Distribution area three-phase impedance matrix identification method based on self-supervised graph attention network

Abstract

The invention provides a power distribution area three-phase impedance matrix identification method based on a self-supervised graph attention network, and relates to the technical field of power distribution network impedance matrix identification. According to the method, a node set including a transformer area distribution total secondary side and at least one user electric meter of a to-be-identified power distribution transformer area can be constructed, and a three-phase impedance matrix of the to-be-identified power distribution transformer area is identified based on a trained power distribution transformer area three-phase impedance matrix identification model by using feature vector data, collected in real time, of each node, so that the identification accuracy of the three-phase impedance matrix of the to-be-identified power distribution transformer area is improved. The problems that an existing method depends on known network topology, physical law fitting is insufficient and the like are solved, the method is suitable for a power distribution area with an unknown topological structure or dynamic change, and the requirements for refined operation and control of a power distribution network are met.

Description

Technical Field

[0001] This application relates to the technical field of distribution network impedance matrix identification, and particularly to a three-phase impedance matrix identification method for distribution transformer areas based on a self-supervised graph attention network. Background Art

[0002] In an intelligent distribution network, the three-phase impedance matrix of a distribution transformer area is a core parameter describing the electrical characteristics of the distribution transformer area, and its accurate identification is crucial for fault diagnosis, power flow calculation, and power quality analysis. Existing methods rely on known network topologies and component parameters (such as line lengths, unit resistances, reactances, etc.) to derive the impedance matrix through Kirchhoff's laws. Figure 1 Schematically shows the topological structure between multiple nodes in a distribution transformer area. Node 1 is a distribution transformer, and nodes 2 to 33 are user ends on the low-voltage line. Existing methods derive the impedance matrix through Kirchhoff's laws based on the topological structure of the nodes and the element parameters of the nodes.

[0003] However, in most distribution transformer areas, it is often difficult to obtain line parameters, and there may be situations where the topology is unknown or dynamically changing. Here, the distribution transformer area refers to the power supply area between the low-voltage side (usually 0.4 kV or 380 V) of the distribution transformer and the user electricity meter in the distribution network, which is the end link of the power network and directly supplies power to users.

[0004] Therefore, it is necessary to provide a new three-phase impedance matrix identification method for distribution transformer areas to solve the above technical problems. Summary of the Invention

[0005] In view of the above problems, this application is proposed to provide a three-phase impedance matrix identification method, device, and electronic device for distribution transformer areas based on a self-supervised graph attention network that overcome the above problems or at least partially solve the above problems. The technical solutions are as follows: In a first aspect, a three-phase impedance matrix identification method for a distribution transformer area based on a self-supervised graph attention network is provided. The method includes: Taking the secondary side of the total distribution transformer in the area and the user electricity meters as nodes, constructing a node set of the distribution transformer area to be identified, including the secondary side of the total distribution transformer in the area and at least one user electricity meter; Obtaining the three-phase voltage amplitudes, three-phase injected current amplitudes, three-phase injected active powers, and three-phase injected reactive powers of the secondary side of the total distribution transformer in the distribution transformer area to be identified collected in real time, and obtaining the three-phase voltage amplitudes, three-phase injected current amplitudes, three-phase injected active powers, and three-phase injected reactive powers of each user electricity meter in the distribution transformer area to be identified collected in real time; Constructing feature vectors of each node in the node set according to the three-phase voltage amplitudes, three-phase injected current amplitudes, three-phase injected active powers, and three-phase injected reactive powers of the secondary side of the total distribution transformer in the distribution transformer area to be identified and each user electricity meter; Input the feature vectors of each node into a pre-trained three-phase impedance matrix identification model for the distribution transformer area, output the three-phase impedance matrix of the distribution transformer area to be identified, and label it as , is a 3 n ×3 n complex matrix. In each element of the complex matrix, the real part is the resistance value and the imaginary part is the reactance value. n is the total number of nodes; among them, the three-phase impedance matrix identification model for the distribution transformer area is trained based on a self-supervised graph attention network and combines a fusion loss function of three-phase voltage consistency, three-phase injected active power and three-phase injected reactive power error constraints, and consistency regularization loss; the consistency regularization loss is used to constrain the model to maintain consistency before and after when the input is subject to noise disturbance.

[0006] In a possible implementation, the three-phase impedance matrix identification model for the distribution transformer area is trained through the following steps: Use the secondary side of the total distribution transformer and at least one user meter of the sample distribution transformer area as each sample node to construct a sample node set of the sample distribution transformer area; Select the historical three-phase voltage amplitudes, historical three-phase injected current amplitudes, historical three-phase injected active powers, and historical three-phase injected reactive powers of the secondary side of the total distribution transformer and each user meter of the sample distribution transformer area to construct the feature vectors of each sample node in the sample node set; Establish the model structure of the self-supervised graph attention network. Among them, the graph attention part is composed of graph attention layers. Multiple attention heads are set in each graph attention layer for parallel calculation, and deep feature extraction is realized by stacking multiple graph attention layers; Construct a fusion loss function that combines three-phase voltage consistency, three-phase injected active power and three-phase injected reactive power error constraints, and consistency regularization loss; Apply a variety of preset neural network parameter initialization strategies to perform differential initialization for different types of network layers. Input the feature vectors of each sample node in the sample node set into the model structure of the self-supervised graph attention network for training. Use the preset optimizer to minimize the fusion loss function on the training nodes to achieve gradient optimization of the model structure of the self-supervised graph attention network. At the same time, use the preset early stopping technique on the validation set to automatically trigger the training termination condition to obtain the three-phase impedance matrix identification model for the distribution transformer area based on the self-supervised graph attention network.

[0007] In a possible implementation, use the secondary side of the total distribution transformer and at least one user meter of the sample distribution transformer area as each sample node to construct a sample node set of the sample distribution transformer area, including: Taking the secondary side of the total distribution transformer in the sample distribution transformer area and at least one user electricity meter as independent sample nodes respectively, and excluding the switches on the line between the secondary side of the total distribution transformer and at least one user electricity meter, a sample node set of the sample distribution transformer area is constructed.

[0008] In a possible implementation, the adjacency matrix of the model structure of the self-supervised graph attention network established is fully connected. The model structure of the self-supervised graph attention network does not require the topological relationship identification step of the sample distribution transformer area, that is, it does not need to obtain the topological structure of the sample distribution transformer area, and defines the initial impedance parameter between any two nodes in the sample node set as the edge feature of the model structure of the self-supervised graph attention network.

