Method and system for predicting remaining life of capacitor based on machine learning

By synthesizing recursive graphs and graph convolutions and gated cyclic units, the heterogeneous graph integrated neural network is solved by combining genetic simulation annealing algorithm, the problems of low efficiency, insufficient accuracy and slow convergence in the residual life prediction of capacitors are solved, and the accuracy of prediction is improved and model performance.

CN118484680BActive Publication Date: 2024-11-15ZHEJIANG SHUANGFENG ELECTRIC
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410792821.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-06-19
Publication Date
2024-11-15
Estimated Expiration
2044-06-19

AI Technical Summary

Technical Problem

The existing capacitor residual life prediction method is inefficient and the accuracy needs to be improved. Single use of sequence data or recursive graph data leads to a decrease in accuracy and efficiency. The local optimal parameter search capability of traditional annealing algorithms is insufficient and the convergence speed is slow, which affects the overall accuracy of the prediction. .

Method used

The synthetic recursive graph is used to combine the original data for the data generation layer construction, and the heterogeneous graph integrated neural network of the gating cycle unit is used to build the lifetime prediction layer, and the model parameters are optimized through the genetic simulation annealing algorithm.

Benefits of technology

The data dimensions and quality are improved, the breadth and dimensions of feature extraction are optimized, the accuracy, accuracy and scalability of predictions are improved, the diversity and convergence speed of the initial solutions are enhanced, and the overall performance of the model is optimized.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118484680B_ABST
    Figure CN118484680B_ABST
Patent Text Reader

Abstract

The present invention discloses a method and system for predicting the remaining life of a capacitor based on machine learning, which belongs to the technical field of power equipment operation and maintenance. The method includes remaining life prediction modeling, data generation layer construction, life prediction layer construction, model optimization layer construction and capacitor remaining life prediction. The present invention adopts a method of combining a synthetic recursive graph with raw data to construct a data generation layer, and utilizes the advantages of nonlinear feature extraction, dimensionality reduction feature representation, noise resistance, visualization and interpretability of the recursive graph to improve the data dimension and quality; a heterogeneous graph integrated neural network combining graph convolution and gated recurrent units is adopted to optimize the breadth and dimension of feature extraction, and improve the precision, accuracy and scalability of prediction; a genetic simulated annealing algorithm is used to optimize model parameters, enhance the diversity of initial solutions, and generate new candidate solutions through crossover operations, improve the convergence speed and local search capability, and optimize the overall performance of the model.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of power equipment operation and maintenance, and specifically refers to a method and system for predicting the remaining life of a capacitor based on machine learning. Background Art

[0002] Capacitor remaining life prediction is a method that uses machine learning technology to analyze the historical operation data of capacitors to predict the remaining life of capacitors. It aims to predict the failure and degradation time of capacitors in advance, thereby optimizing maintenance strategies, reducing maintenance costs, and further improving equipment reliability and system safety. However, in the existing capacitor remaining life prediction process, there are technical problems that the capacitor degradation process has complex nonlinear characteristics, and the traditional prediction method uses process data for prediction, which is inefficient and the prediction accuracy needs to be improved; there is a technical problem that the single use of sequence data will lead to a decrease in accuracy and efficiency, and the single use of recursive graph data is prone to missing time series features. Although the prediction efficiency is improved, it affects the prediction accuracy, so the graph structure data and time series data are needed for comprehensive feature extraction; there is a technical problem that the prediction of capacitor remaining life prediction has high requirements for prediction accuracy, while the traditional annealing algorithm has insufficient local optimal parameter search capabilities and slow convergence speed, and has a high dependence on the initial solution, resulting in insufficient parameter optimization performance and affecting the overall accuracy of the prediction. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a method and system for predicting the remaining life of a capacitor based on machine learning. In view of the technical problem that in the process of predicting the remaining life of a capacitor, the degradation process of the capacitor has complex nonlinear characteristics, and the traditional prediction method uses process data for prediction, which is inefficient and the prediction accuracy needs to be improved, this scheme creatively uses a synthetic recursive graph combined with original data to construct a data generation layer for predicting the remaining life of a battery. By utilizing the advantages of nonlinear feature extraction, dimensionality reduction feature representation, noise resistance, visualization and interpretability of the recursive graph, the dimension and data quality of the data are improved; in the process of predicting the remaining life of a capacitor, the use of a single sequence data will lead to a decrease in accuracy and efficiency, and the use of a single recursive graph data is prone to missing time series features, which improves the prediction efficiency but affects the prediction accuracy, so a graph structure data is required. In order to solve the technical problem of comprehensive feature extraction from data and time series data, this scheme creatively adopts a heterogeneous graph integrated neural network combining graph convolution and gated recurrent units, and combines it with recursive graph prediction to construct a life prediction layer, which optimizes the breadth and dimension of feature extraction and improves the precision, accuracy and scalability of the prediction; in the process of capacitor remaining life prediction, there is a technical problem that the capacitor remaining life prediction has high requirements for prediction accuracy, while the traditional annealing algorithm has insufficient local optimal parameter search ability and slow convergence speed, and has a high dependence on the initial solution, resulting in insufficient parameter optimization performance and affecting the overall accuracy of the prediction. This scheme creatively adopts a genetic simulated annealing algorithm to optimize the model parameters, and enhances the diversity of the initial solution through genetic algorithm mixing, and generates new candidate solutions through crossover operations, which improves the convergence speed and local search ability and optimizes the overall performance of the model.

