Power distribution network topology identification method, device and equipment based on ARAM graph convolution and medium

Abstract

The invention discloses a power distribution network topology identification method based on ARAM graph convolution, and the method comprises the steps: 1, taking lines in a power distribution network as nodes, taking the lines with power connection between the lines as edges, and abstracting the power distribution network as a line graph structure; 2, generating various types of topological structures according to the basic topological structure of the power distribution network system, and establishing a data set by taking electrical quantities such as voltage difference of a power distribution network line as node characteristics of a line graph and taking an on-off state vector of the line as a label; 3, constructing an ARMA-based graph convolutional network model for power distribution network topology identification; 4, training an ARMA-based graph convolutional network model and storing the trained model; and 5, collecting real-time measurement data of the power distribution network, and inputting the normalized data into the trained optimal ARMA graph convolutional network model to carry out real-time identification on the topology of the power distribution network. According to the method, accurate and real-time topology identification can be realized in a large-scale power distribution network with uncertain scenes such as distributed renewable power supply fluctuation.

Description

Technical Field

[0001] The present invention relates to the field of distribution network topology identification, and in particular to a distribution network topology identification method, device, equipment and medium based on ARAM graph convolution. Background Art

[0002] With the rapid development of modern power systems and renewable energy technologies, a large number of photovoltaic, wind power generation and energy storage equipment have been gradually connected to the distribution network, resulting in a significant increase in the complexity and uncertainty of the dispatching and operation of the distribution network. The volatility and randomness of renewable energy affect the stable operation of the distribution network, requiring the automation system to frequently adjust the topology of the distribution network to improve power balance, reduce power line losses and isolate faults. These factors make the topology of the distribution network change frequently and complex, which is not only closely related to advanced analysis functions such as power flow calculation, state estimation, and fault location, but also provides a solution for the optimized operation of distribution networks with a high proportion of renewable energy. However, due to the wide geographical distribution of the distribution network, which consists of a large number of lines, cables and switches, and the lines and cables may be re-laid due to local development, the traditional topology information may not be updated in time, resulting in the switch status no longer reflecting the real topology. Therefore, accurately identifying the actual topology of the distribution network has become a challenging key task in distribution network management.

[0003] At present, the distribution network topology identification methods are mainly divided into two categories: model-driven methods and data-driven methods. Model-driven methods rely on the physical model and parameter information of the distribution network, and identify the network topology by using distribution system state estimation, bad data detection, mathematical optimization, and graph theory analysis. This type of method usually derives the current network topology by measuring physical quantities such as voltage phasors and power flows, combined with the network's admittance matrix or other physical models. Model-driven methods can provide high-precision topology identification under the conditions of high-quality data and accurate models. However, they have high requirements for data quality and model accuracy, and have high computational complexity when dealing with complex nonlinear problems, especially when facing active distribution networks with a large number of distributed energy sources. These methods face challenges. Data-driven methods do not rely on the specific physical model of the network, but identify the topology by analyzing a large amount of historical data and using techniques such as machine learning, deep learning, unsupervised learning, and time series analysis. This type of method achieves topology identification by establishing a mapping relationship between inputs (such as voltage, current, power, etc.) and outputs (such as topology). Data-driven methods have strong adaptability and flexibility in dealing with topology recognition in dynamic and complex environments.

[0004] Graph neural networks are a type of deep learning model developed in recent years that can operate directly on graph structured data. They have shown excellent performance in processing complex topological structures and relationships between nodes. Although existing graph neural network models have shown good performance in dealing with the problem of identifying the topological structure of distribution networks, they still have some limitations when facing complex scenarios with a large number of distributed energy sources. The access of a large number of distributed renewable energy sources in the distribution network has brought volatility and uncertainty, which have a significant impact on the stability of the power grid and power flow. The existing graph neural network model has weak noise resistance when dealing with such uncertainty, resulting in low recognition accuracy of the model. In addition, in order to capture the dependency relationship between long-distance nodes in the distribution network, traditional graph convolutional networks usually use deeper network layers. As the number of convolutional layers increases, the graph convolutional network model is prone to over-smoothing, resulting in the homogenization of different node features and the difficulty in distinguishing the node status, which makes it impossible to accurately identify the topological structure of the distribution network. Summary of the invention

