Search

CN-122026284-A - Relay protection method based on transient energy entropy and graph neural network

CN122026284ACN 122026284 ACN122026284 ACN 122026284ACN-122026284-A

Abstract

The application provides a relay protection method based on transient energy entropy and a graph neural network, which comprises the steps of collecting real-time electrical characteristic data from electric equipment, calculating a current energy entropy value through a self-adaptive algorithm, comparing the current energy entropy value with a historical benchmark to obtain a benchmark variation trend, classifying an energy entropy drifting mode according to the benchmark variation trend by adopting a clustering algorithm, judging whether the drifting mode is from equipment aging, determining the drifting mode as aging drifting and updating a reference value if the classifying result shows a gradual change mode, correcting a real-time energy entropy calculation formula through a drifting compensation model to obtain a calibrated abnormal recognition result, verifying the calibrated abnormal recognition result by adopting the clustering algorithm, judging whether protection misjudgment risks are eliminated, and determining the adaptive output of final algorithm.

Inventors

  • ZHANG QIANG

Assignees

  • 国投钦州发电有限公司

Dates

Publication Date
20260512
Application Date
20260129

Claims (9)

  1. 1. A relay protection method based on transient energy entropy and graph neural network is characterized by comprising the following steps: The method comprises the steps of collecting real-time electrical characteristic data from electrical equipment, calculating a current energy entropy value through an adaptive algorithm, comparing the current energy entropy value with a historical reference to obtain a reference variation trend, classifying an energy entropy drift mode according to the reference variation trend, judging whether the drift mode is from equipment aging or not by adopting a clustering algorithm, determining aging drift and updating a reference value if a classification result shows the gradual change mode, obtaining an abnormal variation signal in real-time operation data through the updated reference value, judging the deviation degree of the abnormal variation signal by adopting a threshold comparison method to obtain potential fault characteristics, extracting a similar abnormal identification sample from historical data aiming at the potential fault characteristics, judging a real fault and triggering protection action if the characteristic matching degree is higher than a preset threshold, obtaining triggered protection action feedback data, adjusting algorithm logic parameters through the adaptive algorithm, determining an optimization reference for next judgment, continuously monitoring the subsequent electrical characteristic data according to the optimization reference, judging drift accumulation in a normal state, generating a drift compensation model if the drift accumulation exceeds a warning line, correcting the energy comparison method to obtain a calibrated abnormal identification result, verifying whether the calibrated abnormal risk is judged by adopting the algorithm, finally verifying the abnormal judgment result, and judging whether the abnormal risk is judged by adopting the clustering algorithm.
  2. 2. The method of claim 1, further comprising extracting an environmental impact accumulation index of an aging factor node and a real-time electrical signal chaos degree of the energy entropy value according to the adaptive output of the final algorithm logic, judging a service impact chain under the association of the environmental impact accumulation index of the aging factor node and the energy entropy value, determining a quantization contribution of aging to the chaos degree and updating the drift mode classification parameter if the service impact chain displays degradation mapping, fusing the operation duration association parameter of the aging factor node and the power fluctuation statistical variance of the energy entropy value through the updated drift mode classification parameter to obtain a fused service dependency index, extracting a service dynamic chain under the feedback association between a node embedding capture aging dynamic characteristic of a graph neural network and a relay signal transmission link device through the fused service dependency index, judging feedback aggregation under the aging embedding, and determining a protection response chain under the association and updating the reference change trend if the feedback aggregation shows degradation accumulation.
  3. 3. The method according to claim 2, wherein the collecting real-time electrical characteristic data from the power equipment, calculating a current energy entropy value through an adaptive algorithm, and comparing the current energy entropy value with a historical reference to obtain a reference change trend, includes: acquiring electrical data from the power equipment in real time, constructing an initial data set, and preprocessing the data by adopting automatic tools to obtain a standardized electrical data set; According to the standardized electrical data set, processing the data by using a self-adaptive algorithm, calculating a current energy entropy value, and determining the numerical expression of the current energy entropy value; if the current energy entropy value exceeds a preset threshold range, triggering a comparison process with a historical reference value, and obtaining a reference value comparison result; Analyzing the specific direction of the variation trend through the reference value comparison result, and judging whether abnormal fluctuation exists or not; If the change trend shows abnormal fluctuation, extracting relevant operation monitoring records by combining equipment state data, and determining potential abnormal points; According to the operation monitoring record and the change trend, generating detailed analysis data of the equipment state, and acquiring a targeted state evaluation result; and updating the historical base database through the state evaluation result to obtain the latest base reference data.
