Power system transient state and voltage stability collaborative prediction method based on time-space diagram neural network

Abstract

The invention relates to an electric power system transient state and voltage stability collaborative prediction method based on a space-time diagram neural network, and belongs to the field of electric power system stability evaluation. The method comprises the following steps: constructing a power grid information adjacency matrix according to a topological structure and electrical parameters of a power system; based on the power grid information adjacency matrix, extracting topological features and time features of the power system from the PMU data; and fusing the extracted topological features and time features, and outputting prediction results of transient state and voltage stability of the power system. According to the method, the accuracy and robustness of stability prediction of the power system can be effectively improved, the real-time requirement is met, and technical support is provided for safe and stable operation of the power system.

Description

Technical Field

[0001] The present invention belongs to the field of power system stability assessment, and in particular relates to a method for collaboratively predicting power system transient and voltage stability based on a spatiotemporal graph neural network. Background Art

[0002] Power system stability refers to the ability of a power system to recover and maintain a stable operating state after a disturbance.

[0003] With economic development, more and more power systems need to operate under high load conditions. At the same time, the use of a large amount of renewable energy and the introduction of power electronic equipment have brought new challenges to the stable operation of power systems. Therefore, the problem of power system stability has become increasingly important.

[0004] The stability issues of power systems are mainly divided into transient stability and voltage stability. Transient stability refers to the ability of the power system to maintain synchronous operation after suffering transient disturbances such as short circuits and faults; voltage stability refers to the ability of the power system to maintain voltage within the allowable range after suffering disturbances such as load changes and faults.

[0005] Traditional power system stability assessment methods are mainly based on physical models and numerical calculations, which require a lot of computing resources and expertise and are difficult to meet the requirements of real-time and accuracy. In recent years, deep learning technology has been increasingly used in the field of power system stability assessment, but most of the existing deep learning models predict a single type of stability, ignoring the coupling relationship between different types of stability, making it difficult to accurately assess the overall stability of the power system. Summary of the invention

[0006] The purpose of the present invention is to provide a method for collaborative prediction of power system transient and voltage stability based on spatiotemporal graph neural network, which can simultaneously predict transient stability and voltage stability and has the characteristics of real-time prediction and robustness to data noise.

[0007] To achieve the above object, the technical solution of the present invention is: a method for collaborative prediction of power system transient and voltage stability based on spatiotemporal graph neural network, comprising:

[0008] According to the topological structure and electrical parameters of the power system, the grid information adjacency matrix is ​​constructed;

[0009] Based on the grid information adjacency matrix, the topological and temporal characteristics of the power system are extracted from the PMU data;

[0010] The extracted topological features and time features are fused, and the prediction results of power system transient and voltage stability are output.

[0011] In one embodiment of the present invention, a grid information adjacency matrix B is constructed based on the topological structure and electrical parameters of the power system. ij , power grid information adjacency matrix B ij is the adjacency matrix containing active power flow information, which is specifically constructed as follows:.

[0012]

[0013] in, represents the set of transmission lines operating under normal conditions, represents the set of transmission lines damaged during the emergency event, (i, j) represents the transmission line of nodes i and j in the interconnected network, and S ij represents the maximum transmission capacity of the connection between nodes i and j, δ i and δ j Represent the voltage phase angles of nodes i and j respectively.

[0014] In one embodiment of the present invention, a graph convolutional network module is used to extract topological features of the power system from PMU data.

