Dynamic equipment quality evaluation method based on multi-agent collaborative reinforcement learning

Abstract

The invention relates to the technical field of intelligent equipment evaluation and fault prediction, and discloses a dynamic equipment quality evaluation method based on multi-agent collaborative reinforcement learning, and the method comprises the steps: obtaining original data from equipment and a sensor, and carrying out the preprocessing; analyzing the preprocessed data, and identifying key factors influencing the quality of the equipment; constructing a plurality of expert models, and predicting and evaluating an analysis result; fusing the weight of each expert model to obtain a comprehensive evaluation result; and performing risk assessment on the comprehensive assessment result, extracting key influence factors and decision rules, and generating an assessment result. Efficient fusion of multi-source data and automatic identification of key factors are realized, and the accuracy and reliability of equipment quality evaluation are remarkably improved; a model fusion result is optimized through dynamic weight distribution, so that the prediction performance is improved; the problems that in the prior art, data fusion is difficult, key factor identification depends on experience, and the interpretability of an evaluation result is insufficient are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent device evaluation and fault prediction, and specifically to a dynamic device quality evaluation method based on multi-agent collaborative reinforcement learning. Background Art

[0002] In recent years, with the development of automation and intelligence of industrial equipment, equipment status monitoring and quality assessment technology has become an important part of ensuring the safe and efficient operation of the system. Traditional equipment quality assessment methods mostly rely on single sensor data or fixed rule models to evaluate the equipment status through linear analysis and preset fault thresholds. However, with the increase in equipment complexity, the operating data presents the characteristics of multi-dimensionality, high nonlinearity and dynamic changes. Traditional methods show obvious limitations in dealing with the fusion analysis and nonlinear dynamic changes of multi-source heterogeneous data. In recent years, the rapid development of machine learning technology has provided new solutions for equipment quality assessment, especially the adaptive decision-making ability of reinforcement learning in complex environments, which has enabled it to be initially applied in dynamic quality assessment and anomaly detection. In addition, causal reasoning methods have also been gradually applied to equipment fault tracing and key factor identification, which can effectively improve the interpretability of the model. However, the existing technology still has many problems in dealing with multi-source data fusion, key causal relationship extraction and interpretability of evaluation results.

[0003] The main deficiencies of existing technologies are reflected in the following aspects: First, the fusion analysis of multimodal data makes it difficult to achieve unified modeling of time series data, image data and environmental data, resulting in insufficient evaluation accuracy; Second, in the evaluation process, the identification of key influencing factors and the modeling of causal relationships rely more on manual experience and lack systematic automatic analysis capabilities; Third, existing multi-model fusion methods are usually based on simple averaging or fixed weights, ignoring the differences in confidence levels of models in different scenarios, making it difficult to give full play to the advantages of each model; Fourth, the interpretability of equipment quality evaluation results is poor, and it is difficult for users to clearly understand the main factors affecting equipment quality and the optimization direction from the results. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is: the existing cloud service platform calculation method has large power consumption, load rate rewards and penalties, high cost, and the optimization problem of how to distribute task requests to each host to achieve load balancing.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: a dynamic equipment quality assessment method based on multi-agent collaborative reinforcement learning, comprising:

[0007] Get raw data from devices and sensors and perform preprocessing;

[0008] Analyze the pre-processed data to identify the key factors affecting equipment quality;

[0009] Build multiple expert models to predict and evaluate the analysis results;

[0010] The weights of each expert model are integrated to obtain a comprehensive evaluation result;

[0011] Conduct risk assessment on the comprehensive assessment results, extract key influencing factors and decision-making rules, and generate assessment results.

[0012] As a preferred solution of the dynamic equipment quality assessment method based on multi-agent collaborative reinforcement learning described in the present invention, wherein: the original data includes equipment operation status data, sensor collection data, environmental data, equipment internal data, external condition data, historical performance data and control system data;

[0013] The preprocessing includes data cleaning, feature extraction and standardization of the original data;

[0014] The feature extraction includes extracting time series features, frequency domain features, environmental features, energy consumption load features and historical fault features.

