Intelligent fault diagnosis method and system for electrical equipment

Abstract

The invention relates to the technical field of electrical equipment fault diagnosis, in particular to an intelligent fault diagnosis method and system for electrical equipment, and the method comprises the steps: constructing a multi-dimensional tensor model, uniformly fusing the equipment state information, electrical distance weighted connection and phase dynamic coupling relation, and extracting an abnormal propagation mode through high-order singular value decomposition; designing a space-time-frequency coupling interference stripping mechanism, and combining structure guide disturbance deconstruction, multi-scale dictionary learning and sparse low-rank decomposition to accurately separate transmissible and non-transmissible interferences; reconstructing a fault trajectory based on a generative adversarial mechanism, coupling a graph structure dynamic encoder, a topology consistency discriminator and a time controllable generator, and restoring a real propagation path; and finally, tensor semantic compression, a three-view graph neural network and fault label back projection interpretation are integrated through a multi-source semantic fusion mechanism. According to the method, cross-space-time and cross-structure fault diagnosis and traceability are realized, and the accuracy and interpretability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electrical equipment fault diagnosis, and in particular to an intelligent fault diagnosis method and system for electrical equipment. Background Art

[0002] The research background of intelligent fault diagnosis technology for electrical equipment focuses on the objective development needs of the continuous expansion of the scale of modern power systems and the significant increase in the complexity of equipment coupling. With the increasing complexity of power grid topology and the diversification of equipment status monitoring dimensions, traditional diagnostic methods based on single indicator thresholds or local state analysis face challenges in their ability to characterize fault propagation across time and space.

[0003] A Chinese invention patent with announcement number CN119881542B discloses a hybrid sensing-based intelligent location and diagnosis method for optical fiber composite faults in distribution networks. The method involves acquiring and preprocessing optical fiber and electrical sensor data, extracting time-frequency domain features through a multi-channel convolutional neural network, using an attention mechanism to calculate dynamic weight coefficients for feature fusion, constructing a dynamic fault propagation graph, and combining graph convolution and causal convolution to extract spatiotemporal features to achieve fault location and type recognition.

[0004] The strong electrical distance correlation, dynamic phase coupling and nonlinear interference superposition between electrical equipment require the diagnostic model to have the ability to efficiently integrate multi-dimensional heterogeneous data and accurately model the laws of topological propagation. At the same time, the industrial site's demand for explainability of fault tracing and proactive operation and maintenance decision support further promotes the evolution of diagnostic technology towards an intelligent collaborative paradigm that integrates physical mechanisms and data-driven. In this context, it is urgent to build a multi-dimensional analysis framework that can uniformly characterize the evolution of equipment status, topological connection weights and phase dynamic responses, develop interference stripping and trajectory reconstruction technologies to cope with data loss and noise interference scenarios, and achieve high-precision and strong interpretability of diagnostic results through cross-domain semantic fusion mechanisms, thereby forming an intelligent diagnostic system to support the safe operation of the power grid. Summary of the Invention

[0005] The purpose of the present invention is to address the problems existing in the background technology and to provide an intelligent fault diagnosis method and system for electrical equipment.

[0006] The technical solution of the present invention is an intelligent fault diagnosis method for electrical equipment, comprising the following specific implementation steps: S1. Collect electrical equipment operating data through a sensor network, construct a normalized state vector, construct a topological weight matrix based on electrical distance and physical connection, define a phase coupling matrix using phase difference cosine response, fuse the normalized state vector, topological weight matrix and phase coupling matrix, construct a three-dimensional dynamic tensor for topological evolution modeling, and extract the dominant anomaly propagation mode through high-order singular value decomposition; S2. Extract the node disturbance intensity and change rate sequence from the three-dimensional dynamic tensor, obtain the basis function response and residual through wavelet residual multi-scale decomposition, construct the disturbance response similarity matrix using topology guidance, identify the propagation interference, generate the interference frequency band set, and remove the system disturbance through low-rank sparse decomposition of the residual matrix, finally generating the purified topological evolution tensor; S3: Based on the topological evolution tensor, a sliding window topology-aware adjacency matrix is ​​constructed to generate a structural embedding. The generator is used to combine device static features and time offset to recover missing trajectories. Discriminator adversarial training is used to ensure trajectory structure consistency. Candidate paths are fused based on similarity and propagation scores to optimize the candidate trajectory reconstruction output. S4. The dynamic trajectory of the equipment is compressed into a semantic vector by weighting the response field intensity. The three-view adjacency matrix is ​​constructed by integrating structural topology, semantic similarity, and attribute homogeneity. The node embedding representation is generated through the graph attention network, and the failure probability is mapped and the gradient-sensitive dimension is located to trace back to the source.

[0007] Preferably, the construction process of the three-dimensional dynamic tensor is as follows: Deploy a sensor network to collect the operating data of any node of the electrical equipment at a certain moment, including voltage amplitude, current amplitude, phase angle, frequency and temperature, and construct a normalized state vector; Based on physical connection marks and electrical distances, a topological weight matrix between devices is defined using a natural exponential decay function to quantify connection density and impact intensity. According to the phase coupling characteristics between electrical devices, a phase response matrix is ​​defined, whose elements are driven by the phase difference between devices. The phase coupling response value is calculated using the cosine function to nonlinearly compress the phase difference. Fuse the normalized state vector, topological weight matrix, and phase coupling matrix to define a three-dimensional tensor : ; in, Represents the three-dimensional dynamic state tensor within N nodes and T time steps; Indicates that at time t k , the comprehensive state influence strength of node i on node j; Represents the state vector Perform transpose operation; Represents the strength weight of the electrical connection relationship between devices i and j; Indicates that at time t k , the phase coupling response value between device i and device j.

