An intelligent production scheduling and exception management method based on large language models

Abstract

The present application provides an intelligent production scheduling and exception management method based on large language models, which relates to the field of artificial intelligence technology and includes: converting user requests into structured data through a semantic parser and complementing fuzzy information in combination with the context; modeling production data using an attention-based relational graph neural network and retrieving relevant context information in combination with a retrieval-augmented generation framework; adjusting the scheduling strategy in real time according to user feedback and changes in the production environment; and generating scheduling instructions using a large language model in combination with the structured data and the scheduling strategy. Through dynamic knowledge base construction, hierarchical memory mechanism design, attention-based relational graph neural network modeling, and the semantic understanding ability of large language models, the intelligence level, logic, and adaptability of the production scheduling system are significantly improved.

Description

Technical Field

[0001] This application belongs to the field of artificial intelligence technology, and particularly relates to an intelligent production scheduling and anomaly management method based on large language models. Background Art

[0002] With the rapid development of intelligent manufacturing technology, the complexity of production tasks has been increasing day by day. How to achieve efficient production scheduling has become a key issue in the field of intelligent manufacturing. Traditional production scheduling methods usually rely on static rules and predefined processes. Although they perform well in structured and less-changing environments, they often show obvious limitations when faced with dynamic and changeable production scenarios. These methods are difficult to quickly adapt to sudden changes in production tasks, such as equipment failures, process adjustments, or temporarily inserted emergency tasks, resulting in reduced production efficiency and resource waste.

[0003] In recent years, the application of artificial intelligence technology, especially natural language processing (NLP) and deep learning methods based on large models, in production scheduling has gradually increased. These methods can automatically generate scheduling plans by analyzing massive data, improving scheduling efficiency. However, the current intelligent scheduling systems still face the following main problems:

[0004] In complex production environments, there are high dependencies and sequential logics between different tasks. However, existing scheduling systems often have difficulty capturing these relationships, resulting in task conflicts or logical chaos in the generated scheduling plans. Production tasks usually need to refer to multiple rounds of instructions and historical data, and existing methods have relatively limited handling of context, especially it is difficult to capture key information hidden in the context. For example, key conditions or restrictions may have been clearly stated in early communication of production instructions, but due to the way of expression becoming gradually blurred or colloquial in subsequent communication, the system ignores these key information, thus misleading the generated scheduling plan. Existing scheduling systems are difficult to respond to changes in the production environment in real time, such as task priority adjustments, equipment status changes, or sudden anomalies, lacking flexibility and dynamic adjustment capabilities. Production data is usually scattered in multiple heterogeneous systems, with complex data structures and insufficient semantic relevance. Existing methods are difficult to effectively integrate and manage these data, resulting in limited scheduling accuracy. Summary of the Invention

[0005] This application provides an intelligent production scheduling and anomaly management method based on large language models to solve one of the above technical problems.

[0006] The technical solution adopted in this application is as follows:

[0007] An embodiment of this application provides an intelligent production scheduling and anomaly management method based on large language models, including:

[0008] Convert the user request into structured data through a semantic parser, and complete the fuzzy information in combination with the context;

[0009] Adopt an attention-based relational graph neural network to model the production data, and retrieve relevant context information in combination with a retrieval-augmented generation framework;

[0010] Adjust the scheduling strategy in real time according to user feedback and production environment changes;

[0011] Combine the structured data and the scheduling strategy, and use a large language model to generate a scheduling instruction.

[0012] According to an embodiment of the present application, the context information is stored and managed through a three-layer structure of a global planning layer, a process logic layer, and an operation step layer.

[0013] According to an embodiment of the present application, the conversion of the user request into structured data through a semantic parser and the completion of fuzzy information in combination with the context are specifically as follows:

[0014] Extract key parameters through a semantic parser, and complete the missing conditions in the user request in combination with the historical context information in the hierarchical memory mechanism;

[0015] If the user request is insufficient, guide the user to supplement details through a prompt mechanism;

[0016] Output the supplemented key parameters as the structured data.

[0017] According to an embodiment of the present application, the modeling of the production data by adopting an attention-based relational graph neural network and the retrieval of relevant context information in combination with a retrieval-augmented generation framework are specifically as follows:

[0018] Integrate the production data from different sources into a knowledge graph, where the nodes represent production entities and the edges represent the relationships between entities;

[0019] Use a vectorized storage model to match the user request with the nodes in the knowledge graph;

[0020] The attention-based relational graph neural network captures the implicit associations and key dependencies between tasks.

[0021] According to an embodiment of the present application, the real-time adjustment of the scheduling strategy according to user feedback and production environment changes is specifically as follows:

[0022] Construct a reward signal through the user feedback and the production environment changes to guide the optimization of the scheduling strategy;

[0023] Use an algorithm based on policy gradients to dynamically adjust the scheduling policy parameters according to real-time data to maximize the cumulative reward.

[0024] According to an embodiment of the present application, when combining the structured data and the scheduling policy, a large language model is used to generate scheduling instructions, specifically:

[0025] Use the large language model to combine the structured data and the scheduling policy to generate the scheduling instructions;

[0026] Automatically adjust the instruction content according to changes in user requirements or sudden environmental situations.

[0027] According to an embodiment of the present application, it further includes:

[0028] Real-time monitor the production status, quickly identify and classify anomalies, and dynamically adjust the scheduling plan.

[0029] According to an embodiment of the present application, the real-time monitoring of the production status, quickly identifying and classifying anomalies, and dynamically adjusting the scheduling plan are specifically:

[0030] Track the status of production equipment, task execution, and environmental parameters;

[0031] Identify and classify anomalies in the production process, and take corresponding handling measures according to the severity;

[0032] Automatically generate a response plan, dynamically adjust task scheduling, and minimize the impact of anomalies on production.

[0033] A computer-readable storage medium, on which a program is stored, and when the program is executed by a processor, the steps in the described method are implemented.

[0034] An electronic device, including a memory, a processor, and a program stored on the memory and executable on the processor, and when the processor executes the program, the steps in the described method are implemented.

[0035] Due to the adoption of the above technical solutions, the beneficial effects obtained by the present application are:

[0036] Through the construction of a dynamic knowledge base, the design of a hierarchical memory mechanism, the modeling of a relational graph neural network (R-GAT) based on attention, and the semantic understanding ability of a large language model (LLM), the present application significantly improves the intelligence level, logic, and adaptability of the production scheduling system.

[0037] Through the hierarchical memory mechanism of hierarchical storage, the production tasks are divided into three levels: global planning, process logic, and operation steps, ensuring that the sequential relationship, dependency structure, and execution conditions of the scheduling tasks are clearly controllable, and avoiding problems such as task conflicts or logical confusion.

[0038] Combined with the context understanding ability of the large language model, it dynamically manages the multi-round interaction of production tasks, stores context information hierarchically through a hierarchical memory mechanism, captures the logical conditions and background information implicit in the early instructions, and avoids scheduling errors caused by the loss of context information.

[0039] The reinforcement learning optimization module collects user feedback and production environment change information in real time, dynamically adjusts task priorities and scheduling strategies, enabling the system to flexibly handle emergencies such as equipment failures and insertion of urgent tasks, thereby enhancing the adaptability and efficiency of the production system.

[0040] The dynamic knowledge base combines the attention-based relational graph neural network (R-GAT) to deeply model the knowledge graph, capturing the implicit associations and key dependencies between tasks. Through the retrieval-augmented generation (RAG) framework, the dynamic knowledge base supports real-time updates and efficient retrieval, providing accurate and reliable context data for complex scheduling tasks.

[0041] Utilizing the semantic generation ability of the large language model, combined with the dynamic knowledge base and the results of reinforcement learning optimization, it dynamically generates production instructions with rigorous logic and detailed content. It supports semantic completion of fuzzy inputs and priority adjustment of emergency tasks, ensuring that the generated instructions can be directly applied to actual production scenarios.

