Intelligent operation and maintenance management method and system for power system

Abstract

The present application relates to the field of electric power operation and maintenance technology, and specifically discloses an intelligent operation and maintenance management method and system for an electric power system, which uses artificial intelligence technology based on deep learning to perform real-time monitoring and data analysis on the operating status parameters of each node of the electric power system, and respectively captures the time series joint feature representation of the node status parameters of each node, and performs a granular matching analysis between nodes to dig out the characteristic differences of the operating status of each node relative to the whole, thereby intelligently identifying abnormal nodes in the electric power system. In this way, the accuracy and efficiency of abnormality detection can be improved, the risk of power supply interruption caused by equipment failure can be reduced, the stable operation of the electric power system can be ensured, and the operation and maintenance process can be optimized.

Description

Technical Field

[0001] The present application relates to the field of power operation and maintenance technology, and more specifically, to an intelligent operation and maintenance management method and system for a power system. Background Art

[0002] As an important infrastructure of modern society, the stable, safe and efficient operation of the power system plays a vital role in economic development and social stability. With the rapid development of the power industry, the scale of the power system continues to expand, and its operation and maintenance work has become extremely complex and cumbersome.

[0003] Traditional power system operation and maintenance mainly relies on manual monitoring and regular inspections, and relies on a large amount of human resource investment. It is not only inefficient, but also has problems such as untimely fault detection and high operation and maintenance costs. Especially when dealing with large-scale and complex power networks, it is often impossible to quickly identify and properly handle potential abnormal conditions, which may lead to serious problems such as equipment failure and power outages, posing a substantial threat to the stable operation of the power system.

[0004] With the development of Internet of Things, big data and artificial intelligence technologies, the operation and maintenance management of power systems is gradually transforming towards intelligence. Therefore, an intelligent operation and maintenance management solution for power systems is expected. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides an intelligent operation and maintenance management method and system for an electric power system, which uses artificial intelligence technology based on deep learning to perform real-time monitoring and data analysis on the operating status parameters of each node of the electric power system, and respectively captures the time series joint feature representation of the node status parameters of each node, and performs granular matching analysis between nodes to dig out the characteristic differences of the operating status of each node relative to the whole, so as to intelligently identify abnormal nodes in the electric power system. In this way, the accuracy and efficiency of anomaly detection can be improved, the risk of power supply interruption caused by equipment failure can be reduced, the stable operation of the electric power system can be ensured, and the operation and maintenance process can be optimized.

[0006] Accordingly, according to one aspect of the present application, a method for intelligent operation and maintenance management of an electric power system is provided, which includes: obtaining a time series of node state parameters collected by a sensor network deployed at each node of the electric power system, wherein the node state parameters include current values, voltage values ​​and temperature values; respectively extracting parameter time series joint features of the time series of node state parameters of the respective nodes to obtain a sequence of node state parameter time series joint representation feature graphs; respectively performing feature enhancement on each node state parameter time series joint representation feature graph in the sequence of node state parameter time series joint representation feature graphs to obtain a sequence of enhanced node state parameter time series joint representation feature vectors; using each enhanced node state parameter time series joint representation feature vector in the sequence of enhanced node state parameter time series joint representation feature vectors as a query feature vector, inputting the query feature vector and the sequence of enhanced node state parameter time series joint representation feature vectors into a granular unidirectional attention interactive scanning matching module to obtain multiple unidirectional matching result representation vectors; and determining whether the node state of each node is abnormal based on the multiple unidirectional matching result representation vectors.

[0007] In the above-mentioned intelligent operation and maintenance management method of the power system, the parameter time series joint features of the time series of the node state parameters of the nodes are extracted respectively to obtain a sequence of node state parameter time series joint representation feature graphs, including: arranging the time series of the node state parameters of the nodes into a node state parameter time series joint matrix according to the time dimension and the parameter sample dimension respectively, and then passing the node state parameter time series feature extractor based on the void convolutional neural network model to obtain the sequence of node state parameter time series joint representation feature graphs.

[0008] In the above-mentioned intelligent operation and maintenance management method of the power system, each node state parameter time series joint representation feature graph in the sequence of the node state parameter time series joint representation feature graph is feature enhanced to obtain a sequence of enhanced node state parameter time series joint representation feature vectors, including: inputting each node state parameter time series joint representation feature graph in the sequence of the node state parameter time series joint representation feature graph into a feature attention selection enhancement module based on the compression-suppression structure to obtain a sequence of enhanced node state parameter time series joint representation feature graphs; calculating the global mean vector of each enhanced node state parameter time series joint representation feature graph in the sequence of the enhanced node state parameter time series joint representation feature graph to obtain a sequence of enhanced node state parameter time series joint representation feature vectors.

[0009] In the above-mentioned intelligent operation and maintenance management method for the power system, each node state parameter time series joint representation feature graph in the sequence of the node state parameter time series joint representation feature graph is respectively input into the feature attention selection and enhancement module based on the compression-suppression structure to obtain a sequence of enhanced node state parameter time series joint representation feature graphs, including: calculating the global mean of each feature matrix along the channel dimension of the node state parameter time series joint representation feature graph to obtain a node state parameter time series joint compression information representation vector; performing one-dimensional convolution encoding on the node state parameter time series joint compression information representation vector to obtain a node state parameter time series joint compression information association representation feature vector; cascading the node state parameter time series joint compression information representation vector and the node state parameter time series joint compression information association representation feature vector to obtain a node state parameter time series joint compression information association representation feature vector. The node state parameter timing joint compressed information multi-scale representation vector is input into the compressed information feature extraction module based on the multi-layer perceptron model to obtain the node state parameter timing joint compressed information multi-scale correlation feature vector, wherein the multi-layer perceptron model comprises two fully connected layers and a SiLU activation function; the node state parameter timing joint compressed information multi-scale correlation feature vector is normalized by using the Sigmoid function to obtain the node state parameter timing joint representation feature weight vector; the node state parameter timing joint representation feature weight vector is used as the weight vector, and each eigenvalue of the node state parameter timing joint representation feature weight vector is calculated and multiplied with each feature matrix of the node state parameter timing joint representation feature graph along the channel dimension to obtain the enhanced node state parameter timing joint representation feature graph.

[0010] In the above-mentioned intelligent operation and maintenance management method for the power system, each enhanced node state parameter time series joint representation feature vector in the sequence of the enhanced node state parameter time series joint representation feature vector is used as a query feature vector, and the query feature vector and the sequence of the enhanced node state parameter time series joint representation feature vector are input into a granular unidirectional attention interactive scanning matching module to obtain multiple unidirectional matching result representation vectors, including: calculating the semantic measurement coefficient between the query feature vector and each enhanced node state parameter time series joint representation feature vector in the sequence of the enhanced node state parameter time series joint representation feature vector to obtain a sequence of unidirectional matching semantic measurement coefficients; normalizing the sequence of unidirectional matching semantic measurement coefficients through a softmax function to obtain a sequence of unidirectional matching semantic measurement weights; using the sequence of unidirectional matching semantic measurement weights as a weight to calculate the weighted sum of the sequence of the enhanced node state parameter time series joint representation feature vector to obtain a unidirectional matching interactive reshaping feature vector; calculating the differential feature vector between the query feature vector and the unidirectional matching interactive reshaping feature vector as the unidirectional matching result representation vector.

