Substation Secondary Equipment Fault Location Method Based on Spatiotemporal Graph Convolutional Network Model

Abstract

The present invention provides a method for fault location of secondary equipment in a substation based on a spatio-temporal graph convolutional network model. Feature information is extracted from the historical fault data of the secondary equipment in the substation, and the operating state of the secondary equipment, the receiving state of SV / GOOSE signals, and the sampled values are arranged in a time series to generate a fault feature set. Based on the change trend of the fault feature set in the time series, multi-modal feature fusion is performed on the fault feature data. An STGCN model is constructed based on the fault features. The STGCN model is trained using historical fault samples, and features highly relevant to the fault are extracted through data preprocessing and feature screening to construct an optimized feature set, continuously improving the fault location accuracy of the model and comprehensively establishing the correlation mapping relationship between the fault feature information and the fault categories. Combining with the fault inference rules, the feature set is input into the STGCN model, and the spatio-temporal graph convolutional ability of the model is used to accurately locate the faults of the secondary equipment, realizing the function of improving the accuracy and efficiency of fault diagnosis of secondary equipment in the substation.

Description

Technical Field

[0001] The present invention belongs to the technical field of secondary equipment in substations, and particularly relates to a method for fault location of secondary equipment in substations based on a spatio-temporal graph convolutional network model. Background Art

[0002] In modern power systems, secondary equipment in substations undertakes key monitoring, measurement, protection, and regulation tasks, and is a core component to ensure the safe, stable, and economic operation of the power system. Secondary equipment is responsible for collecting and transmitting a large amount of real-time information of the power system, and remotely operating, protecting, and automatically controlling primary equipment in the substation. Its stable operation is crucial for the reliability of the power system. However, with the continuous expansion of the scale of the power system and the improvement of the automation level, the types and quantities of secondary equipment have increased significantly, and the equipment structure and functions have become more complex, which has brought new technical challenges to the fault diagnosis and location of secondary equipment.

[0003] Traditional methods for fault diagnosis and location of secondary equipment usually rely on the experience of operation and maintenance personnel or the support of equipment manufacturers. Especially for complex or sporadic faults, it is often difficult to quickly determine the accurate fault location through experience. For operation and maintenance personnel with less experience, the risk of misdiagnosis is higher, which is likely to lead to carrying wrong maintenance equipment, affecting the efficiency of fault handling, and even may delay the protection of primary equipment, thus affecting the safe operation of the entire power system. Therefore, there is an urgent need for a more scientific and efficient fault diagnosis and location method that can quickly and accurately identify secondary equipment faults, reduce the burden on operation and maintenance personnel, and improve the intelligent operation and maintenance level of substations.

[0004] In recent years, with the continuous development of big data technology and artificial intelligence algorithms, data-driven intelligent fault diagnosis methods have gradually been applied to the fault location of secondary equipment. Especially the proposed Spatial-Temporal Graph Convolutional Network (STGCN) model provides a new solution idea for fault location problems with spatio-temporal characteristics in complex systems. STGCN can effectively process temporal and spatial information, and through analyzing the spatio-temporal correlation characteristics between different devices, achieve accurate fault location. This method can not only greatly improve the location accuracy, but also shorten the fault diagnosis time, and enhance the stability and reliability of the power system. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method for fault location of secondary equipment in substations based on a spatio-temporal graph convolutional network model, which is used to improve the accuracy and efficiency of fault diagnosis of secondary equipment in substations.

[0006] The technical solution adopted by the present invention to solve the above technical problems is as follows: A substation secondary equipment fault location method based on a spatio-temporal graph convolutional network model, comprising the following steps:

[0007] S1: Obtain the historical data of the secondary equipment, and construct a fault feature set including operation status information, SV / GOOSE reception status information, and sampled values; sort the fault feature data according to the time series and correspond one by one with the corresponding secondary equipment faults;

[0008] S2: Extract and screen highly relevant fault features to construct a screened fault feature set;

[0009] S3: Perform multi-modal feature fusion processing on the data of the fault feature set based on the time series to obtain fault feature fusion data;

[0010] S4: Construct an STGCN model including a three-layer 1D-CNN network and a one-layer GCN network to simultaneously learn the spatial and temporal features of graph data; divide the fault feature fusion data into a training set and a test set, where the training set is used for model learning, and the test set is used to evaluate the performance and accuracy of the model;

[0011] S5: Set a dynamic threshold for the number of fault feature information under adaptive weather conditions to detect secondary equipment faults; when a secondary equipment fault is detected, diagnose and locate the fault location according to the fault inference rules; if the inference rules cannot accurately locate the fault location, input the fault feature fusion data into the STGCN model, and identify and locate the fault location of the secondary equipment through the model.

[0012] According to the above solution, in step S1, the operation status information of the secondary equipment includes the self-check status of the merging unit, line protection, and intelligent terminal, which is used to reflect the equipment operation status through multi-dimensional data integration, support accurate fault location and early warning, and effectively improve the safety and reliability of the system.

