Traction system fault detection method based on ppdm and expert prompting learning

Abstract

This invention provides a traction system fault detection method based on PPDM and expert prompting learning, including the following steps: S1) Obtaining the fault-free dataset U from the simulation platform. N and noisy fault dataset U F S2) Construct a reconstruction model based on PPDM, including PDM and ICM; S3) Construct a domain generalization target network clustering model to obtain P meta ;S4) Change U N Input into PDM, based on P meta Get P i The extracted features are compared with P i Reconstructed into U f1 Input into ICM, and calculate J. th ;S5) P i As "K" and "V", U F The feature is designated as "Q"; S6) will use U F Input into PDM and cross-attention network to generate P m Feature extraction and P m Reconstructed into U f2 The input is fed into ICM, j is calculated, and compared with Jth. If j > Jth, then... th There is a fault, if j <J th No faults.

Description

Technical Field

[0001] This invention relates to the field of train fault detection technology, and in particular to a traction system fault detection method based on PPDM and expert prompting learning. Background Technology

[0002] High-speed trains, as an important component of modern transportation, provide people with a fast, convenient, and comfortable mode of travel. The traction control system (TCS) of a high-speed train is one of the core devices ensuring the normal operation of the train. The main function of this system is to control the train's traction speed by controlling the motor speed and motor output torque.

[0003] With the development of technology, the traction control system of high-speed trains has been continuously improved and perfected. However, due to the complex operating environment and various reasons, high-speed trains may still encounter momentary faults during operation. The symptoms are very minor and can be easily masked by external noise and interference. However, these faults may evolve into major faults, leading to emergency shutdowns or other serious accidents.

[0004] There are many reasons for minor malfunctions in the traction control system of high-speed trains, including manufacturing defects in the equipment itself, poor manufacturing processes, mechanical damage and physical damage caused by long-term wear and use, and aging, damage, or loosening of key components of the system. These factors can all lead to minor malfunctions in the traction control system, thereby affecting passenger safety and travel experience.

[0005] In recent years, data-driven fault detection (FD) has been widely used in the safe operation of traction control systems due to its unique advantages. However, data-driven weak fault detection technology usually requires a large amount of historical data to build and train the unsupervised fault detection network, which is difficult to obtain in actual high-speed train operation. Currently, most data-driven fault detection methods for high-speed trains only consider the steady-state operation of the train, and their performance under traction and braking conditions is poor. However, traction and braking conditions are crucial aspects of train operation and also affect the safe operation of the train. Summary of the Invention

[0006] This invention provides a traction system fault detection method based on PPDM and expert prompting learning, which can effectively improve the model's ability to detect minor faults in complex scenarios and provide strong support for the safe operation of the traction system.

[0007] In a first aspect, embodiments of the present invention provide a traction system fault detection method based on PPDM and expert prompting learning, comprising the following steps:

[0008] S1) Obtain the fault-free dataset U from the simulation platform respectively. N and noisy fault dataset U F And according to a certain proportion, the fault-free dataset U N and noisy fault dataset U F The dataset is divided into a training set and a test set, and a fault-free dataset U is defined. N For the source domain, the noisy fault dataset U F For the target domain;

[0009] S2) Build a data reconstruction model based on the Pre-Pyramid Deformation Module, wherein the model includes two modules, namely the Pyramid Deformation Module (PDM Module) and the Information Compression Module (ICM Module);

[0010] S3) Construct a target network based on an expert-guided domain generalization model. The source domain dataset is used to obtain the initial meta-prompt P through clustering. meta ={P1……P N};

[0011] S4) Transfer the fault-free dataset U N The input is sent to the PDM module, and the PDM module processes U. N Feature extraction is performed on the target network. Based on metaprompt P meta Training yields expert prompt P i ={P E1 ...P EN}, change U N The extracted features and expert prompts are reconstructed into dataset U. f1 The data is input to the ICM module to compress anomalies and remove noise and redundancy. After the ICM module, an evaluation function is used to calculate the reconstruction error of the dataset and obtain the threshold J. th ;

[0012] S5) Design a network based on the cross-attention mechanism, which will generate a fault-free dataset U N The expert prompt P obtained through training i The fault dataset U is used as the "key" (K) and "value" (V). F The features are used as "queries" (Q);

[0013] S6) Transfer the fault dataset U FThe data are input into the PDM module and the cross-attention network, respectively. The PDM module extracts features from the fault dataset, and the cross-attention network generates attention weights by calculating the similarity between "Q" and "V". Based on these weights, the expert prompts are weighted and combined, and these weighted prompts are then input into the target network. In the middle, generate expert prompt P m ={P FE1 ...P FEN}, extract features from the fault dataset and combine them with expert prompts P m Reconstructing dataset U f2 The input is fed into the ICM module for noise removal of the dataset. The reconstruction error of the dataset is calculated using an evaluation function to obtain the output value j, which is then compared with the threshold Jth obtained during the training phase. If j > Jth, the output value is determined. th If j <J th If so, the system is considered to be fault-free.

