CN-121996948-A - Method for realizing fault prediction based on power distribution network space diagram

CN121996948ACN 121996948 ACN121996948 ACN 121996948ACN-121996948-A

Abstract

The invention discloses a method for realizing fault prediction based on a power distribution network space diagram, which comprises the steps of collecting original multi-source data to generate multi-source data, carrying out data format conversion on the multi-source data to generate multi-dimensional time sequence data, wherein the data format is diagram structure data and time sequence data, constructing a power distribution network space diagram model to obtain space feature vectors, the power distribution network space diagram model is represented by an adjacent matrix A and a node feature matrix X, the adjacent matrix A represents space features among nodes, the node feature matrix X represents node features, the space feature vectors are used for judging fault risk values of all the nodes, generating a power distribution network fault space diagram by combining the power distribution network space diagram, fusing the space features and time features, and realizing fault prediction based on the power distribution network space diagram. According to the technical scheme, the fault area can be accurately positioned, positioning errors are reduced, the high-risk area identification accuracy is improved, the fault prediction accuracy is improved, and the method is suitable for dynamic adjustment of geographical environment and meteorological characteristics of the area.

Inventors

ZHANG YONG
YANG JIANHUA
LIU DANDAN
Wei Siti
XU JING
ZHAO JIAN
CAO SHUANGHE
WANG DEHONG
LIU XIANGBIN
WANG HONGSHENG

Assignees

中国电建集团贵州电力设计研究院有限公司

Dates

Publication Date: 20260508
Application Date: 20251230

Claims (9)

1. The method for realizing fault prediction based on the space diagram of the power distribution network is characterized by comprising the following steps of: The method comprises the steps of collecting original multi-source data, preprocessing the original multi-source data to generate multi-source data, wherein the multi-source data comprises power grid topology data, equipment position data, real-time sensing monitoring data, meteorological and geographic data and historical fault data; The data format of the multi-dimensional time sequence data is graph structure data and time sequence data, wherein the graph structure data consists of nodes and edges with attributes, the nodes are equipment/buses, the edges are line connection, and the attributes of the edges comprise line parameters and environmental characteristics; constructing a power distribution network space diagram model to obtain a space feature vector, wherein the power distribution network space diagram model is represented by an adjacent matrix A and a node feature matrix X, the adjacent matrix A expresses and expresses the space feature associated with each node, and the node feature matrix X expresses the node feature; Generating a power distribution network fault space distribution diagram by combining the power distribution network space diagram; and the space features and the time features are fused, and the fault prediction of the nodes is realized based on the space diagram of the power distribution network.
2. The method for implementing fault prediction based on a space diagram of a power distribution network according to claim 1, wherein the constructing a space diagram model of the power distribution network to obtain a space feature vector comprises the following steps: Constructing a two-dimensional space model of the power distribution network, and overlapping part of contents of the multi-source data to generate a basic space map of the power distribution network; on the basis of the two-dimensional space model of the power distribution network, a space diagram model of the power distribution network is constructed according to structural data, wherein the structural data refers to data related to a power grid line; And (3) learning the spatial dependency relationship of the nodes in the power distribution network space diagram model by using the GCN, and extracting local and global spatial feature vectors of the nodes as the spatial dependency relationship to serve as the spatial feature vectors.
3. The method for realizing fault prediction based on the power distribution network space diagram according to claim 1, wherein the node characteristics comprise equipment self attribute, regional environment parameters where the node is located and historical fault attribute corresponding to the node; The spatial features comprise topological association features, spatial proximity influence features and regional commonality features, wherein the topological association features comprise connection strength of nodes and adjacent equipment and fault risk conduction coefficients of the adjacent nodes, the spatial proximity influence features comprise equipment states of peripheral nodes and spatial consistency of regional environments, and the regional commonality features comprise equipment aging degree average values, spatial coverage of regional weather disasters and fault susceptibility clusters caused by topography in the same topological partition.
4. The method for realizing fault prediction based on the power distribution network space diagram according to claim 1, wherein the generating of the power distribution network fault space diagram is to fuse a space risk level with the power distribution network space diagram, and label regional risks by adopting different colors to generate the power distribution network fault space diagram; and receiving real-time sensing monitoring data, weather and geographic data in real time, and updating the space risk level and the distribution network fault space distribution diagram according to a certain period.
5. The method for realizing fault prediction based on the power distribution network space diagram according to claim 1, wherein the fault prediction of the nodes is that the probability of occurrence of faults of each node/line within 1-24 hours in the future is output through a GCN-BiLSTM integrated model.
6. The method for implementing fault prediction based on a power distribution network space diagram according to claim 5, wherein constructing the GCN-BiLSTM integration model comprises the steps of: Extracting time stamps and corresponding meteorological data from the multidimensional time sequence data as feature sets, extracting the time stamps and corresponding historical fault attributes as tag sets, and forming a fault prediction data set by the feature sets and the tag sets; Defining a GCN-BiLSTM integrated model, wherein the GCN-BiLSTM integrated model comprises an input layer, a GCN feature enhancement layer, a BiLSTM time sequence modeling layer and an output layer; Dividing a fault prediction data set into a training set, a verification set and a test set, wherein the training set is used for parameter learning of the GCN-BiLSTM integrated model, the verification set is used for super-parameter tuning of the GCN-BiLSTM integrated model, and the test set is used for performance evaluation of the GCN-BiLSTM integrated model; And (5) performing model evaluation and optimization to complete GCN-BiLSTM integrated model construction.
7. The method for realizing fault prediction based on the power distribution network space diagram according to claim 6, wherein the input layer is used for splicing the space feature vector and the multidimensional time sequence data to form a space-time combined input matrix, the dimensions of the space-time combined input matrix comprise a time step length T, a node number N and a feature number F, wherein T=24, F is more than or equal to 12, and the feature corresponding to the feature number F covers node features and space features.
8. The method for realizing fault prediction based on the power distribution network space diagram according to claim 1, wherein the preprocessing comprises outlier detection and rejection, missing value filling and data normalization; The abnormal value detection and rejection means is used for setting an abnormal threshold value by combining the data characteristics of the distribution network, identifying and rejecting the sensing data exceeding the abnormal threshold value and the topology data with logic contradiction, the missing value filling means is used for filling space correlation and historical time sequence data of adjacent nodes by adopting an interpolation method based on graph attention aiming at space-time missing data, and the data normalization means is used for mapping the data to a [ -1,1] interval by adopting Z-Score normalization processing on the digital data.
9. The method for realizing fault prediction based on the space diagram of the power distribution network according to claim 1, wherein the multi-source data are classified when the data format is converted, the data associated with the power grid line are used as diagram structure data, and the data associated with the time stamp are used as time sequence data.