[0005] The purpose of the present invention is to solve the deficiencies in the prior art and propose a distribution network topology identification method based on ARAM graph convolution. This method uses the ARAM graph convolution layer to capture the complex relationships between distribution network nodes, and can achieve accurate and real-time topology identification in large-scale distribution networks with uncertain scenarios such as fluctuations in distributed renewable power sources.

[0006] The technical solution adopted by the present invention is: a distribution network topology identification method based on ARAM graph convolution, comprising the following steps: Import distribution network data and abstract the distribution network into a line graph structure; Generate various types of topological structures according to the line graph structure, establish node measurement data sample data sets according to the various types of topological structures and divide them into training sets, validation sets and test sets; Construct an ARMA-based graph convolutional network model for distribution network topology identification; Use the training set to train the ARMA-based graph convolutional network model, use the validation set to select the optimal model, and use the test set to evaluate the prediction accuracy of the model; The trained optimal ARMA graph convolutional network model is used to perform real-time identification of distribution network topology.

[0007] Preferably, the step of abstracting the distribution network into a line graph structure is as follows: The lines in the distribution network are taken as nodes, and the lines with power connections between the lines are taken as edges, so as to obtain the line graph structure of the distribution network, which is recorded as ,in, is the set of nodes in G, represents the i-th line unit, N represents the number of nodes, is the set of edges in G, Represents the i-th line unit and the jth line unit There are edge connections between them; like and If there is an edge connection, let the element in the i-th row and j-th column be ; Otherwise, let the element in row i and column j be , thus obtaining G Order adjacency matrix ; The electrical quantities are used as node features of the line graph structure, and the electrical quantities include the voltage difference between the two ends of the line, active power and reactive power.

[0008] The method for establishing a node measurement data sample data set is as follows: Based on the line diagram structure of the distribution network, various topological structures of the distribution network are generated by adjusting the switch status of the branches and the on-off of the tie lines; Under each topology of the distribution network, different load distribution and output of distributed power sources are configured to simulate the state changes of the system under various operating conditions. The power flow analysis tool is used to simulate and calculate each topology to obtain the measurement data of the nodes. Under each topological structure of the distribution network, node measurement data of several time sections are taken to construct a node measurement data set, and the node measurement data set is divided into a training set, a validation set, and a test set in a ratio of 8:1:1.

[0009] The steps of constructing the ARMA-based graph convolutional network model for distribution network topology identification are as follows: The ARMA filter-based graph convolutional network model for distribution network topology identification includes a preprocessing layer, L ARMA graph convolutional layers, and a post-processing layer; The preprocessing layer is used to perform preliminary processing on the input node features and map the original input features to a high-dimensional feature space. The method is: the input node features are transformed through the fully connected layer, and then a new node feature representation is generated through the activation function. The calculation method is as follows: ; Among them, V represents the node features of the original input, W is the learnable weight matrix, b is the bias term, is the activation function, It is a new node feature representation; The ARMA graph convolution layer consists of K stacks, each of which performs T propagation steps, calculated as: ; in, is the input node feature, is the node feature of the kth ARMA convolution matrix layer at step t, is a nonlinear activation function, , , , is a trainable weight matrix, is the symmetric normalized adjacency matrix, is the degree matrix of the adjacency matrix A; The final output Y of the ARMA graph convolutional layer is the average of all stack outputs: ; The post-processing layer first maps the node feature Y output by the ARMA graph convolution layer into a vector Z with a dimension of n*2 through a fully connected layer. The calculation method is as follows: ; Next, use the softmax function to convert the score output by the fully connected layer into the probability of the line being on or off: ; in, , They represent the probabilities of node i belonging to a connected line and a disconnected line respectively, and the category with a larger probability is taken as the category to which node i belongs.

