Factory production decision management method and system based on MES system

Abstract

The invention relates to the technical field of production decision management, and discloses a factory production decision management method and system based on an MES (Manufacturing Execution System), and the method comprises the steps: obtaining monitoring data through a sensing module, carrying out the analysis and classification, generating a production monitoring information flow, calling a production decision record, and carrying out the multi-dimensional analysis with an overall production target, so as to carry out the prediction simulation; the method comprises the steps of obtaining a reference decision chain, carrying out preliminary material scheduling and task allocation, analyzing decision computing power requirements, identifying idle computing power time periods, positioning abnormal feedback information, carrying out bidirectional equilibrium analysis, generating an expansion decision chain, constructing an expected decision tree and optimizing production scheduling and task allocation. According to the invention, resource allocation is optimized, the capability of dealing with abnormities of factories is improved, intelligent production management is promoted, and the problem of low efficiency of production decision making in the prior art is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of production decision management, and particularly relates to a factory production decision management method and system based on an MES system. BACKGROUND

[0002] In modern intelligent manufacturing, real-time monitoring of factory equipment and production environment is crucial. Through the deployment of various sensors and monitoring devices, factories can obtain a large amount of production data in real time. These data come from a wide range of sources, including equipment status, production progress, material inventory, environmental conditions, and other aspects. As the complexity of factory production increases, the amount of decision-making calculations also increases, and conventional decision-making methods are difficult to meet the needs of high efficiency. SUMMARY

[0003] The present application aims to provide a factory production decision management method and system based on an MES system, which aims to solve the problem of low efficiency of production decision in the prior art.

[0004] The present application is implemented as follows: in a first aspect, the present application provides a factory production decision management method based on an MES system, comprising: obtaining monitoring data of a factory, and inducing the monitoring data to obtain a production monitoring information flow; calling production decision records and production target information of the factory, and analyzing decision value of the production decision records according to the production monitoring information flow and the production target information, to extract decision value information in the production decision records; based on the decision value information, simulating the development of the production decision records adapted to the production target information to obtain a benchmark decision chain, to preliminarily schedule materials and allocate tasks for factory equipment; determining a number of idle computing power time periods on the benchmark decision chain, and expanding the benchmark decision chain according to the production monitoring information flow in the idle computing power time periods, to construct an expected decision tree to schedule materials and allocate tasks for factory equipment.

[0005] In a second aspect, the present application provides a factory production decision management system based on an MES system, for implementing the factory production decision management method based on an MES system of any one of the first aspect, comprising: an information monitoring module, configured to obtain monitoring data of a factory, and induce the monitoring data to obtain a production monitoring information flow; a decision analysis module, configured to call production decision records and production target information of the factory, and analyze decision value of the production decision records according to the production monitoring information flow and the production target information, to extract decision value information in the production decision records; a benchmark decision module configured to simulate a predicted development of the production decision record based on the decision value information and adapted to the production target information to obtain a benchmark decision chain for preliminary material scheduling and task allocation of the factory equipment; a benchmark decision module configured to simulate a predicted development of the production decision record based on the decision value information and adapted to the production target information to obtain a benchmark decision chain for preliminary material scheduling and task allocation of the factory equipment.

[0006] The present application provides a factory production decision management method based on an MES system, which has the following beneficial effects: The present application obtains monitoring data through a sensing module, analyzes and classifies the data, generates production monitoring information flow, calls production decision records and overall production targets, performs multi-dimensional analysis, extracts advantage and risk decision factors, performs prediction simulation based on the decision factors, obtains a benchmark decision chain and performs preliminary material scheduling and task allocation, analyzes decision computing power requirements, identifies idle computing power time periods, locates abnormal feedback information, performs two-way balanced analysis, generates an expanded decision chain, constructs an expected decision tree, and optimizes production scheduling and task allocation. BRIEF DESCRIPTION OF DRAWINGS

[0007] Figure 1 is a step schematic diagram of a factory production decision management method based on an MES system provided by an embodiment of the present application. Figure 2 is a structure schematic diagram of a factory production decision management system based on an MES system provided by an embodiment of the present application. DETAILED DESCRIPTION

[0008] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0009] The implementation of the present application is described in detail below with reference to specific embodiments.

[0010] Referring to Figure 1 , Figure 2 , a preferred embodiment of the present application is provided.

[0011] In a first aspect, the present application provides a factory production decision management method based on an MES system, comprising: S1: obtaining monitoring data of the factory through a pre-deployed sensing module, and classifying the monitoring data to obtain a production monitoring information flow of the factory; S2: calling production decision records and production target information of the factory, and analyzing decision values of the production decision records according to the production monitoring information flow and the production target information, to extract advantage decision factors and risk decision factors in the production decision records; S3: based on the advantage decision factors and the risk decision factors, performing a prediction development simulation of the production decision records that is adapted to the production target information, to obtain a benchmark decision chain, and performing preliminary material scheduling and task allocation of the factory equipment according to the benchmark decision chain; S4: performing decision algorithm demand analysis on the benchmark decision chain to determine a plurality of idle algorithm time periods on the benchmark decision chain, and performing abnormal feedback information tracing positioning on the production monitoring information flow at a time node corresponding to the idle algorithm time period; S5: based on the benchmark decision chain, performing a two-way balanced analysis of abnormal potential risks and abnormal troubleshooting impacts on the abnormal feedback information, to generate an extended decision chain on the benchmark decision chain, and jointly construct an expected decision tree to perform material scheduling and task allocation of the factory equipment.

[0012] Specifically, in step S1 of the embodiments provided by the present application, different types of sensors (temperature, humidity, pressure, vibration, illumination, etc.) are deployed at various key positions (such as production lines, equipment, warehouses, etc.) in the factory for real-time collection of various monitoring data in the production process. These sensing modules can transmit data to a central data processing system through wireless communication protocols (such as Wi-Fi, LoRa, Zigbee, etc.), ensuring real-time and accuracy of data, and achieving comprehensive multi-source data collection covering the entire production process of the factory, providing a basis for subsequent data analysis and decision-making. Real-time data transmission can ensure timely reflection of various changes in factory production, improving the response speed of the monitoring system.

[0013] More specifically, the data collected from each sensing module is preliminarily aggregated, sorted according to time sequence and category, and preprocessed, including data cleaning, missing value filling, noise filtering, etc., to ensure data quality and effectiveness. Different data sources (such as different types of sensor data) are normalized and standardized for subsequent unified analysis. Data aggregation can centrally manage production data from different sources, ensuring data consistency. The preprocessed data has high accuracy and stability, which is helpful for subsequent data analysis and modeling.

[0014] More specifically, the pre-processed data is parsed to identify the role and impact of the data. Specifically, the state of production equipment, production environment conditions, raw material usage, personnel operation, etc. can be classified in detail, and different data streams can be classified according to the needs of the factory production process, such as distinguishing between equipment state data, environmental data, production progress data, etc. to form different monitoring data streams. Machine learning algorithms such as clustering analysis and classification models are used to further classify the data to ensure that each data stream has clear and traceable characteristics. Through data parsing and classification, the data can be clearly mapped to different production links and resource conditions in the factory, and the classified data stream is more clear and understandable, which can provide strong support for subsequent decision analysis and optimization, avoiding confusion and interference of information.

[0015] More specifically, the parsed and classified data streams are summarized to form a complete factory production monitoring information stream, which not only includes equipment state data, production environment data, etc., but also includes worker operation, production plan execution, etc. Through big data platforms or industrial internet technology, these information streams are transmitted to the central decision system for real-time monitoring, analysis and decision-making. The generated production monitoring information stream provides comprehensive and real-time production data for the factory management layer, which can accurately reflect the current production status. Through a unified monitoring information stream, managers can obtain real-time feedback on production efficiency, equipment health, energy consumption, etc. to make more accurate production decisions.

[0016] Specifically, in step S2 of the embodiments provided by the present application, production decision records are retrieved from the production management system (such as MES system) of the factory. These information usually includes production plan, scheduling decision, resource allocation, equipment maintenance strategy, etc. According to the production target of the factory (such as yield target, quality requirement, cost control, etc.), the corresponding target information is collected, which may come from the long-term planning and short-term tasks of the factory. Combined with real-time production monitoring information stream, the running status of various equipment in the production process, the use of resources, the production environment, etc. are understood, providing comprehensive data of historical production decisions of the factory for subsequent analysis, providing detailed decision records. By obtaining production target information, the long-term strategic goals and short-term production tasks of the factory can be clearly defined to ensure consistency of goals in the analysis process. Through the introduction of real-time production monitoring information stream, real-time changes and interference factors occurring in the production process can be considered in the analysis to ensure the accuracy of the decision analysis.

