CN-121984231-A - Substation switching operation intelligent monitoring method and system based on multiple data

CN121984231ACN 121984231 ACN121984231 ACN 121984231ACN-121984231-A

Abstract

The application discloses a substation switching operation intelligent monitoring method and system based on multiple data, wherein the method comprises the steps of constructing a state node comprising a topology matrix, an electric quantity vector, a step mark and a time stamp according to a state change event in the switching operation process, and forming a time sequence topology evolution diagram; and simultaneously, constructing an equivalent impedance network based on the current topology and the boundary of the protection section, carrying out virtual fault calculation by combining key fault points, generating a protection expected action matrix, and obtaining a semantic deviation score according to the virtual fault calculation. On the basis, the evolution embedded features and the protection semantic features are fused to construct a joint feature vector, a joint deviation score is obtained through a joint consistency model, and finally, each deviation score is integrated to form a risk assessment result. The method realizes the analysis of the structural and semantic consistency of the whole switching operation process, and improves the accuracy and the instantaneity of anomaly identification.

Inventors

HUANG SONGLIN
Ji Guiwen
LI CHUHAO
LI JIA

Assignees

广东安总电力建设有限公司

Dates

Publication Date: 20260505
Application Date: 20260408

Claims (10)

1. A substation switching operation intelligent monitoring method based on multiple data is characterized by sequentially generating state nodes comprising a current topology matrix, a current electrical quantity vector, a current step mark and a current time stamp according to state change events in a switching operation process, constructing a time sequence topology evolution diagram based on adjacent state nodes, inputting the time sequence topology evolution diagram into a pre-trained graph neural network model to obtain a current evolution embedded vector, matching the current evolution embedded vector with a legal topology evolution model library to obtain a structure deviation score, constructing an equivalent impedance network according to a current topology matrix and a current protection section boundary set, performing virtual fault calculation based on a key fault point set to generate a current protection expected action matrix, generating a combined feature vector according to the current evolution embedded vector and the current protection expected action matrix, inputting a combined consistency model to obtain a combined deviation score, generating a comprehensive risk value according to the structure deviation score, the semantic deviation score and the combined deviation score, and outputting a monitoring result corresponding to the comprehensive risk value.
2. The method of claim 1, wherein constructing the time sequence topology evolution graph based on adjacent state nodes comprises calculating topology matrix differences of any adjacent previous state node and current state node to obtain topology variation, calculating electric quantity vector differences of the previous state node and current state node to obtain electric quantity variation, calculating time stamp differences of the previous state node and the current state node to obtain time intervals, taking each state node as a graph node, and taking corresponding topology variation, electric quantity variation and time intervals as state transition edge attributes to construct the directed time sequence topology evolution graph.
3. The method of claim 1, wherein the obtaining the structure deviation score includes performing distance calculation on the current evolution embedded vector and each cluster center vector in the legal topology evolution pattern library, determining a target pattern cluster corresponding to a minimum distance, reading an intra-cluster distance threshold corresponding to the target pattern cluster, generating the structure deviation score according to a ratio of the minimum distance to the intra-cluster distance threshold, and determining a structure deviation source node and a structure deviation source edge according to contribution degrees of state nodes and state transition edges forming the time sequence topology evolution diagram.
4. The method of claim 1, wherein constructing an equivalent impedance network according to a current topology matrix and a current protection section boundary set, performing virtual fault calculation based on a key fault point set, and generating a current protection expected action matrix, comprises determining a current network node set and a branch set according to the current topology matrix, generating a node admittance matrix by combining branch impedance parameters, solving the equivalent impedance matrix by the node admittance matrix, determining a key fault point set according to the current topology matrix, a current operation step and the current protection section boundary set, performing virtual fault calculation for each key fault point to obtain a fault response quantity corresponding to each protection device, determining an action mark of each protection device under each key fault point according to a fixed value parameter and a direction discrimination condition of each protection device, and generating the current protection expected action matrix.
5. The method of claim 4, wherein the current protection zone boundary set is generated by pre-storing an initial boundary template corresponding to each protection device, under the current state node, starting from the direction reference node of each protection device, performing limited reachable search based on the current topology matrix, terminating the search in the corresponding direction when encountering a breaker or a disconnecting switch in the disconnection state during the search, intercepting the search result according to the zone cut-off condition when encountering the adjacent protection zone boundary node, and updating the effective boundary set corresponding to each protection device according to the search result to form the current protection zone boundary set.
6. The method of claim 1, wherein obtaining the semantic deviation score according to the current protection expected action matrix and the previous protection expected action matrix comprises differentiating the current protection expected action matrix and the previous protection expected action matrix to obtain a matrix change result, counting change elements in the matrix change result, determining element weights according to protection device types, key fault point types and change types corresponding to the change elements, and generating the semantic deviation score according to the change elements and the corresponding element weights.
7. The method of claim 1, wherein generating a joint feature vector from the current evolution embedded vector and the current protection expected action matrix, inputting a joint consistency model to obtain a joint deviation score, comprises expanding the current protection expected action matrix into a matrix vector, extracting the number of corresponding action elements, the number of change elements and a key protection device change ratio to generate a semantic feature vector, splicing the current evolution embedded vector and the semantic feature vector into a joint feature vector, inputting the joint feature vector into the joint consistency model to obtain a joint consistency probability, and generating the joint deviation score according to the joint consistency probability.
8. The method of claim 7, wherein the joint consistency model comprises an input layer, a first fully connected layer, a second fully connected layer and an output layer which are sequentially connected, wherein the input layer receives joint feature vectors, the first fully connected layer and the second fully connected layer respectively perform nonlinear mapping on the joint feature vectors, the output layer outputs joint consistency probabilities, and the joint consistency model is trained by extracting joint feature vectors corresponding to historical legal switching samples as positive samples, extracting joint feature vectors generated after limited disturbance is applied to the historical legal switching samples as negative samples through historical abnormal samples, and training model parameters.
9. The method of claim 1, wherein the graph neural network model comprises an input encoding module, a first graph rolling module, a second graph rolling module, a time sequence aggregation module and an embedded output module which are sequentially connected, wherein the input encoding module receives node characteristics of state nodes and edge characteristics of state transition edges, the first graph rolling module and the second graph rolling module execute neighborhood aggregation on the node characteristics, the time sequence aggregation module receives node representations arranged in time sequence and outputs time sequence structure characteristics, and the embedded output module maps the time sequence structure characteristics into evolution embedded vectors.
10. A substation switching operation intelligent monitoring system based on multiple data is characterized by comprising a structural deviation analysis module, a comprehensive evaluation and monitoring result output module, a combined feature vector and a combined deviation score, wherein the structural deviation analysis module is used for sequentially generating a state node comprising a current topology matrix, a current electrical quantity vector, a current step mark and a current timestamp according to a state change event in the switching operation process, constructing a time sequence topology evolution graph based on adjacent state nodes, inputting the time sequence topology evolution graph into a pre-trained graph neural network model to obtain a current evolution embedded vector, matching the current evolution embedded vector with a legal topology evolution mode library to obtain a structural deviation score, the semantic deviation analysis module is used for constructing an equivalent impedance network according to a current topology matrix and a current protection section boundary set, performing virtual fault calculation based on a key fault point set, generating a current protection expected action matrix, obtaining a semantic deviation score according to the current protection expected action matrix and a previous protection expected action matrix, inputting the combined feature vector according to the current evolution embedded vector and the current protection expected action matrix, obtaining the combined feature vector, obtaining the combined deviation score according to the structural deviation score, the deviation score and the combined risk score, and outputting a comprehensive risk value corresponding to the monitored result.

