Multi-agent collaborative analysis system based on industrial large model

Abstract

The invention discloses a multi-agent collaborative analysis system based on an industrial large model. The multi-agent collaborative analysis system comprises a demand analysis module, a task triggering module, a task scheduling module, an information transmission module, a conflict processing module and a task execution module. According to the method, the user demand is analyzed, the task is triggered and transmitted to each agent to execute and output the analysis data, so that cooperation among multiple agents is realized, various input data is converted into multi-dimensional and real-time deep analysis of data related to system decision making, the utilization rate of the data is improved, different production scenes can be quickly adapted, and the production efficiency is improved. The full-automatic process from data input to decision output is realized, and flexible and efficient support is provided for complex production tasks.

Description

Technical Field

[0001] The present invention relates to the field of manufacturing information systems, and in particular to a multi-agent collaborative analysis system based on an industrial large model. Background Art

[0002] The intelligent application of manufacturing information systems has developed rapidly in recent years. The manufacturing industry has accumulated a large amount of business data, but the analysis and utilization of this data face major challenges. Although the current manufacturing information system can store and display data, in actual applications, the proportion of data that can be deeply analyzed and efficiently utilized is extremely low. The display of data is mainly in the form of fixed reports, lacking multi-dimensional data insights and in-depth analysis, and unable to provide sufficient support for complex production decisions.

[0003] In the prior art, there are the following problems:

[0004] 1. Current data acquisition and display have limitations. Data query and acquisition mainly rely on fixed statements, with a single display format and lack of dynamic adjustment capabilities, resulting in low data utilization and difficulty in supporting refined management and complex decision-making.

[0005] 2. Data analysis in existing technologies usually relies on preset rule models. Analysis or alerts are only given when data triggers specific rule conditions, which increases the difficulty of implementation and is difficult to expand to diverse production scenarios. In addition, existing analysis methods require specially designed algorithms for certain specific data and are highly dependent on domain knowledge, increasing development costs and implementation complexity. Summary of the invention

[0006] In order to solve the problems of low data utilization and insufficient analysis depth in the existing manufacturing information system, the present invention provides a multi-agent collaborative analysis system based on industrial big model. To this end, the technical solution adopted by the present invention is:

[0007] A multi-agent collaborative analysis system based on industrial large models is provided, including:

[0008] The demand analysis module is used to use the user intention analysis agent to analyze the user's input and obtain the user's needs;

[0009] The task triggering module is used to trigger system task requirements according to user requirements or system anomalies, and distribute tasks to each agent according to the adaptability of resources, the logical relationship between tasks, and the load balancing strategy;

[0010] The task scheduling module is used to analyze the execution order of tasks by using the central management agent to dynamically plan according to the task requirements, system resource status and logical dependencies between tasks;

[0011] The information transmission module is used to transmit the task execution order information to each agent to complete the task allocation;

[0012] The conflict handling module is used to perform conflict detection after task assignment. After detecting the task conflict between agents, task coordination and optimization are performed to obtain the optimized task flow.

[0013] The task execution module is used to enable each intelligent agent to perform tasks according to the optimized task flow, output the data analyzed by each intelligent agent, and make production decisions based on the analyzed data.

[0014] According to the above scheme, the trigger system task requirements include user-driven triggering and exception-driven triggering. The user-driven triggering triggers the task specifically according to the user's needs; the exception-driven triggering triggers the task specifically according to the system's real-time monitoring of the running status.

[0015] According to the above scheme, the execution order of the dynamic programming analysis task specifically includes:

[0016] Receive user requirements and system status information, and describe the task attributes with quantitative parameters; the quantitative parameters include urgency, priority, computing power requirements, and task relevance;

[0017] Build a task dependency graph to model the tasks;

[0018] The execution priority of the task is calculated based on the correlation degree of each task in the task dependency graph and the corresponding weight coefficient and the urgency calculation result in the quantitative parameter, and a task execution sequence is generated; the urgency calculation result is calculated based on the task priority, time limit and the corresponding weights of the task priority and time limit.