[0013] According to the above solution, in step S1, the SV / GOOSE message reception status information includes the message reception status of the merging unit, protection device, and intelligent terminal, which is used to judge the abnormality of sampled values and switch quantities; when an event occurs in the equipment, the GOOSE data set is immediately updated and data frames are sent at specific time intervals: first, the second and third frames are sent at an interval of T1, and then the fourth and fifth frames are sent at intervals of T2 and T3, where T2 is twice T1 and T3 is twice T2; after sending 5 frames, the sending interval returns to T0 and switches to the heartbeat message state, and the message survival time is 2 times T0; if the receiving party does not receive the message within 4 times T0, it is determined that the connection is interrupted and a disconnection alarm is triggered.

[0014] According to the above solution, in step S2, the specific steps are as follows:

[0015] S21: Preprocess the historical data to eliminate the influence of dimension through normalization;

[0016] S22: Analyze the linear correlation between each feature and the fault using the Pearson correlation coefficient, remove the features with low correlation, retain the features with high correlation, and construct a new fault feature set.

[0017] According to the above solution, in step S3, the specific steps are as follows:

[0018] S31: Construct a fault feature fusion weight matrix based on the data values, change trends, and change rates of the fault feature set in the historical data;

[0019] S32: Multiply the fault feature set by the fault feature fusion weight matrix to obtain the fault feature fusion data.

[0020] According to the above solution, in step S4, the 1D-CNN network is used to perform convolution operations in the time dimension of the data to extract time series information; the 1D-CNN network includes two convolutional layers and a flattening operation, and each convolutional layer includes a convolution operation and a pooling operation; after the second pooling operation, there is also a fully connected layer for flattening the high-dimensional data into one-dimensional data; the specific steps are as follows:

[0021] Take the output feature matrix of the previous layer as the input feature matrix of the current layer, extract the local patterns in the input features through the weight function, introduce nonlinearity through the activation function, and obtain the output of the current layer, that is, the feature map or activation value of this layer.

[0022] According to the above solution, in step S4, the GCN network processes the data after convolution operations, combines spatial convolution and graph structure to aggregate the information of adjacent nodes to update the feature representation of each node, and is used to extract the spatial structure relationship of the data to process the nonlinear features in the multivariate time series; the specific steps are as follows:

[0023] Multiply the output feature matrix of the previous layer by the weight matrix to obtain the learnable parameters of the graph convolutional layer, multiply by the adjacency matrix and normalize to obtain the connection relationship between the nodes in the feature map, and then adjust the output of the network through the bias top, introduce nonlinearity through the activation function, and obtain the output feature matrix of the current layer.

[0024] According to the above solution, in step S5, the specific steps are as follows:

[0025] S51: Adaptively set the dynamic threshold of the number of fault feature information according to the basic threshold and weather influence factors to improve the sensitivity of detecting secondary equipment faults; when the total number of fault feature information exceeds the set threshold, trigger the fault location function;

[0026] S52: Locate the fault position through the secondary equipment fault reasoning rules; if it fails, extract the filtered fault feature set obtained in step S2, fuse it according to step S3, and input it into the STGCN model trained in step S4 to obtain the final secondary equipment fault location result.

[0027] A substation secondary equipment fault location system based on a spatio-temporal graph convolutional network model

[0028] A data acquisition sub-module, used to obtain the historical data of secondary equipment, construct a fault feature set including operating status information, SV / GOOSE reception status information, and sampled values; sort the fault feature data according to the time series and correspond it one by one with the corresponding secondary equipment faults;

[0029] A data screening sub-module, used to extract and screen highly correlated fault features to construct a filtered fault feature set;

[0030] A model training and testing sub-module, used to construct an STGCN model including three layers of 1D-CNN networks and one layer of GCN network to simultaneously learn the spatial and temporal features of graph data; divide the data in the fault feature set into a training set and a testing set, the training set is used for model learning, and the testing set is used to evaluate the performance and accuracy of the model;

[0031] A fault location sub-module, used to set a threshold for the number of fault feature information to detect secondary equipment faults; when a secondary equipment fault is detected, diagnose and locate the fault position according to the fault reasoning rules; if the reasoning rules cannot accurately locate the fault position, input the fault feature set into the STGCN model, and identify and locate the fault position of the secondary equipment through the model.

[0032] The STGCN model for substation secondary equipment fault location includes multiple stacked 1D-CNN networks for learning and capturing the time dependence between data variables, and a GCN network for extracting the spatial dependence between data variables; it also includes a fully connected layer for outputting the final result.