[0014] Preferably, in step S1), the source domain training set U N and the target domain test dataset U F Normalization is performed in the range [-1, 1] to accelerate the convergence of the training network, resulting in a one-dimensional real vector:

[0015]

[0016] In the formula, a is the normal operation data training set U. N The number of sample data in the middle; U N (i) represents the i-th group of normal operation data training set; n represents the dimension of the signal; b is the fault operation test dataset U. F The number of sample data in the middle; U F (i) represents the i-th signal in the fault operation test dataset. Indicates signal U F (i) Signal amplitude in the 1st, ..., nth dimension, where n represents the dimension of the signal.

[0017] Preferably, in step S2), an ICM information compression module is designed:

[0018] The information compression module ICM adopts the VQ-Layer design method. After reconstruction, the dataset is passed through an encoder to extract high-dimensional feature embeddings, resulting in the feature representation z. e By searching for z e The most recent memory item is quantized into a single memory feature cube z. qThe ICM information compression module learns the reconstructed dataset output by the data reconstruction diversity perception module based on the embedded noise data information, and finds the features of the noise data by calculating the L2 minimum distance. The ICM information compression module includes two steps: searching and compressing. It searches for noise information in the dataset and compresses the noise information with constraints. (It includes a hyperparameter β, a stopping gradient operator SG, and a compression loss Lcom).

[0019] The two steps of searching and compressing the ICM information compression module are as follows:

[0020] The S21)ICM module receives the input dataset and identifies noise information in the data by finding the feature vector with the minimum L2 distance to the memory cell. The calculation formula is as follows:

[0021]

[0022] Among them, z eh,w z is the feature embedded in the dataset. n These are the feature vectors within the memory unit;

[0023] S22) Compression loss is calculated using VQ-Layer to reduce the impact of noise data. The calculation of compression loss is as follows:

[0024] L com =||SG(Z) e )-Z q ||2+β||ze-SG(Z q )||2;

[0025] Where SG(·) denotes stopping the gradient operation, and Z e It is an embedded representation, Z q These are the quantized features. The hyperparameter β is used to balance the compression effect and the noise removal effect.

[0026] Preferably, step S3) includes the following steps:

[0027] S31) Transfer the source domain dataset U N The data is input into the PDM module to extract a set of feature vectors, which represent the main characteristics of the data.

[0028] S32) Cluster the extracted feature vectors to obtain subsets with multiple similar features. Use meta-learning methods to find commonalities from these clustered subsets and generate meta-prompts.

[0029] Preferably, step S4) includes the following steps:

[0030] The PDM module handles the fault-free dataset U.N and noisy fault dataset U F Feature extraction and reconstruction are performed. The PDM module calculates the feature differences between normal and faulty data by setting a specific deformation function φ(x), and then reconstructs the data. The feature extraction uses the following function:

[0031] φ(x)=f deforn (x)+f norm (x);

[0032] Among them, f deforn f represents a convolution operation used to extract local features from time series data. norm Data normalization is performed to ensure the stability of the training network and accelerate convergence. φ(x) represents the process of converting the original input X into a feature representation.

[0033] As a preferred option, the PDM module supports the fault-free dataset U. N and noisy fault dataset U F Feature extraction and reconstruction are performed. The PDM module reconstruction process is represented as follows:

[0034] The PDM module is used in a given training set of samples X = {X1……X2} N In the case of}, the normal dataset U obtained by the learning digital simulation platform N and noisy fault dataset U F The different information between the data and the data reconstruction are performed; the diversity between the dataset acquired by the digital simulation platform and the reconstructed dataset is represented as follows: The underestimation error generated during the reconstruction process is passed through Compensation will be provided again.

[0035] Preferably, in steps S4) and S6), training the expert prompt includes the following steps:

[0036] S41) Initialize the expert prompt for each source domain X. i First, set the expert prompt P i Initialized as a pre-trained meta-prompt P meta Ensure that the expert prompts for each source domain are tuned and optimized based on the same initial prompt:

[0037] P i ←P meta ;

[0038] S42) Optimize the expert prompt by minimizing KL and divergence loss L. KL For each sample (x) in the source domain data i,j ,y i,j Perform gradient descent optimization:

[0039]

[0040] Where α A L is the learning rate. KL Measure the difference between the model output and the target label;

[0041] S43) Optimize each expert prompt P i Normalization is performed to maintain consistency:

[0042]

[0043] This normalization step ensures that expert prompts maintain the same scale across different source domains.