[0011] In the above-mentioned intelligent operation and maintenance management method of the power system, the semantic metric coefficients between the query feature vector and each enhanced node state parameter time series joint representation feature vector in the sequence of the enhanced node state parameter time series joint representation feature vector are calculated to obtain a sequence of one-way matching semantic metric coefficients, including: calculating the squares of each eigenvalue of each enhanced node state parameter time series joint representation feature vector in the sequence of the enhanced node state parameter time series joint representation feature vector to obtain a sequence of nonlinear transformation enhanced node state parameter time series joint representation feature vectors; cascading the query feature vector and each nonlinear transformation enhanced node state parameter time series joint representation feature vector in the sequence of the nonlinear transformation enhanced node state parameter time series joint representation feature vector to obtain a sequence of semantic association feature vectors; using a preset transformation vector to multiply each semantic association feature vector in the sequence of the semantic association feature vector to obtain a sequence of semantic similarities; adding a bias term to each semantic similarity in the sequence of semantic similarities and passing it through a sigmoid function to obtain a sequence of one-way matching semantic metric coefficients.

[0012] In the above-mentioned intelligent operation and maintenance management method of the power system, based on the multiple one-way matching result representation vectors, it is determined whether the node status of each node is abnormal, including: inputting the multiple one-way matching result representation vectors into the classifier-based operation and maintenance management generator respectively to obtain multiple operation and maintenance management results, and the operation and maintenance management results indicate whether the node status is abnormal.

[0013] The above-mentioned intelligent operation and maintenance management method for the power system also includes a training step: training the node state parameter time series feature extractor based on the void convolutional neural network model, the feature attention selection and enhancement module based on the compression-suppression structure, the granular unidirectional attention interactive scanning matching module and the classifier-based operation and maintenance management generator.

[0014] In the above-mentioned intelligent operation and maintenance management method for the power system, the training step includes: obtaining training data, the training data including the time series of training node state parameters collected by the sensor network deployed at each node of the power system, wherein the training node state parameters include training current values, training voltage values ​​and training temperature values, and whether the node state of each node has an abnormal real value; arranging the time series of the training node state parameters of each node into a training node state parameter time series joint matrix according to the time dimension and the parameter sample dimension respectively, and then passing it through the node state parameter time series feature extractor based on the void convolutional neural network model to obtain a sequence of training node state parameter time series joint representation feature graphs; inputting each training node state parameter time series joint representation feature graph in the sequence of the training node state parameter time series joint representation feature graph into the feature attention selection and enhancement module based on the compression-suppression structure respectively to obtain a sequence of training enhanced node state parameter time series joint representation feature graphs; calculating each training enhanced node state parameter time series joint table in the sequence of the training enhanced node state parameter time series joint representation feature graphs The global mean vector of the feature graph is shown to obtain a sequence of the training enhanced node state parameter time series joint representation feature vectors; each training enhanced node state parameter time series joint representation feature vector in the sequence of the training enhanced node state parameter time series joint representation feature vector is used as a training query feature vector, and the training query feature vector and the sequence of the training enhanced node state parameter time series joint representation feature vector are input into the granular unidirectional attention interactive scanning matching module to obtain a plurality of training one-way matching result representation vectors; the plurality of training one-way matching result representation vectors are respectively input into the classifier-based operation and maintenance management generator to obtain a classification loss function value; the one-way matching result loss function values ​​of the plurality of training one-way matching result representation vectors are calculated; and the weighted sum of the classification loss function value and the one-way matching result loss function value is used as the loss function value to train the node state parameter time series feature extractor based on the void convolutional neural network model, the feature attention selection and enhancement module based on the compression-suppression structure, the granular unidirectional attention interactive scanning matching module and the classifier-based operation and maintenance management generator.

[0015] According to another aspect of the present application, an intelligent operation and maintenance management system for an electric power system is provided, which includes: an electric power system node state monitoring module, which is used to obtain a time series of node state parameters collected by a sensor network deployed at each node of the electric power system, wherein the node state parameters include current values, voltage values ​​and temperature values; a node state parameter time series joint encoding module, which is used to respectively extract parameter time series joint features of the time series of node state parameters of each node to obtain a sequence of node state parameter time series joint representation feature graphs; a feature enhancement module, which is used to enhance the time series joint representation feature graphs of each node state parameter in the sequence of the node state parameter time series joint representation feature graphs. The combined representation feature graphs are respectively feature enhanced to obtain a sequence of enhanced node state parameter time series joint representation feature vectors; a granular interactive matching module is used to use each enhanced node state parameter time series joint representation feature vector in the sequence of enhanced node state parameter time series joint representation feature vectors as a query feature vector, and input the query feature vector and the sequence of enhanced node state parameter time series joint representation feature vectors into a granular unidirectional attention interactive scanning matching module to obtain multiple unidirectional matching result representation vectors; a node state anomaly detection module is used to determine whether the node state of each node is abnormal based on the multiple unidirectional matching result representation vectors.

[0016] Compared with the prior art, the intelligent operation and maintenance management method and system of the power system provided by the present application adopts artificial intelligence technology based on deep learning to perform real-time monitoring and data analysis on the operating status parameters of each node of the power system, respectively captures the joint feature representation of the node status parameters of each node in time series, and performs granular matching analysis between nodes to dig out the characteristic differences of the operating status of each node relative to the whole, so as to intelligently identify abnormal nodes in the power system. In this way, the accuracy and efficiency of abnormality detection can be improved, the risk of power supply interruption caused by equipment failure can be reduced, the stable operation of the power system can be ensured, and the operation and maintenance process can be optimized. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0018] Figure 1 It is a flow chart of the intelligent operation and maintenance management method of the power system according to the embodiment of the present application.

[0019] Figure 2Schematic diagram of the architecture of the intelligent operation and maintenance management method of the power system according to the embodiment of the present application.

[0020] Figure 3 The present invention is a flowchart of performing feature enhancement on each node state parameter time series joint representation feature graph in the sequence of node state parameter time series joint representation feature graphs in the intelligent operation and maintenance management method of the power system according to an embodiment of the present application to obtain a sequence of enhanced node state parameter time series joint representation feature vectors.

[0021] Figure 4 A flowchart of the method for intelligent operation and maintenance management of power systems according to an embodiment of the present application is provided for inputting each node state parameter time series joint representation feature graph in the sequence of the node state parameter time series joint representation feature graph into a feature attention selection enhancement module based on a compression-suppression structure to obtain a sequence of enhanced node state parameter time series joint representation feature graphs.

[0022] Figure 5 In the method for intelligent operation and maintenance management of the power system according to an embodiment of the present application, each enhanced node state parameter time series joint representation feature vector in the sequence of the enhanced node state parameter time series joint representation feature vector is used as a query feature vector, and the query feature vector and the sequence of the enhanced node state parameter time series joint representation feature vector are input into a granular unidirectional attention interactive scanning matching module to obtain a flowchart of multiple unidirectional matching result representation vectors.

[0023] Figure 6 It is a flowchart of the training steps in the intelligent operation and maintenance management method of the power system according to the embodiment of the present application.