[0033] The beneficial effects of the present invention are:

[0034] 1. The substation secondary equipment fault location method based on the spatio-temporal graph convolutional network model of the present invention extracts feature information from the historical fault data of the substation secondary equipment, arranges the operating status of the secondary equipment, the SV / GOOSE signal reception status, and the sampled values in a time series, and generates a fault feature set; based on the change trend of the fault feature set in the time series, multi-modal feature fusion is performed on the fault feature data; based on these fault features, a spatio-temporal graph convolutional network model (STGCN) is constructed; the STGCN model is trained using historical fault samples, features highly relevant to the fault are extracted through data preprocessing and feature screening, an optimized feature set is constructed, the fault location accuracy of the model is continuously improved, and the correlation mapping relationship between the fault feature information and the fault category is comprehensively established; combined with the fault inference rule, the feature set is input into the STGCN model, and the spatio-temporal graph convolutional ability of the model is used to accurately locate the secondary equipment fault, realizing the function of improving the accuracy and efficiency of the substation secondary equipment fault diagnosis.

[0035] 2. Through the automatic reasoning and training of the deep learning model of the present invention, the efficiency and accuracy of fault location are greatly improved, providing a solid technical support for the safe and reliable operation of the power system.

[0036] 3. The present invention solves the problems of a wide variety of secondary equipment in the substation, and the over-reliance on the experience of operation and maintenance personnel or the technical support of manufacturers for fault diagnosis and repair; reduces the misdiagnosis caused by the lack of experience of operation and maintenance personnel, and avoids carrying wrong maintenance equipment, thereby improving the timely processing efficiency of faults, and ensuring the safe production, economic operation and stable power supply of the power system.

[0037] Of course, it is not necessary for any product implementing the present invention to achieve all the above-mentioned advantages simultaneously. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0039] Figure 1 is the flowchart of the embodiment of the present invention.

[0040] Figure 2 is the structural diagram of the 1D-CNN network of the embodiment of the present invention.

[0041] Figure 3 is the structural diagram of the GCN network of the embodiment of the present invention.

[0042] Figure 4It is the structural diagram of the STGCN neural network model of the embodiment of the present invention. Specific Embodiments

[0043] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0044] Embodiment 1

[0045] See Figure 1 , the specific steps of the substation secondary equipment fault location method based on the spatio-temporal graph convolutional network model are as follows:

[0046] S1: Obtain the historical data of the secondary equipment, and construct a fault feature set including operating status information, SV / GOOSE reception status information, and sampled values; sort the fault feature data according to the time series and correspond it one by one with the corresponding secondary equipment faults;

[0047] S2: Extract and screen highly correlated fault features to construct a screened fault feature set;

[0048] S3: Perform multi-modal feature fusion processing on the data of the fault feature set based on the time series to obtain fault feature fusion data;

[0049] S4: Construct an STGCN model including a three-layer 1D-CNN network and a one-layer GCN network to simultaneously learn the spatial and temporal features of graph data; divide the fault feature fusion data into a training set and a test set, the training set is used for model learning, and the test set is used to evaluate the performance and accuracy of the model; obtain the trained STGCN model;

[0050] S5: Set a dynamic threshold for the number of fault feature information under adaptive weather conditions to detect secondary equipment faults; when a secondary equipment fault is detected, diagnose and locate the fault location according to the fault inference rules; if the inference rules cannot accurately locate the fault location, input the fault feature fusion data into the trained STGCN model, and identify and locate the fault location of the secondary equipment through the model.

[0051] Furthermore, in step S1, the operating status information of the secondary equipment includes the self-check status of the merging unit, line protection, and intelligent terminal, which is used to reflect the equipment operating status through multi-dimensional data integration, support accurate fault location and early warning, and effectively improve the safety and reliability of the system.

[0052] In step S1, the SV / GOOSE message reception status information includes the message reception status of the merging unit, protection device, and intelligent terminal, and is used to judge the abnormality of sampled values and switch quantities; when an event occurs in the device, the GOOSE data set is immediately updated and data frames are sent at specific time intervals: first, the second and third frames are sent at an interval of T1, and then the fourth and fifth frames are sent at intervals of T2 and T3, where T2 is twice that of T1 and T3 is twice that of T2; after sending 5 frames, the sending interval resumes to T0 and switches to the heartbeat message state, and the message survival time is twice that of T0; if the receiver does not receive the message within 4 times of T0, it is determined that the connection is interrupted and a disconnection alarm is triggered.

[0053] In step S2, the specific steps are as follows:

[0054] S21: Preprocess the historical data to eliminate the influence of dimensions through normalization;

[0055] S22: Use the Pearson correlation coefficient to analyze the linear correlation between each feature and the fault, remove the features with low correlation, retain the features with high correlation, and construct a new fault feature set.

[0056] In step S3, the specific steps are as follows:

[0057] S31: Construct a fault feature fusion weight matrix according to the data values, change trends, and change rates of the fault feature set in the historical data;

[0058] S32: Multiply the fault feature set by the fault feature fusion weight matrix to obtain the fault feature fusion data.

[0059] In step S4, the 1D-CNN network is used to perform convolution operations in the time dimension of the data to extract time series information; the 1D-CNN network includes two convolutional layers and a flattening operation, and each convolutional layer includes a convolution operation and a pooling operation; after the second pooling operation, there is also a fully connected layer used to flatten the high-dimensional data into one-dimensional data; the specific steps are as follows:

[0060] Use the output feature matrix of the previous layer as the input feature matrix of the current layer, extract the local patterns in the input features through the weight function, introduce non-linearity through the activation function, and obtain the output of the current layer, that is, the feature map or activation value of this layer.

