Power transformer state evaluation method based on multi-metric residual image convolutional network
The power transformer status evaluation is carried out through a multi-metric residual graph convolution network, which solves the problem of insufficient variable correlation mining in the existing technology, and achieves higher evaluation accuracy and timely online monitoring to ensure the safe and stable operation of the transformer.
Patent Information
- Application Number
- CN202411074769.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-07
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-08-07
AI Technical Summary
The existing power transformer state monitoring method based on deep learning has insufficient variable correlation mining when processing multiple monitoring data, and it is difficult to effectively capture the time-varying and nonlinear characteristics of the transformer operating state, resulting in insufficient evaluation accuracy.
Using multi-metric residual graph convolution network, through graph structure modeling and multi-metric fusion composition methods, k-nearest neighbor graphs are constructed, and combined with graph convolution layer, batch layer and ReLU activation function, residual graph convolution network is constructed for features. Extraction to improve the accuracy of status evaluation.
It improves the accuracy of power transformer status evaluation, can timely monitor the transformer operating status online, and reduces the risk of equipment damage.
Smart Images

Figure CN118940120A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of power transformer status monitoring and health assessment, and more specifically, relates to a power transformer health status assessment method. Background Art
[0002] As the core equipment of modern power grids, transformers play an important role in tasks such as power transmission and voltage conversion. Their safe and stable operation is crucial to the reliability and stability of the power grid. The operating conditions of transformers are relatively complex. Once a failure occurs, it will affect the normal operation of the power system and even cause serious economic losses. Therefore, monitoring the operating status of transformers is of great significance to ensure the safety and stability of the power grid and reduce equipment damage.
[0003] Transformer condition monitoring methods can be mainly divided into offline preventive monitoring and online monitoring methods. Offline monitoring methods include electrical tests such as partial discharge, but such methods have a certain lag. In recent years, deep learning-based methods have been widely used in the field of condition assessment due to their outstanding advantages in extracting deep feature representations. However, deep learning-based models must use regular Euclidean data as input. When the transformer operating state is abnormal, multiple monitoring data affect each other and change at the same time, with time-varying and nonlinear characteristics. These methods are still insufficient in variable correlation mining.
[0004] As the transformer operating environment changes, the dissolved gas in the oil and other monitoring data will decompose, synthesize and interact with each other. Therefore, mining the correlation of multivariate data is crucial to improving model performance. Graph neural networks use non-Euclidean data as input and can better model the spatial correlation of variables. To this end, graph structure modeling is performed on the power transformer operation monitoring data to fully mine the spatial correlation of the data. A state monitoring scheme based on multi-metric residual graph convolution is proposed to improve the accuracy of state assessment. Summary of the invention
[0005] In view of the above defects or improvement needs of the prior art, the present invention proposes a power transformer state assessment method based on a multi-metric residual graph convolutional network to improve the accuracy of state assessment.
[0006] To achieve the above object, the present invention provides a method for evaluating the state of a power transformer based on a multi-metric residual graph convolutional network, comprising:
[0007] Obtain the data of dissolved gas in oil under different health conditions of the transformer, divide the data into a training set and a test set, and then perform balanced preprocessing on the data in the training set;
[0008] The data in the training set and the test set after balanced preprocessing are subjected to maximum and minimum normalization processing, and then the multi-metric fusion graph construction method is used to construct a k-nearest neighbor graph from the normalized data in the training set and the test set;
[0009] A feature extraction unit is constructed based on a graph convolution layer, a batch layer, a ReLU activation function, and a graph pooling layer, and a residual graph convolution network is constructed based on multiple feature extraction units as a state evaluation model;
[0010] The k-nearest neighbor graph of the training set and the test set data is input into the constructed residual graph convolutional network, and the state evaluation model is trained to evaluate the operating state of the transformer through the trained state evaluation model.
[0011] In some optional implementation schemes, the step of obtaining the oil dissolved gas data under different health states of the transformer, dividing the data into a training set and a test set, and then performing a balanced preprocessing on the data in the training set includes:
[0012] Obtain the dissolved gas data in oil of the transformer's historical operating data, which are H2, CH4, C2H6, C2H4, and C2H2. Divide the dissolved gas data in oil into multiple operating states according to the size of the health index, and divide them into training sets and test sets according to the proportion.
