned for each dimension, including industry sector, process type, application scenario, technical standard, actual reasoning scheme, and user evaluation. By connecting the process management system with the user feedback platform, we continuously update the reasoning results, conflict feedback, and implementation effect information of historical solution cases; A historical solution case information database is built based on key dimensions and data fields.

[0013] Preferably, the establishment of a multi-dimensional weight configuration model involves combining factors such as process complexity, technological maturity, cost constraints, and environmental adaptability. The Delphi method is used to assign values ​​to each indicator, and a weighted fusion approach is employed to construct an inference model to calculate the scheme matching score, thereby generating a priority sequence for scheme recommendations. Specifically: Based on process complexity, technology maturity, and cost constraints, the importance of the reasoning is classified, with the demand for high-complexity processes or mature technologies set as high importance, and the demand for low-complexity processes or emerging technologies set as low importance. The process types are divided into discrete manufacturing, process manufacturing, and assembly processes. Discrete manufacturing has a high complexity level, while assembly processes have a low complexity level. Environmental adaptation is divided into standard environment, high temperature environment, and humid environment; Based on the Delphi method, weights are assigned to the importance level, complexity level, and environmental adaptability factors of reasoning; A weighted fusion approach is used to construct the inference model. Each index value is multiplied by its corresponding weight and then summed to obtain the matching score for each inference task. Based on the matching score calculated by the reasoning model, all reasoning tasks are ranked, and tasks with higher scores are recommended with higher priority, thus generating a priority sequence for recommended solutions.

[0014] Preferably, the optimization objective of using an improved ant colony algorithm to solve the optimal solution reasoning result, in order to improve matching coverage, reduce conflict probability, and accelerate the reasoning process, specifically includes: A set of candidate solutions is randomly generated as the initial ant colony; Calculate the fitness value for each candidate solution based on the optimization objective; The optimal solution is selected as the head ant based on the fitness value, and the movement paths and pheromone concentrations of other ants are updated. By simulating the group cooperation and pheromone transmission behavior of ant colonies, candidate solutions are continuously updated iteratively until the termination condition is met. Select a set of optimal solutions from the final ant colony as the result of the optimal solution reasoning.

[0015] Preferably, the matching degree score of the solution is as follows: The matching score is jointly determined by the weights of reasoning importance level, complexity level, environment adaptability, and time factor. The weight of reasoning importance level corresponds to the degree of influence of process complexity and technology maturity; the weight of complexity level corresponds to the degree of influence of process type; the weight of environment adaptability corresponds to the degree of influence of environmental conditions; and the weight of time factor corresponds to the urgency of the solution requirement. The final matching score is obtained by multiplying each weight by the corresponding level score and summing them up.

[0016] Preferably, the optimization objective is specifically: The optimization objectives include improving matching coverage, reducing conflict probability, and accelerating the reasoning process. The matching coverage objective is achieved by calculating the breadth of association between candidate solutions and target knowledge nodes. The conflict probability objective is achieved by limiting the range of contradictions between candidate solutions and existing solutions. The reasoning process objective is achieved by reducing the number of algorithm iterations and computation time. The final optimization objective is the comprehensive optimization of the three sub-objectives.

[0017] Preferably, the step of continuously updating the reasoning results, conflict feedback, and implementation effect information of historical solution cases through data integration with the process management system and user feedback platform specifically involves: Establish data interface specifications and define the data transmission format between the process management system and the user feedback platform, including inference result fields in JSON format and feedback evaluation fields in XML format; Set a scheduled task to synchronize data every day at midnight, and obtain the newly added solution and case data in the previous 24 hours; Perform validity checks on the synchronized data and remove invalid data that is missing process types or does not record the actual reasoning scheme; The verified data is categorized and stored according to industry sector and process type, and the reasoning records in the historical solution case information database are updated.

[0018] Compared with the prior art, the present invention has the following beneficial effects: In the data processing and knowledge extraction stage, by constructing a process knowledge element extraction system, preprocessing such as deduplication, noise filtering, and semantic cleaning is performed on multi-source process data. Combined with factor analysis to screen candidate elements and construct a multi-level element system, and using a recurrent neural network model to complete the training of the semantic parsing module, standardized knowledge items can be extracted from massive multi-source data efficiently and accurately. This effectively solves the problems of difficult data processing and inaccurate knowledge extraction in traditional methods, laying a solid foundation for the subsequent construction of process knowledge graphs and scheme reasoning.

[0019] In terms of process knowledge representation and management, a process knowledge graph is established based on knowledge representation technology. Processes are abstracted into data nodes, and logical relationships are abstracted into data links. A node-link data structure is constructed, and a graph architecture including graph database definitions, attribute settings, and link relationships is established, enabling a clear and systematic representation of process knowledge and its relationships. When a solution reasoning requirement is detected, a logical association algorithm is used to calculate the association distance using the nearest neighbor algorithm. Combined with process complexity and technical standard information, the matching range of the target knowledge node is determined, and association relationship labels are generated and visualized. This achieves rapid and accurate positioning of the target knowledge node, greatly improving the efficiency of knowledge retrieval and utilization, and overcoming the shortcomings of traditional knowledge representation methods.

[0020] For the management of historical solution cases, a historical solution case information database is constructed, clearly defining key dimensions such as process type, application scenario, and technical parameters, as well as information fields such as industry domain and actual reasoning solutions. Through data integration with the process management system and user feedback platform, data is synchronized regularly, and its validity is verified and categorized for storage, achieving dynamic recording and continuous updating of historical solution case information. This ensures the full utilization of historical reasoning and generation records, providing rich reference data for current solution generation and effectively improving knowledge reusability.