[0004] The technical solution adopted by the present invention is as follows: The method for predicting the remaining life of a capacitor based on machine learning provided by the present invention comprises the following steps:

[0005] Step S1: Remaining life prediction modeling;

[0006] Step S2: data generation layer construction;

[0007] Step S3: constructing life prediction layer;

[0008] Step S4: Model optimization layer construction;

[0009] Step S5: predicting the remaining life of the capacitor.

[0010] Furthermore, in step S1, the remaining life prediction modeling is used to design and construct the basic architecture of the capacitor remaining life prediction model, specifically to build a prediction model for the remaining life of the capacitor. The prediction model specifically includes a data generation layer, a life prediction layer and a model optimization layer.

[0011] Furthermore, in step S2, the data generation layer is constructed to construct a capacitor experimental system, collect capacitor experimental data, and perform data preprocessing and synthesis. Specifically, a capacitor charging and discharging experimental system is constructed, and capacitor indicator data of the experimental system is collected. A recursive graph generation method is used to perform data preprocessing and recursive graph data generation to obtain capacitor experimental recursive graph data.

[0012] Further, in step S3, the life prediction layer is constructed to construct a prediction layer of a prediction model, specifically using a heterogeneous graph integrated neural network combining graph convolution and gated recurrent units, and combining recursive graph prediction to construct a life prediction layer to obtain a capacitor remaining life prediction model, including the following steps:

[0013] Step S31: construct a graph convolution layer for extracting spatial features, specifically, stacking two graph convolution sublayers to obtain a graph convolution layer, and extracting features of the capacitor experiment recursive graph data through the graph convolution layer to obtain a convolution feature map;

[0014] The graph convolution sublayer extracts spatial features from graph nodes and their neighboring nodes through graph convolution operations and S-type activation functions;

[0015] Step S32: constructing a gated recurrent unit for extracting time features, specifically extracting features from the capacitor experiment recursive graph data through the gated recurrent unit to obtain a gated feature graph, including the following steps:

[0016] Step S321: construct a reset gate to control the degree of forgetting of the hidden state. The calculation formula is:

[0017] ;

[0018] In the formula, r t is the reset gate output vector, Sig(·) is the S-type activation function, W r is to reset the gate weight, h t-1 is the hidden state of the previous moment, t is the time step index used to represent the current moment, x t is the model input at the current moment, specifically the capacitor experiment recursion graph data, b r is the reset gate bias term;

[0019] Step S322: construct an update gate to control the degree of retention of the hidden state. The calculation formula is:

[0020] ;

[0021] In the formula, z t is the update gate output vector, W z is the update gate weight, b zis the update gate bias term;

[0022] Step S323: Calculate candidate hidden states for updating hidden states. The calculation formula is:

[0023] ;

[0024] In the formula, is the candidate hidden state at the current moment, tanh(·) is the hyperbolic tangent activation function, W h is the hidden state weight, b h is the hidden state bias term;

[0025] Step S324: Calculate the final hidden state, the calculation formula is:

[0026] ;

[0027] In the formula, h t is the final hidden state, specifically the hidden state at the current moment;

[0028] Step S33: heterogeneous graph integration, which is used to integrate the output feature graphs of the graph convolution layer and the gated recurrent unit, specifically, concatenating the convolution feature graph and the gated feature graph in the feature dimension to obtain integrated features;

[0029] Step S34: recursive graph prediction, specifically using a K-means clustering method to cluster the convolutional feature graph and the gated feature graph in the integrated feature, and construct a fully connected layer to perform remaining life prediction;

[0030] Step S35: life prediction model training, specifically, performing model training through the construction of graph convolutional layer, the construction of gated recurrent unit, the heterogeneous graph integration and the recursive graph prediction to obtain the capacitor remaining life prediction model.