[0015] In one embodiment of the present invention, the specific implementation method of extracting the topological features of the power system from the PMU data using the graph convolutional network module is as follows:

[0016] PMU data is used to capture the dynamic behavior of the power system, including the voltage phase angle θ, voltage amplitude |U|, and generator speed ω of the node, which is expressed as a transient dynamic time series data X = {U θ ,|U|,θ,ω},U θ Represents the voltage vector of the node, which contains multiple PMU measurement values ​​of the power system nodes. The PMU data stream of each node contains time series data for a period of time after the fault, that is, The i-th data in the PMU data stream of each node is expressed as:

[0017]

[0018] Where i = 1, 2, ..., F, F is the number of features in the PMU data stream, N is the number of nodes in the power system, l is the length of the time series, represents the PMU measurement value of the j-th node coupled with the current node at the k-th moment, j = 1, 2, ..., N, k = 0, 2, ..., l-1;

[0019] The graph convolutional network module consists of two graph convolutional layers and one fully connected layer. In the graph convolutional network module, the feature vector of each node is updated by multiplying it with the adjacency matrix to integrate the information of neighboring nodes into each node. By learning the adjacency matrix and node features, the topological relationship between nodes and node attributes are extracted. The whole process is expressed as follows:

[0020]

[0021] in, is the adjacency matrix A and the identity matrix I N The sum of , means adding a self-join; yes is the degree matrix of the node; X is the node feature matrix; W and b are weights and biases respectively; σ is the ReLU activation function.

[0022] In one embodiment of the present invention, a temporal convolutional network module is used to extract topological features and temporal features of the power system from PMU data.

[0023] In one embodiment of the present invention, the specific implementation method of extracting the time characteristics of the power system from the PMU data using the time convolution network module is as follows:

[0024] The topological features extracted by the graph convolutional network module serve as the input of the temporal convolutional network module;

[0025] The temporal convolutional network module consists of multiple sequential residual blocks, each of which consists of a 1D fully convolutional network using causal convolution and dilated convolution techniques and a residual connection; specifically, the dilated convolution operation F(·) is applied to the jth element x in the one-dimensional sequence x. j By filter Defined as:

[0026]

[0027] Where k represents the filter size and d represents the dilation factor. A 1x1 convolutional layer is applied to solve the problem of inconsistent input and output dimensions. Its operation is defined as Conv1D(·). Therefore, the output of the residual block is defined as:

[0028] o = σ(Conv1D(x+LF(F(x))))

[0029] where x and o represent the input and output of the residual block, respectively, and LF(·) denotes the layer normalization technique.

[0030] In one embodiment of the present invention, a multi-layer perceptron layer is used to fuse the extracted topological features and time features, and output the prediction results of power system transient and voltage stability.

[0031] In one embodiment of the present invention, the multi-layer perceptron layer is used to fuse the extracted topological features and time features, and the specific implementation method of outputting the prediction results of the transient state and voltage stability of the power system is as follows:

[0032] The topological features extracted by the graph convolution module and the temporal features extracted by the temporal convolution module are spliced, and the spliced ​​features are input into the multi-layer perceptron layer for fusion and classification.

[0033] The multi-layer perceptron layers are represented as follows:

[0034] h (l) =σ(W (l) h (l-1) +b (l) )

[0035] y=W (L) h (L-1) +b (L)

[0036] Among them, h (l) is the output of layer l, and h (0) It is the concatenation of the topological features extracted by the graph convolution module and the temporal features extracted by the temporal convolution module; W (l) and b (l) are the weight and bias of the lth layer respectively; L is the total number of layers of the multilayer perceptron; σ is the ReLU activation function; y is the final output, which is the result of multi-task prediction of transient and voltage stability at the same time. It uses one-hot encoding to represent the four different stability states that the power system may be in after experiencing a fault, which are specifically expressed as follows:

[0037] [1,0,0,0]: indicates that the system is in both transient stability and voltage stability states;

[0038] [0,1,0,0]: indicates that the system is only in a transient stable state;

[0039] [0,0,1,0]: indicates that the system is only in voltage stabilization state;

[0040] [0,0,0,1]: Indicates that the system is in transient instability and voltage instability states at the same time.

[0041] In one embodiment of the present invention, the method uses a class-weighted cross entropy loss function to perform model training to handle class imbalance problems in different stability states.

