Load transfer configuration method and device based on large model of power grid topological graph

Abstract

The invention relates to the technical field of power grid operation and maintenance, and particularly provides a load transfer configuration method and device based on a power grid topological graph large model, and the method comprises the steps: enabling a fault region power grid topological graph to serve as the input of a pre-trained graph neural network model, and obtaining an adjusted power grid topological graph outputted by the pre-trained graph neural network model; and performing power load transfer configuration on the power grid in the fault area by using the adjusted power grid topological graph. According to the technical scheme provided by the invention, the power grid topological graph large model with the power grid topological graph analysis capability is used as an auxiliary tool for selecting the power load transfer point, so that the working efficiency and accuracy of power grid operation personnel can be improved, and reliable and stable operation of a power system is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid operation and maintenance, and specifically relates to a load transfer configuration method and device based on a large model of a power grid topology diagram. Background Art

[0002] In a power system, when a fault occurs at a certain node, the load connected to this node will lose power supply. To ensure the normal power consumption of these loads, it is necessary to select a suitable power load transfer point to transfer the load of the fault node to other normal lines, so as to avoid overloading and damage of the faulty equipment, prevent the spread of the fault, and ensure power supply stability. A power load transfer point refers to a connection point in the power grid (such as a sectionalizing switch, a backup power access point) that can receive and transfer the load in the fault area through switch operations.

[0003] In the prior art, the selection of power load transfer points mainly relies on empirical rules, that is, grid operators manually operate switches to select transfer points based on historical experience according to the fault type, location, and grid status. In a simple power grid or common fault scenarios, this method can quickly formulate a transfer plan; however, in a complex power grid (such as an active distribution network with distributed power sources), due to the existence of multiple power sources (such as the main grid, distributed power sources) and load points in the complex power grid, and their interaction relationships are complex, it is difficult for empirical rules to comprehensively consider the interactions between multiple sources and multiple loads, which may lead to improper selection of transfer paths, and manual intervention takes a long time and is prone to errors. Summary of the Invention

[0004] To overcome the above defects, the present invention proposes a load transfer configuration method and device based on a large model of a power grid topology diagram.

[0005] In the first aspect, a load transfer configuration method based on a large model of a power grid topology diagram is provided. The load transfer configuration method based on a large model of a power grid topology diagram includes:

[0006] Taking the power grid topology diagram of the fault area as the input of a pre-trained graph neural network model, and obtaining an adjusted power grid topology diagram output by the pre-trained graph neural network model;

[0007] Using the adjusted power grid topology diagram to perform power load transfer configuration on the power grid in the fault area.

[0008] Preferably, the edges of the power grid topology graph represent the physical connection relationships between nodes. The attributes of the edges of the power grid topology graph include: connection relationship, on-off attribute, and power loss. The on-off attribute includes: conduction and disconnection; the attributes of the nodes of the power grid topology graph include: node type, device parameters, fault attribute, transfer supply attribute, and on-off attribute. The node types include: source node, intermediate node, and load node. The device parameters include: power capacity and power loss. The fault attribute includes: fault and normal; the transfer supply attribute includes: candidate transfer supply point and non-candidate transfer supply point. The on-off attribute includes conduction and disconnection.

[0009] Further, the training process of the pre-trained graph neural network model includes:

[0010] Constructing training data by using a power grid topology graph with node faults and its corresponding power grid topology graph for repairing the node faults;

[0011] Training an initial graph neural network model by using the training data to obtain the pre-trained graph neural network model.

[0012] Further, during the process of training the initial graph neural network model by using the training data, an attribute update rule is embedded in the forward propagation process of the model to update the on-off attributes of each node and edge in the input power grid topology graph.

[0013] Further, the attribute update rule includes: for a node with multiple child nodes, when one of its child nodes fails, if the conduction probability of the edge between it and its unfailed child nodes exceeds 0.5, then adjust the on-off attribute of this edge to conduction; otherwise, adjust the on-off attribute of this edge to disconnection; where the conduction probability of the edge is as follows:

[0014]

[0015] In the above formula, D is the conduction probability of the edge, e is the natural constant, z is the conduction coefficient, and the conduction coefficient is as follows:

[0016]

[0017] In the above formula, k0 is a constant coefficient, x is the coefficient corresponding to the faulty child node, y is the power loss of the faulty child node, k i is the coefficient corresponding to the i-th unfailed child node, P i is the power loss of the i-th unfailed child node, and I is the total number of unfailed child nodes.