[0015] As a preferred solution of the dynamic equipment quality assessment method based on multi-agent collaborative reinforcement learning described in the present invention, wherein: identifying key factors affecting equipment quality includes analyzing the causal relationship between the time dimension and the feature dimension using a dual attention causal reasoning network;

[0016] The dual attention causal reasoning network includes an input layer, an embedding layer, a time dimension attention module, a feature dimension attention module, a causal reasoning module and an output layer;

[0017] The input layer includes receiving the pre-processed raw data;

[0018] The embedding layer includes performing high-dimensional mapping on the data of the input layer to capture complex feature interactions;

[0019] The time dimension attention module includes, for each time step in the time series, calculating the corresponding attention weight, performing weighted summation on the time series data according to the calculated weight, and obtaining a comprehensive time feature representation;

[0020] The feature dimension attention module includes calculating the attention weight of each feature at the current time step, performing weighted summation on the feature data according to the calculated weight, and obtaining a comprehensive feature representation;

[0021] The causal reasoning module includes constructing a causal graph based on a causal discovery algorithm and using a structural equation model to quantify the strength of causal relationships;

[0022] Through counterfactual analysis, observe the impact of feature changes on equipment quality to verify the reliability of causal relationships;

[0023] The output layer includes the output device quality assessment level, the attention weight distribution in the time dimension, the most influential features and the causal influence strength of each feature, and provides the confidence interval and causal matrix representation of the results.

[0024] As a preferred solution of the dynamic equipment quality assessment method based on multi-agent collaborative reinforcement learning described in the present invention, wherein: constructing multiple expert models, predicting and evaluating the analysis results includes selecting the expert model type, and constructing an independent input feature set for each model;

[0025] Train each model individually, calculate the performance of each model on the validation set, and generate a base confidence score.

[0026] As a preferred solution of the dynamic equipment quality assessment method based on multi-agent collaborative reinforcement learning described in the present invention, wherein: the selection of the expert model type includes, according to the needs of equipment quality assessment, selecting LSTM model, random forest, expert system rule set, convolutional neural network and ARIMA as expert models;

[0027] The constructing of an independent input feature set for each model includes constructing an independent input feature set.

[0028] Based on the characteristics of different models, we design independent input feature sets to ensure that each model can play its best advantages:

[0029] The LSTM input feature set includes continuous time series data such as vibration frequency, temperature, pressure, current, and voltage, and the time window is the past week;

[0030] The input feature set of the random forest includes,extracting statistical features, historical information and current operating status from the,original data;

[0031] The input feature set of the expert system rule set includes selecting temperature, vibration frequency and operating status, and mapping them as inputs of logical rules.

[0032] The convolutional neural network input feature set includes,visible light images and thermal images;

[0033] The ARIMA input feature set includes continuous time series data to capture the long-term trend of equipment operation; the continuous time series data includes current, voltage and speed;

[0034] Train each model individually, calculate the performance of each model on the validation set, and generate basic confidence. This includes independently training each expert model and optimizing model parameters for different types of data sets, evaluating the prediction accuracy, precision, and recall of each model on the validation set, and generating basic confidence for each model.

[0035] As a preferred solution of the dynamic equipment quality assessment method based on multi-agent collaborative reinforcement learning described in the present invention, wherein:

[0036] The weights of the expert models are integrated to obtain a comprehensive evaluation result, including: i , its weight is calculated by the normalization formula, which is expressed as:

[0037]

[0038] The weighted summation formula is used to obtain the comprehensive evaluation result Y, which is expressed as:

[0039]

[0040] Among them, w i represents the weight of the i-th expert model, c i represents the confidence of the i-th expert model, Y represents the comprehensive evaluation result, and y i Represents the prediction result of the i-th expert model.