[0009] In a possible implementation, a fusion loss function combining three-phase voltage consistency, three-phase injected active power and three-phase injected reactive power error constraints, and consistency regularization loss is constructed, including: The formula for constructing the fusion loss function combining three-phase voltage consistency, three-phase injected active power and three-phase injected reactive power error constraints, and consistency regularization loss is: (1) In formula (1), is the three-phase voltage consistency loss function, is the three-phase injected active power and three-phase injected reactive power error constraint function, is the consistency regularization loss function, are the preset weight coefficients respectively.

[0010] In a possible implementation, represents the constraint on the voltage amplitude error of each phase, and the total loss is the three-phase average: (2) In formula (2), n is the total number of nodes, p is the phase, A 、 B 、 C are the phase values, and A 、 B 、 C constitute the three phases, is the node i obtained by acquisition p phase voltage amplitude of, and are the real part and imaginary part of the predicted voltage of node i respectively, and the calculation is as follows: According to Kirchhoff's law, is the node i of pThe phase complex voltage satisfies Equation (3): (3) In Equation (3), is the phase impedance value between nodes i, j and pm ; is the complex injected current of phase j at node m ; is the complex injected power of phase j at node m and satisfies Equations (4) and (5): (4) (5) In Equation (4), is the phase complex voltage of phase j at node m ; represents the conjugate; In Equation (5), is the active power of phase j injected at the collected node m ; w is the imaginary unit; is the reactive power of phase j injected at the collected node m ; According to Equations (4) and (5), we get: (6) In Equation (6), is the amplitude of the phase voltage of phase j at the collected node m ; is the phase angle of phase j at node m ; Let the secondary side of the total distribution transformer in the substation area be the reference node, i = 1; The user's electricity meter is a non-reference node, i = 2, 3,... n ; Define p = A the phase voltage phase angle as 0, that is, in Equation (6) = 0, then: , then Equation (6) gives the following Equation (7): (7) Substitute Equation (7) into Equation (3), separate the real and imaginary parts of , and get Equation (8): (8) In Equation (8), is the node output by the model structure of the self-supervised graph attention network i, j between pm phase resistance values, is the node output by the model structure of the self-supervised graph attention network i, j between pm phase reactance values; Equation (9) is obtained from Equation (8): (9) Equations (10) and (11) are obtained from Equation (9): (10) (11).

[0011] In a possible implementation, constrain the three-phase injected active power and the three-phase injected reactive power error, and the formula is as follows (12): (12) In Equation (12), is the predicted i of the node p phase injected active power, is the collected i of the node p phase injected active power, is the predicted i of the node p phase injected reactive power, is the collected i of the node p phase injected reactive power; and are obtained through the preset known three-phase power flow calculation formula.

[0012] In a possible implementation, constrain the model to maintain consistency before and after the input is disturbed by noise, and enhance the robustness of the model to noise. The formulas are as follows (13) and (14): (13) (14) In Equation (13), represents the model prediction result, represents the model prediction result after perturbation, represents the node feature perturbation, x represents the feature vector of the node; In Equation (14), It conforms to a Gaussian distribution with a mean of 0.

[0013] In a second aspect, a device for identifying the three-phase impedance matrix of a distribution transformer area based on a self-supervised graph attention network is provided. The device includes: A first construction unit for constructing a node set of the distribution transformer area to be identified, including the secondary side of the total distribution transformer in the area and at least one user meter, with the secondary side of the total distribution transformer in the area and the user meters as nodes. An acquisition unit for acquiring the three-phase voltage amplitudes, three-phase injected current amplitudes, three-phase injected active powers, and three-phase injected reactive powers of the secondary side of the total distribution transformer in the distribution transformer area to be identified collected in real time, and acquiring the three-phase voltage amplitudes, three-phase injected current amplitudes, three-phase injected active powers, and three-phase injected reactive powers of each user meter in the distribution transformer area to be identified collected in real time. A second construction unit for constructing the feature vectors of each node in the node set according to the three-phase voltage amplitudes, three-phase injected current amplitudes, three-phase injected active powers, and three-phase injected reactive powers of the secondary side of the total distribution transformer in the distribution transformer area to be identified and each user meter. An identification unit for inputting the feature vectors of each node into a pre-trained three-phase impedance matrix identification model for the distribution transformer area, outputting the three-phase impedance matrix of the distribution transformer area to be identified, and marking it as , is a 3 n ×3 n complex matrix. In each element of the complex matrix, the real part is the resistance value and the imaginary part is the reactance value. n is the total number of nodes; among them, the three-phase impedance matrix identification model for the distribution transformer area is based on a self-supervised graph attention network and is trained with a fusion loss function that combines three-phase voltage consistency, three-phase injected active power and three-phase injected reactive power error constraints, and consistency regularization loss; the consistency regularization loss is used to constrain the model to maintain consistency before and after when there is noise disturbance in the input.

[0014] In a third aspect, an electronic device is provided. The electronic device includes a processor and a memory. Among them, a computer program is stored in the memory, and the processor is configured to run the computer program to execute the method for identifying the three-phase impedance matrix of the distribution transformer area based on the self-supervised graph attention network described in any one of the above.

[0015] With the above technical solutions, the method and device for identifying the three-phase impedance matrix of a distribution transformer area based on a self-supervised graph attention network and an electronic device provided by the embodiments of the present application can construct a node set of the distribution transformer area to be identified, including the secondary side of the total distribution transformer of the area and at least one user electricity meter. By using the feature vector data of each node collected in real time and based on the trained identification model of the three-phase impedance matrix of the distribution transformer area, the three-phase impedance matrix of the distribution transformer area to be identified is identified, avoiding problems such as the existing methods relying on known network topologies and insufficient physical law fitting, being applicable to distribution transformer areas with unknown or dynamically changing topologies, and meeting the requirements of refined operation and control of the distribution network. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] To more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for the description of the embodiments of the present application will be briefly introduced below.

[0017] Figure 1 Schematically shows the topological structure among multiple nodes of a distribution transformer area; Figure 2 Shows the flowchart of the method for identifying the three-phase impedance matrix of a distribution transformer area based on a self-supervised graph attention network provided by the embodiments of the present application; Figure 3 Shows an example of the change process of the feature vector provided by the embodiments of the present application; Figure 4 Shows the loss function of the training set of the model provided by the embodiments of the present application; Figure 5 Shows the structural diagram of the device for identifying the three-phase impedance matrix of a distribution transformer area based on a self-supervised graph attention network provided by the embodiments of the present application; Figure 6 Shows the structural diagram of the device for identifying the three-phase impedance matrix of a distribution transformer area based on a self-supervised graph attention network provided by another embodiment of the present application; Figure 7 Shows the structural diagram of an electronic device provided by the embodiments of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] The exemplary embodiments of the present application will be described in more detail below with reference to the drawings. Although the exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.