[0018] Further, the attribute update rule includes: for a node with only one child node, if the conduction probability of the edge between it and its child node exceeds 0.5, then adjust the on-off attribute of this edge to conduction; otherwise, adjust the on-off attribute of this edge to disconnection; where the conduction probability of the edge is as follows:

[0019]

[0020] In the above formula, D is the conduction probability of the edge, e is the natural constant, and z is the conduction coefficient. The conduction coefficient is as follows:

[0021] Z = k0 + mn

[0022] In the above formula, k0 is the constant coefficient, m is the coefficient corresponding to the node with only one child node, and n is the power loss of the node with only one child node.

[0023] Further, the attribute update rule includes: for a node with only one parent node, if the conduction probability of this node exceeds 0.5, then adjust the on-off attribute of this node to conduction; otherwise, adjust the on-off attribute of this node to disconnection; where the conduction probability of the edge is as follows:

[0024]

[0025] In the above formula, D is the conduction probability of the edge, e is the natural constant, and z is the conduction coefficient. The conduction coefficient is as follows:

[0026] Z = k0 + we

[0027] In the above formula, k0 is the constant coefficient, w is the coefficient corresponding to the edge between the node with only one parent node and its parent node, and e is the power loss of the edge between the node with only one parent node and its parent node.

[0028] Further, the attribute update rule includes: for a node with multiple parent nodes, if the conduction probability of this node exceeds 0.5, then adjust the on-off attribute of this node to conduction; otherwise, adjust the on-off attribute of this node to disconnection; where the conduction probability of the edge is as follows:

[0029]

[0030] In the above formula, D is the conduction probability of the edge, e is the natural constant, and z is the conduction coefficient. The conduction coefficient is as follows:

[0031]

[0032] In the above formula, k0 is the constant coefficient, k jis the coefficient corresponding to the edge between a node with multiple parent nodes and its j-th parent node, P j is the power loss of the edge between a node with multiple parent nodes and its j-th parent node, and J is the total number of edges between a node with multiple parent nodes and its parent nodes.

[0033] Furthermore, during the process of training the initial graph neural network model using the training data, the loss function is as follows:

[0034] L total = αL node + βL edge

[0035]

[0036] In the above formula, L total is the total loss function value, α is the first coefficient, L node is the first loss function value, β is the second coefficient, L edge is the second loss function value, N is the total number of nodes, Y n is the true label of the on / off attribute of node n, is the predicted label of the on / off attribute of node n, E is the total number of edges, Y e is the true label of the on / off attribute of edge e, is the predicted label of the on / off attribute of edge e, and α + β = 1.

[0037] In a second aspect, a load transfer configuration device based on a large power grid topology graph model is provided. The load transfer configuration device based on the large power grid topology graph model includes:

[0038] An analysis module, configured to use the power grid topology graph of the fault area as the input of a pre-trained graph neural network model, and obtain an adjusted power grid topology graph output by the pre-trained graph neural network model;

[0039] A configuration module, configured to perform power load transfer configuration on the power grid in the fault area by using the adjusted power grid topology graph.

[0040] One or more of the above technical solutions of the present invention have at least one or more of the following beneficial effects:

[0041] The present invention provides a method and device for load transfer configuration based on a large model of a power grid topology map, including: using the power grid topology map of a fault area as the input of a pre-trained graph neural network model to obtain an adjusted power grid topology map output by the pre-trained graph neural network model; and using the adjusted power grid topology map to perform power load transfer configuration on the power grid in the fault area. The technical solution provided by the present invention provides an auxiliary selection for power load transfer points for power grid operators, effectively improving work efficiency and accuracy, thereby ensuring the reliable and stable operation of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is a schematic flow chart of the main steps of the load transfer configuration method based on the large model of the power grid topology map according to an embodiment of the present invention;

[0043] Figure 2 is the power grid topology map according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0044] The following further describes in detail the specific embodiments of the present invention with reference to the drawings.

[0045] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0046] Embodiment 1

[0047] Refer to the attached Figure 1 , Figure 1 which is a schematic flow chart of the main steps of the load transfer configuration method based on the large model of the power grid topology map according to an embodiment of the present invention. As Figure 1 shown, the load transfer configuration method based on the large model of the power grid topology map in the embodiment of the present invention mainly includes the following steps:

[0048] Step S101: Use the power grid topology map of the fault area as the input of the pre-trained graph neural network model to obtain an adjusted power grid topology map output by the pre-trained graph neural network model;

[0049] Step S102: Use the adjusted power grid topology map to perform power load transfer configuration on the power grid in the fault area.

[0050] In this embodiment, it is considered that the power grid topology has the following characteristics: The nodes in the power grid topology represent various facilities in the power grid, including generator equipment, substations, and terminal equipment (loads), etc. The edges in the power grid topology represent the connection relationships between devices, specifically the connection relationships through wires. The edges have unidirectionality, usually flowing from the generator to the load terminal (except for energy storage devices), which reflects the physical connection relationships between nodes. Moreover, both nodes and edges have certain power losses (where different types of nodes have different loss calculation functions, which will not be elaborated here).