[0024] Figure 7 It is a block diagram of an intelligent operation and maintenance management system of an electric power system according to an embodiment of the present application. DETAILED DESCRIPTION

[0025] Below, the embodiments of the present application will be described in more detail in conjunction with the accompanying drawings, and the above and other purposes, features and advantages of the present application will become more apparent. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the example embodiments described herein.

[0026] As mentioned in the above background technology, the traditional power system operation and maintenance mainly relies on manual monitoring and regular inspections, and relies on a large amount of human resources investment. It is not only inefficient, but also has problems such as untimely fault detection and high operation and maintenance costs. Especially when dealing with large-scale and complex power networks, it is often impossible to quickly identify and properly handle potential abnormal conditions, which may lead to serious problems such as equipment failure and power outages, posing a substantial threat to the stable operation of the power system. In response to the above technical problems, the technical concept of this application is to use artificial intelligence technology based on deep learning to perform real-time monitoring and data analysis on the operating status parameters of each node of the power system, respectively capture the time series joint feature representation of the node status parameters of each node, and perform granular matching analysis between nodes to dig out the characteristic differences of the operating status of each node relative to the whole, so as to intelligently identify abnormal nodes in the power system. In this way, the accuracy and efficiency of anomaly detection can be improved, the risk of power outages caused by equipment failures can be reduced, the stable operation of the power system can be ensured, and the operation and maintenance process can be optimized.

[0027] Figure 1 It is a flow chart of the intelligent operation and maintenance management method of the power system according to the embodiment of the present application. Figure 2 FIG. 1 is a schematic diagram of the architecture of the intelligent operation and maintenance management method of the power system according to the embodiment of the present application. Figure 1 and Figure 2 As shown, according to the embodiment of the present application, the intelligent operation and maintenance management method of the power system includes the following steps: S110, obtaining the time series of node state parameters collected by the sensor network deployed at each node of the power system, wherein the node state parameters include current value, voltage value and temperature value; S120, respectively extracting the parameter time series joint features of the time series of the node state parameters of the each node to obtain a sequence of node state parameter time series joint representation feature graphs; S130, respectively performing feature enhancement on each node state parameter time series joint representation feature graph in the sequence of the node state parameter time series joint representation feature graph to obtain a sequence of enhanced node state parameter time series joint representation feature vectors; S140, using each enhanced node state parameter time series joint representation feature vector in the sequence of the enhanced node state parameter time series joint representation feature vector as a query feature vector, and inputting the query feature vector and the sequence of the enhanced node state parameter time series joint representation feature vector into a granular unidirectional attention interactive scanning matching module to obtain multiple unidirectional matching result representation vectors; S150, based on the multiple unidirectional matching result representation vectors, determining whether the node state of each node is abnormal.

[0028] In the above-mentioned intelligent operation and maintenance management method of the power system, the step S110 obtains the time series of node state parameters collected by the sensor network deployed at each node of the power system, wherein the node state parameters include current value, voltage value and temperature value. It should be understood that the current value, voltage value and temperature value are key indicators reflecting the state of the power system nodes, and the three respectively represent the transmission of electric energy, the quality of electric energy and the thermal state of the equipment. In the technical solution of the present application, by real-time monitoring and analysis of the three, the operating status of the power grid nodes can be deeply analyzed from multiple dimensions, and continuous and uninterrupted monitoring of the operating status of the power system can be achieved, so as to identify any potential problems or abnormal phenomena as early as possible, thereby improving the efficiency of fault prediction and processing.

[0029] In the above-mentioned intelligent operation and maintenance management method of the power system, the step S120 extracts the parameter time series joint features of the time series of the node state parameters of each node to obtain a sequence of node state parameter time series joint representation feature graphs. In a specific example of the present application, the processing method of extracting the parameter time series joint features of the time series of the node state parameters of each node is to arrange the time series of the node state parameters of each node into a node state parameter time series joint matrix according to the time dimension and the parameter sample dimension, and then pass it through a node state parameter time series feature extractor based on a void convolutional neural network model to obtain a sequence of node state parameter time series joint representation feature graphs. It should be understood that, considering that during the operation of the power grid system, the current, voltage and temperature parameters of its nodes usually show a certain spatiotemporal correlation, that is, each parameter shows a certain dynamic change law in the time dimension, and there is a certain mutual influence relationship between different parameters. Based on this, in the technical solution of the present application, the time series of the node state parameters of each node are further arranged into a node state parameter time series joint matrix according to the time dimension and the parameter sample dimension, so as to realize the time series integration of multi-dimensional parameters, which is helpful to maintain the time and space correlation between the parameters and more comprehensively reflect the operation status of the node. Furthermore, the node state parameter time series joint matrix of each node is convolutionally encoded through the hole convolution neural network model, so as to utilize the unique hole structure of the hole convolution neural network to capture the larger-scale time and space correlation features in the node state parameter time series joint matrix, and mine the deep-level time series joint information of the node state parameters, so as to obtain a sequence of node state parameter time series joint representation feature maps, so as to reflect the time series change pattern of the operation status of each node.

[0030] In the above-mentioned intelligent operation and maintenance management method for the power system, the step S130 is to perform feature enhancement on each node state parameter time series joint representation feature graph in the sequence of node state parameter time series joint representation feature graphs to obtain a sequence of enhanced node state parameter time series joint representation feature vectors. Figure 3 The present invention is a flowchart of performing feature enhancement on each node state parameter time series joint representation feature graph in the sequence of node state parameter time series joint representation feature graphs in the intelligent operation and maintenance management method of the power system according to the embodiment of the present application to obtain a sequence of enhanced node state parameter time series joint representation feature vectors. Figure 3 As shown, the step S130 includes: S131, inputting each node state parameter time series joint representation feature graph in the sequence of the node state parameter time series joint representation feature graph into the feature attention selection enhancement module based on the compression-suppression structure to obtain a sequence of enhanced node state parameter time series joint representation feature graphs; S132, calculating the global mean vector of each enhanced node state parameter time series joint representation feature graph in the sequence of the enhanced node state parameter time series joint representation feature graphs to obtain a sequence of enhanced node state parameter time series joint representation feature vectors.

[0031] Specifically, in step S131, each node state parameter time series joint representation feature graph in the sequence of the node state parameter time series joint representation feature graph is input into a feature attention selection enhancement module based on a compression-suppression structure to obtain a sequence of enhanced node state parameter time series joint representation feature graphs. It should be understood that, considering that in the enhanced node state parameter time series joint representation feature graph, not all feature information contributes equally to the abnormal detection of node state. Therefore, in order to further highlight the abnormal fluctuation characteristics of the node state in the feature graph and suppress irrelevant information such as redundant noise therein to improve the accuracy of abnormal detection, in the technical solution of the present application, a feature attention selection enhancement module based on a compression-suppression structure is introduced to process each node state parameter time series joint representation feature graph separately. Among them, the feature attention selection and enhancement module compresses the spatial dimension of the node state parameter time series joint representation feature map to achieve global information integration, and then captures the multi-scale correlation between the global compressed information through a series of convolution, activation and other operations on the global compressed information, and thereby generates the attention weights of each feature channel corresponding to the node state parameter time series joint representation feature map, and further adaptively recalibrates the feature response of each channel through the suppression operation, focusing attention on key information, thereby achieving effective screening and enhancement of feature information, providing a more reliable basis for subsequent anomaly detection.