[0008] Preferably, the process of extracting the dominant anomaly propagation mode is: ; in, represents the singular value of the rth principal component; Represents the eigenvector of the rth mode in the device dimension; Represents the eigenvector of the rth pattern in the connection relationship dimension; represents the eigenvector of the rth mode in the time dimension; R represents the number of retained main propagation modes; Represents the outer product operation of a tensor.

[0009] Preferably, the propagation interference identification process is: Extract the comprehensive influence of disturbance of each node from the three-dimensional dynamic tensor, introduce the first-order difference disturbance change rate, project the tensor onto the node-time plane to obtain the disturbance sequence, and quantify the overall disturbance intensity and perturbation intensity: ; ; in, Represents node i at time t k the perceived overall intensity of the disturbance; represents the disturbance change rate of node i at adjacent moments; The perturbation sequence Decompose into multi-scale perturbation subspace and non-interpretable residual, and construct a sparse representation of node perturbation: ; Where M represents the total number of decomposed scale levels; represents the mth perturbation basis function; represents the response coefficient of node i to the m-th layer scale perturbation basis function; represents the disturbance residual; Based on the similarity propagation matching of the node perturbation basis function components and the adjacent node states, the topological information is introduced to construct the response similarity matrix: ; like If the value is greater than the set threshold, the frequency band is marked. For propagation interference, record it in the interference frequency band set ; in, represents the perturbation basis function component of device i at scale m; represents the perturbation basis function component of device j at scale m; represents the topology-guided similarity of the perturbation morphology between device i and device j at scale m; represents the perturbation shape similarity metric function; Represents the electrical topology connection weight between device node i and node j.

[0010] Preferably, the generation process of the purified topological evolution tensor is as follows: Collect node perturbation residuals Combine two-dimensional matrices, constrain the nuclear norm and L1 norm through low-rank sparse joint decomposition, ensure stability through time-differential gradient, strip off systematic interference to capture shared trends, and retain sparse local mutation anomalies: ; in, represents the sparse term adjustment coefficient; represents the trend smoothing adjustment coefficient; represents the nuclear norm, which is the sum of the singular values ​​of the matrix; represents the L1 norm; represents the time difference gradient of L; L represents the low rank term of the residual matrix; S represents the sparse term of the residual matrix; Represents the perturbation residual matrix, each element , that is, the residual value; Remove the identified interference components from the original evolution tensor and construct the purified tensor: ; in, represents the purified device topology evolution tensor; represents the low-rank background perturbation term.

[0011] Preferably, the optimization process of candidate trajectory reconstruction output is: Construct a sliding window topology-aware adjacency matrix to quantify the propagation probability between nodes, evolve the topology based on the time window, and combine the node feature input graph attention network to generate structural embedding to capture the local structural semantics and propagation impact; A generator network integrating temporal logic is constructed based on the extracted structural embedding representation. The network inputs the historical structural embedding vector, the device static feature vector, and the time offset variable, and outputs the device state reconstruction vector. The network uses a graph neural network and a gating unit to collaboratively model the temporal evolution trend and topological semantics, restoring the evolution trajectory consistent with the physical logic of the missing segment. A topological consistency discriminator is constructed to evaluate the structural consistency between the generated trajectory and the real propagation path through an adversarial training mechanism. The reconstruction state and adjacency matrix are input to output the authenticity probability, and the generator is optimized to output a physically logically reasonable trajectory. The confidence score of each candidate trajectory is calculated based on structural similarity and propagation flow intensity, and the weighted fusion output is the optimal state estimate, which is the reconstructed output of the candidate trajectory.

[0012] Preferably, the adversarial loss function of the adversarial training mechanism is: ; in, represents the adversarial loss function; represents the real observed device state vector; represents the pseudo sample device state vector generated by the generator; and represents the expected value calculation; A represents the topological adjacency matrix at the current time point.

[0013] Preferably, the process of constructing the three-view adjacency matrix is ​​as follows: The semantic compression vector of each device node i is defined as: ; Weight Defined as: ; in, Represents the semantic compression vector of device i; represents the semantic contribution weight of time step t to device i; represents the reconstructed state vector of device i at time t; Represents the L2 norm of the vector; represents the average state of the entire cycle of device i; T represents the set number of time steps; Construct the three-view adjacency matrix: Structure diagram G S : Multi-level electrical topology evolution structure, adjacency matrix A S ; Semantic graph G E : Calculate the cosine similarity of any two device semantic vectors: ; Property graph G A : Meta-attribute matching to construct binary adjacency: ; Construct the fused adjacency matrix: ; Node embedding using graph attention network: ; in, Represents the similarity between nodes i and j in the semantic space; Indicates performing a transpose operation on the semantic compression vector of node i; Represents the attribute graph adjacency matrix; I represents the indicator function, which is 1 when the condition is met and 0 otherwise; type(i) represents the type identifier of node i; Represents the adjacency matrix of the multi-view fusion graph; 、 and represents the weight factor; A S Represents the adjacency matrix of the topological structure graph; A E Represents the semantic similarity graph, that is, the similarity of device evolution trends; A A Represents a device attribute graph; represents the fused semantic representation of node i after the graph attention network; represents the k-th layer of graph attention network.