[0019] According to the above scheme, after the task is completed, the weight coefficients corresponding to the degree of association of each task in the task dependency graph and the corresponding weights of the task priority and time limit in the urgency calculation result are adjusted through the feedback mechanism according to the execution time and resource consumption of the task, so as to optimize the execution order of subsequent tasks.

[0020] According to the above scheme, the conflict detection includes resource competition conflict detection and logic conflict detection;

[0021] The resource competition conflict detection is specifically: monitoring and analyzing the resource requirements of the task and the current resource status of the system, and when the difference between the resource requirements of the task and the current resource status of the system is less than 0, it is determined that there is a resource competition conflict;

[0022] The logic conflict detection is specifically as follows: performing a logic analysis on the tasks of each intelligent agent, and determining that a logic conflict exists when there is a logic contradiction between multiple tasks.

[0023] According to the above scheme, the resource optimization and logic rearrangement after detecting the task conflict between the agents specifically include:

[0024] Quantitative assessment of detected conflicts;

[0025] According to the conflict quantitative assessment results, the conflicting tasks are coordinated according to the preset resolution strategy; the preset resolution strategy includes priority strategy, resource allocation strategy and logic rearrangement strategy;

[0026] Verify the results of the coordinated task execution sequence and resource allocation plan.

[0027] According to the above scheme, the intelligent agents include: predictive maintenance intelligent agent, quality prediction and improvement intelligent agent, process parameter optimization intelligent agent, report generation intelligent agent, and SQL intelligent agent.

[0028] According to the above scheme, the quality prediction and improvement intelligent agent receives real-time production data and historical production data as input, and outputs production optimization suggestion data; the real-time production data includes sensor data, equipment status, and process parameters.

[0029] According to the above scheme, it is characterized in that the predictive maintenance intelligent agent receives real-time production data stream and historical maintenance data as input, and outputs maintenance recommendation data; the real-time production data stream includes equipment operating parameters and sensor data.

[0030] According to the above scheme, the process parameter optimization agent receives real-time production data, historical production data and process models as input, and outputs process parameter optimization recommendation data.

[0031] The beneficial effects produced by the present invention are:

[0032] 1. The present invention realizes collaboration among multiple intelligent agents by analyzing user needs, triggering tasks and passing them to each intelligent agent for execution and outputting analysis data, converting various types of input data into multi-dimensional, real-time in-depth analysis of system decision-related data, improving data utilization, and being able to quickly adapt to different production scenarios. It realizes a fully automated process from data input to decision output, providing flexible and efficient support for complex production tasks.

[0033] 2. The present invention optimizes the execution order of subsequent tasks through a feedback mechanism, realizes dynamic optimization and long-term improvement of the task execution process, effectively reduces the error rate, and improves the applicability of the system in a variety of scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a schematic diagram of the structure of a multi-agent collaborative analysis system based on an industrial large model according to an embodiment of the present invention;

[0035] Figure 2 It is a system technology architecture diagram of an embodiment of the present invention. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0037] The present invention provides a multi-agent collaborative analysis system based on industrial large models, such as Figure 1 As shown, the system includes a demand analysis module, a task triggering module, a task scheduling module, an information transmission module, a conflict handling module, and a task execution module, wherein:

[0038] The demand analysis module is used to use the user intention analysis agent to analyze the user's input and obtain the user's needs.