[0041] As a preferred solution of the dynamic equipment quality assessment method based on multi-agent collaborative reinforcement learning described in the present invention, wherein: generating the assessment result includes obtaining the risk value by using the risk calculation formula according to the equipment quality assessment level and the confidence interval;

[0042] According to the preset risk value threshold, the current status of the equipment is divided into corresponding risk levels;

[0043] From the attention weights in the time and feature dimensions, we select the time steps that account for more than 10% of the total weight as the key influencing factors. The formula is:

[0044]

[0045] Among them, P t Indicates the weight ratio of time step t, A t represents the attention weight at time step t, represents the sum of attention weights of all time steps;

[0046] From the causal matrix, extract the causal relationships with greater impact intensity and identify the main causal paths;

[0047] Generate decision rules based on the results of structural equation modeling and counterfactual analysis;

[0048] Use historical data and counterfactual analysis to verify the effectiveness of the rules and output an evaluation report.

[0049] A dynamic equipment quality assessment system based on multi-agent collaborative reinforcement learning, where:

[0050] Data acquisition and preprocessing module, which acquires raw data from devices and sensors and performs preprocessing;

[0051] Dual attention causal reasoning analysis module, including input layer, embedding layer, time dimension attention module, feature dimension attention module and causal reasoning module;

[0052] Expert model building module, including selecting models such as LSTM, random forest, convolutional neural network, ARIMA and expert system rule set, designing independent input feature sets, and independently training each model to generate basic confidence;

[0053] The model fusion module calculates the weight based on the confidence of each expert model through a normalization formula, and fuses the prediction results of the expert models using a weighted sum formula to generate a comprehensive evaluation result;

[0054] The risk assessment module calculates the risk value based on the comprehensive assessment results and confidence interval, and divides the equipment status into corresponding risk levels according to the preset risk value threshold;

[0055] The key influencing factor extraction module selects time steps that account for more than 10% of the attention weight in the time dimension, and extracts causal relationships with greater influence from the causal matrix to identify the key time periods and main causal paths for equipment quality.

[0056] The decision rule generation and evaluation report output module generates decision rules based on structural equation models and counterfactual analysis, verifies the effectiveness of the rules, and integrates risk levels, key factors, and decision rules to generate evaluation reports.

[0057] A computer device comprises: a memory and a processor; the memory stores a computer program, wherein the processor implements the steps of any one of the methods of the present invention when executing the computer program.

[0058] A computer-readable storage medium stores a computer program, wherein the computer program, when executed by a processor, implements the steps of any one of the methods of the present invention.

[0059] Beneficial effects of the invention: The dynamic equipment quality assessment method based on multi-agent collaborative reinforcement learning provided by the invention realizes efficient fusion of multi-source data and automatic identification of key factors through multi-agent collaborative reinforcement learning combined with a dual-attention causal reasoning network, significantly improving the accuracy and reliability of equipment quality assessment; optimizes model fusion results through dynamic weight allocation to improve prediction performance; and combines causal analysis with counterfactual verification to generate explainable decision rules, providing a scientific basis for equipment management, solving the problems of data fusion difficulties, key factor identification dependence on experience, and insufficient interpretability of evaluation results in the prior art, and has the advantages of strong real-time performance, high adaptability, and precise decision support. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0061] Figure 1 An overall flow chart of a dynamic equipment quality assessment method based on multi-agent collaborative reinforcement learning provided in the first embodiment of the present invention. DETAILED DESCRIPTION

[0062] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0063] Example 1, reference Figure 1 , which is an embodiment of the present invention, provides a dynamic equipment quality assessment method based on multi-agent collaborative reinforcement learning, including:

[0064] S1: Get raw data from devices and sensors and perform preprocessing.

[0065] The original data includes equipment operating status data, sensor collection data, environmental data, equipment internal data, external condition data, historical performance data and control system data.

[0066] The preprocessing includes data cleaning, feature extraction and standardization of the original data;

[0067] The feature extraction includes extracting time series features, frequency domain features, environmental features, energy consumption load features and historical fault features.