[0019] It should be noted that the terms "first", "second", etc. in the description, claims, and the above-mentioned drawings of this application are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that such use can be interchanged under appropriate circumstances so that the embodiments of this application described herein can be implemented in an order other than those illustrated or described herein. In addition, the term "including" and its variants are to be interpreted as open-ended terms meaning "including but not limited to".

[0020] To solve the above technical problems, an embodiment of this application provides a method for identifying the three-phase impedance matrix of a distribution transformer area based on a self-supervised graph attention network, as Figure 2 shown. The method for identifying the three-phase impedance matrix of a distribution transformer area based on a self-supervised graph attention network may include the following steps S201 to S204: Step S201: Using the secondary side of the distribution transformer at the substation and the user meters as nodes, construct a node set of the distribution transformer area to be identified, including the secondary side of the distribution transformer at the substation and at least one user meter.

[0021] In this step, the distribution transformer area refers to the area range powered by a distribution transformer, and the secondary side of the distribution transformer at the substation refers to the low-voltage output terminal of the distribution transformer. The distribution transformer area to be identified here refers to the distribution transformer area for which the three-phase impedance matrix is to be identified.

[0022] Here, using the secondary side of the distribution transformer at the substation and the user meters as independent nodes, without including the switches on the lines between the secondary side of the distribution transformer at the substation and the user meters, construct a node set of the distribution transformer area to be identified, which does not require obtaining the topological structure between the secondary side of the distribution transformer at the substation and at least one user meter, and is applicable to the distribution transformer area to be identified with an unknown or dynamically changing topological structure.

[0023] Step S202: Obtain the three-phase voltage amplitudes, three-phase injected current amplitudes, three-phase injected active powers, and three-phase injected reactive powers of the secondary side of the distribution transformer at the substation of the distribution transformer area to be identified collected in real time, and obtain the three-phase voltage amplitudes, three-phase injected current amplitudes, three-phase injected active powers, and three-phase injected reactive powers of each user meter of the distribution transformer area to be identified collected in real time.

[0024] In this step, the three phases are A phase, B phase, C phase, which are composed of three sinusoidal alternating currents with the same frequency and a phase difference of 120°.

[0025] Step S203: According to the three-phase voltage amplitudes, three-phase injected current amplitudes, three-phase injected active powers, and three-phase injected reactive powers of the secondary side of the distribution transformer at the substation and each user meter of the distribution transformer area to be identified, construct the feature vectors of each node in the node set.

[0026] In this step, the feature vectors of each node x are expressed as: , where , , ; is the three-phase (i.e., i phase, A phase, B phase, C phase) voltage amplitude of the node collected; i is the three-phase injected current amplitude of the node collected; i is the three-phase injected active power of the node collected; i is the three-phase injected reactive power of the node

[0027] Step S204: Input the feature vectors of each node into a pre-trained three-phase impedance matrix identification model for a distribution transformer area, output the three-phase impedance matrix of the distribution transformer area to be identified, and mark it as , is a 3 n ×3 n complex matrix. In each element of the complex matrix, the real part is the resistance value and the imaginary part is the reactance value. n is the total number of nodes. Among them, the three-phase impedance matrix identification model for the distribution transformer area is trained based on a self-supervised graph attention network and a fusion loss function that combines three-phase voltage consistency, three-phase injected active power and three-phase injected reactive power error constraints, and consistency regularization loss. The consistency regularization loss is used to constrain the model to maintain consistency before and after when noise perturbation occurs in the input.

[0028] In this embodiment, a node set including the secondary side of the total distribution transformer and at least one user meter of the distribution transformer area to be identified can be constructed. Using the feature vector data of each node collected in real time, based on the trained three-phase impedance matrix identification model for the distribution transformer area, the three-phase impedance matrix of the distribution transformer area to be identified is identified, avoiding problems such as the existing method relying on known network topologies and insufficient physical law fitting, being applicable to distribution transformer areas with unknown or dynamically changing topologies, and meeting the requirements of refined operation and control of the distribution network.

[0029] A possible implementation manner is provided in the embodiment of the present application. The three-phase impedance matrix identification model for the distribution transformer area mentioned in step S204 above can be trained through the following steps a1 to a5: Step a1: Use the secondary side of the total distribution transformer and at least one user meter of the sample distribution transformer area as each sample node to construct a sample node set of the sample distribution transformer area.​

[0030] In this step, the secondary side of the total distribution transformer in the sample distribution transformer area and at least one user electricity meter can be used as independent sample nodes, excluding the switches on the line between the secondary side of the total distribution transformer in the area and at least one user electricity meter, to construct a sample node set of the sample distribution transformer area, without obtaining the topological structure between the secondary side of the total distribution transformer in the area and at least one user electricity meter. Here, the sample distribution transformer area refers to the distribution transformer area used as a sample.

[0031] Step a2: Select the historical three-phase voltage amplitudes, historical three-phase injected current amplitudes, historical three-phase injected active powers, and historical three-phase injected reactive powers of the secondary side of the total distribution transformer in the sample distribution transformer area and each user electricity meter, and construct the feature vectors of each sample node in the sample node set.

[0032] Step a3: Establish the model structure of the self-supervised graph attention network. Among them, the graph attention part is composed of graph attention layers. Multiple attention heads are set in each graph attention layer for parallel calculation, and deep feature extraction is realized by stacking multiple graph attention layers.

[0033] In this step, the adjacency matrix of the model structure of the established self-supervised graph attention network is fully connected. The model structure of the self-supervised graph attention network does not require the topological relationship identification step of the sample distribution transformer area, that is, it does not need to obtain the topological structure of the sample distribution transformer area, and the initial impedance parameter between any two nodes in the sample node set is defined as the edge feature of the model structure of the self-supervised graph attention network.

[0034] The model stacks multiple attention layers, and at the same time retains the original feature information through residual connections to prevent gradient disappearance, and extracts the local features and global correlation relationships of the node neighborhood layer by layer; the bottom layer network focuses on the feature extraction of the original measurement data, and the high layer network realizes the abstract representation of the cross-node electrical coupling relationship, forming a hierarchical feature expression system.

