A charging pile fault diagnosis method based on network model

Abstract

This invention discloses a charging pile fault diagnosis method based on a network model, belonging to the field of fault diagnosis technology. The method includes: extracting single-modal features from preprocessed operational data; concatenating and fusing these single-modal features; constructing a spatiotemporal graph neural network model; generating node-level features after spatiotemporal feature extraction based on the fused single-modal features; classifying faults based on node-level features through group collaborative fault diagnosis; and providing feedback and optimization based on the fault classification results. This invention achieves accurate characterization of complex operating states of charging piles and improves the accuracy and global applicability of fault classification. The overall invention, from multimodal data processing to group collaborative analysis and dynamic optimization, forms a complete, accurate, and efficient charging pile fault diagnosis process, significantly improving accuracy, robustness, and applicability.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology, and in particular to a fault diagnosis method for charging piles based on a network model. Background Technology

[0002] With the rapid popularization of new energy vehicles, charging piles, as a critical infrastructure, are increasing in number and usage frequency. However, charging piles are prone to failure due to various factors during long-term operation, such as hardware aging, software anomalies, or external environmental influences. These failures not only affect the lifespan of the charging piles themselves but may also reduce charging efficiency and even pose safety hazards. Therefore, the diagnosis and maintenance of charging pile failures are of significant practical importance. Currently, charging pile failure diagnosis mainly relies on traditional methods, including rule-based diagnostic methods and statistical analysis-based predictive models. Rule-based methods achieve diagnosis by manually setting fault feature thresholds or logical rules, but this method relies too heavily on human experience and struggles to handle the implicit relationships between features when faced with complex multimodal data, resulting in low diagnostic accuracy and robustness. On the other hand, statistical analysis-based predictive models, such as Support Vector Machines (SVM) and Random Forests (RF), while capable of fault prediction through data training, are still insufficient when dealing with the high dimensionality, temporal sequence, and complex spatial correlations of charging pile operation data. Furthermore, these methods typically ignore the collaborative relationships between charging stations and fail to fully utilize group collaborative information, thus limiting their ability to perform global fault analysis and accurate classification.

[0003] In recent years, deep learning-based diagnostic methods have gained increasing attention due to their shortcomings in addressing the deficiencies of existing technologies. Graph Neural Networks (GNNs), in particular, have become a research hotspot due to their advantages in handling complex topological relationships and spatiotemporal data. However, existing GNN-based fault diagnosis methods still have limitations. First, traditional GNNs are mostly used for static graph structures, making it difficult to adapt to the temporal characteristics and dynamic changes of charging pile operation data, resulting in inaccurate capture of spatiotemporal dependencies. Second, current research largely focuses on single-modal data (such as voltage and current), failing to effectively integrate the characteristics of multimodal operation data and thus failing to comprehensively reflect the operating status of charging piles. Therefore, in the field of charging pile fault diagnosis, there is an urgent need for a diagnostic method that can comprehensively consider multimodal features, spatiotemporal correlations, and group collaborative relationships to improve the accuracy of fault classification and the reliability of diagnostic results. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a charging pile fault diagnosis method based on a network model, which solves the problems in the prior art of being unable to adapt to the temporal characteristics of charging pile operation data, being unable to effectively integrate multimodal data, and ignoring spatiotemporal correlation.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a charging pile fault diagnosis method based on a network model, comprising: collecting charging pile operation data and preprocessing the collected operation data; extracting single-modal features based on the preprocessed operation data and fusing the single-modal features by concatenation; constructing a spatiotemporal graph neural network model and generating node-level features after spatiotemporal feature extraction based on the fused single-modal features; classifying the node-level features for faults through group collaborative fault diagnosis; and providing feedback and optimization of the diagnosis results based on the fault classification results.

[0008] As a preferred embodiment of the charging pile fault diagnosis method based on the network model described in this invention, the specific steps for collecting the charging pile's operational data are as follows:

[0009] The charging pile collects electrical signals, temperature signals, vibration signals, sound signals, and image data through its internal sensors and external data acquisition equipment.

