Subway electrical equipment partial discharge signal intelligent analysis and diagnosis method and system

Abstract

The invention provides a subway electrical equipment partial discharge signal intelligent analysis and diagnosis method and system, and belongs to the technical field of electrical equipment fault diagnosis. And processing the acquired partial discharge signal data by using a pre-trained diagnosis model to obtain an electrical equipment fault diagnosis result. According to the method, a partial discharge signal is finely decomposed by using a symplectic geometric mode decomposition (SGMD) method introducing sparse regular optimization, and hyper-parameter adaptive adjustment is performed on an LSTM model in combination with a near-end strategy optimization (PPO) algorithm, so that accurate extraction and type identification of the partial discharge signal are realized. The method can effectively improve the precision and intelligent level of partial discharge diagnosis, and provides reliable technical support for operation and maintenance of subway electrical equipment.

Description

Technical Field

[0001] This invention relates to the field of electrical equipment fault diagnosis technology, specifically to an intelligent analysis and diagnosis method and system for partial discharge signals of subway electrical equipment. Background Technology

[0002] Subway electrical equipment (such as transformers, rectifiers, and switchgear) is a key component of urban rail transit power supply systems, and its operating status directly affects the system's safety and stability. Partial discharge, as an early manifestation of insulation system degradation, is one of the main causes of electrical equipment failure. Therefore, conducting research on efficient and intelligent partial discharge signal processing and identification is of great significance for ensuring the safe operation of subway power supply systems.

[0003] High-frequency pulse current detection has become the mainstream method for partial discharge detection in subway electrical equipment due to its high sensitivity, strong anti-interference ability, and non-contact structure. However, in the signal processing and recognition stage, traditional methods such as wavelet transform, EMD, and VMD suffer from severe mode aliasing, reliance on manual parameters, and limited feature representation capabilities when processing high-frequency partial discharge signals, making them difficult to adapt to the complex non-stationary characteristics of partial discharge signals. In addition, existing recognition algorithms mostly rely on manual features and traditional machine learning methods, making it difficult to fully explore the deep time-frequency features in the signal; while deep learning methods have stronger modeling capabilities, they face bottlenecks such as a lack of fault samples and difficulty in globally optimizing hyperparameters, limiting their practicality and generalization ability. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent analysis and diagnosis method and system for partial discharge signals of subway electrical equipment that combines sparse regularized optimized symplectic geometric mode decomposition (SGMD) with reinforcement learning optimized LSTM network, thereby improving the accuracy and reliability of partial discharge identification and solving at least one of the technical problems existing in the background art.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] In a first aspect, the present invention provides an intelligent analysis and diagnosis method for partial discharge signals of subway electrical equipment, comprising:

[0007] Acquire partial discharge signal data of electrical equipment;

[0008] The acquired partial discharge signal data is processed using a pre-trained diagnostic model to obtain electrical equipment fault diagnosis results. The training of the diagnostic model includes: collecting partial discharge fault data during the operation of subway electrical equipment to construct an original dataset; performing denoising preprocessing on the original dataset and extracting effective signals; performing feature analysis and extraction on the effective signals; using reinforcement learning to dynamically optimize the hyperparameters of the LSTM; outputting the optimal hyperparameter combination and using a test set to evaluate the model performance to obtain the trained diagnostic model.

[0009] As a further limitation of the first aspect of the present invention, wavelet denoising is used to preprocess the original dataset, and db4 wavelet is used to perform wavelet threshold denoising on the signal; then the effective signal of the dataset is extracted to reduce the influence of environmental interference, and the effective modes of the denoised partial discharge signal are extracted using symplectic geometric mode decomposition; wherein, in order to optimize the decomposition effect, a sparse regularization term is introduced into the objective function of the symplectic geometric mode decomposition method, which is applicable to the sparse characteristics of the partial discharge signal in the time and frequency domain, highlights the main signal, suppresses noise, and reduces mode mixing.

[0010] As a further limitation of the first aspect of the present invention, partial discharge signals from various devices are collected and labeled to construct a preliminary dataset. Data augmentation techniques such as time-frequency perturbation are used to expand the number of samples, and finally a raw dataset of partial discharge signals with high time-frequency resolution and good engineering adaptability is formed.