[0042] The present invention also provides a computer-readable storage medium, on which computer program instructions that can be executed by a processor are stored. When the processor executes the computer program instructions, the method steps described above can be implemented.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] 1. Improved prediction accuracy: It can simultaneously predict the transient stability and voltage stability of the power system, which are often conducted separately in previous studies. By collaboratively predicting transient and voltage stability, the model can better capture the coupling relationship between different types of stability, thereby improving prediction accuracy.

[0045] 2. Real-time prediction: The model can use PMU data to make rapid predictions to meet the needs of real-time monitoring and evaluation of the power system.

[0046] 3. Enhanced robustness: The model is highly robust to noise in PMU data and can operate stably in practical applications.

[0047] 4. Integration of spatiotemporal characteristics: By integrating the topological structure and timing characteristics of the power system, the spatiotemporal embedding graph neural network can effectively improve the accuracy of power system stability prediction.

[0048] The present invention is applicable to the fields of power system stability assessment, fault diagnosis, control decision-making, etc., and can effectively improve the safe and stable operation level of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Attached Figure 1 It is the spatiotemporal embedded graph neural network joint prediction framework adopted by the method of the present invention.

[0050] Attached Figure 2 It is the stability that jointly predicts the accuracy and loss on the training and validation sets.

[0051] Attached Figure 3 It is a visualization of the high-dimensional features of the hidden layer.

[0052] Attached Figure 4 is the confusion matrix for the test set. DETAILED DESCRIPTION

[0053] The technical solution of the present invention is described in detail below in conjunction with the accompanying drawings.

[0054] The present invention provides a method for collaboratively predicting power system transient and voltage stability based on a spatiotemporal graph neural network, comprising:

[0055] According to the topological structure and electrical parameters of the power system, the grid information adjacency matrix is ​​constructed;

[0056] Based on the grid information adjacency matrix, the topological and temporal characteristics of the power system are extracted from the PMU data;

[0057] The extracted topological features and time features are fused, and the prediction results of power system transient and voltage stability are output.

[0058] The following is the specific implementation process of the present invention.

[0059] The present invention provides a method for collaborative prediction of power system transient and voltage stability based on spatiotemporal graph neural network, and proposes a spatiotemporal embedded graph neural network joint prediction framework, which is a deep learning model combining graph convolutional network and time series convolutional network, and is specifically used for power system stability assessment. It can simultaneously extract the topological features and time series features of the power system, and make joint predictions, thereby improving the prediction accuracy.

[0060] Figure 1 The structure of the spatiotemporal embedded graph neural network joint prediction framework is presented for predicting transient stability and voltage stability of power systems. The framework is mainly divided into three parts:

[0061] 1. Graph convolution module: This module is responsible for extracting the topological features of the power system. Each module contains two graph convolution layers and one fully connected layer. The graph convolution layer uses the adjacency matrix to extract the connection relationship between nodes from the PMU data to learn the topological structure of the system. The fully connected layer further extracts and integrates the output of the graph convolution layer.

[0062] 2. Temporal convolution module: This module is responsible for extracting the time series characteristics of the power system. This module consists of multiple residual blocks, each of which contains a full convolutional network and layer normalization technology.

[0063] 3. Multi-layer perceptron prediction layer: This layer is responsible for integrating the features extracted by the graph convolution module and the temporal convolution module and performing classification prediction. This layer contains multiple fully connected layers and outputs the prediction results through the softmax activation function, that is, the predicted probability of the system being in four stable states.

[0064] The framework's workflow is as follows:

[0065] 1) Input data: Input PMU data into the graph convolution module.

[0066] 2) Feature extraction: The graph convolution module extracts the topological features of the system. The temporal convolution module extracts the temporal features of the system.

[0067] 3) Feature integration: Integrate the features extracted by the graph convolution module and the temporal convolution module.

[0068] 4) Prediction: The multi-layer perceptron prediction layer predicts the probability of the system being in four stable states based on the integrated features.

[0069] In order to evaluate the performance of the proposed spatiotemporal embedding graph neural network model in power system transient stability and voltage stability prediction, simulations were carried out on the IEEE-118Bus System to further illustrate the model.