[0017] More specifically, by combining production decision records, production target information, and production monitoring information flows, multi-dimensional data fusion is performed, key factors in the decision execution process (such as resource consumption, equipment efficiency, production progress, etc.) are compared with the factory's target information (such as yield target, cost control, etc.), and based on multi-dimensional data, the effect of each historical decision is evaluated, for example, by analyzing whether the expected yield is achieved after the implementation of a certain decision, whether the production cost is controlled, whether the equipment failure rate is reduced, etc., to measure its decision value, using data mining techniques (such as cluster analysis, principal component analysis, regression analysis, etc.) to build an optimization model, to analyze the correlation between different decision factors, to identify which factors in the decision process have played a positive role in achieving the target, and which factors may pose risks. Multi-dimensional analysis can comprehensively evaluate the effect of historical production decisions, so that the decision value is not limited to a single dimension (such as cost or yield), but is evaluated from multiple angles. Through data fusion and effect evaluation, it can be identified which factors play the greatest role in the production process and which factors lead to unsatisfactory target achievement, providing a basis for subsequent decision optimization. The optimization model can discover the internal relationship between different decision factors, thereby helping to identify key decision factors that should be given priority in actual production processes.

[0018] More specifically, according to the results of multi-dimensional analysis, decision factors that have played a positive role in historical production decisions are identified, these factors usually effectively improve production efficiency, reduce costs, and ensure quality, etc. Through comparative analysis, decision factors that may lead to production problems, delays, or cost increases are identified, for example, overly tight production plans may lead to high equipment load, resulting in increased failure rates; improper resource scheduling may result in material waste, etc. Advantageous decision factors and risk decision factors are classified, and each factor is assigned a weight. The influence of each factor can be valued based on historical data and actual production results, helping decision-makers prioritize factors with the greatest positive effect or high risk. By extracting advantageous decision factors, clear reference can be provided for future production decisions, helping managers prioritize factors that help improve production efficiency, quality, and reduce costs when formulating new decisions. Identifying and classifying risk decision factors can help factories avoid potential risks in advance, reduce uncertainty and potential losses in the production process, and ensure the smooth implementation of production plans. The weighted factor model can help factories establish a data-driven decision framework, making production scheduling and resource allocation more scientific and accurate.

[0019] More specifically, based on the extracted advantage decision factors and risk decision factors, feedback is given to the production management system to help managers adjust production plans and decision-making strategies. According to the analysis results, specific decision optimization suggestions are proposed, such as adjusting production scheduling plans, optimizing resource allocation, and strengthening equipment maintenance. Reinforcement learning models can be combined to simulate and train historical decisions, and decision-making strategies can be continuously adjusted according to actual production conditions to improve future decision-making effectiveness. The feedback mechanism can timely deliver analysis results to decision-makers to help them make decisions based on historical experience and multi-dimensional analysis results in actual production. Decision optimization suggestions can directly guide specific decisions in the factory production process, improve production efficiency, reduce risks, and improve the achievement rate of overall production goals. Reinforcement learning models can continuously optimize the decision-making process through continuous adaptive adjustment, thereby improving long-term production efficiency and decision-making accuracy.

[0020] Specifically, in step S3 of the embodiments provided by the present application, historical production decision data, including production plans, equipment status, production tasks, resource usage, etc., are sorted and cleaned. These data will be used as input to ensure that the simulation accurately reflects the implementation of decisions in the historical production process. Advantage decision factors and risk decision factors are used to build a prediction model (such as regression analysis, machine learning model, etc.) based on historical data to simulate the development of production processes under different decision-making scenarios. Production target information will be introduced into the model to ensure that the prediction is consistent with the overall goals of the factory (such as yield, quality, cost, etc.).

[0021] More specifically, through the prediction ability of the model, historical decisions are simulated in multiple scenarios to evaluate the results under different decision-making strategies. By comparing the deviation between predicted results and actual results, the model is further adjusted to better predict and optimize future decisions. This step can accurately predict and simulate based on actual historical data, helping to identify possible production trends under different decision-making scenarios. The influence of advantage decision factors and risk decision factors is introduced to ensure that the prediction not only reflects historical data but also considers risks and advantage elements in the production process, thereby improving the accuracy and reliability of the prediction. Through simulation of the decision chain, the production process of the factory can be better understood and optimized to ensure that each decision is executed in line with the overall production goals.

[0022] More specifically, according to the simulation results, the decision chain that highly matches the overall production target of the factory is extracted, which refers to the decision sequence and content of each link in the production process, such as equipment scheduling, production planning, material distribution, and final product testing. By analyzing the prediction results under different scenarios, the best decision path is selected, and the benchmark decision chain is established, which will serve as a reference standard for future production processes, helping decision-makers make adjustments and optimizations. By combining historical data and simulation results, the structure of the decision chain is continuously optimized to maximize the role of advantageous decision factors and minimize the negative impact of risky decision factors.

[0023] More specifically, according to the production process determined in the benchmark decision chain, combined with the production plan and actual inventory situation, material scheduling is carried out. The scheduling system needs to consider the raw materials, tools, and equipment required for each production stage, reasonably allocate materials to ensure that the production process is not delayed due to material shortages, and allocate production tasks according to the availability, capacity, and production requirements of equipment. Intelligent allocation of tasks is achieved using scheduling optimization algorithms such as genetic algorithms, ant colony algorithms, etc., based on factors such as task priority, equipment load, and worker capacity.

[0024] More specifically, during task allocation and material scheduling, the system needs to monitor the status of equipment in real-time to ensure that equipment can complete designated tasks on time. If equipment failure or production line congestion occurs, the system will automatically adjust tasks or material scheduling to maintain production continuity. Through preliminary material scheduling and task allocation, optimal allocation of production resources can be achieved, improving production efficiency and reducing bottlenecks caused by insufficient resources or uneven task allocation. This step provides a preliminary framework for subsequent production scheduling and real-time adjustment, ensuring smooth execution of production plans. Based on the benchmark decision chain, scheduling and allocation can effectively reduce production costs, improve material and equipment utilization, and avoid resource waste and excessive load.

[0025] More specifically, by monitoring factory production data (such as production progress, equipment status, material usage, etc.) in real-time, the system compares the deviation in execution with the prediction results and adjusts production tasks and scheduling plans in a timely manner to ensure consistent alignment with production targets. Based on feedback data, the benchmark decision chain is continuously optimized to improve material scheduling and task allocation strategies. Through machine learning or reinforcement learning techniques, the system continuously optimizes itself in repeated execution, improving the accuracy of prediction and decision-making. The real-time feedback mechanism helps the factory quickly adjust for deviations in production, ensuring the smooth realization of production targets. The feedback and optimization process enables dynamic adjustment of production plans and scheduling, allowing the factory to operate efficiently and consistently in complex and changing production environments. This optimization process promotes the self-improvement of the intelligent decision-making system, making the system increasingly intelligent and more adaptable over time.

[0026] Specifically, in step S4 of the embodiment provided by the present invention, the computational complexity and computational requirements of each decision node in the baseline decision chain are analyzed to determine the computational power requirements. Each decision node needs to perform analysis and calculation based on actual production data (such as equipment status, production tasks, resource scheduling, etc.), which will occupy certain computing resources. By using performance analysis tools, the computational power required by each decision node during processing (such as CPU / GPU utilization, memory usage, storage requirements, etc.) can be quantified. This will help analyze the computational power requirements of each node in the entire decision chain, as well as the load on computing resources at each stage.

[0027] More specifically, based on production plans and historical data, fluctuations in computing demand at different time points can be predicted. For example, some decision-making stages may require more computing power (such as complex material scheduling and optimization calculations), while other stages may have idle computing power. Through load forecasting, idle computing power periods in the system can be identified. Through decision-making computing power demand analysis, a clear understanding of the computing resource requirements of each production stage can be ensured, thereby enabling more efficient planning and allocation of computing power. By quantifying computing power consumption, it is possible to accurately determine which periods of computing resources may be overloaded and which periods may have idle computing power, thus providing a basis for subsequent task allocation.

[0028] More specifically, by utilizing historical production data and analysis of decision-making computing power requirements, a computing power load curve is plotted throughout the entire production process. Through the scheduling of production tasks and analysis of computing power requirements, fluctuations in computing power during certain periods are identified. The computing power load curve reveals which periods have lower computing power loads, representing idle computing power periods. These periods may be times when the computing power demand of decision nodes is low during production, or they may be temporary idle times caused by external factors (such as equipment maintenance, material delivery delays, etc.). During these idle computing power periods, some resources (such as computing resources, memory, etc.) can be allocated to other tasks, especially those requiring higher computing resources, to ensure the overall computing power utilization efficiency of the system. Identifying idle computing power periods enables efficient resource utilization and avoids wasting computing power. Determining idle computing power periods helps to rationally arrange additional work tasks, such as task scheduling, data analysis, and anomaly detection, when computing resources are sufficient, thereby improving the flexibility and responsiveness of the overall production system.