[0051] Therefore, for the convenience of analysis, the nodes in the power grid topology diagram are divided into three categories: source nodes, intermediate nodes, and load nodes. A source node refers to the starting point that provides electric energy in the power grid, usually the outlet of a power plant or a substation, and is responsible for injecting electric energy into the power grid. An intermediate node refers to the node that transmits and distributes electric energy in the power grid, responsible for transmitting electric energy from the source node to the load node, and playing a role of a bridge and link in the power grid, such as the transfer point of a substation and the power load transfer point. Among them, considering that the role of the power load transfer point is to transfer the load of a faulty node to other normal lines when a certain node fails, the power load transfer point can be mapped to an intermediate node according to the connection relationship. A load node refers to the node that consumes electric energy in the power grid, usually various electrical equipment or users.

[0052] Since the edges in the power grid topology structure represent the physical connection relationships between nodes, and this physical connection relationship will not change due to the failure of a certain node, in order to adapt to the power load transfer scenario, it is necessary to redefine the attributes of nodes and edges. Specifically, the node attributes include node type (such as source node, intermediate node, load node), device parameters (such as power capacity, power loss, etc.), fault attribute (1 represents faulty, 0 represents normal), transfer attribute (1 represents candidate transfer point, 0 represents non-candidate transfer point), and on-off attribute (1 represents conducting, 0 represents disconnecting); the edge attributes include connection relationship (such as wire connection), on-off attribute (1 represents conducting, 0 represents disconnecting), and power loss.

[0053] It should be noted that the on-off attribute is used to indicate whether the current node / edge is conducting electricity normally, that is, when the on-off attribute is in the conducting state, it proves that the current node / edge can conduct electricity normally, and when the on-off attribute is in the disconnecting state, it proves that the current node / edge cannot conduct electricity normally.

[0054] In the present invention, the edges of the power grid topology diagram represent the physical connection relationships between nodes. The attributes of the edges of the power grid topology diagram include: connection relationship, on-off attribute, and power loss. The on-off attribute includes: conduction and disconnection; the attributes of the nodes of the power grid topology diagram include: node type, device parameters, fault attribute, transfer supply attribute, and on-off attribute. The node type includes: source node, intermediate node, and load node. The device parameters include: power capacity and power loss. The fault attribute includes: fault and normal; the transfer supply attribute includes: candidate transfer supply point and non-candidate transfer supply point. The on-off attribute includes conduction and disconnection.

[0055] In one embodiment, the training process of the pre-trained graph neural network model includes:

[0056] Constructing training data by using a power grid topology diagram with node faults and its corresponding power grid topology diagram for repairing the node faults;

[0057] Training an initial graph neural network model by using the training data to obtain the pre-trained graph neural network model.

[0058] In one embodiment, during the process of training the initial graph neural network model by using the training data, an attribute update rule is embedded in the forward propagation process of the model to update the on-off attributes of each node and edge in the input power grid topology diagram.

[0059] In one embodiment, the attribute update rule includes: for a node with multiple child nodes, when one of its child nodes fails, if the conduction probability of the edge between it and its unfailed child nodes exceeds 0.5, then adjust the on-off attribute of the edge to conduction; otherwise, adjust the on-off attribute of the edge to disconnection. Wherein, the conduction probability of the edge is as follows:

[0060]

[0061] In the above formula, D is the conduction probability of the edge, e is the natural constant, z is the conduction coefficient, and the conduction coefficient is as follows:

[0062]

[0063] In the above formula, k0 is a constant coefficient, x is the coefficient corresponding to the faulty child node, y is the power loss of the faulty child node, k i is the coefficient corresponding to the i-th unfailed child node, P i is the power loss of the i-th unfailed child node, and I is the total number of unfailed child nodes.

[0064] In one embodiment, the attribute update rule includes: for a node with only one child node, if the conduction probability of the edge between it and its child node exceeds 0.5, then adjust the on-off attribute of this edge to conduction; otherwise, adjust the on-off attribute of this edge to disconnection; wherein, the conduction probability of the edge is as follows:

[0065]

[0066] In the above formula, D is the conduction probability of the edge, e is the natural constant, z is the conduction coefficient, and the conduction coefficient is as follows:

[0067] Z = k0 + mn

[0068] In the above formula, k0 is a constant coefficient, m is the coefficient corresponding to the node with only one child node, and n is the power loss of the node with only one child node.