[0070] PSSE is used to simulate the transient process of the power system, taking into account N-1 fault conditions, including randomly changing the load level of some buses in the system to simulate the sudden increase or decrease of load; after the system fails, the fault clearing time is randomly set to simulate the fault handling process in the actual power system; three-phase grounding faults are set at the ends of 118 buses and 177 transmission lines to simulate the types of faults that may occur in the actual power system. The simulation time is set to 5 seconds. The simulation generates 59,000 samples and divides them into training, validation and test sets with proportions of 80%, 20% and 20% respectively, and the performance of the model is evaluated using indicators such as accuracy, false positive rate, false negative rate and F-score.

[0071] The accuracy and loss curves during training and validation are as follows Figure 2 As shown in Figure 2, the prediction accuracy finally reached 95.53%, indicating that the training has converged. The confusion matrix of the test set is shown in Figure 2. Figure 3 shown.

[0072] Figure 4 The visualization results of the high-dimensional features extracted by the model from the hidden layer are shown. It can be seen from the figure that the model prediction results are basically consistent with the real data, and the number of error points is within an acceptable range. This further verifies that the model can effectively learn data features and make accurate predictions.

[0073] The simulation results show that the spatiotemporal embedded graph neural network model has good performance in predicting power system transient stability and voltage stability. The model can effectively utilize spatial features and time series data, and can capture the correlation information between the two stability states, thereby improving the prediction accuracy.

[0074] The present invention also provides a computer-readable storage medium, on which computer program instructions that can be executed by a processor are stored. When the processor executes the computer program instructions, the method steps described above can be implemented.

[0075] The above are preferred embodiments of the present invention. Any changes made according to the technical solution of the present invention, as long as the resulting functions do not exceed the scope of the technical solution of the present invention, belong to the protection scope of the present invention.

Claims

1. A method for collaborative prediction of power system transient and voltage stability based on spatiotemporal graph neural network, characterized in that: include: According to the topological structure and electrical parameters of the power system, the grid information adjacency matrix is ​​constructed; Based on the grid information adjacency matrix, the topological and temporal characteristics of the power system are extracted from the PMU data; The extracted topological features and time features are fused, and the prediction results of power system transient and voltage stability are output.

2. The method for collaborative prediction of power system transient and voltage stability based on spatiotemporal graph neural network according to claim 1 is characterized in that: According to the topological structure and electrical parameters of the power system, the grid information adjacency matrix B is constructed. ij , power grid information adjacency matrix B ij is the adjacency matrix containing active power flow information, which is specifically constructed as follows:. in, represents the set of transmission lines operating under normal conditions, represents the set of transmission lines damaged during the emergency event, (i, j) represents the transmission line of nodes i and j in the interconnected network, and S ij represents the maximum transmission capacity of the connection between nodes i and j, δ i and δ j Represent the voltage phase angles of nodes i and j respectively.

3. The method for collaborative prediction of power system transient and voltage stability based on spatiotemporal graph neural network according to claim 1 is characterized in that: The graph convolutional network module is used to extract the topological features of the power system from PMU data.

4. The method for collaborative prediction of power system transient and voltage stability based on spatiotemporal graph neural network according to claim 3 is characterized in that: The specific implementation method of extracting the topological features of the power system from PMU data using the graph convolutional network module is as follows: PMU data is used to capture the dynamic behavior of the power system, including the voltage phase angle θ, voltage amplitude |U|, and generator speed ω of the node, which is expressed as a transient dynamic time series data X = {U θ ,|U|,θ,ω},U θ Represents the voltage vector of the node, which contains multiple PMU measurement values ​​of the power system nodes. The PMU data stream of each node contains time series data for a period of time after the fault, that is, The i-th data in the PMU data stream of each node is expressed as: Where i = 1, 2, ..., F, F is the number of features in the PMU data stream, N is the number of nodes in the power system, l is the length of the time series, represents the PMU measurement value of the j-th node coupled with the current node at the k-th moment, j = 1, 2, ..., N, k = 0, 2, ..., l-1; The graph convolutional network module consists of two graph convolutional layers and one fully connected layer. In the graph convolutional network module, the feature vector of each node is updated by multiplying it with the adjacency matrix to integrate the information of neighboring nodes into each node. By learning the adjacency matrix and node features, the topological relationship between nodes and node attributes are extracted. The whole process is expressed as follows: in, is the adjacency matrix A and the identity matrix I N The sum of , means adding a self-join; yes is the degree matrix of the node; X is the node feature matrix; W and b are weights and biases respectively; σ is the ReLU activation function.