[0010] The steps of using the training set to train the ARMA-based graph convolutional network model, using the validation set to select the optimal model, and using the test set to evaluate the prediction accuracy of the model are as follows: The loss value is calculated by the cross entropy loss function: ; in, is the category of line i, with a value of 1 representing a connected line and a value of 0 representing a disconnected line; Update the neural network parameters through the Adam optimizer: ; In the formula, is the updated network parameter after the tth round of iteration, is the learning rate, is the bias-corrected first-order moment estimate, is the bias-corrected second-order moment estimate, is the smoothing term, is the exponential decay rate of the first-order moment estimate, is the first-order moment estimate in momentum form, is the exponential decay rate of the second-order moment estimate, is the second-order moment estimate in momentum form, is the loss function The partial derivative of Set the number of iterations n, select the appropriate batch size bs and learning rate lr for iterative training: input node feature data, pass through the preprocessing layer, L ARMA graph convolution layers, and post-processing layers to obtain the prediction results of line connection and disconnection, calculate the loss between the prediction results and the true labels, calculate the gradient of network parameters according to the cross entropy loss function, and use the Adam optimizer to adjust the weights and biases; During the training process, the model is regularly evaluated using the validation set to select the network model that performs best on the validation set until the predetermined number of iterations is reached; Save the trained optimal model and use the test set to test the optimal model.

[0011] The steps of using the trained optimal ARMA graph convolutional network model to perform real-time identification of the distribution network topology are as follows: Collect real-time measurement data of the distribution network, including voltage, active power, and reactive power; Normalize the real-time measurement data: ; in is the original data, and are the minimum and maximum values ​​of the data respectively; The normalized data is input into the trained optimal ARMA graph convolutional network model to perform real-time identification of the distribution network topology.

[0012] A distribution network topology identification device based on ARAM graph convolution, comprising: Input module, used to import distribution network data; A data processing module is used to abstract the imported distribution network data into a line graph structure; A node measurement data sample data set construction module is used to construct a node measurement data sample data set; An ARMA-based graph convolutional network model building module, which is used to build an ARMA-based graph convolutional network model for distribution network topology identification; A training module, used to train an ARMA-based graph convolutional network model for distribution network topology identification through node measurement data sample data; The distribution network topology real-time identification module identifies the real-time data on both sides of the distribution network through the trained ARMA-based graph convolutional network model for distribution network topology identification.

[0013] An electronic device comprises: at least one processor; and a memory connected to the at least one processor in communication; wherein: The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the distribution network topology identification method based on ARAM graph convolution.

[0014] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the distribution network topology identification method based on ARAM graph convolution is implemented.

[0015] Aiming at the problem of topology identification, the present invention proposes a distribution network topology identification method based on autoregressive moving average model (ARMA) convolutional network, which can effectively meet the requirements of algorithm speed and accuracy, and specifically includes the following effects: The present invention recursively aggregates the information of multi-hop neighbor nodes through the autoregressive (AR) operation of the ARMA layer, which enhances the model's ability to capture the dependency relationship between distant nodes and ensures accurate perception and recognition of topological structures in large-scale networks. The present invention smoothes node features through the sliding average (MA) method of the ARMA layer, suppresses high-frequency noise, and enhances the robustness of the model in uncertain scenarios such as fluctuations in renewable energy. The present invention adopts ARMA graph convolutional network to effectively prevent node features from becoming homogenized in deep networks. By designing a multi-level recursive propagation mechanism, the feature differences between nodes can be maintained in deeper networks, thereby improving the recognition accuracy under complex topological structures. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 The topology identification process of the present invention; Figure 2 A schematic diagram of an IEEE33 node distribution network connected to a distributed power source in an embodiment of the present invention; Figure 3 It is the process of abstracting the distribution network into a line graph structure model in the present invention; Figure 4 It is the ARMA graph convolutional network model in the present invention; Figure 5 It is the ARMA graph convolution layer in the present invention; Figure 6 It is a schematic diagram of the structure of the electronic device of the present invention; Figure 7 It is a schematic diagram of the computer-readable storage medium structure of the present invention. DETAILED DESCRIPTION

[0017] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art may think of other obvious variations.