[0039] In a preferred embodiment of the present invention, the user intention analysis agent receives the operation log D of the production manager. 操作日志 , including data query, parameter adjustment, alarm processing and other behavioral data, as well as historical operation records 历史 As input, it is used to generate response plans. By providing accurate response plans, it can improve the efficiency of system production management and reduce the complexity of manual operations. Its calculation steps include:

[0040] Receive the operation log of production management personnel D 操作日志 , for the operation log data D 操作日志 Preprocess and extract key feature vectors: F 操作 =Φ 操作 (D 操作日志 );

[0041] Using AI big model M AI For the eigenvector F 操作 Analyze and combine historical records 历史 Predict the user’s current operation intention I 预测 , where the prediction model of operation intention is:

[0042] I 预测 =M 意图预测 (F 操作 ,D 历史 )

[0043] Among them, user intention I 预测 It may include data query (such as production report, equipment status), alarm processing (such as abnormal equipment location) or process parameter adjustment (such as optimizing production plan);

[0044] Based on the predicted user intent I 预测 , the agent generates a response plan G 响应 , including automatically locating data sources, generating relevant reports, or providing parameter optimization suggestions:

[0045] G 响应 =M 响应 (I 预测 ,D 操作日志 ).

[0046] The task triggering module is used to trigger system task requirements based on user requirements or system anomalies, and distribute tasks to each agent based on the adaptability of resources, the logical relationship between tasks, and the load balancing strategy.

[0047] Among them, the triggering system task requirements include user-driven triggering and exception-driven triggering. Among them, in user-driven triggering, the task is initiated by user demand, and the user input is analyzed by the user intention analysis agent. The system converts the analysis results into user needs and determines the task type, goal and related parameters to be performed. For example, when the user requests to analyze the data of a certain production link, the system will trigger the corresponding data collection and analysis tasks after analysis. This mode is suitable for clear user request scenarios;

[0048] The abnormality-driven trigger triggers the task based on the real-time monitoring of the system's operating status. The system captures abnormal events (such as sensor readings exceeding the threshold or abnormal equipment status) through the monitoring module. When an abnormal event is detected, the system analyzes the abnormality type and selects the appropriate task processing logic. In a preferred embodiment of the present invention, when the system detects an abnormal increase in the temperature of the equipment, it may trigger a fault diagnosis task or a shutdown maintenance task. For multiple abnormal events occurring at the same time, the system supports event correlation analysis, identifies the logical relationship between events, and triggers a comprehensive task to avoid duplication of tasks or omission of key issues.

[0049] The task scheduling module is used to use the central management agent to dynamically plan and analyze the task execution sequence according to task requirements, system resource status and logical dependencies between tasks.

[0050] Specifically, the execution sequence of dynamic programming analysis tasks includes:

[0051] S11. Receive user requirements and system status information, and describe the task attributes with quantitative parameters; the quantitative parameters include urgency, priority, computing power requirements, and task relevance.

[0052] S12. Build a task dependency graph and model the tasks.

[0053] S13. Calculate the execution priority of the task based on the task dependency graph and the urgency calculation result, and generate a task execution sequence.

[0054] In a preferred embodiment of the present invention, the central management agent receives user demand D 需求 and system status information D 状态 , quantitatively describe the attributes of the analysis task, each task T i The description includes the urgency level 紧急,i , Priority P 优先级,i 、Computing power requirements C 算力,i And the correlation between tasks R i,j ; and build the task dependency graph G 任务 Model the analysis task. The nodes of the task graph represent specific tasks, the edges represent the dependencies between tasks, and the weight of the dependency is R. i,j Represents task T i and Task T j The degree of association is determined by the analysis of the task content and the intensity of data interaction between agents; finally, based on the task dependency graph G 任务 and the urgency calculation result, calculate the execution priority S of the task 执行,i , and generate the task execution sequence Q accordingly 任务 .

[0055] In addition, during the task execution, the central management agent monitors the system resource status in real time. 资源 , including the current resource load L, resource occupancy U and availability C. When there is insufficient task resources, task timeout or new high-priority tasks are added, the system recalculates the priority score S 执行,i and update task sequence Q 任务 , dynamically adjust the scheduling plan.

[0056] Among them, the urgency level U 紧急,i The calculation of takes into account the priority and time limit of the task, and its calculation formula is:

[0057]

[0058] Among them, w1 and w2 are the priority weight and time limit weight respectively, which can be dynamically adjusted according to the task scenario.