[0011] As a further definition of the first aspect of the present invention, the key features extracted are peak factor, kurtosis factor, impulse factor, high-frequency energy, spectral centroid, sample entropy and energy entropy, and a feature matrix is ​​established; wherein, peak factor, kurtosis factor and impulse factor are time-domain features, high-frequency energy and spectral centroid are frequency-domain spectral features, and sample entropy and energy entropy are entropy value features.

[0012] As a further limitation of the first aspect of the present invention, the parameters of the initial reinforcement learning are optimized by using the nearest neighbor policy, the "state-action-reward" range of PPO is defined, and the loss function and related parameters of PPO are defined. The extracted feature matrix is ​​divided into training set and test set in a 7:3 ratio. The LSTM hyperparameter sampling is continuously adjusted and trained by updating the policy network and value network of PPO, thereby outputting the optimal combination of LSTM hyperparameters and weights.

[0013] As a further limitation of the first aspect of the present invention, under the reinforcement learning framework, the hyperparameters of LSTM and their training states are regarded as state variables of PPO, the action is defined as the adjustment method of LSTM hyperparameters, and the reward function is set as the weighted sum of LSTM recognition accuracy and hyperparameter regularization term.

[0014] Secondly, the present invention provides an intelligent analysis and diagnosis system for partial discharge signals of subway electrical equipment, comprising:

[0015] The acquisition module is used to acquire partial discharge signal data of electrical equipment;

[0016] The processing module is used to process the acquired partial discharge signal data using a pre-trained diagnostic model to obtain electrical equipment fault diagnosis results. Training the diagnostic model includes: collecting partial discharge fault data during the operation of subway electrical equipment to construct an original dataset; performing denoising preprocessing on the original dataset and extracting effective signals; performing feature analysis and extraction on the effective signals; using reinforcement learning to dynamically optimize the hyperparameters of the LSTM; outputting the optimal hyperparameter combination and using a test set to evaluate the model performance to obtain the trained diagnostic model.

[0017] Thirdly, the present invention provides a non-transitory computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the intelligent analysis and diagnosis method for partial discharge signals of subway electrical equipment as described in the first aspect.

[0018] Fourthly, the present invention provides a computer device including a memory and a processor, wherein the processor and the memory communicate with each other, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the intelligent analysis and diagnosis method for partial discharge signals of subway electrical equipment as described in the first aspect.

[0019] Fifthly, the present invention provides an electronic device, comprising: a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions for implementing the intelligent analysis and diagnosis method for partial discharge signals of subway electrical equipment as described in the first aspect.

[0020] The beneficial effects of this invention are as follows: It employs a symplectic geometric mode decomposition (SGMD) method with sparse regularization optimization to perform fine decomposition of partial discharge (PD) signals, and combines this with a near-end policy optimization (PPO) algorithm to adaptively adjust the hyperparameters of the LSTM model, achieving accurate extraction and type identification of PD signals. This method effectively improves the accuracy and intelligence level of PD diagnosis, providing reliable technical support for the operation and maintenance of subway electrical equipment.

[0021] The advantages of additional aspects of the invention will be set forth more clearly in the following description or will be learned by practice of the invention. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart of the intelligent analysis and diagnosis method for partial discharge signals of subway electrical equipment according to an embodiment of the present invention.

[0024] Figure 2 This is a flowchart of the improved SGMD algorithm described in an embodiment of the present invention.

[0025] Figure 3 This is a framework diagram of the electrical equipment partial discharge fault diagnosis model according to an embodiment of the present invention.

[0026] Figure 4 This is a schematic diagram of the confusion matrix for test set classification according to an embodiment of the present invention. Detailed Implementation

[0027] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0028] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0029] It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as here.

[0030] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, and / or groups thereof.

[0031] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0032] To facilitate understanding of the present invention, the present invention will be further explained and described below with reference to the accompanying drawings and specific embodiments. However, the specific embodiments do not constitute a limitation on the embodiments of the present invention.

[0033] Those skilled in the art should understand that the accompanying drawings are merely schematic diagrams of embodiments, and the components in the drawings are not necessarily essential for implementing the present invention.

