Search

CN-121683142-B - Distribution network topology graph real difference diagnosis method based on graph neural network

CN121683142BCN 121683142 BCN121683142 BCN 121683142BCN-121683142-B

Abstract

The invention belongs to the technical field of operation and maintenance of a power distribution network, and relates to a power distribution network topological graph real difference diagnosis method based on a graph neural network, which comprises the following steps of injecting an initial detection signal sequence into the power distribution network, collecting and integrating responses generated by a collaborative sensing terminal, generating a graph node original feature set, executing cross-validation, generating an enhanced initial topological connection feature graph, inputting the initial topological connection feature graph into a preset graph neural network model, and generating a topological diagnosis probability graph; according to the topology diagnosis probability map, a secondary detection signal sequence is generated, the secondary detection signal sequence is injected into the power distribution network, a focusing response data set is generated in an abnormal area, iterative diagnosis is carried out on the abnormal area based on the focusing response data set, and a final topology difference positioning report is generated, and the problem that the modern power grid is difficult to support the large-scale and normalized topology verification requirements under the operation and maintenance requirements of instantaneity, accuracy and economy is solved.

Inventors

  • LIU GUANGHUA

Assignees

  • 湖南力光信息技术有限公司

Dates

Publication Date
20260508
Application Date
20260205

Claims (10)

  1. 1. The utility model provides a distribution network topological graph real difference diagnosis method based on a graph neural network, which is characterized by comprising the following steps: S1, injecting an initial detection signal sequence comprising a time sequence code and a specific frequency component into a power distribution network, and exciting a plurality of cooperative sensing terminals distributed in the power distribution network to generate response, wherein the specific frequency component refers to a frequency which is an integer multiple of the power frequency avoiding the power distribution network; S2, acquiring and integrating responses generated by the collaborative sensing terminal to generate a structured map node original feature set; S3, performing cross-validation based on the original feature set of the structured map nodes, and generating an increased-information initial topological connection feature diagram for theoretical topological connection of the power distribution network; S4, inputting the initial topological connection feature map into a preset map neural network model, and identifying and generating a topological diagnosis probability map marking an abnormal region; S5, generating a secondary detection signal sequence according to the abnormal region marked in the topology diagnosis probability map; s6, injecting a secondary detection signal sequence into the power distribution network, and generating a focusing response data set in an abnormal area; And S7, performing iterative diagnosis on the abnormal region based on the focusing response data set, and generating a final topology difference positioning report.
  2. 2. The method for diagnosing real differences of distribution network topology based on a graph neural network according to claim 1, wherein the step S1 comprises the following steps: Injecting an initial detection signal sequence into a power grid through an injection unit arranged at the head end of a feeder line of the power distribution network or at a key contact point; The initial detection signal sequence propagates along the physical path of the power grid, and excites each cooperative sensing terminal along the line to synchronously acquire the transient electrical parameters and the physical environment parameters of the response.
  3. 3. The method for diagnosing real differences of distribution network topology map based on the graph neural network according to claim 1, wherein the generation of the original feature set of the map nodes comprises the following steps: The collaborative sensing terminal synchronously collects transient electrical response and environmental internet of things data of the position of the collaborative sensing terminal; Performing time stamp alignment on the acquired transient electrical response and the environmental internet of things data to generate a data unit; Collecting data units generated by all collaborative sensing terminals to form a structured map node original feature set; The transient electrical response comprises voltage, current transient waveforms and phase changes, and the environmental internet of things data comprises infrared temperature measurement readings of the equipment surface and micro-vibration waveforms of the switch box body.
  4. 4. The method for diagnosing real differences of distribution network topology diagrams based on the graphic neural network according to claim 1, wherein the method for generating the initial topology connection characteristic diagram of the added message comprises the following steps: Aiming at two connected associated nodes in the theoretical topology, extracting corresponding transient electrical responses from the original characteristic set of the structured map nodes, calculating attenuation and time delay in the signal propagation process, and forming a preliminary criterion of electrical connectivity; synchronously extracting and comparing the environmental internet of things data of the two associated nodes to form an auxiliary criterion of a physical connection state; When the primary criterion and the auxiliary criterion are mutually verified, the connection is subjected to enhanced assignment, otherwise, the connection is subjected to weight reduction treatment to generate an initial topological connection characteristic diagram with increased information.
  5. 5. The method for diagnosing real differences in distribution network topology based on a graphic neural network as set forth in claim 4, wherein the forming of the auxiliary criteria includes the steps of: extracting respective infrared temperature measurement readings of two associated nodes from environment internet of things data, and calculating the temperature difference between the two; and comparing the temperature difference with a preset abnormal temperature difference threshold value to judge whether the abnormal temperature difference exists or not, and taking the judging result as an auxiliary criterion of the physical connection state.
  6. 6. The method for diagnosing real differences of distribution network topology based on the graph neural network according to claim 1, wherein the generation of the topology diagnosis probability map comprises the following steps: the graph neural network model compares and analyzes the input added information initial topological connection feature graph with a preset health topological feature library; The graph neural network model infers and outputs graph-real difference probability values existing in each connection in the network; and performing visual rendering on all the connected graph real difference probability values to form a topology diagnosis probability map.
  7. 7. The method for diagnosing real differences of distribution network topology based on a graph neural network according to claim 1, wherein the generating of the secondary detection signal sequence comprises the following steps: analyzing a topological diagnosis probability map, locking connection of the map real difference probability value exceeding a preset probability threshold value, and defining the connection as a target diagnosis area; selecting signal frequency and time sequence combination matched with the potential fault type of the region from a preset waveform library according to the electrical characteristics of the target diagnosis region; The selected signal frequency and timing combination is constructed as a secondary probing signal sequence.
  8. 8. The method for real-time differential diagnosis of distribution network topology based on a graph neural network according to claim 1, wherein generating a focus response data set comprises the steps of: Reinjecting the secondary detection signal sequence into the power distribution network; The energy of the secondary detection signal sequence is concentrated to act on a target diagnosis area, and the response is excited on a cooperative sensing terminal of the area; and acquiring response data of each terminal in the target diagnosis area again to form a focusing response data set.
  9. 9. The method for diagnosing real differences of distribution network topology map based on a graphic neural network according to claim 1, wherein the step of generating a final topology difference positioning report comprises the following steps: repeatedly inputting the focus response data set into the cross-validation of the step S3 and the diagnosis flow of the step S4, and reasoning the topology state of the target diagnosis area again to obtain an updated graph-real difference probability value; Obtaining a confidence conclusion about fault points and abnormal states in a target diagnosis area by comparing the change of the graph real difference probability value of the primary diagnosis and the reasoner reasoning; and locking and outputting a root cause causing the real difference of the graph according to the confidence conclusion to form a final topology difference positioning report.
  10. 10. The method for diagnosing a real difference in a distribution network topology based on a graphic neural network as recited in claim 9, wherein the final topology difference positioning report includes a fault location, a fault type, and a confidence level evaluation calculated based on a change in a probability value of the real difference.