[0059] The execution priority of the task S 执行,i :

[0060]

[0061] Among them, λ represents the weight coefficient of the dependency, which is used to adjust the impact of relevance on task sorting.

[0062] After completing the task, the system records the task execution time, resource consumption and other information, and uses the feedback mechanism to optimize the weight parameters w1, w2, λ in the scheduling algorithm to improve the efficiency and accuracy of subsequent task scheduling.

[0063] The information transmission module is used to transmit the task execution order information to each agent to complete the task allocation.

[0064] The information transmission module is implemented by the information sharing mechanism. The information sharing mechanism realizes efficient data sharing and result transmission between intelligent agents through message queues, middleware or data buses, ensuring the timeliness and accuracy of information flow.

[0065] Specifically, the core functions of information sharing include message management, priority scheduling, consistency assurance, and error recovery. The system organizes and manages information through message queues. Each message contains data content, sender ID, receiver ID, and timestamp, which is used to mark the generation time of the message to ensure that the information is processed in order. The message format is unified and supports different types of data transmission and compatibility requirements. The system dynamically schedules messages through message priority queues to ensure that important information is processed first. The system implements many-to-many message distribution through the data bus. The data bus uniformly manages the metadata of the data (such as type, size, and status), and uses status tags to monitor whether the message is successfully delivered and processed. If the message delivery fails or is delayed, the system will mark it as a retransmission state and trigger a retransmission operation. Data In terms of consistency assurance, the system adopts a distributed locking mechanism and version control method to ensure that there will be no conflict when multiple agents concurrently access or modify the same data. For example, when an agent tries to modify the data, the system limits the access rights of other agents through a locking mechanism, and updates the status and version number of the data after the modification is completed. In this way, the system ensures that each access is based on the latest data version; for possible message loss or delay problems, the information sharing mechanism provides error recovery function. When the recipient does not receive the expected message, it will send a retransmission request, and the system will retransmit it according to the status record of the message to ensure that all necessary information can accurately reach the target agent.

[0066] The conflict handling module is used to perform conflict detection after task allocation. After detecting the task conflict between the agents, task coordination and optimization are performed to obtain the optimized task process.

[0067] Among them, the conflict handling module is implemented through a conflict handling mechanism, which solves the contradictions between intelligent agents caused by task requirements, resource competition or decision differences, and ensures the coordination and consistency of the overall decision of the system.

[0068] Specifically, conflict detection includes resource contention conflict detection and logic conflict detection. Resource contention conflict is detected by analyzing the resource requirements of tasks.任务 and the current resource status R 可用 To determine whether it exists, the judgment formula is: ΔR = R 任务 -R 可用 ,When ΔR>0, it means that the current resources are insufficient, which may lead to task conflicts; logical conflicts perform logical analysis on the agent's task decisions. If there are logical contradictions between multiple tasks (such as opposite operation requirements for the same resource), it is determined to be a logical conflict.

[0069] Among them, task coordination after detecting task conflicts between agents includes:

[0070] S21. Perform quantitative evaluation on the detected conflicts.

[0071] S22. Coordinate the conflicting tasks according to a preset resolution strategy; the preset resolution strategy includes a priority strategy, a resource allocation strategy, and a logic rearrangement strategy.

[0072] S23. Verify the results of the coordinated task execution sequence and resource allocation plan.

[0073] Specifically, in S21, the system determines the urgency of the task according to the U 紧急,i and the priority of the task P 优先级,i Calculate the impact weight of conflicting tasks:

[0074] W 冲突,i =w1·U 紧急,i +w2·P 优先级,i

[0075] Among them, w1 and w2 are weight parameters, which are used to reflect the impact of urgency and priority on conflict resolution.

[0076] In S22, the system coordinates the conflicting tasks according to the preset resolution strategy. Common conflict resolution strategies include: priority strategy: prioritize high-priority tasks and postpone or suspend low-priority tasks; resource allocation strategy: dynamically adjust the resource allocation ratio to meet the minimum requirements of conflicting tasks; logic reordering strategy: adjust the logical order of tasks to avoid mutually exclusive operations. The choice of conflict resolution solution is based on the following optimization goals:

[0077]

[0078] Among them, C 冲突,i Represents the cost of conflict resolution, and the system prioritizes the solution with the lowest cost.