[0010] As a preferred embodiment of the charging pile fault diagnosis method based on the network model described in this invention, the preprocessing of the collected operational data includes the following specific steps:

[0011] The collected operational data is processed through data cleaning, data synchronization and alignment, and data normalization.

[0012] As a preferred embodiment of the charging pile fault diagnosis method based on the network model described in this invention, the step of extracting single-modal features based on preprocessed operating data includes the following specific steps.

[0013] Based on multi-scale entropy, signal complexity features are extracted through signal segmentation, multi-scale interpretation, entropy calculation, and feature aggregation. The expression is as follows:

[0014]

[0015] Among them, F elec For the multi-scale complexity characteristics of electrical signals, S l Let x be the sample entropy at the l-th scale. l Let T be the signal at the l-th scale, β be the nonlinear adjustment parameter, and T be the signal at the l-th scale. lLet t represent the time range corresponding to the l-th scale, L represent the number of layers in the multi-scale decomposition, t represent time, and l represent the scale number in the multi-scale analysis.

[0016] Based on variational mode decomposition, modal energy features are extracted through signal decomposition, modal energy calculation, and modal energy distribution feature extraction. The expression is as follows:

[0017]

[0018] Among them, F vib E represents the modal energy characteristics of the vibration signal. k For the energy of the k-th IMF, PSD k Let α be the power spectral density of the k-th IMF, where k represents the k-th mode generated by the vibration signal in variational mode decomposition, K represents the total number of modes, and α is the smoothing parameter.

[0019] Based on the Choi-Williams distribution, time-frequency distribution features are extracted through signal framing, calculation of the autocorrelation function, calculation of time-frequency distribution, and feature aggregation. The expression is as follows:

[0020]

[0021] Among them, F sound Let S(t,f) represent the energy concentration characteristics of the sound signal, where S(t,f) is the time-frequency distribution of the sound signal, f is the frequency, and σ is the energy concentration characteristic of the sound signal. f f is the frequency smoothing parameter. max and f min Frequency range;

[0022] High-dimensional image features are extracted based on convolutional neural networks, expressed as follows:

[0023] F image =ReLU(W*I+b);

[0024] Among them, F image Let I be the image feature vector, W be the convolution kernel matrix, I be the high-dimensional image, b be the bias term of the image feature vector, and * be the convolution operation.

[0025] As a preferred embodiment of the charging pile fault diagnosis method based on the network model described in this invention, the specific steps for splicing and fusing single-modal features are as follows:

[0026] The extracted electrical signal features, vibration signal features, sound signal features, and image features are concatenated into a joint feature vector F in element-wise order;

[0027] By constructing a dynamic attention map and decoupling from a nonlinear tensor, a fused feature is generated, expressed as:

[0028]

[0029] Among them, F ′ i T represents the i-th feature after fusion. ′ ijk This represents the interaction value between the i, j, and k features, where i, j, and k represent the indexes of the features.

[0030] As a preferred embodiment of the charging pile fault diagnosis method based on a network model according to the present invention, the specific steps for constructing the spatiotemporal graph neural network model, and generating node-level features after spatiotemporal feature extraction based on the fused single-modal features, are as follows:

[0031] Nodes are defined based on charging pile components. The feature vector of a node is represented by fused single-modal features. Edges are constructed based on the physical connection relationship and spatial proximity relationship between components, as well as the dynamic evolution in the time dimension.

[0032] By aggregating the features of a node itself and its neighboring nodes through graph convolution operations, local spatial dependencies are extracted. By aggregating the features of adjacent nodes and simultaneously utilizing edge weights and dynamic normalization adjustment of features, spatial representation of node features is achieved.