[0042] By real-time monitoring of equipment status, task progress, and environmental parameters, combined with anomaly detection algorithms to quickly identify and classify problems, and adjusting the scheduling plan through the hierarchical memory mechanism and the dynamic knowledge base, it ensures the coherence of production tasks and minimizes the impact of anomalies. Brief Description of the Drawings

[0043] The drawings described herein are used to provide a further understanding of the present application and form a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation of the present application. In the drawings:

[0044] Figure 1 It is a schematic flow chart of an intelligent production scheduling and anomaly management method based on a large language model provided by an embodiment of the present application. Detailed Embodiments

[0045] To more clearly illustrate the overall concept of the present application, the following will be described in detail by way of examples in conjunction with the drawings of the specification.

[0046] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application. However, the present application may be implemented in other ways different from those described herein. Therefore, the scope of protection of the present application is not limited by the specific embodiments disclosed below. It should be noted that, without conflict, the embodiments of the present application and the features in each embodiment may be combined with each other.

[0047] In the present application, unless otherwise clearly specified and limited, the first feature being "on" or "under" the second feature may mean that the first and second features are in direct contact, or the first and second features are indirectly in contact through an intermediate medium. In the description of this specification, the description with reference to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0048] Embodiment 1

[0049] As Figure 1 shown, an intelligent production scheduling and anomaly management method based on a large language model includes:

[0050] Converting a user request into structured data through a semantic parser and complementing fuzzy information in combination with the context.

[0051] Specifically, first, various forms of input from users are received, including but not limited to text, voice commands, and direct scheduling instructions, etc. These inputs may contain content regarding scheduling tasks (e.g., "Arrange the maintenance of device A"), status queries (e.g., "The current status of device B"), or process adjustments (such as "Optimize the production line process"), etc. Through semantic parsing technology, key elements are extracted from the user input, including task types (such as repair, inspection, production, etc.), parameters (such as device name, time requirement, priority, etc.), and the specific intention of the request is determined. Considering the integrity of complex instructions, a hierarchical memory mechanism is utilized to extract background information from the previous rounds of interactions to supplement key conditions or dependencies that may be omitted in the user's current input. For example, in a multi-round conversation, the prerequisite conditions or limitations of certain operations may have been clarified in the early exchanges, but these details are not explicitly mentioned in subsequent instructions. At this time, the context completion mechanism will come into play to ensure that all necessary background information is taken into account, avoiding inaccurate instructions or logical confusion caused by information missing. After the above steps, the user's natural language request is transformed into a structured instruction format, such as "Task type: Device inspection; Device name: Device A; Priority: High; Time: Immediately". This structured representation facilitates further processing and use by subsequent modules.

[0052] After the user submits a request, the semantic parser first conducts a preliminary analysis of the natural language request, attempting to extract key information such as the task type, parameter conditions, and priority. If it is found that some necessary information is missing or the expression is not clear enough (for example, the device name, time requirement, etc. are missing), then it enters the next step. Based on preset rules or a knowledge base of historical data, it is determined which information is necessary but not provided. For example, when arranging a maintenance task, if the specific device name is not specified, this is considered an information gap. For the identified information gap, the system automatically generates targeted prompts or questions to guide the user to provide the required specific information. These prompts can be in the form of direct questions, such as "Please specify the specific device name that needs to be maintained", or more detailed guiding instructions. The prompts can also be based on the background information provided by the context completion mechanism. For example, if a certain device has been previously mentioned as possibly having problems, and the current request seems related to it but does not clearly indicate, the prompt may include something like "Do you mean device A mentioned last time?" After the user receives the prompt, they can fill in these information blanks through further input. This may include providing the missing specific parameter values, clarifying the intention, or confirming previous assumptions. The system will parse the user's new input again and repeat the above process until all necessary information is accurately obtained. Finally, the system will conduct a comprehensive check on all the information collected to ensure that there are no omissions and that all the information provided is clear and consistent with each other. If there are still doubts or conflict points, the system will continue to request the user to clarify through the feedback mechanism. Once it is confirmed that all relevant information is complete and accurate, the system will convert this information into a structured instruction format for subsequent task execution or other operations.

[0053] The process of converting the user request into structured data through the semantic parser and complementing fuzzy information in combination with the context is mainly to ensure that the system can accurately understand the user's intention and generate precise and executable instructions based on the complete context information. This process not only improves the accuracy of the instructions but also enhances the system's ability to cope with changes in complex production environments.

[0054] For example, assume that a dispatcher in a manufacturing enterprise hopes to arrange a device maintenance task, but due to being busy or negligent, not all necessary detailed information is provided in the initial request. The following is a detailed expansion of this process:

[0055] Initial user request: "Arrange device maintenance."

[0056] The system first attempts to extract the key elements from this sentence. In this example, "Task type: Device maintenance" is identified, but important information such as the specific device name to be maintained, priority, and the expected completion time is missing.

[0057] Given the incomplete information, the system uses a hierarchical memory mechanism to search for relevant background information. For example, if a potential problem with a specific device (such as Device A) was discussed in a previous interaction, the system might infer that the current request could be related to this device.

[0058] Meanwhile, the system might also refer to recent task records or known production plans to deduce which device is most likely in need of maintenance.

[0059] The system automatically generates targeted questions or suggestions to supplement the missing information. "Do you mean to perform maintenance on Device A? If so, please confirm the priority and the desired completion time."

[0060] Suppose the user replies, "Yes, it's the maintenance of Device A, with high priority and preferably completed today."

[0061] Based on the user's supplementary information, the system transforms the original request into a detailed structured instruction: "Task type: Equipment inspection; Equipment name: Device A; Priority: High; Time requirement: To be completed today."

[0062] The system conducts a comprehensive check on this structured instruction to ensure that all necessary information is included and consistent. If everything is normal, this instruction will be passed to the dynamic knowledge base module for further retrieval of relevant context information to support subsequent task execution.

[0063] Through the above steps, even if the user's initial request is very brief or ambiguous, the intelligent production scheduling system can still accurately capture the user's true intention through effective semantic parsing and context completion mechanisms, and generate precise and executable instructions. This approach not only improves the flexibility and adaptability of the system but also greatly enhances the user experience.

[0064] Furthermore, in addition to the basic semantic parser, more advanced natural language processing (NLP) technologies, such as large language models (LLMs) based on the Transformer architecture, can be introduced to improve the ability to understand complex sentence structures and contexts. Utilize external knowledge bases or industry-specific knowledge graphs to enrich the system's context understanding ability, enabling the system to not only understand information in the current domain but also understand and apply relevant knowledge across domains.

[0065] By analyzing the user's historical behavior and preferences, automatically adjust the instruction generation strategy to provide more personalized services. For example, for users who often focus on efficiency, the system can preferentially recommend task arrangements for quick execution. Based on the current production environment status, historical data, and predictive analysis, provide the user with optimal operation suggestions or alternative solutions to help the user make better decisions.

[0066] Model the production data using an attention-based relational graph neural network and retrieve relevant context information in combination with a retrieval-augmented generation framework.

[0067] Specifically, use an embedding model (such as Transformer) to convert the nodes and relationships of the knowledge graph into vector representations, enabling the user's request to quickly locate relevant nodes and context information through vector matching.

[0068] The formulas used in this process include a query embedding model , and an embedding model that converts nodes or documents in the knowledge base into vectors . Then use cosine similarity to measure the relevance between the user's query and the content in the knowledge base.

[0069] In the retrieval stage, the RAG framework selects the k nodes most relevant to the user's request to form a context supplementary dataset.

[0070] In the generation stage, combine the context information and the user input to generate production scheduling instructions that better meet the actual scenario requirements. This method effectively avoids the hallucination problem in traditional generation methods and improves the accuracy and relevance of the generated content.

[0071] The dynamic knowledge base works closely with R-GAT to form a complete closed loop from data organization to semantic modeling. The dynamic knowledge base ensures that the knowledge graph can accurately reflect the latest state in the production scenario by updating node attributes and relationship weights in real time; while R-GAT performs in-depth modeling based on the updated graph to capture implicit associations and task dependencies, thereby enhancing the semantic expression ability of the data.