[0021] In constructing the solution reasoning and decision-making model, a multi-dimensional weight configuration model is established, comprehensively considering factors such as process complexity, technological maturity, cost constraints, and environmental adaptability. The Delphi method is used to assign values ​​to each indicator, and a weighted fusion approach is employed to construct the reasoning model to calculate the solution matching score, generating a scientifically sound priority sequence for recommended solutions. This model comprehensively considers multiple key factors, avoiding the one-sidedness of traditional decision-making methods, and can provide more reliable decision support for solution generation.

[0022] In finding the optimal solution, an improved ant colony algorithm is adopted, with the optimization goals of increasing matching coverage, reducing conflict probability, and accelerating the inference process. Through iterative processes such as randomly generating the initial ant colony, calculating fitness values, selecting head ants, and updating paths and pheromone concentrations, the algorithm can efficiently find the optimal solution inference result. This algorithm effectively improves matching coverage, reduces conflict probability, and shortens the inference process, meeting the industrial production demand for efficient and accurate solution generation.

[0023] In the interactive verification and feedback optimization phase, the final solution reasoning result is output through the interactive verification platform, and feedback on the actual application effect is continuously tracked. Data interface specifications are established, data is synchronized and verified regularly, and the historical solution case information database is updated. This comprehensive interactive feedback mechanism enables the solution generation method to be continuously optimized and improved based on actual application conditions, enhancing the adaptability of the solution generation method to industrial production needs and ensuring that the generated solutions can be better applied to actual production scenarios, thereby improving production efficiency and quality. Attached Figure Description

[0024] Figure 1 This is a schematic diagram illustrating the working principle of the intelligent reasoning and generation method for solutions based on process knowledge graphs as described in this invention. Figure 2 A workflow diagram for multi-source process data acquisition and standardized knowledge entry generation; Figure 3 A flowchart for target knowledge node localization and tag generation based on a logical association algorithm; Figure 4 A flowchart for building and dynamically recording a database of historical solution cases. Detailed Implementation

[0025] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0026] Please see Figures 1-4 This invention provides a method for intelligent reasoning and generating solutions based on process knowledge graphs, the method comprising: Collect multi-source process data, construct a process knowledge element extraction system, and output standardized knowledge items through a semantic parsing module.

[0027] A process knowledge graph is established based on knowledge representation technology. When a solution reasoning requirement is detected, the target knowledge node is located and a relationship label is generated through a logical association algorithm.

[0028] Build a historical solution case information database to dynamically record the reasoning and generation records of different process types, application scenarios, and technical parameters.

[0029] A multi-dimensional weight configuration model is established. By combining factors such as process complexity, technological maturity, cost constraints, and environmental adaptability, the Delphi method is used to assign values ​​to each indicator. A weighted fusion method is adopted to construct an inference model to calculate the matching degree score of the scheme, thereby generating a priority sequence of recommended schemes.

[0030] An improved ant colony algorithm is used to solve the inference results of the optimal solution, with the optimization goal of improving matching coverage, reducing the probability of conflict and accelerating the inference process.

[0031] The final solution reasoning results are output through the interactive verification platform, and feedback on the actual application effect is continuously tracked.

[0032] Example 1: In the process of collecting multi-source process data, constructing a process knowledge element extraction system, and outputting standardized knowledge entries through a semantic parsing module, the first step is to acquire multi-source data such as process documents, equipment logs, and expert experience based on an industrial-grade data interface. This industrial-grade data interface must meet high reliability and stability requirements, enabling seamless integration with different types of data sources, such as internal production management systems, equipment monitoring systems, and expert knowledge bases. After receiving the raw data from the interface, the data acquisition terminal performs preprocessing operations such as deduplication, noise filtering, and semantic cleaning. Deduplication aims to eliminate duplicate data records and prevent data redundancy from interfering with subsequent processing. A hash-based deduplication algorithm based on data features can be used, performing hash calculations on key data features and comparing hash values ​​to determine if data is duplicated. Noise filtering removes irrelevant noise information from the data, such as abnormal fluctuations generated during equipment operation or interference signals introduced during transmission. Wavelet transform filtering and other methods can be used to smooth the data. Semantic cleaning is the process of cleaning data at the semantic level, correcting semantic errors or ambiguities in the data, and making the data more in line with semantic norms. It can utilize semantic analysis techniques in natural language processing to parse and correct the semantics of the data and remove redundant and interfering information.

[0033] Based on the semantic features and logical relationships of process knowledge, factor analysis is used to screen parameters closely related to knowledge elements as candidate elements. Factor analysis is a dimensionality reduction statistical method that identifies common factors that reflect the essential characteristics of the data by studying the correlation coefficient matrix between variables, thereby selecting parameters that have a significant impact on knowledge elements. When conducting factor analysis, it is first necessary to calculate the correlation coefficient matrix between variables, then determine the number of common factors, and then use methods such as factor rotation to make the common factors more interpretable, ultimately obtaining candidate parameters closely related to knowledge elements.

[0034] The candidate elements are then categorized into process parameters, technical indicators, and operating procedures, constructing a multi-level element system. The overall knowledge relevance is used as the top-level element, which is further decomposed into several first-level elements, and each first-level element is further subdivided into several second-level elements. For example, the top-level element is the overall knowledge relevance; first-level elements may include process parameter relevance, technical indicator relevance, and operating procedure relevance. The second-level elements under each first-level element can be further subdivided according to specific process characteristics; for example, process parameter relevance can be subdivided into temperature parameter relevance, pressure parameter relevance, and time parameter relevance. This multi-level element system more clearly reflects the structure and internal connections of process knowledge, providing a more explicit framework for subsequent semantic analysis.

[0035] Based on a multi-level element system, a recurrent neural network (RNN) model is used to train the semantic parsing module. An RNN is a neural network model capable of processing sequential data. Its internal recurrent structure can capture time-series information and long-term dependencies in the data, making it highly suitable for semantic parsing tasks. During training, elements from the multi-level element system are used as input data. Through training the RNN model, it learns the semantic features and logical relationships of process knowledge, thereby achieving semantic parsing of process data. During training, it is necessary to appropriately set the model parameters, such as the number of hidden layers, the number of neurons, and the learning rate, and to employ suitable loss functions and optimization algorithms to improve the model's training effect and the accuracy of semantic parsing.