[0033] By combining information from historical time steps to capture temporal dependencies, temporal features are extracted through temporal convolution. After spatial and temporal feature extraction, the two are integrated to generate node-level features after spatiotemporal feature extraction, expressed as:

[0034]

[0035] in, These are the node-level features extracted from the spatiotemporal features. For node v i Time characteristics, Aggregate the spatial features of neighboring nodes. To integrate learnable weight matrices of spatial and temporal features, For feature splicing operations, is the bias term, and g is the hyperbolic tangent activation function.

[0036] As a preferred embodiment of the charging pile fault diagnosis method based on the network model described in this invention, the specific steps of classifying and analyzing node-level features through group collaborative fault diagnosis are as follows:

[0037] Spatiotemporal graph neural networks extract the spatiotemporal features of nodes and calculate the mutual influence weights between nodes by combining the feature similarity and topological relationship between nodes through a dynamic neighborhood collaborative weight mechanism.

[0038] Based on the label propagation mechanism, the fault information of the initially labeled nodes is gradually propagated to the unlabeled nodes. At the same time, feature guidance factors are introduced to directly participate in classification optimization, and the probability distribution of fault categories of each node is obtained through iterative calculation.

[0039] Based on the actual application scenario, the fault categories are divided into normal state, minor fault, moderate fault and severe fault.

[0040] As a preferred embodiment of the charging pile fault diagnosis method based on the network model described in this invention, the specific steps for feedback and optimization of diagnostic results based on fault classification and analysis results are as follows:

[0041] Error assessment is performed on the fault classification results, weighted cross-entropy loss is calculated to quantify the classification error, and the set of misclassified nodes and their distribution characteristics are screened to analyze the categories and graph topological locations in the error set.

[0042] Based on a gradient-guided feature importance evaluation method, the feature matrix is ​​optimized, and the quality of effective features is enhanced by eliminating low-contribution feature dimensions and neighborhood-weighted smoothing.

[0043] By combining error feedback to adjust model parameters, optimizing the smoothing parameters and feature guidance factors of collaborative weights, adding regularization terms to control model complexity, and updating the objective optimization function to iteratively improve classification performance.

[0044] In a second aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the charging pile fault diagnosis method based on a network model as described in the first aspect of the present invention.

[0045] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the charging pile fault diagnosis method based on a network model as described in the first aspect of the present invention.

[0046] The beneficial effects of this invention are as follows: By extracting and fusing features from multimodal data, it achieves accurate characterization of the complex operating states of charging piles; through the construction of a spatiotemporal graph neural network and the generation of node-level features, it captures the spatiotemporal correlation characteristics of the equipment during operation; and through group collaborative diagnosis and dynamic optimization, it improves the accuracy and global applicability of fault classification. The overall invention, from multimodal data processing to group collaborative analysis and then to dynamic optimization, forms a complete, accurate, and efficient charging pile fault diagnosis process, significantly improving accuracy, robustness, and applicability. Attached Figure Description

[0047] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. 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.

[0048] Figure 1 This is a flowchart of the charging pile fault diagnosis method based on a network model in Example 1.

[0049] Figure 2 This is a schematic diagram of the node-level features generated after spatiotemporal feature extraction in Example 1. Detailed Implementation

[0050] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0051] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0052] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0053] Example 1, referring to Figure 1 and Figure 2 This is the first embodiment of the present invention, which provides a charging pile fault diagnosis method based on a network model, including the following steps:

[0054] S1: Collect the operating data of the charging piles and preprocess the collected operating data;

[0055] Furthermore, electrical signals, temperature signals, vibration signals, sound signals, and image data are collected through the internal sensors of the charging pile and external data acquisition equipment.

[0056] It should be noted that the electrical signals include current, voltage, and power, with a sampling frequency of 1kHz to ensure high-precision capture of dynamic changes during operation; the temperature signal has a sampling frequency of 1Hz, suitable for monitoring slowly changing temperatures; the vibration signal has a sampling frequency of 10kHz, used to capture subtle mechanical vibrations; the sound signal has a sampling frequency of 44.1kHz; and the image data acquisition frequency is 1 frame per second, suitable for monitoring the appearance of the equipment and the environment.