Description

Method for realizing fault prediction based on power distribution network space diagram Technical Field The invention relates to the crossing field of operation and maintenance of a power system and artificial intelligence, in particular to a method for realizing fault prediction based on a power distribution network space diagram. Background The distribution network is an end link of the power system, directly faces to users for power supply, and the running state of the distribution network is closely related to social production and life. The distribution network has the characteristics of wide line distribution, complex geographical environment and various equipment types, is easily influenced by multiple factors such as extreme weather (storm, thunder and lightning, icing), topography conditions (mountain land and vegetation coverage), equipment aging and the like, has frequent faults and presents obvious spatial heterogeneity and time randomness. If the fault happens, if the fault area cannot be positioned quickly and early warning is carried out in advance, the power failure range is enlarged, the rush repair efficiency is low, and serious economic loss and social influence are caused. According to the method, dynamic fusion of real-time equipment states and environmental factors is lacking, a fault space distribution rule and a potential risk area cannot be intuitively presented, a drawing result is lagged and the practicability is limited, on the other hand, the fault prediction method focuses on time sequence analysis of a single time dimension in multiple ways (such as LSTM only captures time dependence), spatial association characteristics of power distribution network topology are ignored, and fault evolution rules under 'space-time-environment' multidimensional coupling are difficult to effectively mine. Meanwhile, the problems of format isomerism and complex relevance of multi-source data (power grid topology, sensing data and meteorological geographic data) exist, the prior art lacks an efficient fusion mechanism, so that the data value is not fully mined, and the prediction precision and drawing accuracy are difficult to meet the actual operation and maintenance requirements. Preliminary attempts of combining GIS and machine learning have been made in the prior art, but the method has obvious defects that firstly, the spatial characteristics are not extracted sufficiently, a special model is not designed aiming at the graph structural characteristics of the power distribution network topology, the spatial dependence relationship among devices cannot be effectively captured, secondly, the drawing and prediction are disjointed, the spatial distribution visualization is not dynamically linked with the time sequence prediction result, the whole process operation and maintenance of early warning-positioning-disposal is difficult to support, thirdly, the dynamic adaptability is poor, the drawing result and the prediction model cannot be updated according to real-time data, and the rapidly-changing operation environment is difficult to deal with. Therefore, there is a need in the industry for an intelligent technology that can integrate multi-source heterogeneous data, deeply mine spatio-temporal coupling features, and realize "spatial visualization-temporal prediction" integration. Disclosure of Invention In order to achieve the above purpose, the present application provides a method for implementing fault prediction based on a space diagram of a power distribution network, comprising the following steps: The method comprises the steps of collecting original multi-source data, preprocessing the original multi-source data to generate multi-source data, wherein the multi-source data comprises power grid topology data, equipment position data, real-time sensing monitoring data, meteorological and geographic data and historical fault data; The data format of the multi-dimensional time sequence data is graph structure data and time sequence data, wherein the graph structure data consists of nodes and edges with attributes, the nodes are equipment/buses, the edges are line connection, and the attributes of the edges comprise line parameters and environmental characteristics; constructing a power distribution network space diagram model to obtain a space feature vector, wherein the power distribution network space diagram model is represented by an adjacent matrix A and a node feature matrix X, the adjacent matrix A expresses and expresses the space feature associated with each node, and the node feature matrix X expresses the node feature; Generating a power distribution network fault space distribution diagram by combining the power distribution network space diagram; and the space features and the time features are fused, and the fault prediction of the nodes is realized based on the space diagram of the power distribution network. The method for constructing the space map model of the power distrib