[0031] Further, in step S4, the model optimization layer is constructed to optimize model parameters and optimize model performance, specifically through model validation evaluation and iterative training, and using genetic simulated annealing algorithm to optimize model parameters to obtain the optimal remaining life prediction model, including the following steps:

[0032] Step S41: constructing verification data, specifically using the capacitor remaining life prediction model to predict the capacitor remaining life to obtain a prediction data combination, and combining it with the actual capacitor experimental data to verify the effect to obtain a verification data combination;

[0033] Step S42: Model verification, specifically, model verification is performed through accuracy, precision, recall, F score and error rate to obtain model performance reference data;

[0034] Step S43: model iterative training, specifically, performing model iterative training based on the model performance reference data;

[0035] Step S44: Model parameter optimization, specifically using a genetic simulated annealing algorithm to optimize the model parameters, including the following steps:

[0036] Step S441: randomly generate initial model parameters, use the initial model parameters as the initial solution, and use the model loss function as the objective function;

[0037] Step S442: setting the optimal solution as the initial solution, setting the initial temperature, the maximum number of iterations of the inner loop and the minimum temperature, and setting the current temperature as the initial temperature;

[0038] Step S443: Generate a domain solution, specifically generate a candidate solution from the optimal solution, calculate the objective function value of the candidate solution, and calculate the objective function increment. The calculation formula is:

[0039] ;

[0040] In the formula, is the increment of the objective function, which is used to measure the quality of the candidate solution, E(·) is the objective function, X new is a candidate solution, X best is the optimal solution;

[0041] Step S444: updating the optimal solution according to the increment of the objective function, including the following steps:

[0042] Step S4441: When the increment of the objective function is less than 0, the optimal solution is set as the candidate solution, and the calculation formula is:

[0043] ;

[0044] In the formula, is the updated optimal solution;

[0045] Step S4442: When the objective function increment is not less than 0, a new candidate solution is generated through a crossover operation, the objective function value of the new candidate solution is calculated, and the new objective function increment is calculated to update the optimal solution. The calculation formula is:

[0046] ;

[0047] ;

[0048] In the formula, S new is a new candidate solution, crossover(·) is the crossover operation function, is the new objective function increment;

[0049] Step S445: performing inner loop iteration by repeatedly executing the operations described in steps S443 to S444 until the maximum number of inner loop iterations is reached;

[0050] Step S446: Use geometric cooling to update the current temperature. The calculation formula is:

[0051] ;

[0052] Where i is the index of the number of iterations of the outer loop, temp i is the temperature of the i-th outer loop iteration, specifically the current temperature after the update, temp i-1 is the temperature of the i-1th outer loop iteration, specifically the current temperature before the update, and α is the cooling coefficient;

[0053] Step S447: Perform outer loop iteration by repeatedly executing the operations described in step S443 to step S446 until the current temperature is less than the minimum temperature;

[0054] Step S45: Model optimization, specifically, constructing the model optimization layer through the verification data construction, the model verification, the model iterative training and the model parameter optimization to obtain the optimal remaining life prediction model.

[0055] Further, in step S5, the capacitor remaining life prediction is specifically performed by constructing the data generation layer, the life prediction layer and the model optimization layer to predict the capacitor remaining life, and obtaining the capacitor remaining life information by using the optimal remaining life prediction model.

[0056] The capacitor remaining life prediction system based on machine learning provided by the present invention comprises: a remaining life prediction modeling module, a data generation layer construction module, a life prediction layer construction module, a model optimization layer construction module and a capacitor remaining life prediction module;

[0057] The remaining life prediction modeling module is used for remaining life prediction modeling, and the basic architecture of the capacitor remaining life prediction model is obtained through the remaining life prediction modeling, and the data generation layer, the life prediction layer and the model optimization layer are constructed according to the basic architecture of the capacitor remaining life prediction model;

[0058] The data generation layer construction module is used for data generation layer construction, through which capacitor experimental recursive graph data is obtained, and the capacitor experimental recursive graph data is sent to the life prediction layer construction module and the model optimization layer construction module;

[0059] The life prediction layer construction module is used for life prediction layer construction, and obtains the capacitor remaining life prediction model through life prediction layer construction, and sends the obtained capacitor remaining life prediction model to the model optimization layer construction module;

[0060] The model optimization layer construction module is used for model optimization layer construction, and obtains the optimal remaining life prediction model through model optimization layer construction, and sends the optimal remaining life prediction model to the capacitor remaining life prediction module;

[0061] The capacitor remaining life prediction module is used for capacitor remaining life prediction. It predicts the remaining life of the capacitor by constructing the data generation layer, the life prediction layer and the model optimization layer, and obtains the remaining life information of the capacitor by using the optimal remaining life prediction model.