[0013] The SMOTE Tomek mixed sampling algorithm is used to process the data in the training set to balance the various types of data in the training set.
[0014] In some optional implementation schemes, the data in the training set and the test set after balanced preprocessing are subjected to maximum and minimum normalization processing, and then a multi-metric fusion graph construction method is used to construct a k-nearest neighbor graph of the normalized data in the training set and the test set, including:
[0015] The maximum and minimum normalization method is used to normalize the data in the training set and the test set after balanced preprocessing;
[0016] Each monitoring variable is regarded as a node, and the correlation between monitoring variables is regarded as an edge graph, where the monitoring variable represents the dissolved gas in the oil;
[0017] The correlation between each node is calculated by Gaussian kernel function, cosine similarity and Pearson correlation coefficient, and the three calculated adjacency matrices are averaged to fuse multiple metrics to obtain the fused adjacency matrix.
[0018] For each node, only the k nodes with the greatest correlation with the node are retained as the final correlation matrix A to construct the k-nearest neighbor graph.
[0019] In some optional implementation schemes, the feature extraction unit is constructed based on the graph convolution layer, the batch processing layer, the ReLU activation function, and the graph pooling layer, and a residual graph convolution network is constructed based on multiple feature extraction units as a state evaluation model, including:
[0020] A feature extraction unit is constructed based on the graph convolution layer, batch processing layer, and ReLU activation function. Each feature extraction unit is connected to the EdgePool graph pooling layer.
[0021] Several feature extraction units are stacked, and residual connections are added to every other feature extraction unit to construct a feature extraction network. The extracted features are input to the fully connected layer and classified by the softmax function to output the state evaluation results.
[0022] The constructed k-nearest neighbor graph is input into the residual graph convolutional network to obtain the result of state assessment and complete the state assessment model training.
[0023] In some optional implementations, the EdgePool graph pooling layer output is: output =EdgePool(X′), where X′=Relu(BN(GConv(AXW))), A is the adjacency matrix, X is the input data, and W is the trainable weight matrix.
[0024] In some optional implementation schemes, evaluating the operating state of the transformer by using a trained state evaluation model includes:
[0025] After the maximum and minimum normalization processing is performed on the transformer operation data, the multi-metric fusion graph construction method is used to construct a k-nearest neighbor graph of the transformer operation data after balance preprocessing. The k-nearest neighbor graph data of the transformer operation data is input into the trained state assessment model to obtain the online monitoring results of the transformer operation status.
[0026] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art:
[0027] The present invention proposes a method for evaluating the state of power transformers based on a multi-metric residual graph convolutional network, which models the graph structure of dissolved gas in power transformer oil and fully explores the spatial correlation of data. A multi-metric fusion graph construction method is used to construct a k-nearest neighbor graph of multiple monitoring variables, which is then input into the constructed residual graph convolutional network for feature extraction, thereby realizing online state evaluation of power transformers, processing and maintaining transformers in a timely manner, and avoiding deep damage. BRIEF DESCRIPTION OF THE DRAWINGS
[0028] Figure 1 It is a schematic diagram of a method flow provided by an embodiment of the present invention;
[0029] Figure 2 is a schematic diagram of a multi-metric fusion composition method provided by an embodiment of the present invention;
[0030] Figure 3 It is a schematic diagram of a residual graph convolutional network provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0031] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with 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 intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.
[0032] like Figure 1 As shown, a method for evaluating the state of a power transformer based on a multi-metric residual graph convolutional network includes five parts: data balance preprocessing, multi-metric fusion composition, spatiotemporal feature learning, offline training model, and online state evaluation. Specifically, the method includes the following steps:
[0033] Step S1: Obtain the data of dissolved gas in oil under different health states of the transformer, divide the data into a training set and a test set, and then perform balanced preprocessing on the data in the training set;
[0034] Furthermore, step S1 can be implemented in the following manner:
[0035] Step S11: Obtain the dissolved gas data in the oil of the historical operation data of the transformer, which are H2, CH4, C2H6, C2H4, and C2H2, and divide them into five operating states: very good, good, moderate, poor, and very poor according to the size of the health index, and divide them into a training set and a test set at a ratio of 70% and 30%;
[0036] For example, the data set is divided into five operating states: very good, good, moderate, poor, and very poor according to the health index of 0.8 to 1, 0.6 to 0.8, 0.4 to 0.6, 0.2 to 0.4, and 0 to 0.2.