[0044] As a preferred method, the method for training expert prompts using the cross-attention mechanism in step S5) is as follows:

[0045] A weighted averaging method is used to combine prompts that match subsets; a cross-attention mechanism is used to assign weights and the required number of expert prompts to subsets of the training dataset, with expert prompts becoming "keys" (K) and "values" (V). The normal operation dataset obtained from the digital simulation platform becomes the target attention of the "query" (Q), and expert prompts P are trained and generated. m ={P FE1 ...P FEN}, calculate the attention weights q1...q, which are the same as the number of expert prompts corresponding to the subset of training data. n ;(The method for calculating weights q) n =VQK T ).

[0046] Preferably, step S6) specifically includes the following steps:

[0047] The expert prompt P generated by the expert prompt module is... i ={P E1 ...P EN The training dataset, obtained from the digital simulation platform, serves as input to the fault detection framework. This training dataset is then fed into the diversity perception module for feature amplification, extraction, and dataset reconstruction. Expert prompts are assigned to the reconstructed dataset, and the similarity between the reconstructed and training datasets is controlled through these prompts. The reconstructed training dataset and the matched expert prompts are then used as input to the ICM information compression module. This module restricts and compresses anomalous information within the reconstructed dataset, using a loss function L... com This indicates that the fault detection threshold j is calculated using the (statistical method);

[0048] The test dataset obtained from the digital simulation platform is used as the input to the data reconstruction fault detection framework, and the output is the calculated evaluation criterion J. th Evaluation criteria J th The fault detection results are obtained by comparing them with the fault detection threshold j:

[0049]

[0050] Preferably, in step S2), an ICM information compression module is designed:

[0051] The information compression module ICM adopts the VQ-Layer design method. After reconstruction, the dataset is passed through an encoder to extract high-dimensional feature embeddings, resulting in the feature representation z. e By searching for z e The most recent memory item is quantized into a single memory feature cube z. q The ICM information compression module learns the reconstructed dataset output by the data reconstruction diversity perception module based on the embedded noise data information, and finds the features of the noise data by calculating the L2 minimum distance. The ICM information compression module includes two steps: searching and compressing. It searches for noise information in the dataset and compresses the noise information with constraints. (It includes a hyperparameter β, a stopping gradient operator SG, and a compression loss Lcom).

[0052] The two steps of searching and compressing the ICM information compression module are as follows:

[0053] The S21)ICM module receives the input dataset and identifies noise information in the data by finding the feature vector with the minimum L2 distance to the memory cell. The calculation formula is as follows:

[0054]

[0055] in, z is the feature embedded in the dataset. n These are the feature vectors within the memory unit;

[0056] S22) Compression loss is calculated using VQ-Layer to reduce the impact of noise data. The calculation of compression loss is as follows:

[0057] L com =||SG(z) e )-z q ||2+β||z e -SG(z q )||2;

[0058] Where SG(·) denotes stopping the gradient operation, and Z eIt is an embedded representation, Z q These are the quantized features. The hyperparameter β is used to balance the compression effect and the noise removal effect.

[0059] Preferably, in step S6), the fault detection threshold is calculated, and the specific method is as follows:

[0060] The formula for calculating the mean residual is:

[0061]

[0062] Where x i It is the raw data. It involves reconstructing the data, where N is the number of data points;

[0063] The formula for calculating the standard deviation of residuals is:

[0064]

[0065] The threshold calculation formula is:

[0066] j = μ res +η·σ res +βL com ;

[0067] Where, μ res Let σ be the mean of the residuals. res η is the standard deviation of the residuals, η is the confidence parameter, and β is the adjustment hyperparameter.

[0068] The beneficial effects of this invention are as follows:

[0069] 1. This invention is based on a traction system fault detection method based on PPDM and expert prompting learning. PPDM realizes efficient feature extraction and processing of fault-free data and noisy fault data. This method effectively solves the problem of insufficient processing of noisy data in traditional methods, thereby significantly improving the accuracy and robustness of fault detection.

[0070] 2. This invention innovatively proposes a domain generalization model based on expert prompts. By using expert prompts obtained through training and combining them with a cross-attention mechanism, the "Q" and "K" features of fault-free data and the "V" features of faulty data are weighted and combined. The expert prompts serve as auxiliary information. After being fused with the fault features, accurate fault location and classification can be achieved. This method effectively improves the model's ability to detect subtle faults in complex scenarios and provides strong support for the safe operation of the traction system. Attached Figure Description

[0071] Figure 1 This is a flowchart of a traction system fault detection method based on PPDM and expert prompting learning provided in one embodiment of the present invention;

[0072] Figure 2 This is a pyramid deformation module and target network provided in one embodiment of the present invention. Structural diagram;

[0073] Figure 3 This is a structural diagram of the ICM module in the data reconstruction stage provided in one embodiment of the present invention;

[0074] Figure 4 This is a graph showing the total detection loss of the overall framework provided in one embodiment of the present invention. Detailed Implementation

[0075] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0076] In this embodiment of the invention, terms such as "furthermore," "exemplarily," or "preferredly" are used as examples, illustrations, or descriptions and should not be construed as being more preferred or advantageous than other embodiments or designs. The use of terms such as "furthermore," "exemplarily," or "optionally" is intended to present the relevant concepts in a specific manner.

