CN-121983904-A - Relay protection testing method based on panoramic parameter prediction of power system

CN121983904ACN 121983904 ACN121983904 ACN 121983904ACN-121983904-A

Abstract

The invention discloses a relay protection test method based on panoramic parameter prediction of an electric power system, which relates to the field of relay protection and intelligent operation and maintenance of electric power equipment and aims to solve the technical problem that the original main protection history data is not effectively utilized in the traditional mode, in the invention, firstly, a XGBoost model is initialized and a super parameter search space is defined, then an optimal super parameter combination of XGBoost is intelligently searched by utilizing an adaptive moth flame optimization algorithm (MFO), the prediction precision and generalization capability of the system are improved, an optimized prediction model is finally output and deployed, after the model is deployed, the system enters a real-time dual protection logic judgment stage, the main protection unit and the standby protection unit work in parallel, the monitored real-time power values of the main protection unit and the standby protection unit are respectively compared with the set values of the main protection unit and the standby protection unit, and only when the two protection units simultaneously monitor the power exceeding limit, the system can finally send a tripping instruction to turn off the switch.

Inventors

LI BIN
WANG XIAODONG
XU ZHENG
HE HUAN
ZHAO SIYAN
CHEN HUIWEN
LI XINYU
ZHAO JINHUI

Assignees

国网辽宁省电力有限公司鞍山供电公司

Dates

Publication Date: 20260505
Application Date: 20251224

Claims (8)

1. The relay protection testing method based on the panoramic parameter prediction of the power system is characterized by comprising the following steps of: setting an original data set and carrying out data standardization processing; step two, XGBoost model construction, namely establishing XGBoost regression model, wherein an objective function of the model construction comprises a loss function of a prediction error and a regularization term for controlling complexity of the model; step three, self-adaptive MFO optimization, wherein key super parameters of XGBoost are used as optimization variables, the population of the moths is initialized, and each moths represents a group of super parameter combinations; step four, iterative optimization, wherein in each iteration, the moth updates its own position according to the distance between the moth and the flame; fifthly, the adaptability function ensures the stability of the evaluation result through cross verification; Step six, a flame position updating mechanism sorts the moths according to the fitness value, and selects an individual with optimal performance as a new flame position; step seven, self-adaptive parameter adjustment, namely dynamically adjusting search parameters by adopting a learning rate attenuation mechanism and a Levy flight strategy; Step eight, judging convergence conditions, setting dual convergence conditions including relative error change and maximum iteration times, terminating an algorithm and outputting an optimal solution when the optimization process reaches a preset precision requirement or the maximum iteration times; and step nine, protecting a parameter prediction model, training a final XGBoost prediction model by using an optimal super parameter combination obtained by optimization, wherein the model can accurately predict optimal protection parameter values under various operation conditions.
2. The relay protection testing method based on panoramic parameter prediction of a power system according to claim 1, wherein the objective function is: ; Wherein, the As a loss function; Is the predicted value of the previous t-1 round; a predicted value of a t-th tree; is a regularization term.
3. The relay protection testing method based on panoramic parameter prediction of an electric power system according to claim 1, wherein the position of the moth is initialized: ; Wherein eta is learning rate, gamma is minimum splitting loss, lambda is L2 regularization coefficient, d is maximum tree depth, n is number of trees, M is population scale of moths; Adjusting the self-adaptive flame quantity: ; Wherein T is the current iteration number, and T is the maximum iteration number; the moth position update formula: ; Wherein, the B (t) is a self-adaptive spiral constant; The position of the j-th flame; adaptive spiral constant: ; wherein b0 is the initial spiral constant and alpha 1 is the attenuation coefficient.
4. The relay protection testing method based on panoramic parameter prediction of a power system according to claim 1, wherein the cross validation function: ; Wherein ω1, ω2, ω3 are weight coefficients, K is a cross validation fold number, MSEk is a mean square error of the kth fold, MAEk is a mean absolute error of the kth fold, and RMSEk is a root mean square error of the kth fold.
5. The relay protection testing method based on panoramic parameter prediction of an electric power system according to claim 1, wherein the flame position is updated: ; Wherein, the And (5) sorting the j-th moth positions according to the fitness value.
6. The relay protection testing method based on panoramic parameter prediction of an electric power system according to claim 1, wherein the learning rate attenuation mechanism is as follows: ; wherein eta 0 is the initial learning rate and beta is the attenuation rate; levy flight mechanism: ; Wherein alpha 2 is a step control parameter; Is the Levy random step size and fbest is the current optimal flame position.
7. The relay protection testing method based on panoramic parameter prediction of a power system according to claim 1, wherein the relative error condition is as follows: ; Wherein ε 1 is the convergence threshold; Maximum iteration number condition: 。
8. The relay protection testing method based on panoramic parameter prediction of a power system according to claim 1, wherein the final prediction model is as follows: ; Wherein n is the optimal tree number; a predictive function for the kth decision tree; Real-time prediction and protection fixed value adjustment In actual operation, dynamically adjusting a protection rated value according to real-time monitoring data and output of a prediction model; and (3) dynamically adjusting a protection fixed value: ; Wherein g is a safety coefficient; to predict confidence; is the predicted value of the current moment.