[0072] The retrieved results and node embeddings are directly passed to the hierarchical memory mechanism as the basic input for context management, enhancing the coherence of task logic and improving the system's adaptability to complex production tasks.

[0073] The combination of the attention-based relational graph neural network (R-GAT) and the retrieval-augmented generation (RAG) framework enables the system to not only capture deep semantic associations when processing complex production data, but also efficiently retrieve and generate relevant context information according to the specific needs of the user, greatly improving the intelligence level and flexibility of production scheduling.

[0074] For example, assume that in a manufacturing plant, a dispatcher needs to arrange a new production line task, and this task depends on the status of other equipment and some previous process steps. Due to the complex production environment, involving multiple equipment, processes, and their interrelationships, an intelligent system is needed to help the dispatcher make optimal decisions.

[0075] First, the dynamic knowledge base integrates heterogeneous data from different sources into a production-based knowledge graph. For example: Nodes: represent different production entities, such as Equipment A, Equipment B, Process Step X, etc. Edges: represent the relationships between these entities, such as Equipment A must start working after Equipment B completes a certain operation; Process Step X is a key link in the production process, etc.

[0076] Next, use R-GAT to perform in-depth modeling on this knowledge graph:

[0077] Calculate the attention weights between each node and its neighbor nodes to highlight the connections crucial for the current task. For example, when arranging a new task, if the status of Equipment A directly affects the smooth progress of the new task, then a higher attention weight is assigned to Equipment A. Based on the results of the attention weights, R-GAT will perform multiple embedding updates on each node, considering not only the information of local neighbors but also reflecting the relevance between global tasks. For example, considering that the status of Equipment A may affect the efficiency of the entire production chain, the system will comprehensively analyze the status of all relevant nodes to ensure the optimization of the overall scheduling plan.

[0078] When the user submits a new request, such as "arrange the best task sequence for a new production line", the system will perform the following steps:

[0079] Use an embedding model (such as Transformer) to convert the user's request and the nodes in the knowledge graph into vector representations. Then, find the nodes and context information most relevant to the user's query through cosine similarity matching. For example, identify which equipment is currently idle, and which process steps are not completed but are crucial for the new task, etc. The k most relevant nodes selected from the retrieval stage form a context supplementary dataset. Then, the system combines this context information and the user input to generate a detailed production scheduling instruction. For example, "First start Equipment A, and immediately start the operation of Equipment B after it is completed; the estimated total time does not exceed 4 hours." In this process, R-GAT captures the implicit task dependency chain, such as the completion of Equipment A is a prerequisite for the operation of Equipment B; while RAG uses this information to efficiently locate and generate scheduling instructions that meet the actual needs.

[0080] As can be seen from the above example, the combination of the attention-based relational graph neural network (R-GAT) and the retrieval-augmented generation (RAG) framework enables the intelligent production scheduling system to not only deeply understand the complex production environment and various internal relationships, but also quickly respond to the specific needs of users, providing accurate and efficient scheduling suggestions. This approach greatly improves the intelligent level and adaptability of production scheduling, and helps to achieve more flexible and efficient production management.

[0081] Furthermore, in addition to the existing structured modeling and semantic association analysis, technologies such as time series analysis and machine learning models can be introduced to predict future possible events, such as equipment failure early warning and raw material demand forecasting, so as to make preparations in advance and reduce downtime and costs. Integrate external knowledge bases from different fields (such as supply chain management, market demand analysis, etc.) into the current system, and capture the implicit relationships between these fields through R-GAT to provide more comprehensive support for production scheduling. In the instruction generation module, the format of the generated instructions (concise or detailed) can be customized according to the preferences of different users, or a special instruction template can be designed for the needs of specific industries to improve the applicability and user experience of the instructions.

[0082] Adjust the scheduling strategy in real time according to user feedback and production environment changes.

[0083] Specifically, the system will collect direct feedback from users, including subjective opinions such as the evaluation of the logic of instructions and the rationality of priority settings. In addition to user feedback, the system will also automatically collect actual execution data of production tasks, such as objective indicators such as task completion time and equipment utilization rate. Design a reward function based on the above information to quantitatively evaluate the effect of the current scheduling strategy. For example:

[0084]

[0085] Among them, represents the task completion efficiency, represents the resource utilization rate, represents the user satisfaction, , , are weight coefficients used to balance the importance of various indicators.

[0086] In the reinforcement learning framework, define the production environment and task status as the state , while the scheduling instruction is used as the action .

[0087] Adopt a policy gradient-based method (such as the REINFORCE algorithm) to optimize the scheduling strategy. Specifically, the policy selects actions according to the current state to maximize the cumulative reward . The policy update follows the following formula:

[0088]

[0089] Among them, is the objective function of the policy, is the discount factor, is the time tImmediate rewards obtained.

[0090] Continuously adjust scheduling parameters such as task priorities and resource allocation plans using the latest feedback data to adapt to changes in the current production environment. When facing different types of production environments, through policy migration technology, it can quickly adapt to the needs of the new environment while maintaining the core scheduling principles unchanged. During the multi-round task generation process, if inefficiencies or errors are found in the existing scheduling scheme, the system can correct them through a dynamic optimization mechanism to ensure the consistency between the instruction logic and the actual production requirements.

[0091] Continuously monitor the status of production equipment, task progress, and environmental parameters to identify any factors that may affect the production process. Once an abnormal situation (such as equipment failure, task delay) is detected, immediately initiate a predefined response plan and dynamically adjust the task scheduling order. Regularly update the parameters in the reinforcement learning model based on the newly collected data and user feedback to make the scheduling strategy more in line with the actual situation and improve the adaptability and efficiency of the overall system.

[0092] Through this mechanism, the intelligent production scheduling system can not only formulate a reasonable scheduling plan in the initial stage but also make timely adjustments according to changes in external conditions during operation, ensuring the flexibility and stability of production. This method greatly improves the intelligence level of the system, enabling it to operate efficiently in a complex and changing industrial environment.

[0093] For example, assume that in a manufacturing factory, the task arrangement on the production line depends on the status of multiple devices, the requirements of the process flow, and external factors (such as raw material supply). The system needs to be able to dynamically adjust the scheduling strategy according to real-time user feedback and changes in the production environment to ensure the efficient operation of production.

[0094] Current task arrangement: The production line is planned to complete three main processes in sequence: A (prepare materials), B (assemble parts), C (quality inspection).

[0095] Resource allocation: Equipment X is responsible for process A, equipment Y is responsible for process B, and equipment Z is responsible for process C.

[0096] Expected goal: All processes need to be completed within the next 8 hours, and seamless connection between each link should be ensured.

[0097] The dispatcher provided feedback through the system interface: "The actual execution time of process B is 30% longer than expected, which may affect the timely completion of subsequent processes."

[0098] Production environment changes: The system monitoring module detected a minor fault in device Y, resulting in a 25% reduction in its working efficiency. At the same time, the warehouse management system reported sufficient raw material supply, but the transportation vehicle was delayed due to traffic congestion, which might cause the specific materials required for process A to arrive 1 hour late.

[0099] After receiving user feedback and information on production environment changes, the reinforcement learning optimization module begins to evaluate the impact of these changes on the overall production plan. For example, calculate the time delay of the entire production chain due to the reduced efficiency of device Y and the delayed arrival of raw materials. Adjust the parameters in the reward function according to the new situation. For instance, increase the weight for the on-time completion of process B, and at the same time, considering that process A may start late, appropriately relax the strict requirements for the initial stage.

[0100] The system re-evaluates the priorities of each process and decides to allocate more resources to process B to speed up the progress, while considering whether part of the preparatory work for process C can be started in advance.

[0101] If possible, call spare equipment or personnel from other production lines to support the work of process B, or adjust the workload of device Y to reduce unnecessary downtime.

[0102] Based on the latest prediction data, re-plan the specific execution time of each process. For example, postpone the start time of process A to match the actual arrival time of raw materials and adjust the plans of subsequent processes accordingly.