[0077] To facilitate a more convenient description of the working principle of the embodiments of the present invention, the following introduction of relevant technical scenarios is given first.

[0078] In recent years, data-driven fault detection (FD) has been widely used in the safe operation of traction control systems due to its unique advantages. However, data-driven weak fault detection technology usually requires a large amount of historical data to build and train the unsupervised fault detection network, which is difficult to obtain in actual high-speed train operation. Currently, most data-driven fault detection methods for high-speed trains only consider the steady-state operation of the train, and their performance under traction and braking conditions is poor. However, traction and braking conditions are crucial aspects of train operation and also affect the safe operation of the train.

[0079] Based on this, the present invention provides a fault detection method for traction systems based on PPDM and expert prompting learning, including the following steps: S1) Obtaining a fault-free dataset U from a digital simulation platform. N and noisy fault dataset u F The dataset is divided into training and test sets according to a certain ratio, and a fault-free dataset U is set. N For the source domain, the noisy fault dataset U FS2) Construct a data reconstruction model based on the Pre-Pyramid Deformation Module (PDM) for the target domain; S3) Construct a target network based on an expert-guided domain generalization model. The source domain dataset is used to obtain the initial meta-prompt P through clustering. meta ={P1……P N}; S4), Transfer the fault-free dataset U N The input is sent to the PDM module, and the PDM module processes U. N Feature extraction is performed on the target network. Based on metaprompt P meta Training yields expert prompt P i ={P E1 ...P EN}, change U N Extracted features and expert prompt P i Reconstructed into dataset U f1 The data is input into the ICM module to compress anomalies and remove noise and redundancy. After the ICM module, an evaluation function is used to calculate the reconstruction error of the dataset and obtain the threshold J. th S5) Design a network based on the cross-attention mechanism to process the fault-free dataset U N The expert prompt P obtained through training i The fault dataset U is used as the "key" (K) and "value" (V). F The features are used as "queries" (Query, Q); S6), the fault dataset U F The data are input into the PDM module and the cross-attention network, respectively. The PDM module extracts features from the fault dataset, and the cross-attention network generates attention weights by calculating the similarity between "Q" and "V". Based on these weights, the expert prompts are weighted and combined, and these weighted prompts are then input into the target network. In the middle, generate expert prompt P m ={P fE1 ...P FEN}, extract features from the fault dataset and combine them with expert prompts P m Reconstructing dataset U f2 The input is fed into the ICM module for noise removal of the dataset. The reconstruction error of the dataset is calculated using an evaluation function to obtain the output value j, which is then compared with the threshold Jth obtained during the training phase. If j > Jth, the output value is determined. th If j <J thIf the system is found to be fault-free, then it is determined that there is no fault. Based on this, the embodiments of the present invention can effectively improve the model's ability to detect minor faults in complex scenarios, providing strong support for the safe operation of the traction system.

[0080] The embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0081] like Figure 1 As shown, this embodiment provides a traction system fault detection method based on PPDM and expert prompting learning, including the following steps:

[0082] S1) Obtain the fault-free dataset and the noisy fault dataset of the digital simulation platform, and divide the fault-free dataset and the noisy fault dataset into a data training set and a test dataset;

[0083] The datasets include both fault-free and noisy fault operation datasets, comprising the transformer secondary voltage output U. n and current output I n The voltage output U of the capacitor supporting the DC link cd1 The voltage output U of the supporting capacitor under DC link cd2 The inverter's three-phase current output {I} is... sa ,I sb ,I sc}, Motor speed signal V;

[0084] The normal operation data training set acquired by the digital simulation platform is defined as U. N ,in

[0085] [U n ,I n U cd1 U cd2 ,I sa ,I sb ,I sc ,V]∈U N ;

[0086] Define the fault operation test dataset obtained by the digital simulation platform as U F ,in

[0087] [U n ,I n U cd1 U cd2 ,I sa ,I sb ,I sc ,V]∈U F ;

[0088] Training set U of normal operating data N and fault run test dataset U FNormalization is performed in the range [-1, 1] to accelerate the convergence of the training network, resulting in a one-dimensional real vector:

[0089]

[0090] In the formula, a is the normal operation data training set U. N The number of sample data in the middle; U N (i) represents the i-th set of normal operation data training set; n represents the dimension of the signal;

[0091] b represents the fault operation test dataset U. F The number of sample data in the middle; U F (i) represents the i-th signal in the fault operation test dataset. Indicates signal U F (i) Signal amplitude in the 1st, ..., nth dimension, where n represents the dimension of the signal;

[0092] b represents the fault operation test dataset U. F The number of sample data in the middle; U F (i) represents the i-th signal in the fault operation test dataset. Indicates signal U F (i) Signal amplitude in the 1st, ..., nth dimension, where n represents the dimension of the signal.