[0014] Preferably, the positioning process of tracing back the source of the positioning gradient sensitive dimension is: Define the final classification head as a linear projector: ; Set the threshold Th. If , then the diagnostic tag is triggered; in, represents the failure probability vector of node i, with a total of C categories; Represents the weight matrix of the classification head, mapping the fused features to C-dimensional output; Represents the classification head bias; The inverse activation weight analysis is used to locate the most diagnostically weighted dimension in the fused semantic vector, that is, to calculate the gradient of the classification output with respect to each dimension in the input semantic vector: ; Based on this, we find the top k dimensions with the highest activation and associate them with their upstream adjacent nodes in the graph structure to assist in identifying potential fault sources. in, Represents the semantic vector F of the fault category c i Sensitivity of each dimension.

[0015] The technical solution of the present invention is an intelligent fault diagnosis system for electrical equipment, which is used to execute the above-mentioned intelligent fault diagnosis method for electrical equipment, including: The multi-level topology evolution modeling module collects operating data of electrical equipment based on the deployed sensor network. Based on the operating data of electrical equipment and the system topology structure, it constructs a multi-level electrical topology evolution tensor to accurately characterize the state changes of equipment and the abnormal propagation path within the topology; The heterogeneous interference deconstruction and multi-scale disturbance stripping module strips non-fault-related interference from the original state tensor, retaining the key disturbance features that truly reflect the fault evolution; A structure-guided fault evolution trajectory reconstruction module is used to reconstruct a complete, coherent, and structurally consistent fault evolution trajectory based on the state data after disturbance stripping; The multi-graph semantic fusion and fault output module performs deep semantic fusion analysis on the reconstructed state trajectory and completes the final fault type determination and reverse interpretation.

[0016] Compared with the prior art, the above technical solution of the present invention has the following beneficial technical effects: The present invention designs an intelligent fault diagnosis method and system for electrical equipment. Through multi-dimensional tensor modeling, it unifies and integrates electrical equipment status information, topological weights and phase dynamic coupling relationships, significantly improving the ability to characterize abnormal propagation in complex coupling scenarios; interference stripping technology based on space-time-frequency coupling mechanism is combined with topology-guided similarity analysis to accurately separate propagation and non-propagation interference sources, and greatly suppress misjudgment caused by non-fault disturbances; it integrates graph structure dynamic encoder, topological consistency discriminator and time-controllable generator to achieve physical and logical consistent reconstruction of fault trajectory, effectively solving the diagnostic blind spots of data missing or ambiguous scenarios; multi-source semantic fusion mechanism integrates tensor compression, three-view graph neural network and gradient back projection interpretation, and simultaneously improves accuracy and interpretability in cross-time, space and structure diagnosis, providing intelligent decision support for electrical equipment fault tracing and active operation and maintenance that takes into account both theoretical rigor and engineering practicality. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a flow chart of an intelligent fault diagnosis method for electrical equipment proposed by the present invention; Figure 2 This is a system architecture diagram of an intelligent fault diagnosis system for electrical equipment proposed by the present invention. DETAILED DESCRIPTION