[0034] Example 1

[0035] In this embodiment 1, a smart analysis and diagnosis system for partial discharge signals of subway electrical equipment is first provided, including: an acquisition module for acquiring partial discharge signal data of electrical equipment; a processing module for processing the acquired partial discharge signal data using a pre-trained diagnostic model to obtain fault diagnosis results of electrical equipment; training the diagnostic model includes: collecting partial discharge fault data during the operation of subway electrical equipment to construct an original dataset; performing denoising preprocessing on the original dataset and extracting effective signals; performing feature analysis and extraction on the effective signals; using reinforcement learning to dynamically optimize the hyperparameters of LSTM; outputting the optimal hyperparameter combination and using a test set to evaluate the model performance to obtain a trained diagnostic model.

[0036] In this embodiment, as Figure 1 As shown, the above system realizes an intelligent analysis and diagnosis method for partial discharge signals of subway electrical equipment, including: acquiring partial discharge signal data of electrical equipment using an acquisition module; processing the acquired partial discharge signal data using a pre-trained diagnostic model using a processing module to obtain fault diagnosis results of electrical equipment; training the diagnostic model includes: collecting partial discharge fault data during the operation of subway electrical equipment to construct an original dataset; performing denoising preprocessing on the original dataset and extracting effective signals; performing feature analysis and extraction on the effective signals; dynamically optimizing the hyperparameters of LSTM using reinforcement learning; outputting the optimal hyperparameter combination and evaluating the model performance using a test set to obtain a trained diagnostic model.

[0037] The original dataset was preprocessed using wavelet denoising, and the signal was denoised using wavelet thresholding with db4 wavelet. Then, the effective signals of the dataset were extracted to reduce the impact of environmental interference. The effective modes of the denoised partial discharge signal were extracted using symplectic geometric mode decomposition. In order to optimize the decomposition effect, a sparse regularization term was introduced into the objective function of the symplectic geometric mode decomposition method. This term is suitable for the sparse characteristics of the partial discharge signal in the time and frequency domain, highlighting the main signal, suppressing noise, and reducing mode mixing.

[0038] Specifically, Figure 2 The flowchart of the improved SGMD algorithm provided in this embodiment is shown.

[0039] First, the raw partial discharge (PD) data is preprocessed using wavelet denoising. A db4 wavelet basis is selected, and the data is decomposed into 5 levels. After removing noise using a soft thresholding method, the signal is reconstructed. Then, the processed data x is subjected to sparse regularized optimized symplectic geometric decomposition. The specific decomposition steps are as follows:

[0040] 1) Construct the trajectory matrix

[0041] Let the original time series be x = {x1, x2, ..., xn}. n Let n be the length of the time series, and construct the trajectory matrix X.

[0042] 2) Construct the Hamilton matrix

[0043] Using the trajectory matrix X, we can calculate the covariance matrix A = X. T X, then the Hamiltonian matrix M can be constructed:

[0044]

[0045] And calculate W=M 2 , where W is also a Hamiltonian matrix.

[0046] 3) Calculate the symplectic geometric similarity transformation

[0047] Construct a symplectic orthogonal matrix Q that satisfies:

[0048]

[0049] In the formula, B is an upper triangular matrix.

[0050] 4) Calculate the symplectic geometric components

[0051] Calculate the coefficient matrix:

[0052]

[0053] Then, the reconstructed matrix Z is obtained using matrix S and the symplectic orthogonal vector matrix Q:

[0054] Z i =Q i S i

[0055] Obtain the single-component matrix Z i The reconstructed matrix is ​​obtained as Z = Z1 + Z2 + ... + Z d .

[0056] 5) Introduce sparse regularization optimization

[0057] To enhance the sparsity of the decomposition results, the objective function is optimized as follows:

[0058]

[0059] In the formula, the first term is the original objective function, ensuring the accuracy of the decomposition; the second term is the L1 regularization term, which promotes S... i Sparsity is used to filter out key feature components; λ is a regularization parameter that can be optimized through cross-validation.