[0103] The system generates detailed scheduling instructions through a large language model, clearly indicating the new time arrangements, required resources, and precautions for each step. For example, "Please start process A immediately after the arrival of raw materials, with an expected delay of 1 hour; increase the manpower input for process B to ensure completion within the original time; process C can conduct pre-inspections of some non-critical items in advance."

[0104] During the implementation of the new scheduling strategy, the system continues to monitor changes in the production environment. If any new problems or opportunities are found (such as the repair speed of device Y exceeding expectations), the strategy is immediately adjusted again to ensure continuous development in the most optimized direction.

[0105] Furthermore, use historical data and machine learning models to conduct predictive analysis of the device status, identify potential failure risks in advance, and arrange maintenance work before problems occur. Combine the requirements of the current production task and the actual operating conditions of the device to intelligently adjust the maintenance plan and avoid production interruptions caused by over-maintenance or neglect of maintenance.

[0106] By monitoring the raw material inventory in real time, combined with market demand forecasting, dynamically adjust the procurement plan to ensure that there is always sufficient raw material supply for the production line while reducing inventory backlogs. Based on external factors such as traffic information and weather forecasts, optimize the raw material transportation routes and schedules to ensure that the materials arrive at the factory on time.

[0107] Create a digital twin model of the physical production line, test different scheduling schemes and their effects in a virtual environment, and apply the optimal solution to actual production after selection. The digital twin model is kept in real-time synchronization with the actual production line, and any changes in the production environment can be immediately reflected in the digital model, facilitating timely responses.

[0108] Combine the structured data and the scheduling strategy, and use a large language model to generate scheduling instructions.

[0109] Specifically, through user request input and processing, the user's natural language request is transformed into structured data containing key information such as task type, target entity, priority, etc. The dynamic knowledge base combines a relational graph attention network (R-GAT) to model multi-source heterogeneous production data, and obtains relevant context background information through a retrieval-augmented generation (RAG) framework.

[0110] Utilize reinforcement learning optimization to adjust the scheduling strategy in real time according to user feedback and changes in the production environment, ensuring that task priorities, resource allocations, etc. are all optimally set. The scheduling strategy not only considers the current task requirements, but also comprehensively considers factors such as historical data, equipment status, and process flow to ensure the efficiency and continuity of production.

[0111] The large language model runs through the whole process as the core technology, especially playing an important role in parsing user requests, managing context, reasoning, and instruction generation. Instruction generation uses the powerful semantic understanding and generation capabilities of the large language model to integrate the information obtained from the dynamic knowledge base, R-GAT, and hierarchical memory mechanism to generate specific executable scheduling instructions. These instructions usually include, but are not limited to, detailed information such as task type, target entity, execution conditions, priority, time requirements, and subsequent steps. For example, "Check the running status of equipment C, give priority to checking the power module, and notify the maintenance team after completion. If any abnormalities are found, record the log and send a report."

[0112] When generating instructions, the large language model will refer to the hierarchical storage information (global planning, process logic, and operation steps) provided by the hierarchical memory mechanism to ensure the consistency of task sequence, dependency relationships, and completion conditions.

[0113] Meanwhile, by integrating the in-depth reasoning results provided by R-GAT, the large language model can understand the task background, thereby improving the applicability and accuracy of the generated instructions. According to the results of the reinforcement learning optimization module, the large language model can also adjust the instruction format and priority. For example, in an emergency task scenario, high-priority tasks will be generated first.

[0114] When the user input is vague or some details are omitted, the large language model can automatically complete these missing parts based on the context to ensure that the generated instructions are complete and logical. In the face of emergencies or changes in user requirements, the system can quickly respond, adjust the priority of existing instructions, insert new urgent tasks, and regenerate relevant instructions.

[0115] Suppose the user enters a relatively vague request: "Adjust production line 1." The system first converts it into structured data, then combines the current state of the production environment (such as which processes are currently being carried out on production line 1 and which equipment is idle), and determines the optimal scheduling strategy through the reinforcement learning optimization module. Then, the large language model is used to generate specific scheduling instructions, such as: "Adjust the priority of the task 'assembly' on production line 1 to high and it needs to be completed within 2 hours. Immediately start the 'testing' task after completion."

[0116] For example, assume that in a manufacturing factory, the task arrangement of the production line depends on the status of multiple devices, the requirements of the process flow, and external factors (such as raw material supply). The system needs to be able to dynamically adjust the scheduling strategy according to real-time user requests and changes in the production environment, and generate detailed scheduling instructions through the large language model.

[0117] Current task arrangement: The production line is planned to complete three main processes in sequence: A (prepare materials), B (assemble parts), C (quality inspection).

[0118] Resource allocation: Device X is responsible for process A, device Y is responsible for process B, and device Z is responsible for process C.

[0119] Expected goal: All processes need to be completed within the next 8 hours, and seamless connection between each link should be ensured.

[0120] User request: The dispatcher enters a natural language request: "Please optimize the task sequence of production line 2 and give priority to completing the 'packaging' task."

[0121] Semantic parsing and structured conversion: The system converts the above request into structured data through a semantic parser: "Task type: optimization; Production line number: 2; Priority task: packaging."

[0122] Retrieving Context Information: The dynamic knowledge base combines an attention-based relational graph neural network (R-GAT) to model production data and retrieve relevant context information. For example, it is known that the current task on production line 2 is "assembly" and it is expected to take another 2 hours to complete; meanwhile, the "packaging" task depends on the completion of the "assembly" task.

[0123] Formulating Scheduling Strategies: The reinforcement learning optimization module adjusts the scheduling strategies in real-time according to the latest changes in the production environment (such as equipment status, task progress, etc.). Considering that the "packaging" task has been assigned a high priority, the system decides to start the "packaging" task immediately after the "assembly" task is completed and postpone the scheduling of other non-critical tasks.

[0124] Comprehensive Analysis and Decision Support: The large language model receives the retrieval results from the dynamic knowledge base, the in-depth reasoning output provided by R-GAT, and the context information in the hierarchical memory mechanism. It synthesizes this information, considering factors such as the overall goals and priorities of the global planning layer, the sequential relationships and dependency structures of the process logic layer, and the specific states and execution conditions of the operation step layer.

[0125] Generating Specific Executable Scheduling Instructions: Based on the above analysis, the large language model generates the following specific scheduling instructions:

[0126] Equipment Maintenance Task Instruction: Check the operating status of equipment C, prioritize checking the power module, and notify the maintenance team after completion. If any abnormalities are found, record the log and send a report.

[0127] Production Process Instruction: Adjust the priority of the "assembly" task on production line 2 to the normal level, which is expected to take another 2 hours to complete. Immediately start the "packaging" task after completion, with a high priority and an expected duration of 1 hour. Then continue with the remaining "testing" tasks.

[0128] Task Assignment Instruction: Assign the "material transportation" task to operator Li Si, use forklift No. 2, with an expected completion time of 1 hour, and the status needs to be updated after the task is completed.

[0129] Content and Logical Consistency of Instructions: During the process of generating instructions, the large language model ensures the logical clarity and adaptability of the instruction content. For example, in the production process instruction, it is clearly stated that the "packaging" task starts after the "assembly" task is completed, avoiding potential task conflicts or logical confusion.

[0130] At the same time, for ambiguous or omitted information, the large language model can automatically complete it based on the context. For example, if the user only mentions "optimize production line 2" but does not clearly state which specific tasks need to be adjusted, the system can infer the most likely parts to be adjusted based on historical data and the current production situation and generate detailed instructions accordingly.

[0131] Real-time monitoring and feedback: Once the scheduling instructions are generated and issued, the system will continue to monitor changes in the production environment and be ready to adjust the scheduling strategy and instruction content again according to new user feedback or environmental changes at any time to ensure the continuity and efficiency of production.