[0061] In step S4, the GCN network processes the data after convolution operations, combines spatial convolution with graph structure to aggregate the information of adjacent nodes to update the feature representation of each node, and is used to extract the spatial structure relationship of the data to process the non-linear features in the multivariate time series; the specific steps are as follows:

[0062] Multiply the output feature matrix of the previous layer by the weight matrix to obtain the learnable parameters of the graph convolutional layer, multiply by the adjacency matrix and normalize to obtain the connection relationship between the nodes in the feature map, and then adjust the output of the network through the bias top, introduce non-linearity through the activation function, and obtain the output feature matrix of the current layer.

[0063] In step S5, the specific steps are as follows:

[0064] S51: Adaptively set the dynamic threshold of the number of fault feature information according to the basic threshold and weather influence factors to improve the sensitivity of detecting secondary equipment faults; when the total number of fault feature information exceeds the set threshold, trigger the fault location function;

[0065] S52: Locate the fault location through the secondary equipment fault inference rule; if it fails, extract the filtered fault feature set obtained in step S2, fuse it according to step S3, and input it into the STGCN model trained in step S4 to obtain the final secondary equipment fault location result.

[0066] In this embodiment, based on the substation fault inference rule, the operating state of the secondary equipment, the SV / GOOSE information reception state, and the sampling value are selected to construct a fault feature set, and a complete fault feature representation is formed by matching the fault type code. Combining a large amount of historical fault data of secondary equipment, it is proposed to use the spatio-temporal graph convolutional neural network (STGCN) model for training, so as to establish an association mapping relationship between comprehensive fault feature information and fault categories, realize the accurate location of secondary equipment faults, effectively shorten the location time, and significantly improve the accuracy of fault location.

[0067] Embodiment 2

[0068] The steps of this embodiment are the same as those of Embodiment 1, except that each step is applied to a specific example. The specific steps are as follows:

[0069] S1: Select the characteristic information of the fault type to form the characteristic information set of the secondary equipment fault section X i . The selection of secondary equipment fault characteristic information should be based on multi-dimensional data such as equipment operating state, instrument readings, indicator light states, message reception conditions, voltage and current sampling values, and historical alarm information to comprehensively and accurately identify faults. The above fault characteristic information can be classified according to whether the device issues an abnormal alarm, the abnormal conditions of sampling and switch quantities, and the actual sampling values, so the fault information is divided into the operating state information of the secondary equipment X ZTi , SV / GOOSE (sampling value / Generic Object Oriented Substation Event) reception state information X SGi , sampling value XCYi Sort the fault feature information in time series and correspond it one by one with the corresponding secondary equipment faults.

[0070] Operating status information of secondary equipment X ZTi Summarize the self-check status of merging units, line protections, and intelligent terminals. If the self-check is abnormal T , synchronization anomaly S and device locking A . When the monitoring host receives these feature information, the element at the corresponding position is "1", otherwise it is "0". The fault feature set formed by this information reflects the device operating status through multi-dimensional data integration, supports accurate fault location and early warning, and effectively improves the security and reliability of the system.

[0071] The SV / GOOSE message receiving information of secondary equipment is an important basis for judging the abnormality of sampled values and digital inputs. When an event occurs in the device (such as a protection operation), the GOOSE data set is immediately updated and data frames are sent at specific time intervals: first, the second and third frames are sent at an interval of T1, and then the fourth and fifth frames are sent at intervals of T2 and T3, where T2 is twice that of T1 and T3 is twice that of T2. After five frames are completed, the sending interval returns to T0 and switches to the heartbeat message state, and the message survival time is twice that of T0. If the receiving party does not receive the message within 4 times of T0, it is determined that the connection is interrupted and a link break alarm is triggered. SV / GOOSE receiving status information X SGi Integrate the message receiving status of devices such as merging units, protection devices, and intelligent terminals. If the message is not received, then X SGi the corresponding position in

[0072] Sampled value X CYi Represents the three-phase voltage and current sampling data of the two channels of the protection device.

[0073] Fault feature set X i As shown in (6), the operating status information of secondary equipment X ZTi , SV / GOOSE acceptance status information X SGi , sampled value X CYi Are shown in (7), (8), and (9) respectively.