[0018] Example 1, as Figure 1 As shown, the present invention proposes an intelligent fault diagnosis method for electrical equipment, including the following specific implementation steps: S1. Construct a multidimensional tensor model to integrate multidimensional state information, electrical distance weighted connections, and phase dynamic coupling into a spatiotemporal evolution tensor. By decomposing the tensor, we can reveal abnormal propagation and dynamically and comprehensively reflect the complex coupling relationship between electrical equipment and its temporal changes, thus achieving efficient capture of abnormal fault information and characterization of propagation paths. The specific implementation process is as follows: S11, deploy a sensor network, collect data on the electrical device node i at time t k The running status indicator is used to construct the state vector : ; in, Represents the electrical equipment node i at time t k The multi-dimensional normalized operating state vector of Denotes device i at time t k The voltage amplitude reflects the stability and integrity of the power supply; Denotes device i at time t k The current amplitude is an important indicator for measuring equipment load and power consumption behavior; Denotes device i at time t kThe phase angle is used to analyze the phase difference between voltage and current; Denotes device i at time t k The frequency reflects the stability of the power grid frequency; Denotes device i at time t k The temperature is an important external status signal of the equipment's healthy operation; Represents a transpose operation; S12. Define the topological relationship weight matrix between devices based on the physical connection and electrical distance of the devices , where each element: ; in, The electrical connection strength weight between devices i and j is used to express the connection density and influence of each device in the topology. Represents the electrical distance between devices i and j; Represents the distance attenuation control parameter, which controls The degree of exponential decay of the mid-range term; Indicates the physical connection flag, binary value (0 or 1), 1 if there is a physical connection between device i and device j, otherwise 0; represents the natural exponential function; S13. Define the phase response matrix based on the phase coupling characteristics between electrical equipment , whose elements are driven by the inter-device phase difference: ; in, Indicates that at time t k , the phase coupling response value between device i and device j reflects the phase coordination relationship between the two devices. The cosine function is used to perform nonlinear compression on the phase difference. The closer the value is to 1, the more synchronized the phase is. Represents devices i and j at time t k The phase angle difference; N represents the total number of electrical equipment nodes, that is, the number of devices to be monitored, modeled and diagnosed in the entire electrical network; S14. The above normalized state vector, topological weight matrix and phase coupling matrix are combined to define a three-dimensional tensor: ; Specifically, the tensor calculation process is: for each time point t k , traverse all device pairs (i, j) and calculate the normalized state vector 、 、 The product of , forming a tensor element , in the continuous time sliding window, update , realize topology dynamic evolution modeling; in, Represents a three-dimensional dynamic state tensor at N nodes and T time steps, which is used to model the multivariate state evolution process and abnormal propagation law in high-dimensional electrical systems; Indicates that at time t k , the comprehensive state influence strength of node i on node j; Represents the state vector Perform transpose operation; S15. Through tensors The dominant anomaly propagation mode is extracted by using the tensor decomposition method (this example uses high-order singular value decomposition HOSVD) based on the time series changes of ; in, Represents the singular value of the rth principal component, which measures the weight or significance of the propagation mode in the overall tensor. The larger the value, the more important the mode. The eigenvector of the rth mode in the device dimension (node ​​dimension) indicates which devices are most affected by the mode; It represents the eigenvector of the rth pattern in the connection relationship dimension, reflecting which connection relationships the propagation is mainly concentrated on; represents the eigenvector of the rth mode in the time dimension, describing the temporal evolution of the abnormal propagation mode; R represents the number of retained main propagation modes, which is usually set based on the cumulative contribution rate of singular values ​​(95% energy coverage is used in this embodiment) to determine how many high-order propagation modes are retained for anomaly extraction; Represents the outer product operation of a tensor, which is used to combine vectors of different dimensions to form a tensor approximation.

[0019] S2. Design an interference modeling and stripping method based on the space-time-frequency coupling mechanism, through structure-guided disturbance deconstruction + multi-scale disturbance dictionary learning + sparse low-rank anomaly recognition. The specific implementation process is as follows: S21. From the constructed three-dimensional dynamic state tensor , extract the comprehensive impact of disturbance on each node (i.e., the aggregate value of topological propagation disturbance): ; To further characterize the volatility of the disturbance, the disturbance change rate (i.e., first-order difference) is introduced: ; Based on this: the tensor is projected from three dimensions to the node-time plane to obtain a perturbation sequence with physical meaning; in, Represents node i at time t k the perceived overall intensity of the disturbance; represents the perturbation change rate (perturbation intensity) of node i at adjacent moments; S22, the perturbation sequence Input to the data-driven wavelet residual deconstructor, based on the data-driven perturbation dictionary learning mechanism, the perturbation sequence Decompose into multi-scale perturbation subspace and non-interpretable residual, and construct a sparse representation of node perturbation: ; Where M represents the total number of scale levels of wavelet / perturbation decomposition; represents the mth disturbance basis function, which characterizes the typical disturbance form at this scale and is adaptively learned from historical samples through data-driven or sparse learning (such as transformer switching signals, lightning harmonic waveforms, etc.); It represents the response coefficient of node i to the m-th scale perturbation basis function, reflecting the significance or activity of the perturbation form on the current node; represents the perturbation residual, i.e., the “unexplained component” that does not belong to the existing perturbation dictionary; S23. Perform similarity propagation matching on the disturbance feature of node i and the state of its adjacent nodes, introduce structural topology information to enhance the accuracy of disturbance identification, and construct a disturbance response similarity matrix: ; in, represents the perturbation basis function component of device i at scale m; represents the perturbation basis function component of device j at scale m; represents the topology-guided similarity of the perturbation morphology between device i and device j at scale m; represents the perturbation shape similarity measurement function, and this embodiment adopts cosine similarity; Represents the electrical topological connection weight between device node i and node j; If there are multiple disturbance components between a group of strongly connected devices (i.e.: greater than the set threshold) appear at the same time, the frequency band is marked For propagation interference, record it in the interference frequency band set ; S24, collect the perturbation residual term of each node i , combined into a two-dimensional matrix ; Then perform low-rank sparse joint decomposition: ; Based on this, the estimated low-rank term L is stripped from the perturbation response of the original tensor, retaining a more realistic non-propagation mutation anomaly; in, represents the sparse term adjustment coefficient; represents the trend smoothing adjustment coefficient; represents the nuclear norm, that is, the sum of the singular values ​​of the matrix, which is used to constrain the low rank of L; represents the L1 norm, the sum of the absolute values ​​of all elements, used to constrain the sparsity of S; represents the time-differential gradient of L, that is, the changing trend of L on the time axis, which is used to constrain its stability (i.e., there should be no sharp jumps); L represents the low-rank term of the residual matrix, which captures the shared disturbance trend existing in all nodes, such as thermal drift, long-period noise, equipment aging trend and other systematic factors; S represents the sparse term of the residual matrix, that is, sparse local jump abnormal signals, such as short-term arcing, contact faults, electromagnetic interference, etc. Represents the perturbation residual matrix, the rows represent nodes, the columns represent time; each element , that is, the residual value; S25. Finally, integrate the above analysis results, remove the identified interference components from the original evolution tensor, and construct the purified tensor: ; in, represents the purified device topology evolution tensor; Represents the low-rank background disturbance term, that is, device i at time t k The systematic long-term trend or periodic disturbance component of the market.