[0057] The collected operational data is processed through data cleaning, data synchronization and alignment, and data normalization.

[0058] It should be noted that data cleaning includes removing noise from electrical signals, vibration signals, and sound signals during acquisition. For data lost during the acquisition process, linear interpolation is used to fill in missing values, and outlier detection and removal are performed. Data synchronization and alignment include timestamp standardization, interpolation alignment, and data alignment.

[0059] S2: Based on the preprocessed running data, perform single-modal feature extraction, and then merge the single-modal features after splicing them together;

[0060] Based on multi-scale entropy, signal complexity features are extracted through signal segmentation, multi-scale interpretation, entropy calculation, and feature aggregation. The expression is as follows:

[0061]

[0062] Among them, F elec For the multi-scale complexity characteristics of electrical signals, S l Let x be the sample entropy at the l-th scale. l Let T be the signal at the l-th scale, β be the nonlinear adjustment parameter, and T be the signal at the l-th scale. l Let t represent the time range corresponding to the l-th scale, L represent the number of layers in the multi-scale decomposition, t represent time, and l represent the scale number in the multi-scale analysis.

[0063] It should be noted that signal segmentation divides electrical signals into time segments, forming multiple fixed-length interfaces; multi-scale processing uses a method based on segmented mean to perform multi-scale processing on the signal, with the signal at each scale generated through downsampling; at each scale, the sample entropy of the signal is calculated to measure the signal complexity, expressed as:

[0064]

[0065] Among them, S l Let p be the sample entropy at the l-th scale. l,j M is the probability of the signal in the j-th state at the l-th scale. lLet represent the number of states at the l-th scale, where l is the scale number in the multi-scale entropy analysis, and j represents the state number at the l-th scale.

[0066] The final multi-scale entropy feature is obtained by weighted averaging of the entropy values ​​at all scales.

[0067] Based on variational mode decomposition, modal energy features are extracted through signal decomposition, modal energy calculation, and modal energy distribution feature extraction. The expression is as follows:

[0068]

[0069] Among them, F vib E represents the modal energy characteristics of the vibration signal. k For the energy of the k-th IMF, PSD k Let α be the power spectral density of the k-th IMF, where k represents the k-th mode generated by the vibration signal in variational mode decomposition, K represents the total number of modes, and α is the smoothing parameter.

[0070] It should be noted that signal decomposition involves inputting the vibration signal into the VKM algorithm, decomposing it into K intrinsic mode functions (IMFs). IMFs represent the vibration modes of the signal. The energy characteristics of the IMFs in the time domain are calculated using the following expression:

[0071]

[0072] Among them, E k For the energy characteristics of the IMF in the time domain, u k (t) represents the kth IMF, where t0 and t1 are time ranges;

[0073] Based on the Choi-Williams distribution, time-frequency distribution features are extracted through signal framing, calculation of the autocorrelation function, calculation of time-frequency distribution, and feature aggregation. The expression is as follows:

[0074]

[0075] Among them, F sound Let S(t,f) represent the energy concentration characteristics of the sound signal, where S(t,f) is the time-frequency distribution of the sound signal, f is the frequency, and σ is the energy concentration characteristic of the sound signal. f f is the frequency smoothing parameter. max and f min Frequency range;

[0076] It should be noted that signal framing involves dividing the audio signal into several frames, and calculating the autocorrelation function for each frame. The expression is as follows:

[0077]

[0078] Where R(τ) is the autocorrelation function, A(t) is the sound signal, and τ is the time delay;

[0079] The time-frequency distribution is calculated using the Choi-Williams distribution formula, and the expression is:

[0080]

[0081] Where σ is the time smoothing parameter;

[0082] Feature aggregation calculates the energy concentration characteristics of the time-frequency distribution to obtain the final sound signal characteristics;

[0083] High-dimensional image features are extracted based on convolutional neural networks, expressed as follows:

[0084] F image =ReLU(W*I+b);

[0085] Among them, F image Let I be the image feature vector, W be the convolution kernel matrix, I be the high-dimensional image, b be the bias term of the image feature vector, and * be the convolution operation.