[0069] In one embodiment, the attribute update rule includes: for a node with only one parent node, if the conduction probability of this node exceeds 0.5, then adjust the on-off attribute of this node to conduction; otherwise, adjust the on-off attribute of this node to disconnection; wherein, the conduction probability of the edge is as follows:

[0070]

[0071] In the above formula, D is the conduction probability of the edge, e is the natural constant, z is the conduction coefficient, and the conduction coefficient is as follows:

[0072] Z = k0 + we

[0073] In the above formula, k0 is a constant coefficient, w is the coefficient corresponding to the edge between the node with only one parent node and its parent node, and e is the power loss of the edge between the node with only one parent node and its parent node.

[0074] In one embodiment, the attribute update rule includes: for a node with multiple parent nodes, if the conduction probability of this node exceeds 0.5, then adjust the on-off attribute of this node to conduction; otherwise, adjust the on-off attribute of this node to disconnection; wherein, the conduction probability of the edge is as follows:

[0075]

[0076] In the above formula, D is the conduction probability of the edge, e is the natural constant, z is the conduction coefficient, and the conduction coefficient is as follows:

[0077]

[0078] In the above formula, k0 is a constant coefficient, k jis the coefficient corresponding to the edge between the node with multiple parent nodes and its j-th parent node, P j is the power loss of the edge between the node with multiple parent nodes and its j-th parent node, and J is the total number of edges between the node with multiple parent nodes and its parent nodes.

[0079] In one embodiment, during the process of training the initial graph neural network model using the training data, the loss function is as follows:

[0080] L total = αL node + βL edge

[0081]

[0082] In the above formula, L total is the total loss function value, α is the first coefficient, L node is the first loss function value, β is the second coefficient, L edge is the second loss function value, N is the total number of nodes, Y n is the true label of the on / off attribute of node n, is the predicted label of the on / off attribute of node n, E is the total number of edges, Y e is the true label of the on / off attribute of edge e, is the predicted label of the on / off attribute of edge e, and α + β = 1.

[0083] In a specific embodiment, the technical solution disclosed in this embodiment mainly includes key steps such as power grid topology graph data preparation, graph large model construction, model training and fine-tuning, and model evaluation. The overall solution will be described in sequence below:

[0084] 1. Power grid topology graph data preparation:

[0085] Use power system simulation tools (such as MATLAB / Simulink, OpenDSS) to generate power grid topology graph data applicable to the power load transfer scenario according to the above-mentioned representation rules. The power grid topology graph data includes the node attributes of all nodes in the graph, the edge attributes of all edges, and the connection relationships between nodes. Among them, Figure 2 is a schematic diagram of the power grid topology graph. The node N1 in the graph is the source node, N2 to N15 are intermediate nodes, N16 to N19 are load nodes, and E1 to E28 are edges.

[0086] Simulate different fault scenarios and adjust the power grid topology graph data. Taking Figure 1Taking the power grid topology diagram described above as an example, nodes N8, N10, and N12 in the diagram are candidate transfer supply points. Suppose node N9 fails. At this time, the fault attribute in the node attributes of node N9 needs to be adjusted from 0 to 1. The adjusted power grid topology diagram data is used as a training sample, and then the correct transfer supply path is marked. The on-off attribute of the transfer supply points (such as N8) included in this transfer supply path is adjusted from 0 to 1. The marked and adjusted power grid topology diagram data is used as a validation sample.

[0087] The generated data is divided into a training set, a validation set, and a test set. Among them, the training set is used for model training, the validation set is used for parameter tuning, and the test set is used for final evaluation. The division ratio is, for example but not limited to, 70% training set, 15% validation set, and 15% test set.

[0088] Through the above method, it is possible to generate power grid topology diagram data that meets the requirements according to the needs of the power load transfer supply scenario, and label the data for subsequent model training and validation.

[0089] 2: Construction of the graph large model:

[0090] The goal of this stage is to construct a large power grid topology graph model based on the graph neural network (GNN), and it is required that the constructed graph large model can parse the power grid topology graph and update the on-off attributes of nodes and edges based on pre-established rules. The specific steps are as follows:

[0091] 2-1: Select the graph neural network (GNN) as the core model;

[0092] 2-2: Design the architecture of the large power grid topology graph model, including: input layer, GNN layer, and output layer;

[0093] 2-3: Embed the attribute update rules in the forward propagation process of the model to update the on-off attributes of each node and edge in the input data (power grid topology graph data);

[0094] Specifically, the attribute update rule is: if a node fails, the on-off attributes of this node and the edges connected to it are directly updated to 0, that is, in the disconnected state. For the remaining nodes in the power grid topology graph (that is, other nodes except the faulty node) and the edges that have no connection relationship with them, the on-off attributes need to be calculated and updated according to the connection relationship and power loss according to the following rules:

[0095] 1) For a node with multiple downward connections, the on-off attribute of each edge can be determined using binary classification:

[0096] Taking the connection relationship shown by N5, N8, N9, and N10 in the appendix Figure 1 as an example, the specific determination process is as follows:

[0097] When N9 fails, edge E9 is disconnected. At this time, the conduction probability of edge E8 can be calculated using logistic regression. When the conduction probability is greater than or equal to 0.5, its on-off attribute takes the value of 1; otherwise, it takes the value of 0. The logistic regression calculation formula is as follows:

[0098]

[0099] Among them, D(E8) represents the conduction probability of edge E8; Z = k0 + k1P1(N5) + k2P2(E9, N9) + k3P3(E10, N10); P1(N5) is the power loss of node N5; P2(E9, N9) is the total power loss of edge E9 and node N9; P3(E10, N10) is the total power loss of edge E10 and node N10.

[0100] Similarly, the calculation method of the on-off attribute of edge E10 is the same as that of edge E8.

[0101] 2) For a node that outputs only a single edge, the on-off attribute of the edge is only related to whether the previous node is conducting:

[0102] Taking the connection relationship shown by N2 and E3 in the appendix Figure 1 as an example, the specific determination process is as follows:

[0103] For edge E3, its on-off attribute is only related to node N2. At this time, the conduction probability of edge E3 can be calculated using logistic regression. When the conduction probability is greater than or equal to 0.5, its on-off attribute takes the value of 1; otherwise, it takes the value of 0. The logistic regression calculation formula is as follows:

[0104]

[0105] Among them, D((E3) represents the conduction probability of edge E3. When the conduction probability is greater than or equal to 0.5, the on-off attribute takes the value of 1; otherwise, it takes the value of 0; Z = k0 + k1P1(N2); P1(N2) is the power loss of node N2.

[0106] 3) For an edge that connects to a node downward, the conduction attribute of the node is only related to this edge:

[0107] Taking the connection relationship shown by E1 and N2 in the appendix Figure 1 as an example, the specific determination process is as follows:

[0108] For node N2, its on-off attribute is only related to edge E1. At this time, the conduction probability of node N2 can be calculated using logistic regression. When the conduction probability is greater than or equal to 0.5, its on-off attribute takes the value of 1; otherwise, it takes the value of 0. The logistic regression calculation formula is as follows:

[0109]

[0110] Among them, D(N2) represents the conduction probability of node N2; Z = k0 + k1P1(E1); P1(E1) is the power loss of edge E1.

[0111] 4) For a node connected by multiple edges, the conduction attribute of this node is related to multiple edges:

[0112] Take the connection relationship shown by E15, E16, E17 and N14 in the appendix Figure 1 as an example, the specific determination process is as follows:

[0113] For node N14, its on-off attribute is related to edges E15, E16, and E17. At this time, the conduction probability of node N14 can be calculated using logistic regression. When the conduction probability is greater than or equal to 0.5, its on-off attribute value is 1, otherwise it is 0; among them, the logistic regression calculation formula is as follows:

[0114]

[0115] Among them, D(N14) represents the conduction probability of node N14; Z = k0 + k1P1(E15) + k2P2(E16) + k3P3(E17); P1(E15) represents the power loss of edge E15, P2(E16) is the power loss of edge E16, and P3(E17) is the power loss of edge E17;

[0116] It should be noted that in the above attribute update rule, the power losses of nodes and edges are known parameters, and when a node fails, the power losses of the failed node and the edges connected to it take the value of 0 when substituted into any of the above formulas for calculation. k0, k1, and k3 are randomly generated values and need to be continuously optimized and adjusted during the backpropagation process of the model.

[0117] 3. Model Training and Fine-tuning

[0118] The goal of this stage is to train a graph large model so that it can accurately predict the on-off attributes of nodes and edges, and fine-tune the model under specific power scenarios to improve its adaptability.

[0119] 3.1 Select Optimizer

[0120] Use the Adam optimizer with the learning rate set to 0.01. Among them, a learning rate scheduler (such as ReduceLROnPlateau) can be used to dynamically adjust the learning rate.

[0121] 3.2 Training Process

[0122] 1) Data Loading: Load the training set data into the graph data structure;

[0123] 2) Forward propagation: Calculate and update the on / off attributes of nodes and edges according to the rules formulated in Step 2-3 of the GNN model above.

[0124] 3) Calculate the loss according to the loss function designed in Step 3.1 above to obtain the total loss of the model.

[0125] 4) Perform backpropagation based on the total loss: that is, update the model parameters.

[0126] 5) Evaluate the model performance on the validation set and adjust the hyperparameters.