[0068] By acquiring multi-dimensional, multi-modal raw data from devices and sensors, and performing cleaning, feature extraction, and standardization, a high-quality data foundation is laid for subsequent analysis. Equipment operating status data, environmental data, historical performance data, etc. cover the dynamic characteristics and environmental impact of equipment operation, and feature extraction (such as time series features, frequency domain features, energy consumption load features, etc.) further mines the key information contained in the data. This preprocessing can effectively remove noise and redundancy in the data, enhance the expressiveness of features, and provide accurate and reliable input for subsequent causal analysis, model building, and comprehensive evaluation, ensuring the scientificity and robustness of the entire evaluation process.

[0069] S2: Analyze the preprocessed data to identify the key factors affecting equipment quality.

[0070] Identifying the key factors affecting device quality involves analyzing the causal relationships in the time and feature dimensions using a dual-attention causal inference network.

[0071] The dual attention causal reasoning network includes an input layer, an embedding layer, a time dimension attention module, a feature dimension attention module, a causal reasoning module and an output layer.

[0072] The input layer includes receiving the pre-processed raw data.

[0073] The embedding layer includes high-dimensional mapping of the data in the input layer to capture complex feature interactions.

[0074] The time dimension attention module includes calculating the corresponding attention weight for each time step in the time series, performing weighted summation on the time series data according to the calculated weight, and obtaining a comprehensive time feature representation.

[0075] The feature dimension attention module includes calculating the attention weight of each feature at the current time step, weighted summing the feature data according to the calculated weight, and obtaining a comprehensive feature representation.

[0076] The causal reasoning module includes constructing a causal graph based on a causal discovery algorithm and using a structural equation model to quantify the strength of the causal relationship.

[0077] Through counterfactual analysis, the impact of feature changes on equipment quality is observed to verify the reliability of causal relationships.

[0078] The output layer includes the output device quality assessment level, the attention weight distribution in the time dimension, the most influential features and the causal influence strength of each feature, and provides the confidence interval and causal matrix representation of the results.

[0079] Furthermore, through the dual attention causal reasoning network, the causal relationship between the time dimension and feature dimension of the data is comprehensively analyzed to accurately identify the key factors affecting the quality of the equipment. The time dimension attention module captures the dynamic contribution of different time steps in the time series to the evaluation results, the feature dimension attention module explores the impact of each feature on the equipment quality at a specific time point, and the causal reasoning module uses causal discovery algorithms and counterfactual analysis to verify the causal relationship between key features and equipment quality. This design not only improves the accuracy and depth of data analysis, but also provides the reliability and interpretability of the evaluation results by outputting the equipment quality evaluation level, attention weight distribution and causal matrix, providing a scientific basis for equipment quality optimization and management.

[0080] S3: Build multiple expert models to predict and evaluate the analysis results.

[0081] Constructing multiple expert models to predict and evaluate the analysis results includes selecting the expert model type and constructing an independent input feature set for each model.

[0082] Train each model individually, calculate the performance of each model on the validation set, and generate a base confidence score.

[0083] The selecting of the expert model type includes selecting LSTM model, random forest, expert system rule set, convolutional neural network and ARIMA as the expert model according to the requirements of equipment quality assessment.

[0084] The constructing of an independent input feature set for each model includes constructing an independent input feature set.

[0085] Based on the characteristics of different models, we design independent input feature sets to ensure that each model can play its best advantages:

[0086] The LSTM input feature set includes continuous time series data such as vibration frequency, temperature, pressure, current, and voltage, and the time window is the past week.

[0087] The random forest input feature set includes statistical features, historical information, and current operating status extracted from the original data.

[0088] The input feature set of the expert system rule set includes selecting temperature, vibration frequency and operating status, and mapping them as inputs of logical rules.

[0089] The convolutional neural network input feature set includes visible light images and thermal images.

[0090] The ARIMA input feature set includes continuous time series data to capture the long-term trend of equipment operation; the continuous time series data includes current, voltage and rotation speed.