[0029] More specifically, during the production process, monitoring information of various devices and production links (such as temperature, pressure, running state, production progress, etc.) is collected in real time, and real-time monitoring is performed on these information. If the monitoring information is abnormal (for example, device failure, production interruption, etc.), the system will generate abnormal feedback information. According to the generated abnormal feedback information, through the time stamp, device number, decision chain information, etc., the abnormality is associated with the corresponding decision node, and the specific time and link where the abnormality may occur are located. In the idle computing power time period, the generated abnormal feedback information is analyzed in detail. This analysis will involve the causes of abnormal events, such as device failure, material problems, or production plan errors, etc. Through data mining or artificial intelligence models, the root cause of the abnormality is located. Based on the traceability and positioning, appropriate processing strategies are adopted, such as adjusting the production plan, scheduling the device, repairing the fault or adjusting the operation process, etc. to ensure that the abnormality does not affect the production efficiency. Through the abnormal information traceability analysis in the idle computing power time period, the abnormal problems occurring in the production process can be quickly found and located, the production downtime is reduced, the traceability and positioning help to determine the root cause of the abnormality, thereby providing accurate basis for subsequent decision and optimization, avoiding the repeated occurrence of similar problems, and the process can effectively improve the intelligent level of the production line, so that the production system can quickly respond and adjust when an abnormality occurs, thereby ensuring the smooth achievement of the production target.

[0030] More specifically, during the abnormal feedback information traceability process, real-time monitoring is continuously performed to ensure that new abnormalities can be responded to in a timely manner. If it is found that the idle computing power time period is not effectively utilized, the system can dynamically adjust resource allocation, prioritize abnormal feedback analysis tasks, and continuously optimize related nodes and task allocation strategies in the decision chain according to the results of the traceability analysis. For example, if it is found that a certain decision link is prone to abnormality, the decision logic of this link can be optimized in the next round of scheduling to reduce the risk. The system can intelligently learn from the experience of each abnormality handling, gradually improve the accuracy of decision-making computing power demand analysis and abnormality traceability, and realize self-optimization. The real-time feedback and optimization mechanism can ensure that the abnormalities occurring in the production process can be quickly identified and handled, reducing the impact on the overall production efficiency. The ability of the system to continuously optimize and intelligentize can gradually improve the accuracy of decision-making and abnormality handling during a long running process, thereby realizing continuous performance improvement.

[0031] Specifically, in step S5 of the embodiments provided by the present application, the system needs to continuously monitor the equipment and task status of each link in the production process, collect abnormal feedback information in real time, which includes equipment failure, production delay, material shortage, etc., analyze the collected abnormal information, identify potential risk sources, for example, equipment failure may cause production line downtime, material shortage may cause task delay, etc., quantify the impact of each potential risk, including the impact on production tasks, production efficiency, resource consumption, etc., simulate production fluctuations under different scenarios, and evaluate the adverse effects of potential risks on the overall production process.

[0032] More specifically, the investigation process of abnormal feedback information itself may have an impact on the system's computing resources and production tasks, and it is necessary to evaluate whether the investigation process will cause additional resource consumption, production delay or other adverse effects, for example, the investigation process requires inspection or downtime of equipment, which may affect the smooth progress of other tasks, through comprehensive evaluation of the impact of abnormal investigation, determine the equipment, personnel, resource scheduling, etc. that may be involved in the repair process, and then quantify the repair time and the impact on other tasks.

[0033] More specifically, based on the evaluation of abnormal potential risks and abnormal investigation impact, a two-way balance analysis is conducted, the goal is to consider the risks and investigation impact comprehensively, ensure that resources are allocated at the optimal time and area, minimize the negative effects brought by risks and investigation process, according to the results of two-way balance analysis, adjust and optimize the decision strategy, ensure that when an abnormality occurs, resources can be reasonably allocated for rapid investigation, and the normal production process can be restored in the shortest time.

[0034] More specifically, through comprehensive analysis of the potential risks of abnormal feedback information, risk factors that may have a significant impact on the production process can be identified in advance, so that preventive measures or emergency plans can be taken, through analysis of the impact of the investigation process, the resources and time required for abnormal investigation can be accurately evaluated, to avoid excessive interference with the production process and reduce the impact of investigation on the overall production efficiency, two-way balance analysis helps to reasonably allocate resources in the production process, to ensure that when an abnormality occurs, adjustments can be made quickly to minimize losses.

[0035] More specifically, the existing benchmark decision chain is analyzed to understand the functions and interrelationships of each decision-making link. Based on the benchmark decision chain, an exception handling module is added, which should consider how to adjust production plans, scheduling tasks, and device configurations based on exception feedback information. Expanding the decision chain needs to cover decision nodes such as exception response, resource adjustment, and production optimization. According to the exception risk analysis and investigation, a new decision chain is generated to ensure effective resource scheduling and task allocation when exceptions occur. At the same time, the expanded decision chain should support flexible decision paths to adopt different response strategies under different exception situations.

[0036] More specifically, based on the expanded decision chain, an expected decision tree is constructed. The nodes of the decision tree represent different decision options, and the edges represent the interrelationships and possible results between decisions. Each decision node needs to be weighed according to actual production needs, resource allocation, task priority, and other factors. By simulating various possible production exception scenarios (such as device failure, material shortage, etc.), the performance of the decision tree under different conditions is analyzed to find the optimal task scheduling and material allocation strategy. Based on the simulation results, the structure of the decision tree is optimized to minimize redundant decision nodes and ensure that the decision tree can make optimal decisions within a short time.

[0037] More specifically, by generating an expanded decision chain, more flexibility and scalability can be added to the benchmark decision chain, improving the system's response to abnormal situations and reducing the interference of exceptions on the production process. Through the construction of an expected decision tree, the system can intelligently adjust production tasks and material scheduling based on the current production status, exception feedback, and risk assessment, ensuring the optimization of task priority and resource allocation. The application of the decision tree can help the factory make quick decisions when facing unexpected situations, especially in terms of task arrangement, device scheduling, and material replenishment, reducing decision delays and improving production efficiency.

[0038] More specifically, based on the extended decision chain, the material scheduling is carried out: according to the coping strategy of the abnormality in the extended decision chain, the material scheduling plan is adjusted, the material may be short due to equipment failure or production delay, therefore, the system needs to dynamically adjust the material scheduling strategy under the framework of the decision tree, ensure that the material supply can meet the production demand in time, according to the current production status and the urgency of the task, combined with the decision path in the expected decision tree, dynamically adjust the task allocation. If a device or production line has a problem, the system can quickly adjust the task of other devices or production lines to ensure that the production target is not affected, through the optimized material scheduling strategy, the risk of material shortage in production can be reduced, and the smooth progress of the production process can be ensured, the dynamic task allocation can quickly adjust the production arrangement when the emergency occurs, maximize the use of existing resources, reduce the production stagnation time, the overall optimized scheduling and allocation system makes the production line more intelligent and flexible when facing abnormality, which helps to improve the production efficiency and response speed.

[0039] The application provides a factory production decision management method based on an MES system, which has the following beneficial effects: The application obtains monitoring data through a sensing module, analyzes and classifies the monitoring data, generates a production monitoring information flow, calls production decision records and overall production targets, performs multi-dimensional analysis, extracts advantage and risk decision factors, performs prediction simulation based on the decision factors, obtains a benchmark decision chain, and performs preliminary material scheduling and task allocation, analyzes decision algorithm demand, identifies idle algorithm time periods, locates abnormal feedback information, performs two-way balance analysis, generates an extended decision chain, constructs an expected decision tree, and optimizes production scheduling and task allocation. The method can improve the accuracy and flexibility of production decisions, optimize resource allocation, improve the ability of the factory to respond to abnormalities, promote intelligent production management, and solve the problem of low production decision efficiency in the prior art.

[0040] Preferably, the step of obtaining monitoring data of the factory through a pre-deployed sensing module and classifying the monitoring data to obtain a production monitoring information flow of the factory comprises: S11: deploying sensing modules on factory equipment, product monitoring areas, and material storage areas respectively to assign corresponding data characteristic tags according to the deployment positions of the sensing modules; S12: generating a monitoring network composed of various interconnected monitoring nodes according to the data characteristic tags of the various sensing modules, and substituting the sensing monitoring data of the various sensing modules into the corresponding monitoring nodes in the monitoring network to obtain monitoring data of the factory; S13: performing data role analysis of the monitoring data on the material scheduling level, the equipment production level, and the quality detection level according to the monitoring network to obtain the data mapping relationship of each data part in the monitoring data with respect to each analysis level; S14: feedback the monitoring data relative to the data content of each analysis layer based on the data mapping relationship, to obtain the monitoring information of the factory at the material scheduling level, the equipment production level, and the quality detection level; S15: track the product production trajectory of the monitoring data according to the monitoring network, to obtain the product production trajectory feature; wherein the product production trajectory feature is used for the time relationship of the product at each link of material scheduling, equipment production, and quality detection; S16: based on the product production trajectory feature, the monitoring information of the factory at the material scheduling level, the equipment production level, and the quality detection level of each time node is connected, to obtain the production monitoring information flow of the factory.