[0020] S3, based on the purified topological evolution tensor A novel trajectory reconstruction method is constructed by integrating structural priors, time dependence, and generative adversarial mechanisms. This method couples a graph-guided dynamic encoder, a structural consistency discriminator, and a time-controllable trajectory generator. This method accurately reconstructs missing or ambiguous segments of the electrical equipment's historical state while restoring the true fault propagation trajectory and path evolution sequence. The specific implementation process is as follows: S31. By constructing dynamic structural embedding, we extract the evolutionary relationship of the actual propagation paths between devices over time, thereby providing accurate topological guidance for subsequent trajectory generation. Specifically: For each moment t k Construct a sliding window topology-aware adjacency matrix: ; in, represents the propagation probability from node i to node j in the topological adjacency matrix at time point k, i.e., the possible impact strength of the failure from node i to j at that moment; Represents the width of the sliding time window, that is, the time span considered from the current moment forward, which is used to construct time-aware topological evolution; The adjacency matrix and node features at each moment are input into the graph neural network module (this embodiment uses a graph attention network) to generate a structural embedding: ; in, represents the state feature matrix of each device at time k, , N is the total number of devices, d is the state dimension; It represents the structural embedding representation extracted under the Graph Attention Network (GAT), reflecting the local structural semantics and context propagation influence of each node at the current moment; S32, based on the extracted structural embedding representation, a generator network based on fusion time logic is used to restore the evolution trajectory on state missing or abnormal segments to make it conform to the physical evolution order and propagation direction, and define the generator function The output is: ; in, represents the state reconstruction vector of device i output by the trajectory generator at time t; The structural embedding vector of device i at time t-1 reflects the semantics of the topological propagation state of the device before the current moment; The static structural feature vector representing device i, including device type (transformer, switch, etc.), importance rating, historical fault frequency, etc., is used for personalized trajectory generation; Indicates that the current reconstruction time point t is relative to the starting point t of the fault segment s Time offset, used to simulate time evolution trend; It should be noted that the generator It is the core module of the structure-guided fault evolution trajectory reconstruction method, which aims to dynamically restore the state evolution trajectory of electrical equipment in the missing or abnormal section by using historical structure embedding information, equipment static characteristics and time offset variables. Its input includes the structural embedding vector of the previous moment , device static attribute vector and the time offset between the current moment and the missing starting point of the trajectory By fusing the structural semantics extracted by the Graph Attention Network (GAT) with the modeling of temporal evolution trends by GRU-like gating units, the generator can output state estimates that are continuous, structurally consistent, and have temporal evolution logic; S33. To ensure that the generated trajectory is structurally consistent with the real propagation path, we constructed a topological consistency discriminator. , its core goal is to improve the structural credibility of the trajectory in the generative-discriminative adversarial mechanism, specifically: Define the discriminant probability: ; Its output value indicates whether the reconstruction state conforms to the propagation logic of the current topology. If it is close to 1, it means that the reconstruction state is consistent with the actual topology propagation trend and conforms to the fault evolution logic. If it is close to 0, it means that the state may deviate from the network propagation path and is a false reconstruction. The training objective is to minimize the following adversarial loss: ; in, Represents the credibility score (between 0 and 1) output by the discriminator at time t, indicating the reconstruction state Whether it conforms to the topological propagation law; Denotes the adversarial loss function, which is used to optimize the training process of the generator and discriminator to make the generator's generated results more realistic and credible. The loss function consists of two parts: the first is to make the discriminator mark the real samples as "true" to the greatest extent (output close to 1), and the second is to make the discriminator accept the generated samples to the minimum extent (output close to 0); Represents the real observed device state vector, which is used to train the positive samples in the discriminator; represents the pseudo sample device state vector generated by the generator, which is used to train the negative samples in the discriminator; and represents the expected value calculation, that is, solving the average loss of the sample; A represents the topological adjacency matrix at the current time point, which is used to constrain whether the generated state is consistent with the physical connection between devices and serves as a structural prior injection; It should be noted that the topological consistency discriminator This discriminant mechanism is used to evaluate the consistency between fault evolution trajectories and the actual propagation structure of electrical networks. Its core function is to ensure that the state sequence constructed by the trajectory generator is not only numerically reasonable but also structurally consistent with the physical topology and propagation logic. The discriminator takes the generated state vector and the corresponding topological adjacency matrix at each moment as input and outputs a probability value representing the authenticity of the trajectory. Through adversarial training, the generated trajectory is forced to strictly adhere to the laws of network propagation while maintaining continuity. Specifically, the discriminator uses a structure-aware neural network to encode the topological features of the input state and combines the historical evolution semantics of the propagation path to extract the topological contours of the fault evolution, thereby effectively identifying trajectories that do not conform to physical logic, such as structural drift and forged paths. S34. Introduce a path confidence fusion mechanism to perform weighted optimization on multiple candidate trajectories based on structural similarity and propagation power scores. Specifically: Define the consistency scoring function: ; The final output fusion result: ; in, represents the reconstruction output of the rth path of the candidate trajectory; represents the consistency confidence score of the state estimate generated by the rth candidate trajectory of device i at time t; represents the topological adjacency probability vector of device i at time t, that is, the probability distribution of its outgoing edge propagation in the graph structure; It represents the propagation flow intensity score of device i at time t, which measures its structural activity or signal flux as a propagation source / mediator.