[0091] Train each model individually, calculate the performance of each model on the validation set, and generate basic confidence. This includes independently training each expert model and optimizing model parameters for different types of data sets, evaluating the prediction accuracy, precision, and recall of each model on the validation set, and generating basic confidence for each model.

[0092] By building multiple expert models and designing independent input feature sets based on the characteristics of each model, we can fully tap the feature advantages of different types of data to ensure that each model can perform at its best in the data dimension it is good at. For example, LSTM is suitable for capturing dynamic trends in time series data, random forests have strong classification and regression capabilities for multi-dimensional statistical features, convolutional neural networks can extract spatial features from images, and ARIMA is good at long-term trend analysis. In addition, by independently training and verifying each model, the basic confidence is generated in combination with the performance indicators on the verification set, providing a quantitative basis for subsequent model fusion. This multi-model collaborative design improves the robustness and evaluation accuracy of the overall system, and effectively solves the limitations of a single model in a multi-source data environment.

[0093] S4: Integrate the weights of each expert model to obtain a comprehensive evaluation result.

[0094] According to the confidence c of each expert model i , its weight is calculated by the normalization formula, which is expressed as:

[0095]

[0096] The weighted summation formula is used to obtain the comprehensive evaluation result Y, which is expressed as:

[0097]

[0098] Among them, w i represents the weight of the i-th expert model, c i represents the confidence of the i-th expert model, Y represents the comprehensive evaluation result, and y i Represents the prediction result of the i-th expert model.

[0099] Furthermore, by integrating the prediction results of each expert model and using the confidence normalization formula to dynamically assign model weights, the advantages of different models are effectively combined to improve the accuracy and reliability of the evaluation results. The confidence reflects the prediction ability of each model in a specific scenario, avoiding the imbalance of weight distribution caused by simple averaging, making the contribution of high-confidence models greater, and the influence of low-confidence models is appropriately weakened. In addition, this weighted summation method takes into account the diversity of models and the complexity of data, and can more flexibly adapt to the dynamically changing equipment operating environment, thereby achieving more accurate dynamic quality evaluation.

[0100] S5: Conduct risk assessment on the comprehensive assessment results, extract key influencing factors and decision rules, and generate assessment results.

[0101] According to the equipment quality assessment level and confidence interval, the risk value is obtained using the risk calculation formula.

[0102] According to the preset risk value threshold, the current status of the equipment is divided into corresponding risk levels.

[0103] From the attention weights in the time and feature dimensions, we select the time steps that account for more than 10% of the total weight as the key influencing factors. The formula is:

[0104]

[0105] represents the sum of attention weights over all time steps.

[0106] From the causal matrix, extract the causal relationships with greater impact intensity and identify the main causal paths.

[0107] Based on the results of structural equation modeling and counterfactual analysis, decision rules are generated.

[0108] Use historical data and counterfactual analysis to verify the effectiveness of the rules and output an evaluation report.

[0109] Furthermore, through in-depth analysis of the comprehensive evaluation results, accurate quantification of equipment risks and effective extraction of key influencing factors can be achieved, and then scientific decision rules can be generated and evaluation reports can be output. On the one hand, the risk value is calculated based on the equipment quality assessment level and confidence interval, and the risk level is divided according to the preset threshold, which can provide a clear risk warning for the equipment operation status. On the other hand, by analyzing the attention weights and causal matrices of the time and feature dimensions, the key time periods and causal paths that have the greatest impact on equipment quality are identified, which helps to understand the root causes of equipment performance fluctuations. In addition, combining structural equation models and counterfactual analysis to generate actionable decision rules and verify their effectiveness can provide users with targeted optimization suggestions, ultimately achieving intelligent, precise and efficient equipment operation management.

[0110] Embodiment 2 is an embodiment of the present invention, which provides a dynamic equipment quality assessment system based on multi-agent collaborative reinforcement learning, including:

[0111] The data acquisition and preprocessing module acquires raw data from devices and sensors and performs preprocessing.