[0032] Figure 4 The present invention is a flowchart of inputting each node state parameter time series joint representation feature graph in the sequence of the node state parameter time series joint representation feature graph into the feature attention selection enhancement module based on the compression-suppression structure to obtain a sequence of enhanced node state parameter time series joint representation feature graphs in the power system intelligent operation and maintenance management method according to the embodiment of the present application. Figure 4 As shown, the step S131 includes: S1311, calculating the global mean of each feature matrix along the channel dimension of the node state parameter time series joint representation feature graph to obtain a node state parameter time series joint compressed information representation vector; S1312, performing one-dimensional convolution encoding on the node state parameter time series joint compressed information representation vector to obtain a node state parameter time series joint compressed information association representation feature vector; S1313, cascading the node state parameter time series joint compressed information representation vector and the node state parameter time series joint compressed information association representation feature vector to obtain a node state parameter time series joint compressed information multi-scale representation vector; S1314, inputting the node state parameter time series joint compressed information multi-scale representation vector into a multi-layer perceptron-based representation vector. The compressed information feature extraction module of the model is used to obtain a node state parameter time series joint compressed information multi-scale associated feature vector, wherein the multi-layer perceptron model includes two fully connected layers and a SiLU activation function; S1315, using a Sigmoid function to normalize the node state parameter time series joint compressed information multi-scale associated feature vector to obtain a node state parameter time series joint representation feature weight vector; S1316, using the node state parameter time series joint representation feature weight vector as a weight vector, calculating the corresponding multiplication of each eigenvalue of the node state parameter time series joint representation feature weight vector and each feature matrix of the node state parameter time series joint representation feature graph along the channel dimension to obtain the enhanced node state parameter time series joint representation feature graph.

[0033] That is, the step S131 includes: performing attention selection on the node state parameter time series joint representation feature graph using the following feature attention selection enhancement formula to obtain the enhanced node state parameter time series joint representation feature graph, wherein the feature attention selection enhancement formula is: ;in, represents the node state parameter time series joint representation feature graph, The node state parameter time series joint representation feature graph is represented by The coordinates in the channel are The eigenvalues ​​of a point, represents the height of the characteristic graph jointly represented by the time series of the node state parameters, represents the width of the characteristic graph represented by the time series of the node state parameters, is the node state parameter time series joint compression information representation vector eigenvalues, represents the node state parameter time series joint compression information representation vector, represents a one-dimensional convolution operation, Indicates cascade operation, Represents the multi-scale representation vector of the node state parameter temporal joint compression information, represents a multi-layer perceptron, represents the sigmoid function, represents dot product, A time series joint representation feature graph representing the enhanced node state parameters.

[0034] Specifically, the step S132 calculates the global mean vector of each enhanced node state parameter time series joint representation feature graph in the sequence of the enhanced node state parameter time series joint representation feature graph to obtain the sequence of the enhanced node state parameter time series joint representation feature vector. That is, in order to grasp the operating status of each node as a whole and facilitate the subsequent comparative analysis and abnormality detection of the operating status between nodes, in the technical solution of the present application, the global mean vector of each enhanced node state parameter time series joint representation feature graph is further calculated, thereby effectively reducing the feature dimension, simplifying the complexity of subsequent calculations, and retaining the main information of the original feature graph, so as to achieve more efficient abnormal node detection.

[0035] In the above-mentioned intelligent operation and maintenance management method of the power system, the step S140 uses each enhanced node state parameter time series joint representation feature vector in the sequence of the enhanced node state parameter time series joint representation feature vector as a query feature vector, and inputs the query feature vector and the sequence of the enhanced node state parameter time series joint representation feature vector into a granular unidirectional attention interactive scanning matching module to obtain multiple unidirectional matching result representation vectors. It should be understood that, considering that in the power system, each node works together, and when the operating state of a node is significantly deviated or abnormal from that of other nodes, it often means that the node may have a potential failure risk. Therefore, in the technical solution of the present application, a granular unidirectional attention interactive scanning matching module is further introduced to calculate the matching degree of each enhanced node state parameter time series joint representation feature vector relative to the overall node set, so as to evaluate the consistency of the operating state of each node in the overall power system. Specifically, the interactive scanning matching module captures the potential association between nodes by calculating the semantic similarity between the query feature vector and each enhanced node state parameter time series joint representation feature vector in the sequence, and uses this to weighted fuse the sequence of the enhanced node state parameter time series joint representation feature vector, and then further reveals the difference between the query node and the overall node state at the semantic level by calculating the feature difference between the query feature vector and the overall node state feature after weighted fusion, thereby obtaining the one-way matching result representation vector of the query node. In this way, nodes in the power system that are significantly different from the normal operating mode can be accurately identified, providing strong support for subsequent anomaly detection and fault location.

[0036] Figure 5 A flowchart of using each enhanced node state parameter time series joint representation feature vector in the sequence of the enhanced node state parameter time series joint representation feature vector as a query feature vector in the power system intelligent operation and maintenance management method according to the embodiment of the present application, inputting the query feature vector and the sequence of the enhanced node state parameter time series joint representation feature vector into the granular unidirectional attention interactive scanning matching module to obtain multiple unidirectional matching result representation vectors. Figure 5 As shown, the step S140 includes: S141, calculating the semantic measurement coefficients between the query feature vector and each enhanced node state parameter time series joint representation feature vector in the sequence of the enhanced node state parameter time series joint representation feature vector to obtain a sequence of one-way matching semantic measurement coefficients; S142, normalizing the sequence of one-way matching semantic measurement coefficients through a softmax function to obtain a sequence of one-way matching semantic measurement weights; S143, using the sequence of one-way matching semantic measurement weights as weights to calculate the weighted sum of the sequence of the enhanced node state parameter time series joint representation feature vector to obtain a one-way matching interactive reshaping feature vector; S144, calculating the differential feature vector between the query feature vector and the one-way matching interactive reshaping feature vector as the one-way matching result representation vector.

[0037] More specifically, the step S141 includes: calculating the square of each eigenvalue of each enhanced node state parameter time series joint representation feature vector in the sequence of the enhanced node state parameter time series joint representation feature vector to obtain a sequence of nonlinear transformation enhanced node state parameter time series joint representation feature vectors; cascading the query feature vector with each nonlinear transformation enhanced node state parameter time series joint representation feature vector in the sequence of the nonlinear transformation enhanced node state parameter time series joint representation feature vector to obtain a sequence of semantic association feature vectors; using a preset transformation vector to multiply each semantic association feature vector in the sequence of semantic association feature vectors to obtain a sequence of semantic similarities; adding a bias term to each semantic similarity in the sequence of semantic similarities and then passing it through a sigmoid function to obtain a sequence of one-way matching semantic measurement coefficients.