[0079] Finally, in S23, after resolving the conflict, the system verifies the coordinated task execution order and resource allocation plan to ensure that the adjusted task execution order and resource allocation plan conform to the overall decision logic. If the verification fails, the conflict resolution method is re-evaluated and adjusted through the feedback mechanism until the conflict is completely eliminated.

[0080] The task execution module is used to enable each intelligent agent to perform tasks according to the optimized task flow, output the data analyzed by each intelligent agent, and make production decisions based on the analyzed data.

[0081] Specifically, the task triggering module and the task scheduling module use the task triggering mechanism and the task scheduling mechanism respectively. After the task is triggered, the tasks are distributed to each agent according to resource adaptability, logical relationship between tasks and load balancing strategy. During the task execution process, the system continuously monitors the execution status of the task. If a task execution failure or abnormal result is detected, the task triggering mechanism will generate supplementary tasks to fix the problem and ensure the integrity of the task chain. Through the collaboration with the task scheduling mechanism and the user intention analysis agent, the task triggering mechanism realizes efficient and reliable task initiation and distribution.

[0082] Specifically, resource adaptability is as follows: the system dynamically evaluates the matching of the resource requirements of the task with the currently available resources. If the resources required by the task significantly exceed the currently available resources of the system, the task will be postponed or decomposed into multiple subtasks to adapt the resources.

[0083] Specifically, the logical relationship between tasks is as follows: the execution order of tasks is controlled by the task scheduling mechanism, and the task triggering mechanism ensures that tasks can be assigned and executed in a logical order by providing dependency information between tasks. For example, if a task depends on the analysis results of other tasks before it can be started, the triggering mechanism needs to wait for the dependent tasks to be completed before executing the current task.

[0084] Specifically, the load balancing strategy is as follows: when the system allocates tasks among multiple agents, it gives priority to agents with lower current loads. The task triggering mechanism monitors the resource usage of the agents in real time to avoid overloading a single agent and causing a decrease in overall efficiency.

[0085] Specifically, message priority is calculated based on urgency and importance:

[0086] P 优先级,i =w1·U 紧急,i +w2·I 重要,i

[0087] Among them, U 紧急,i Indicates the urgency of the message, I 重要,i Indicates the importance of the message. w1 and w2 are weight parameters that can be adjusted according to system requirements.

[0088] In addition, in a preferred embodiment of the present invention, the intelligent agents participating in the system also include: quality prediction and improvement intelligent agent, predictive maintenance intelligent agent, process parameter optimization intelligent agent, report generation intelligent agent, and SQL intelligent agent.

[0089] Among them, the quality prediction and improvement agent receives real-time production data D 实时 , including sensor data, equipment status, process parameters, etc., as well as historical production data D 历史 As input, it is used to generate production optimization suggestions and provide specific guidance for quality improvement in the production process. Its calculation steps include:

[0090] For real-time data 实时 Perform preprocessing and feature extraction to obtain the key feature vector: F 实时 =Φ 实时 (D 实时 ), where Φ 实时 For data preprocessing and feature extraction functions;

[0091] Using AI big model M AI For the eigenvector F 实时 Conduct qualitative analysis and combine historical production data 历史 , predict the quality Q of the product 预测 , the quality prediction model is:

[0092] Q 预测 =M 预测 (F 实时 ,D 历史 ), where M 预测 The quality prediction model is trained based on historical data, which predicts product quality by analyzing real-time data and historical data;

[0093] Combined with the process parameter optimization agent, according to the prediction results Q 预测 and target quality Q 目标 Calculation optimization suggestions G 改进 :

[0094]

[0095] in is the quality loss function, the goal is to minimize the error between the predicted quality and the target quality, and improve the production process by adjusting the process parameters P;

[0096] Based on the improved results G 改进 and the current production status D 实时 , generate production optimization suggestions G 优化 :

[0097] G优化 =M 优化 (G 改进 ,D 实时 ).