[0060] Iterative solution

[0061] 1. Initialization:

[0062] Let t = 0, initialize the symplectic orthogonal matrix. sum coefficient matrix

[0063] 2. Using proximal gradient iteration:

[0064] A) Update the coefficient matrix S i :

[0065]

[0066] Use soft thresholding:

[0067]

[0068] B) Update the symplectic orthogonal vector matrix Q i Maintain symplectic orthogonality:

[0069]

[0070] C) Check the convergence condition; if it is satisfied... The iteration terminates if the iteration terminates; otherwise, return to step A.

[0071] 3. Output the optimal solution and the sparse representation of the reconstruction:

[0072]

[0073] In this embodiment, partial discharge (PD) signals from various devices are collected and labeled to construct a preliminary dataset. Time-frequency perturbation data augmentation techniques are then used to expand the sample size, ultimately forming a raw PD signal dataset with high time-frequency resolution and good engineering adaptability. The key extracted features are peak factor, kurtosis factor, impulse factor, high-frequency energy, spectral centroid, sample entropy, and energy entropy, which form a feature matrix. Among these, peak factor, kurtosis factor, and impulse factor are time-domain features; high-frequency energy and spectral centroid are frequency-domain spectral features; and sample entropy and energy entropy are entropy value features.

[0074] Specifically, signal reconstruction includes: converting matrix Z... i The element in the middle is defined as z ij (1≤i≤m, 1≤j≤d). Let d * =min(m,d),m * =max(m,d) and 1≤i≤m, 1≤j≤d, if m<d, we have otherwise Single-component reconstruction matrix Z i The expression for converting it to a one-dimensional time series is:

[0075]

[0076] Then use sparsity to Divide into N groups, and sum the results of each group to obtain N modal SGCs. h (n), 1≤h≤N and the remaining residual term g (N+1) (n).

[0077] Subsequently, feature extraction is performed on the extracted effective signals. Due to the special characteristics of urban rail transit, the signals vary with the periodic changes in load. Therefore, selecting appropriate features can help improve the accuracy of diagnosis. This invention chooses to combine multiple features for comprehensive analysis. The seven selected features form a feature matrix, namely: Crest Factor (CF), Kurtosis (K), Impulse Factor (IF), High-Frequency Energy (HFE), Spectral Centroid (SC), Sample Entropy (SampEn), and Energy Entropy (EE). The calculation formulas for each feature index are listed below.

[0078] In the formula, X max It is the peak value of the signal, X rms It is the root mean square value of the signal; X i For signal samples, The mean of the signal; X is the average absolute value of the signal; f f is the amplitude of the Fourier transform spectrum of the signal. c is the high-frequency threshold; f is the frequency component; m is the embedding dimension; r is the similarity tolerance; B m Let A be the number of subsequences of length m whose Euclidean distance is less than r. m P represents the number of subsequences of length m+1 whose Euclidean distance is less than r. i E represents the normalized energy of the signal in the i-th frequency band. i Let M be the energy of the i-th frequency band, and M be the total number of frequency bands. The calculation formulas for each parameter are as follows:

[0079]

[0080] In this embodiment, the nearest neighbor policy is selected to optimize the parameters of the initial reinforcement learning. The "state-action-reward" range of the PPO is defined, and the loss function and related parameters of the PPO are defined. The extracted feature matrix is ​​divided into training and test sets in a 7:3 ratio. By updating the policy network and value network of the PPO, the LSTM hyperparameter sampling is continuously adjusted and trained to output the optimal LSTM hyperparameter and weight combination. Under the reinforcement learning framework, the LSTM hyperparameters and their training states are regarded as the state variables of the PPO, the action is defined as the adjustment method of the LSTM hyperparameters, and the reward function is set as the weighted sum of the LSTM recognition accuracy and the hyperparameter regularization term.

[0081] Specifically, LSTM is an improved recurrent neural network that can effectively capture long-term dependencies in time-series data and alleviate the vanishing gradient problem through input, forget, and output gating mechanisms, making it particularly suitable for processing highly nonlinear high-frequency partial discharge signals. However, traditional LSTM requires manual determination of hyperparameter combinations, which can easily lead to local optima, thus affecting recognition performance.