[0018] See also Figures 1 to 7 The present invention provides a technical solution: a distribution network topology identification method based on ARAM convolution, which is performed in the following steps: Step 1: Import distribution network data and abstract the distribution network into a line diagram structure model; Step 1.1: The schematic diagram of the IEEE33 node distribution network connected to the distributed power source in this embodiment is as follows: Figure 2 As shown in the figure, the process of abstracting the distribution network into a line diagram structure model is as follows Figure 3 As shown in the figure, the lines in the distribution network are taken as nodes, and the lines with power connections between the lines are taken as edges, so as to obtain the line graph structure of the distribution network, which is recorded as ,in, is the set of nodes in G, represents the i-th line unit, N represents the number of nodes, is the set of edges in G, Represents the i-th line unit and the jth line unit There are edge connections between them; in the specific example, N= 33; Step 1.2: If and If there is an edge connection, let the element in the i-th row and j-th column be ; Otherwise, let the element in row i and column j be , thus obtaining G Order adjacency matrix ; Step 1.3: Use electrical quantities as node features of the network graph model, wherein the electrical quantities include voltage difference at both ends of the line, active power, and reactive power.

[0019] Step 2: Generate a data set from various types of topological structures based on the line diagram structure of the distribution network obtained in step 1 and divide it into a training set, a validation set, and a test set; Step 2.1 Obtain the initial structure based on the distribution network, and generate various topological structures by adjusting the switch status of the branch and the on-off of the tie line; in the specific example, Figure 2 Based on the IEEE33-node distribution network topology shown, the line connections are changed to generate 30 topological structures; Step 2.2 Under each topology, configure different load distributions and outputs of distributed power sources to simulate the state changes of the system under various operating conditions; use the power flow analysis tool to simulate and calculate each topology to obtain the voltage, current, power and other measurement data of the nodes; in the specific example, select nodes 7, 10, 14, and 33 as distributed power generation access, and use MATLAB software to simulate each topology to obtain 400 groups of sample data, a total of 12,000 groups, and each group of sample data is the voltage amplitude and injection power of 33 nodes; Step 2.3 Take several time-section node measurement data as samples in each topological structure of the distribution network, construct a node measurement data sample data set, and divide the node measurement data sample data set into a training set, a validation set, and a test set in a ratio of 8:1:1. Specifically, the method for constructing the node measurement data sample data set is as follows: take several time-section node measurement data in each topological structure of the distribution network, and divide the on-off state vector of the line into As a label, where The node corresponding to the line graph structure is a binary variable, 1 represents a connected line and 0 represents a disconnected line, and a sample data set of node measurement data is constructed.