[0021] S4. Based on structure-guided fault trajectory reconstruction, and oriented towards the final diagnostic output goal, a multi-source semantic fusion and fault mode output method is constructed. This method integrates tensor semantic compression, three-view graph neural network fusion, and fault label back-projection interpretation mechanism to achieve cross-temporal, cross-indicator, and cross-structural information integration and reasoning in electrical systems. This ensures that the diagnostic results are not only accurate but also highly interpretable and generalizable. The specific implementation process is as follows: S41. Construct a tensor semantic compression mechanism based on response field strength weighting to compress the dynamic trajectories of multiple time steps and multiple devices into a joint semantic vector in the response domain, and extract the main line of semantic response of the device at the critical moment. Specifically: The semantic compression vector of each device node i is defined as: ; Weight Defined as: ; in, The semantic compression vector representing device i represents the comprehensive expression of the important state evolution of the device in the entire time period T, which is the input feature of the subsequent diagnosis model; represents the semantic contribution weight of time step t to device i, reflecting the degree of deviation of the state from its mean at that moment; represents the reconstructed state vector of device i at time t, which comes from the fault trajectory reconstruction output in step S3; Represents the L2 norm of the vector, that is, the Euclidean distance, which represents the degree of difference between the state at a certain time point and the mean; It represents the average state of the whole cycle of device i, which is an estimate of the center of the normal state; T represents the set time step number; S42. Construct a multi-view graph fusion mechanism to map structural topology, attribute homogeneity, and state semantic similarity into a unified graph representation, and learn the final fused representation based on the graph attention mechanism. Specifically: Construct the three-view adjacency matrix: Structure diagram G S : The multi-level electrical topology evolution structure from step S1, the adjacency matrix A S ; Semantic graph G E : Calculate the cosine similarity of any two device semantic vectors: ; Property graph G A : Construct binary adjacency by matching meta-attributes such as equipment category (such as transformer, circuit breaker), installation environment (indoor / outdoor), etc. ; Construct the fused adjacency matrix: ; Node embedding using graph attention network: ; in, Represents the similarity between nodes i and j in the semantic space. The larger the value, the more similar the state evolution. Indicates performing a transpose operation on the semantic compression vector of node i; When the attributes of devices i and j are the same (such as type, installation environment, etc.), it is 1, otherwise it is 0, indicating structural similarity; I represents the indicator function, which is 1 when the condition is met, otherwise it is 0; type(i) represents the type identifier of node i (such as circuit breaker, busbar, transformer, etc.); Represents the adjacency matrix of the multi-view fusion graph; 、 and represents the weight factor; A S Represents the adjacency matrix of the topological structure graph; A E Represents the semantic similarity graph, that is, the similarity of device evolution trends; A A Represents a device attribute graph; It represents the fused semantic representation of node i after the graph attention network, which is used for final fault identification; represents the k-th layer graph attention network, which is used to learn weighted information propagation between nodes and their neighbors; S43, after obtaining the fusion semantic representation F i After that, it is mapped to specific fault label output (multi-label, multi-level), and comes with a reverse interpretation mechanism to enhance the comprehensibility and traceability of the diagnosis results. Specifically: (1) Define the final classification head as a linear projector: ; Set the threshold Th. If , then the diagnostic tag is triggered; in, It represents the failure probability vector of node i, which has C categories, and each dimension represents the probability of occurrence of that category; Represents the weight matrix of the classification head, mapping the fused features to C-dimensional output; Represents the classification head bias; (2) Use reverse activation weight analysis to locate the most diagnostically weighted dimension in the fused semantic vector, that is, calculate the gradient of the classification output with respect to each dimension in the input semantic vector: ; Based on this, we find the top k dimensions with the highest activation and associate them with their upstream adjacent nodes in the graph structure to assist in identifying potential fault sources. in, Represents the semantic vector F of the fault category c i Sensitivity (gradient) of each dimension.

[0022] Example 2, as Figure 2 As shown, the present invention proposes an intelligent fault diagnosis system for electrical equipment, which is used to execute an intelligent fault diagnosis method for electrical equipment proposed in Example 1, including: a multi-level topology evolution modeling module, a heterogeneous interference deconstruction and multi-scale disturbance stripping module, a structure-guided fault evolution trajectory reconstruction module, and a multi-graph semantic fusion and fault output module.

[0023] The multi-level topology evolution modeling module collects operating data of electrical equipment based on the deployed sensor network. Based on the operating data of electrical equipment and the system topology structure, it constructs a multi-level electrical topology evolution tensor to accurately characterize the state changes of equipment and the abnormal propagation path within the topology; The heterogeneous interference deconstruction and multi-scale disturbance stripping module strips non-fault-related interference (including but not limited to scheduling fluctuations, operational disturbances, and temperature drift) from the original state tensor, retaining the key disturbance features that truly reflect the fault evolution; A structure-guided fault evolution trajectory reconstruction module is used to reconstruct a complete, coherent, and structurally consistent fault evolution trajectory based on the state data after disturbance stripping; The multi-graph semantic fusion and fault output module performs deep semantic fusion analysis on the reconstructed state trajectory and completes the final fault type determination and reverse interpretation.