[0041] Specifically, different sensing modules are deployed at each key position of the factory (such as production line equipment, product monitoring area, and material storage area), which can monitor various production parameters in real time, including temperature, humidity, pressure, vibration, and material flow state, etc. According to the deployment position of the sensing module, a data characteristic label is assigned: according to the position and function of the sensing module, a unique data characteristic label is assigned to each sensing module, for example, a device sensor may be assigned a "device monitoring label", a material storage area sensor may be assigned a "material storage label", and a product monitoring area sensor may be assigned a "product monitoring label". These labels help to distinguish the monitoring data of different areas and enable better mapping to specific levels during data analysis. Through reasonable deployment of sensing modules, real-time data of each link of the factory can be accurately captured, providing comprehensive information for subsequent monitoring and analysis. Each data source is clearly labeled with a characteristic label to ensure accurate identification of data source and characteristics during data processing and analysis, reducing information confusion.

[0042] More specifically, according to the deployment position and data characteristic label of each sensing module, a monitoring network composed of multiple interconnected monitoring nodes is constructed, each node in the monitoring network represents a sensor or a group of sensors, and the function of each node is to receive and process real-time data of the corresponding area. The monitoring data collected by each sensing module is substituted into the corresponding node in the monitoring network, which can include device status, material storage condition, product quality parameter, etc. By assigning data to the correct monitoring node, the accuracy and effectiveness of the data can be ensured. Through the establishment of the monitoring network, the rapid transmission and processing of multi-sensor data can be realized, so that the production information of the factory can flow quickly and accurately to each analysis module. The monitoring network ensures that all data from different sensors can be integrated and processed on a centralized platform, avoiding the problem of data silos.

[0043] More specifically, according to the different areas represented by each node in the monitoring network (such as material scheduling, equipment production, quality detection, etc.), the monitoring data obtained from each sensing module is analyzed, specifically: the material scheduling layer analyzes data related to material flow, storage, demand, etc., identifies information such as the status of material supply, demand changes, etc., the equipment production layer analyzes data related to equipment operation status, efficiency, fault information, etc., understands whether the equipment is running normally, whether the production efficiency meets the expectations, the quality detection layer analyzes information related to product quality detection results, detection equipment data, etc., analyzes whether the product quality meets the standards.

[0044] More specifically, through analysis of the monitoring data, a mapping relationship between each data part and different analysis levels is established, for example, the status change of a certain device may affect the status of material scheduling and quality detection, and effective mapping is needed during analysis. Through data analysis at different levels, comprehensive monitoring of the production process of the factory can be achieved, helping managers to grasp the status changes in the production process in real time and make quick responses. Independent analysis of data at different levels ensures that the details of each production link can be accurately captured and optimized.

[0045] More specifically, according to the mapping relationship established in the previous step, each part of the monitoring data is fed back to each analysis level, for example, in the material scheduling layer, material distribution may need to be adjusted based on production progress; in the equipment production layer, maintenance or replacement may be needed according to the equipment status; in the quality detection layer, unqualified products may need to be removed. According to the above data feedback, monitoring information at each level is generated, for example, the monitoring information at the material scheduling level can include the current material inventory status, future material demand forecast, etc.; the monitoring information at the equipment production level can include the running status of the equipment, maintenance needs, etc.; the monitoring information at the quality detection level can include the product quality pass rate, detection result analysis, etc. Through the data feedback mechanism, the factory can quickly respond to problems at different levels, thereby optimizing material scheduling, equipment maintenance and quality detection. These feedback information provides effective decision support for management personnel, helping to improve production efficiency, reduce resource waste and ensure product quality.

[0046] More specifically, based on monitoring the network and data feedback, the production track of the product is tracked, which includes the whole process of the product from material scheduling, equipment production to quality detection. By tracking the production path of each product, analyzing the time relationship of each link, identifying possible bottlenecks or delays, and based on the production track characteristics of the product, the data feedback of each time node is connected, for example, when the product enters the quality detection link from the production line, the system needs to seamlessly connect the monitoring information of material scheduling, equipment production and quality detection through the time relationship, to ensure smooth and seamless connection of information flow. Through tracking the production track of the product, the time flow of the product in each production link can be comprehensively understood, which helps to analyze the delays or bottlenecks in the production process. Through connection processing, information sharing and cooperation between materials, equipment and quality detection can be ensured, time loss and errors in the production link can be reduced, and production efficiency can be improved.

[0047] More specifically, according to the time relationship and layer feedback, the monitoring information of each link of material scheduling, equipment production and quality detection is integrated in time sequence to generate the production monitoring information flow of the factory. Finally, the data of all links is integrated into a clear production monitoring information flow, which helps managers to master the overall state of production in real time, including material supply, equipment status, quality detection and other information. By generating a complete production monitoring information flow, real-time monitoring data can be provided for factory managers to ensure control of the whole production process. The generated monitoring information flow can provide a key basis for production scheduling and decision-making, helping managers to optimize production processes, schedule resources, and improve quality detection and other links according to real-time data.

[0048] Preferably, the step of calling the production decision record and production target information of the factory, and analyzing the decision value of the production decision record according to the production monitoring information flow and the production target information, to extract the advantage decision factor and risk decision factor in the production decision record, comprises: S21: calling the production decision information executed in the past time through the MES processing platform to jointly serve as the production decision record, and calling the existing production target information; S22: matching processing the production monitoring information flow and the production decision record to construct a time mapping relationship between the production decision record and the production monitoring information flow; wherein the time mapping relationship is used to describe the information part corresponding to the production decision record of a time node in the production monitoring information flow; S23: constructing a multi-dimensional analysis space of the production efficiency dimension, the production quality dimension and the production safety dimension of the production decision record according to the time mapping relationship of the production monitoring information flow, to obtain an original version of the value analysis space; S24: dynamically assigning priority weights to each analysis dimension in the original version of the value analysis space according to the production target information to obtain an advanced version of the value analysis space; S25: extracting decision patterns and evaluating value scores based on the advanced version of the value analysis space to obtain a decision pattern value graph based on the value analysis space; S26: mining strong association rules between decision patterns and high value scores from the decision pattern value graph, and performing variable analysis of decision parameters and value scores on the strong association rules through a pre-trained random forest model to obtain dominant decision factors in the decision patterns; S27: clustering low-value-score decision patterns in the decision pattern value graph to extract common features, and feeding the common features back to the decision pattern value graph for causal inspection processing, while constraining the causal inspection processing based on a pre-constructed production scheduling knowledge graph to obtain risk decision factors in the decision patterns.

[0049] Specifically, the production decision records are retrieved through the MES platform. The MES (Manufacturing Execution System) platform stores the production decision records of the factory, including production planning, resource allocation, production scheduling, equipment usage, personnel configuration, etc. These historical production decision data are retrieved through the MES platform and used as the basis for subsequent analysis. The production target information is the production target set by the factory in a certain period, including production efficiency, product quality, safety production, etc. This information is crucial for setting the decision target in subsequent analysis. By retrieving the MES platform and the overall production target, the production decision records and target information are fully acquired, providing complete data support for subsequent analysis. The retrieved data can be consistent with the subsequent production monitoring information flow, laying a foundation for the accuracy and reliability of data analysis.

[0050] More specifically, the production decision records are matched with the production monitoring information flow. According to the time dimension, the production decision record part corresponding to each time node is identified in the production monitoring information flow. Through the time mapping relationship, it is described how the historical decision at each moment affects the changes of each part in the production monitoring information flow, for example, a certain production decision may affect the equipment running state or quality detection result at a specific time node. By establishing the time mapping relationship between the production decision record and the monitoring information flow, the decision information can be connected with the monitoring data in the production process, ensuring the timeliness and relevance of data analysis, and ensuring that the historical production decision and real-time production monitoring data can be closely related, providing a clear reference framework for subsequent analysis.

[0051] More specifically, based on the time mapping relationship between production decision records and production monitoring information flow, multiple dimensions such as production efficiency, production quality and production safety are analyzed, for example: the production efficiency dimension analyzes the relationship between production decisions and resource use, production time, yield, etc. in the production process, the production quality dimension analyzes the impact of production decisions on product quality, pass rate, defect rate, etc., the production safety dimension analyzes the impact of production decisions on safety production, equipment failure, worker health, etc. A preliminary multi-dimensional analysis space is constructed to evaluate the production value of production decision records. By analyzing production decisions in multiple dimensions, the impact of decisions on various aspects of the production process can be comprehensively evaluated, avoiding focusing on only one dimension, providing a multi-angle analysis framework, which helps to reveal the relationship between different dimensions and obtain more comprehensive production decision information.