[0082] To improve model performance, this embodiment uses the PPO algorithm to optimize LSTM hyperparameters. PPO dynamically adjusts parameters through reinforcement learning, efficiently explores the parameter space, and utilizes entropy regularization S(π) to maintain exploration capability and avoid getting trapped in local optima. The specific steps are as follows:

[0083] Define the PPO optimization problem

[0084] (a) State (S) t )

[0085] Within the reinforcement learning framework, the hyperparameters of the LSTM and its training state are considered as the state variables of the PPO, defined as follows:

[0086] S t= {learningrate, hidden size, batch size, dropoutrate, number of training epochs, current loss value}

[0087] Where: learning rate is the learning rate, which determines the gradient update step size; hidden size is the number of neurons in the LSTM hidden layer, which affects model complexity; batch size is the batch size, which affects training stability; dropout rate is used for regularization to prevent overfitting; training epochs is the number of epochs in the current training; and the current loss value is used to measure the performance of the LSTM under the current hyperparameter combination.

[0088] (b) Action t )

[0089] Action is defined as the way to adjust the LSTM hyperparameters:

[0090] a t ={Δlearningrate,Δhiddensize,ΔBatch Size,Δdropoutrate}

[0091] PPO uses the policy network π θ (a|s) generates a new combination of hyperparameters, and the step size is adjusted by the policy network for adaptive optimization.

[0092] (c) Reward function (Reward, R) t )

[0093] The reward function is set as a weighted sum of the LSTM recognition accuracy (ACC) and the hyperparameter regularization term (C), defined as follows:

[0094] R t =ACC-κ||θ||1-β1max(0,HH) max )-β2max(0,η-η max )

[0095] In the formula, ACC is the classification accuracy on the test set, used to measure the model's performance; κ||θ||1 is the L1 regularization term to reduce redundant weights and optimize the network structure; the last two terms constrain the hidden layer size H and the learning rate η to ensure that the model is not too complex or unstable; β1 and β2 are penalty coefficients. R is then calculated. t The cumulative reward calculation formula is as follows:

[0096] G t =R t +γR t+1 +γ 2 R t+2+,,,+γ T-t R T

[0097] In the formula, γ is a discount factor used to control the degree of attention paid to long-term rewards during PPO training.

[0098] Initialize the PPO and LSTM network:

[0099] Initialize the policy network π of PPO θ (a|s) and value network The hyperparameters of the LSTM are randomly initialized, and an entropy regularization term S(π) is set to control the explore-exploitation balance. The loss function of PPO is set as follows:

[0100] L PPO =L Policy -c1L value +c2S(π)

[0101] The formula for the policy loss function is as follows:

[0102]

[0103] δ t =R t +γV(s t+1 )-V(s t )

[0104] In the formula, r t (θ) is the ratio of the new strategy to the old strategy, used to measure the degree of change of the new strategy relative to the old strategy; A t λ is the advantage function used to measure the relative performance of the current policy compared to the baseline policy; λ is the GAE parameter used to control the smoothness of the advantage estimation; T is the termination time of a trajectory; ε is the pruning coefficient used to control the policy update magnitude; c1 is the value function weight, and c2 is the entropy coefficient; L value The value function loss measures the prediction error of the value network, and its formula is as follows:

[0105] L value =E[(V φ (s t )-R t ) 2 ]

[0106] PPO samples LSTM hyperparameters and trains the model.

[0107] Using the current strategy π θ (a t |S t Sample new LSTM hyperparameter combinations, train the LSTM network under the current hyperparameter combinations, and calculate the corresponding reward value R. t .

[0108] Update PPO policy network and value network

[0109] Update PPO policy network π θ (a|s) determines how to adjust the LSTM hyperparameters in the next step, where θ is the parameter of the policy network, and α is the update speed and learning rate.

[0110] θ′=θ+α▽ θ L PPO

[0111] Updating the value network To predict the current state and accumulate rewards over the long term.

[0112] φ′=φ-α▽ φ L value

[0113] 5) Iterate through the PPO process to adjust the LSTM hyperparameters until the stopping condition is met.

[0114] By continuously optimizing the PPO policy network and adjusting the LSTM hyperparameters, the recognition accuracy is improved. When the model performance meets the requirements or the number of iterations reaches the upper limit, the optimal hyperparameters are output.