[0037] Step S12: Use the SMOTE Tomek mixed sampling algorithm to process the training set data so that the various types of data in the training set are balanced. The specific steps are as follows:
[0038] Step S121: randomly select some samples x in the minority class, calculate the Euclidean distance between them and other samples in the minority class sample set, and obtain their k nearest neighbors;
[0039] Step S122: Randomly select several samples x from the k nearest neighbors n, for a randomly selected neighbor sample x n , according to x new =x+rand(0,1)*|xx n |Construct new samples and add them to the minority class samples;
[0040] Step S123: For most categories, the nearest neighbor of each sample is calculated by Euclidean distance, and it is determined whether the category labels of each pair of samples are the same;
[0041] Step S124: For samples of different categories, a Tomek Link pair is established and marked as noise and removed from the data set.
[0042] Step S2: Perform maximum and minimum normalization processing on the data in the training set and the test set after balanced preprocessing, and then use the multi-metric fusion graph construction method to construct a k-nearest neighbor graph for the normalized training set and test set data;
[0043] Furthermore, step S2 can be implemented in the following manner:
[0044] Step S21: Use the maximum and minimum normalization method to normalize the divided training set and test set data. The calculation is shown as follows: Among them, x′ represents the value of a single data, min is the minimum value of the column where the data is located, and max is the maximum value of the column where the data is located;
[0045] Step S22: Calculate the correlation between nodes by Gaussian kernel function, cosine similarity and Pearson correlation coefficient respectively, and take the average of the three adjacency matrices obtained. For each node, retain the k nodes with the greatest correlation with the node as the final correlation matrix A to construct a k-nearest neighbor graph.
[0046] Among them, multi-metric fusion composition methods such as Figure 2 As shown. For the edge weights between every two nodes, the similarity between each node is calculated using the Gaussian kernel function, cosine similarity, and Pearson correlation coefficient. The Gaussian kernel function measures similarity by projecting vectors into high-order space, the cosine distance is used to measure the difference between sample vectors, and the Pearson correlation coefficient is used to measure the linear relationship between two variables. i With node x j The calculation formulas of Gaussian kernel function, cosine similarity and Pearson correlation coefficient are as follows:
[0047]
[0048] where ξ is the bandwidth of the Gaussian kernel.
[0049]
[0050] The three calculated adjacency matrices are added together and the average is taken as the final correlation matrix. To reduce redundant connections, for each node, the k nodes with the largest weights are selected as its neighbor nodes, and the edges connected to them are retained. Finally, the redundant edges are deleted, that is, their weights are set to 0, and the correlation matrix A of the k-nearest neighbor graph can be obtained.
[0051] Step S3: constructing a feature extraction unit based on the graph convolution layer, the batch processing layer, the ReLU activation function, and the graph pooling layer, and constructing a residual graph convolution network as a state evaluation model based on multiple feature extraction units;
[0052] like Figure 3 The schematic diagram of the residual graph convolutional network is shown as follows: the proposed residual graph convolutional network consists of 5 stacked feature extraction modules, each of which consists of a graph convolutional network (GConv), a batch layer, and a ReLU activation function. The output of a single feature extraction unit can be expressed as X′=Relu(BN(GConv(AXW))), where A is the adjacency matrix, X is the input data, and W is the trainable weight matrix.
[0053] Each feature extraction unit is connected to the EdgePool graph pooling layer. The EdgePool graph pooling layer can reduce the number of nodes, reduce computing requirements, and improve model performance. Therefore, the output data is input to the EdgePool layer for pooling, and its output can be expressed as X output =EdgePool(X′). Since the Laplace smoothing is easily applied repeatedly to different clusters when the graph convolution network has a deeper number of layers, resulting in a decrease in classification performance, in order to prevent the repeated application of Laplace smoothing to different clusters when the graph convolution network has a deeper number of layers, a residual connection is added every other feature extraction unit. After feature extraction, it is input to two fully connected layers and classified by the softmax classifier to obtain the final state evaluation result.
[0054] Step S4: input the k-nearest neighbor graph of the training set and the test set data obtained in step S2 into the residual graph convolutional network constructed in step S3, train the state evaluation model, and evaluate the operating state of the transformer;
[0055] Step S5: For the online monitoring data of the transformer, the data is preprocessed by step S1, and then the k-nearest neighbor graph is constructed by step S2. The graph structure data of the online monitoring data is input into the trained state assessment model to realize the online state assessment of the power transformer.