[0074] (6)

[0075] (7)

[0076] (8)

[0077] (9)

[0078] In the formula, represents the feature set of the i th fault event, represents the operating state information of secondary equipment, represents the receiving status of SV / GOOSE messages, is the sampled value of secondary equipment. The total number of fault events is N , where k , i , j are the numbers of merging units, protection devices, and intelligent terminal devices respectively. The parameters , , correspond to the self-check information of the merging unit a , protection device b , and intelligent terminal c , covering abnormal status G , synchronization anomaly H , locking L , etc. When the self-check information is abnormal during a fault, the value of the abnormal position is 1, otherwise it is 0. The SV / GOOSE reception status set represents the message reception status of the merging unit, measuring and controlling device, bus / line protection, and intelligent terminal. The total number of messages is all , is the reception status of the i th message, and the reception status of the n devices subscribing to this message is E . When a fault causes a disconnection, if the secondary equipment j does not receive the message x , then Exj is 1, otherwise it is 0. The sampled value is the three-phase voltage and current sampling of the protection device's dual channels. Channels 1 and 2 are respectively C 1 and C 2 , I , U represent three-phase current and voltage.

[0079] S2: Perform fault feature extraction and screening. First, normalize the historical data to eliminate the influence of different dimensions. Then, use the Pearson correlation coefficient to analyze the linear correlation between each feature and the fault result, screen out features with low correlation, and retain features with higher correlation to construct a refined fault feature set.

[0080] To improve the model accuracy and solve the problem of inconsistent data dimensions, the Min-Max normalization method is used to process the fault feature data. The conversion function of Min-Max normalization is shown in Equation (1).

[0081] (1)

[0082] In the formula X max and X min are the maximum and minimum values in the dataset, respectively.

[0083] During the fault location process, the Pearson correlation coefficient formula is used to quantitatively evaluate the linear correlation between each fault feature and the fault result, so as to clarify the relationship between the feature and the fault and provide a basis for feature screening.

[0084] (2)

[0085] In the formula: is the correlation coefficient between various factors; and are the values of the two factors at the i th data point; and are the means of the two factors; n is the number of data points.

[0086] Through the Pearson correlation coefficient, low-correlation fault features can be screened out, the applicability of the invention can be improved, the data input volume can be reduced, and the analysis accuracy can be improved.

[0087] S3: Perform multi-modal feature fusion processing on the data of the fault feature set based on the time series to obtain the fault feature fusion data; the specific steps are as follows:

[0088] S31: Construct a fault feature fusion weight matrix according to the data values, change trends, and change rates of the fault feature set in the historical data;

[0089] Let the data in the fault feature set be ; a is a constant greater than 1; the fault feature over-limit coefficient is:

[0090]

[0091] Let b be a constant less than 1, represents the change amount of the fault feature within the time period; the change trend coefficient is:

[0092]

[0093] Let be the basic coefficient of fault features, which is adjusted according to the importance of fault features and usually takes the value of 1; then the fault feature fusion coefficient is as follows:

[0094]

[0095] The fusion coefficient of fault features is obtained by multiplying the three, highlighting the importance degree of the value and change trend of this fault feature in a certain time period, which is beneficial to improving the recognition rate of the model for abnormal changes of this fault feature.

[0096] Combining the fault feature fusion coefficients corresponding to all fault features to obtain the fault feature fusion weight matrix.

[0097] S32: Multiply the fault feature set by the fault feature fusion weight matrix to obtain the fault feature fusion data.

[0098] S4: Construct the corresponding STGCN model, select some of the fault feature fusion data for training the model, and use the remaining fault feature fusion data as the test set to verify the model performance.

[0099] The graph embedding method of traditional GCN performs well in learning spatial relationships, but has limited capture of time dependencies. As shown in the appendix Figure 4 This embodiment uses an STGCN module, which consists of three layers of one-dimensional convolutional neural network 1D-CNN and one layer of GCN, and can learn the spatial and temporal features of graph data simultaneously. Among them, 1D-CNN is used to extract time series information, and GCN is used to capture spatial structure relationships. Through the collaborative operation of these two operations, STGCN can effectively model the spatio-temporal dependence relationships in the data, so as to make more accurate predictions and analyses on the spatio-temporal dynamics of graph structure data.

[0100] In STGCN, the parameter sharing between different channels is non-parallel. This design choice helps to maintain the independence of graph structure learning, thereby enhancing the generalization ability and interpretability of the model.

[0101] 1D-CNN is a model constructed based on the biological visual perception mechanism, which can perform supervised learning and unsupervised learning. 1D-CNN utilizes the convolutional kernel parameter sharing in the hidden layer and the sparsity of the inter-layer connections. With less computational complexity, it can extract deep local features from high-dimensional data and generate effective representations through convolutional layers and pooling layers. The network structure of 1D-CNN usually includes two convolutional layers and a flattening operation, and each convolutional layer consists of a convolutional operation and a pooling operation. After the second pooling operation, the fully connected layer can be used to flatten the high-dimensional data into one-dimensional data for more convenient subsequent processing. The 1D-CNN structure is as shown in the appendixFigure 2 as shown

[0102] The time dependence inside the nodes is extracted by using 1D-CNN in the STGCN model. In this embodiment, the convolution operation is only performed on the time dimension of the data, and the specific method is as follows:

[0103]

[0104] In the formula: is the output of the l th layer, which is the feature map or activation value of this layer; is the activation function, whose role is to introduce nonlinearity to ensure that the neural network can fit complex functions; is the weight matrix, which is used to extract local patterns in the input features; is the output of the th layer and is also the input of the l th layer.