[0086] The extracted electrical signal features, vibration signal features, sound signal features, and image features are concatenated into a joint feature vector F in element-wise order;

[0087] By constructing a dynamic attention map and decoupling from a nonlinear tensor, a fused feature is generated, expressed as:

[0088]

[0089] Among them, F ′ i T represents the i-th feature after fusion. ′ ijk This represents the interaction value between the i, j, and k features, where i, j, and k represent the indexes of the features.

[0090] It should be noted that, based on the concatenated joint features, a dynamic attention map between modalities is constructed, and the mutual influence between modalities is built through a nonlinear function, the expression of which is:

[0091]

[0092] Among them, A ij Represents the dynamic attention weights between the i-th feature and the j-th feature, |F i -F j | represents the absolute difference between the two features, and σ(F) is the normalization factor;

[0093] Based on the dynamic attention graph, a joint feature tensor is constructed, and the tensor is nonlinearly decoupled to generate a decoupled feature tensor. The final fused feature is generated by weighting the elements of the tensor.

[0094] S3: Construct a spatiotemporal graph neural network model, and generate node-level features after spatiotemporal feature extraction based on the fused single-modal features;

[0095] Nodes are defined based on charging pile components. The feature vector of a node is represented by fused single-modal features. Edges are constructed based on the physical connection relationship and spatial proximity relationship between components, as well as the dynamic evolution in the time dimension.

[0096] By aggregating the features of a node itself and its neighboring nodes through graph convolution operations, local spatial dependencies are extracted. By aggregating the features of adjacent nodes and simultaneously utilizing edge weights and dynamic feature normalization adjustments, a spatial representation of node features is achieved, expressed as:

[0097]

[0098] in, For node v i Spatial features at layer l+1 and time t, w ij (t) represents v at time t. i and v j edge weight, F j (t) is the node v j The characteristics of time t, Let F be a learnable weight matrix in the spatial direction. j (t)∥2 is the node v j The L2 norm of the eigenvectors Here, g is the bias term, and g is the hyperbolic tangent activation function.

[0099] By combining information from historical time steps to capture temporal dependencies, temporal features are extracted through temporal convolution, expressed as:

[0100]

[0101] in, Let v be the node at time t. i The time features at layer l+1 Let v be the node at time q. i Features Let be the learnable weight matrix in the time direction, and Δt be the length of the time window. For node v i The L1 norm of the eigenvectors at time τ, where η is an adjustment parameter controlling the smoothness of the time features. Here, h is the bias term, h is the rectified linear unit, and q represents the time step from time t-Δt to t;

[0102] After spatial and temporal feature extraction, the two are integrated to generate node-level features after spatiotemporal feature extraction, expressed as:

[0103]

[0104] in, These are the node-level features extracted from the spatiotemporal features. For node v i Time characteristics, Aggregate the spatial features of neighboring nodes. To integrate learnable weight matrices of spatial and temporal features, For feature splicing operations, is the bias term, and g is the hyperbolic tangent activation function.

[0105] S4: Through group collaborative fault diagnosis, fault classification is performed on node-level features;

[0106] Spatiotemporal graph neural networks extract the spatiotemporal features of nodes and calculate the mutual influence weights between nodes by combining the feature similarity and topological relationship between nodes through a dynamic neighborhood collaborative weight mechanism.

[0107] Based on the label propagation mechanism, the fault information of the initially labeled nodes is gradually propagated to the unlabeled nodes. At the same time, feature guidance factors are introduced to directly participate in classification optimization, and the probability distribution of fault categories of each node is obtained through iterative calculation.