[0093] Construct a target network based on an expert-guided domain generalization model The source domain dataset is used to obtain the initial meta-prompt P through clustering. meta ={P1……P N}

[0094] A data reconstruction model based on the Pre-Pyramid Deformation Module is constructed, which includes two modules: the PDM module and the ICM module.

[0095] The fault-free dataset U N The input is sent to the PDM module, and the PDM module processes U. N Feature extraction is performed on the target network. Based on metaprompt P meta Training yields expert prompt P i ={P E1 ...P EN}, change U N Extracted features and expert prompt P i Reconstructed into dataset U f1 The data is input into the ICM module to compress anomalies and remove noise and redundancy. After the ICM module, an evaluation function is used to calculate the reconstruction error of the dataset and obtain the threshold J.th .

[0096] Design a network based on the cross-attention mechanism, starting from a fault-free dataset U N The expert prompt P obtained through training i The fault dataset U is used as the "key" (K) and "value" (V). F The feature is referred to as "Query" (Q).

[0097] Fault dataset U F The data are input into the PDM module and the cross-attention network, respectively. The PDM module extracts features from the fault dataset, and the cross-attention network generates attention weights by calculating the similarity between "Q" and "V". Based on these weights, the expert prompts are weighted and combined, and these weighted prompts are then input into the target network. In the middle, generate expert prompt P m ={P FE1 ...P FEN}, extract features from the fault dataset and combine them with expert prompts P m Reconstructing dataset U f2 The input is fed into the ICM module for noise removal of the dataset. The reconstruction error of the dataset is calculated using an evaluation function to obtain the output value j, which is then compared with the threshold Jth obtained during the training phase. If j > Jth, the output value is determined. th If j <J th If so, the system is considered to be fault-free.

[0098] like Figure 2 As shown, the pyramid module and the target network Its functions are as follows:

[0099] In preferred steps S4) and S6), the PDM module processes the fault-free dataset U. N and noisy fault dataset U F Feature extraction and reconstruction are performed. The PDM module calculates the feature differences between normal and faulty data by setting a specific deformation function φ(x), and then reconstructs the data. The feature extraction uses the following function:

[0100] φ(x)=f deforn (x)+f norm (x);

[0101] Among them, f deforn f represents a convolution operation used to extract local features from time series data. norm Data normalization is performed to ensure the stability of the training network and accelerate convergence. φ(x) represents the process of converting the original input X into a feature representation.

[0102] The PDM module reconstruction process is represented as follows:

[0103] The PDM module is used in a given training set of samples X = {X1……X2} N In the case of}, the normal dataset U obtained by the learning digital simulation platform N and noisy fault dataset U F The different information between the data and the data reconstruction are performed; the diversity between the dataset acquired by the digital simulation platform and the reconstructed dataset is represented as follows: The underestimation error generated during the reconstruction process is passed through Compensation will be provided again;

[0104] S31) Transfer the source domain dataset U N The data is input into the PDM module to extract a set of feature vectors, which represent the main characteristics of the data.

[0105] S32) Cluster the extracted feature vectors to obtain subsets with multiple similar features. Use meta-learning methods to find commonalities from these clustered subsets and generate meta-cues.

[0106] S41) Initialize the expert prompt for each source domain X. i First, set the expert prompt P i Initialized as a pre-trained meta-prompt P meta Ensure that the expert prompts for each source domain are tuned and optimized based on the same initial prompt:

[0107] P i ←[ meta ;

[0108] S42) Optimize the expert prompt by minimizing KL and divergence loss L. kL For each sample (x) in the source domain data i,j ,y i,j Perform gradient descent optimization:

[0109]

[0110] Where α A L is the learning rate. KL Measure the difference between the model output and the target label;

[0111] S43) Optimize each expert prompt P i Normalization is performed to maintain consistency:

[0112]

[0113] This normalization step ensures that expert prompts maintain the same scale across different source domains.

[0114] In the preferred step S5), the cross-attention mechanism is applied as follows:

[0115] A weighted averaging method is used to combine prompts that match subsets; a cross-attention mechanism is used to assign weights and the required number of expert prompts to subsets of the training dataset, with expert prompts becoming "keys" (K) and "values" (V). The normal operation dataset obtained from the digital simulation platform becomes the target attention of the "query" (Q), and expert prompts P are trained and generated. m ={P FE1 ...P FEN}, calculate the attention weights q1...q, which are the same as the number of expert prompts corresponding to the subset of training data. n ;(The method for calculating weights q) n =VQK T ).