[0105] In the GCN module of the STGCN model, the convolutional neural network is used to extract the spatial dependence of the data, aiming to process the nonlinear features in the multivariate time series. By processing the data after the convolution operation, the GCN module aggregates the information of adjacent nodes, thereby updating the feature representation of each node. This is achieved by the combination of spatial convolution and the graph structure.

[0106]

[0107] In the formula: is the output feature matrix of the l th layer, representing the feature representation of all nodes in the graph at this layer; is the activation function, which can introduce nonlinearity; is the normalized degree matrix, mainly to prevent numerical instability and weight deviation; is the adjacency matrix, representing the connection relationship between nodes in the graph; is the output feature matrix of the previous layer; is the weight matrix of the l th layer, that is, the learnable parameter of the graph convolutional layer; is the bias of the l th layer, which is used to adjust the output of the network during calculation. The GCN structure is as shown in Appendix Figure 3 as shown

[0108] S5: When a secondary device fails, first use the fault inference rule to perform inference to determine the location of the fault. If the inference cannot accurately locate, input the fault feature set into the model, and further use the model to locate the fault of the secondary device. The specific steps are as follows:

[0109] S51: Adaptively set the dynamic threshold of the number of fault feature information according to the basic threshold and weather influence factors to improve the sensitivity of detecting secondary equipment faults; when the total number of fault feature information exceeds the set dynamic threshold, trigger the fault location function.

[0110] Let the weather influence factor coefficient be:

[0111]

[0112] Let the basic threshold be , then the dynamic threshold is:

[0113]

[0114] When the weather conditions are severe, the threshold will automatically decrease to improve the sensitivity of fault detection.

[0115] S52: First, try to locate through the secondary equipment fault inference rules. If it fails, extract and screen the fused feature information and sampling values to form a feature set X i , and input it into the trained STGCN fault location model to obtain the final location result. Part of the inference rules are shown in Table 1.

[0116] Table 1 Partial Inference Rule Table

[0117] Number Fault type Characteristic information 1 Fault of the merging unit DSP or sampling DSP module Abnormal sampling of the merging unit, abnormal synchronization of the merging unit, total alarm of the merging unit / protection SV, protection blocking, etc. 2 Configuration error of the merging unit Configuration error of the board, error in the GOOSE configuration file, merging unit / protection blocking, etc. 3 Configuration error of the intelligent terminal Incorrect input wiring, error in the GOOSE configuration file, configuration error of the board, intelligent terminal blocking, etc. 4 Fault of the power board of the intelligent terminal Power loss alarm of the intelligent terminal 5 Fault of the power module of the merging unit Power loss alarm of the merging unit 6 Fault of the I / O board of the intelligent terminal Communication interruption of the merging unit / measurement and control / intelligent terminal / protection GOOSE, total alarm of the merging unit / intelligent terminal / protection GOOSE, etc. 7 Configuration error of the line protection Error in the SV / GOOSE configuration file, configuration error of the board, version error alarm, protection blocking, etc. 8 Fault of the I / O, SV, GOOSE plug-ins of the line protection Total alarm of the protection SV / GOOSE, total alarm of the intelligent terminal GOOSE, interruption of the protection SV / GOOSE, protection blocking, reclosing blocking, etc. 9 Fault of the power plug-in of the line protection Power loss alarm of the line protection

[0118] It should be understood that the magnitudes of the sequence numbers of the steps in the above embodiments do not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0119] Embodiment 3

[0120] This embodiment is used to implement the principle of the above method embodiment to construct a substation secondary equipment fault location system based on a spatio-temporal graph convolutional network model, including a data acquisition sub-module, a data screening sub-module, a model training and testing sub-module, and a fault location sub-module;

[0121] The data acquisition sub-module is used to acquire the historical data of secondary equipment, construct a fault feature set including operation status information, SV / GOOSE reception status information, and sampling values; sort the fault feature data according to the time series and correspond it to the corresponding secondary equipment faults one by one;

[0122] The data screening sub-module is used to extract and screen highly correlated fault features to construct a screened fault feature set;

[0123] The model training and testing sub-module is used to build an STGCN model including a three-layer 1D-CNN network and a one-layer GCN network to simultaneously learn the spatial and temporal features of graph data; divide the data in the fault feature set into a training set and a testing set, where the training set is used for model learning, and the testing set is used to evaluate the performance and accuracy of the model.

[0124] The fault location sub-module is used to set a threshold for the number of fault feature information to detect secondary equipment faults; when a secondary equipment fault is detected, diagnose and locate the fault location according to the fault reasoning rules; if the reasoning rules cannot accurately locate the fault location, input the fault feature set into the STGCN model to identify and locate the fault location of the secondary equipment through the model.

[0125] Each sub-module is mainly used to implement each step of the method embodiment, which will not be elaborated here.