[0035] The evolution mechanism from node features to edge features samples the adjacent edges by obtaining the source node and the target node according to the index, splices the adjacent node features and inputs them into a multi-layer perceptron, and uses the softplus activation function (a smooth, non-linear activation function) to constrain the non-negativity of the real part of the impedance, and the imaginary part is allowed to be freely adjusted.

[0036] In the impedance matrix construction process, the edge features are filled into a complex matrix with a preset dimension in sequence according to the source node and the target node, and the reciprocity theorem is forced to be satisfied through matrix symmetrization processing. The preset dimension here can be, for example, 99×99 or 90×90, etc., and this embodiment does not limit this.

[0037] Step a4: Construct a fusion loss function that combines the three-phase voltage consistency, the constraints of the three-phase injected active power and the three-phase injected reactive power errors, and the consistency regularization loss.

[0038] Step a5: Apply a variety of preset neural network parameter initialization strategies to perform differential initialization for different types of network layers. Input the feature vectors of each sample node in the sample node set into the model structure of the self-supervised graph attention network for training. Use the preset optimizer to minimize the fusion loss function on the training nodes to achieve gradient optimization of the model structure of the self-supervised graph attention network. At the same time, use the preset early stopping technique on the validation set to automatically trigger the training termination condition, and obtain a three-phase impedance matrix identification model for the distribution transformer area based on the self-supervised graph attention network.

[0039] In this step, the variety of preset neural network parameter initialization strategies can be, for example, Xavier or Glorot initialization (an initialization that adaptively adjusts according to the input and output dimensions of the network layer), which is applicable to the weight matrix of the attention mechanism and adjusts the initialization range according to the input and output dimensions to avoid gradient vanishing or explosion; it can also be He initialization (an adaptive variance initialization). If ReLU (a rectified linear unit function) or its variant is used as the activation function, He initialization is adopted to alleviate the asymmetric problem of neuron output.

[0040] The preset optimizer can be, for example, the Adam optimizer, an adaptive moment estimation that can adaptively adjust the learning rate and is suitable for the non-convex optimization problem of the graph attention network.

[0041] The preset early stopping technique can be, for example, early stopping based on the validation loss or early stopping based on the task metric, etc. This embodiment does not limit this.

[0042] In this embodiment, the secondary side of the distribution transformer total and at least one user electricity meter in the sample distribution transformer area are used as each sample node to construct a sample node set of the sample distribution transformer area. The historical three-phase voltage amplitudes, historical three-phase injected current amplitudes, historical three-phase injected active powers, and historical three-phase injected reactive powers of the secondary side of the distribution transformer total and each user electricity meter in the sample distribution transformer area are used as the feature vectors of each sample node in the sample node set to construct an input sample, and the model structure of the self-supervised graph attention network is used for training; during the training process, the model combines the fusion loss function of the three-phase voltage consistency, the constraints of the three-phase injected active power and the three-phase injected reactive power errors, and the consistency regularization loss, and continuously adjusts the network parameters to learn the electrical coupling relationship between nodes, and finally obtains a three-phase impedance matrix identification model for the distribution transformer area based on the self-supervised graph attention network.

[0043] In an embodiment of the present application, a possible implementation is provided. In step a4 above, a fusion loss function combining three-phase voltage consistency, three-phase injected active power and three-phase injected reactive power error constraints, and consistency regularization loss is constructed. Specifically, it can be: The formula for constructing a fusion loss function combining three-phase voltage consistency, three-phase injected active power and three-phase injected reactive power error constraints, and consistency regularization loss is: (1) In formula (1), is the three-phase voltage consistency loss function, is the three-phase injected active power and three-phase injected reactive power error constraint function, is the consistency regularization loss function, are preset weight coefficients respectively. The preset weight coefficients here can be set according to actual needs, and this embodiment does not limit this.

[0044] In an embodiment of the present application, a possible implementation is provided. represents constraining the voltage amplitude error of each phase, and the total loss is the three-phase average: (2) In formula (2), n is the total number of nodes, p is the phase, A 、 B 、 C are phase values, and A 、 B 、 C constitute the three phases, is the node i obtained by acquisition p phase voltage amplitude, and are the real part and imaginary part of the predicted voltage of node i respectively, and the calculation is as follows: According to Kirchhoff's law, is the i node p phase complex voltage, satisfying formula (3): (3) In formula (3), is the i, j between nodes pm phase impedance value, is the j node m phase injected complex current; is the j node mThe phase injected complex power satisfies equations (4) and (5): (4) (5) In formula (4), For Node j of m Phase complex voltage, represents conjugation; In formula (5), The nodes collected j of m Phase injected active power, w is an imaginary unit, The nodes collected j of m Phase injected reactive power; According to formula (4) and formula (5), we can get: (6) In formula (6), The nodes collected j of m Phase voltage amplitude, For Node j of m Phase angle; Assume the secondary side of the distribution center is the reference node. i =1; the user's meter is a non-reference node, i =2,3,… n ; definition p = A The phase angle of the phase voltage is 0, that is, =0, then: , then formula (6) is converted into formula (7): (7) Substituting equation (7) into equation (3), we can separate The real and imaginary parts of , we get formula (8): (8) In formula (8), Nodes output by the model structure of the self-supervised graph attention network i, j between pm Phase resistance value, Nodes output by the model structure of the self-supervised graph attention network i, j between pm Phase reactance value; According to formula (8), we get formula (9): (9) Equations (10) and (11) are obtained according to Equation (9): (10) (11).

[0045] In this embodiment, according to Kirchhoff's law, the real part i of the predicted voltage and the imaginary part are obtained, and then, according to Equation (2), is calculated to constrain the voltage amplitude error of each phase, so that the fusion loss function has physical constraints, and there is no need to train the model with the true three-phase impedance matrix label.

[0046] A possible implementation is provided in the embodiment of the present application, constraining the three-phase injected active power and the three-phase injected reactive power errors, and the formula is as follows (12): (12) In Equation (12), is the predicted i phase injected active power of node p , is the measured i phase injected active power of node p , is the predicted i phase injected reactive power of node p , is the measured i phase injected reactive power of node p ; and are obtained through a preset known three-phase power flow calculation formula.