[0024] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. An intelligent fault diagnosis method for electrical equipment, characterized in that: The specific implementation steps include the following: S1. Collect electrical equipment operating data through a sensor network, construct a normalized state vector, construct a topological weight matrix based on electrical distance and physical connection, define a phase coupling matrix using phase difference cosine response, fuse the normalized state vector, topological weight matrix and phase coupling matrix, construct a three-dimensional dynamic tensor for topological evolution modeling, and extract the dominant anomaly propagation mode through high-order singular value decomposition; S2. Extract the node disturbance intensity and change rate sequence from the three-dimensional dynamic tensor, obtain the basis function response and residual through wavelet residual multi-scale decomposition, construct the disturbance response similarity matrix using topology guidance, identify the propagation interference, generate the interference frequency band set, and remove the system disturbance through low-rank sparse decomposition of the residual matrix, finally generating the purified topological evolution tensor; S3: Based on the topological evolution tensor, a sliding window topology-aware adjacency matrix is ​​constructed to generate a structural embedding. The generator is used to combine device static features and time offset to recover missing trajectories. Discriminator adversarial training is used to ensure trajectory structure consistency. Candidate paths are fused based on similarity and propagation scores to optimize the candidate trajectory reconstruction output. S4. The dynamic trajectory of the equipment is compressed into a semantic vector by weighting the response field intensity. The three-view adjacency matrix is ​​constructed by integrating structural topology, semantic similarity, and attribute homogeneity. The node embedding representation is generated through the graph attention network, and the failure probability is mapped and the gradient-sensitive dimension is located to trace back to the source.

2. The intelligent fault diagnosis method for electrical equipment according to claim 1, characterized in that: The construction process of the three-dimensional dynamic tensor is as follows: Deploy a sensor network to collect the operating data of any node of the electrical equipment at a certain moment, including voltage amplitude, current amplitude, phase angle, frequency and temperature, and construct a normalized state vector; Based on physical connection marks and electrical distances, a topological weight matrix between devices is defined using a natural exponential decay function to quantify connection density and impact intensity. According to the phase coupling characteristics between electrical devices, a phase response matrix is ​​defined, whose elements are driven by the phase difference between devices. The phase coupling response value is calculated using the cosine function to nonlinearly compress the phase difference. Fuse the normalized state vector, topological weight matrix, and phase coupling matrix to define a three-dimensional tensor : ； in, Represents the three-dimensional dynamic state tensor within N nodes and T time steps; Indicates that at time t k , the comprehensive state influence strength of node i on node j; Represents the state vector Perform transpose operation; Represents the strength weight of the electrical connection relationship between devices i and j; Indicates that at time t k , the phase coupling response value between device i and device j.

3. The intelligent fault diagnosis method for electrical equipment according to claim 2, characterized in that: The process of extracting the dominant anomaly propagation pattern is as follows: ; in, represents the singular value of the rth principal component; Represents the eigenvector of the rth mode in the device dimension; Represents the eigenvector of the rth pattern in the connection relationship dimension; represents the eigenvector of the rth mode in the time dimension; R represents the number of retained main propagation modes; Represents the outer product operation of a tensor.

4. The intelligent fault diagnosis method for electrical equipment according to claim 3, characterized in that: The process of identifying propagation interference is as follows: Extract the comprehensive influence of disturbance of each node from the three-dimensional dynamic tensor, introduce the first-order difference disturbance change rate, project the tensor onto the node-time plane to obtain the disturbance sequence, and quantify the overall disturbance intensity and perturbation intensity: ； ； in, Represents node i at time t k the perceived overall intensity of the disturbance; represents the disturbance change rate of node i at adjacent moments; The perturbation sequence Decompose into multi-scale perturbation subspace and non-interpretable residual, and construct a sparse representation of node perturbation: ; Where M represents the total number of decomposed scale levels; represents the mth perturbation basis function; represents the response coefficient of node i to the m-th layer scale perturbation basis function; represents the disturbance residual; Based on the similarity propagation matching of the node perturbation basis function components and the adjacent node states, the topological information is introduced to construct the response similarity matrix: ； like If the value is greater than the set threshold, the frequency band is marked. For propagation interference, record it in the interference frequency band set ; in, represents the perturbation basis function component of device i at scale m; represents the perturbation basis function component of device j at scale m; represents the topology-guided similarity of the perturbation morphology between device i and device j at scale m; represents the perturbation shape similarity metric function; Represents the electrical topology connection weight between device node i and node j.