[0115] like Figure 4 The image shows the confusion matrix for the test set classification provided in this embodiment.

[0116] In this embodiment, taking a transformer as an example, four typical partial discharge datasets of transformers are obtained, and the feature matrix is ​​extracted as the input of the diagnostic model. The model is trained through the training set, and the main parameters are set as shown in Tables 1 and 2 below.

[0117] Table 1. Setting of PPO-related parameters

[0118]

[0119] Table 2. LSTM-related parameter settings

[0120]

[0121] The training model was then evaluated using a test set. The confusion matrix analysis showed that the intelligent diagnostic method for partial discharge signals in subway electrical equipment exhibited high classification accuracy in identifying transformer partial discharge fault types. Only a very small number of samples showed identification errors, and these errors were all within acceptable limits, not significantly affecting the overall diagnostic results. This demonstrates that the method in this embodiment can effectively distinguish different types of partial discharge signals, possessing strong stability and robustness. It further verifies the accuracy and reliability of the proposed method in diagnosing partial discharge faults in subway electrical equipment in practical applications.

[0122] Example 2

[0123] This embodiment 2 provides a non-transitory computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, they implement the intelligent analysis and diagnosis method for partial discharge signals of subway electrical equipment as described above. The method includes:

[0124] Acquire partial discharge signal data of electrical equipment;

[0125] The acquired partial discharge signal data is processed using a pre-trained diagnostic model to obtain electrical equipment fault diagnosis results. The training of the diagnostic model includes: collecting partial discharge fault data during the operation of subway electrical equipment to construct an original dataset; performing denoising preprocessing on the original dataset and extracting effective signals; performing feature analysis and extraction on the effective signals; using reinforcement learning to dynamically optimize the hyperparameters of the LSTM; outputting the optimal hyperparameter combination and using a test set to evaluate the model performance to obtain the trained diagnostic model.

[0126] Example 3

[0127] This embodiment 3 provides a computer device, including a memory and a processor, wherein the processor and the memory communicate with each other, and the memory stores program instructions that can be executed by the processor. The processor calls the program instructions to execute the intelligent analysis and diagnosis method for partial discharge signals of subway electrical equipment as described above, the method including:

[0128] Acquire partial discharge signal data of electrical equipment;

[0129] The acquired partial discharge signal data is processed using a pre-trained diagnostic model to obtain electrical equipment fault diagnosis results. The training of the diagnostic model includes: collecting partial discharge fault data during the operation of subway electrical equipment to construct an original dataset; performing denoising preprocessing on the original dataset and extracting effective signals; performing feature analysis and extraction on the effective signals; using reinforcement learning to dynamically optimize the hyperparameters of the LSTM; outputting the optimal hyperparameter combination and using a test set to evaluate the model performance to obtain the trained diagnostic model.

[0130] Example 4

[0131] This embodiment 4 provides an electronic device, including: a processor, a memory, and a computer program; wherein, the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes instructions to implement the intelligent analysis and diagnosis method for partial discharge signals of subway electrical equipment as described above. The method includes:

[0132] Acquire partial discharge signal data of electrical equipment;

[0133] The acquired partial discharge signal data is processed using a pre-trained diagnostic model to obtain electrical equipment fault diagnosis results. The training of the diagnostic model includes: collecting partial discharge fault data during the operation of subway electrical equipment to construct an original dataset; performing denoising preprocessing on the original dataset and extracting effective signals; performing feature analysis and extraction on the effective signals; using reinforcement learning to dynamically optimize the hyperparameters of the LSTM; outputting the optimal hyperparameter combination and using a test set to evaluate the model performance to obtain the trained diagnostic model.