[0126] Embodiment 4

[0127] This embodiment is used to implement the principle of the above method embodiment to build an STGCN model for fault location of secondary equipment in a substation, including multiple stacked 1D-CNN networks for learning and capturing the temporal dependence between data variables, and a GCN network for extracting the spatial dependence between data variables; it also includes a fully connected layer for outputting the final result.

[0128] It should be noted that according to the needs of implementation, each step / component described in this application can be split into more steps / components, or two or more steps / components or partial operations of steps / components can be combined into new steps / components to achieve the purpose of the present invention.

[0129] This embodiment also includes a processor, a communication interface, a memory, and a communication bus; where the processor, communication interface, and memory complete communication with each other through the communication bus; a computer program is stored in the memory, and when the program is executed by the processor, the processor is caused to execute the steps of the method for fault location of secondary equipment in a substation based on a spatio-temporal graph convolutional network model.

[0130] This embodiment also provides a computer-readable storage medium, on which executable instructions are stored, and when the instructions are executed by the processor, the processor is caused to implement the method for fault location of secondary equipment in a substation based on a spatio-temporal graph convolutional network model.

[0131] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects.

[0132] Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) that contain computer-usable program code.

[0133] The present application is described with reference to the flowchart of the method and computer program product according to Embodiment 1 of the present application and the block diagram of the system according to Embodiment 3. It should be understood that each process or block in the flowchart or block diagram can be implemented by computer program instructions, as well as the combination of processes or blocks in the flowchart or block diagram.

[0134] These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a substation secondary equipment fault location system based on a spatio-temporal graph convolutional network model for implementing the functions specified in one process Figure 1 one process or multiple processes or blocks Figure 1 or multiple blocks.

[0135] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing devices to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device implements the functions specified in one process Figure 1 one process or multiple processes or blocks Figure 1 or multiple blocks.

[0136] These computer program instructions can also be loaded onto a computer or other programmable data processing devices, so that a series of operation steps are executed on the computer or other programmable devices to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable devices provide the steps of a substation secondary equipment fault location method based on a spatio-temporal graph convolutional network model for implementing the functions specified in one process Figure 1 one process or multiple processes or blocks Figure 1 or multiple blocks.

[0137] The above embodiments are only used to illustrate the design concept and features of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The protection scope of the present invention is not limited to the above embodiments. Therefore, all equivalent changes or modifications made according to the principles and design concepts disclosed by the present invention are within the protection scope of the present invention.

Claims

1. A substation secondary equipment fault location method based on a spatiotemporal graph convolutional network model, characterized by: The following steps are involved: S1: Obtain historical data of secondary equipment, construct a fault feature set including operating status information, SV / GOOSE receiving status information and sampling values; sort the fault feature data in time series and correspond them to the corresponding secondary equipment faults one by one; S2: Extract and filter high-correlation fault features and construct a filtered fault feature set; S3: Perform multi-modal feature fusion processing on the data of the fault feature set based on the time series to obtain fault feature fusion data; The specific steps are: S31: construct a fault feature fusion weight matrix according to the data values, change trends and change rates of the fault feature set in the historical data; Assume that the data in the fault feature set is ; a is a constant greater than 1; Fault characteristic over-limit coefficient for: ； Let b be a constant less than 1, Indicates that the fault characteristics are The amount of change within a period; the coefficient of change trend for: ； set up is the basic coefficient of fault characteristics, which is adjusted according to the importance of fault characteristics; then the fault characteristic fusion coefficient for: ； The fault feature fusion coefficients corresponding to all fault features are combined to obtain a fault feature fusion weight matrix; S32: multiplying the fault feature set by the fault feature fusion weight matrix to obtain fault feature fusion data; S4: Construct an STGCN model consisting of three layers of 1D-CNN networks and one layer of GCN network to simultaneously learn the spatial and temporal features of feature map data; The fault feature fusion data is divided into a training set and a test set. The training set is used for model learning, and the test set is used to evaluate the performance and accuracy of the model. The trained STGCN model is obtained. S5: Set a dynamic threshold for the number of fault feature information that is adaptive to weather conditions to detect secondary equipment failures; when a secondary equipment failure is detected, diagnose and locate the fault location based on the fault reasoning rules; if the reasoning rules cannot accurately locate the fault location, input the fault feature fusion data into the trained STGCN model, and use the model to identify and locate the fault location of the secondary equipment.

2. The method for locating faults of secondary equipment in a substation based on a spatiotemporal graph convolutional network model according to claim 1 is characterized in that: In step S1, the operating status information of the secondary equipment includes the self-test status of the merging unit, line protection and intelligent terminal, which is used to reflect the equipment operating status through multi-dimensional data integration, support accurate fault location and early warning, and effectively improve the safety and reliability of the system.