[0062] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0063] (1) In the process of predicting the remaining life of capacitors, the degradation process of capacitors has complex nonlinear characteristics, while the traditional prediction method uses process data for prediction, which is inefficient and has a need to improve the prediction accuracy. This solution creatively uses a synthetic recursive graph combined with raw data to construct a data generation layer for battery remaining life prediction. By taking advantage of the nonlinear feature extraction, dimensionality reduction feature representation, noise resistance, visualization and interpretability of the recursive graph, the data dimension and data quality are improved.

[0064] (2) In the process of predicting the remaining life of capacitors, the use of sequence data alone will lead to a decrease in accuracy and efficiency, while the use of recursive graph data alone is prone to missing time series features. Although it improves the prediction efficiency, it affects the prediction accuracy. Therefore, the technical problem of comprehensive feature extraction of graph structure data and time series data is required. This solution creatively adopts a heterogeneous graph integrated neural network that combines graph convolution and gated recurrent units, and combines recursive graph prediction to construct a life prediction layer, which optimizes the breadth and dimension of feature extraction and improves the precision, accuracy and scalability of prediction.

[0065] (3) In the process of predicting the remaining life of capacitors, there is a technical problem that the prediction accuracy of the remaining life of capacitors is high, while the traditional annealing algorithm has insufficient local optimal parameter search capability and slow convergence speed, and is highly dependent on the initial solution, resulting in insufficient parameter optimization performance and affecting the overall accuracy of the prediction. This scheme creatively uses a genetic simulated annealing algorithm to optimize the model parameters. Through the hybrid of genetic algorithms, the diversity of the initial solution is enhanced, and new candidate solutions are generated through crossover operations, which improves the convergence speed and local search capability and optimizes the overall performance of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 A schematic diagram of the flow chart of the method for predicting the remaining life of a capacitor based on machine learning provided by the present invention;

[0067] Figure 2 A schematic diagram of a capacitor remaining life prediction system based on machine learning provided by the present invention;

[0068] Figure 3 is a schematic flow chart of step S2;

[0069] Figure 4 is a schematic flow chart of step S3;

[0070] Figure 5 It is a schematic diagram of the process of step S4.

[0071] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0072] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0073] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0074] Example 1, see Figure 1 The present invention provides a method for predicting the remaining life of a capacitor based on machine learning, the method comprising the following steps:

[0075] Step S1: Remaining life prediction modeling;

[0076] Step S2: data generation layer construction;

[0077] Step S3: constructing life prediction layer;

[0078] Step S4: Model optimization layer construction;

[0079] Step S5: predicting the remaining life of the capacitor.

[0080] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, the remaining life prediction modeling is used to design and construct the basic architecture of the capacitor remaining life prediction model, specifically to build a prediction model for the remaining life of the capacitor. The prediction model specifically includes a data generation layer, a life prediction layer and a model optimization layer.

[0081] Example 3, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S2, the data generation layer is constructed to construct a capacitor experimental system, collect capacitor experimental data, and perform data preprocessing and synthesis. Specifically, a capacitor charging and discharging experimental system is constructed, and capacitor index data of the experimental system is collected. A recursive graph generation method is used to perform data preprocessing and recursive graph data generation to obtain capacitor experimental recursive graph data, including the following steps:

[0082] Step S21: constructing a capacitor charging and discharging experimental system, specifically initializing the capacitor and the charging and discharging equipment, and setting up an experimental plan to construct the capacitor charging and discharging experimental system;

[0083] The experimental scheme specifically refers to a capacitor charge and discharge experiment with rest, charging operation and discharging operation cycles;

[0084] Step S22: experimental data collection, specifically, carrying out the experimental scheme, and monitoring and recording the experimental data in real time by setting sensors, and constructing an experimental raw data set by repeated experiments, and denoising and removing outliers from the experimental raw data set to obtain an optimized experimental data set;

[0085] Step S23: recursive graph synthesis, including the following steps:

[0086] Step S231: constructing time series data, specifically, arranging the optimization experiment data set into continuous time series data according to the time series to obtain a time series experiment data set;

[0087] Step S232: constructing a recursive graph generation algorithm, specifically, constructing a recursive graph generation algorithm through phase space reconstruction, Euclidean distance calculation and binarization processing;

[0088] The phase space reconstruction is used to select appropriate data embedding dimensions and embedding delay time, and to perform phase space reconstruction on the data in the time series experiment data set to obtain high-dimensional space trajectory data;

[0089] The Euclidean distance calculation is used to calculate the Euclidean distance between the trajectory points in the high-dimensional space trajectory data to generate a distance matrix;

[0090] The binarization process is used to set a binarization threshold and binarize the distance matrix to obtain binarized recursive graph data;