5. The intelligent fault diagnosis method for electrical equipment according to claim 4, characterized in that: The generation process of the purified topological evolution tensor is as follows: Collect node perturbation residuals Combine two-dimensional matrices, constrain the nuclear norm and L1 norm through low-rank sparse joint decomposition, ensure stability through time-differential gradient, strip off systematic interference to capture shared trends, and retain sparse local mutation anomalies: ； in, represents the sparse term adjustment coefficient; represents the trend smoothing adjustment coefficient; represents the nuclear norm, which is the sum of the singular values ​​of the matrix; represents the L1 norm; represents the time difference gradient of L; L represents the low rank term of the residual matrix; S represents the sparse term of the residual matrix; Represents the perturbation residual matrix, each element , that is, the residual value; Remove the identified interference components from the original evolution tensor and construct the purified tensor: ； in, represents the purified device topology evolution tensor; represents the low-rank background perturbation term.

6. The intelligent fault diagnosis method for electrical equipment according to claim 5, characterized in that: The optimization process of candidate trajectory reconstruction output is: Construct a sliding window topology-aware adjacency matrix to quantify the propagation probability between nodes, evolve the topology based on the time window, and combine the node feature input graph attention network to generate structural embedding to capture the local structural semantics and propagation impact; A generator network integrating temporal logic is constructed based on the extracted structural embedding representation. The network inputs the historical structural embedding vector, the device static feature vector, and the time offset variable, and outputs the device state reconstruction vector. The network uses a graph neural network and a gating unit to collaboratively model the temporal evolution trend and topological semantics, restoring the evolution trajectory consistent with the physical logic of the missing segment. A topological consistency discriminator is constructed to evaluate the structural consistency between the generated trajectory and the real propagation path through an adversarial training mechanism. The reconstruction state and adjacency matrix are input to output the authenticity probability, and the generator is optimized to output a physically logically reasonable trajectory. The confidence score of each candidate trajectory is calculated based on structural similarity and propagation flow intensity, and the weighted fusion output is the optimal state estimate, which is the reconstructed output of the candidate trajectory.

7. The intelligent fault diagnosis method for electrical equipment according to claim 6, characterized in that: The adversarial loss function of the adversarial training mechanism is: ； in, represents the adversarial loss function; Represents the real observed device state vector; represents the pseudo sample device state vector generated by the generator; and represents the expected value calculation; A represents the topological adjacency matrix at the current time point.

8. The intelligent fault diagnosis method for electrical equipment according to claim 7, characterized in that: The process of constructing the three-view adjacency matrix is ​​as follows: The semantic compression vector of each device node i is defined as: ; Weight Defined as: ; in, Represents the semantic compression vector of device i; represents the semantic contribution weight of time step t to device i; represents the reconstructed state vector of device i at time t; Represents the L2 norm of the vector; represents the average state of the entire cycle of device i; T represents the set number of time steps; Construct the three-view adjacency matrix: Structure diagram G S : Multi-level electrical topology evolution structure, adjacency matrix A S ; Semantic graph G E : Calculate the cosine similarity of any two device semantic vectors: ； Property graph G A : Meta-attribute matching to construct binary adjacency: ; Construct the fused adjacency matrix: ; Node embedding using graph attention network: ； in, Represents the similarity between nodes i and j in the semantic space; Indicates performing a transpose operation on the semantic compression vector of node i; Represents the attribute graph adjacency matrix; I represents the indicator function, which is 1 when the condition is met and 0 otherwise; type(i) represents the type identifier of node i; Represents the adjacency matrix of the multi-view fusion graph; 、 and represents the weight factor; A S Represents the adjacency matrix of the topological structure graph; A E Represents the semantic similarity graph, that is, the similarity of device evolution trends; A A Represents a device attribute graph; represents the fused semantic representation of node i after the graph attention network; represents the k-th layer of graph attention network.

9. The intelligent fault diagnosis method for electrical equipment according to claim 8, characterized in that: The process of tracing back the source of the localization gradient sensitive dimension is as follows: Define the final classification head as a linear projector: ; Set the threshold Th. If , then the diagnostic tag is triggered; in, represents the failure probability vector of node i, with a total of C categories; Represents the weight matrix of the classification head, mapping the fused features to C-dimensional output; Represents the classification head bias; The inverse activation weight analysis is used to locate the most diagnostically weighted dimension in the fused semantic vector, that is, to calculate the gradient of the classification output with respect to each dimension in the input semantic vector: ; Based on this, we find the top k dimensions with the highest activation and associate them with their upstream adjacent nodes in the graph structure to assist in identifying potential fault sources. in, Represents the semantic vector F of the fault category c i Sensitivity of each dimension.

10. An intelligent fault diagnosis system for electrical equipment, used to execute the intelligent fault diagnosis method for electrical equipment according to any one of claims 1 to 9, characterized in that: include: The multi-level topology evolution modeling module collects operating data of electrical equipment based on the deployed sensor network. Based on the operating data of electrical equipment and the system topology structure, it constructs a multi-level electrical topology evolution tensor to accurately characterize the state changes of equipment and the abnormal propagation path within the topology; The heterogeneous interference deconstruction and multi-scale disturbance stripping module strips non-fault-related interference from the original state tensor, retaining the key disturbance features that truly reflect the fault evolution; A structure-guided fault evolution trajectory reconstruction module is used to reconstruct a complete, coherent, and structurally consistent fault evolution trajectory based on the state data after disturbance stripping; The multi-graph semantic fusion and fault output module performs deep semantic fusion analysis on the reconstructed state trajectory and completes the final fault type determination and reverse interpretation.

Images