[0098] In addition, the predictive maintenance agent receives real-time production data stream D 实时 , including equipment operating parameters, sensor data, and historical maintenance data D 历史 As input, it is used to generate equipment failure probability and maintenance strategy. Based on the predicted failure probability and optimized maintenance strategy, the predictive maintenance agent can provide decision support for the maintenance team, reduce sudden failures and improve equipment operation efficiency. Its calculation steps include:

[0099] The real-time data is preprocessed through the streaming data processing module to obtain the real-time feature vector F 实时 =Φ 流式 (D 实时 );

[0100] Combine multiple prediction models to estimate the probability of equipment failure. First, use the machine learning-based AI model M AI Make a prediction and get the preliminary probability P of equipment failure AI :

[0101] P AI =M AI (F 实时 )

[0102] At the same time, combined with the physical model M 物理 Model the equipment status and obtain the failure probability P predicted by the physical model 物理 :

[0103] P 物理 =M 物理 (F 实时 ,D 历史 )

[0104] Finally, the two prediction results are combined to obtain the equipment failure probability P 故障 :

[0105] P 故障 =α·P AI +(1-α)·P 物理

[0106] Among them, α is the weight coefficient used to balance the two prediction models;

[0107] According to the prediction results P 故障 and threshold θ to determine whether the equipment needs maintenance. If P 故障 >θ, it is judged as equipment failure and a maintenance suggestion G is generated 维护 :

[0108] G 维护 =M 维护 (P 故障 ,D 实时 ), where maintenance recommendations are dynamically adjusted based on factors such as equipment usage status, failure history, and equipment importance;

[0109] Based on the current equipment status and historical data, a maintenance priority list is generated and optimized. Through real-time feedback mechanism and monitoring of equipment health, the intelligent agent can continuously optimize the maintenance strategy and adaptively adjust the maintenance plan according to the load, health status and environmental factors of different equipment.

[0110] The process parameter optimization agent receives real-time production data D 实时 , historical production data D 历史 , and process model M 工艺 As input, it is used to generate optimization suggestions and provide guidance for process optimization in the production process. Its calculation steps include:

[0111] For real-time data 实时 Perform preprocessing and feature extraction to obtain the key feature vector F 实时 =Φ 实时 (D 实时 ), where Φ 实时 For data preprocessing and feature extraction functions;

[0112] Using AI big model M AI For the eigenvector F 实时 Conduct qualitative analysis and combine historical production data 历史 and process model M 工艺 , optimize the process parameters P 优化 , where the optimization objective function of process parameters is:

[0113]

[0114] in, It is a loss function based on real-time data and historical data. The goal is to minimize the loss. By optimizing the objective function, the optimal process parameter P is obtained. 优化 ;

[0115] Based on the process parameter optimization results P 优化 and the current production status D 实时 , generate optimization suggestions G 优化 :

[0116] G 优化 =M 优化 (P 优化 ,D 实时 ).

[0117] Specifically, the SQL agent includes SQL builder, SQL executor and SQL validator. Among them, the SQL builder is responsible for converting the natural language query input by the user into a standard SQL statement. Relying on the natural language processing capability of the big model, the builder can parse the user's query intention, such as "query the latest operation record of a certain device" or "statistical production efficiency of a certain process", and generate the corresponding SQL statement in combination with the table structure definition of the flexible manufacturing database, through context understanding and dynamic adaptation capabilities. The SQL builder supports multi-table joint query, nested query and dynamic generation of complex conditions, so as to meet the diverse needs of the flexible manufacturing field;