[0112] The dual attention causal reasoning analysis module includes an input layer, an embedding layer, a time dimension attention module, a feature dimension attention module, and a causal reasoning module.

[0113] Expert model building module, including selecting models such as LSTM, random forest, convolutional neural network, ARIMA and expert system rule set, designing independent input feature sets, and independently training each model to generate basic confidence.

[0114] The model fusion module calculates the weight based on the confidence of each expert model through a normalization formula, and uses the weighted sum formula to fuse the prediction results of the expert models to generate a comprehensive evaluation result.

[0115] The risk assessment module calculates the risk value based on the comprehensive assessment results and confidence interval, and divides the equipment status into corresponding risk levels according to the preset risk value threshold.

[0116] The key influencing factor extraction module selects time steps that account for more than 10% of the attention weight in the time dimension, and extracts causal relationships with greater influence from the causal matrix to identify the key time periods and main causal paths for equipment quality.

[0117] The decision rule generation and evaluation report output module generates decision rules based on structural equation models and counterfactual analysis, verifies the effectiveness of the rules, and integrates risk levels, key factors, and decision rules to generate evaluation reports.

[0118] Embodiment 3, an embodiment of the present invention, is different from the first two embodiments in that:

[0119] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0120] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0121] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0122] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0123] Example 4: Select a set of industrial equipment (such as fan equipment) as the test object, and collect equipment operation data in real time through the sensor network, including vibration frequency, temperature, current, voltage, speed, etc., to simulate the actual operation scenario. The test data includes three equipment states: normal operation, slight abnormality and failure.

[0124] Obtain raw data from devices and sensors, including equipment operating status data, environmental data, and historical performance data, totaling 7 dimensions, with a data frequency of once per second and continuous monitoring for 2 weeks.

[0125] Clean the data to remove outliers and missing values; extract time series features (such as average value, volatility), frequency domain features (such as spectrum power), environmental features (such as temperature and humidity changes), energy load features (such as power consumption fluctuations) and historical fault features.

[0126] After normalization, the data were input into the subsequent model.

[0127] The preprocessed data is analyzed using a dual attention causal inference network. In the time dimension attention module, the weight distribution of key time steps in the equipment operation cycle is extracted; in the feature dimension module, the main influence weights of vibration frequency, temperature and current on equipment quality are identified.

[0128] Combined with the causal reasoning module, a causal graph is constructed to quantify the strength of causal relationships and verify the reliability of the results using counterfactual analysis.

[0129] Construct five expert models: LSTM model, random forest, CNN, ARIMA and expert system rule set, and train each model independently:

[0130] LSTM: Input time series data (vibration frequency, temperature, current, etc.) from the past week.

[0131] Random Forest: Input statistical features and current running status data.

[0132] CNN: Input visible light image and thermal image of the device.

[0133] ARIMA: Capturing long-term trends in equipment operation.

[0134] Expert system rule set: Based on logical rules, analyzes threshold changes of vibration and temperature.

[0135] The precision and recall of each model are evaluated on the validation set, and the base confidence is calculated.

[0136] According to the confidence of the model, the weights are calculated using the normalization formula and weighted summation is performed to generate a comprehensive evaluation result.

[0137] The comprehensive evaluation results output the quality level of the equipment status (such as normal, slightly abnormal, and faulty).

[0138] Based on the comprehensive assessment results and combined with the confidence interval, the equipment risk value is calculated and the risk level is divided.

[0139] From the time and feature attention weights, the time steps with the highest proportion and the key influencing factors are extracted; the main causal paths are extracted from the causal matrix to generate decision rules in the form of "if-then" (such as "when the temperature exceeds 75°C and the vibration amplitude exceeds 0.2g, the risk of equipment failure increases by 20%").

[0140] Output the final evaluation report, including risk level, key influencing factor analysis and optimization suggestions. The experimental results are shown in Table 1.