5. The method for collaborative prediction of power system transient and voltage stability based on spatiotemporal graph neural network according to claim 4 is characterized in that: The temporal convolutional network module is used to extract the topological and temporal characteristics of the power system from PMU data.

6. A method for collaborative prediction of power system transient and voltage stability based on spatiotemporal graph neural network according to claim 5, characterized in that: The specific implementation method of extracting the time characteristics of the power system from PMU data using the time convolutional network module is as follows: The topological features extracted by the graph convolutional network module serve as the input of the temporal convolutional network module; The temporal convolutional network module consists of multiple sequential residual blocks, each of which consists of a 1D fully convolutional network using causal convolution and dilated convolution techniques and a residual connection; specifically, the dilated convolution operation F(·) is applied to the jth element x in the one-dimensional sequence x. j By filter f: Defined as: Where k represents the filter size and d represents the dilation factor. A 1x1 convolutional layer is applied to solve the problem of inconsistent input and output dimensions. Its operation is defined as Conv1D(·). Therefore, the output of the residual block is defined as: o = σ(Conv1D(x+LF(F(x)))) where x and o represent the input and output of the residual block, respectively, and LF(·) denotes the layer normalization technique.

7. The method for collaborative prediction of power system transient and voltage stability based on spatiotemporal graph neural network according to claim 6 is characterized in that: The extracted topological features and time features are fused by using a multi-layer perceptron layer, and the prediction results of power system transient and voltage stability are output.

8. The method for collaborative prediction of power system transient and voltage stability based on spatiotemporal graph neural network according to claim 7 is characterized in that: The specific implementation method of using the multi-layer perceptron layer to fuse the extracted topological features and time features and output the prediction results of power system transient and voltage stability is as follows: The topological features extracted by the graph convolution module and the temporal features extracted by the temporal convolution module are spliced, and the spliced ​​features are input into the multi-layer perceptron layer for fusion and classification. The multi-layer perceptron layers are represented as follows: h (l) =σ(W (l) h (l-1) +b (l) ) y=W (L) h (L-1) +b (L) Among them, h (l) is the output of layer l, and h (0) It is the concatenation of the topological features extracted by the graph convolution module and the temporal features extracted by the temporal convolution module; W (l) and b (l) are the weight and bias of the lth layer respectively; L is the total number of layers of the multilayer perceptron; σ is the ReLU activation function; y is the final output, which is the result of multi-task prediction of transient and voltage stability at the same time. It uses one-hot encoding to represent the four different stability states that the power system may be in after experiencing a fault, which are specifically expressed as follows: [1,0,0,0]: indicates that the system is in both transient stability and voltage stability states; [0,1,0,0]: indicates that the system is only in a transient stable state; [0,0,1,0]: indicates that the system is only in voltage stabilization state; [0,0,0,1]: Indicates that the system is in transient instability and voltage instability states at the same time.

9. The method for collaborative prediction of power system transient and voltage stability based on spatiotemporal graph neural network according to claim 1, characterized in that: The method uses a class-weighted cross entropy loss function for model training to handle class imbalance problems with different stability states.

10. A computer-readable storage medium having stored thereon computer program instructions that can be executed by a processor, and when the processor executes the computer program instructions, the method steps according to any one of claims 1 to 9 can be implemented.

Images