[0056] It should be pointed out that, according to the needs of implementation, the various steps / components described in this application can be split into more steps / components, and 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.
[0057] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for power transformer state assessment based on multi-metric residual graph convolutional network, characterized in that: include: Obtain the data of dissolved gas in oil under different health conditions of the transformer, divide the data into a training set and a test set, and then perform balanced preprocessing on the data in the training set; The data in the training set and the test set after balanced preprocessing are subjected to maximum and minimum normalization processing, and then the multi-metric fusion graph construction method is used to construct a k-nearest neighbor graph from the normalized data in the training set and the test set; A feature extraction unit is constructed based on a graph convolution layer, a batch layer, a ReLU activation function, and a graph pooling layer, and a residual graph convolution network is constructed based on multiple feature extraction units as a state evaluation model; The k-nearest neighbor graph of the training set and the test set data is input into the constructed residual graph convolutional network, and the state evaluation model is trained to evaluate the operating state of the transformer through the trained state evaluation model.
2. The method according to claim 1, characterized in that The method obtains the data of dissolved gas in oil under different health states of the transformer, divides the data into a training set and a test set, and then performs a balanced preprocessing on the data in the training set, including: Obtain the dissolved gas data in oil of the transformer's historical operating data, which are H2, CH4, C2H6, C2H4, and C2H2. Divide the dissolved gas data in oil into multiple operating states according to the size of the health index, and divide them into training sets and test sets according to the proportion. The SMOTE Tomek mixed sampling algorithm is used to process the data in the training set to balance the various types of data in the training set.
3. The method according to claim 1 or 2, characterized in that: The method performs maximum and minimum normalization processing on the data in the training set and the test set after the balanced preprocessing, and then uses a multi-metric fusion graph construction method to construct a k-nearest neighbor graph from the normalized data in the training set and the test set, including: The maximum and minimum normalization method is used to normalize the data in the training set and the test set after balanced preprocessing; Each monitoring variable is regarded as a node, and the correlation between monitoring variables is regarded as an edge graph, where the monitoring variable represents the dissolved gas in the oil; The correlation between each node is calculated by Gaussian kernel function, cosine similarity and Pearson correlation coefficient, and the three calculated adjacency matrices are averaged to fuse multiple metrics to obtain the fused adjacency matrix. For the fused adjacency matrix, each node retains the k nodes with the greatest correlation with the node as the final correlation matrix A to construct the k-nearest neighbor graph.
4. The method according to claim 3, characterized in that The method constructs a feature extraction unit based on a graph convolution layer, a batch processing layer, a ReLU activation function, and a graph pooling layer, and constructs a residual graph convolution network as a state evaluation model based on multiple feature extraction units, including: A feature extraction unit is constructed based on the graph convolution layer, batch processing layer, and ReLU activation function. Each feature extraction unit is connected to the EdgePool graph pooling layer. Several feature extraction units are stacked, and residual connections are added to every other feature extraction unit to construct a feature extraction network. The extracted features are input to the fully connected layer and classified by the softmax function to output the state evaluation results. The constructed k-nearest neighbor graph is input into the residual graph convolutional network to obtain the result of state assessment and complete the state assessment model training.
5. The method according to claim 4, characterized in that The output of the EdgePool graph pooling layer is: X output =EdgePool(X′), where X′=Relu(BN(GConv(AXW))), A is the adjacency matrix, X is the input data, and W is the trainable weight matrix.
6. The method according to claim 1, characterized in that The step of evaluating the operating state of the transformer by using the trained state evaluation model includes: After the maximum and minimum normalization processing is performed on the transformer operation data, the multi-metric fusion graph construction method is used to construct a k-nearest neighbor graph of the transformer operation data after balance preprocessing. The k-nearest neighbor graph data of the transformer operation data is input into the trained state assessment model to obtain the online monitoring results of the transformer operation status.
Citation Information
Patent Citations
Power transformer health state assessment method based on multi-mode neural network
CN114462508A
Transformer fault identification method and system based on residual image convolutional neural network
CN118378178A
Deep parallel fault diagnosis method and system for dissolved gas in transformer oil
US20210278478A1