[0141] Table 1 Experimental data table

[0142]

[0143] The test data table shows that the key parameters of the equipment in different states show significant differences. Under normal conditions, the vibration frequency, temperature and current fluctuate slightly, while under minor abnormalities and fault conditions, these parameters deviate significantly from the normal range.

[0144] Existing equipment evaluation methods are difficult to quickly identify abnormal conditions through a single parameter or simple rules, especially for comprehensive analysis of multi-dimensional data. The present invention can effectively extract key influencing factors in time and feature dimensions through a dual-attention causal reasoning network. For example, in this experiment, the model identified "temperature" and "vibration frequency" as the main influencing factors, and quantified their causal influence on the fault through a causal matrix, making up for the shortcomings of key factor identification in the prior art.

[0145] Evaluation on the validation set showed that the accuracy of the LSTM model was 92%, the random forest was 89%, the CNN was 88%, the ARIMA was 85%, and the fusion model had an accuracy of 96%, which was significantly higher than that of a single model.

[0146] After comprehensively evaluating the results, the present invention further combines risk assessment and decision rule generation to achieve a complete closed-loop process from status assessment to optimization recommendations, which not only improves the accuracy and real-time performance of the assessment, but also provides a scientific basis for equipment maintenance and operation. Compared with the prior art, the present invention significantly enhances the interpretability and practicality of the assessment results, and has significant advantages in industrial equipment quality assessment.

[0147] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A dynamic equipment quality assessment method based on multi-agent collaborative reinforcement learning, characterized in that: include: Get raw data from devices and sensors and perform preprocessing; Analyze the pre-processed data to identify the key factors affecting equipment quality; Build multiple expert models to predict and evaluate the analysis results; The weights of each expert model are integrated to obtain a comprehensive evaluation result; Conduct risk assessment on the comprehensive assessment results, extract key influencing factors and decision-making rules, and generate assessment results.

2. The dynamic equipment quality assessment method based on multi-agent collaborative reinforcement learning as claimed in claim 1, characterized in that: The raw data includes equipment operation status data, sensor collection data, environmental data, equipment internal data, external condition data, historical performance data and control system data; The preprocessing includes data cleaning, feature extraction and standardization of the original data; The feature extraction includes extracting time series features, frequency domain features, environmental features, energy consumption load features and historical fault features.

3. The dynamic equipment quality assessment method based on multi-agent collaborative reinforcement learning as claimed in claim 2, characterized in that: Identifying the key factors that affect device quality includes using a dual-attention causal inference network to analyze the causal relationship between the time dimension and the feature dimension; The dual attention causal reasoning network includes an input layer, an embedding layer, a time dimension attention module, a feature dimension attention module, a causal reasoning module and an output layer; The input layer includes receiving the pre-processed raw data; The embedding layer includes performing high-dimensional mapping on the data of the input layer to capture complex feature interactions; The time dimension attention module includes, for each time step in the time series, calculating the corresponding attention weight, performing weighted summation on the time series data according to the calculated weight, and obtaining a comprehensive time feature representation; The feature dimension attention module includes calculating the attention weight of each feature at the current time step, The feature data is weighted and summed according to the calculated weights to obtain a comprehensive feature representation; The causal reasoning module includes constructing a causal graph based on a causal discovery algorithm and using a structural equation model to quantify the strength of causal relationships; Through counterfactual analysis, observe the impact of feature changes on equipment quality to verify the reliability of causal relationships; The output layer includes the output device quality assessment level, the attention weight distribution in the time dimension, the most influential features and the causal influence strength of each feature, and provides the confidence interval and causal matrix representation of the results.

4. The dynamic equipment quality assessment method based on multi-agent collaborative reinforcement learning as claimed in claim 3, characterized in that: Constructing multiple expert models to predict and evaluate the analysis results includes selecting the expert model type and constructing an independent input feature set for each model; Train each model individually, calculate the performance of each model on the validation set, and generate a base confidence score.