[0036] This involves associating process flow information with knowledge elements in preprocessed data to establish a process-knowledge element mapping relationship. Process flow information includes the sequence of processes and operational requirements for each step. Linking this information with knowledge elements clarifies the knowledge elements involved in each process step, providing more accurate knowledge support for subsequent solution reasoning. Establishing the mapping relationship can combine manual annotation with automatic matching. First, process experts annotate some process steps and knowledge elements to establish an initial mapping relationship. Then, machine learning algorithms are used to automatically match the remaining process steps and knowledge elements. Through continuous optimization and adjustment, the accuracy and completeness of the mapping relationship are improved.

[0037] Finally, knowledge management software was selected as the semantic analysis platform. The interface was annotated with each step according to the actual process flow, and different identifiers were used for visualization based on the differences in knowledge elements. The knowledge management software needed to have powerful data management and visualization capabilities, enabling effective storage, management, and display of process data. In terms of interface design, each step was labeled according to the actual process flow sequence, allowing users to clearly understand the entire process. Steps with significant differences in knowledge elements were displayed with eye-catching identifiers, such as using different colors, icons, or font sizes, to attract user attention; steps with minor differences in knowledge elements were displayed with conventional identifiers, making the interface layout clearer and more logical. This visualization method helps users more intuitively understand the distribution and differences of process knowledge elements, improving their understanding and application efficiency of process knowledge.

[0038] Example 2: When building a process knowledge graph based on knowledge representation technology, the first step is to systematically collect basic process data from common industry sectors, process types, and technical standards. This data covers process names, parameter ranges, and standard descriptions of operating procedures. During the collection process, it is crucial to ensure the completeness and accuracy of the data. For example, standard parameter ranges for general processes should be obtained from industry standard databases, and detailed descriptions of specific operating procedures should be extracted from historical process documents of enterprises. For process data from different industry sectors, it is necessary to organize the data according to unified specifications to avoid difficulties in subsequent processing due to inconsistent data formats.

[0039] Obtaining the entity attribute information of the process is also a crucial step. This information includes process category, applicable materials, and equipment requirement parameters. Taking process category as an example, it is necessary to clearly distinguish between the major and minor categories to which different processes belong. For instance, machining processes can be subdivided into turning, milling, grinding, etc. Applicable material information needs to record in detail which types of materials the process is suitable for, including the material grade and performance indicators. Equipment requirement parameters involve the model, precision requirements, and rated power of the processing equipment. This attribute information must be collected from authoritative sources such as equipment manuals and process specifications to ensure data reliability.

[0040] The process is abstracted into data nodes, and the logical relationships between processes are abstracted into data links, constructing a node-link data structure. During abstraction, nodes must be defined based on the essential characteristics of the process, and each node must contain a unique identifier, process name, key attributes, and other information. The determination of logical relationships must be based on the actual production flow or technological dependencies between processes. For example, the output of one process may be the input of another, or the two processes may have technological similarities. These relationships need to be established through analysis and judgment by process experts to ensure that the data structure accurately reflects the inherent connections of process knowledge.

[0041] When establishing a graph architecture, it is necessary to define the graph database, set its attributes, and establish links to store process information, attribute information, and links. The selection of the graph database should consider factors such as data scale and query efficiency; professional graph database management systems such as Neo4j can be used. When defining the database, the node types, relationship types, and their respective attributes must be clearly defined. For example, node types can be divided into process nodes, material nodes, equipment nodes, etc., and relationship types can include "belongs to," "depends on," "similar to," etc. Each relationship type needs to be assigned corresponding weights or attributes to represent the strength or characteristics of the association. Attribute settings must follow database design specifications to ensure data consistency and integrity.

[0042] The preprocessed data is imported into the graph database of the knowledge management software to establish a process knowledge graph. Before importing, the data needs to be formatted and validated to ensure that it meets the input requirements of the graph database. For example, structured process data should be converted into JSON format containing nodes and relationships. During the import process, the progress and status of the data import need to be monitored, and any errors or anomalies that may occur should be handled promptly to ensure the correct construction of the process knowledge graph.

[0043] When a solution reasoning requirement is detected, each process node is treated as a sample point in the dataset based on the process knowledge graph, and the logical links between processes are considered as the correlations between samples. This method of transforming the knowledge graph into a dataset provides a data foundation for subsequent correlation analysis. The characteristics of each sample point are composed of the attributes of the node and the relationships connected to it, which can comprehensively reflect the knowledge characteristics of the process.

[0044] Using the knowledge requirement of the proposed solution as the query point, the nearest neighbor algorithm is used to calculate the association distance to each sample point. When the sample point with the highest association degree is encountered, its feature information is recorded. The core of the nearest neighbor algorithm is defining the distance metric between sample points, which can be Euclidean distance, cosine similarity, etc., and an appropriate distance metric should be selected based on the characteristics of the process knowledge. During the calculation process, all nodes in the knowledge graph need to be traversed, and the association distance between the query point and each node needs to be calculated. By comparison, the node with the smallest distance, i.e., the sample point with the highest association degree, is found. The feature information of this node includes process name, parameter range, operation steps, etc., which will serve as an important basis for subsequent reasoning.

[0045] The calculated association distance is analyzed and combined with information on process complexity and technical standards to determine the matching range of the target knowledge node. Process complexity can be measured by indicators such as the number of process steps and the difficulty of parameter adjustment. Technical standard information includes the industry or enterprise standards followed by the process. When analyzing the association distance, the impact of different process complexities and technical standards on the matching results must be considered. For example, for highly complex processes, the matching range can be appropriately broadened to ensure that enough candidate solutions are found; for processes that strictly adhere to specific technical standards, the matching range should be limited to nodes that conform to those standards. Through this comprehensive analysis, the range of target knowledge nodes can be determined more accurately, improving the efficiency and accuracy of reasoning.