[0127] 3.4 Fine-tuning

[0128] Use the GNN model trained in Step 3.3 to perform fine-tuning in a specific power scenario. The specific steps are as follows:

[0129] 1) Prepare data for a specific scenario

[0130] First, it is necessary to collect data related to a specific power scenario, which includes node features, edge information, and on / off attribute labels of nodes and edges. Then, preprocess the collected data (including steps such as data cleaning, feature engineering, and label encoding) to ensure that the format, quality, and consistency of the data meet the requirements of the model input.

[0131] 2) Load the pre-trained model

[0132] Load the pre-trained weights from the GNN model trained in Step 3.3 for subsequent fine-tuning.

[0133] 3) Set the fine-tuning parameters

[0134] Select a smaller learning rate for fine-tuning, such as 0.001; select an appropriate batch size for training according to the computing resources and memory limitations;

[0135] 4) Perform fine-tuning

[0136] Use the data of the specific scenario to train the model and monitor the changes in the loss function and evaluation metrics. During the fine-tuning process, the model weights can be saved regularly to restore the best model when needed.

[0137] In addition, to prevent overfitting, an early stopping strategy can be adopted. When the performance on the validation set no longer improves, stop the training process.

[0138] 5) Evaluation and verification

[0139] Run the evaluation script on the fine-tuned model, and use the test set of specific scenario data to evaluate the performance of the model. This includes calculating metrics such as accuracy, recall, and F1 score. Analyze the evaluation results to understand the performance of the model in a specific scenario, and adjust the fine-tuning strategy as needed.

[0140] Thus, through the above steps, the construction and fine-tuning of the large model of the power grid topology map are completed, and a large model of the power grid topology map that can adapt to complex power grid scenarios and has the ability to analyze the power grid topology structure is obtained.

[0141] When a fault occurs in a certain section of the power system, the power grid operator can input the power grid topology map including the node fault into the model, and the model will output the power grid topology map with updated on-off attributes. After that, a graph visualization tool (such as NetworkX, PyVis) can be used to display the updated power grid topology map, so that the power grid operator can more intuitively understand the power grid state. Specifically, by viewing the transfer supply attributes in the updated power grid topology map, the power grid operator can quickly determine the candidate transfer supply points and make a final choice based on the on-off attributes of the candidate transfer supply points. For example, when the on-off attribute of a certain candidate transfer supply point is 1, it means that the node is determined to have the transfer supply ability or is suitable as a transfer supply point after power loss calculation, so it can be preferentially considered as the final power load transfer supply point.

[0142] In this way, it effectively assists the power grid operator in selecting the power load transfer supply point, not only improving the work efficiency but also ensuring the accuracy of the selection, thus ensuring the reliable and stable operation of the power system.

[0143] Embodiment 2

[0144] Based on the same inventive concept, the present invention also provides a load transfer configuration device based on the large model of the power grid topology map. The load transfer configuration device based on the large model of the power grid topology map includes:

[0145] An analysis module for taking the power grid topology map of the fault area as the input of a pre-trained graph neural network model and obtaining the adjusted power grid topology map output by the pre-trained graph neural network model;

[0146] A configuration module for performing power load transfer configuration on the power grid in the fault area by using the adjusted power grid topology map.

[0147] Preferably, the edges of the power grid topology map represent the physical connection relationships between nodes. The attributes of the edges of the power grid topology map include: connection relationship, on-off attribute, and power loss. The on-off attribute includes: conduction and disconnection; the attributes of the nodes of the power grid topology map include: node type, device parameters, fault attribute, transfer supply attribute, and on-off attribute. The node types include: source node, intermediate node, and load node. The device parameters include: power capacity and power loss. The fault attribute includes: fault and normal; the transfer supply attribute includes: candidate transfer supply point and non-candidate transfer supply point. The on-off attribute includes conduction and disconnection.

[0148] Further, the training process of the pre-trained graph neural network model includes:

[0149] Constructing training data by using a power grid topology map with node faults and its corresponding power grid topology map for repairing the node faults;

[0150] Training an initial graph neural network model by using the training data to obtain the pre-trained graph neural network model.

[0151] Further, during the process of training the initial graph neural network model by using the training data, an attribute update rule is embedded in the forward propagation process of the model to update the on-off attributes of each node and edge in the input power grid topology map.

[0152] Further, the attribute update rule includes: for a node with multiple child nodes, when one of its child nodes fails, if the conduction probability of the edge between it and its non-failed child nodes exceeds 0.5, then adjust the on-off attribute of this edge to conduction, otherwise, adjust the on-off attribute of this edge to disconnection; where the conduction probability of the edge is as follows:

[0153]

[0154] In the above formula, D is the conduction probability of the edge, e is the natural constant, z is the conduction coefficient, and the conduction coefficient is as follows:

[0155]

[0156] In the above formula, k0 is a constant coefficient, x is the coefficient corresponding to the faulty child node, u is the power loss of the faulty child node, k i is the coefficient corresponding to the i-th non-faulty child node, P i is the power loss of the i-th non-faulty child node, and I is the total number of non-faulty child nodes.