[0116] like Figure 3 As shown, the ICM module functions as follows:

[0117] The expert prompt P generated by the expert prompt module is... i ={P E1 ...P EN The training dataset, obtained from the digital simulation platform, serves as input to the fault detection framework. This training dataset is then fed into the diversity perception module for feature amplification, extraction, and dataset reconstruction. Expert prompts are assigned to the reconstructed dataset, and the similarity between the reconstructed and training datasets is controlled through these prompts. The reconstructed training dataset and the matched expert prompts are then used as input to the ICM information compression module. This module restricts and compresses anomalous information within the reconstructed dataset, using a loss function L... com This indicates that the calculated L com The fault detection threshold j is obtained by calculating the weight values ​​of the expert prompts using a statistical method.

[0118] The test dataset obtained from the digital simulation platform is used as the input to the data reconstruction fault detection framework, and the output is the calculated evaluation criterion J. th Evaluation criteria J th The fault detection results are obtained by comparing them with the fault detection threshold j:

[0119]

[0120] The Information Compression Module (ICM) adopts the VQ-Layer design method. Based on the embedded noisy data information, the ICM learns the reconstructed dataset output by the Data Reconstruction Diversity Awareness Module and finds the features of the noisy data by calculating the L2 minimum distance. The ICM information compression module includes two steps: "Search" and "Comp". It searches for noisy information in the dataset and performs restrictive compression on the noisy information (including a hyperparameter β, a stopping gradient operator SG, and a compression loss Lcom).

[0121] The two steps of searching and compressing the ICM information compression module are as follows:

[0122] The S21)ICM module receives the input dataset and identifies noise information in the data by finding the feature vector with the minimum L2 distance to the memory cell. The calculation formula is as follows:

[0123]

[0124] Among them, z eh,w z is the feature embedded in the dataset. n These are the feature vectors within the memory unit;

[0125] S22) Compression loss is calculated using VQ-Layer to reduce the impact of noise data. The calculation of compression loss is as follows:

[0126] L com =||SG(Z) e )-Z q ||2+β||ze-SG(Z q )||2;

[0127] Where SG(·) denotes stopping the gradient operation, and Z e It is an embedded representation, Z q These are the quantized features. The hyperparameter β is used to balance the compression effect and the noise removal effect.

[0128] In the preferred step S6), the fault detection threshold is calculated, and the specific method is as follows:

[0129] The formula for calculating the mean residual is:

[0130]

[0131] Where x i It is the raw data. It involves reconstructing the data, where N is the number of data points;

[0132] The formula for calculating the standard deviation of residuals is:

[0133]

[0134] The threshold calculation formula is:

[0135] j = μ res +η·σ res +βL com ;

[0136] Where, μ res Let σ be the mean of the residuals. res η is the standard deviation of the residuals, η is the confidence parameter, and β is the adjustment hyperparameter.

[0137] like Figure 4 As shown, the detection performance of this model is as follows:

[0138] The validation loss of the model tends to stabilize during training. This loss includes all training losses. In the initial stage, the validation loss drops rapidly, indicating that the model has effectively learned the data features in the early stages. As the number of training epochs increases, the validation loss tends to stabilize around 10 epochs, eventually remaining at around 1.65, indicating that the model's generalization ability is relatively stable after training. These results verify the robustness and reliability of the model in handling fault detection tasks, making it suitable for practical application scenarios.

[0139] In addition, an embodiment of the present invention discloses an electronic device, comprising: at least one processor; at least one memory for storing at least one program; and when the at least one program is executed by the at least one processor, implementing the traction system fault detection method based on PPDM and expert prompting learning as in any of the preceding embodiments.

[0140] In addition, one embodiment of the present invention discloses a computer-readable storage medium storing computer-executable instructions for performing the traction system fault detection method based on PPDM and expert prompting learning as described in any of the preceding embodiments.

[0141] The system architecture and application scenarios described in the embodiments of this invention are for the purpose of more clearly illustrating the technical solutions of the embodiments of this invention, and do not constitute a limitation on the technical solutions provided by the embodiments of this invention. As those skilled in the art will know, with the evolution of system architecture and the emergence of new application scenarios, the technical solutions provided by the embodiments of this invention are also applicable to similar technical problems.

[0142] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0143] In hardware implementations, the division between functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0144] The terms “component,” “module,” “system,” etc., used in this specification are used to refer to computer-related entities, hardware, firmware, combinations of hardware and software, software, or software in execution. For example, a component can be, but is not limited to, a process running on a processor, a processor, an object, an executable file, an execution thread, a program, or a computer. As illustrated, applications running on computing devices and computing devices can both be components. One or more components may reside in a process or execution thread, and components may be located on a single computer or distributed among two or more computers. Furthermore, these components can be executed from various computer-readable media on which various data structures are stored. Components can communicate, for example, via local or remote processes based on signals having one or more data packets (e.g., data from two components interacting with another component between a local system, a distributed system, or a network, such as the Internet interacting with other systems via signals).