[0020] Step 3: Construct an ARMA-based graph convolutional network model for distribution network topology identification; Step 3.1: Construct an ARMA-based graph convolutional network for distribution network topology identification. Figure 4 As shown, the ARMA graph convolution network includes a preprocessing layer, L ARMA graph convolution layers and a post-processing layer; in the specific example, L=3; The main purpose of the preprocessing layer in step 3.2 is to perform preliminary processing on the input node features and map the original input features to a higher-dimensional feature space to facilitate processing by the subsequent graph convolution layer. Specifically, the input node features are transformed through the fully connected layer, and then a new node feature representation is generated through the activation function. The calculation method is as follows: ; Where V represents the node features of the original input, W is the learnable weight matrix, and b is the bias term. is the activation function, It is a new node feature representation; Step 3.3: If Figure 5 As shown, the ARMA filter-based graph convolution layer consists of K stacks, each stack performs T propagation steps, and the calculation method is: ; in, is the input node feature, is the node feature of the kth ARMA convolution matrix layer at step t, is a nonlinear activation function, , , , is a trainable weight matrix, is the symmetric normalized adjacency matrix, is the degree matrix of the adjacency matrix A; in the specific example, K=5, T=8; The final output Y of the ARMA graph convolutional layer is the average of all stack outputs: ; Step 3.4 The post-processing layer first maps the node feature Y output by the ARMA graph convolution layer into a vector Z with a dimension of n*2 through a fully connected layer. The calculation method is as follows: ; Next, use the softmax function to convert the score output by the fully connected layer into the probability of the line being on or off: ; in, , They represent the probabilities of node i belonging to a connected line and a disconnected line respectively, and the category with a larger probability is taken as the category to which node i belongs.

[0021] Step 4: Train the ARMA-based graph convolutional network model, use the validation set to select the optimal model, and use the test set to evaluate the prediction accuracy of the model; Step 4.1 Calculate the loss value through the cross entropy loss function: ; in, is the category of line i, with a value of 1 representing a connected line and a value of 0 representing a disconnected line; Step 4.2 Update the neural network parameters through the Adam optimizer: ; In the formula, is the updated network parameter after the tth round of iteration, is the learning rate, is the bias-corrected first-order moment estimate, is the bias-corrected second-order moment estimate, is the smoothing term, is the exponential decay rate of the first-order moment estimate, is the first-order moment estimate in momentum form, is the exponential decay rate of the second-order moment estimate, is the second-order moment estimate in momentum form, is the loss function The partial derivative of ; In the specific example, ; Step 4.3: Set the number of iterations n, select the appropriate batch size bs and learning rate lr for iterative training: input node feature data, pass through the preprocessing layer, L ARMA graph convolution layers, and post-processing layers to obtain the prediction results of line connection and disconnection, calculate the loss between the prediction results and the true labels, calculate the gradient of network parameters according to the cross entropy loss function, and use the Adam optimizer to adjust the weights and biases; In the specific example, n=2000, bs=64, L=3; Step 4.4: Repeat step 4.3. During the training process, the model is evaluated regularly using the validation set to select the network model with the best performance on the validation set until the predetermined number of iterations is reached. Step 4.5: Save the trained optimal model and use the test set to test the optimal model.

[0022] Step 5: Real-time identification of distribution network topology.

[0023] Step 5.1: Collect real-time measurement data of the distribution network, including voltage, active power, and reactive power; Step 5.2: Normalize the real-time measurement data: ; in is the original data, and are the minimum and maximum values ​​of the data respectively; Step 5.3: Input the normalized data into the trained optimal ARMA graph convolutional network model to perform real-time identification of the distribution network topology.

[0024] The present invention adopts ARMA graph convolutional network to effectively prevent node features from becoming homogenized in deep networks. By designing a multi-level recursive propagation mechanism, the feature differences between nodes can be maintained in deeper networks, thereby improving the recognition accuracy under complex topological structures.

[0025] Furthermore, the present invention also proposes a distribution network topology identification device based on ARAM graph convolution, comprising: Input module, used to import distribution network data; A data processing module is used to abstract the imported distribution network data into a line graph structure; A node measurement data sample data set construction module is used to construct a node measurement data sample data set; An ARMA-based graph convolutional network model building module, which is used to build an ARMA-based graph convolutional network model for distribution network topology identification; A training module, used to train an ARMA-based graph convolutional network model for distribution network topology identification through node measurement data sample data; The distribution network topology real-time identification module identifies the real-time data on both sides of the distribution network through the trained ARMA-based graph convolutional network model for distribution network topology identification.