[0091] Step S233: generating image data, specifically converting the binary recursive graph data into an image form, and performing image normalization to obtain image experimental data;

[0092] The image normalization specifically refers to unifying each image data into 256×256 pixels;

[0093] Step S234: data division and labeling, specifically, performing manual data classification and label labeling on the imaged experimental data to obtain divided image data, and dividing the divided image data into training data and test data to obtain capacitor experimental recursive graph data;

[0094] Step S24: data generation, specifically, generating data by constructing the capacitor charging and discharging experimental system, collecting the experimental data, and synthesizing the recursion graph to obtain capacitor experimental recursion graph data;

[0095] By performing the above operations, in order to solve the technical problem that in the process of predicting the remaining life of capacitors, the capacitor degradation process has complex nonlinear characteristics, and the traditional prediction method uses process data for prediction, which is inefficient and the prediction accuracy needs to be improved, this scheme creatively uses a synthetic recursive graph combined with original data to construct a data generation layer for battery remaining life prediction. By utilizing the advantages of nonlinear feature extraction, dimensionality reduction feature representation, noise resistance, visualization and interpretability of the recursive graph, the dimension and data quality of the data are improved.

[0096] Example 4, see Figure 1 and Figure 4 This embodiment is based on the above embodiment. In step S3, the life prediction layer is constructed to construct the prediction layer of the prediction model. Specifically, a heterogeneous graph integrated neural network combining graph convolution and gated recurrent unit is used, and recursive graph prediction is combined to construct the life prediction layer to obtain the capacitor remaining life prediction model, including the following steps:

[0097] Step S31: construct a graph convolution layer for extracting spatial features, specifically, stacking two graph convolution sublayers to obtain a graph convolution layer, and extracting features of the capacitor experiment recursive graph data through the graph convolution layer to obtain a convolution feature map;

[0098] The graph convolution sublayer extracts spatial features from graph nodes and their neighboring nodes through graph convolution operations and S-type activation functions;

[0099] Step S32: constructing a gated recurrent unit for extracting time features, specifically extracting features from the capacitor experiment recursive graph data through the gated recurrent unit to obtain a gated feature graph, including the following steps:

[0100] Step S321: construct a reset gate to control the degree of forgetting of the hidden state. The calculation formula is:

[0101] ;

[0102] In the formula, r t is the reset gate output vector, Sig(·) is the S-type activation function, W r is to reset the gate weight, h t-1 is the hidden state of the previous moment, t is the time step index used to represent the current moment, x t is the model input at the current moment, specifically the capacitor experiment recursion graph data, b r is the reset gate bias term;

[0103] Step S322: construct an update gate to control the degree of retention of the hidden state. The calculation formula is:

[0104] ;

[0105] In the formula, z t is the update gate output vector, W z is the update gate weight, b z is the update gate bias term;

[0106] Step S323: Calculate candidate hidden states for updating hidden states. The calculation formula is:

[0107] ;

[0108] In the formula, is the candidate hidden state at the current moment, tanh(·) is the hyperbolic tangent activation function, W h is the hidden state weight, b h is the hidden state bias term;

[0109] Step S324: Calculate the final hidden state, the calculation formula is:

[0110] ;

[0111] In the formula, h t is the final hidden state, specifically the hidden state at the current moment;

[0112] Step S33: heterogeneous graph integration, which is used to integrate the output feature graphs of the graph convolution layer and the gated recurrent unit, specifically, concatenating the convolution feature graph and the gated feature graph in the feature dimension to obtain integrated features;

[0113] Step S34: recursive graph prediction, specifically using a K-means clustering method to cluster the convolutional feature graph and the gated feature graph in the integrated feature, and construct a fully connected layer to perform remaining life prediction;

[0114] Step S35: life prediction model training, specifically, performing model training through the construction of graph convolution layer, the construction of gated recurrent unit, the heterogeneous graph integration and the recursive graph prediction to obtain a capacitor remaining life prediction model;

[0115] By performing the above operations, in the process of capacitor remaining life prediction, the use of single sequence data will lead to a decrease in accuracy and efficiency, while the use of single recursive graph data is prone to missing time series features. Although it improves the prediction efficiency, it affects the prediction accuracy. Therefore, the technical problem of comprehensive feature extraction of graph structure data and time series data is required. This solution creatively adopts a heterogeneous graph integrated neural network that combines graph convolution and gated recurrent units, and combines it with recursive graph prediction to construct a life prediction layer, which optimizes the breadth and dimension of feature extraction and improves the precision, accuracy and scalability of the prediction.