Claims

1. A traction system fault detection method based on PPDM and expert-guided learning, characterized in that: Includes the following steps: S1) Obtain the fault-free dataset U from the digital simulation platform. N and noisy fault dataset U F And according to a certain proportion, the fault-free dataset U N and noisy fault dataset U F The dataset is divided into a training set and a test set, and a fault-free dataset U is defined. N For the source domain, the noisy fault dataset U F For the target domain; S2) Construct a data reconstruction model based on the Pre-Pyramid Deformation Module, wherein the data reconstruction model includes a pyramid deformation module and an information compression module; S3) Construct a target network based on an expert-guided domain generalization model. The source domain dataset is used to obtain the initial meta-prompt P through clustering. meta ={P1……P N }; S4) Transfer the fault-free dataset U N The input is fed into the pyramid deformation module, which then processes the U... N Feature extraction is performed on the target network. Based on metaprompt P meta Training yields expert prompt P i ={P E1 ...P EN }, change U N Extracted features and expert prompt P i Reconstructed into dataset U f1 , will dataset U f1 The data is input into the information compression module to remove noise and redundant information. After the information compression module, an evaluation function is used to calculate the reconstruction error of the dataset and obtain the threshold Jt. h ; S5) Design a network based on the cross-attention mechanism, which will generate a fault-free dataset U N The expert prompt P obtained through training i As "keys" and "values", the fault dataset U F The features are used as "queries"; S6) Transfer the fault dataset U F The data are input into the pyramid deformation module and the cross-attention network, respectively. The pyramid deformation module extracts features from the fault dataset, while the cross-attention network generates attention weights by calculating the similarity between "queries" and "values," and then weights and combines the expert prompts based on these weights. The weighted expert prompts are then input into the target network. In the middle, generate expert prompt P m ={P FE1 ...P FEN }, extract features from the fault dataset and combine them with expert prompts P m Reconstructing dataset U f2 The data is input to the information compression module for noise removal, and the reconstruction error of the dataset is calculated using an evaluation function to obtain the output value j, which is then compared with the threshold J obtained during the training phase. th Compare, if j>J th If j <J th If so, the system is considered to be fault-free.

2. The traction system fault detection method based on PPDM and expert prompting learning according to claim 1, characterized in that: In step S1), the fault-free dataset U N and the aforementioned noisy fault dataset U F Including transformer secondary voltage output U n and current output I n The voltage output U of the capacitor supporting the DC link cd1 The voltage output U of the supporting capacitor under DC link cd2 Inverter a, b, c three-phase current output {I sa I sb I sc }, Motor speed signal V; The normal operation data training set acquired by the digital simulation platform is defined as U. N , where [U n I n U cd1 U cd2 I sa I sb I sc ,V]∈U N ; Define the fault operation test dataset obtained by the digital simulation platform as U F , where [U n I n U cd1 U cd2 I sa I sb I sc ,V]∈U F ; Training set U of normal operating data N and fault run test dataset U F Normalization is performed within the range [-1, 1], resulting in a one-dimensional real vector: In the formula, a is the normal operation data training set U. N The number of sample data in the middle; U N (i) represents the i-th set of normal operation data training set; n represents the dimension of the signal; c represents the fault operation test dataset U. F The number of sample data in the middle; U F (i) represents the i-th signal in the fault operation test dataset. Indicates signal U F (i) Signal amplitude in the 1st, ..., nth dimension, where n represents the dimension of the signal.

3. The traction system fault detection method based on PPDM and expert prompting learning according to claim 1, characterized in that: Step S3) involves obtaining the initial meta-prompt from the source domain dataset using clustering methods, including the following steps: S31) Transfer the source domain dataset U N The input is fed into the pyramid deformation module to extract a set of feature vectors, which represent the main characteristics of the data. S32) Cluster the extracted feature vectors to obtain subsets with multiple similar features. Use a meta-learning method to find commonalities from the clustered subsets and generate meta-prompts.

4. The traction system fault detection method based on PPDM and expert prompting learning according to claim 1, characterized in that: In steps S4) and S6), the pyramid deformation module processes the fault-free dataset U. N and noisy fault dataset U F Feature extraction and reconstruction are performed. The pyramid deformation module calculates the feature differences between normal and faulty data by setting a specific deformation function φ(x), and then reconstructs the data. The feature extraction uses the following function: φ(x)=f conv (x)+f deform (x) ; (1) Among them, f conv f represents a convolution operation used to extract local features from time series data. deform Used for data normalization, φ(x) represents the process of converting the original input X into a feature representation.