[0026] Furthermore, the method according to the embodiment of the present application can also be performed by Figure 6 The electronic device architecture shown in FIG. Figure 6 As shown, the electronic device 500 may include a bus 501, one or more CPUs 502, a read-only memory (ROM) 503, a random access memory (RAM) 504, a communication port 505 connected to a network, an input / output component 506, a hard disk 507, etc. The storage device in the electronic device 500, such as the ROM 503 or the hard disk 507, may store the distribution network topology identification method based on ARAM graph convolution provided in the present application. The electronic device 500 may also include a user interface 508. Of course, Figure 6 The architecture shown is only exemplary and can be omitted according to actual needs when implementing different devices. Figure 6 One or more components of an electronic device are shown.

[0027] Figure 7 Schematic diagram of a computer-readable storage medium structure provided by an embodiment of the present application. Figure 7 As shown, a computer-readable storage medium 600 according to an embodiment of the present application is shown. Computer-readable instructions are stored on the computer-readable storage medium 600. When the computer-readable instructions are executed by the processor, the distribution network topology identification method based on ARAM graph convolution according to the embodiment of the present application described with reference to the above figures can be executed. The storage medium 600 includes, but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory (cache), etc. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0028] In summary, the advantages of the present invention are that it can realize accurate and real-time topology identification in a large-scale distribution network with uncertain scenarios such as fluctuations in distributed renewable power sources.

[0029] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions only describe the principles of the present invention. The present invention may be subject to various changes and improvements without departing from the spirit and scope of the present invention. These changes and improvements fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the attached claims and their equivalents.

Claims

1. A distribution network topology identification method based on ARAM graph convolution, characterized in that: The steps include: Import distribution network data and abstract the distribution network into a line graph structure; Generate various types of topological structures according to the line graph structure, establish node measurement data sample data sets according to the various types of topological structures and divide them into training sets, validation sets and test sets; Construct an ARMA-based graph convolutional network model for distribution network topology identification; Use the training set to train the ARMA-based graph convolutional network model, use the validation set to select the optimal model, and use the test set to evaluate the prediction accuracy of the model; The trained optimal ARMA graph convolutional network model is used to perform real-time identification of distribution network topology.

2. The distribution network topology identification method based on ARAM graph convolution according to claim 1 is characterized in that: The steps of abstracting the distribution network into a line graph structure are as follows: The lines in the distribution network are taken as nodes, and the lines with power connections between the lines are taken as edges, so as to obtain the line graph structure of the distribution network, which is recorded as ,in, is the set of nodes in G, represents the i-th line unit, N represents the number of nodes, is the set of edges in G, Represents the i-th line unit and the jth line unit There are edge connections between them; like and If there is an edge connection, let the element in the i-th row and j-th column be ; Otherwise, let the element in row i and column j be , thus obtaining G Order adjacency matrix ; The electrical quantities are used as node features of the line graph structure, and the electrical quantities include the voltage difference between the two ends of the line, active power and reactive power.

3. The distribution network topology identification method based on ARAM graph convolution according to claim 2 is characterized in that: The method for establishing a node measurement data sample data set is as follows: Based on the line diagram structure of the distribution network, various topological structures of the distribution network are generated by adjusting the switch status of the branches and the on-off of the tie lines; Under each topology of the distribution network, different load distribution and output of distributed power sources are configured to simulate the state changes of the system under various operating conditions. The power flow analysis tool is used to simulate and calculate each topology to obtain the measurement data of the nodes. Under each topological structure of the distribution network, node measurement data of several time sections are taken to construct a node measurement data set, and the node measurement data set is divided into a training set, a validation set, and a test set in a ratio of 8:1:

1.