[0038] That is, the step S140 includes: processing the query feature vector and the sequence of the enhanced node state parameter time series joint representation feature vector using the following one-way interactive matching formula to obtain the one-way matching result representation vector, wherein the one-way interactive matching formula is: ;in, represents the query feature vector, The first in the sequence of the joint representation feature vector of the time series of the enhanced node state parameters is represented The time series of enhanced node state parameters jointly represent the feature vector, Indicates cascade operation, is the preset transformation vector, is the bias term, express function, Represents the query feature vector and the The time series of enhanced node state parameters jointly represent the one-way matching semantic metric coefficient between feature vectors. The value of is the number of characteristic vectors jointly represented by the time series of the enhanced node state parameters, represents the exponential function operation with e as the base, Represents the query feature vector and the The time series of enhanced node state parameters jointly represent the one-way matching semantic metric weight between feature vectors. represents the one-way matching interaction reshaping feature vector, It means subtracting by position. Indicates that the one-way matching result represents a vector.

[0039] In the above-mentioned intelligent operation and maintenance management method for the power system, the step S150 determines whether the node state of each node is abnormal based on the multiple one-way matching result representation vectors. In a specific example of the present application, the implementation method of determining whether the node state of each node is abnormal based on the multiple one-way matching result representation vectors is to input the multiple one-way matching result representation vectors into the operation and maintenance management generator based on the classifier to obtain multiple operation and maintenance management results, and the operation and maintenance management results indicate whether the node state is abnormal. That is, in the technical solution of the present application, the operation and maintenance management generator is constructed based on the machine learning classification algorithm, and the deep learning and reasoning of each one-way matching result representation vector is performed by the pre-trained classifier, and according to the classification decision rules established by the classifier model during the training process, each one-way matching result representation vector is mapped to the corresponding classification label, that is, "node state is normal" and "node state is abnormal", so as to generate the operation and maintenance management results corresponding to each node, so as to help the operation and maintenance personnel quickly locate the abnormal node, evaluate the fault risk, and take corresponding maintenance and management measures.

[0040] In a preferred example, each enhanced node state parameter time series joint representation feature graph expresses the sample-time series cross-dimensional correlation characteristics of the node state parameters of each node respectively. Therefore, after calculating the global mean vector of each enhanced node state parameter time series joint representation feature graph to obtain a sequence of enhanced node state parameter time series joint representation feature vectors, the enhanced node state parameter time series joint representation feature vector will have distribution difference significance relative to the distribution difference of the feature graph corresponding to each node through further mean enhancement. In this way, after further inputting the query feature vector and the sequence of the enhanced node state parameter time series joint representation feature vector into the granular unidirectional attention interactive scanning matching module, the multiple unidirectional matching result representation vectors obtained will also lead to differences in unidirectional attention weights of interactive scanning matching based on the distribution difference significance, so that the multiple unidirectional matching result representation vectors as a whole will have classification regression consistency explanation recognition difficulties attributed to complex semantic weighted interactive aggregation, thereby affecting the consistency of the multiple operation and maintenance management results.

[0041] Based on this, the present application considers jointly optimizing the multiple one-way matching result representation vectors to obtain an optimized one-way matching result cascade representation vector, including the steps of: cascading the multiple one-way matching result representation vectors into a one-way matching result cascade representation vector; calculating the autocorrelation matrix of the one-way matching result cascade representation vector, that is: ;in, represents the cascade representation vector of the one-way matching result, represents the transpose of a vector, represents matrix multiplication, Denotes the autocorrelation matrix.

[0042] The one-way matching result cascade representation vector is multiplied by the self-association matrix in a query manner to obtain an intermediate feature vector, that is: ;in, represents the cascade representation vector of the one-way matching result, represents the autocorrelation matrix, represents matrix multiplication, Denotes the intermediate eigenvector.

[0043] Calculate the first Row vector and Inner product sum of row vectors distance, to obtain the first weight matrix and the second weight matrix , expressed as: ;in, and The autocorrelation matrix is ​​represented by Row vector and row vector, represents the transpose of a vector, represents matrix multiplication, represents the autocorrelation matrix, represents the first weight matrix, represents the first weight matrix The eigenvalues ​​of the positions, represents the second weight matrix, represents the second weight matrix The eigenvalues ​​of the positions, Represents the calculation vector distance.

[0044] The first weight matrix and the second weight matrix After matrix multiplication, further matrix multiplication is performed with the intermediate feature vector to obtain an optimized one-way matching result cascade representation vector, namely: ;in, represents the first weight matrix, represents the second weight matrix, represents the intermediate eigenvector, represents matrix multiplication, The optimized one-way matching result cascade representation vector is represented.

[0045] Then the optimized one-way matching result cascade representation vector is split into multiple optimized one-way matching result representation vectors. Finally, the multiple optimized one-way matching result representation vectors are respectively input into the classifier-based operation and maintenance management generator to obtain multiple operation and maintenance management results.

[0046] That is, for the bit-by-bit scattering association response space of the eigenvalues ​​of the multiple one-way matching result representation vectors, a deep embedding structure in the high-order association space is obtained through the joint embedding space of the vector inner product embedding space and the vector independent embedding space, and the decomposable association response sub-dimensions of the bit-by-bit scattering association response space of the eigenvalues ​​of the multiple one-way matching result representation vectors are represented as a set of spatial sub-dimensions to further perform dependency space mapping of the decomposable association response sub-dimensions of the deep embedding structure of the multiple one-way matching result representation vectors, so as to improve the interpretable consistency of the solution space of the regression solutions of the multiple one-way matching result representation vectors based on the association distribution through the complex dimensional representation of the association response, thereby improving the consistency of the multiple operation and maintenance management results while maintaining their accuracy.

[0047] It should be understood that before using the above neural network model, it needs to be trained. That is to say, in the intelligent operation and maintenance management method of the power system of the present application, it also includes a training step: training the node state parameter time series feature extractor based on the hole convolutional neural network model, the feature attention selection and enhancement module based on the compression-suppression structure, the granular unidirectional attention interactive scanning matching module and the classifier-based operation and maintenance management generator.

[0048] Figure 6 FIG. 1 is a flow chart of the training steps in the intelligent operation and maintenance management method of the power system according to an embodiment of the present application. Figure 6As shown, the training step S200 includes: S210, obtaining training data, wherein the training data includes a time series of training node state parameters collected by a sensor network deployed at each node of the power system, wherein the training node state parameters include training current values, training voltage values ​​and training temperature values, and whether the node state of each node has an abnormal true value; S220, arranging the time series of the training node state parameters of each node into a training node state parameter time series joint matrix according to the time dimension and the parameter sample dimension, respectively, and then passing through the node state parameter time series feature extractor based on the void convolutional neural network model to obtain a sequence of training node state parameter time series joint representation feature graphs; S230, inputting each training node state parameter time series joint representation feature graph in the sequence of the training node state parameter time series joint representation feature graph into the feature attention selection and enhancement module based on the compression-suppression structure to obtain a sequence of training enhanced node state parameter time series joint representation feature graphs; S240, calculating the full representation of each training enhanced node state parameter time series joint representation feature graph in the sequence of the training enhanced node state parameter time series joint representation feature graphs. The local mean vector is used to obtain a sequence of the training reinforcement node state parameter time series joint representation feature vectors; S250, each training reinforcement node state parameter time series joint representation feature vector in the sequence of the training reinforcement node state parameter time series joint representation feature vector is used as a training query feature vector, and the training query feature vector and the sequence of the training reinforcement node state parameter time series joint representation feature vector are input into the granular unidirectional attention interactive scanning matching module to obtain multiple training unidirectional matching result representation vectors; S260, the multiple training unidirectional matching result representation vectors are respectively Input the classifier-based operation and maintenance management generator to obtain the classification loss function value; S270, calculate the one-way matching result loss function value of the multiple training one-way matching result representation vectors; S280, use the weighted sum of the classification loss function value and the one-way matching result loss function value as the loss function value to train the node state parameter temporal feature extractor based on the void convolutional neural network model, the feature attention selection and enhancement module based on the compression-suppression structure, the granular one-way attention interactive scanning matching module and the classifier-based operation and maintenance management generator.