[0046] Extract the process name, parameter range, and operation step description of the target node from the process knowledge graph to generate relationships. The extraction process must traverse and query according to the structure of the knowledge graph to ensure the completeness and accuracy of the extracted information. The generated relationships must clearly define the logical connections between the target node and other nodes, such as the dependency relationship between the target node and raw material nodes, and the adaptation relationship with equipment nodes. These relationships can provide more comprehensive knowledge support for solution reasoning.

[0047] The relationships between target nodes, along with process types and application scenarios, are combined into a tag and visualized on the knowledge management software interface. Tag generation must adhere to certain standards to ensure conciseness and readability, such as using the format "Process Type_Application Scenario_Relationship". The visualization on the knowledge management software interface should employ an intuitive graphical approach, such as using icons of different shapes to represent nodes, lines of different colors to represent relationships, and tags displayed as floating prompts. This allows users to clearly understand the relevant information about the target nodes, assisting them in solution reasoning and decision-making.

[0048] Example 3: When building a historical solution case database that dynamically records the reasoning and generation of different process types, application scenarios, and technical parameters, the first step is to clearly define the main dimensions for constructing the database. These dimensions form the basic framework of the database and mainly include the reasoning and generation of different process types, application scenarios, and technical parameters. The classification of process types should be based on industry standards and the actual production situation of the enterprise. For example, it can be divided into broad categories such as casting, forging, and welding processes, with each category further subdivided into specific process types. Application scenarios should be combined with the actual usage environment of the process, such as high-temperature and high-pressure scenarios in the aerospace field, or precision machining scenarios in the electronics manufacturing field. The technical parameter dimension should cover the key parameters involved in the process, such as temperature, pressure, and speed.

[0049] Simultaneously, specific information fields are planned for each dimension. These information fields are the concrete content carriers of the information database, including industry sectors, process types, application scenarios, technical standards, actual reasoning solutions, and user evaluations. The industry sector needs to clearly define the industry category to which the solution belongs, such as machinery manufacturing, automotive industry, aerospace, etc.; the technical standards need to record the national, industry, or enterprise standards followed by the solution, including the standard number and name; the actual reasoning solution needs to describe in detail the specific process solution generated through reasoning, including process steps, parameter settings, etc.; and the user evaluations collect user feedback on the solution's effectiveness after actual application, including advantages, shortcomings, and suggestions for improvement.

[0050] Establishing data interface specifications is a crucial step in achieving data integration between the process management system and the user feedback platform. This requires defining the data transmission format between the two systems, including JSON-formatted inference result fields and XML-formatted feedback evaluation fields. For JSON-formatted inference result fields, the name, data type, and meaning of each field must be clearly defined. For example, "process_type" represents the process type, and "parameters" represents process parameters, ensuring the standardization and consistency of the data structure. XML-formatted feedback evaluation fields must adhere to XML tag definition rules and establish a reasonable hierarchical structure, such as a root tag of "feedback" with child tags including "user_id," "evaluation," and "suggestion," to facilitate data parsing and processing.

[0051] Set up a scheduled task to synchronize data daily at midnight, retrieving newly added solution case data from the previous 24 hours. This scheduled task can be implemented using the operating system's task scheduler or the scheduling function of an application; for example, in Linux systems, the `crontab` command can be used to set a data synchronization script to run at midnight every day. During data synchronization, a reliable network connection must be established to ensure the stability and security of data transmission. Information such as the time and quantity of data synchronized should be recorded for later querying and management.

[0052] Validating synchronized data is a crucial step in ensuring the quality of the database. Invalid data, such as data lacking process types or failing to record actual reasoning schemes, must be removed. Validation can be implemented using a data validation script. The script first checks if each data entry contains a process type field; if missing, it is marked as invalid. Then, it checks if the actual reasoning scheme field is present; if empty, it is also marked as invalid. For invalid data, its ID and reason must be recorded, and it should be stored or deleted separately to prevent invalid data from contaminating the database.

[0053] The validated data is categorized and stored according to industry sector and process type, updating the inference records in the historical solution case information database. Categorized storage can employ database table partitioning or file system directory structures. For example, a separate table can be created for each industry sector in the database, further partitioned by process type within that table; or multi-level directories can be created in the file system based on industry sector and process type, storing data files in the corresponding directories. When updating inference records, data integrity and consistency must be ensured. New data is directly inserted into the appropriate storage location; for updates to existing data, the corresponding record must be located and modified based on the primary key or unique identifier, recording the update time and operator information.

[0054] In building the information repository, storage capacity and query efficiency must also be considered. As the amount of solution and case data increases over time, the repository needs to be optimized regularly, such as cleaning up expired data and creating indexes. When creating indexes, appropriate fields should be selected based on commonly used query conditions, such as process type, application scenario, and technical parameters, to improve query speed. Simultaneously, a data backup strategy must be established, and the information repository should be backed up regularly to prevent data loss.

[0055] Furthermore, to ensure the scalability of the database, certain extension fields and interfaces must be reserved when designing the database structure. When new process types, application scenarios, or technical parameters emerge, corresponding records can be easily added to the database; when data interaction with other systems is required, data import and export can be achieved through the reserved interfaces.

[0056] When dynamically recording reasoning and generation processes, it is essential to ensure the timeliness and accuracy of the records. Whenever a new solution is generated, relevant information must be promptly entered into the database, including the process knowledge, technical parameters, and generated solution content used in the reasoning process. User feedback must also be collected and organized promptly, and entered into the user evaluation field of the database for subsequent optimization and improvement of the solution.