[0052] More specifically, based on the production target information, the priority weights of each analysis dimension in the original version of the multi-dimensional analysis space are allocated. Different production targets may have different emphasis on each dimension, so the weights need to be dynamically adjusted. For example, if production efficiency is the focus of the current production target, a higher weight is allocated to the production efficiency dimension in the analysis. According to the production target of the factory and the actual demand at the time, the weights of each analysis dimension are dynamically adjusted so that the analysis results can closely match the current production target. This process is flexible and can be adjusted according to actual production targets and conditions to ensure the accuracy of the analysis.

[0053] More specifically, based on the advanced version of the multi-dimensional analysis space, decision patterns taken by the factory at different historical time points are extracted. Each decision pattern represents a specific production decision behavior and its corresponding production results. By evaluating the value score of each decision pattern and considering multiple dimensions such as production efficiency, quality, safety, etc., the comprehensive benefits of decisions are evaluated. By extracting decision patterns, common decision-making methods of the factory can be identified to provide reference for subsequent optimization. Comprehensive evaluation of decision patterns can provide quantitative evaluation of decision benefits to help management make more scientific decisions.

[0054] More specifically, strong association rules are mined from the decision pattern value graph to find strong associations between decision patterns and high value scores. For example, a certain decision pattern may be highly related to higher production efficiency or lower quality problems. Using a pre-trained random forest model, variable analysis of decision parameters and value scores is performed on these strong association rules. Through model analysis, it is identified which factors play a decisive role in the decision pattern. Through strong association rules and random forest models, key factors affecting decision effectiveness can be found, such as equipment utilization efficiency, personnel scheduling, etc., providing important clues for subsequent decision optimization. The use of random forest models can automatically perform complex decision factor analysis, improving analysis efficiency and reducing human interference.

[0055] More specifically, the low-value score decision patterns in the decision pattern value graph are clustered to identify which decision patterns perform poorly overall and need improvement. Through cluster analysis, common features of these low-value score patterns are extracted to provide a basis for further improvement. The extracted common features are fed back to the decision pattern value graph for causal testing to verify whether these common features are indeed the root cause of inefficient decision-making. Based on the pre-constructed production scheduling knowledge graph, the causal testing is constrained by prior knowledge to ensure logical consistency and rationality of the causal relationship analysis. Through clustering and causal testing, the root cause of inefficient decision-making can be found to provide substantive recommendations for subsequent decision optimization and improvement. Using the production scheduling knowledge graph, experience knowledge and data analysis can be combined to further improve the accuracy and effectiveness of causal testing.

[0056] Preferably, the step of mining strong association rules between decision patterns and high-value scores in the decision pattern value graph, and performing variable analysis of decision parameters and value scores on the strong association rules through a pre-trained random forest model to obtain dominant decision factors in the decision pattern includes: S261: Analyzing the decision pattern value graph in terms of time window and spatial topology to configure dynamic weights based on temporal proximity and device distance relationships for the decision pattern value graph; S262: Mining wide association features of the decision pattern value graph configured with dynamic weights through a reinforcement learning model for the first round to obtain wide association rules between decision patterns and high-value scores fed back by the decision pattern value graph; S263: Configuring a graph mining focus weight for the reinforcement learning model according to the wide association rules to perform subsequent rounds of detailed association feature mining on the decision pattern value graph through the reinforcement learning model to obtain strong association rules between decision patterns and high-value scores fed back by the decision pattern value graph; S264: Analyzing the strong association rules in terms of decision parameter basic elements and performing binary feature encoding on the analyzed decision parameter basic elements as analysis objects input to a pre-trained random forest model; S265: Letting the random forest model reorganize the analysis objects in terms of each decision parameter basic element and predict the value scores to generate a rule influence matrix composed of value feature vectors of each decision parameter specific element; S266: Analyzing the rule influence matrix in terms of decision executability and decision mutual conflict to perform high-value oriented pattern analysis of decision parameter specific elements on the rule influence matrix based on the analysis results to generate a number of dominant decision factors.

[0057] Specifically, according to the time factor in the production process, the decision mode value graph is divided into several time windows, each of which represents a period of production time, analyzes how the relationship between the effect of the decision mode and the value score changes with the change of time, performs spatial topology analysis on the decision mode value graph according to the physical layout of the production equipment and the distance relationship between the equipment, configures dynamic weights based on the distance of the equipment for the decision mode in the graph by considering the distribution and connection relationship of the equipment, and further affects the evaluation of the decision mode. Through the analysis of time and space, different weights can be allocated to each decision mode, so that the analysis of the graph is more in line with the actual production situation. The dynamic weight can reflect the influence of time distance relationship and equipment distance and other factors on decision-making, increase the accuracy of analysis, and through the analysis of time window and spatial topology, a more flexible and practical decision evaluation framework can be provided to improve the authenticity and reliability of the decision-making process.

[0058] More specifically, the decision mode value graph configured with dynamic weights is subjected to first-round correlation feature mining by the reinforcement learning model. This process is broad, that is, potential correlation rules between decision modes and high value scores are found under relatively loose standards. According to the mining results of the first round, the reinforcement learning model configures a focus weight for graph mining of the decision mode, and focuses on analyzing decision features and parameters closely related to high value scores. Through the first round of broad mining, it can be initially found that there is a certain correlation between some decision modes and high value scores, laying a foundation for subsequent refined analysis. By configuring the focus weight, the reinforcement learning model will more accurately guide the subsequent feature mining, improving the efficiency and accuracy of rule mining.

[0059] More specifically, on the basis of the first round of broad correlation rules, the reinforcement learning model performs subsequent rounds of detailed feature mining on the decision mode value graph through the focus weight. At this time, the model will analyze the detailed relationship between each decision mode and high value scores in depth according to the previous findings, so as to extract stronger correlation rules. Through multiple rounds of detailed feature mining, the close correlation between decision modes and high value scores can be more accurately captured, providing strong evidence for the extraction of decision factors. Through the continuous optimization of the reinforcement learning model, more accurate strong correlation rules can be obtained, so as to accurately identify the core decision modes that affect high value scores.

[0060] More specifically, for each decision pattern and value score in the obtained strong association rules, the basic elements of the decision parameters such as production equipment type, production time, resource allocation, etc. are extracted, and these decision parameter basic elements are binary feature coded to facilitate subsequent data processing and analysis, and input into a pre-trained random forest model for further analysis. Through binary feature coding, the decision parameters are converted into a data format that can be directly input into a machine learning model, ensuring consistency in analysis and standardization of data. Binary feature coding can simplify the subsequent analysis process and improve the efficiency of the model in processing large-scale data.

[0061] More specifically, the random forest model reorganizes the binary feature coded decision parameters, analyzes the relationship between each decision parameter basic element and the value score, and predicts the value through multiple decision trees of the random forest. Finally, a rule influence matrix composed of value feature vectors of each decision parameter specific element is generated. The random forest model can handle complex nonlinear relationships between multiple decision parameters, accurately predict value scores under different decision combinations, and clearly show the contribution of different decision parameters to the final decision result, facilitating subsequent analysis.

[0062] More specifically, decision executability analysis is performed on the rule influence matrix to evaluate whether each decision pattern can be executed in actual production and the effect after execution, as well as whether there are conflicts between each decision parameter. For example, certain decision objectives may contradict each other (such as the conflict between improving production efficiency and ensuring product quality). Through executability analysis, decision patterns that are feasible in actual production can be selected, avoiding wasting effort on unfeasible patterns. Through conflict analysis, potential contradictions in the decision-making process can be discovered, helping management make trade-offs and optimizations when making decisions.

[0063] More specifically, based on the results of decision executability and conflict analysis, high-value decisions in the rule influence matrix are analyzed, and the most valuable decision patterns are selected. According to the results of high-value oriented pattern analysis, several advantage decision factors are generated, which represent core elements that have performed well in historical decision-making and can provide guidance for future production decisions. Through high-value oriented analysis, key advantage decision factors can be extracted from complex decision parameters, providing an operable improvement direction for future production decisions. The generated advantage decision factors will directly affect the optimization of subsequent production processes, providing strong data support for improving production efficiency.

[0064] Preferably, the decision mode value graph is subjected to clustering processing of low-value score decision modes to extract common features, and the common features are fed back to the decision mode value graph for causal inspection processing, while the causal inspection processing is subjected to prior knowledge constraint processing based on the pre-constructed production scheduling knowledge graph, to obtain the risk decision factor in the decision mode. S271: The decision mode value graph is subjected to feature decoupling processing of maximizing the similarity of similar samples and minimizing the similarity of different samples between the decision mode and the low-value score, and a time-dependent relationship attention mechanism is introduced to dynamically allocate weights to the feature decoupling processing, to obtain a low-value common feature analysis matrix of the decision mode value graph; S272: The low-value common feature analysis matrix is subjected to identification of potential causal chains by a causal discovery algorithm, and the low-value common feature analysis matrix is subjected to verification of causal effect strength based on the identification result, to construct a data verification driven path; S273: While constructing the data verification driven path, a knowledge verification driven path corresponding to the data verification driven path is generated based on the pre-constructed production scheduling knowledge graph; S274: The potential causal chains in the low-value common feature analysis matrix are subjected to mutual constraint analysis of double confidence based on the knowledge verification driven path and the data verification driven path, to obtain a causal effect network of the low-value common feature analysis matrix; S275: The low-value common feature analysis matrix is subjected to matrix dimension compression based on the causal effect network, to generate a number of risk decision factors.

