CN-122024428-A - Substation equipment state early warning telemetry signal simulation test method

CN122024428ACN 122024428 ACN122024428 ACN 122024428ACN-122024428-A

Abstract

The invention relates to the technical field of intelligent operation and maintenance and test of a power system and discloses a substation equipment state early-warning telemetry signal simulation test method, which comprises the following steps of establishing a reinforcement learning framework, taking a centralized control system and an early-warning function module thereof as environments, constructing an intelligent body as a signal generation and decision unit, defining a state space as early-warning result information returned by the centralized control system, representing the state space as a plurality of groups comprising early-warning levels, early-warning types, time stamps and equipment identifiers, defining an action space as a telemetry signal value set generated by the intelligent body, and designing a reward function And evaluating action effectiveness. The substation equipment state early warning telemetry signal simulation test method aims at solving the problems that in the prior art, a test process lacks self-adaptive learning and closed loop optimization capabilities, and a test strategy cannot be dynamically adjusted according to system response.

Inventors

CHEN YONGXIN
MA DI
LIU XIN
DU ZHENAN
SHI BING
LI HENGHUI
ZHANG ZANZAN
TENG JIE
XIU LIANCHENG
Niu Chenmeng

Assignees

国网湖北省电力有限公司电力科学研究院

Dates

Publication Date: 20260512
Application Date: 20260206

Claims (10)

1. The simulation test method for the state early warning telemetry signal of the power transformation equipment is characterized by comprising the following steps of: s1, establishing a reinforcement learning framework, taking a centralized control system and an early warning function module thereof as environments, constructing an intelligent body as a signal generation and decision unit, defining a state space as early warning result information returned by the centralized control system, representing the state space as a plurality of groups comprising early warning level, early warning type, time stamp and equipment identifier, defining an action space as a telemetry signal value set generated by the intelligent body, and designing a reward function Evaluating action effectiveness, wherein the reward function comprises an early warning success reward item, an accuracy reward item, an timeliness reward item and a cost penalty item; s2, performing closed loop iterative training of the agent and the environment, wherein the agent is in accordance with the current state through a strategy network Generating test actions The virtualization test platform sends the telemetry signal to the centralized control system, and the centralized control system returns to a new state after responding And rewards Sampling historical data from the buffer area, calculating long-term accumulated expected return through a strategy gradient loss function, and updating strategy network parameters; S3, executing the next round of test based on the updated strategy network to form an execution-feedback-learning-optimizing closed loop, and recording test cases causing missing report, false report or delay as a centralized control system early warning rule optimizing sample; And S4, outputting a test report containing defect case libraries, performance quantitative analysis and early warning rule optimization suggestions.
2. The simulation test method for the state early warning telemetry signal of the power transformation equipment according to claim 1, wherein the state of the state space in the S1 is that The specific characteristics are as follows: ; wherein the early warning level Includes four levels of normal, concerned, abnormal and serious, early warning type Corresponding to event rules and state early warning types, and time stamps For the calculation of the response delay, To trigger a specific substation device number or name for the pre-alarm, A sequence of historical states is provided to provide timing context information.
3. The simulation test method for the state early warning telemetry signal of the power transformation equipment according to claim 1, wherein the action of the action space in the step S1 is that A numerical vector comprising a plurality of telemetry signal channels, each telemetry signal channel corresponding to a particular physical measurement point of the substation device, the actions expressed as: ; Wherein the method comprises the steps of Is the first The number of telemetry channels at time Is used for the signal amplitude of (a), For the rate of change or duration of the channel signal, For the total number of telemetry channels.
4. The simulation test method for the state early warning telemetry signal of the power transformation equipment according to claim 2 or 3, wherein the calculation formula of the S1 winning function is as follows: ; Wherein: giving a base value rewarding when triggering the expected target early warning, otherwise, giving zero; Giving negative rewards when triggering non-target false alarm early warning; According to the early warning response delay time Calculation using decreasing function Wherein the method comprises the steps of Is a positive constant; The signal complexity penalty is positively correlated with the action sequence length or signal variation amplitude.
5. The simulation test method for the state early warning telemetry signal of the power transformation equipment according to claim 1, wherein the S2 strategy network adopts a deep neural network structure, an input layer receives a current state vector, features are extracted through a plurality of full-connection layers or cyclic neural network layers, an output layer generates motion probability distribution or motion parameters, and the strategy network generates motion probability distribution or motion parameters through a mapping function Expressed in parameters In the lower position The probability of the action is selected.
6. The substation equipment state early warning telemetry signal simulation test method according to claim 5, wherein the step S2 adopts a near-end strategy optimization algorithm, and the loss function is: ; Wherein the method comprises the steps of For the policy ratio to be the same, As an estimate of the merit function, Is a clipping parameter, the dominance function is calculated as Wherein The return is accumulated for the discount, As a discount factor, the number of times the discount is calculated, Is a state cost function.
7. The simulation test method for the early warning telemetry signal of the state of the power transformation equipment according to claim 1, wherein the experience playback buffer zone in the step S2 adopts a priority experience playback mechanism according to a time sequence differential error And giving priority to each experience, and preferentially selecting the experience with larger error for training during sampling, so that the learning efficiency is improved.
8. The substation equipment state early warning telemetry signal simulation test method according to claim 1, wherein the closed loop iterative process in S2 sets a multi-stage training strategy: increasing the randomness of the strategy network output in the initial exploration stage to enlarge the coverage range of the state space; The stability optimization stage reduces randomness to fine tune the test strategy; Fixing the strategy network in the convergence verification stage and performing continuous tests to evaluate the performance stability; The phase switch is automatically triggered based on a moving average or variance indicator of the jackpot.
9. The substation equipment state early-warning telemetry signal simulation test method according to claim 1 is characterized in that a centralized control system performance index quantitative evaluation system is established in the step S3, wherein the system comprises early-warning recall rate, accuracy rate, F1 fraction and average response time delay; The defect test cases recorded in the step S3 comprise complete signal time sequence data, centralized control system response data and defect classification labels, the defect cases are subjected to cluster analysis, signal mode features triggering specific defects are identified, and early warning rule optimization suggestions are generated based on a clustering result and comprise threshold adjustment schemes, newly added early warning logic rules or equipment association relation correction.
10. The substation equipment state early warning telemetry signal simulation test method according to claim 1 is characterized in that the test report output by the step S4 further comprises the steps of training a convergence graph to show the change of a cumulative prize along with iteration times, testing a coverage thermodynamic diagram to identify the triggering frequency distribution of various early warning rules, a signal waveform visualization diagram of a key defect use case, and a strategy network model file and a calling interface thereof after training for subsequent regression test and continuous integration flow.

