CN-121997039-A - Power distribution network topology detection method and system based on double-graph structure

CN121997039ACN 121997039 ACN121997039 ACN 121997039ACN-121997039-A

Abstract

The invention provides a power distribution network topology detection method and system based on a double-graph structure, wherein the method comprises the steps of acquiring measurement data of each node of a power distribution network at intervals of a first preset time, and extracting a first feature and a second feature from the measurement data; the method comprises the steps of calculating comprehensive electrical distance indexes according to first features and second features, constructing a comprehensive distance matrix according to electrical distances, constructing a node attribute similarity graph according to the comprehensive distance matrix, inputting the node attribute similarity graph into a pre-trained dual-graph structure diagram neural network model, generating node embedded vectors, performing supervision training on the node embedded vectors through a structure perception loss function by using an actual topological graph, calculating connection scores of all potential sides, judging whether all sides are connected according to the connection scores, and outputting the current topological structure of the power distribution network according to a judging result. The invention can realize the power distribution network topology inference without relying on priori topology structure information.

Inventors

CAI MULIANG
LI YINGZHENG
CUI MINGJIAN
ZHONG XIAO
QIU HAICHAO
LAI XINHUI
LIU BEI

Assignees

国网江西省电力有限公司电力科学研究院

Dates

Publication Date: 20260508
Application Date: 20251219

Claims (10)

1. The utility model provides a distribution network topology detection method based on double-graph structure, which is characterized in that the method comprises the following steps: Acquiring measurement data of each node of the power distribution network at intervals of a first preset time, and extracting a first characteristic and a second characteristic from the measurement data; Calculating a comprehensive electrical distance index according to the first feature and the second feature, constructing a comprehensive distance matrix according to the electrical distance, and constructing a node attribute similarity graph according to the comprehensive distance matrix; inputting the node attribute similarity graph into a pre-trained double-graph structure graph neural network model, generating a node embedding vector, and performing supervised training on the node embedding vector through a structure perception loss function by using an actual topological graph to align the node embedding with the actual topology so as to update graph neural network parameters; and calculating the connection scores of all the potential sides, judging whether all the sides are connected according to the connection scores, and outputting the current topological structure of the power distribution network according to the judgment result.
2. The method for detecting the topology of the power distribution network based on the double-graph structure according to claim 1, wherein the steps of obtaining measurement data of each node of the power distribution network at intervals of a first preset time and extracting the first feature and the second feature from the measurement data comprise: The measured data comprise voltage amplitude and voltage phase angle, and the measured data of the nodes of the power distribution network are defined as a time sequence matrix , wherein, The time window length is N, the number of nodes is N, and F is the characteristic dimension; two nodes are calculated according to the following formula And In the time window Inner pearson coefficients: ; the normalized Euclidean distance between nodes is calculated according to the following formula: ; Wherein, the Is a node And The pearson coefficient between them, 、 The measured value of the f characteristic of the node i at the time t and the measured value of the f characteristic of the node j at the time t are respectively, For the data mean value of node i in the current time window, For the data mean of node j in the current time window, Is a node And The normalized euclidean distance between them, Is a node And The original euclidean distance between them, Is the maximum of the original euclidean distances.
3. The method for detecting a topology of a power distribution network based on a double-graph structure according to claim 2, wherein the step of calculating a comprehensive electrical distance index from the first feature and the second feature and constructing a comprehensive distance matrix from the electrical distances comprises: The integrated electrical distance is calculated according to the following formula: ; Wherein, the Is a node And The integrated electrical distance between the two electrodes, Is a regulatory factor; And constructing a fully-connected comprehensive distance matrix based on the comprehensive electrical distances of all the node pairs.
4. A method for detecting a topology of a power distribution network based on a dual graph structure as recited in claim 3, wherein said step of constructing a node attribute similarity graph from said integrated distance matrix comprises: Sequencing all potential edges in the comprehensive distance matrix from small to large according to comprehensive electrical distances, initializing a graph containing N nodes, and setting a maximum allowable degree threshold value of each node; Iterative selection of ordered edges, when nodes And After the node is not connected in the current graph and the edge is added, the node And If the degree of the edge does not exceed the maximum allowable degree threshold, adding the edge to the graph; When the number of edges added in the graph reaches And stopping iteration when the bar or the edges meeting the conditions cannot be continuously added, and outputting the node attribute similarity graph.
5. The method for detecting a topology of a power distribution network based on a dual graph structure according to claim 4, wherein the step of inputting the node attribute similarity graph into a pre-trained dual graph structure graph neural network model, generating a node embedding vector comprises: the attention weight of the neighbor node is calculated according to the following formula: ; Wherein, the Is a node And The weight of the attention between them, As a distance decay coefficient that can be learned, As a set of neighbors, Is a node And A comprehensive electrical distance therebetween; generating the aggregation characteristics of the current layer according to the attention weight And node embedding is carried out: ; ; Wherein, the 、 Respectively embedding vectors for the nodes of the adjacent node j in the upper layer and the nodes of the node j in the k+1th layer, To activate the function (RECTIFIED LINEAR Unit), Is the trainable weight matrix of the k+1 layer, Is a splicing operation.
6. The method for detecting the topology of a power distribution network based on a dual graph structure according to claim 5, wherein the step of performing supervised training on the node embedding vector by using the actual topology graph through the structure-aware loss function to align the node embedding with the actual topology to update the parameters of the graph neural network comprises: The loss function is constructed according to the following formula: ; Wherein, the As a function of the composite loss, 、 Are all the balance coefficients of the two-dimensional space, For the binary cross entropy classification loss, In order to be based on the ordering penalty of the interval, For the loss of the physical constraint of the voltage phase angle, For the edge between node i and node j Is used to determine the connection probability prediction value of (1), 、 Respectively nodes And Is used for the voltage phase angle of (a), A maximum phase angle difference threshold allowed for proper operation of the distribution network, As a true label of the edge, For the connection probability predictor of any edge e, For the interval threshold value, Is the predictive score of the positive sample edge, Is the predictive score of the negative sample edge.
7. The method for detecting the topology of the power distribution network based on the double-graph structure according to claim 6, wherein the steps of calculating the connection scores of all the potential sides, judging whether the sides are connected according to the connection scores, and outputting the current topology of the power distribution network according to the judgment result comprise: Setting a connectivity probability threshold Each potential edge The connection score of (2) And threshold value Comparing one by one; If it is Judging the side to be in a connected/closed state, if so The edge is determined to be in an open/off state.
8. A power distribution network topology detection system based on a double-graph structure, the system comprising: The measurement data acquisition module is used for acquiring measurement data of each node of the power distribution network at intervals of a first preset time, and extracting a first characteristic and a second characteristic from the measurement data; The node diagram construction module is used for calculating a comprehensive electrical distance index according to the first feature and the second feature, constructing a comprehensive distance matrix according to the electrical distance, and constructing a node attribute similarity diagram according to the comprehensive distance matrix; The parameter updating module is used for inputting the node attribute similarity graph into a pre-trained double-graph structure chart neural network model, generating a node embedding vector, and performing supervision training on the node embedding vector through a structure perception loss function by using an actual topological graph to enable node embedding to be aligned with the actual topology so as to update the graph neural network parameters; And the communication detection module is used for calculating the connection scores of all the potential sides, judging whether all the sides are communicated according to the connection scores, and outputting the current topological structure of the power distribution network according to the judgment result.
9. A storage medium storing one or more programs which when executed by a processor implement the dual graph structure based power distribution network topology detection method of any of claims 1-7.
10. An electronic device comprising a memory and a processor, wherein: The memory is used for storing a computer program; The processor is configured to implement the power distribution network topology detection method based on the dual graph structure according to any one of claims 1 to 7 when executing the computer program stored on the memory.