[0065] Specifically, a deep clustering framework based on contrastive learning is used to process the decision mode value graph. In this process, feature decoupling processing is performed by maximizing the similarity between similar samples and minimizing the similarity between different samples. In other words, the framework optimizes clustering so that the features of the same class of decision mode are more similar, while the features of different classes of decision mode are more different. At the same time, an attention mechanism is introduced to consider the time-dependent relationship of the decision mode, and dynamic weight allocation is used to fully consider the influence of time on the decision mode, improving the sensitivity to time factors in the clustering process. Through the deep clustering framework based on contrastive learning, the low-value score decision mode can be effectively divided into different categories, so that the decision modes within each category have high similarity, thereby helping to more accurately identify low-value score modes. The introduction of time-dependent relationship makes the time-dependent changes of the decision mode be fully captured, so that the time factor is considered when decoupling the features, which is particularly important for time-sensitive decisions such as production scheduling.

[0066] More specifically, through the above clustering results, the common characteristics of low-value score patterns are obtained, and on this basis, a low-value common characteristic analysis matrix is generated, which shows the main characteristics and their weights of low-value score patterns. The introduction of attention mechanism makes the feature weights in the matrix not only reflect static data, but also dynamically respond to changes in time sequence. This step enables the low-value common characteristics of decision patterns to be systematically extracted and represented, revealing the universally existing characteristics in these patterns and providing a basis for subsequent causal analysis. Dynamic weight allocation ensures adaptation to time sequence information, helping to identify potential rules that change over time, thereby providing more timely analysis results for production scheduling.

[0067] More specifically, the low-value common characteristic analysis matrix is analyzed using a causal discovery algorithm, which can reveal potential causal relationships between features and identify potential risk factors in low-value score decision patterns. After identifying potential causal chains, the strength of causal effects is verified based on existing data. This process ensures that the identified causal chains are real and strong, providing a reliable basis for subsequent decision analysis. The causal discovery algorithm can reveal potential causal relationships in decision patterns, identifying factors that may lead to low-value scores and providing scientific evidence for decision improvement. Verifying the strength of causal effects makes the identified causal chains more robust and credible, ensuring their effectiveness for decision analysis.

[0068] More specifically, based on the prior knowledge graph of production scheduling, the data verification driven path is subjected to knowledge constraint processing. Through the prior information in the knowledge graph, the possibility of causal chains can be limited, allowing only those causal relationships that conform to the knowledge of production scheduling to exist. Combining data verification driven paths and knowledge verification driven paths, the potential causal chains in the low-value common characteristic analysis matrix are subjected to mutual constraint analysis of double confidence. This process ensures that causal chains are not only effective in data but also meet the actual needs of production scheduling. By introducing the production scheduling knowledge graph, the causal discovery process is effectively constrained, avoiding the identification of causal chains that do not conform to actual situations and increasing the credibility of analysis results. With the support of the knowledge graph, existing production scheduling experience can be better utilized, improving the accuracy of analysis and the practical application value.

[0069] More specifically, according to the results of the double confidence analysis, a causal effect network of the low-value common feature analysis matrix is constructed, which shows the causal relationship between the features and its strength, helps to identify which causal relationship has important influence on the low-value score decision pattern, dimensionally compresses the low-value common feature analysis matrix based on the causal effect network, filters out the most risky decision factors, reduces the redundant information in the decision analysis, makes the number of decision factors more concise and powerful, through dimensionally compression, reduces the complexity of the causal effect network, makes the analysis more efficient, can focus on processing the most important decision factors, and finally generates the risk decision factors which can effectively identify the potential risks behind the low-value score, and provides accurate decision support for optimizing production scheduling and improving production efficiency.

[0070] Preferably, the step of performing a prediction development simulation of the production decision records based on the advantage decision factors and the risk decision factors and adapted to the production target information to obtain a benchmark decision chain comprises: S31: constructing a spatio-temporal graph attention network of the production decision records according to the advantage decision factors and the risk decision factors, and performing decision scheduling on the spatio-temporal graph attention network through a multi-objective deep reinforcement learning algorithm to generate a plurality of preliminary production decisions; S32: performing digital simulation of the production process of the factory through digital simulation technology to simulate the production process of each type of preliminary production decision towards the production target information to obtain an expected development simulation sequence of each type of preliminary production decision; S33: constructing an original decision tree by combining the expected development simulation sequences of each type of preliminary production decision, and extracting key decision nodes through a decision tree distillation algorithm; S34: performing optimization resource allocation of each key decision node under hybrid scheduling through a quantum annealing algorithm to generate a benchmark decision chain composed of a plurality of optimization decision nodes with optimal effects.

[0071] Specifically, based on the aforementioned identified advantage decision factors and risk decision factors, a spatio-temporal graph attention network is constructed, which can capture the temporal relationship and spatial characteristics in production decision records, thereby providing more accurate background information for subsequent decisions. Through the attention mechanism of the spatio-temporal graph, the importance of decisions can be dynamically adjusted according to the context of the current decision. In certain time periods or specific spatial conditions, certain decision factors may become more important. This mechanism helps to highlight the most critical factors in historical decisions, improving prediction accuracy. Through the spatio-temporal graph attention network, the temporal and spatial characteristics of historical decisions can be better captured, i.e., the changes of decision factors under different time and space conditions. This enhances the flexibility and adaptability of the prediction model. The attention mechanism can automatically adjust the weights of each decision factor under different time and space conditions, thereby helping the model better understand and predict the impact of production decisions.

[0072] More specifically, a multi-objective deep reinforcement learning (DRL) algorithm is used to make scheduling decisions for the spatio-temporal graph attention network. This algorithm can find the most suitable production decisions for different production goals under a multi-objective optimization framework. Through the feedback mechanism of reinforcement learning, the model continuously adjusts to meet the needs of different goals. Through reinforcement learning training, multiple preliminary production decision schemes are generated, each considering historical decisions, production goals, resource constraints, and other factors. Through multi-objective deep reinforcement learning, trade-offs and optimizations can be made among multiple goals to ensure that the generated production decisions can achieve optimal results while meeting multiple production goals. The reinforcement learning algorithm can adjust the decision strategy based on feedback, thereby generating flexible and adaptable preliminary production decisions.

[0073] More specifically, digital simulation technology is used to simulate the production process of the factory and simulate the implementation effects of multiple preliminary production decisions. The simulation process takes into account production goals, resource limitations, equipment conditions, and other factors. The simulation simulates the execution process and effects of each decision scheme in a real environment. For each preliminary production decision, its expected development simulation sequence is generated, which shows the evolution of the production process after executing the decision. Digital simulation can simulate real production processes in a virtual environment, helping decision-makers evaluate the implementation effects of different decision schemes and reduce risks in actual production. Through simulation technology, the impact of different decision schemes on production goals can be visually observed, providing a basis for subsequent decisions.

[0074] More specifically, according to the expected development simulation sequence of various types of preliminary production decision-making, a raw decision tree is constructed, each decision node represents a specific production decision and its corresponding expected development sequence, and the key decision nodes in the raw decision tree are extracted through a decision tree distillation algorithm. The purpose of distillation is to simplify the complex decision tree and extract the most important nodes, thereby reducing the redundant information in the decision path. Through decision tree distillation, the most critical decision nodes can be extracted from the huge decision tree, simplifying the decision chain and improving decision efficiency. The distillation algorithm ensures that only the most important decision nodes for the production target are retained, which helps to reduce the decision error rate and improve the decision quality.

[0075] More specifically, quantum annealing is used to optimize resource allocation for key decision nodes in the decision tree. Quantum annealing can find the optimal resource allocation scheme among multiple decision nodes through the powerful parallel processing capability of quantum computing. Through quantum annealing optimization, a benchmark decision chain composed of decision nodes with optimal effects is finally generated. Quantum annealing algorithm can find the optimal solution among multiple possible decision resource configurations, significantly improving resource allocation efficiency. By optimizing the resource allocation of decision nodes, the benchmark decision chain can maximize the overall production target, reduce resource waste, and improve production efficiency.