[0108] Based on the actual application scenario, the fault categories are divided into normal state, minor fault, moderate fault and severe fault.

[0109] It should be noted that the fault classification is defined based on the degree of characteristic changes in the equipment's operating status. It is determined through feature extraction and engineering experience combined with machine learning algorithms. Each type of fault corresponds to a specific set of characteristic patterns and probability distributions.

[0110] S5: Based on the fault classification results, provide feedback and optimization of diagnostic results.

[0111] Error assessment is performed on the fault classification results, weighted cross-entropy loss is calculated to quantify the classification error, and the set of misclassified nodes and their distribution characteristics are screened to analyze the categories and graph topological locations in the error set.

[0112] Based on a gradient-guided feature importance evaluation method, the feature matrix is ​​optimized, and the quality of effective features is enhanced by eliminating low-contribution feature dimensions and neighborhood-weighted smoothing.

[0113] The model parameters are adjusted by combining error feedback, including optimizing the smoothing parameters and feature guidance factors of the collaborative weights, adding regularization terms to control model complexity, and updating the objective optimization function to iteratively improve classification performance.

[0114] It should be noted that the re-evaluation of the optimized classification results generates a diagnostic report, which includes overall accuracy, category distribution, error node analysis, and optimization suggestions. This enables more accurate diagnostic results and continuous model optimization, thereby improving the reliability and robustness of fault classification.

[0115] This embodiment also provides a computer device applicable to the charging pile fault diagnosis method based on a network model, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the charging pile fault diagnosis method based on a network model as proposed in the above embodiment.

[0116] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0117] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the charging pile fault diagnosis method based on a network model as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0118] In summary, this invention achieves accurate characterization of the complex operating states of charging piles through: feature extraction and fusion of multimodal data; captures the spatiotemporal correlation characteristics of equipment operation through the construction of a spatiotemporal graph neural network and node-level feature generation; and improves the accuracy and global applicability of fault classification through group collaborative diagnosis and dynamic optimization. The overall invention solution, from multimodal data processing to group collaborative analysis and dynamic optimization, forms a complete, accurate, and efficient charging pile fault diagnosis process, significantly improving accuracy, robustness, and applicability.

[0119] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A charging pile fault diagnosis method based on a network model, characterized in that: include, Collect operational data from charging piles and preprocess the collected data. Based on the preprocessed runtime data, single-modal features are extracted, and then concatenated and fused. The specific steps are as follows. Based on multi-scale entropy, signal complexity features are extracted through signal segmentation, multi-scale interpretation, entropy calculation, and feature aggregation. The expression is as follows: ； in, The multi-scale complexity characteristics of electrical signals For the Sample entropy at each scale, For the Signals at various scales It is a non-linear adjustment parameter. For the The time range corresponding to each scale The number of layers in the multi-scale decomposition. Indicates time, Indicates the scale number in multiscale analysis; Based on variational mode decomposition, modal energy features are extracted through signal decomposition, modal energy calculation, and modal energy distribution feature extraction. The expression is as follows: ； in, The modal energy characteristics of the vibration signal, For the The energy of the IMF For the The power spectral density of each IMF This represents the first vibration signal generated in variational mode decomposition. One modality, Indicates the total number of modes. For smoothing parameters; Based on the Choi-Williams distribution, time-frequency distribution features are extracted through signal framing, calculation of the autocorrelation function, calculation of time-frequency distribution, and feature aggregation. The expression is as follows: ； in, The energy concentration characteristic of sound signals. This represents the time-frequency distribution of the sound signal. For frequency, For frequency smoothing parameters, and Frequency range; High-dimensional image features are extracted based on convolutional neural networks, expressed as follows: ； in, For image feature vectors, The convolution kernel matrix, For high-dimensional images, This is the bias term of the image feature vector. This is a convolution operation; The extracted electrical signal features, vibration signal features, sound signal features, and image features are concatenated into a joint feature vector in element-wise order. ; By constructing a dynamic attention map and decoupling from a nonlinear tensor, a fused feature is generated, expressed as: ； in, Indicates the fused first One characteristic, Indicates the first , and Interaction values ​​between features , and Index representing the feature; A spatiotemporal graph neural network model is constructed, and node-level features are generated after spatiotemporal feature extraction based on the fused single-modal features. Fault classification is performed on node-level features through group collaborative fault diagnosis; Based on the fault classification results, diagnostic results are fed back and optimized.