[0118] The SQL executor receives the SQL statements generated by the SQL builder and executes query operations in the flexible manufacturing database. The executor is responsible for mapping the query statements to the corresponding data tables and returning the query results. When an abnormal situation (such as syntax errors, table structure changes, or missing data) is encountered during operation, the executor will automatically capture and feedback the error information to prompt the user for possible correction directions. In addition, the SQL executor has built-in query optimization strategies that can dynamically adjust the execution mode according to the complexity of the query statement and the load status of the database to improve query efficiency;

[0119] The SQL validator verifies the syntax and logic of the SQL statement before querying to ensure that the statement conforms to the definition and specifications of the flexible manufacturing database. The validator verifies whether the generated SQL is consistent with the user's query intention. Figure 1 After the query is completed, the validator will also verify the returned results to check the integrity and rationality of the data. With the context understanding ability of the big model, the validator can dynamically adjust the validation rules and improve its adaptability to complex query scenarios.

[0120] Specifically, the table structure definition support of the SQL builder and the data access capability of the SQL executor are provided by the flexible manufacturing database. The flexible manufacturing database is the data storage core of the system, recording the full process data such as production equipment, process flow, quality inspection, etc. The database consists of real-time production data and historical records, and supports dynamic expansion to adapt to the diverse data needs in the flexible manufacturing environment. Through the structured storage and real-time interaction of the database, the flexible manufacturing database ensures the efficient operation of the intelligent body and the accuracy of the data.

[0121] Specifically, the report generation agent dynamically generates data processing workflows to achieve visual display of query data and in-depth data analysis of complex scenarios. Its functions include extracting features from user needs, dynamically generating scripts, executing tool flows, and generating analysis reports. The entire process is implemented through modular design and feedback mechanisms.

[0122] Specifically, in terms of the visualization of query data, the report generation agent extracts the features required for the target chart from the prompt word entered by the user, including the x-axis, y-axis and chart type, and matches the appropriate chart type with the query results of the SQL agent. The system then generates a visualization script for dynamically processing data and generating corresponding charts. During execution, the system checks whether the generated chart meets expectations. If an error is detected, the relevant information is injected into the prompt word, the script is regenerated and the chart generation logic is adjusted.

[0123] Specifically, in terms of in-depth data analysis in complex scenarios, the report generation agent handles multi-step data analysis tasks, such as anomaly detection, trend analysis, and cluster analysis. The system extracts analysis features from user needs, selects appropriate data processing modules, dynamically builds workflows, and executes analysis tasks; the tool flow is generated by model reasoning guided by prompt words to adapt to different analysis needs. During the execution process, the system monitors possible problems in real time and adjusts the generation logic through a feedback mechanism to ensure that the output results meet the analysis requirements. Finally, the system integrates the analysis results and outputs customized reports for in-depth data insights.

[0124] The system architecture diagram of this embodiment is as follows Figure 2 As shown, in summary, this embodiment realizes collaboration among multiple intelligent agents by analyzing user needs, triggering tasks and passing them to various intelligent agents for execution, converting various types of input data into multi-dimensional, real-time in-depth analysis of system decisions, improving data utilization, and being able to quickly adapt to different production scenarios. It realizes a fully automated process from data input to decision output, providing flexible and efficient support for complex production tasks; and through a feedback mechanism, it realizes dynamic optimization and long-term improvement of the task execution process, effectively reducing the error rate and improving the applicability of the system in a variety of scenarios.

[0125] It should be pointed out that, according to the needs of implementation, the various steps / components described in this application can be split into more steps / components, and two or more steps / components or partial operations of steps / components can be combined into new steps / components to achieve the purpose of the present invention.

[0126] The order of execution of each step in the above embodiment does not mean the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

[0127] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all these improvements and changes should fall within the scope of protection of the appended claims of the present invention.