[0116] Example 5, see Figure 1 and Figure 5 This embodiment is based on the above embodiment. In step S4, the model optimization layer is constructed to optimize model parameters and optimize model performance. Specifically, the model parameters are optimized by model verification evaluation and iterative training, and the genetic simulated annealing algorithm is used to obtain the optimal remaining life prediction model, including the following steps:

[0117] Step S41: constructing verification data, specifically using the capacitor remaining life prediction model to predict the capacitor remaining life to obtain a prediction data combination, and combining it with the actual capacitor experimental data to verify the effect to obtain a verification data combination;

[0118] Step S42: Model verification, specifically, model verification is performed through accuracy, precision, recall, F score and error rate to obtain model performance reference data;

[0119] Step S43: model iterative training, specifically, performing model iterative training based on the model performance reference data;

[0120] Step S44: Model parameter optimization, specifically using a genetic simulated annealing algorithm to optimize the model parameters, including the following steps:

[0121] Step S441: randomly generate initial model parameters, use the initial model parameters as the initial solution, and use the model loss function as the objective function;

[0122] Step S442: setting the optimal solution as the initial solution, setting the initial temperature, the maximum number of iterations of the inner loop and the minimum temperature, and setting the current temperature as the initial temperature;

[0123] Step S443: Generate a domain solution, specifically generate a candidate solution from the optimal solution, calculate the objective function value of the candidate solution, and calculate the objective function increment. The calculation formula is:

[0124] ;

[0125] In the formula, is the increment of the objective function, which is used to measure the quality of the candidate solution, E(·) is the objective function, X new is a candidate solution, X best is the optimal solution;

[0126] Step S444: updating the optimal solution according to the increment of the objective function, including the following steps:

[0127] Step S4441: When the increment of the objective function is less than 0, the optimal solution is set as the candidate solution, and the calculation formula is:

[0128] ;

[0129] In the formula, is the updated optimal solution;

[0130] Step S4442: When the objective function increment is not less than 0, a new candidate solution is generated through a crossover operation, the objective function value of the new candidate solution is calculated, and the new objective function increment is calculated to update the optimal solution. The calculation formula is:

[0131] ;

[0132] ;

[0133] In the formula, S new is a new candidate solution, crossover(·) is the crossover operation function, is the new objective function increment;

[0134] Step S445: performing inner loop iteration by repeatedly executing the operations described in steps S443 to S444 until the maximum number of inner loop iterations is reached;

[0135] Step S446: Use geometric cooling to update the current temperature. The calculation formula is:

[0136] ;

[0137] Where i is the index of the number of iterations of the outer loop, temp i is the temperature of the i-th outer loop iteration, specifically the current temperature after the update, temp i-1 is the temperature of the i-1th outer loop iteration, specifically the current temperature before the update, and α is the cooling coefficient;

[0138] Step S447: Perform outer loop iteration by repeatedly executing the operations described in step S443 to step S446 until the current temperature is less than the minimum temperature;

[0139] Step S45: model optimization, specifically, constructing the model optimization layer through the verification data construction, the model verification, the model iterative training and the model parameter optimization to obtain the optimal remaining life prediction model;

[0140] By executing the above operations, in order to solve the technical problem that in the process of capacitor remaining life prediction, the prediction of capacitor remaining life has high requirements for prediction accuracy, while the traditional annealing algorithm has insufficient local optimal parameter search ability and slow convergence speed, and has high dependence on the initial solution, resulting in insufficient parameter optimization performance and affecting the overall accuracy of the prediction, this scheme creatively adopts the genetic simulated annealing algorithm to optimize the model parameters, enhances the diversity of the initial solution through the hybrid of genetic algorithms, generates new candidate solutions through crossover operations, improves the convergence speed and local search ability, and optimizes the overall performance of the model.

[0141] Example 6, see Figure 1 This embodiment is based on the above embodiment. In step S5, the remaining life of the capacitor is predicted by constructing the data generation layer, the life prediction layer and the model optimization layer to predict the remaining life of the capacitor, and the remaining life information of the capacitor is obtained by using the optimal remaining life prediction model.

[0142] Embodiment 7, see Figure 2 , this embodiment is based on the above embodiment, and the capacitor remaining life prediction system based on machine learning provided by the present invention includes: a remaining life prediction modeling module, a data generation layer construction module, a life prediction layer construction module, a model optimization layer construction module and a capacitor remaining life prediction module;

[0143] The remaining life prediction modeling module is used for remaining life prediction modeling, and the basic architecture of the capacitor remaining life prediction model is obtained through the remaining life prediction modeling, and the data generation layer, the life prediction layer and the model optimization layer are constructed according to the basic architecture of the capacitor remaining life prediction model;

[0144] The data generation layer construction module is used for data generation layer construction, through which capacitor experimental recursive graph data is obtained, and the capacitor experimental recursive graph data is sent to the life prediction layer construction module and the model optimization layer construction module;