[0057] Example 4: In establishing a multi-dimensional weighted configuration model, which combines factors such as process complexity, technological maturity, cost constraints, and environmental adaptability, uses the Delphi method to assign values ​​to each indicator, and employs a weighted fusion approach to construct an inference model to calculate the matching degree score of the solutions, thereby generating a priority sequence for recommended solutions, the first step is to classify the importance level of the inference based on process complexity, technological maturity, and cost constraints. For high-complexity processes, such as the precision casting process for aero-engine blades, which involves numerous process steps and strict parameter control requirements, the requirements for such processes should be set to a high importance level; while for low-complexity processes, such as the stamping process for simple parts, which is relatively simple, the requirements can be set to a low importance level. Regarding technological maturity, mature technologies, such as traditional turning technology, are widely used and have stable processes, and their corresponding requirements should be set to a high importance level; emerging technologies, such as new metal printing technology in additive manufacturing, are still in the development stage and their requirements should be set to a low importance level. Cost constraints also need to be comprehensively considered. For high-cost processes, such as processing processes using rare materials, cost control is crucial, and the related requirements should be upgraded in importance; low-cost processes can have their importance level reduced accordingly.

[0058] Manufacturing processes are categorized into three main types: discrete manufacturing, process manufacturing, and assembly manufacturing. Discrete manufacturing, such as the processing of automotive parts, has a high level of reasoning complexity. Each part has its own independent processing flow and technical requirements, involving multiple different processes and equipment. The logical relationships between processes are complex, and numerous factors need to be considered during reasoning, thus resulting in a high level of complexity. Process manufacturing, such as the production of chemical products, involves continuous production processes, but requires high control of process parameters and stability of the process, resulting in a lower level of reasoning complexity. Assembly manufacturing, such as the assembly of electronic products, mainly involves combining various components into a finished product; the processes are relatively fixed, resulting in a low level of complexity.

[0059] Environmental adaptability factors are categorized into standard environments, high-temperature environments, and humid environments. Standard environments refer to conditions such as temperature and humidity within normal ranges, under which most processes can operate normally. High-temperature environments, such as those in the smelting processes of the metallurgical industry, significantly impact process parameters and equipment performance, requiring special consideration. Humid environments, such as manufacturing workshops in coastal areas, can cause equipment corrosion and changes in material properties, affecting process effectiveness. Therefore, weighting different environmental adaptability factors is necessary.

[0060] Based on the Delphi method, weights are assigned to the importance level, complexity level, and environmental adaptability factor of reasoning. The implementation process of the Delphi method is as follows: First, an expert group is formed, consisting of several process experts, technical managers, and cost analysts with rich industry experience. Then, a questionnaire is designed, providing the experts with the definitions and descriptions of the importance level, complexity level, and environmental adaptability factor of reasoning. Experts are required to independently score the weight of each factor based on their professional knowledge and experience. After collecting the experts' scores, statistical analysis is performed to calculate the average and standard deviation of the weights of each factor. The statistical results are fed back to the experts, who then reconsider their scores. If there is a large deviation from the average, the reasons must be explained. The above process is repeated, with multiple rounds of feedback and adjustment, until the experts' opinions tend to be consistent, and the final weight values ​​of each factor are determined. For example, after multiple rounds of Delphi method consultation, the weight of the importance level of reasoning is determined to be 0.4, the weight of the complexity level to be 0.3, the weight of the environmental adaptability factor to be 0.2, and the weight of the time factor to be 0.1 (the time factor corresponds to the urgency of the solution requirements).

[0061] A weighted fusion approach is used to construct the inference model. Each indicator value is multiplied by its corresponding weight and then summed to obtain a matching score for each inference task. Specifically, the scoring criteria for each indicator are first determined. For example, the importance level of inference is rated as high (10 points), medium (7 points), and low (4 points); in terms of complexity, discrete manufacturing is rated as 10 points, process manufacturing as 7 points, and assembly process as 4 points; in terms of environmental adaptability, high-temperature environment is rated as 10 points, humid environment as 7 points, and standard environment as 4 points; and the time factor is categorized into urgent, moderate, and non-urgent, rated as 10 points, 7 points, and 4 points respectively. Then, for each inference task, its score on each indicator is determined, multiplied by its corresponding weight, and finally, the weighted scores of all indicators are summed to obtain the final matching score.

[0062] The matching score is jointly determined by the weights of reasoning importance level, complexity level, environmental adaptability, and time factor. The reasoning importance level weight corresponds to the influence of process complexity and technology maturity; a higher weight indicates a greater impact of process complexity and technology maturity on the matching score. The complexity level weight corresponds to the influence of process type, reflecting the differences in difficulty between different process types during the reasoning process. The environmental adaptability weight corresponds to the influence of environmental conditions, reflecting the applicability requirements of environmental factors to the process solution. The time factor weight corresponds to the urgency of the solution requirement, used to handle reasoning tasks with different time requirements. The final matching score is obtained by multiplying each weight by its corresponding level score and summing the results. This calculation method comprehensively considers multiple dimensions of factors to fully evaluate the matching degree of the solution reasoning task.

[0063] Based on the matching score calculated by the inference model, all solution inference tasks are ranked, with tasks having higher scores receiving higher priority, thus generating a solution recommendation priority sequence. The ranking process employs efficient sorting algorithms, such as quicksort or heapsort, to ensure rapid generation of the priority sequence when handling a large number of solution inference tasks. The generated priority sequence can be visualized in knowledge management software, presented in list or chart form, allowing users to intuitively understand the priority of each solution inference task, thereby allocating resources rationally, prioritizing high-priority tasks, and improving the efficiency and relevance of solution inference generation.