[0132] Furthermore, use a machine learning model to perform predictive analysis on the production process, identify risk factors that may affect the production progress in advance (such as equipment failures, raw material shortages, etc.), and automatically generate preventive instructions through a large language model. Combining real-time monitoring data and prediction results, the system can dynamically adjust the scheduling strategy and generate corresponding adjustment instructions through the large language model to ensure the continuity and efficiency of production.

[0133] By analyzing the feedback and behavior patterns of different users, the system can learn the preferences of users and customize and generate more personalized scheduling instructions accordingly. For example, some users may prefer a concise instruction format, while others need detailed step-by-step instructions. Support the generation of scheduling instructions in multiple formats (such as concise version, detailed version) to adapt to different user needs and application scenarios.

[0134] According to an embodiment of the present application, the context information is stored and managed through a three-layer structure of a global planning layer, a process logic layer, and an operation step layer.

[0135] Specifically, the global planning layer, function: Record the overall goals and priorities of production tasks, and provide global directional guidance for the scheduling instructions generated by the system.

[0136] Specific content: Overall schedule of tasks: Define the general arrangement of each task on the time axis, for example, which tasks need to be completed first and which can be processed later. Resource allocation strategy: Determine how resources required for different tasks (such as manpower, equipment, raw materials, etc.) are allocated to maximize resource utilization and avoid conflicts. Priority setting: Sort tasks based on business requirements or urgency to ensure that high-priority tasks can be processed in a timely manner.

[0137] The process logic layer, function: Store the sequential relationship and dependency structure of tasks to ensure the logical consistency of tasks in the process flow.

[0138] Specific content: Task dependencies: Define the sequence and dependencies between tasks. For example, task A must start after task B is completed, or task C needs to meet the conditions that both task D and task E are completed before it can be started. Process flow rules: Specify the standard operation steps in a specific production process flow and their logical connections with each other to ensure that each process is executed in the correct order. Exception handling rules: Preset the countermeasures when a certain task cannot be completed on time, such as skipping the task or adjusting the time arrangements of subsequent tasks.

[0139] The operation step layer, function: Track the specific status and execution conditions of each task, and reflect the completion progress and waiting conditions of the task in real time (such as equipment status or resource availability).

[0140] Specific content: Task status update: Continuously monitor and update the status of each task (such as not started, in progress, completed, etc.) to facilitate timely understanding of the current production progress. Execution condition check: Check whether all necessary prerequisite conditions are met before each task starts (such as whether the required equipment is idle, whether the raw materials are in place, etc.), and trigger the task start when the conditions are met. Dynamic adjustment: Automatically adjust the associated task plan according to the real-time monitored changes in task status (such as a task delay caused by equipment failure), and reduce the impact on the overall production process.

[0141] In order to ensure that the system can respond to changes in the production environment in a timely manner, the hierarchical memory mechanism adopts a dynamic management and update mechanism:

[0142] Incremental update rule: Adopt the incremental update method to adjust the memory content by real-time monitoring of task status changes (such as equipment failure, task delay, etc.). For example, when the task progress is updated, the update formula for the memory status is:

[0143]

[0144] Among them, represents the memory status at time t, is the update weight, is the task status feature at time t, is the mapping function of the task status feature.

[0145] During the multi-round task generation process, the module will dynamically fuse the user input, the retrieval results of the dynamic knowledge base, and the R-GAT inference output, adjust the content of each hierarchical memory, and ensure that the generated instructions are consistent with the context logic. For example, when the user modifies the task priority or adds a new task, the module can re-adjust the scheduling order and logical dependencies of each task to avoid instruction conflicts or logical errors.

[0146] Through this design of hierarchical storage and dynamic management, the intelligent production scheduling system can maintain logical coherence and real-time adaptability in complex task chains, providing context support and decision-making basis for the efficient operation of intelligent production scheduling. This method not only improves the flexibility and adaptability of the system, but also enhances the accuracy and efficiency of production scheduling.

[0147] According to an embodiment of the present application, the user request is converted into structured data by the semantic parser, and the fuzzy information is complemented in combination with the context, specifically as follows:

[0148] Extract key parameters through a semantic parser, and complete the missing conditions in the user request by combining the historical context information in the hierarchical memory mechanism;

[0149] If the user request is insufficient, guide the user to supplement details through a prompting mechanism;

[0150] Output the supplemented key parameters as the structured data.

[0151] Specifically, receive a user request: The system first receives a user's natural language request or instruction, which can be a text input, a voice command, or a direct scheduling instruction, etc.

[0152] For example, a user may input: "Arrange the maintenance of device A".

[0153] Semantic parsing: Use a semantic parser to analyze the user input and identify and extract key elements. These elements usually include the task type (such as repair, inspection, production), the target entity (such as the device name), the priority, the time requirement, etc.

[0154] In the above example, the semantic parser will identify "task type: device maintenance" and "target entity: device A". However, if the user does not explicitly provide the priority or time requirement, these information are regarded as missing.

[0155] Combine the historical context information in the hierarchical memory mechanism

[0156] Context completion: The hierarchical memory mechanism includes a global planning layer, a process logic layer, and an operation step layer, and stores the historical context information of past interactions. The system will use this information to complete the missing conditions in the current request.

[0157] For example, assume that the priority of all emergency maintenance tasks has been clarified as high and needs to be completed as soon as possible in the previous conversation. Then, the system can infer that the priority in the current request is high based on this background information, and infer that the time requirement is "as soon as possible" or a specific time limit.

[0158] Dynamic context fusion: The system not only relies on static historical records, but also updates and fuses the latest production environment status in real time. For example, if the current production line is in a high-load operation stage, the system may automatically adjust the priority or suggest that the user choose another time period for maintenance to reduce the interference to production.

[0159] If the user request is insufficient, guide the user to supplement details through a prompting mechanism

[0160] Identify information gaps: If it is found that some necessary information is still missing (such as the specific time requirement) after preliminary parsing, the system will enter the next step.

[0161] Generate prompt feedback: The system automatically generates targeted questions or prompts to guide the user to provide more details. For example, "When do you want the maintenance work of Device A to be completed?" or "Please confirm whether the maintenance of Device A has the highest priority."

[0162] User response and supplementation: The user provides additional information according to the prompt, such as "I hope to complete the maintenance of Device A within the next two hours and set it as the highest priority."

[0163] Output the supplemented key parameters as the structured data

[0164] Integrate information: The system combines the original request information provided by the user with the supplementary information obtained through the context completion and prompt mechanism to form a complete task description.

[0165] Structured conversion: Finally, the integrated information is converted into a structured data format for further processing by subsequent modules. For example, "Task type: Equipment maintenance; Target entity: Device A; Priority: High; Time requirement: Complete within two hours."

[0166] Verification and optimization: The system will comprehensively check the generated structured data to ensure that all necessary information is included and consistent with each other. If there are any doubts or conflicts, the system will continue to request the user for further clarification through the feedback mechanism.

[0167] Suppose the user enters a relatively brief request: "Adjust the task order of production line 1." The system first attempts to extract the key parameters from it but finds that the specific task name to be adjusted and the new priority setting are missing. So, the system accesses the hierarchical memory mechanism to search for early interaction records related to production line 1 and discovers that the priority of the "packaging" task was recently discussed and needs to be increased. Based on this background, the system generates a prompt: "Do you want to adjust the priority of the 'packaging' task on production line 1 to high? If so, please confirm the estimated completion time."

[0168] User reply: "Yes, set the priority of the 'packaging' task to high and hope to complete it within the next 3 hours." The system then converts this request into structured data: "Task type: Adjust order; Production line number: 1; Priority task: Packaging; Priority: High; Time requirement: Complete within 3 hours."

[0169] In this way, the system can not only accurately understand the user's intention but also ensure that the generated instructions are both complete and logical, greatly improving the efficiency and accuracy of production scheduling.

[0170] According to an embodiment of the present application, the production data is modeled using an attention-based relational graph neural network, and relevant context information is retrieved in combination with a retrieval-augmented generation framework. Specifically:

[0171] Integrate the production data from different sources into a knowledge graph, where nodes represent production entities and edges represent the relationships between entities.