Claims

1. A multi-agent collaborative analysis system based on industrial large models, characterized by: include: The demand analysis module is used to use the user intention analysis agent to analyze the user's input and obtain the user's needs; The task triggering module is used to trigger system task requirements according to user requirements or system anomalies, and distribute tasks to each agent according to the adaptability of resources, the logical relationship between tasks, and the load balancing strategy; The task scheduling module is used to analyze the execution order of tasks by using the central management agent to dynamically plan according to the task requirements, system resource status and logical dependencies between tasks; The information transmission module is used to transmit the task execution order information to each agent to complete the task allocation; The conflict handling module is used to perform conflict detection after task assignment. After detecting the task conflict between agents, task coordination and optimization are performed to obtain the optimized task flow. The task execution module is used to enable each intelligent agent to perform tasks according to the optimized task flow, output the data analyzed by each intelligent agent, and make production decisions based on the analyzed data.

2. According to claim 1, a multi-agent collaborative analysis system based on industrial big model is characterized in that: The trigger system task requirements include user-driven triggering and exception-driven triggering. The user-driven triggering triggers tasks specifically according to user requirements; the exception-driven triggering triggers tasks specifically according to the system's real-time monitoring of the operating status.

3. The multi-agent collaborative analysis system based on industrial big model according to claim 1 is characterized in that: The execution order of the dynamic programming analysis task specifically includes: Receive user requirements and system status information, and describe the task attributes with quantitative parameters; the quantitative parameters include urgency, priority, computing power requirements, and task relevance; Build a task dependency graph to model the tasks; The execution priority of the task is calculated based on the correlation degree of each task in the task dependency graph and the corresponding weight coefficient and the urgency calculation result in the quantitative parameter, and a task execution sequence is generated; the urgency calculation result is calculated based on the task priority, time limit and the corresponding weights of the task priority and time limit.

4. A multi-agent collaborative analysis system based on industrial big model according to claim 1 or 3, characterized in that: After the task is completed, the weight coefficient corresponding to the degree of association of each task in the task dependency graph and the corresponding weight of the task priority and time limit in the urgency calculation result are adjusted through the feedback mechanism according to the task execution time and resource consumption, so as to optimize the execution order of subsequent tasks.

5. The multi-agent collaborative analysis system based on industrial big model according to claim 1 is characterized in that: The conflict detection includes resource competition conflict detection and logic conflict detection; The resource competition conflict detection is specifically: monitoring and analyzing the resource requirements of the task and the current resource status of the system, and when the difference between the resource requirements of the task and the current resource status of the system is less than 0, it is determined that there is a resource competition conflict; The logic conflict detection is specifically as follows: performing a logic analysis on the tasks of each intelligent agent, and determining that a logic conflict exists when there is a logic contradiction between multiple tasks.

6. The multi-agent collaborative analysis system based on industrial big model according to claim 1 is characterized in that: The resource optimization and logic rearrangement after detecting the task conflicts among the agents specifically include: Quantitative assessment of detected conflicts; According to the conflict quantitative assessment results, the conflicting tasks are coordinated according to the preset resolution strategy; the preset resolution strategy includes priority strategy, resource allocation strategy and logic rearrangement strategy; Verify the results of the coordinated task execution sequence and resource allocation plan.

7. The multi-agent collaborative analysis system based on industrial big model according to claim 1 is characterized in that: The intelligent agents include: predictive maintenance intelligent agent, quality prediction and improvement intelligent agent, process parameter optimization intelligent agent, report generation intelligent agent, and SQL intelligent agent.

8. The multi-agent collaborative analysis system based on industrial big model according to claim 7 is characterized in that: The quality prediction and improvement agent receives real-time production data and historical production data as input, and outputs production optimization suggestion data; the real-time production data includes sensor data, equipment status, and process parameters.

9. The multi-agent collaborative analysis system based on industrial big model according to claim 7 is characterized in that: The predictive maintenance agent receives real-time production data stream and historical maintenance data as input, and outputs maintenance suggestion data; the real-time production data stream includes equipment operating parameters and sensor data.

10. The multi-agent collaborative analysis system based on industrial big model according to claim 7 is characterized in that: The process parameter optimization agent receives real-time production data, historical production data and process models as input, and outputs process parameter optimization suggestion data.

Images