CN-122020098-A - Power distribution network topology intelligent identification method based on deep convolution network

CN122020098ACN 122020098 ACN122020098 ACN 122020098ACN-122020098-A

Abstract

The invention discloses an intelligent identification method of a power distribution network topology based on a deep convolutional network, which comprises the steps of collecting electrical measurement data of a plurality of nodes in the power distribution network; the method for constructing the deep convolution network model comprises the steps of introducing a physical confidence gating factor into an attention mechanism of the network model to modulate attention weights between any two nodes so as to inhibit false correlation generated based on accidental similarity of data, calculating the discrete degree of dynamic virtual impedance in a time window to obtain a virtual impedance jitter index, and inputting the virtual impedance jitter index into a preset nonlinear gating function to obtain the physical confidence gating factor. According to the scheme provided by the invention, the accuracy and the robustness of the topology identification of the power distribution network are effectively improved.

Inventors

LUO HUAFENG
CAI JUNYU
CHEN XIN
HAN JIAJIA
YAO YING
LIU RONGHAO
WANG ZIXIANG
WANG LIUWANG

Assignees

国网浙江省电力有限公司电力科学研究院

Dates

Publication Date: 20260512
Application Date: 20260410

Claims (10)

1. The intelligent identification method for the topology of the power distribution network based on the deep convolution network is characterized by comprising the following steps of: Collecting electrical measurement data of a plurality of nodes in the power distribution network, wherein the electrical measurement data comprise node voltage and injection power; The method for constructing the depth convolution network model comprises the steps of introducing a physical confidence gating factor into an attention mechanism of the network model to modulate attention weights between any two nodes so as to inhibit false correlation generated based on accidental similarity of data, wherein the method for calculating the physical confidence gating factor comprises the following steps: Acquiring the ratio of the voltage difference of two nodes in a preset time window to the injection power of one node so as to represent the dynamic virtual impedance between the two nodes; calculating the discrete degree of the dynamic virtual impedance in a time window to obtain a virtual impedance jitter index, and And inputting the virtual impedance jitter index into a preset nonlinear gating function to obtain the physical confidence gating factor, wherein the value of the physical confidence gating factor is inversely related to the value of the virtual impedance jitter index.
2. The intelligent identification method for the power distribution network topology based on the deep convolutional network according to claim 1, wherein in the construction method of the deep convolutional network model, the method further comprises the following steps: Performing data enhancement on the original training sample set to expand training data; the data enhancement method comprises at least one of the following: Performing data enhancement based on the generated countermeasure network, generating amplification data similar to the real data distribution; Data enhancement is performed based on continuous power flow calculations, and complementary samples are generated by simulating electrical data at different load levels and modes of operation.
3. A method of intelligent identification of a topology of a power distribution network based on a deep convolutional network according to claim 2, wherein said generating an enhancement of the countermeasure network comprises: constructing a generation countermeasure network model comprising a generator and a discriminator, wherein the generator receives random noise and outputs a simulation electric sample; through game training of the generator and the discriminator, the generator generates augmentation data which approximates to real distribution.
4. The intelligent identification method for the topology of the power distribution network based on the deep convolutional network according to claim 2, wherein the enhancement based on continuous power flow calculation comprises the following steps: under the condition of insufficient training sample quantity, simulating electric data under different load levels and running modes by using continuous tide calculation, and generating supplementary samples to cover more running scenes.
5. The intelligent identification method for the power distribution network topology based on the deep convolutional network according to claim 2, wherein the network structure of the deep convolutional network model sequentially comprises an input layer for receiving an electrical feature matrix, at least one convolutional layer for extracting local electrical features, a pooling layer for compressing feature dimensions and a full connection layer for mapping features to topology categories; The convolution layer performs convolution operation by utilizing a convolution check input matrix, extracts local electrical characteristics of the power distribution network, and introduces a nonlinear activation function to increase fitting capacity of a model; the pooling layer performs downsampling operation on the local electrical characteristics extracted by convolution, compresses characteristic dimensions, reduces calculation complexity and improves generalization capability of the model; The full connection layer flattens the multidimensional features, and the features are mapped to the sample marking space through the full connection structure.
6. The intelligent identification method for the power distribution network topology based on the deep convolutional network according to claim 1, wherein the nonlinear gating function is a Sigmoid variant function, the value of the physical confidence gating factor approaches to 1 when the virtual impedance jitter index is lower than a preset threshold value, and the value of the physical confidence gating factor approaches to 0 when the virtual impedance jitter index is higher than the preset threshold value.
7. The intelligent identification method for the power distribution network topology based on the deep convolutional network according to claim 1, wherein the calculation formula of the physical confidence gating factor is as follows: In the formula, Represents a physical confidence gating factor, and has a dimensionless coefficient with a value range of (0, 1), Represents an impedance jitter tolerance threshold derived based on historical operating data statistics, Representing the attenuation sensitivity coefficient, controlling the steepness of the nonlinear gating function, Representing the degree of dispersion in the physical consistency of the electrical connection between nodes i, j.
8. The intelligent identification method for the power distribution network topology based on the deep convolutional network according to claim 7, wherein the calculation formula of the attention weight is as follows: In the formula, Representing the normalized attention weights ultimately used for feature aggregation, And Respectively representing the high-dimensional feature vectors of the nodes i and j after feature transformation, As a matrix of weights that can be learned, A parameter vector representing the mechanism of attention, The activation function is represented as a function of the activation, The vector concatenation operation is represented by a vector, Representing a collection of nodes.
9. The intelligent identification method of the power distribution network topology based on the deep convolutional network according to claim 7, wherein a calculation formula of the discrete degree of the physical consistency of the electrical connection between the nodes is as follows: In the formula, Representing the degree of dispersion in the physical consistency of the electrical connection between nodes i, j, And The voltage vectors at node i and node j at time t are shown, And Active and reactive injection powers of time t node i are indicated respectively, Representing a small constant that prevents the denominator from being zero, Representing the mean value of the calculated virtual impedance modulus values within the time window T.
10. The intelligent identification method for the power distribution network topology based on the deep convolutional network according to claim 1, wherein the electrical measurement data are derived from a SCADA system, an AMI system or PMU equipment, and the construction step of the intelligent identification method for the power distribution network topology further comprises the steps of cleaning and semantically analyzing the collected unstructured text data.