[0064] When constructing a multi-dimensional weighting model, it is also necessary to consider the mutual influence and dynamic adjustment among various factors. For example, as process technology continues to develop and emerging technologies gradually mature, their technology maturity levels may need to be reassessed, and the corresponding weights need to be adjusted accordingly. When a company's cost strategy changes and the emphasis on cost constraints increases, the weight of cost constraint factors in the inference importance level should also increase accordingly. Therefore, it is necessary to establish a regular weight review mechanism to reassess and adjust the weights of each factor annually or quarterly to ensure that the model can adapt to changes in actual conditions.

[0065] Furthermore, to improve the model's accuracy and applicability, historical data on the actual performance of reasoning tasks can be collected to analyze the correlation between matching scores and actual results. If unreasonable weighting of certain factors is found, leading to significant discrepancies between scores and actual performance, the weights can be optimized through expert discussion or by reapplying the Delphi method. Simultaneously, machine learning algorithms, such as regression analysis or neural networks, can be introduced to automatically learn and optimize weights based on historical data, further enhancing model performance.

[0066] The entire process of establishing and applying the multi-dimensional weight configuration model involves scientifically classifying the importance levels of reasoning, reasonably categorizing process types and environmental adaptation factors, determining weights using the Delphi method, constructing a reasoning model using weighted fusion, and calculating the matching degree score. Ultimately, it generates a priority sequence of recommended solutions, achieving a scientific evaluation and priority ranking of solution reasoning tasks. This provides an important decision-making basis for intelligent solution reasoning generation, helps improve the quality and efficiency of solution generation, and meets the requirements of different process needs and application scenarios.

[0067] Example 5: When using the improved ant colony algorithm to solve for the optimal solution inference result, the first step is to clarify the optimization objectives as improving matching coverage, reducing conflict probability, and accelerating the inference process. Specifically, the matching coverage objective is achieved by calculating the breadth of association between candidate solutions and target knowledge nodes. This requires statistically analyzing the number and types of target knowledge nodes involved in the candidate solutions to ensure that the inference result covers more key knowledge elements. The conflict probability objective is achieved by limiting the range of contradictions between candidate solutions and existing solutions. This requires comparing the differences between candidate solutions and historical solutions in terms of process parameters, technical indicators, etc., to avoid significant conflicts between the inference result and existing feasible solutions. The inference process objective is achieved by reducing the number of algorithm iterations and computation time. This requires optimizing algorithm parameters and processes to improve computational efficiency. The ultimate optimization objective is the comprehensive optimization of these three sub-objectives.

[0068] The specific solution process begins with the random generation of a set of candidate solutions as the initial ant colony. The generation of candidate solutions is based on the node and relationship data of the process knowledge graph. Each candidate solution can be represented as a set containing several process nodes and their associated relationships. The selection of nodes and relationships needs to cover the main aspects of the target knowledge requirements. For example, for a machining scheme reasoning task, the initial candidate solutions can randomly select process nodes such as turning and milling, along with their corresponding parameter ranges, operation steps, and other associated relationships, forming the initial candidate set for scheme reasoning. The size of the initial ant colony needs to be determined based on the problem complexity and computational resources, typically set to 30 to 100 candidate solutions to achieve a balance between computational efficiency and search capability.

[0069] Next, the fitness value of each candidate solution is calculated based on the optimization objectives. The design of the fitness function is crucial, requiring comprehensive consideration of three sub-objectives: matching coverage, conflict probability, and inference process. Matching coverage can be defined as the proportion of valid nodes in a candidate solution associated with the target knowledge nodes to the total number of target nodes. Conflict probability can be measured by calculating the number and severity of conflict parameters between the candidate solution and historical solutions; fewer conflict parameters and lower severity result in a lower conflict probability. Inference process can be referenced by the number of iterations already performed and the current computation time; fewer iterations and shorter computation time indicate a better inference process. These three sub-objectives are then weighted and summed according to preset weights (e.g., matching coverage weight 0.4, conflict probability weight 0.3, and inference process weight 0.3) to obtain the fitness value of each candidate solution.

[0070] The optimal solution is selected as the leader ant based on its fitness value, and the movement paths and pheromone concentrations of other ants are updated. The leader ant selection must ensure its fitness value is the highest in the current ant colony, representing the current optimal solution for inference. The pheromone concentration update rule must reflect a positive feedback mechanism, i.e., the pheromone concentration on the path of a candidate solution with high fitness increases more, guiding other ants to search towards a better solution. The specific update formula is: Pheromones Concentration = Original Concentration + Volatilization Coefficient × Fitness Value Contribution, where the volatile coefficient is used to simulate the natural evaporation of pheromones and is typically set to 0.1 to 0.3 to avoid the algorithm getting trapped in local optima too early.

[0071] By simulating the collective collaboration and pheromone transmission behavior of ant colonies, candidate solutions are iteratively updated until a termination condition is met. During the iteration process, each ant selects the next process node to visit based on pheromone concentration and heuristic information (such as the correlation between the target knowledge node and the target knowledge node), constructing a new candidate solution. Heuristic information can be determined by calculating the correlation distance between the current node and the target knowledge node; the smaller the correlation distance, the greater the heuristic information, and the higher the probability that the ant will choose that node. After each iteration, the fitness values ​​of all candidate solutions are recalculated, and the head ant and pheromone concentration are updated. This process is repeated until a preset termination condition is met, such as the number of iterations reaching a maximum value (usually 200 to 500 times), the fitness value no longer significantly improving, or the computation time exceeding a threshold.

[0072] A set of optimal solutions is selected from the final ant colony as the optimal solution inference result. The candidate solutions in the final ant colony have undergone multiple rounds of iterative optimization, achieving a good balance between matching coverage, conflict probability, and inference process. The set of solutions with the highest fitness value is selected as the final result. This set of solutions must contain complete information such as process nodes, parameter ranges, operation steps, and correlations, meeting the specific requirements of solution inference. For example, for a certain aerospace part machining solution inference task, the final optimal solution might include five-axis simultaneous milling process nodes, specific cutting parameter ranges, detailed clamping steps, and correlations with material heat treatment processes.