[0047] A possible implementation is provided in the embodiment of the present application, constraining the model to maintain consistency before and after the input is disturbed by noise, and enhancing the robustness of the model to noise. The formulas are as follows (13) and (14): (13) (14) In Equation (13), represents the model prediction result, represents the model prediction result after perturbation, represents the node feature perturbation, x represents the feature vector of the node; In Equation (14), conforms to a Gaussian distribution with a mean of 0.

[0048] In an embodiment of the present application, a possible implementation is provided. In step a2, the historical three-phase voltage amplitudes, historical three-phase injected current amplitudes, historical three-phase injected active powers, and historical three-phase injected reactive powers of the secondary side of the total distribution transformer in the sample distribution transformer area and each user meter are selected to construct the feature vectors of each sample node in the sample node set. Here, data preprocessing can be performed on the historical three-phase voltage amplitudes, historical three-phase injected current amplitudes, historical three-phase injected active powers, and historical three-phase injected reactive powers of the secondary side of the total distribution transformer in the sample distribution transformer area and each user meter, and then the feature vectors of each sample node in the sample node set are constructed.

[0049] Data preprocessing includes two parts: data cleaning and data normalization, which are used to process the original data to improve the accuracy of model prediction.

[0050] (1) Data cleaning When the measuring devices in the distribution network collect and transmit measurement data, the devices themselves and the external environment may affect the data. Therefore, it is necessary to perform data cleaning on the original measurement data to restore the real data as much as possible and ensure the quality of the input data of the self-supervised graph attention network model structure. For missing data at a certain moment, the linear interpolation method can be used for filling to avoid inconsistent dimensions of the feature vectors.

[0051] In this embodiment, an approximately symmetric compactly supported orthogonal wavelet denoising function can be used to perform preliminary processing on the measurement data. The approximately symmetric compactly supported orthogonal wavelet has good symmetry and can reduce distortion when performing data denoising.

[0052] (2) Data normalization processing In the analysis of power system data, using per-unit values to normalize three-phase electrical quantities can effectively eliminate the dimension difference, improve the model convergence efficiency, and enhance the comparability of data at different voltage levels. The per-unit value here is a dimensionless relative value representation method, which realizes data normalization by dividing the actual physical quantity (such as voltage, current, power, etc.) by a pre-selected reference value.

[0053] Next, a specific embodiment is used to introduce the training process of the three-phase impedance matrix identification model for the distribution transformer area.

[0054] In this specific embodiment, a 0.4 kV distribution system in a certain distribution area is selected as the experimental object, and the three-phase measurement data of the secondary side outlet of the distribution transformer in the substation area and the user electricity meters (32) in the affiliated substation area from October 10th to October 14th, 2024 are selected as the sample data set, specifically including: three-phase voltage amplitude (unit: V), three-phase injected current amplitude (unit: A), three-phase injected active power (unit: kW), three-phase injected reactive power (unit: kvar), the sampling interval is 60 minutes, and 24 sample points are collected every day, that is, the number of data samples included is 120. Among them, the first 80% of the data is used as the training set and the validation set, and the last 20% of the data is used as the test set to verify the model performance.

[0055] 1) Data preprocessing Each sample node is divided into 3 nodes according to A , B , C phase. Each sample has a total of 99 nodes. The feature vector of the sample node (taking the i phase of node A as an example), , the feature matrix is 4×99 dimensions, and the per-unit value is used to normalize the three-phase electrical quantities.

[0056] 2) Edge index It is set as a fully connected adjacency matrix (9081 dimensions). An example of its edge index sequence is as follows. The first dimension represents the source node, and the second dimension represents the target node:

[0057] 3) Model construction The model structure of the self-supervised graph attention network is set with 2 attention layers. The regularization coefficient is set to 0.4 for both (that is, 40% of the data in the sample is randomly set to zero to prevent the model from overfitting), and a dynamic residual mapping layer is set after each attention layer to reduce the loss of the original features. The node mask ratio is set to 20%, the initial value of the learning rate is 0.005, and a learning rate scheduler is set for continuous adjustment. The change process of the feature vector x of each node is as follows Figure 3 shown. In Figure 3 , " x before SSGAT1" means the feature vector x before passing through the first attention layer of the model structure of the Self Supervised Graph Attention Network (SSGAT); " x after SSGAT1" means the feature vector x after passing through the first attention layer of the self-supervised graph attention network; " x"before SSGAT2" represents the feature vector x before the second attention layer of the model structure of the self-supervised graph attention network; " x "after SSGAT2" represents the feature vector x after the second attention layer of the model structure of the self-supervised graph attention network, where tensor represents the tensor format.

[0058] 4) Result analysis The loss function of the training set of the model is as Figure 4 shown. In Figure 4 , the black curve represents the fusion loss (TotalLoss) , the blue curve represents the three-phase voltage consistency loss (Vol Loss) , the red curve represents the error constraint loss of the three-phase injected active power and the three-phase injected reactive power (P Loss) , and the green curve represents the consistency regularization loss (Cons Loss) . The ordinate represents the loss value (Loss Value), and the abscissa represents the number of training epochs (Epoch). The abscissa scales of the black curve, the blue curve, the red curve, and the green curve are the same. It can be seen from Figure 4 that as the number of training epochs increases, the loss value decreases and tends to be stable, and the training can be stopped, and finally a three-phase impedance matrix identification model for the distribution transformer area based on the self-supervised graph attention network is obtained.

[0059] It should be noted 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. In practical applications, all the above possible implementation manners can be combined in any combination to form possible embodiments of the present application, which will not be elaborated herein one by one.

[0060] Based on the three-phase impedance matrix identification method for the distribution transformer area based on the self-supervised graph attention network provided in the above embodiments, based on the same inventive concept, the embodiments of the present application also provide a three-phase impedance matrix identification device for the distribution transformer area based on the self-supervised graph attention network.

[0061] Figure 5 is the structural diagram of the three-phase impedance matrix identification device for the distribution transformer area based on the self-supervised graph attention network provided by the embodiments of the present application. As Figure 5 shown, the three-phase impedance matrix identification device for the distribution transformer area based on the self-supervised graph attention network may specifically include a first construction unit 510, an acquisition unit 520, a second construction unit 530, and an identification unit 540.