[0049] In the technical solution of the present application, each training enhanced node state parameter time series joint representation feature graph respectively expresses the sample-time series cross-dimensional correlation characteristics of the training node state parameters of each node. Therefore, after calculating the global mean vector of each training enhanced node state parameter time series joint representation feature graph to obtain a sequence of training enhanced node state parameter time series joint representation feature vectors, the training enhanced node state parameter time series joint representation feature vector will have distribution difference significance relative to the distribution difference of the feature graph corresponding to each node through further mean enhancement. In this way, after further inputting the training query feature vector and the sequence of the training enhanced node state parameter time series joint representation feature vector into the granular unidirectional attention interactive scanning matching module, the multiple training unidirectional matching result representation vectors obtained will also lead to differences in unidirectional attention weights of interactive scanning matching based on the significance of distribution differences, thereby causing the multiple training unidirectional matching result representation vectors as a whole to have classification regression consistency recognition difficulties attributed to complex semantic weighted interactive aggregation, thereby affecting the consistency of the multiple operation and maintenance management results.

[0050] Therefore, in a preferred example of the present application, in the model training process, a loss function other than the classification loss function is further introduced, that is, training the model based on the loss function value through gradient back propagation includes the following steps: cascading the multiple training one-way matching result representation vectors into a training one-way matching result cascade representation vector; calculating a first one-way matching result cascade weight matrix and a second one-way matching result cascade weight matrix based on the training one-way matching result cascade representation vector; wherein the first one-way matching result cascade weight matrix The feature value of the position is the first cascade representation vector of the training one-way matching result. Eigenvalues ​​and The mean of the eigenvalues, and the second one-way matching result cascade weight matrix The feature value of the position is the first cascade representation vector of the training one-way matching result. Eigenvalues ​​and Half of the absolute value of the difference in eigenvalues; perform query-type matrix multiplication on the training one-way matching result cascade representation vector with the first one-way matching result cascade weight matrix and the second one-way matching result cascade weight matrix to obtain the first one-way matching result cascade intermediate vector and the second one-way matching result cascade intermediate vector; calculate the vector inner product of the first one-way matching result cascade intermediate vector and the second one-way matching result cascade intermediate vector to obtain the first one-way matching result cascade loss term; perform matrix multiplication on the first one-way matching result cascade weight matrix and the second one-way matching result cascade weight matrix, and calculate the result matrix norm to obtain a second one-way matching result cascade loss term; subtract the product of a predetermined weight hyperparameter and the second one-way matching result cascade loss term from the first one-way matching result cascade loss term to obtain a one-way matching result loss function value; based on the weighted sum of the one-way matching result loss function value and the classification loss function value, optimize the model parameters through gradient back propagation.

[0051] That is, the one-way matching result loss function values ​​of the multiple training one-way matching result representation vectors are calculated using the following one-way matching result loss function, wherein the one-way matching result loss function is specifically expressed as: ;in, represents a training one-way matching result cascade representation vector obtained by cascading the multiple training one-way matching result representation vectors, represents the first cascade representation vector of the training one-way matching result eigenvalues, represents the first cascade representation vector of the training one-way matching result eigenvalues, is the first one-way matching result cascade weight matrix, the first one-way matching result cascade weight matrix The eigenvalues ​​of the positions, represents the first cascade representation vector of the training one-way matching result The eigenvalues ​​and The mean of the eigenvalues, is the second one-way matching result cascade weight matrix, the first The eigenvalues ​​of the positions, represents the first cascade representation vector of the training one-way matching result The eigenvalues ​​and One-half the absolute value of the difference of the eigenvalues, represents the matrix multiplication operation, is the predetermined weight hyperparameter, Represents the matrix norm, Represents the one-way matching result loss function value.

[0052] That is, in the above preferred example, the one-way matching result loss function performs the query composition of the detail inner product space within the multiple training one-way matching result representation vectors through the structured feature representation of the short-range and long-range cross-scale detail links of the training one-way matching result cascade representation vectors obtained by cascading the multiple training one-way matching result representation vectors, so as to approximate the low-rank independent observable composition of the link detail composition provided by the structured detail interaction of the multiple training one-way matching result representation vectors. In this way, by training with the one-way matching result loss function, the distributed detail groups of the multiple training one-way matching result representation vectors can be used to perform detail group decomposition on the basis of detail complexity, so as to promote the classification regression consistency decomposition and recognition of the complex feature structures of the multiple training one-way matching result representation vectors, and improve the consistency of the multiple operation and maintenance management results.

[0053] In summary, the intelligent operation and maintenance management method of the power system according to the embodiment of the present application is explained, which uses artificial intelligence technology based on deep learning to perform real-time monitoring and data analysis on the operating status parameters of each node of the power system, and captures the time series joint feature representation of the node status parameters of each node respectively, and performs granular matching analysis between nodes to dig out the characteristic differences of the operating status of each node relative to the whole, so as to intelligently identify abnormal nodes in the power system. In this way, the accuracy and efficiency of abnormality detection can be improved, the risk of power supply interruption caused by equipment failure can be reduced, the stable operation of the power system can be ensured, and the operation and maintenance process can be optimized.

[0054] Figure 7 FIG. 1 is a block diagram of an intelligent operation and maintenance management system for a power system according to an embodiment of the present application. Figure 7As shown, according to the embodiment of the present application, the intelligent operation and maintenance management system 100 of the power system includes: a power system node state monitoring module 110, which is used to obtain the time series of node state parameters collected by the sensor network deployed at each node of the power system, wherein the node state parameters include current value, voltage value and temperature value; a node state parameter time series joint encoding module 120, which is used to respectively extract the parameter time series joint features of the time series of the node state parameters of each node to obtain a sequence of node state parameter time series joint representation feature graphs; a feature enhancement module 130, which is used to perform a time series joint encoding of each node state parameter in the sequence of the node state parameter time series joint representation feature graphs. The combined representation feature graphs are respectively feature enhanced to obtain a sequence of enhanced node state parameter time series joint representation feature vectors; a granular interactive matching module 140 is used to use each enhanced node state parameter time series joint representation feature vector in the sequence of enhanced node state parameter time series joint representation feature vectors as a query feature vector, and input the query feature vector and the sequence of enhanced node state parameter time series joint representation feature vectors into a granular unidirectional attention interactive scanning matching module to obtain multiple unidirectional matching result representation vectors; a node state anomaly detection module 150 is used to determine whether the node state of each node is abnormal based on the multiple unidirectional matching result representation vectors.