[0172] Use a vectorized storage model to match the user request with the nodes in the knowledge graph.

[0173] The attention-based relational graph neural network captures implicit associations and key dependencies between tasks.

[0174] Specifically, multi-source heterogeneous data integration:

[0175] Dynamic knowledge base: First, the system integrates heterogeneous production data from different sources into a production-based knowledge graph. This data may include device information, process flows, task records, and environmental data, etc.

[0176] Nodes and edges: In this knowledge graph, nodes represent production entities (such as device A, task B, process step C, etc.), and edges represent the relationships between these entities (such as task dependencies, device linkages, process sequences, etc.). For example, nodes can be specific production devices or process steps, and edges indicate whether there is a certain operation dependency between these devices.

[0177] Standardization and real-time update:

[0178] Semantic labeling and data standardization techniques: To ensure data consistency and comparability, semantic labeling and data standardization techniques are used to convert the original data into a unified structured form.

[0179] Dynamic update: The dynamic knowledge base supports real-time updates and can adjust node attributes (such as task priorities) and relationship weights (such as task dependency strengths) according to the latest production status to reflect changes in the current production environment.

[0180] Use a vectorized storage model for matching

[0181] Vectorized storage and retrieval:

[0182] Embedding model: Use an embedding model (such as Transformer) to convert the user request and the nodes and relationships in the knowledge graph into high-dimensional vector representations. This enables the user request to quickly locate relevant nodes and context information through vector matching.

[0183] Cosine similarity matching: Measure their relevance by calculating the cosine similarity between the user request vector and the knowledge graph node embeddings. The formula is as follows:

[0184]

[0185] Among them, is a query embedding model that converts the user request q into a vector, is an embedding model that converts the node in the knowledge base into a vector.

[0186] In the retrieval stage, the RAG framework will select the k nodes most relevant to the user request to form a context supplement data set. This method effectively avoids the hallucination problem in traditional generation methods and improves the accuracy and relevance of the generated content.

[0187] Generation stage: Combine the context information and the user input to generate production scheduling instructions that better meet the actual scenario requirements. This context-based generation method helps to improve the quality and applicability of the instructions.

[0188] Attention-based relational graph neural network captures implicit associations

[0189] Relational attention mechanism:

[0190] Dynamic weight allocation: Dynamically allocate the weights of different relationships in the knowledge graph through the relational attention mechanism to highlight the semantic associations between key tasks. Specifically, for the attention weight of nodes u and 𝑣 under relationship 𝑟

[0191]

[0192] Among them, and are the embedding representations of nodes u and v respectively, is the feature weight matrix of relationship 𝑟.

[0193] Combining the results of the attention mechanism, R-GAT performs multiple embedding updates on each node to capture its semantic associations with neighbor nodes. The update formula is as follows:

[0194]

[0195] Among them, is the embedding representation of node 𝑣 at the l+1 layer, is the activation function (such as ReLU).

[0196] Deep modeling ability: The multi-level modeling ability of R-GAT ensures that the system can capture implicit task dependency chains and optimize task sequences in process flows in complex task scenarios. For example, in a multi-task dependency chain, the delay of a certain task may affect the scheduling of a series of subsequent tasks, and R-GAT can identify and handle such complex dependencies.

[0197] Suppose the user submits a natural language request: "Schedule the maintenance of device A". The system will perform the following steps:

[0198] Data integration: Obtain all information related to device A from the knowledge graph, including the device status, last maintenance time, current task dependencies, etc.

[0199] Vectorized storage and retrieval: Convert the user request "Schedule the maintenance of device A" into a vector and match it with the nodes in the knowledge graph to find all nodes related to device A.

[0200] Attention mechanism application: Through the relational attention mechanism, the system identifies the key dependencies between device A and other devices or tasks, such as which tasks need to wait for the maintenance of device A to be completed before starting.

[0201] Generate scheduling instructions: Based on the above analysis results, the system generates detailed scheduling instructions: "Please immediately schedule the maintenance work of device A, which is expected to take 2 hours. During this period, please suspend all tasks that depend on device A."

[0202] In this way, the system can not only deeply understand the complex production environment and various internal relationships, but also efficiently respond to the specific needs of users, provide accurate and efficient scheduling suggestions, thus greatly improving the intelligent level and adaptability of production scheduling.

[0203] According to an embodiment of the present application, the scheduling strategy is adjusted in real time according to user feedback and production environment changes, specifically:

[0204] Construct a reward signal through the user feedback and the production environment changes to guide the optimization of the scheduling strategy;

[0205] Use an algorithm based on policy gradient to dynamically adjust the scheduling strategy parameters according to real-time data to maximize the cumulative reward.

[0206] Specifically, subjective evaluation: The system will collect direct feedback from users, including subjective opinions such as the evaluation of the logic of the instructions and the rationality of the priority setting.

[0207] Objective data: In addition to user feedback, the system will also automatically collect actual execution data of production tasks, such as objective indicators such as task completion time and equipment utilization rate.

[0208] Construct the reward function:

[0209] Quantitative evaluation: Design a reward function based on the above information to quantitatively evaluate the effect of the current scheduling strategy. For example, the following formula can be defined to measure the performance of the scheduling strategy:

[0210]

[0211] where represents the task completion efficiency, represents the resource utilization rate, represents the user satisfaction, , , are weight coefficients used to balance the importance of various indicators.

[0212] According to different production goals or business requirements, these weight coefficients can be dynamically adjusted to adapt to different optimization focuses (such as improving efficiency or increasing user satisfaction).

[0213] State representation: Define the production environment and task status as state . For example, the state can include the status of the device (running, idle, faulty), the task progress (not started, in progress, completed), the resource availability, etc.

[0214] Action representation: The scheduling instruction is used as the action . For example, the action can be to start a certain device, pause a certain task, adjust the task priority, etc.

[0215] Policy gradient method: Adopt a policy gradient-based method (such as the REINFORCE algorithm) to optimize the scheduling strategy. Specifically, the policy , selects actions according to the current state to maximize the cumulative reward , and the policy update follows the following formula:

[0216]

[0217] where is the objective function of the policy, is the discount factor used to consider the time value of future rewards, is the immediate reward obtained at time 𝑡.

[0218] Real-time data-driven: The system uses real-time monitored data (such as equipment status changes, task completion status, user feedback, etc.) to continuously adjust the policy parameters to better fit the current production environment. For example, if a minor fault is detected in a certain device, the system can dynamically adjust the priorities of related tasks, reduce the dependence on this device, or schedule maintenance work in advance.

[0219] Policy iteration and optimization: Through continuous experimentation and feedback loops, the system gradually optimizes its scheduling policy. After each adjustment, the system evaluates the effectiveness of the new policy and further fine-tunes the parameters based on the results to achieve the optimal production effect.

[0220] Specific implementation steps for real-time adjustment of the scheduling policy

[0221] Monitoring and data analysis: Continuously monitor the status of production equipment, task progress, and environmental parameters to identify any factors that may affect the production process. For example, monitor the operating conditions of equipment, raw material inventory levels, task execution progress, etc.

[0222] Anomaly detection and response: Once an anomaly (such as equipment failure, task delay) is detected, immediately initiate a predefined response plan and dynamically adjust the task scheduling order. For example, when it is detected that equipment Y fails, the system can reassign tasks to backup equipment and at the same time adjust the time arrangements of subsequent tasks to minimize the impact on the overall production.

[0223] Generate new scheduling instructions: Based on the latest prediction data and optimized policies, the system generates detailed scheduling instructions through a large language model. For example, "Please start process B immediately after equipment Y is repaired, which is expected to take 2 hours; during this period, please suspend all tasks that depend on equipment Y and reschedule other executable tasks."

[0224] Continuous monitoring and further adjustment: During the implementation of the new scheduling policy, the system continues to monitor changes in the production environment. If any new problems or opportunities (such as the repair speed of equipment Y exceeding expectations) are found, the policy is immediately adjusted again to ensure continuous development in the direction of optimization.