Description

Relay protection testing method based on panoramic parameter prediction of power system Technical Field The invention relates to the field of relay protection and intelligent operation and maintenance of power equipment, in particular to a relay protection test method based on panoramic parameter prediction of a power system. Background In the single-chip microcomputer protection performance test and load operation protection work of an industrial control system relay protection device, the traditional protection and test mode has obvious limitations, and at present, the performance test of the single-chip microcomputer protection module mainly depends on acquisition signals of an existing current transformer and a voltage transformer on site, and whether the protection module is normal is judged by calculating power parameters. However, no additional protection mechanism is arranged for the running state of the motor, if the current and voltage rise, the power overload and other anomalies occur in the test or running process, the motor is extremely easy to damage due to the lack of a secondary protection means, the maintenance cost of the equipment is increased, the production flow is interrupted due to the shutdown of the motor, and the on-site continuous operation efficiency is influenced. In addition, the historical data of the original main protection is not effectively utilized in the traditional mode, and even if the standby protection is additionally arranged, the rated power parameter of the device is set by manual experience, so that the suitability of the parameter and the actual running characteristic of the motor is poor, the problems of untimely overload protection of the motor, unreasonable protection action logic and the like are caused, and the safe running of the device is caused to face potential risks. Therefore, it is necessary to develop a relay protection optimization scheme which does not destroy the original protection loop, has double protection redundancy and is based on panoramic parameter prediction of historical data, so as to improve the reliability of relay protection and the safety of motor operation, and adapt to the requirements of continuous production and precise protection of industrial sites. Disclosure of Invention The invention aims to solve the problems and provides a relay protection test method based on panoramic parameter prediction of an electric power system. The invention aims at realizing the relay protection test method based on panoramic parameter prediction of the electric power system, which comprises the following steps: setting an original data set and carrying out data standardization processing; step two, XGBoost model construction, namely establishing XGBoost regression model, wherein an objective function of the model construction comprises a loss function of a prediction error and a regularization term for controlling complexity of the model; step three, self-adaptive MFO optimization, wherein key super parameters of XGBoost are used as optimization variables, the population of the moths is initialized, and each moths represents a group of super parameter combinations; step four, iterative optimization, wherein in each iteration, the moth updates its own position according to the distance between the moth and the flame; fifthly, the adaptability function ensures the stability of the evaluation result through cross verification; Step six, a flame position updating mechanism sorts the moths according to the fitness value, and selects an individual with optimal performance as a new flame position; step seven, self-adaptive parameter adjustment, namely dynamically adjusting search parameters by adopting a learning rate attenuation mechanism and a Levy flight strategy; Step eight, judging convergence conditions, setting dual convergence conditions including relative error change and maximum iteration times, terminating an algorithm and outputting an optimal solution when the optimization process reaches a preset precision requirement or the maximum iteration times; and step nine, protecting a parameter prediction model, training a final XGBoost prediction model by using an optimal super parameter combination obtained by optimization, wherein the model can accurately predict optimal protection parameter values under various operation conditions. Further, the objective function: ; Wherein, the As a loss function; Is the predicted value of the previous t-1 round; a predicted value of a t-th tree; is a regularization term. Further, the moth position is initialized: ; Wherein eta is learning rate, gamma is minimum splitting loss, lambda is L2 regularization coefficient, d is maximum tree depth, n is number of trees, M is population scale of moths; Adjusting the self-adaptive flame quantity: ; Wherein T is the current iteration number, and T is the maximum iteration number; the moth position update formula: ; Wherein, the B (t) is a self-adaptive spiral c