2. The charging pile fault diagnosis method based on a network model as described in claim 1, characterized in that: The specific steps for collecting the operating data of the charging piles are as follows: The charging pile collects electrical signals, temperature signals, vibration signals, sound signals, and image data through its internal sensors and external data acquisition equipment.

3. The charging pile fault diagnosis method based on a network model as described in claim 2, characterized in that: The preprocessing of the collected operational data involves the following specific steps: The collected operational data is processed through data cleaning, data synchronization and alignment, and data normalization.

4. The charging pile fault diagnosis method based on a network model as described in claim 1, characterized in that: The construction of the spatiotemporal graph neural network model involves generating node-level features after spatiotemporal feature extraction based on the fused single-modal features. The specific steps are as follows: Nodes are defined based on charging pile components. The feature vector of a node is represented by fused single-modal features. Edges are constructed based on the physical connection relationship and spatial proximity relationship between components, as well as the dynamic evolution in the time dimension. By aggregating the features of a node itself and its neighboring nodes through graph convolution operations, local spatial dependencies are extracted. By aggregating the features of adjacent nodes and simultaneously utilizing edge weights and dynamic normalization adjustment of features, spatial representation of node features is achieved. By combining information from historical time steps to capture temporal dependencies, temporal features are extracted through temporal convolution. After spatial and temporal feature extraction, the two are integrated to generate node-level features after spatiotemporal feature extraction, expressed as: ； in, These are the node-level features extracted from the spatiotemporal features. For nodes Time characteristics, Aggregate the spatial features of neighboring nodes. To integrate learnable weight matrices of spatial and temporal features, For feature splicing operations, is the bias term, and g is the hyperbolic tangent activation function.

5. The charging pile fault diagnosis method based on a network model as described in claim 4, characterized in that: The method of classifying faults based on node-level features through group collaborative fault diagnosis involves the following specific steps. Spatiotemporal graph neural networks extract the spatiotemporal features of nodes and calculate the mutual influence weights between nodes by combining the feature similarity and topological relationship between nodes through a dynamic neighborhood collaborative weighting mechanism. Based on the label propagation mechanism, the fault information of the initially labeled nodes is gradually propagated to the unlabeled nodes. At the same time, feature guidance factors are introduced to directly participate in classification optimization, and the probability distribution of fault categories of each node is obtained through iterative calculation. Based on the actual application scenario, the fault categories are divided into normal state, minor fault, moderate fault and severe fault.

6. The charging pile fault diagnosis method based on a network model as described in claim 5, characterized in that: The diagnostic results feedback and optimization based on the fault classification results are described in the following steps. Error assessment is performed on the fault classification results, weighted cross-entropy loss is calculated to quantify the classification error, and the set of misclassified nodes and their distribution characteristics are screened to analyze the categories and graph topological locations in the error set. Based on a gradient-guided feature importance evaluation method, the feature matrix is ​​optimized, and the quality of effective features is enhanced by eliminating low-contribution feature dimensions and neighborhood-weighted smoothing. By combining error feedback to adjust model parameters, optimizing the smoothing parameters and feature guidance factors of collaborative weights, adding regularization terms to control model complexity, and updating the objective optimization function to iteratively improve classification performance.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the charging pile fault diagnosis method based on the network model as described in any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the charging pile fault diagnosis method based on the network model as described in any one of claims 1 to 6.

Images