5. The traction system fault detection method based on PPDM and expert prompting learning according to claim 4, characterized in that: The pyramid deformation module described above is for the fault-free dataset U N and noisy fault dataset U F Feature extraction and reconstruction are performed. The reconstruction process of the pyramid deformation module is represented as follows: The pyramid deformation module is based on a given training set of samples X = {X1……X}. N In the case of}, the normal dataset U obtained by the learning digital simulation platform N and noisy fault dataset U F The different information between the data and the data reconstruction are performed; the diversity between the dataset acquired by the digital simulation platform and the reconstructed dataset is represented as follows: The underestimation error generated during the reconstruction process is passed through Compensation will be provided again.

6. The traction system fault detection method based on PPDM and expert prompting learning according to claim 4, characterized in that: In steps S4) and S6), the expert prompt P is trained. i Includes the following steps: S41) Initialize the expert prompt for each source domain X. i First, set the expert prompt P i Initialized as a pre-trained meta-prompt P meta Ensure that the expert prompts for each source domain are tuned and optimized based on the same initial prompt: P i ←P meta ; (2) S42) Optimize the expert prompt by minimizing KL and divergence loss L. KL For each sample (x) in the source domain data i,j ,y i,j Perform gradient descent optimization: Where α A L is the learning rate. KL Measure the difference between the model output and the target label; S43) Optimize each expert prompt P i Normalization is performed to maintain consistency: The normalization process ensures that the expert prompts maintain the same scale across different source domains.

7. The traction system fault detection method based on PPDM and expert prompting learning according to claim 1, characterized in that: The method for training expert cues using the cross-attention mechanism in step S5) is as follows: A weighted average method is used to combine prompts that match subsets; a cross-attention mechanism is used to assign weights and the required number of expert prompts to subsets of the training dataset, with expert prompts becoming "keys" and "values," and the normal operation dataset obtained from the digital simulation platform becoming the target attention of the "query," thus training and generating expert prompt P. m ={P FE1 ...P FEN }, calculate the attention weights q1...q, which are the same as the number of expert prompts corresponding to the subset of training data. n Among them, the weight calculation method q n =VQK T .

8. The traction system fault detection method based on PPDM and expert prompting learning according to claim 6, characterized in that: In step S6), the generated expert prompt P i ={P E1 ...P EN The training dataset, obtained from the digital simulation platform, serves as input to the fault detection framework. This training dataset is then fed into a diversity perception module for feature amplification, extraction, and dataset reconstruction. Expert prompts are assigned to the reconstructed dataset, and the similarity between the reconstructed and training datasets is controlled through these prompts. The reconstructed training dataset and the matched expert prompts are then used as input to an information compression module. This module restricts and compresses anomalous information within the reconstructed dataset, using a loss function L... com This indicates that the threshold J is calculated using the statistical method. th ; The test dataset obtained from the digital simulation platform is used as the input for the data reconstruction fault detection framework. The output is the calculated threshold J. th , threshold J th The fault detection results are obtained by comparing the output value j with the reconstruction error:

9. The traction system fault detection method based on PPDM and expert prompting learning according to claim 8, characterized in that: In step S2), design the information compression module: The information compression module adopts the VQ-Layer design method. The reconstructed dataset is passed through an encoder to extract high-dimensional feature embeddings, resulting in the feature representation z. e By searching for z e The most recent memory item is quantized into a single memory feature cube z. q The information compression module learns the reconstructed dataset output by the data reconstruction diversity perception module based on the embedded noise data information, and finds the features of the noise data by calculating the L2 minimum distance. The information compression module includes two steps: "Search" and "Comp". It searches for noise information in the dataset and performs restricted compression on the noise information. The noise information includes a hyperparameter β, a stopping gradient operator SG, and a compression loss Lcom. The information compression module performs two steps of searching and compression, as follows: S21) The information compression module receives the input dataset and identifies noise information in the data by finding the feature vector with the minimum L2 distance to the memory unit. The calculation formula is as follows: in, z is the feature embedded in the dataset. n These are the feature vectors within the memory unit; S22) The compression loss is calculated using VQ-Layer. The calculation of the compression loss is as follows: L com =||SG(z e )-With q ||2+β||z e -SG(z q )||2; (6) 10. The traction system fault detection method based on PPDM and expert prompting learning according to claim 8, characterized in that: In step S6), the output value is calculated, and the specific method is as follows: The formula for calculating the mean residual is: Where x i It is the raw data. It involves reconstructing the data, where N is the number of data points; The formula for calculating the standard deviation of residuals is: The formula for calculating the output value is: j=μ res +η·s res +βL com (9) Where, μ res Let σ be the mean of the residuals. res η is the standard deviation of the residuals, η is the confidence parameter, and β is the adjustment hyperparameter.

Images