[0157] Further, the attribute update rule includes: for a node with only one child node, if the conduction probability of the edge between it and its child node exceeds 0.5, then adjust the on-off attribute of this edge to conduction; otherwise, adjust the on-off attribute of this edge to disconnection; where the conduction probability of the edge is as follows:

[0158]

[0159] In the above formula, D is the conduction probability of the edge, e is the natural constant, z is the conduction coefficient, and the conduction coefficient is as follows:

[0160] Z = k0 + mn

[0161] In the above formula, k0 is the constant coefficient, m is the coefficient corresponding to the node with only one child node, and n is the power loss of the node with only one child node.

[0162] Further, the attribute update rule includes: for a node with only one parent node, if the conduction probability of this node exceeds 0.5, then adjust the on-off attribute of this node to conduction; otherwise, adjust the on-off attribute of this node to disconnection; where the conduction probability of the edge is as follows:

[0163]

[0164] In the above formula, D is the conduction probability of the edge, e is the natural constant, z is the conduction coefficient, and the conduction coefficient is as follows:

[0165] Z = k0 + we

[0166] In the above formula, k0 is the constant coefficient, w is the coefficient corresponding to the edge between the node with only one parent node and its parent node, and e is the power loss of the edge between the node with only one parent node and its parent node.

[0167] Further, the attribute update rule includes: for a node with multiple parent nodes, if the conduction probability of this node exceeds 0.5, then adjust the on-off attribute of this node to conduction; otherwise, adjust the on-off attribute of this node to disconnection; where the conduction probability of the edge is as follows:

[0168]

[0169] In the above formula, D is the conduction probability of the edge, e is the natural constant, z is the conduction coefficient, and the conduction coefficient is as follows:

[0170]

[0171] In the above formula, k0 is the constant coefficient, k jis the coefficient corresponding to the edge between the node with multiple parent nodes and its j-th parent node, P j is the power loss of the edge between the node with multiple parent nodes and its j-th parent node, and J is the total number of edges between the node with multiple parent nodes and its parent nodes.

[0172] Furthermore, during the process of training the initial graph neural network model using the training data, the loss function is as follows:

[0173] L total = αL node + βL edge

[0174]

[0175] In the above formula, L total is the total loss function value, α is the first coefficient, and L node is the first loss function value, β is the second coefficient, and L edge is the second loss function value, N is the total number of nodes, and Y n is the true label of the on / off attribute of node n, is the predicted label of the on / off attribute of node n, E is the total number of edges, and Y e is the true label of the on / off attribute of edge e, is the predicted label of the on / off attribute of edge e, and α + β = 1.

[0176] Embodiment 3

[0177] Based on the same inventive concept, the present invention also provides a computer device, which includes a processor and a memory. The memory is used to store a computer program, and the computer program includes program instructions. The processor is used to execute the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions in the computer storage medium to implement the corresponding method flow or corresponding function, so as to implement the steps of a load transfer configuration method based on a large power grid topology graph model in the above embodiment.

[0178] Example 4

[0179] Based on the same inventive concept, the present invention further provides a storage medium, specifically a computer-readable storage medium (Memory). The computer-readable storage medium is a memory device in a computer device, used to store programs and data. It can be understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and, of course, the extended storage medium supported by the computer device. The computer-readable storage medium provides a storage space, which stores the operating system of the terminal. And, in this storage space, there is also stored one or more instructions suitable for being loaded and executed by the processor. These instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. One or more instructions stored in the computer-readable storage medium can be loaded and executed by the processor to implement the steps of a load transfer configuration method based on a large power grid topology model in the above embodiments.

[0180] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.

[0181] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the specified functions in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0182] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including an instruction means that implements the functions specified in one or more of the processes and / or blocks Figure 1 in one or more of the processes and / or blocks Figure 1 specified in the function.

[0183] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in one or more of the processes and / or blocks Figure 1 in one or more of the processes and / or blocks Figure 1 specified in the function.

[0184] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications or equivalent replacements can still be made to the specific embodiments of the present invention. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention shall be covered by the protection scope of the claims of the present invention.

Claims

1. A load transfer configuration method based on a large model of the power grid topology diagram, characterized in that, The method includes: Taking the power grid topology map of the fault area as the input of a pre-trained graph neural network model, and obtaining an adjusted power grid topology map output by the pre-trained graph neural network model; Using the adjusted power grid topology map to configure the power load transfer of the power grid in the fault area.