Description

Substation switching operation intelligent monitoring method and system based on multiple data Technical Field The application relates to the technical field of operation monitoring and intelligent analysis of power systems, in particular to a substation switching operation intelligent monitoring method and system based on multiple data. Background The switching operation of the transformer substation is a key link in operation management of the power system, mainly relates to state switching of a breaker, a disconnecting switch and related primary equipment, and aims to realize operation requirements such as equipment overhaul, operation mode adjustment, fault processing and the like. In the prior art, a monitoring mode of switching operation mainly depends on a manual checking and rule-based checking method, for example, potential risks are identified through operation ticket checking, five-prevention logic judgment and simple topology consistency checking, the method is usually based on static rules or single data sources, coupling relation of multidimensional information in the switching process is difficult to comprehensively reflect, and particularly, under the conditions of complex wiring mode, temporary operation mode switching and multi-step continuous operation, the problem of insufficient abnormal path identification capability exists. With the popularization of dispatching automation systems and transformer substation comprehensive automation systems, the system can acquire multi-source data such as remote signaling, remote sensing, topology models, protection information and the like. However, the prior art often carries out isolation processing on the data, lacks overall modeling of a topology evolution process and also lacks an analysis means for consistency between protection action expectations and actual topology changes, and in actual engineering application, partial abnormal operation is not recognized under single rule verification, but is deviated from topology evolution paths or protection response angles, so that misoperation or protection misoperation risks are easily caused. The existing method is also insufficient in terms of time sequence characteristics of switching operation, and is generally only judged according to a state at a certain moment, continuous state change information in the operation process is not fully utilized, dynamic evolution characteristics of an operation path are difficult to characterize, and in the case, when the sequence of operation steps is abnormal or the intermediate state is not in line with expectations, the system is difficult to recognize in time and give effective prompts. Therefore, how to perform joint modeling on the structure and semantic level of the whole switching operation process based on the existing multi-source data and realize the fine recognition of the deviation of the operation path becomes a technical problem to be solved in the field. Disclosure of Invention The embodiment of the application provides a substation switching operation intelligent monitoring method and system based on multiple data, which are used for at least solving part of technical problems in the related art. According to a first aspect of the embodiment of the application, an intelligent monitoring method for switching operation of a transformer substation based on multiple data is provided, and the intelligent monitoring method comprises the steps of sequentially generating state nodes comprising a current topology matrix, a current electrical quantity vector, a current step mark and a current time stamp according to a state change event in the switching operation process, constructing a time sequence topology evolution diagram based on adjacent state nodes, inputting the time sequence topology evolution diagram into a pre-trained graph neural network model to obtain a current evolution embedded vector, matching the current evolution embedded vector with a legal topology evolution model library to obtain a structure deviation score, constructing an equivalent impedance network according to a current topology matrix and a current protection section boundary set, performing virtual fault calculation based on a key fault point set to generate a current protection expected action matrix, generating a combined feature vector according to the current evolution vector and the current protection expected action matrix, inputting a combined consistency model to obtain a combined deviation score, generating a comprehensive risk value according to the structure deviation score, the semantic deviation score and the combined risk score, and outputting a monitoring result corresponding to the comprehensive risk value. The method comprises the steps of calculating topology matrix differences of any adjacent previous state node and current state node to obtain topology variation, calculating electric quantity vector differences of the previous state node and the current