[0076] Preferably, the benchmark decision chain is subjected to decision power demand analysis to determine several idle power time periods on the benchmark decision chain, and the step of tracing and positioning the abnormal feedback information of the production monitoring information flow at the time nodes corresponding to the idle power time periods comprises: S41: unit decomposition and power analysis are performed on the benchmark decision chain to obtain several algorithm demand units and corresponding algorithm demand evaluation features; S42: the power dependency relationship between each algorithm demand unit is calculated through a time graph network, and the negative influence analysis of algorithm fluctuation influence is performed on the algorithm demand evaluation feature display of the algorithm demand unit with low power demand based on the power dependency relationship, to determine several idle power time periods according to the analysis result; S43: at the time nodes corresponding to the idle algorithm time periods, the multi-dimensional information features of the production monitoring information flow in the specified range of the past time period are extracted to obtain a multi-dimensional analysis feature matrix; S44: the latent relationship of the multi-dimensional analysis feature matrix is analyzed, and the inverse covariance matrix of the multi-dimensional analysis feature matrix is generated based on the analysis result; S45: the inverse covariance matrix is identified through a graph convolution network model to generate an abnormal information index feature of the multi-dimensional analysis feature matrix; S46: According to the abnormal information index feature, the multi-dimensional analysis feature matrix is located for abnormal data, and the abnormal data in the multi-dimensional analysis feature matrix is associated in the time-space propagation relationship to obtain complete form abnormal feedback information.

[0077] Specifically, the benchmark decision chain is decomposed into multiple smaller algorithm requirement units, each algorithm requirement unit represents the execution requirement of a certain calculation process or module in the decision chain, unit decomposition helps to understand the required computing resources and processing steps of each part in the decision chain, and the computing power of each algorithm requirement unit is analyzed to evaluate its computing resource requirements, including memory, processing capacity and storage, etc. The results of the analysis are used to identify which algorithm units require higher computing power and which are lighter. Through unit decomposition and computing power analysis, the computing power requirements in the benchmark decision chain can be clearly divided, helping subsequent decision-making and optimization of computing power allocation, and the computing resource requirements of each algorithm requirement unit can be accurately evaluated, thereby optimizing the use of computing resources and avoiding waste of computing power.

[0078] More specifically, the computing power dependency relationship between each algorithm requirement unit is analyzed through the time graph network model. The time graph network can describe the dependency relationship and mutual influence of each algorithm unit in the time dimension, thereby constructing a dynamic computing power demand graph. Based on the time graph network, the computing power requirements and dependency relationships between algorithm requirement units are calculated to determine which units have higher computing power resource requirements and which units have lower requirements. The analysis of whether low computing power requirement algorithm units will be affected by algorithm fluctuations, especially the impact on other high demand units in the case of insufficient or unstable computing power, can effectively demonstrate the dependency relationship between algorithm requirement units, helping decision-makers understand the dynamic changes in computing power requirements. By analyzing the negative impact of computing power fluctuations, risks of insufficient or unreasonable allocation of computing power can be identified in advance, and adjustments or optimizations can be made in advance.

[0079] More specifically, according to the computing power dependency relationship analysis results, the time periods with low computing power requirements in the calculation process are identified, especially when the computing requirements of some algorithm requirement units are relatively low. The system can detect these idle computing power time periods as a basis for future computing power resource optimization, and mark these time periods as "idle computing power time periods". In these time periods, other tasks such as production monitoring information flow abnormal feedback information tracing can be performed. Identifying idle computing power time periods helps to achieve reasonable scheduling of resources and ensures that other tasks are processed during idle periods, thereby improving the resource utilization efficiency of the system. Performing other tasks during time periods with low computing power requirements can avoid waste of computing power and improve the computing efficiency of the entire system.

[0080] More specifically, at the time node of the idle computing power time period, the production monitoring information flow in the past time period is extracted for multi-dimensional information features. This step can extract features of multiple dimensions (such as production status, device performance, personnel operation, etc.) from the monitoring information for subsequent anomaly detection. The extracted multi-dimensional features form an analysis feature matrix, providing a basic data structure for further data analysis and anomaly detection. Through multi-dimensional feature extraction, various factors in the production process can be fully reflected, providing rich information for subsequent anomaly detection and analysis. Organizing multi-dimensional features into a feature matrix facilitates further analysis and processing, improving data analysis efficiency.

[0081] More specifically, the multi-dimensional analysis feature matrix is analyzed for potential relationships to find potential correlations between different features. For example, abnormal changes in certain production parameters may affect other related parameters, leading to abnormal behavior. According to the results of potential relationship analysis, an inverse covariance matrix is generated. The inverse covariance matrix can represent the correlation between features, helping to capture and predict potential abnormal behavior. Through potential relationship analysis, the potential relationships between different monitoring parameters can be revealed, providing a deeper perspective for anomaly detection. The inverse covariance matrix provides a mathematical tool to identify patterns of abnormal data, helping to discover potential abnormal behavior earlier.

[0082] More specifically, the inverse covariance matrix is analyzed for abnormal feedback elements using a graph convolution network (GCN) model. GCN can effectively capture the correlation between features from the graph structure, thereby identifying abnormal feedback elements. According to the identified abnormal information index features, the abnormal data in the multi-dimensional analysis feature matrix is located and further analyzed for correlation tracing in the spatio-temporal propagation relationship to obtain abnormal feedback information. Graph convolution network can capture complex spatio-temporal correlations through graph structure, helping to accurately identify and locate abnormal feedback elements. Through the tracing analysis of spatio-temporal propagation relationship, the root cause of abnormal events can be fully traced, providing data support for subsequent improvement.

[0083] Preferably, the abnormal potential risk and abnormal troubleshooting impact of the abnormal feedback information are analyzed in a bidirectional balanced manner based on the benchmark decision chain to generate an expanded decision chain on the benchmark decision chain, and an expected decision tree is constructed to perform material scheduling and task allocation for the factory equipment. S51: Simulate the abnormal potential risk of the abnormal feedback information based on the benchmark decision chain to obtain the potential risk probability characteristics that the abnormal feedback information will bring after continuing to execute the benchmark decision chain; S52: selecting a plurality of abnormality troubleshooting times on the benchmark decision chain, generating an abnormality troubleshooting strategy for each of the abnormality troubleshooting times, and analyzing the positive and negative troubleshooting effects of the strategy execution of the abnormality troubleshooting strategy of each of the abnormality troubleshooting times to obtain abnormality troubleshooting impact characteristics of each of the abnormality troubleshooting times and the corresponding abnormality troubleshooting strategy; S53: performing bidirectional balancing analysis according to the potential risk probability characteristics and the abnormality troubleshooting impact characteristics to determine abnormality troubleshooting time nodes and corresponding abnormality troubleshooting strategies on the benchmark decision chain; S54: predicting abnormal events fed back by the abnormality troubleshooting strategy, and performing subsequent adjustment of the benchmark decision chain based on the abnormal events according to the prediction results to generate an expansion decision chain corresponding to each type of abnormal event at a position corresponding to the abnormality troubleshooting time node, thereby jointly constructing an expected decision tree.

[0084] Specifically, in the benchmark decision chain, the potential risks are simulated and calculated based on the abnormal feedback information, which means that through in-depth analysis of the feedback information, the risk probability that the abnormal feedback may bring in the process of continuing to execute the benchmark decision chain is evaluated, and the probability characteristics of the potential risks are obtained through a risk analysis model, for example, the probability of occurrence of a certain specific abnormal event and the consequences that the event may cause, and then it is identified which risks are the most critical. Through simulation calculation, the risks that may be triggered by abnormal feedback information can be accurately evaluated, helping decision makers to identify potential threats in advance, and through quantifying the risk probability, the prediction ability of potential risks is improved, providing more accurate data support for subsequent decision making.

[0085] More specifically, based on the benchmark decision chain, a plurality of abnormality troubleshooting time nodes are selected, the selection of the troubleshooting time is based on the running state of the system, the characteristics of the abnormal feedback and the resource availability, and the purpose is to troubleshoot the abnormality at the most appropriate time, and appropriate abnormality troubleshooting strategies are formulated for each troubleshooting time node. The troubleshooting strategies may include different detection methods, countermeasures and resource scheduling schemes, and the positive and negative troubleshooting effects of each troubleshooting strategy are analyzed, the positive effect refers to the effective identification and prevention of the troubleshooting strategy on the abnormal event, and the negative effect refers to the negative effects such as resource waste and efficiency loss that the troubleshooting may bring to the system. By selecting appropriate troubleshooting time nodes, the abnormality can be troubleshooted at the appropriate time, avoiding resource waste, and through the positive and negative effect analysis of the troubleshooting strategy, the negative effects in the troubleshooting process can be effectively avoided, thereby improving the troubleshooting efficiency and reducing the potential adverse consequences caused by troubleshooting.

[0086] More specifically, based on the potential risk probability characteristics and the abnormality investigation influence characteristics, a two-way balance analysis is performed, and the purpose of this analysis is to balance the relationship between the potential risk and the abnormality investigation, find the best balance point, pair the possibility of the potential risk with the effect of the investigation strategy, select the most appropriate investigation time node and strategy, and maximize the reduction of risk. Both the possibility of risk occurrence and the effect after the implementation of the investigation strategy are considered, so as to balance between the two and select the optimal scheme. Through the two-way balance analysis, the investigation strategy and the time node in the decision chain can be optimized, so as to ensure risk control while avoiding negative effects in the investigation process. The two-way balance analysis makes the decision more flexible and dynamic, and can adjust the strategy in real time according to the risk and the investigation effect under different conditions.