Description

Distribution network topology graph real difference diagnosis method based on graph neural network Technical Field The invention belongs to the technical field of operation and maintenance of power distribution networks, and relates to a power distribution network topological graph real difference diagnosis method based on a graph neural network. Background The accuracy of the topological structure of the power distribution network is a basis for guaranteeing safe and stable operation of the power grid and realizing efficient operation and maintenance decision, however, in actual operation, the actual physical connection state of the power grid often has a difference from an archived theoretical topological diagram due to various reasons such as construction change, equipment aging, temporary inspection operation and the like, and the difference causes certain interference to key services such as fault positioning, load transfer, line loss calculation and the like. At present, passive data analysis and traditional manual inspection are commonly adopted in the industry. The passive data analysis method generally utilizes daily operation data collected by power distribution automation systems, advanced metering architecture and other systems to infer the connection relationship among nodes by analyzing the statistical correlation of parameters such as voltage, current, power and the like. The manual inspection is to find out the topology difference by manual comparison. The method of relying on passive data analysis has strong dependence on the operation condition of the power grid, when the network load is stable or the change is not obvious, the signal characteristics are weak, the connection relation is difficult to judge accurately, the detection is difficult to initiate actively, the manual field checking mode has low efficiency although the result is reliable, and the requirements of the modern power grid on real-time, accurate and economical operation and maintenance are difficult to meet, and the requirements of large-scale and normalized topology checking are difficult to support. Disclosure of Invention In order to solve the problems, the invention provides a distribution network topology graph real difference diagnosis method based on a graph neural network. A distribution network topology graph real difference diagnosis method based on a graph neural network comprises the following steps: S1, injecting an initial detection signal sequence comprising time sequence codes and specific frequency components into a power distribution network, and exciting a plurality of collaborative sensing terminals distributed in the power distribution network to generate responses; S2, acquiring and integrating responses generated by the collaborative sensing terminal to generate a structured map node original feature set; S3, performing cross-validation based on the original feature set of the structured map nodes, and generating an increased-information initial topological connection feature diagram for theoretical topological connection of the power distribution network; S4, inputting the initial topological connection feature map into a preset map neural network model, and identifying and generating a topological diagnosis probability map marking an abnormal region; S5, generating a secondary detection signal sequence according to the abnormal region marked in the topology diagnosis probability map; s6, injecting a secondary detection signal sequence into the power distribution network, and generating a focusing response data set in an abnormal area; And S7, performing iterative diagnosis on the abnormal region based on the focusing response data set, and generating a final topology difference positioning report. In a further aspect of the present invention, step S1 includes the following steps: Injecting an initial detection signal sequence into a power grid through an injection unit arranged at the head end of a feeder line of the power distribution network or at a key contact point; The initial detection signal sequence propagates along the physical path of the power grid, and excites each cooperative sensing terminal along the line to synchronously acquire the transient electrical parameters and the physical environment parameters of the response. The further scheme of the invention generates an original feature set of the map node, and comprises the following steps: The collaborative sensing terminal synchronously collects transient electrical response and environmental internet of things data of the position of the collaborative sensing terminal; Performing time stamp alignment on the acquired transient electrical response and the environmental internet of things data to generate a data unit; Collecting data units generated by all collaborative sensing terminals to form a structured map node original feature set; The transient electrical response comprises voltage, current transient waveforms and phase cha