[0073] In improving the implementation of the ant colony algorithm, it is also necessary to reasonably set and adjust the key parameters of the algorithm. For example, the magnitude of the pheromone evaporation coefficient affects the convergence speed and global search capability of the algorithm. Too small an evaporation coefficient may cause the algorithm to get stuck in a local optimum, while too large an evaporation coefficient will reduce the convergence speed. The weight factor of heuristic information needs to be adjusted according to the characteristics of the process knowledge graph and the reasoning requirements of the solution. For problems with high knowledge correlation, the weight of heuristic information can be appropriately increased. The number of ants also affects the algorithm performance. Too few ants may result in a limited search range and failure to find the optimal solution, while too many ants will increase computational complexity and prolong the reasoning time. Therefore, before applying the algorithm, these parameters need to be optimized through preliminary experiments to determine the parameter combination suitable for the specific process domain and problem scale.

[0074] Furthermore, to improve the algorithm's search efficiency and avoid getting trapped in local optima, various improvement strategies can be introduced. For example, an elitist strategy can be used to increase the pheromone concentration of historically optimal solutions, strengthening their guiding role; maximum and minimum pheromone limits can be imposed to control the pheromone concentration within a certain range, preventing excessively high or low pheromone concentrations on certain paths; and a local search mechanism can be used to locally optimize candidate solutions after the ants have constructed them, such as adjusting the order of process nodes or optimizing parameter ranges, to improve the quality of the solutions. The combined use of these improvement strategies can effectively enhance the solution-solving ability of the improved ant colony algorithm in solution reasoning and generation.

[0075] The interactive verification platform outputs the final solution reasoning results and continuously tracks feedback on actual application effects. The platform must have a user-friendly interface, capable of visually displaying the optimal solution's process nodes, relationships, and key parameters for easy user understanding and verification. Output includes a detailed description of the solution, its matching with target knowledge requirements, and conflict analysis with historical solutions. Simultaneously, the platform must establish a feedback collection mechanism to record problems encountered by users during practical application and their evaluations of the solution's effectiveness. This feedback will serve as crucial evidence for subsequent algorithm optimization and knowledge graph updates. For example, if a user reports that a solution's processing quality fails to meet standards due to unreasonable parameter settings in practical application, this feedback can be entered into the system for adjusting the algorithm's fitness function or updating the historical solution case database.

[0076] The entire implementation process of the improved ant colony algorithm, through clearly defining optimization objectives, scientifically designing the fitness function, rationally setting algorithm parameters, introducing improvement strategies, and combining with an interactive verification platform, achieved efficient solving of the optimal solution reasoning results. It can quickly find solutions that meet multi-objective optimization requirements within complex process knowledge graphs, providing powerful algorithmic support for intelligent reasoning and generation of process solutions. Furthermore, by continuously tracking feedback from actual application effects, the algorithm and knowledge graph can be continuously optimized to improve the accuracy and applicability of solution reasoning, meeting the changing needs of different process scenarios and applications.

[0077] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0078] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent reasoning and generating solutions based on process knowledge graphs, characterized in that, Includes the following steps: Collect multi-source process data, construct a process knowledge element extraction system, and output standardized knowledge items through a semantic parsing module; A process knowledge graph is established based on knowledge representation technology. When a solution reasoning requirement is detected, the target knowledge node is located and a relationship label is generated through a logical association algorithm. Build a historical solution case information database to dynamically record the reasoning and generation records of different process types, application scenarios, and technical parameters; A multi-dimensional weight configuration model is established. By combining factors such as process complexity, technology maturity, cost constraints, and environmental adaptability, the Delphi method is used to assign values ​​to each indicator. A weighted fusion method is used to construct an inference model to calculate the matching degree score of the scheme, thereby generating a priority sequence of recommended schemes. An improved ant colony algorithm is used to solve the inference results of the optimal solution, with the optimization goal of improving matching coverage, reducing the probability of conflict and accelerating the inference process. The final solution reasoning results are output through the interactive verification platform, and feedback on the actual application effect is continuously tracked.

2. The intelligent reasoning and generation method for schemes based on process knowledge graphs according to claim 1, characterized in that, The process involves collecting multi-source process data, constructing a process knowledge element extraction system, and outputting standardized knowledge items through a semantic parsing module. Specifically: Based on industrial-grade data interfaces, multi-source data such as process documents, equipment logs, and expert experience are acquired. The data acquisition terminal receives the raw data from the interface and performs deduplication, noise filtering, and semantic cleaning preprocessing to remove redundant and interfering information. Based on the semantic features and logical relationships of process knowledge, factor analysis was used to screen parameters closely related to knowledge elements as candidate elements. Candidate elements are divided into process parameters, technical indicators, and operating procedures, and a multi-level element system is constructed. The overall knowledge correlation is taken as the top-level element, and the top-level element is decomposed into several first-level elements, and each first-level element is further subdivided into several second-level elements. The semantic parsing module is trained using a recurrent neural network model based on a multi-level element system. By associating process information of technological steps with knowledge elements of preprocessed data, a process-knowledge element mapping relationship is established. Knowledge management software was selected as the semantic analysis platform, and the process steps were marked on the interface according to the actual process flow. Different labels were used for visualization based on the differences in knowledge elements. Steps with large differences in elements were marked with prominent labels, while steps with small differences in elements were marked with conventional labels.