[0055] Here, those skilled in the art can understand that the specific operations of each module in the above-mentioned power system intelligent operation and maintenance management system have been referred to above. Figures 1 to 6 The invention has been introduced in detail in the description of the intelligent operation and maintenance management method of the power system, and therefore, its repeated description will be omitted.

[0056] The basic principle of the present invention is described above in conjunction with specific embodiments. However, it should be pointed out that the advantages, strengths, effects, etc. mentioned in the present invention are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. must be possessed by each embodiment of the present invention. In addition, the specific details of the above embodiments are only for the purpose of illustration and facilitation of understanding, rather than limitation, and the above details do not limit the present invention to being implemented by adopting the above specific details.

[0057] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, please refer to the relevant description of other embodiments. In the several embodiments provided by the present invention, it should be understood that the disclosed system and method can be implemented in other ways. For example, the system embodiment described above is only schematic. For example, the module division is only a logical function division, and there may be other division methods in actual implementation. The modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0058] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference to a figure in a claim should not be considered as limiting the claim to which it relates.

[0059] In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. The multiple units stated in the system claims can also be implemented by one unit through software or hardware.

[0060] Finally, it should be noted that the above description has been given for the purpose of illustration and description. In addition, the above embodiments are only used to illustrate the technical solution of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for intelligent operation and maintenance management of a power system, characterized in that: include: The method comprises the following steps: obtaining a time series of node state parameters collected by a sensor network deployed at each node of the power system, wherein the node state parameters include current value, voltage value and temperature value; extracting parameter time series joint features of the time series of the node state parameters of each node respectively to obtain a sequence of node state parameter time series joint representation feature graphs; performing feature enhancement on each node state parameter time series joint representation feature graph in the sequence of the node state parameter time series joint representation feature graphs respectively to obtain a sequence of enhanced node state parameter time series joint representation feature vectors; taking each enhanced node state parameter time series joint representation feature vector in the sequence of the enhanced node state parameter time series joint representation feature vectors as a query feature vector, and converting the enhanced node state parameter time series joint representation feature vector into a query feature vector. The query feature vector and the sequence of the enhanced node state parameter time series joint representation feature vector are input into the granular unidirectional attention interactive scanning matching module to obtain multiple unidirectional matching result representation vectors; based on the multiple unidirectional matching result representation vectors, it is determined whether the node state of each node is abnormal; wherein, each node state parameter time series joint representation feature graph in the sequence of the node state parameter time series joint representation feature graph is feature enhanced respectively to obtain a sequence of enhanced node state parameter time series joint representation feature vectors, including: each node state parameter time series joint representation feature graph in the sequence of the node state parameter time series joint representation feature graph is input into the feature attention selection enhancement module based on the compression-suppression structure To obtain a sequence of enhanced node state parameter time series joint representation feature graphs; calculate the global mean vector of each enhanced node state parameter time series joint representation feature graph in the sequence of enhanced node state parameter time series joint representation feature graphs to obtain a sequence of enhanced node state parameter time series joint representation feature vectors; wherein, each node state parameter time series joint representation feature graph in the sequence of node state parameter time series joint representation feature graphs is respectively input into a feature attention selection enhancement module based on a compression-suppression structure to obtain a sequence of enhanced node state parameter time series joint representation feature graphs, including: calculating the global mean of each feature matrix along the channel dimension of the node state parameter time series joint representation feature graph to obtain the node state parameter time series joint representation feature graph. A joint compressed information representation vector; performing one-dimensional convolution encoding on the node state parameter time series joint compressed information representation vector to obtain a node state parameter time series joint compressed information association representation feature vector; cascading the node state parameter time series joint compressed information representation vector and the node state parameter time series joint compressed information association representation feature vector to obtain a node state parameter time series joint compressed information multi-scale representation vector; inputting the node state parameter time series joint compressed information multi-scale representation vector into a compressed information feature extraction module based on a multi-layer perceptron model to obtain a node state parameter time series joint compressed information multi-scale association feature vector, wherein the multi-layer perceptron model includes two fully connected layers and a SiLU activation function;The multi-scale associated feature vector of the node state parameter time series joint compression information is normalized using a Sigmoid function to obtain a node state parameter time series joint representation feature weight vector; the node state parameter time series joint representation feature weight vector is used as a weight vector, and each eigenvalue of the node state parameter time series joint representation feature weight vector is calculated and multiplied with each feature matrix along the channel dimension of the node state parameter time series joint representation feature graph to obtain the enhanced node state parameter time series joint representation feature graph; wherein each enhanced node state parameter time series joint representation feature vector in the sequence of the enhanced node state parameter time series joint representation feature vector is used as a query feature vector, and the query feature vector and the sequence of the enhanced node state parameter time series joint representation feature vector are input into a granular unit. The interactive scanning matching module of the attention is used to obtain multiple one-way matching result representation vectors, including: calculating the semantic metric coefficients between the query feature vector and each of the enhanced node state parameter time series joint representation feature vectors in the sequence of the enhanced node state parameter time series joint representation feature vectors to obtain a sequence of one-way matching semantic metric coefficients; normalizing the sequence of one-way matching semantic metric coefficients through a softmax function to obtain a sequence of one-way matching semantic metric weights; using the sequence of one-way matching semantic metric weights as a weight to calculate the weighted sum of the sequence of the enhanced node state parameter time series joint representation feature vectors to obtain a one-way matching interactive reshaping feature vector; calculating the differential feature vector between the query feature vector and the one-way matching interactive reshaping feature vector as the one-way matching result representation vector. ; 2. The intelligent operation and maintenance management method of the power system according to claim 1 is characterized in that: The parameter time series joint features of the time series of the node state parameters of each node are extracted respectively to obtain a sequence of node state parameter time series joint representation feature graphs, including: arranging the time series of the node state parameters of each node into a node state parameter time series joint matrix according to the time dimension and the parameter sample dimension respectively, and then passing the matrix through a node state parameter time series feature extractor based on a void convolutional neural network model to obtain a sequence of node state parameter time series joint representation feature graphs.

3. The intelligent operation and maintenance management method of the power system according to claim 2 is characterized in that: The semantic metric coefficients between the query feature vector and each of the enhanced node state parameter time series joint representation feature vectors in the sequence of the enhanced node state parameter time series joint representation feature vectors are calculated to obtain a sequence of one-way matching semantic metric coefficients, including: calculating the squares of each eigenvalue of each of the enhanced node state parameter time series joint representation feature vectors in the sequence of the enhanced node state parameter time series joint representation feature vectors to obtain a sequence of nonlinear transformation enhanced node state parameter time series joint representation feature vectors; cascading the query feature vector and each of the nonlinear transformation enhanced node state parameter time series joint representation feature vectors in the sequence of the nonlinear transformation enhanced node state parameter time series joint representation feature vectors to obtain a sequence of semantic association feature vectors; multiplying each of the semantic association feature vectors in the sequence of the semantic association feature vectors with a preset transformation vector to obtain a sequence of semantic similarities; adding an offset term to each of the semantic similarities in the sequence of semantic similarities and passing the result through a sigmoid function to obtain a sequence of one-way matching semantic metric coefficients.