[0134] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0135] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0136] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0137] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment, whereby a series of operational steps are performed to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0138] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that, based on the technical solutions disclosed in the present invention, various modifications or variations that can be made by those skilled in the art without creative effort should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent analysis and diagnosis of partial discharge signals in subway electrical equipment, characterized in that, include: Acquire partial discharge signal data of electrical equipment; The acquired partial discharge signal data is processed using a pre-trained diagnostic model to obtain the fault diagnosis results of electrical equipment. Training the diagnostic model includes: collecting partial discharge fault data during the operation of subway electrical equipment to construct an original dataset; performing denoising preprocessing on the original dataset and extracting effective signals; performing feature analysis and extraction on the effective signals; using reinforcement learning to dynamically optimize the hyperparameters of the LSTM; outputting the optimal hyperparameter combination and using a test set to evaluate the model performance to obtain the trained diagnostic model.

2. The intelligent analysis and diagnosis method for partial discharge signals of subway electrical equipment according to claim 1, characterized in that, The original dataset was preprocessed using wavelet denoising, and the signal was denoised using wavelet thresholding with db4 wavelet. Then, the effective signals of the dataset were extracted to reduce the impact of environmental interference. The effective modes of the denoised partial discharge signal were extracted using symplectic geometric mode decomposition. In order to optimize the decomposition effect, a sparse regularization term was introduced into the objective function of the symplectic geometric mode decomposition method. This term is suitable for the sparse characteristics of the partial discharge signal in the time and frequency domain, highlighting the main signal, suppressing noise, and reducing mode mixing.

3. The intelligent analysis and diagnosis method for partial discharge signals of subway electrical equipment according to claim 1, characterized in that, Partial discharge signals from various devices were collected and labeled to construct a preliminary dataset. Data augmentation techniques such as time-frequency perturbation were used to expand the number of samples, ultimately forming a raw dataset of partial discharge signals with high time-frequency resolution and good engineering adaptability.

4. The intelligent analysis and diagnosis method for partial discharge signals of subway electrical equipment according to claim 1, characterized in that, The key features extracted are peak factor, kurtosis factor, impulse factor, high-frequency energy, spectral centroid, sample entropy, and energy entropy, and a feature matrix is ​​established. Among them, peak factor, kurtosis factor, and impulse factor are time-domain features, high-frequency energy and spectral centroid are frequency-domain spectral features, and sample entropy and energy entropy are entropy value features.

5. The intelligent analysis and diagnosis method for partial discharge signals of subway electrical equipment according to claim 1, characterized in that, The nearest neighbor policy is selected to optimize the parameters of the initial reinforcement learning. The "state-action-reward" range of PPO is defined, and the loss function and related parameters of PPO are defined. The extracted feature matrix is ​​divided into training set and test set in a 7:3 ratio. The LSTM hyperparameter sampling is continuously adjusted and trained by updating the policy network and value network of PPO, so as to output the optimal combination of LSTM hyperparameters and weights.

6. The intelligent analysis and diagnosis method for partial discharge signals of subway electrical equipment according to claim 5, characterized in that, Within the reinforcement learning framework, the hyperparameters of the LSTM and its training state are considered as state variables of the PPO, the action is defined as the adjustment method of the LSTM hyperparameters, and the reward function is set as the weighted sum of the LSTM recognition accuracy and the hyperparameter regularization term.

7. A smart analysis and diagnosis system for partial discharge signals of subway electrical equipment, characterized in that, include: The acquisition module is used to acquire partial discharge signal data of electrical equipment; The processing module is used to process the acquired partial discharge signal data using a pre-trained diagnostic model to obtain the fault diagnosis results of electrical equipment. Training the diagnostic model includes: collecting partial discharge fault data during the operation of subway electrical equipment to construct an original dataset; performing denoising preprocessing on the original dataset and extracting effective signals; performing feature analysis and extraction on the effective signals; using reinforcement learning to dynamically optimize the hyperparameters of the LSTM; outputting the optimal hyperparameter combination and using a test set to evaluate the model performance to obtain the trained diagnostic model.

8. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement the intelligent analysis and diagnosis method for partial discharge signals of subway electrical equipment as described in any one of claims 1-6.

9. A computer device, characterized in that, The method includes a memory and a processor, which communicate with each other. The memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the intelligent analysis and diagnosis method for partial discharge signals of subway electrical equipment as described in any one of claims 1-6.

10. An electronic device, characterized in that, include: The device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions that implement the intelligent analysis and diagnosis method for partial discharge signals of subway electrical equipment as described in any one of claims 1-6.

Images