Description

Power distribution network topology detection method and system based on double-graph structure Technical Field The invention relates to the technical field of topology detection, in particular to a power distribution network topology detection method and system based on a double-graph structure. Background The accurate and real-time power distribution network topological structure is the basis for high-level application efficient operation such as support system state estimation, fault positioning, reactive power optimization, protection coordination and the like. In power transmission networks, the topology is generally directly accessible through a data acquisition and monitoring control system. However, in the power distribution network, due to the large scale of the network, the large number of switches and insufficient communication coverage, and the frequent network frame reconstruction and the access of distributed energy sources, the operators have difficulty in accurately grasping the actual connection relationship of the power grid in real time. The state sensing, control capability and operation efficiency of the power distribution network are seriously affected by the absence or hysteresis of the switch state information. To address this challenge, a variety of power distribution network topology detection methods have been proposed in the industry, which can be largely classified into a state estimation-based method and a data driving method. The method based on state estimation needs to pre-establish a topology base containing all possible topologies, perform state estimation on each topology in the base by using real-time measurement data, and finally identify the real topology according to the criterion of minimum residual error and the like. The accuracy of the method is seriously dependent on the accuracy of the parameters of the power distribution network, the calculation complexity of the method is increased sharply along with the increase of the scale of the topology library, and the method is difficult to realize real-time application in a large-scale power distribution network. The data driving method, especially the deep learning method, predicts the system topology through the historical data training model, and avoids the need of constructing a complete topology base. However, the conventional deep learning model (such as a fully-connected neural network) can only utilize the attribute information (such as voltage and phase angle) of the nodes, but cannot effectively utilize the inherent graph structure information of the power distribution network, so that the generalization capability and the detection precision of the deep learning model are limited. The graph neural network is used as a novel deep learning model capable of processing node attribute and graph structure information simultaneously, and provides potential for solving the problem. However, existing topology detection methods based on graph neural networks rely heavily on a known, as input, baseline graph structure (e.g., partially known line connections) during both training and reasoning phases. This premise of "predictive topology" is often not met in a real distribution network because the operational objective is to detect this unknown topology. Therefore, the existing graph neural network model is difficult to apply to a scene without any priori structural information, and practical application of the graph neural network model in power distribution network topology detection is limited. Disclosure of Invention The invention aims to provide a power distribution network topology detection method and system based on a double-graph structure, and aims to solve the problem that the traditional power distribution network topology detection technology is limited in use due to dependence on priori topology structure information. In a first aspect, the present invention provides a method for detecting a topology of a power distribution network based on a double-graph structure, the method comprising: Acquiring measurement data of each node of the power distribution network at intervals of a first preset time, and extracting a first characteristic and a second characteristic from the measurement data; Calculating a comprehensive electrical distance index according to the first feature and the second feature, constructing a comprehensive distance matrix according to the electrical distance, and constructing a node attribute similarity graph according to the comprehensive distance matrix; inputting the node attribute similarity graph into a pre-trained double-graph structure graph neural network model, generating a node embedding vector, and performing supervised training on the node embedding vector through a structure perception loss function by using an actual topological graph to align the node embedding with the actual topology so as to update graph neural network parameters; and calculating the connection scores of