Description

Power distribution network topology intelligent identification method based on deep convolution network Technical Field The invention relates to the technical field of power systems and automation thereof, in particular to an intelligent identification method for power distribution network topology based on a deep convolution network. Background The power distribution network is used as a terminal link of a power system connection user, and the accurate identification of the topological structure is a basis for carrying out state estimation, fault positioning and load transfer. With the access of a distributed power supply and the expansion of the scale of a power distribution network, the running mode of the network is changeable, and the topological structure is frequently switched. The existing topology identification method mainly has the following problems: 1. the traditional method relies on remote signaling/telemetry data of the SCADA system, and when the data is missing or has errors, the identification accuracy is reduced. 2. The calculation efficiency is low, the method based on mathematical optimization faces the problem of combined explosion when the number of nodes is increased, and the real-time requirement is difficult to meet. 3. Sample data starvation-machine learning based methods require a large number of training samples, but the sample data for faults or specific topologies in actual operation are sparse and single, resulting in insufficient model generalization capability. 4. The prior art has significant limitations in processing actual metrology data. The power distribution network measuring device (such as PMU and smart meter) often causes asynchronous time of uploaded voltage and tide data due to communication delay, clock drift and the like, and electromagnetic noise interference is often accompanied. In conventional attention mechanisms, weight calculation depends only on the similarity of feature vectors in numerical space (e.g., dot product or cosine similarity). When there is a timing misalignment in the data, two physically unconnected or weakly correlated nodes may exhibit occasional numerical synchronization (false correlation), resulting in the attention mechanism erroneously giving high weight, introducing noise features as a topological criterion, resulting in a significant decrease in the recognition accuracy of the model in noisy environments. Based on this, the problem of low accuracy of topology identification in environments with incomplete measurement data, large noise interference and unbalanced samples is needed to be solved. Disclosure of Invention In order to solve the technical problems of incomplete measurement data, large noise interference and low topology identification accuracy in a sample imbalance environment, the invention provides a power distribution network topology intelligent identification method based on a deep convolution network, so as to effectively improve the accuracy and robustness of power distribution network topology identification. The method comprises the steps of collecting electrical measurement data of a plurality of nodes in a power distribution network, including node voltage and injection power, inputting the electrical measurement data into a pre-trained deep convolution network model for topology identification, introducing a physical confidence gating factor into an attention mechanism of the network model to modulate attention weights between any two nodes so as to inhibit false correlation generated based on accidental similarity of data, wherein the calculation method of the physical confidence gating factor comprises the steps of obtaining the ratio of voltage difference of the two nodes in a preset time window to the injection power of one node so as to represent dynamic virtual impedance between the two nodes, calculating the discrete degree of the dynamic virtual impedance in the time window so as to obtain a virtual impedance jitter index, and inputting the virtual impedance jitter index into a preset nonlinear gating function so as to obtain the physical confidence gating factor, wherein the value of the physical confidence gating factor is inversely correlated with the value of the virtual impedance jitter index. Preferably, the method for constructing the deep convolutional network model further comprises the step of carrying out data enhancement on an original training sample set to expand training data, wherein the data enhancement method comprises at least one of the steps of carrying out data enhancement on the basis of generating an countermeasure network, generating amplification data similar to real data distribution, carrying out data enhancement on the basis of continuous tide calculation, and generating complementary samples by simulating electrical data under different load levels and running modes. Preferably, the generating countermeasure network-based enhancement comprises constructing a generating co