3. The intelligent reasoning and generation method for schemes based on process knowledge graphs according to claim 2, characterized in that, The process knowledge graph established based on knowledge representation technology specifically includes: Collect basic process data for common industry sectors, process types, and technical standards, including process names, parameter ranges, and standard descriptions of operating procedures; Obtain the entity attribute information of the process, including process category, applicable materials, and equipment requirement parameters; Abstract the process into data nodes, and the logical relationships between processes into data links, and construct a node-link data structure. Establish a graph architecture, including defining the graph database, setting attributes, and establishing links, and storing process information, attribute information, and links; The preprocessed data is imported into the graph database of the knowledge management software to establish a process knowledge graph.

4. The intelligent reasoning and generation method for schemes based on process knowledge graphs according to claim 3, characterized in that, The process of locating target knowledge nodes and generating association tags using a logical association algorithm specifically involves: Based on the process knowledge graph, each process node is regarded as a sample point in the dataset, and the logical links between processes are regarded as the correlation between samples. Using the knowledge requirement of the solution to be reasoned as the query point, the nearest neighbor algorithm is used to calculate the association distance to each sample point. When the sample point with the highest association degree is traversed, the feature information of the sample point is recorded. By analyzing and calculating the correlation distance, and combining it with process complexity and technical standard information, the matching range of the target knowledge node is obtained; Extract the process name, parameter range, and operation step description of the target node from the process knowledge graph to generate the association relationship; The relationships between target nodes, along with process types and application scenarios, are combined into a tag, which is then visualized on the knowledge management software interface.

5. The intelligent reasoning and generation method for schemes based on process knowledge graphs according to claim 4, characterized in that, The aforementioned construction of a historical solution case information database dynamically records the reasoning and generation records for different process types, application scenarios, and technical parameters, specifically as follows: The main dimensions for constructing the information database are obtained, including inference and generation records for different process types, application scenarios, and technical parameters. At the same time, specific information fields are planned for each dimension, including industry sector, process type, application scenario, technical standard, actual reasoning scheme, and user evaluation. By connecting the process management system with the user feedback platform, we continuously update the reasoning results, conflict feedback, and implementation effect information of historical solution cases; A historical solution case information database is built based on key dimensions and data fields.

6. The intelligent reasoning and generation method for schemes based on process knowledge graphs according to claim 5, characterized in that, The establishment of a multi-dimensional weight configuration model involves combining factors such as process complexity, technological maturity, cost constraints, and environmental adaptability. The Delphi method is used to assign values ​​to each indicator, and a weighted fusion approach is employed to construct an inference model that calculates the scheme matching score, thereby generating a priority sequence for scheme recommendations. Specifically: Based on process complexity, technology maturity, and cost constraints, the importance of the reasoning is classified, with the demand for high-complexity processes or mature technologies set as high importance, and the demand for low-complexity processes or emerging technologies set as low importance. The process types are divided into discrete manufacturing, process manufacturing, and assembly processes. Discrete manufacturing has a high complexity level, while assembly processes have a low complexity level. Environmental adaptation is divided into standard environment, high temperature environment, and humid environment; Based on the Delphi method, weights are assigned to the importance level, complexity level, and environmental adaptability factors of reasoning; A weighted fusion approach is used to construct the inference model. Each index value is multiplied by its corresponding weight and then summed to obtain the matching score for each inference task. Based on the matching score calculated by the reasoning model, all reasoning tasks are ranked, and tasks with higher scores are recommended with higher priority, thus generating a priority sequence for recommended solutions.

7. The intelligent reasoning and generation method for schemes based on process knowledge graphs according to claim 6, characterized in that, The improved ant colony algorithm is used to solve the optimal solution reasoning result. The optimization objectives are set in the direction of improving matching coverage, reducing conflict probability, and accelerating the reasoning process. Specifically: A set of candidate solutions is randomly generated as the initial ant colony; Calculate the fitness value for each candidate solution based on the optimization objective; The optimal solution is selected as the head ant based on the fitness value, and the movement paths and pheromone concentrations of other ants are updated. By simulating the group cooperation and pheromone transmission behavior of ant colonies, candidate solutions are continuously updated iteratively until the termination condition is met. Select a set of optimal solutions from the final ant colony as the result of the optimal solution reasoning.

8. The intelligent reasoning and generation method for schemes based on process knowledge graphs according to claim 7, characterized in that, The matching score of the proposed solution is as follows: The matching score is jointly determined by the weights of reasoning importance level, complexity level, environment adaptability, and time factor. The weight of reasoning importance level corresponds to the degree of influence of process complexity and technology maturity; the weight of complexity level corresponds to the degree of influence of process type; the weight of environment adaptability corresponds to the degree of influence of environmental conditions; and the weight of time factor corresponds to the urgency of the solution requirement. The final matching score is obtained by multiplying each weight by the corresponding level score and summing them up.

9. The intelligent reasoning and generation method for schemes based on process knowledge graphs according to claim 8, characterized in that, The optimization objective is specifically as follows: The optimization objectives include improving matching coverage, reducing conflict probability, and accelerating the reasoning process. The matching coverage objective is achieved by calculating the breadth of association between candidate solutions and target knowledge nodes. The conflict probability objective is achieved by limiting the range of contradictions between candidate solutions and existing solutions. The reasoning process objective is achieved by reducing the number of algorithm iterations and computation time. The final optimization objective is the comprehensive optimization of the three sub-objectives.

10. The intelligent reasoning and generation method for schemes based on process knowledge graphs according to claim 5, characterized in that, The process management system connects with the user feedback platform to continuously update the reasoning results, conflict feedback, and implementation effect information of historical solution cases. Specifically: Establish data interface specifications and define the data transmission format between the process management system and the user feedback platform, including inference result fields in JSON format and feedback evaluation fields in XML format; Set a scheduled task to synchronize data every day at midnight, and obtain the newly added solution and case data in the previous 24 hours; Perform validity checks on the synchronized data and remove invalid data that is missing process types or does not record the actual reasoning scheme; The verified data is categorized and stored according to industry sector and process type, and the reasoning records in the historical solution case information database are updated.

Images