5. The dynamic equipment quality assessment method based on multi-agent collaborative reinforcement learning as claimed in claim 4, characterized in that: The selecting of the expert model type includes selecting LSTM model, random forest, expert system rule set, convolutional neural network and ARIMA as the expert model according to the requirements of equipment quality assessment; The constructing of an independent input feature set for each model includes constructing an independent input feature set. Based on the characteristics of different models, we design independent input feature sets to ensure that each model can play its best advantages: The LSTM input feature set includes continuous time series data such as vibration frequency, temperature, pressure, current, and voltage, and the time window is the past week; The input feature set of the random forest includes,extracting statistical features, historical information and current operating status from the,original data; The input feature set of the expert system rule set includes selecting temperature, vibration frequency and operating status, and mapping them as inputs of logical rules. The convolutional neural network input feature set includes,visible light images and thermal images; The ARIMA input feature set includes,continuous time series data, which captures the long-term trend of equipment operation; The continuous time series data includes current, voltage and rotation speed; Train each model individually, calculate the performance of each model on the validation set, and generate basic confidence. This includes independently training each expert model and optimizing model parameters for different types of data sets, evaluating the prediction accuracy, precision, and recall of each model on the validation set, and generating basic confidence for each model.

6. The dynamic equipment quality assessment method based on multi-agent collaborative reinforcement learning as claimed in claim 5, characterized in that: The weights of the expert models are integrated to obtain a comprehensive evaluation result, including: i , its weight is calculated by the normalization formula, which is expressed as: The weighted summation formula is used to obtain the comprehensive evaluation result Y, which is expressed as: Among them, w i represents the weight of the i-th expert model, c i represents the confidence of the i-th expert model, Y represents the comprehensive evaluation result, and y i Represents the prediction result of the i-th expert model.

7. The dynamic equipment quality assessment method based on multi-agent collaborative reinforcement learning as claimed in claim 6, characterized in that: Generating the assessment results includes obtaining the risk value using the risk calculation formula according to the equipment quality assessment level and the confidence interval; According to the preset risk value threshold, the current status of the equipment is divided into corresponding risk levels; From the attention weights in the time and feature dimensions, we select the time steps that account for more than 10% of the total weight as the key influencing factors. The formula is: Among them, P t Indicates the weight ratio of time step t, A t represents the attention weight at time step t, represents the sum of attention weights of all time steps; From the causal matrix, extract the causal relationships with greater impact intensity and identify the main causal paths; Generate decision rules based on the results of structural equation modeling and counterfactual analysis; Use historical data and counterfactual analysis to verify the effectiveness of the rules and output an evaluation report.

8. A dynamic equipment quality assessment system based on multi-agent collaborative reinforcement learning using the method according to any one of claims 1 to 7, characterized in that: Data acquisition and preprocessing module, which acquires raw data from devices and sensors and performs preprocessing; Dual-attention causal reasoning analysis module, including input layer, embedding layer, time dimension attention module, feature dimension attention module and causal reasoning module; Expert model building module, including selecting models such as LSTM, random forest, convolutional neural network, ARIMA and expert system rule set, designing independent input feature sets, and independently training each model to generate basic confidence; The model fusion module calculates the weight based on the confidence of each expert model through a normalization formula, and fuses the prediction results of the expert models using a weighted sum formula to generate a comprehensive evaluation result; The risk assessment module calculates the risk value based on the comprehensive assessment results and confidence interval, and divides the equipment status into corresponding risk levels according to the preset risk value threshold; The key influencing factor extraction module selects time steps that account for more than 10% of the attention weight in the time dimension, and extracts causal relationships with greater influence from the causal matrix to identify the key time periods and main causal paths for equipment quality. The decision rule generation and evaluation report output module generates decision rules based on structural equation models and counterfactual analysis, verifies the effectiveness of the rules, and integrates risk levels, key factors, and decision rules to generate evaluation reports.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the dynamic equipment quality assessment method based on multi-agent collaborative reinforcement learning as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the dynamic equipment quality assessment method based on multi-agent collaborative reinforcement learning as described in any one of claims 1 to 7 are implemented.

Images