[0062] The first construction unit 510 is used to construct a node set of the distribution substation to be identified, including the secondary side of the total distribution substation in the zone and at least one user meter, with the secondary side of the total distribution substation in the zone and the user meter as nodes; The acquisition unit 520 is used to acquire the three-phase voltage amplitudes, three-phase injected current amplitudes, three-phase injected active powers, and three-phase injected reactive powers of the secondary side of the total distribution substation in the distribution substation to be identified collected in real time, and acquire the three-phase voltage amplitudes, three-phase injected current amplitudes, three-phase injected active powers, and three-phase injected reactive powers of each user meter in the distribution substation to be identified collected in real time; The second construction unit 530 is used to construct the feature vectors of each node in the node set according to the three-phase voltage amplitudes, three-phase injected current amplitudes, three-phase injected active powers, and three-phase injected reactive powers of the secondary side of the total distribution substation in the distribution substation to be identified and each user meter; The identification unit 540 is used to input the feature vectors of each node into a pre-trained three-phase impedance matrix identification model for the distribution substation, output the three-phase impedance matrix of the distribution substation to be identified, and mark it as , is a 3 n ×3 n complex matrix. In each element of the complex matrix, the real part is the resistance value and the imaginary part is the reactance value. n is the total number of nodes; among them, the three-phase impedance matrix identification model for the distribution substation is trained based on a self-supervised graph attention network and a fusion loss function that combines three-phase voltage consistency, three-phase injected active power and three-phase injected reactive power error constraints, and consistency regularization loss; the consistency regularization loss is used to constrain the model to maintain consistency before and after when the input is subject to noise perturbation.

[0063] A possible implementation manner is provided in the embodiments of the present application. As Figure 6 shown, the device shown above Figure 5 may further include a training unit 610, which is used to train a three-phase impedance matrix identification model for the distribution substation through the following steps: Construct a sample node set of the sample distribution substation with the secondary side of the total distribution substation in the sample distribution substation and at least one user meter as each sample node; Select the historical three-phase voltage amplitudes, historical three-phase injected current amplitudes, historical three-phase injected active powers, and historical three-phase injected reactive powers of the secondary side of the total distribution substation in the sample distribution substation and each user meter, and construct the feature vectors of each sample node in the sample node set; Establish the model structure of the self-supervised graph attention network. Among them, the graph attention part is composed of graph attention layers. Multiple attention heads are set in each graph attention layer for parallel calculation, and deep feature extraction is realized by stacking multiple graph attention layers; Construct a fusion loss function that combines the three-phase voltage consistency, the three-phase injected active power and reactive power error constraints, and the consistency regularization loss; Using a variety of preset neural network parameter initialization strategies, perform differential initialization for different types of network layers, input the feature vectors of each sample node in the sample node set into the model structure of the self-supervised graph attention network for training, and use the preset optimizer to minimize the fusion loss function on the training nodes to achieve gradient optimization of the model structure of the self-supervised graph attention network. At the same time, use the preset early stopping technique on the validation set to automatically trigger the training termination condition, and obtain a three-phase impedance matrix identification model for the distribution transformer area based on the self-supervised graph attention network.

[0064] In a possible implementation provided in the embodiments of the present application, the training unit 610 is further configured to: Taking the secondary side of the distribution transformer and at least one user electricity meter in the sample distribution transformer area as independent sample nodes, and not including the switches on the line between the secondary side of the distribution transformer and at least one user electricity meter, construct a sample node set of the sample distribution transformer area.

[0065] In a possible implementation provided in the embodiments of the present application, the adjacency matrix of the model structure of the established self-supervised graph attention network is fully connected. The model structure of the self-supervised graph attention network does not require the topological relationship identification step of the sample distribution transformer area, that is, it does not need to obtain the topological structure of the sample distribution transformer area, and defines the initial impedance parameter between any two nodes in the sample node set as the edge feature of the model structure of the self-supervised graph attention network.

[0066] In a possible implementation provided in the embodiments of the present application, the training unit 610 is further configured to: The formula for constructing a fusion loss function that combines the three-phase voltage consistency, the three-phase injected active power and reactive power error constraints, and the consistency regularization loss is: (1) In formula (1), is the three-phase voltage consistency loss function, is the three-phase injected active power and reactive power error constraint function, is the consistency regularization loss function, are preset weight coefficients respectively.

[0067] In a possible implementation provided in the embodiments of the present application, represents constraining the voltage amplitude error of each phase, and the total loss is the three-phase average: (2) In formula (2), nis the total number of nodes, p is the phase, A , B , C are phase values, and A , B , C constitute a three-phase, is the node obtained by acquisition i 's p phase voltage amplitude, and are respectively the real part and the imaginary part of the predicted voltage of node i , calculated as follows: According to Kirchhoff's law, is the i node's p phase complex voltage, satisfying Equation (3): (3) In Equation (3), is the i, j phase impedance value between nodes pm , is the j node's m phase injected complex current; is the j node's m phase injected complex power, satisfying Equation (4) and Equation (5): (4) (5) In Equation (4), is the j node's m phase complex voltage, represents the conjugate; In Equation (5), is the j node obtained by acquisition's m phase injected active power, w is the imaginary unit, is the j node obtained by acquisition's m phase injected reactive power; According to Equation (4) and Equation (5), we get: (6) In Equation (6), is the j node obtained by acquisition's m phase voltage amplitude, is the j node's m phase phase angle; Set the secondary side of the distribution substation in the power distribution area as the reference node, i = 1; The user's electricity meter is a non-reference node, i = 2, 3, … n ; Define p = A The phase voltage phase angle is 0, that is, in Equation (6) = 0, then: If, then Equation (6) yields the following Equation (7): (7) Substitute Equation (7) into Equation (3), separate the real and imaginary parts of to obtain Equation (8): (8) In Equation (8), is the phase resistance value between nodes i, j output by the model structure of the self-supervised graph attention network, pm ; is the phase reactance value between nodes i, j output by the model structure of the self-supervised graph attention network, pm ; According to Equation (8), Equation (9) is obtained: (9) According to Equation (9), Equation (10) and Equation (11) are obtained: (10) (11).

[0068] A possible implementation is provided in the embodiments of the present application, constrain the three-phase injected active power and three-phase injected reactive power errors, and the formula is as follows (12): (12) In Equation (12), is the predicted i phase injected active power of node p , is the measured i phase injected active power of node p , is the predicted i phase injected reactive power of node p , is the measured i phase injected reactive power of node p ; and are obtained through the preset known three-phase power flow calculation formula.

[0069] In an embodiment of the present application, a possible implementation is provided. The constraint model remains consistent before and after the input is disturbed by noise, enhancing the robustness of the model to noise. The formulas are as follows (13) and (14): (13) (14) In formula (13), represents the model prediction result, represents the model prediction result after perturbation, represents the node feature perturbation, x represents the feature vector of the node; In formula (14), conforms to a Gaussian distribution with a mean of 0.