4. The distribution network topology identification method based on ARAM graph convolution according to claim 3 is characterized in that: The steps of constructing the ARMA-based graph convolutional network model for distribution network topology identification are as follows: The ARMA filter-based graph convolutional network model for distribution network topology identification includes a preprocessing layer, L ARMA graph convolutional layers, and a post-processing layer; The preprocessing layer is used to perform preliminary processing on the input node features and map the original input features to a high-dimensional feature space. The method is: the input node features are transformed through the fully connected layer, and then a new node feature representation is generated through the activation function. The calculation method is as follows: ; Among them, V represents the node features of the original input, W is the learnable weight matrix, b is the bias term, is the activation function, It is a new node feature representation; The ARMA graph convolution layer consists of K stacks, each of which performs T propagation steps, calculated as: ； in, is the input node feature, is the node feature of the kth ARMA convolution matrix layer at step t, is a nonlinear activation function, , , , is a trainable weight matrix, is the symmetric normalized adjacency matrix, is the degree matrix of the adjacency matrix A; The final output Y of the ARMA graph convolutional layer is the average of all stack outputs: ； The post-processing layer first maps the node feature Y output by the ARMA graph convolution layer into a vector Z with a dimension of n*2 through a fully connected layer. The calculation method is as follows: ； Next, use the softmax function to convert the score output by the fully connected layer into the probability of the line being on or off: ； in, , They represent the probabilities of node i belonging to a connected line and a disconnected line respectively, and the category with a larger probability is taken as the category to which node i belongs.

5. The distribution network topology identification method based on ARAM graph convolution according to claim 4 is characterized in that: The steps of using the training set to train the ARMA-based graph convolutional network model, using the validation set to select the optimal model, and using the test set to evaluate the prediction accuracy of the model are as follows: The loss value is calculated by the cross entropy loss function: ； in, is the category of line i, with a value of 1 representing a connected line and a value of 0 representing a disconnected line; Update the neural network parameters through the Adam optimizer: ； In the formula, is the updated network parameter after the tth round of iteration, is the learning rate, is the bias-corrected first-order moment estimate, is the bias-corrected second-order moment estimate, is the smoothing term, is the exponential decay rate of the first-order moment estimate, is the first-order moment estimate in momentum form, is the exponential decay rate of the second-order moment estimate, is the second-order moment estimate in momentum form, is the loss function The partial derivative of Set the number of iterations n, select the appropriate batch size bs and learning rate lr for iterative training: input node feature data, pass through the preprocessing layer, L ARMA graph convolution layers, and post-processing layers to obtain the prediction results of line connection and disconnection, calculate the loss between the prediction results and the true labels, calculate the gradient of network parameters according to the cross entropy loss function, and use the Adam optimizer to adjust the weights and biases; During the training process, the model is regularly evaluated using the validation set to select the network model that performs best on the validation set until the predetermined number of iterations is reached; Save the trained optimal model and use the test set to test the optimal model.

6. The distribution network topology identification method based on ARAM graph convolution according to claim 5 is characterized in that: The steps of using the trained optimal ARMA graph convolutional network model to perform real-time identification of the distribution network topology are as follows: Collect real-time measurement data of the distribution network, including voltage, active power, and reactive power; Normalize the real-time measurement data: ； in is the original data, and are the minimum and maximum values ​​of the data respectively; The normalized data is input into the trained optimal ARMA graph convolutional network model to perform real-time identification of the distribution network topology.

7. A distribution network topology identification device based on ARAM graph convolution, characterized in that: include: Input module, used to import distribution network data; A data processing module is used to abstract the imported distribution network data into a line graph structure; A node measurement data sample data set construction module is used to construct a node measurement data sample data set; An ARMA-based graph convolutional network model building module, which is used to build an ARMA-based graph convolutional network model for distribution network topology identification; A training module, used to train an ARMA-based graph convolutional network model for distribution network topology identification through node measurement data sample data; The distribution network topology real-time identification module identifies the real-time data on both sides of the distribution network through the trained ARMA-based graph convolutional network model for distribution network topology identification.

8. An electronic device, characterized in that: include: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the distribution network topology identification method based on ARAM graph convolution as described in any one of claims 1-6.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the distribution network topology identification method based on ARAM graph convolution described in any one of claims 1 to 6 is implemented.

Images