[0087] More specifically, based on the selected investigation strategy, the abnormal events that may be fed back in the investigation result are predicted. The prediction result can help to judge whether the existing production or equipment state needs to be adjusted. According to the predicted abnormal events, the benchmark decision chain is adjusted. This adjustment can include task reallocation, material scheduling adjustment, or equipment operation mode change, etc. An extended decision chain related to the abnormal events is generated at each abnormality investigation time node. The extended decision chain can add additional decision steps to the benchmark decision chain to handle new abnormal situations. By predicting the investigation result, the state of production or equipment can be adjusted in advance to reduce the impact of abnormal events on production. Based on the prediction of abnormal events, the decision chain can be flexibly adjusted to ensure that each link in the decision tree can make the most appropriate response according to the actual situation. The extended decision chain is generated to provide a more complex decision support system for the factory, which not only relies on the basic decision chain, but also includes the handling and adjustment of abnormal events.

[0088] More specifically, the benchmark decision chain and the extended decision chain are combined to form a complete expected decision tree. The expected decision tree provides a multi-level and multi-dimensional decision framework by considering abnormal feedback information, potential risks, abnormal investigation strategies and their effects. The expected decision tree can provide specific task allocation, material scheduling, equipment operation and other decision basis in the production process to help the factory achieve accurate scheduling and efficient production. The expected decision tree can integrate various types of information to provide a comprehensive decision framework to help decision makers make the best decisions according to the actual situation. The expected decision tree has the ability to dynamically adjust to provide real-time decision support according to different abnormal events and production demands.

[0089] Referring to Figure 2 As shown in FIG. 1, in a second aspect, the present application provides a factory production decision management system based on an MES system, which is used to implement the factory production decision management method based on the MES system in any one of the first aspect. The factory production decision management system based on the MES system comprises: An information monitoring module is configured to acquire monitoring data of the factory through a pre-deployed sensing module, and classify the monitoring data to obtain a production monitoring information flow of the factory. A decision analysis module is configured to call production decision records and production target information of the factory, and analyze decision values of the production decision records according to the production monitoring information flow and the production target information, to extract advantage decision factors and risk decision factors in the production decision records. A benchmark decision module is configured to perform a predicted development simulation on the production decision records based on the advantage decision factors and the risk decision factors, to obtain a benchmark decision chain, and perform preliminary material scheduling and task allocation of the factory equipment according to the benchmark decision chain. An abnormality analysis module is configured to perform decision computing power demand analysis on the benchmark decision chain, to determine a plurality of idle computing power time periods on the benchmark decision chain, and trace and locate abnormal feedback information of the production monitoring information flow at a time node corresponding to the idle computing power time period. An extended decision module is configured to perform bidirectional balance analysis on the abnormal feedback information based on the benchmark decision chain, to generate an extended decision chain on the benchmark decision chain, and jointly construct an expected decision tree to perform material scheduling and task allocation of the factory equipment.

[0090] In the embodiment, the specific implementation of each module in the system embodiment is described above, and will not be repeated here.

[0091] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement and improvement within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A factory production decision management method based on a MES system, characterized in that, include: Acquire factory monitoring data and summarize the monitoring data to obtain a production monitoring information flow; Retrieve the factory's production decision records and production target information, and analyze the decision value of the production decision records based on the production monitoring information flow and the production target information to extract the decision value information from the production decision records; Based on the decision value information, the production decision records are subjected to predictive development simulation adapted to the production target information to obtain a baseline decision chain, so as to perform preliminary material scheduling and task allocation for factory equipment. Several idle computing time periods are determined on the baseline decision chain, and the baseline decision chain is expanded according to the production monitoring information flow during the idle computing time periods to construct a prospective decision tree for material scheduling and task allocation of factory equipment.

2. The factory production decision management method based on the MES system as described in claim 1, characterized in that, The steps of acquiring factory monitoring data and summarizing the monitoring data to obtain a production monitoring information flow include: Collect monitoring data from various areas of the factory and perform data function analysis on the monitoring data to analyze the data content based on the data function of each data part in the monitoring data, so as to obtain the monitoring information of the factory in each process. The monitoring information of each process is connected and processed according to the production trajectory of the product to obtain the production monitoring information flow of the factory.

3. The factory production decision management method based on a MES system as described in claim 1, characterized in that, The steps of retrieving the factory's production decision records and production target information, and analyzing the decision value of the production decision records based on the production monitoring information flow and the production target information to extract decision value information from the production decision records include: The MES processing platform retrieves records of production decisions made in the past, and also retrieves current production target information. The production monitoring information flow and the production decision record are matched to construct a multi-dimensional analysis space of efficiency, quality, and safety of the production decision record based on the matching results, so as to obtain a value analysis space. Based on the production target information, decision-making patterns are extracted and value scores are evaluated in the value analysis space to obtain a value map of decision-making patterns based on the value analysis space. The decision pattern value map is analyzed to mine strong correlation rules between decision patterns and high value scores. Then, a pre-trained random forest model is used to analyze the decision parameters and value scores of the strong correlation rules to obtain decision value information in the decision pattern.

4. The factory production decision management method based on the MES system as described in claim 1, characterized in that, The steps for performing predictive development simulations on the production decision records based on the decision value information to obtain the baseline decision chain include: Based on the decision value information, the production decision records are analyzed to generate several preliminary production decisions; The production process of the factory is digitally simulated using digital simulation technology, so as to arrange the various pre-production decisions into expected development simulation sequences, and to construct an original decision tree based on each expected development simulation sequence. The original decision tree is optimized to obtain the baseline decision chain.

5. The factory production decision management method based on a MES system as described in claim 1, characterized in that, The steps for determining several idle computing power time periods on the benchmark decision chain include: The benchmark decision chain is decomposed into units and analyzed by computing power to obtain several algorithm requirement units and corresponding algorithm requirement evaluation characteristics. The computational power dependency between each algorithm requirement unit is calculated using a temporal graph network. Based on the computational power dependency, the negative impact of algorithm fluctuations on algorithm requirement units that show low computational power requirements in the algorithm requirement assessment characteristics is analyzed, so as to determine several idle computational power time periods based on the analysis results.

6. The factory production decision management method based on a MES system as described in claim 1, characterized in that, The steps of expanding the baseline decision chain based on the production monitoring information flow during idle computing power periods to construct the expected decision tree include: Multi-dimensional information features are extracted from the production monitoring information flow to obtain a multi-dimensional analysis feature matrix and generate the corresponding inverse covariance matrix. The graph convolutional network model is used to identify the abnormal feedback elements of the inverse covariance matrix, so as to locate the abnormal data and trace the correlation of the multi-dimensional analysis feature matrix according to the identification results, so as to obtain abnormal feedback information. Based on the benchmark decision chain, the abnormal feedback information is simulated and calculated to obtain the probability characteristics of the potential risks brought about by the abnormal feedback information after continuing to execute the benchmark decision chain. Several anomaly investigation times are selected on the baseline decision chain, and anomaly investigation strategies are generated for each anomaly investigation time. The positive and negative investigation effects of the anomaly investigation strategies for each anomaly investigation time are analyzed to obtain the anomaly investigation impact characteristics of each anomaly investigation time and the corresponding anomaly investigation strategy. A two-way equilibrium analysis is performed based on the potential risk probability characteristics and the anomaly investigation impact characteristics to determine the anomaly investigation time nodes and corresponding anomaly investigation strategies on the baseline decision chain. The anomaly detection strategy is used to predict the abnormal events fed back by the detection results, and the baseline decision chain is adjusted based on the abnormal events according to the prediction results, so as to generate extended decision chains corresponding to various abnormal events at the positions corresponding to the anomaly detection time nodes, so as to jointly construct the expected decision tree.

7. A factory production decision management system based on a MES system, characterized in that, A factory production decision management method based on a MES system, as described in any one of claims 1-6, includes: The information monitoring module is used to acquire monitoring data from the factory and summarize the monitoring data to obtain a production monitoring information flow. The decision analysis module is used to retrieve the factory's production decision records and production target information, and to analyze the decision value of the production decision records based on the production monitoring information flow and the production target information, so as to extract the decision value information in the production decision records. The baseline decision module is used to perform predictive development simulation on the production decision records based on the decision value information, adapting them to the production target information, to obtain a baseline decision chain for preliminary material scheduling and task allocation of factory equipment. An extended decision-making module is used to perform predictive development simulations on the production decision records based on the decision value information, adapting them to the production target information, to obtain a baseline decision chain for preliminary material scheduling and task allocation of factory equipment.

Images