3. The method for locating faults of secondary equipment in a substation based on a spatiotemporal graph convolutional network model according to claim 1 is characterized in that: In the step S1, the SV / GOOSE message reception status information includes the message reception status of the merging unit, the protection device and the intelligent terminal, which is used to judge the abnormality of the sampling value and the switching value; When an event occurs on the device, the GOOSE data set is immediately updated and data frames are sent at specific time intervals: first, the second and third frames are sent at intervals of T1, and then the fourth and fifth frames are sent at intervals of T2 and T3, where T2 is twice T1 and T3 is twice T2; after sending 5 frames, the sending interval is restored to T0 and it is converted to the heartbeat message state, and the message survival time is twice T0; if the receiver does not receive the message within 4 times T0, it is determined that the connection is interrupted and a link break alarm is triggered.

4. The method for locating faults of secondary equipment in a substation based on a spatiotemporal graph convolutional network model according to claim 1 is characterized in that: In the step S2, the specific steps are: S21: Preprocess the historical data and eliminate the influence of dimension by normalization; S22: The Pearson correlation coefficient is used to analyze the linear correlation between each feature and the fault, remove the low-correlation features, retain the high-correlation features, and construct a new fault feature set.

5. The method for locating faults of secondary equipment in a substation based on a spatiotemporal graph convolutional network model according to claim 1 is characterized in that: In step S4, the 1D-CNN network is used to perform a convolution operation on the time dimension of the data to extract time series information; the 1D-CNN network includes two convolution layers and a flattening operation, each convolution layer includes a convolution operation and a pooling operation; after the second pooling operation, a fully connected layer is also included to flatten the high-dimensional data into one-dimensional data; the specific steps are: The output feature matrix of the previous layer is used as the input feature matrix of the current layer. The local patterns in the input features are extracted through the weight function. Nonlinearity is introduced through the activation function to obtain the output of the current layer, that is, the feature map or activation value of the layer.

6. The method for locating faults of secondary equipment in a substation based on a spatiotemporal graph convolutional network model according to claim 1 is characterized in that: In step S4, the GCN network updates the feature representation of each node by processing the data after the convolution operation, combining spatial convolution with the information of adjacent nodes aggregated by the graph structure, and is used to extract the spatial structural relationship of the data to process the nonlinear features in the multivariate time series; The specific steps are: The output feature matrix of the previous layer is multiplied by the weight matrix to obtain the learnable parameters of the graph convolution layer, and is multiplied by the adjacency matrix and normalized to obtain the connection relationship between the nodes in the feature graph. After adjusting the output of the network through the bias top, nonlinearity is introduced through the activation function to obtain the output feature matrix of the current layer.

7. The method for locating faults of secondary equipment in a substation based on a spatiotemporal graph convolutional network model according to claim 1 is characterized in that: In the step S5, the specific steps are: S51: adaptively setting a dynamic threshold of the number of fault characteristic information according to the basic threshold and weather influencing factors to improve the sensitivity of detecting secondary equipment faults; when the total number of fault characteristic information exceeds the set threshold, triggering the fault location function; S52: Locate the fault location through the secondary equipment fault inference rule; if it fails, extract the filtered fault feature set obtained in step S2, fuse it according to step S3 and input it into the STGCN model trained in step S4 to obtain the final secondary equipment fault location result.

8. The substation secondary equipment fault location system based on the spatiotemporal graph convolutional network model is characterized by: The data acquisition submodule is used to obtain the historical data of the secondary equipment, build a fault feature set including the operating status information, SV / GOOSE receiving status information and sampling values; sort the fault feature data according to the time series and correspond them one by one with the corresponding secondary equipment faults; The data screening submodule is used to extract and screen the fault features with high correlation and construct the screened fault feature set; The feature fusion submodule is used to perform multi-modal feature fusion processing on the data of the fault feature set based on the time series to obtain fault feature fusion data; Specifically include: Construct a fault feature fusion weight matrix based on the data values, change trends and change rates of the fault feature set in the historical data; Assume that the data in the fault feature set is ; a is a constant greater than 1; fault characteristic over-limit coefficient for: ； Let b be a constant less than 1, Indicates that the fault characteristics are The amount of change within a period; the coefficient of change trend for: ； set up is the basic coefficient of fault characteristics, which is adjusted according to the importance of fault characteristics; then the fault characteristic fusion coefficient for: ； The fault feature fusion coefficients corresponding to all fault features are combined to obtain a fault feature fusion weight matrix; Multiplying the fault feature set by the fault feature fusion weight matrix to obtain fault feature fusion data; The model training and testing submodule is used to build an STGCN model including three layers of 1D-CNN network and one layer of GCN network to simultaneously learn the spatial and temporal features of graph data; divide the data in the fault feature set into a training set and a test set, the training set is used for model learning, and the test set is used to evaluate the performance and accuracy of the model; and obtain the trained STGCN model; The fault location submodule is used to set the threshold of the number of fault feature information to detect secondary equipment faults. When a secondary equipment fault is detected, the fault location is diagnosed and located according to the fault reasoning rules. If the reasoning rules cannot accurately locate the fault location, the fault feature set is input into the STGCN model to identify and locate the fault location of the secondary equipment through the model.

Images