  4. 4. The relay protection method based on transient energy entropy and graph neural network according to claim 2, wherein the classifying the energy entropy drift mode by using a clustering algorithm according to the reference change trend, judging whether the drift mode is derived from equipment aging, if the classification result shows a gradual change mode, determining that the drift is aging drift and updating a reference value comprises: acquiring current equipment energy entropy sequence data; calculating the change trend of adjacent time points according to the energy entropy sequence data; classifying energy entropy drift modes corresponding to the change trend by adopting a k-means clustering algorithm to obtain a plurality of mode categories; judging whether the category of the classified drift mode is matched with the gradual change mode, and if the category is matched with the gradual change mode, determining that the equipment ages and drifts; Extracting a characteristic value corresponding to the current reference from the energy entropy sequence determined to be aging drift; replacing a prestored reference value according to the extracted characteristic value to finish updating the reference value; and (3) calculating the change trend of the subsequent energy entropy sequence and the new reference again from the updated reference value.
  5. 5. The relay protection method based on transient energy entropy and graph neural network according to claim 2, wherein the method is characterized in that the abnormal change signal in the real-time operation data is obtained through the updated reference value, the deviation degree of the abnormal change signal is judged by adopting a threshold comparison method, potential fault characteristics are obtained, similar abnormal identification samples are extracted from the historical data aiming at the potential fault characteristics, if the characteristic matching degree is higher than a preset threshold, the method is judged to be a real fault and a protection action is triggered, and the method comprises the following steps: Acquiring real-time operation data and comparing the real-time operation data with an updated reference value to obtain an abnormal change signal; calculating the deviation degree by adopting a threshold comparison mode aiming at the abnormal change signal to obtain potential fault characteristics; extracting similar abnormal samples containing potential fault characteristics from the historical data; calculating the feature matching degree between the potential fault features and similar abnormal samples; If the feature matching degree exceeds a preset threshold, the real fault is determined; starting a protection action aiming at a real fault; and updating the reference value according to the execution result of the protection action.
  6. 6. The relay protection method based on transient energy entropy and graph neural network according to claim 2, wherein the obtaining the triggered protection action feedback data adjusts a calculation logic parameter through a self-adaptive algorithm, determines an optimization reference for next judgment, continuously monitors subsequent electrical characteristic data according to the optimization reference, judges drift accumulation in a normal state, and generates a drift compensation model if the drift accumulation exceeds a warning line, comprising: The method comprises the steps of acquiring feedback data after triggering a protection action, analyzing key indexes of the feedback data, and determining an initial calculation parameter state; according to the acquired feedback data, adopting a self-adaptive algorithm to dynamically adjust the calculated parameters to obtain optimized parameter configuration; establishing a judgment basis aiming at the optimized parameter configuration to form an optimization reference for subsequent monitoring; By continuously monitoring the electrical data, recording the data fluctuation in a normal state, and judging the variation trend of drift accumulation; if the drift accumulated change trend exceeds a preset warning line value, starting a construction flow of a compensation model to acquire compensation parameters; According to the constructed compensation model, adjusting the monitoring logic of the electrical data, and determining a corrected data reference; And continuously analyzing the fluctuation condition of the subsequent electrical data by adopting the corrected data standard to obtain real-time state evaluation.
  7. 7. The relay protection method based on transient energy entropy and graph neural network according to claim 2, wherein the correcting the real-time energy entropy calculation formula through the drift compensation model to obtain the calibrated abnormal recognition result, verifying the calibrated abnormal recognition result by adopting a clustering algorithm, judging whether to eliminate the risk of protection misjudgment, and determining the adaptive output of the final algorithm logic comprises: processing the collected real-time monitoring data through a drift compensation model, and correcting a calculation formula of an energy entropy value to obtain calibrated entropy value data; performing preliminary comparison by adopting a preset threshold value aiming at the calibrated entropy value data, and marking the entropy value data as potential abnormal points if the entropy value data exceeds the threshold value range to obtain a preliminary abnormal marking set; according to the preliminary abnormal marking set, grouping analysis is carried out on the marking points by using a clustering algorithm, the distribution mode of the abnormal points is obtained, and the aggregation area of the abnormal points is determined; Extracting core data from the aggregation area of the abnormal points, and combining with a preset rule of the protection logic, if the core data is inconsistent with the rule, judging that the risk points are misjudged, and obtaining a risk point classification result; Analyzing the matching degree of the algorithm logic through the risk point classification result, and if the matching degree is lower than a preset standard, carrying out parameter adjustment on the algorithm logic to determine an optimized logic framework; acquiring an optimized logic framework, and combining real-time monitoring data to perform dynamic verification, and judging the adaptability performance of the logic framework to obtain a final adaptability result; and generating corresponding protection strategy configuration according to the final adaptability result, outputting strategy configuration data, and completing closed-loop processing of anomaly identification and risk control.
  8. 8. An electronic device comprising a processor, a memory and a program stored on the memory and executable on the processor, characterized in that the program when executed by the processor implements the functions of the method according to any one of claims 1 to 7.
  9. 9. A computer readable storage medium having stored thereon a computer program, which when executed by a processor performs the functions of the method according to any of claims 1 to 7.