[0225] Suppose in a manufacturing factory, the task arrangement of the production line depends on the status of multiple devices, the requirements of the production process, and external factors (such as raw material supply). The system needs to be able to dynamically adjust the scheduling policy according to real-time user feedback and changes in the production environment.

[0226] Initial state

[0227] Current task arrangement: The production line plans to sequentially complete three main processes: A (prepare materials), B (assemble parts), and C (quality inspection).

[0228] Resource Allocation: Equipment X is responsible for Process A, Equipment Y is responsible for Process B, and Equipment Z is responsible for Process C.

[0229] Expected Goal: All processes need to be completed within the next 8 hours, and seamless connection between each link should be ensured.

[0230] User Feedback and Changes in the Production Environment

[0231] User Feedback: The dispatcher provided feedback through the system interface: "The actual execution time of Process B is 30% longer than expected, which may affect the timely completion of subsequent processes."

[0232] Changes in the Production Environment: The system monitoring module detected a minor fault in Equipment Y, resulting in a 25% reduction in its working efficiency.

[0233] Meanwhile, the warehouse management system reported that the raw material supply is sufficient, but the transport vehicle was delayed due to traffic congestion, which may cause a 1-hour delay in the arrival of specific materials required for Process A.

[0234] The Process of Real-time Adjustment of the Scheduling Strategy

[0235] Constructing the Reward Signal: The system constructs the reward signal based on user feedback and changes in the production environment, adjusts the weights in the reward function. For example, increase the weight for the timely completion of Process B, and at the same time, considering that Process A may start late, appropriately relax the strict requirements for the initial stage.

[0236] Policy Update: Use a policy gradient-based method (such as the REINFORCE algorithm) to optimize the scheduling strategy. According to the latest changes in the production environment (such as the reduced efficiency of Equipment Y and the delayed arrival of raw materials), the system re-evaluates the priorities of each process, decides to allocate more resources to Process B to speed up the progress, and also considers whether some preparatory work for Process C can be started in advance.

[0237] Generating New Scheduling Instructions:

[0238] The system generates detailed scheduling instructions through a large language model, clearly indicating the new time arrangements, required resources, and precautions for each step. For example, "Please start Process A immediately after the arrival of raw materials, with an expected delay of 1 hour; increase the manpower input for Process B to ensure completion within the original time; for Process C, some non-critical pre-inspection work can be carried out in advance."

[0239] In this way, "real-time adjustment of the scheduling strategy according to user feedback and changes in the production environment" not only helps solve sudden problems but also improves the flexibility and adaptability of the entire production process, ensuring efficient operation even in the face of uncertainties. This method greatly enhances the intelligence level of the system, enabling it to maintain efficient operation in a complex and changing industrial environment.

[0240] According to an embodiment of the present application, combining the structured data and the scheduling policy, using a large language model to generate a scheduling instruction, specifically:

[0241] Using the large language model to combine the structured data and the scheduling policy to generate the scheduling instruction;

[0242] Automatically adjust the instruction content according to changes in user requirements or sudden environmental situations.

[0243] Specifically, integrating structured data and scheduling policy:

[0244] Structured data preparation: Through the user request input and processing module, the user's natural language request is converted into structured data containing key information such as task type, target entity, priority, etc.

[0245] Scheduling policy formulation: Using the reinforcement learning optimization module, the scheduling policy is adjusted in real time according to user feedback and production environment changes to ensure that task priorities, resource allocations, etc. are all optimized.

[0246] Application of the large language model:

[0247] Instruction content generation: The large language model runs through the entire process as the core technology, especially playing an important role in parsing user requests, managing context, reasoning, and instruction generation. The instruction generation module uses the powerful semantic understanding and generation capabilities of the large language model to integrate the information obtained from the dynamic knowledge base, R-GAT, and hierarchical memory mechanism to generate specific executable scheduling instructions.

[0248] For example, assume the system receives the following structured data: "Task type: equipment maintenance; Equipment name: Equipment A; Priority: high; Time requirement: complete within today." Combining the current scheduling policy (such as Equipment B depends on the normal operation of Equipment A), the large language model can generate a specific scheduling instruction: "Immediately arrange the maintenance work of Equipment A, set the priority to high, and the estimated time consumption is 4 hours. During this period, please suspend all tasks that depend on Equipment A and re-arrange other executable tasks."

[0249] Logical consistency and adaptability:

[0250] Context fusion: When generating instructions, the large language model will refer to the hierarchical storage information (global planning, process logic, and operation steps) provided by the hierarchical memory mechanism to ensure the consistency of task sequence, dependency relationship, and completion conditions.

[0251] Dynamic adjustment: In the face of emergencies or changes in user requirements, the system can respond quickly, adjust the priorities of existing instructions, insert new urgent tasks, and regenerate relevant instructions. For example, if it is suddenly detected that device A has failed, the system can immediately update the scheduling instructions: "Due to the failure of device A, please immediately start the emergency repair procedure and notify the relevant personnel to be present for handling."

[0252] Automatically adjust the instruction content according to changes in user requirements or sudden environmental situations

[0253] Real-time monitoring and feedback loop:

[0254] Continuous monitoring: The system continuously monitors the status of production equipment, task progress, and environmental parameters to identify any factors that may affect the production process.

[0255] Feedback mechanism: Once an abnormal situation (such as equipment failure, task delay) is detected or a new user requirement is received (such as a temporary addition of an urgent task), the system will trigger the feedback mechanism, re-evaluate the current scheduling strategy, and adjust the scheduling instructions accordingly.

[0256] Automated adjustment and instruction update:

[0257] Automatically generate new instructions: Based on the latest scheduling strategy and user requirements, the large language model automatically generates new scheduling instructions. For example, when the raw material supply is delayed, the system can generate the following instructions: "Please postpone the start time of process A until the raw materials arrive, with an expected delay of 1 hour; at the same time, start some preparatory work for process C in advance to reduce the impact on the overall production plan."

[0258] Instruction optimization: In addition to directly generating new instructions, the system can also optimize existing instructions. For example, when it is detected that the working efficiency of a certain device is lower than expected, the system can adjust the time arrangement of all tasks on that device and generate corresponding adjustment instructions.

[0259] Initial state: The production line is planned to complete three main processes in sequence: A (prepare materials), B (assemble parts), C (quality inspection).

[0260] Change in user requirements: The user submits a new request: "Please complete the 'packaging' task as soon as possible."

[0261] The system first converts this request into structured data: "Task type: optimization; production line number: 2; priority task: packaging."

[0262] Combined with the status of the current production environment (such as which processes are being carried out on production line 2 and which equipment is in an idle state, etc.), and determine the optimal scheduling strategy through the reinforcement learning optimization module - decide to immediately start the "packaging" task after the "assembly" task is completed, and postpone the scheduling of other non-critical tasks.

[0263] The large language model generates specific scheduling instructions: "Adjust the priority of the task 'assembly' on production line 2 to the normal level, and it is expected to take another 2 hours to complete. Immediately start the 'packaging' task after completion, with the priority set to high, and it is expected to take 1 hour. Then continue with the remaining 'testing' tasks."

[0264] Respond to emergencies

[0265] Emergency situation: Assume that during the execution of the above instructions, equipment Y suddenly fails.

[0266] The system discovers this problem through real-time monitoring and immediately adjusts the scheduling strategy.

[0267] The large language model generates new scheduling instructions: "Due to the failure of equipment Y, please immediately arrange a maintenance team for inspection and repair. During this period, please suspend all tasks that depend on equipment Y and re-arrange other available resources to execute non-critical tasks."

[0268] In this way, the process of using the large language model to generate scheduling instructions by combining structured data and scheduling strategies not only improves the intelligence level and scenario adaptability of instruction generation, but also ensures the orderliness and practicality of task execution in each link, greatly improving the efficiency and accuracy of production scheduling. This method enables the system to maintain efficient operation in a complex and changing industrial environment and respond to various changes and challenges in a timely manner.