Description

Substation equipment state early warning telemetry signal simulation test method Technical Field The invention relates to the technical field of intelligent operation and maintenance and testing of power systems, in particular to a substation equipment state early-warning telemetry signal simulation testing method. Background The power transformation equipment state early warning system is a core safety guarantee module of the power dispatching centralized control center, and early warning information is timely sent out when equipment is abnormal through collecting equipment telemetry signals in real time and comparing the equipment telemetry signals with a preset threshold value or rules, so that the expansion of accidents is avoided. Along with the expansion of the power grid scale and the increase of the equipment types, the configuration of the early warning rules is more and more complex, hundreds of equipment types and thousands of rule logics are involved, and the reliability verification of the early warning function becomes a rigid requirement. The traditional testing method mainly relies on a tester to manually design a test case according to an early warning rule, and a specific telemetry signal is injected into a centralized control system through a virtualization platform to observe whether early warning response accords with expectations. The method can meet basic test requirements in the scene of less rules and simple logic, but is essentially a static test mode based on manual experience, and the design quality of the test case is highly dependent on the understanding depth of the early warning rules by testers. However, the testing process lacks adaptive learning and closed loop optimization capabilities, and the testing strategy cannot be dynamically adjusted according to the real-time response of the centralized control system. The method comprises the steps of performing solidification execution once design of a test case is completed, even if some signal modes are found to trigger system defects, subsequent tests cannot autonomously mine boundary conditions around the depth of the defects, when a centralized control system fails to report or misreports, the test system can only passively record the defects and cannot actively generate test signals with stronger pertinence to verify problem root causes, more importantly, the test process and the centralized control system are completely split in an optimization mode, the found defect case only serves as a problem report, and cannot be converted into a training sample optimized by system rules, so that test efficiency is low and early warning rule verification requirements related to complex time sequences are difficult to deal with, and improvement of intelligent test levels of an electric power system is restricted. Disclosure of Invention The invention aims to solve the problems that the testing process in the prior art lacks self-adaptive learning and closed-loop optimization capability and a testing strategy cannot be dynamically adjusted according to system response, and provides a substation equipment state early-warning telemetry signal simulation testing method. The technical scheme for solving the technical problems is as follows: A substation equipment state early warning telemetry signal simulation test method comprises the following steps: s1, establishing a reinforcement learning framework, taking a centralized control system and an early warning function module thereof as environments, constructing an intelligent body as a signal generation and decision unit, defining a state space as early warning result information returned by the centralized control system, representing the state space as a plurality of groups comprising early warning level, early warning type, time stamp and equipment identifier, defining an action space as a telemetry signal value set generated by the intelligent body, and designing a reward function Evaluating action effectiveness, wherein the reward function comprises an early warning success reward item, an accuracy reward item, an timeliness reward item and a cost penalty item; s2, performing closed loop iterative training of the agent and the environment, wherein the agent is in accordance with the current state through a strategy network Generating test actionsThe virtualization test platform sends the telemetry signal to the centralized control system, and the centralized control system returns to a new state after respondingAnd rewardsSampling historical data from the buffer area, calculating long-term accumulated expected return through a strategy gradient loss function, and updating strategy network parameters; S3, executing the next round of test based on the updated strategy network to form an execution-feedback-learning-optimizing closed loop, and recording test cases causing missing report, false report or delay as a centralized control system early warning rule optimizing sample; And S4,