[0070] Based on the same inventive concept, an embodiment of the present application also provides an electronic device, including a processor and a memory. A computer program is stored in the memory, and the processor is configured to run the computer program to execute the three-phase impedance matrix identification method of the distribution transformer area based on the self-supervised graph attention network in any one of the above embodiments.

[0071] In an exemplary embodiment, an electronic device is provided, as Figure 7 shown. Figure 7 The electronic device 700 shown includes: a processor 701 and a memory 703. Among them, the processor 701 and the memory 703 are connected, such as connected through a bus 702. Optionally, the electronic device 700 may further include a transceiver 704. It should be noted that in practical applications, the transceiver 704 is not limited to one, and the structure of the electronic device 700 does not constitute a limitation to the embodiments of the present application.

[0072] The processor 701 may be a CPU (Central Processing Unit), GPU (Graphics Processing Unit), DSP (Digital Signal Processor), ASIC (Application Specific Integrated Circuit), FPGA or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. It can implement or execute various exemplary logical blocks, modules and circuits described in combination with the disclosure of the present application. The processor 701 may also be a combination that realizes computing functions, such as a combination including one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0073] The bus 702 may include a path for transmitting information among the above components. The bus 702 can be a PCI (Peripheral Component Interconnect) bus, an EISA (Extended Industry Standard Architecture) bus, or the like. The bus 702 can be divided into an address bus, a data bus, a control bus, etc. For the sake of convenience of representation, Figure 7 it is only represented by a thick line in Figure 7 , but it does not mean that there is only one bus or one type of bus.

[0074] The memory 703 can be a ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, a RAM (Random Access Memory) or other types of dynamic storage devices that can store information and instructions, or it can also be an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory), or other optical disc storage, optical disc storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), magnetic disk storage media, or other magnetic storage devices, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto.

[0075] The memory 703 is used to store the computer program code for executing the solution of this application and is controlled by the processor 701 to execute. The processor 701 is used to execute the computer program code stored in the memory 703 to implement the content shown in the foregoing method embodiments.

[0076] Among them, the electronic device includes but is not limited to: mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Tablet Computers), PMPs (Portable Multimedia Players), in-vehicle terminals (such as in-vehicle navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 7 The shown electronic device is only an example and should not bring any limitations to the functions and usage scopes of the embodiments of this application.

[0077] Those skilled in the art can clearly understand the specific working processes of the above-described systems, devices, and modules. They can refer to the corresponding processes in the foregoing method embodiments. For the sake of brevity, they will not be described in detail here.

[0078] Those of ordinary skill in the art can understand that the technical solution of the present application can essentially be embodied in the form of a software product, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, which includes a number of program instructions for causing an electronic device (such as a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the embodiments of the present application when the program instructions are running. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs.

[0079] Alternatively, all or part of the steps of implementing the foregoing method embodiments can be completed by hardware related to program instructions (such as an electronic device such as a personal computer, a server, or a network device), and the program instructions can be stored in a computer-readable storage medium. When the program instructions are executed by the processor of the electronic device, the electronic device executes all or part of the steps of the methods described in the embodiments of the present application.

[0080] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that within the spirit and principles of the present application, they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the corresponding technical solutions to deviate from the protection scope of the present application.

Claims

1. A three-phase impedance matrix identification method for a distribution transformer area based on a self-supervised graph attention network, characterized in that The method includes: Taking the secondary side of the substation distribution main and the user electricity meters as nodes, constructing a node set of the distribution substation to be identified, including the secondary side of the substation distribution main and at least one user electricity meter; Obtaining the three-phase voltage amplitudes, three-phase injected current amplitudes, three-phase injected active powers, and three-phase injected reactive powers of the secondary side of the substation distribution main of the distribution substation to be identified collected in real time, and obtaining the three-phase voltage amplitudes, three-phase injected current amplitudes, three-phase injected active powers, and three-phase injected reactive powers of each user electricity meter of the distribution substation to be identified collected in real time; Constructing feature vectors of each node in the node set according to the three-phase voltage amplitudes, three-phase injected current amplitudes, three-phase injected active powers, and three-phase injected reactive powers of the secondary side of the substation distribution main and each user electricity meter of the distribution substation to be identified; Input the feature vectors of each node into a pre-trained three-phase impedance matrix identification model for a distribution transformer area, output the three-phase impedance matrix of the distribution transformer area to be identified, and label it as , is a 3 n ×3 n complex matrix. In each element of the complex matrix, the real part is the resistance value and the imaginary part is the reactance value. n is the total number of nodes; among them, the three-phase impedance matrix identification model for the distribution transformer area is trained based on a self-supervised graph attention network and combined with a fusion loss function of three-phase voltage consistency, three-phase injected active power and three-phase injected reactive power error constraints, and consistency regularization loss; the consistency regularization loss is used to constrain the model to maintain consistency before and after when the input is disturbed by noise.

2. The three-phase impedance matrix identification method for a distribution transformer area based on a self-supervised graph attention network according to claim 1, wherein Training to obtain a three-phase impedance matrix identification model of the distribution substation through the following steps: Taking the secondary side of the substation distribution main and at least one user electricity meter of the sample distribution substation as each sample node, constructing a sample node set of the sample distribution substation; Selecting the historical three-phase voltage amplitudes, historical three-phase injected current amplitudes, historical three-phase injected active powers, and historical three-phase injected reactive powers of the secondary side of the substation distribution main and each user electricity meter of the sample distribution substation, and constructing feature vectors of each sample node in the sample node set; Establishing the model structure of the self-supervised graph attention network, where the graph attention part is composed of graph attention layers, multiple attention heads are set in each graph attention layer for parallel calculation, and deep feature extraction is realized by stacking multiple graph attention layers; Constructing a fusion loss function that combines three-phase voltage consistency, three-phase injected active power and three-phase injected reactive power error constraints, and consistency regularization loss; Using a preset variety of neural network parameter initialization strategies, performing differential initialization for different types of network layers, inputting the feature vectors of each sample node in the sample node set into the model structure of the self-supervised graph attention network for training, using a preset optimizer to minimize the fusion loss function on the training nodes, realizing the gradient optimization of the model structure of the self-supervised graph attention network, and at the same time automatically triggering the training termination condition using a preset early stopping technique on the validation set, to obtain a three-phase impedance matrix identification model of the distribution substation based on the self-supervised graph attention network.