[0269] According to an embodiment of the present application, it further includes:

[0270] Real-time monitor the production status, quickly identify and classify anomalies, and dynamically adjust the scheduling plan.

[0271] According to an embodiment of the present application, the real-time monitoring of the production status, quickly identifying and classifying anomalies, and dynamically adjusting the scheduling plan are specifically:

[0272] Track the status of production equipment, task execution, and environmental parameters;

[0273] Identify and classify anomalies in the production process, and take corresponding handling measures according to the severity;

[0274] Automatically generate a response plan, dynamically adjust task scheduling, and minimize the impact of anomalies on production.

[0275] Specifically, real-time monitor the production status

[0276] Tracking the status of production equipment:

[0277] Equipment health monitoring: The system continuously monitors the operating status of production equipment, including but not limited to key parameters such as temperature, pressure, vibration frequency, etc., to evaluate the working condition of the equipment.

[0278] Fault warning: By setting thresholds or using machine learning models to predict potential equipment failures, warning signals are sent in advance to facilitate timely preventive measures.

[0279] Monitoring the progress of task execution:

[0280] Progress tracking: Real-time tracking of the actual execution progress of each production task, comparing and analyzing it with the scheduled plan to ensure that all tasks are advanced according to the established schedule.

[0281] Bottleneck detection: Identify task bottlenecks that may lead to a reduction in production line efficiency and provide optimization suggestions.

[0282] Monitoring environmental parameters:

[0283] Environmental condition monitoring: Monitoring environmental parameters such as temperature, humidity, air quality, etc., to ensure that they are within the range suitable for production and avoid product quality problems or equipment damage caused by environmental factors.

[0284] Quickly identify and classify abnormal situations

[0285] Abnormality detection:

[0286] Multi-source data analysis: Using sensor data, historical records, and other relevant information, automatically detect situations that deviate from the normal operating range through algorithms. For example, when the temperature of a certain piece of equipment exceeds the safe range, the system will mark it as abnormal.

[0287] Pattern recognition: Adopting advanced pattern recognition technologies (such as deep learning) to identify complex abnormal patterns, not limited to single-point abnormalities, but also including associated abnormalities between multiple variables.

[0288] Abnormality classification:

[0289] Severity grading: Classify abnormalities according to the degree of impact on production, such as mild, moderate, and severe levels. This helps to prioritize the handling of problems that have the greatest impact on production.

[0290] Type classification: Classify abnormalities into different types, such as equipment failures, raw material shortages, human errors, etc., to facilitate the formulation of targeted solutions.

[0291] Dynamically adjust the scheduling plan

[0292] Response plan generation:

[0293] Automated Decision Support: Once an anomaly is identified, the system automatically generates a response plan, including specific repair steps, required resources, and estimated recovery time. For example, for a device failure, the system may recommend immediately activating the backup device and arranging for a maintenance team to conduct an inspection.

[0294] Notification Mechanism: Timely notify relevant personnel or departments about the anomaly and its handling solutions to ensure the timeliness and accuracy of information transmission.

[0295] Dynamic Adjustment of Task Scheduling:

[0296] Re-planning Task Sequence: Based on the current anomaly and available resources, the system dynamically adjusts the task scheduling sequence to minimize the impact of the anomaly on overall production. For example, during a device failure, tasks that depend on the faulty device can be paused and other unrelated tasks can be executed instead.

[0297] Resource Reallocation: If a bottleneck occurs in a certain link, the system can reallocate human and material resources according to the actual situation to alleviate the bottleneck effect.

[0298] Example Illustration

[0299] Initial State: Assume a production line is planned to complete three processes in sequence: A (preparing materials), B (assembling parts), and C (quality inspection). Device X is responsible for process A, device Y is responsible for process B, and device Z is responsible for process C.

[0300] Anomaly Occurrence: During production, the system monitors that device Y has a minor fault, resulting in a 25% reduction in its working efficiency.

[0301] Identification and Classification: The system identifies this anomaly and classifies it as "moderate" according to its impact on production because although the efficiency has decreased, the work has not completely stopped.

[0302] Response Plan: The system automatically generates a response plan, suggesting temporarily reducing the task load on device Y and arranging for a maintenance team to conduct an inspection; meanwhile, notify the dispatcher to re-plan the execution sequence of subsequent tasks.

[0303] Dynamic Adjustment: The scheduling plan is dynamically adjusted to decide to postpone some tasks that depend on device Y and initiate some preparatory work for process C in advance to minimize the impact on the overall production plan.

[0304] Through the above mechanism, the system can monitor the production status in real time, quickly identify and classify abnormal situations, and dynamically adjust the scheduling plan accordingly to ensure the continuity and efficiency of production even in the face of unexpected problems. This method greatly improves the flexibility and adaptability of the system, enabling it to operate stably in a complex and changeable industrial environment.

[0305] A computer-readable storage medium, on which a program is stored, and when the program is executed by a processor, it implements the steps in the method as described above.

[0306] An electronic device, including a memory, a processor, and a program stored on the memory and executable on the processor, and when the processor executes the program, it implements the steps in the method as described above.

[0307] What is not described in this application can be achieved by adopting or referring to the existing technology.

[0308] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments.

[0309] The above are only the embodiments of this application and are not intended to limit this application. For those skilled in the art, various changes and modifications can be made to this application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this application shall be included within the scope of the claims of this application.

Claims

1. An intelligent production scheduling and anomaly management method based on large language models, characterized in that Including: Converting the user request into structured data through a semantic parser and complementing fuzzy information in combination with the context; Modeling the production data using an attention-based relational graph neural network and retrieving relevant context information in combination with a retrieval-augmented generation framework; Integrating the production data from different sources into a knowledge graph, where nodes represent production entities and edges represent the relationships between entities; Using a vectorized storage model to match the user request with the nodes in the knowledge graph; The attention-based relational graph neural network captures implicit associations and key dependencies between tasks; According to user feedback and production environment changes, adjust the scheduling strategy in real time; Combining the structured data and the scheduling strategy, using a large language model to generate scheduling instructions.

2. The method according to claim 1, wherein The context information is stored and managed through a three-layer structure of a global planning layer, a process logic layer, and an operation step layer.

3. The method according to claim 1, wherein The conversion of the user request into structured data through the semantic parser and the complementing of fuzzy information in combination with the context are specifically as follows: Extracting key parameters through the semantic parser and complementing the missing conditions in the user request in combination with the historical context information in the hierarchical memory mechanism; If the user request is insufficient, guiding the user to supplement details through a prompting mechanism; Outputting the supplemented key parameters as the structured data.

4. The method according to claim 1, wherein The real-time adjustment of the scheduling strategy according to user feedback and production environment changes is specifically as follows: Constructing a reward signal through the user feedback and the production environment changes to guide the optimization of the scheduling strategy; Using a policy gradient-based algorithm to dynamically adjust the scheduling strategy parameters according to real-time data to maximize the cumulative reward.

5. The method according to claim 1, wherein The generation of scheduling instructions using the large language model in combination with the structured data and the scheduling strategy is specifically as follows: Using the large language model to generate the scheduling instructions in combination with the structured data and the scheduling strategy; Automatically adjusting the instruction content according to changes in user requirements or sudden environmental situations.

6. The method according to claim 1, wherein Also including: Real-time monitoring of the production status, quickly identifying and classifying anomalies, and dynamically adjusting the scheduling plan.

7. The method according to claim 6, wherein The real-time monitoring of the production status, quickly identifying and classifying anomalies, and dynamically adjusting the scheduling plan are specifically as follows: Tracking the status of production equipment, task execution, and environmental parameters; Identifying and classifying abnormal situations in the production process and taking corresponding handling measures according to the severity; Automatically generating a response plan, dynamically adjusting task scheduling, and minimizing the impact of anomalies on production.

8. A computer-readable storage medium having a program stored thereon, characterized in that, When the program is executed by a processor, it implements the steps in the method according to any one of claims 1-7.

9. An electronic device, comprising a memory, a processor, and a program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the method according to any one of claims 1-7.

Images