2. The method according to claim 1, wherein The edges of the power grid topology map represent the physical connection relationships between nodes. The attributes of the edges of the power grid topology map include: connection relationship, on-off attribute, and power loss. The on-off attribute includes: conduction and disconnection; the attributes of the nodes of the power grid topology map include: node type, device parameters, fault attributes, transfer attributes, and on-off attributes. The node types include: source node, intermediate node, and load node. The device parameters include: power capacity and power loss. The fault attributes include: fault and normal; the transfer attributes include: candidate transfer points and non-candidate transfer points. The on-off attributes include conduction and disconnection.

3. The method according to claim 2, wherein The training process of the pre-trained graph neural network model includes: Constructing training data by using the power grid topology map with node faults and its corresponding power grid topology map for repairing the node faults; Training an initial graph neural network model by using the training data to obtain the pre-trained graph neural network model.

4. The method according to claim 3, characterized in that, During the process of training the initial graph neural network model by using the training data, an attribute update rule is embedded in the forward propagation process of the model to update the on-off attributes of each node and edge in the input power grid topology map.

5. The method according to claim 4, wherein The attribute update rule includes: for a node with multiple child nodes, when one of its child nodes has a fault, if the conduction probability of the edge between it and its non-faulty child nodes exceeds 0.5, then adjust the on-off attribute of this edge to conduction, otherwise, adjust the on-off attribute of this edge to disconnection; where the conduction probability of the edge is as follows: In the above formula, D is the conduction probability of the edge, e is the natural constant, and z is the conduction coefficient. The conduction coefficient is as follows: In the above formula, k0 is a constant coefficient, x is the coefficient corresponding to the faulty child node, y is the power loss of the faulty child node, k i is the coefficient corresponding to the i-th non-faulty child node, P i is the power loss of the i-th non-faulty child node, and I is the total number of non-faulty child nodes.

6. The method according to claim 4, wherein The attribute update rule includes: for a node with only one child node, if the conduction probability of the edge between it and its child node exceeds 0.5, then adjust the on-off attribute of this edge to conduction, otherwise, adjust the on-off attribute of this edge to disconnection; where the conduction probability of the edge is as follows: In the above formula, D is the conduction probability of the edge, e is the natural constant, and z is the conduction coefficient. The conduction coefficient is as follows: Z = k0 + mn In the above formula, k0 is a constant coefficient, m is the coefficient corresponding to the node with only one child node, and n is the power loss of the node with only one child node.

7. The method according to claim 4, wherein The attribute update rule includes: for a node with only one parent node, if the conduction probability of this node exceeds 0.5, then adjust the on-off attribute of this node to conduction, otherwise, adjust the on-off attribute of this node to disconnection; where the conduction probability of the edge is as follows: In the above formula, D is the conduction probability of the edge, e is the natural constant, and z is the conduction coefficient. The conduction coefficient is as follows: Z = k0 + we In the above formula, k0 is a constant coefficient, w is the coefficient corresponding to the edge between the node with only one parent node and its parent node, and e is the power loss of the edge between the node with only one parent node and its parent node.

8. The method according to claim 4, characterized in that The attribute update rule includes: for a node with multiple parent nodes, if the conduction probability of the node exceeds 0.5, the on-off attribute of the node is adjusted to on; otherwise, the on-off attribute of the node is adjusted to off; where the conduction probability of the edge is as follows: In the above formula, D is the conduction probability of the edge, e is the natural constant, and z is the conduction coefficient, and the conduction coefficient is as follows: In the above formula, k o is a constant coefficient, and k j is the coefficient corresponding to the edge between the node with multiple parent nodes and its j-th parent node. P j is the power loss of the edge between the node with multiple parent nodes and its j-th parent node, and J is the total number of edges between the node with multiple parent nodes and its parent nodes.

9. The method according to claim 3, wherein During the process of training the initial graph neural network model using the training data, the loss function is as follows: L total = αL node + βL edge In the above formula, L total is the total loss function value, α is the first coefficient, L node is the first loss function value, β is the second coefficient, L edge is the second loss function value, N is the total number of nodes, Y n is the true label of the on / off attribute of node n, is the predicted label of the on / off attribute of node n, E is the total number of edges, Y e is the true label of the on / off attribute of edge e, is the predicted label of the on / off attribute of edge e, α + β = 1.

10. An apparatus for a load transfer configuration method based on the large model of the power grid topology diagram according to any one of claims 1-9, characterized in that, The device includes: An analysis module, configured to use the power grid topology map of the fault area as the input of a pre-trained graph neural network model, and obtain an adjusted power grid topology map output by the pre-trained graph neural network model; A configuration module, configured to perform power load transfer configuration on the power grid in the fault area by using the adjusted power grid topology map.

Images