3. The three-phase impedance matrix identification method for a distribution transformer area based on a self-supervised graph attention network according to claim 2, characterized in that Taking the secondary side of the substation distribution main and at least one user electricity meter of the sample distribution substation as each sample node, constructing a sample node set of the sample distribution substation, including: Taking the secondary side of the substation distribution main and at least one user electricity meter of the sample distribution substation as independent sample nodes, without including the switches on the line between the secondary side of the substation distribution main and at least one user electricity meter, constructing a sample node set of the sample distribution substation.

4. The three-phase impedance matrix identification method for a distribution transformer area based on a self-supervised graph attention network according to claim 2, characterized in that The adjacency matrix of the established model structure of the self-supervised graph attention network is fully connected. The model structure of the self-supervised graph attention network does not require the step of identifying the topological relationship of the sample distribution substation, that is, it does not need to obtain the topological structure of the sample distribution substation, and defines the initial impedance parameter between any two nodes in the sample node set as the edge feature of the model structure of the self-supervised graph attention network.

5. The three-phase impedance matrix identification method for a distribution transformer area based on a self-supervised graph attention network according to claim 2, characterized in that Construct a fusion loss function that combines the three-phase voltage consistency, the error constraints of the three-phase injected active power and the three-phase injected reactive power, and the consistency regularization loss, including: The formula for constructing a fusion loss function that combines the three-phase voltage consistency, the error constraints of the three-phase injected active power and the three-phase injected reactive power, and the consistency regularization loss is: （1） In formula (1), is the three-phase voltage consistency loss function, is the error constraint function of the three-phase injected active power and the three-phase injected reactive power, is the consistency regularization loss function, are respectively the preset weight coefficients.

6. The three-phase impedance matrix identification method for a distribution transformer area based on a self-supervised graph attention network according to claim 5, wherein Represents the constraint on the voltage amplitude error of each phase, and the total loss is the three-phase average: （2） In Equation (2), n is the total number of nodes, p is the phase, A , B , C are phase values, and A , B , C constitute a three-phase, is the node voltage amplitude of the i obtained by acquisition, p phase voltage amplitude, and are the real and imaginary parts of the predicted voltage of node i respectively, and the calculation is as follows: According to Kirchhoff's law, for node i the p complex phase voltage satisfies Equation (3): （3） In formula (3), is the i, j phase impedance value between pm nodes; is the j complex injected current of m phase at node; is the j complex injected power of m phase at node, satisfying formulas (4) and (5): （4） （5） In Equation (4), is the j phase complex voltage of node m , and represents the conjugate. In Equation (5), is the active power injected into the node obtained by acquisition j of m phase, w is the imaginary unit, is the reactive power injected into the node obtained by acquisition j of m phase; According to equations (4) and (5), we get: （6） In Equation (6), is the phase voltage amplitude of the node obtained by acquisition j ; m is the phase phase angle of the node ; j ; m ; Set the secondary side of the distribution substation in the power distribution area as the reference node, i = 1; The user electric meter is a non-reference node, i = 2, 3, … n ; Definition p = A The phase voltage phase angle is 0, that is, in Equation (6) = 0, then: , then formula (6) gives the following formula (7): （7） Substitute Equation (7) into Equation (3), separate out the real and imaginary parts to obtain Equation (8): （8） In formula (8), is the node output by the model structure of the self-supervised graph attention network i, j between pm phase resistance values, is the node output by the model structure of the self-supervised graph attention network i, j between pm phase reactance values; Equation (9) is obtained from equation (8): （9） Equations (10) and (11) are obtained from equation (9): （10） （11）。 7. The three-phase impedance matrix identification method for a distribution transformer area based on a self-supervised graph attention network according to claim 5, characterized in that Constrain the three-phase injected active power and the three-phase injected reactive power errors, as shown in the following formula (12): （12） In formula (12), is the predicted active power injected at node i . p is the measured active power injected at node . i . p is the predicted reactive power injected at node . i . p is the measured reactive power injected at node . i . p They are obtained through a preset known three-phase power flow calculation formula. and .

8. The three-phase impedance matrix identification method for a distribution transformer area based on a self-supervised graph attention network according to claim 5, characterized in that The constraint model remains consistent before and after the input is disturbed by noise, enhancing the model's robustness to noise. The formulas are as follows (13) and (14): （13） （14） In formula (13), represents the model prediction result, represents the model prediction result after perturbation, represents the node feature perturbation, x represents the feature vector of the node; In formula (14), obeys a Gaussian distribution with a mean of 0.

9. A three-phase impedance matrix identification device for a distribution transformer area based on a self-supervised graph attention network, characterized in that, The device includes: A first construction unit, configured to construct a node set of the power distribution substation to be identified, including the secondary side of the total distribution in the substation area and at least one user meter, with the secondary side of the total distribution in the substation area and the user meters as nodes; An acquisition unit, configured to acquire the three-phase voltage amplitudes, the three-phase injected current amplitudes, the three-phase injected active power, and the three-phase injected reactive power of the secondary side of the total distribution in the power distribution substation to be identified collected in real time, and acquire the three-phase voltage amplitudes, the three-phase injected current amplitudes, the three-phase injected active power, and the three-phase injected reactive power of each user meter in the power distribution substation to be identified collected in real time; A second construction unit, configured to construct a feature vector of each node in the node set according to the three-phase voltage amplitudes, the three-phase injected current amplitudes, the three-phase injected active power, and the three-phase injected reactive power of the secondary side of the total distribution in the power distribution substation to be identified and each user meter; An identification unit is configured to input the feature vectors of each node into a pre-trained three-phase impedance matrix identification model for a distribution transformer area, output the three-phase impedance matrix of the distribution transformer area to be identified, and mark it as , which is a 3 n ×3 n complex matrix. In each element of the complex matrix, the real part is the resistance value and the imaginary part is the reactance value. n where is the total number of nodes. Among them, the three-phase impedance matrix identification model for the distribution transformer area is trained based on a self-supervised graph attention network and a fusion loss function that combines three-phase voltage consistency, three-phase injected active power and three-phase injected reactive power error constraints, and consistency regularization loss. The consistency regularization loss is used to constrain the model to maintain consistency before and after when the input is disturbed by noise.

10. An electronic device, characterized in that, It includes a processor and a memory. Among them, the memory stores a computer program, and the processor is configured to run the computer program to execute the method for identifying the three-phase impedance matrix of the power distribution substation based on the self-supervised graph attention network according to any one of claims 1 to 8.

Images