[0145] The life prediction layer construction module is used for life prediction layer construction, and obtains the capacitor remaining life prediction model through life prediction layer construction, and sends the obtained capacitor remaining life prediction model to the model optimization layer construction module;

[0146] The model optimization layer construction module is used for model optimization layer construction, and obtains the optimal remaining life prediction model through model optimization layer construction, and sends the optimal remaining life prediction model to the capacitor remaining life prediction module;

[0147] The capacitor remaining life prediction module is used for capacitor remaining life prediction. It predicts the remaining life of the capacitor by constructing the data generation layer, the life prediction layer and the model optimization layer, and obtains the remaining life information of the capacitor by using the optimal remaining life prediction model.

[0148] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0149] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0150] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A method for predicting the remaining life of a capacitor based on machine learning, characterized in that: The method comprises the following steps: Step S1: Remaining life prediction modeling, used to design and build a basic framework of a capacitor remaining life prediction model; Step S2: constructing a data generation layer, which is used to construct a capacitor experimental system, collect capacitor experimental data, and perform data preprocessing and synthesis; Step S3: constructing a life prediction layer, which is used to construct a prediction layer of a prediction model, specifically using a heterogeneous graph integrated neural network combining graph convolution and gated recurrent units, and combining recursive graph prediction to construct a life prediction layer, and obtain a capacitor remaining life prediction model; The life prediction layer construction includes the following steps: Step S31: constructing a graph convolution layer for extracting spatial features, specifically, stacking two graph convolution sublayers to obtain a graph convolution layer, and extracting features of the capacitor experiment recursive graph data through the graph convolution layer to obtain a convolution feature map; The graph convolution sublayer extracts spatial features from graph nodes and their neighboring nodes through graph convolution operations and S-type activation functions; Step S32: constructing a gated recurrent unit for extracting time features, specifically extracting features from the capacitor experiment recursive graph data through the gated recurrent unit to obtain a gated feature graph, including the following steps: Step S321: construct a reset gate to control the degree of forgetting of the hidden state. The calculation formula is: ; In the formula, r t is the reset gate output vector, Sig(·) is the S-type activation function, W r is to reset the gate weight, h t-1 is the hidden state of the previous moment, t is the time step index used to represent the current moment, x t is the model input at the current moment, specifically the capacitor experiment recursion graph data, b r is the reset gate bias term; Step S322: construct an update gate to control the degree of retention of the hidden state. The calculation formula is: ; In the formula, z t is the update gate output vector, W z is the update gate weight, b z is the update gate bias term; Step S323: Calculate candidate hidden states for updating hidden states. The calculation formula is: ; In the formula, is the candidate hidden state at the current moment, tanh(·) is the hyperbolic tangent activation function, W h is the hidden state weight, b h is the hidden state bias term; Step S324: Calculate the final hidden state, the calculation formula is: ; In the formula, h t is the final hidden state, specifically the hidden state at the current moment; Step S33: heterogeneous graph integration, which is used to integrate the output feature graphs of the graph convolution layer and the gated recurrent unit, specifically, concatenating the convolution feature graph and the gated feature graph in the feature dimension to obtain integrated features; Step S34: recursive graph prediction, specifically using a K-means clustering method to cluster the convolutional feature graph and the gated feature graph in the integrated feature, and construct a fully connected layer to perform remaining life prediction; Step S35: life prediction model training, specifically, performing model training through the construction of graph convolution layer, the construction of gated recurrent unit, the heterogeneous graph integration and the recursive graph prediction to obtain a capacitor remaining life prediction model; Step S4: constructing a model optimization layer, which is used to optimize model parameters and optimize model performance, specifically through model validation evaluation and iterative training, and using a genetic simulated annealing algorithm to optimize model parameters, to obtain an optimal remaining life prediction model; Step S5: predicting the remaining life of the capacitor.

2. The method for predicting the remaining life of a capacitor based on machine learning according to claim 1, characterized in that: In step S4, the model optimization layer is constructed, including the following steps: Step S41: constructing verification data, specifically using the capacitor remaining life prediction model to predict the capacitor remaining life to obtain a prediction data combination, and combining it with the actual capacitor experimental data to verify the effect to obtain a verification data combination; Step S42: Model verification, specifically, model verification is performed through accuracy, precision, recall, F score and error rate to obtain model performance reference data; Step S43: model iterative training, specifically, performing model iterative training based on the model performance reference data; Step S44: optimizing model parameters, specifically using a genetic simulated annealing algorithm to optimize model parameters; Step S45: Model optimization, specifically, constructing the model optimization layer through the verification data construction, the model verification, the model iterative training and the model parameter optimization to obtain the optimal remaining life prediction model.