4. The intelligent operation and maintenance management method of the power system according to claim 3 is characterized in that: Based on the multiple one-way matching result representation vectors, determine whether there is an abnormality in the node status of each node, including: inputting the multiple one-way matching result representation vectors into a classifier-based operation and maintenance management generator respectively to obtain multiple operation and maintenance management results, and the operation and maintenance management results indicate whether there is an abnormality in the node status.

5. The intelligent operation and maintenance management method of the power system according to claim 4 is characterized in that: It also includes a training step: training the node state parameter temporal feature extractor based on the void convolutional neural network model, the feature attention selection and enhancement module based on the compression-suppression structure, the granular unidirectional attention interactive scanning matching module and the classifier-based operation and maintenance management generator.

6. The intelligent operation and maintenance management method of the power system according to claim 5 is characterized in that: The training step includes: obtaining training data, wherein the training data includes a time series of training node state parameters collected by a sensor network deployed at each node of the power system, wherein the training node state parameters include training current values, training voltage values, and training temperature values, as well as a true value of whether the node state of each node is abnormal; arranging the time series of the training node state parameters of each node into a training node state parameter time series joint matrix according to the time dimension and the parameter sample dimension, respectively, and then passing the time series feature extractor of the node state parameter based on the void convolutional neural network model to obtain a sequence of training node state parameter time series joint representation feature graphs; inputting each training node state parameter time series joint representation feature graph in the sequence of the training node state parameter time series joint representation feature graphs into the feature attention selection and enhancement module based on the compression-suppression structure to obtain a sequence of training enhanced node state parameter time series joint representation feature graphs; calculating the global mean of each training enhanced node state parameter time series joint representation feature graph in the sequence of the training enhanced node state parameter time series joint representation feature graphs vector to obtain a sequence of training enhanced node state parameter time series joint representation feature vectors; use each training enhanced node state parameter time series joint representation feature vector in the sequence of training enhanced node state parameter time series joint representation feature vectors as a training query feature vector, input the training query feature vector and the sequence of the training enhanced node state parameter time series joint representation feature vectors into the granular unidirectional attention interactive scanning matching module to obtain multiple training one-way matching result representation vectors; input the multiple training one-way matching result representation vectors into the classifier-based operation and maintenance management generator respectively to obtain a classification loss function value; calculate the one-way matching result loss function values ​​of the multiple training one-way matching result representation vectors; use the weighted sum of the classification loss function value and the one-way matching result loss function value as the loss function value to train the node state parameter time series feature extractor based on the void convolutional neural network model, the feature attention selection and enhancement module based on the compression-suppression structure, the granular unidirectional attention interactive scanning matching module and the classifier-based operation and maintenance management generator.

7. An intelligent operation and maintenance management system for a power system, characterized in that: include: The power system node state monitoring module is used to obtain the time series of node state parameters collected by the sensor network deployed at each node of the power system, wherein the node state parameters include current value, voltage value and temperature value; the node state parameter time series joint encoding module is used to extract the parameter time series joint features of the time series of the node state parameters of each node to obtain a sequence of node state parameter time series joint representation feature graphs; the feature enhancement module is used to perform feature enhancement on each node state parameter time series joint representation feature graph in the sequence of node state parameter time series joint representation feature graphs to obtain a sequence of enhanced node state parameter time series joint representation feature vectors; the granular interactive matching module is used to extract the parameter time series joint representation feature graphs of the node state ... A block is used to use each enhanced node state parameter time series joint representation feature vector in the sequence of the enhanced node state parameter time series joint representation feature vector as a query feature vector, and input the query feature vector and the sequence of the enhanced node state parameter time series joint representation feature vector into a granular unidirectional attention interactive scanning matching module to obtain multiple unidirectional matching result representation vectors; a node state anomaly detection module is used to determine whether there is an abnormality in the node state of each node based on the multiple unidirectional matching result representation vectors; wherein, each node state parameter time series joint representation feature graph in the sequence of the node state parameter time series joint representation feature graph is feature enhanced to obtain an enhanced node state The method comprises the following steps: inputting each node state parameter time series joint representation feature graph in the sequence of the node state parameter time series joint representation feature graph into a feature attention selection and enhancement module based on a compression-suppression structure to obtain a sequence of enhanced node state parameter time series joint representation feature graphs; calculating the global mean vector of each enhanced node state parameter time series joint representation feature graph in the sequence of the enhanced node state parameter time series joint representation feature graphs to obtain a sequence of enhanced node state parameter time series joint representation feature vectors; wherein each node state parameter time series joint representation feature graph in the sequence of the node state parameter time series joint representation feature graphs is input into a feature attention selection and enhancement module based on a compression-suppression structure ... graphs; The feature attention selection and strengthening module of the compression-suppression structure is used to obtain a sequence of strengthened node state parameter temporal joint representation feature graphs, including: calculating the global mean of each feature matrix along the channel dimension of the node state parameter temporal joint representation feature graph to obtain a node state parameter temporal joint compression information representation vector; performing one-dimensional convolution encoding on the node state parameter temporal joint compression information representation vector to obtain a node state parameter temporal joint compression information association representation feature vector; cascading the node state parameter temporal joint compression information representation vector and the node state parameter temporal joint compression information association representation feature vector to obtain a node state parameter temporal joint compression information multi-scale representation vector;The node state parameter time series joint compressed information multi-scale representation vector is input into a compressed information feature extraction module based on a multi-layer perceptron model to obtain a node state parameter time series joint compressed information multi-scale correlation feature vector, wherein the multi-layer perceptron model includes two fully connected layers and a SiLU activation function; the node state parameter time series joint compressed information multi-scale correlation feature vector is normalized using a Sigmoid function to obtain a node state parameter time series joint representation feature weight vector; the node state parameter time series joint representation feature weight vector is used as a weight vector, and each eigenvalue of the node state parameter time series joint representation feature weight vector is calculated and multiplied with the corresponding feature matrices of the node state parameter time series joint representation feature graph along the channel dimension to obtain the enhanced node state parameter time series joint representation feature graph; wherein, each enhanced node state parameter time series joint representation in the sequence of the enhanced node state parameter time series joint representation feature vector is used as a weight vector. The feature vector is used as a query feature vector, and the query feature vector and the sequence of the reinforced node state parameter time series joint representation feature vector are input into the granular unidirectional attention interactive scanning matching module to obtain multiple unidirectional matching result representation vectors, including: calculating the semantic metric coefficients between the query feature vector and each reinforced node state parameter time series joint representation feature vector in the sequence of the reinforced node state parameter time series joint representation feature vector to obtain a sequence of unidirectional matching semantic metric coefficients; normalizing the sequence of unidirectional matching semantic metric coefficients through a softmax function to obtain a sequence of unidirectional matching semantic metric weights; using the sequence of unidirectional matching semantic metric weights as a weight to calculate the weighted sum of the sequence of the reinforced node state parameter time series joint representation feature vector to obtain a unidirectional matching interactive reshaping feature vector; calculating the differential feature vector between the query feature vector and the unidirectional matching interactive reshaping feature vector as the unidirectional matching result representation vector. ;

Images