Description

Relay protection method based on transient energy entropy and graph neural network Technical Field The invention relates to the technical field of information, in particular to a relay protection method based on transient energy entropy and a graph neural network. Definition of terms The energy entropy value is shannon entropy calculated based on the electric signal energy distribution and is used for quantifying the signal confusion degree; the real-time energy entropy value is the energy entropy calculation result of the electrical signal in the current acquisition period; Transient energy entropy, namely an energy entropy change value at the moment of occurrence of a fault (0-50 ms), and is used for rapidly capturing fault characteristics; The drift mode classification parameters are clustering characteristic parameters for distinguishing aging drift from fault abnormality and comprise a change trend slope, a fluctuation amplitude and duration; calculating logic parameters, wherein the parameters comprise an energy entropy calculation threshold, a characteristic matching degree threshold, a drift accumulation warning line and the like for fault judgment; and optimizing the reference, namely dynamically updating the reference value along with the running state of the equipment based on the fault judgment reference value after the protection action feedback data are adjusted. Background The relay protection of the power system is a core technology for guaranteeing the safe and stable operation of the power grid, directly determines whether a fault area can be rapidly and accurately cut off when a fault occurs, avoids the expansion of accidents, and is crucial to the safety of the whole power system in reliability and adaptability. The current relay protection device mainly depends on manually set fixed algorithm rules and threshold values to judge faults, the rules are determined when the equipment leaves a factory, and as the input operation time of the power equipment is prolonged, the electrical characteristics and the insulation performance of the equipment can be gradually changed, so that the originally set energy entropy reference value continuously drifts. The protection device cannot sense the slow and long-term reference change, and still performs criterion calculation according to fixed standards for many years, so that after the equipment ages to a certain extent, the normal running state can be mistaken as a fault feature, or a real slight fault signal is covered by drift caused by aging, so that the risk of protection misoperation or refusal operation is gradually increased. The energy entropy benchmark drift caused by equipment aging is very similar to transient energy abnormal changes caused by power grid faults in terms of numerical expression, the phenomenon that the energy entropy deviates from a normal interval is caused, and the existing protection logic lacks an effective means to distinguish the changes of the two different sources. Especially when equipment ageing degree is deeper, the energy entropy value under normal operation has been showing and is deviating from original benchmark, if not distinguishing, protection device just has hardly kept original correct action characteristic in long-term service in-process, and this becomes the outstanding contradiction that influences relay protection long-term reliability. Therefore, how to enable the relay protection algorithm to sense and adapt to the natural drift of the energy entropy standard of the power equipment caused by long-term operation, and accurately identify the abnormal energy change caused by faults, so that the rationality of criteria and the correctness of actions are maintained in the whole life cycle of the equipment, and the key problem to be solved by the current intelligent reconstruction relay protection algorithm is urgent. Disclosure of Invention The invention provides a relay protection method based on transient energy entropy and a graph neural network, which mainly comprises the following steps: collecting real-time electrical characteristic data from the power equipment, calculating a current energy entropy value through a self-adaptive algorithm, and comparing the current energy entropy value with a historical benchmark to obtain a benchmark variation trend; classifying an energy entropy drift mode according to the standard change trend, judging whether the drift mode is from equipment aging by adopting a clustering algorithm, determining to age drift and update a reference value if a classification result shows the gradual change mode, acquiring an abnormal change signal in real-time operation data by adopting a threshold comparison method to judge the deviation degree of the abnormal change signal to obtain potential fault characteristics, extracting a similar abnormal identification sample from historical data aiming at the potential fault characteristics, judging to be a real fault and triggering a p