3. The method for predicting the remaining life of a capacitor based on machine learning according to claim 2, characterized in that: In step S44, the model parameter optimization is specifically performed using a genetic simulated annealing algorithm, which includes the following steps: Step S441: randomly generate initial model parameters, use the initial model parameters as the initial solution, and use the model loss function as the objective function; Step S442: setting the optimal solution as the initial solution, setting the initial temperature, the maximum number of iterations of the inner loop and the minimum temperature, and setting the current temperature as the initial temperature; Step S443: Generate a domain solution, specifically generate a candidate solution from the optimal solution, calculate the objective function value of the candidate solution, and calculate the objective function increment. The calculation formula is: ; In the formula, is the increment of the objective function, which is used to measure the quality of the candidate solution, E(·) is the objective function, X new is a candidate solution, X best is the optimal solution; Step S444: updating the optimal solution according to the increment of the objective function, including the following steps: Step S4441: When the increment of the objective function is less than 0, the optimal solution is set as the candidate solution, and the calculation formula is: ; In the formula, is the updated optimal solution; Step S4442: When the objective function increment is not less than 0, a new candidate solution is generated through a crossover operation, the objective function value of the new candidate solution is calculated, and the new objective function increment is calculated to update the optimal solution. The calculation formula is: ; ; In the formula, S new is a new candidate solution, crossover(·) is the crossover operation function, is the new objective function increment; Step S445: performing inner loop iteration by repeatedly executing the operations described in steps S443 to S444 until the maximum number of inner loop iterations is reached; Step S446: Use geometric cooling to update the current temperature. The calculation formula is: ; Where i is the index of the number of iterations of the outer loop, temp i is the temperature of the i-th outer loop iteration, specifically the current temperature after the update, temp i-1 is the temperature of the i-1th outer loop iteration, specifically the current temperature before the update, and α is the cooling coefficient; Step S447: Perform outer loop iteration by repeatedly executing the operations described in steps S443 to S446 until the current temperature is less than the minimum temperature.

4. The method for predicting the remaining life of a capacitor based on machine learning according to claim 3, characterized in that: In step S1, the remaining life prediction modeling is used to design and construct the basic architecture of the capacitor remaining life prediction model, specifically to construct a prediction model for the remaining life of the capacitor, and the prediction model specifically includes a data generation layer, a life prediction layer and a model optimization layer; In step S2, the data generation layer is constructed to construct a capacitor experimental system, collect capacitor experimental data, and perform data preprocessing and synthesis. Specifically, a capacitor charging and discharging experimental system is constructed, and capacitor indicator data of the experimental system is collected. A recursive graph generation method is used to perform data preprocessing and recursive graph data generation to obtain capacitor experimental recursive graph data.

5. The method for predicting the remaining life of a capacitor based on machine learning according to claim 4, characterized in that: In step S5, the remaining life prediction of the capacitor is specifically performed by constructing the data generation layer, the life prediction layer and the model optimization layer to predict the remaining life of the capacitor, and obtaining the remaining life information of the capacitor by using the optimal remaining life prediction model.

6. A capacitor remaining life prediction system based on machine learning, used to implement the capacitor remaining life prediction method based on machine learning as described in any one of claims 1 to 5, characterized in that: It includes a remaining life prediction modeling module, a data generation layer building module, a life prediction layer building module, a model optimization layer building module and a capacitor remaining life prediction module.

7. The capacitor remaining life prediction system based on machine learning according to claim 6, characterized in that: The remaining life prediction modeling module is used for remaining life prediction modeling, and the basic architecture of the capacitor remaining life prediction model is obtained through the remaining life prediction modeling, and the data generation layer, the life prediction layer and the model optimization layer are constructed according to the basic architecture of the capacitor remaining life prediction model; The data generation layer construction module is used for data generation layer construction, through which capacitor experimental recursive graph data is obtained, and the capacitor experimental recursive graph data is sent to the life prediction layer construction module and the model optimization layer construction module; The life prediction layer construction module is used for life prediction layer construction, and obtains the capacitor remaining life prediction model through life prediction layer construction, and sends the obtained capacitor remaining life prediction model to the model optimization layer construction module; The model optimization layer construction module is used for model optimization layer construction, and obtains the optimal remaining life prediction model through model optimization layer construction, and sends the optimal remaining life prediction model to the capacitor remaining life prediction module; The capacitor remaining life prediction module is used for capacitor remaining life prediction. It predicts the remaining life of the capacitor by constructing the data generation layer, the life prediction layer and the model optimization layer, and obtains the remaining life information of the capacitor by using the optimal remaining life prediction model.