CN-121980344-A - Electric energy meter abnormal state self-adaptive identification method, system and medium

CN121980344ACN 121980344 ACN121980344 ACN 121980344ACN-121980344-A

Abstract

The invention discloses a self-adaptive identification method, a self-adaptive identification system and a self-adaptive identification medium for an abnormal state of an electric energy meter, which comprise the steps of obtaining basic data of the electric energy meter, inputting XGBoost the basic data into a multi-attention fusion model for training, generating a real model and generating a virtual model consistent with the time step of the real model, calculating the relative deviation rate and trend difference degree of electric characteristic data in the real model and virtual electric characteristic data in the virtual model under the same time step, taking the relative deviation rate as the characteristic difference degree, constructing a three-dimensional evaluation index system, marking the time step exceeding a threshold value as a difference index, inputting the difference index into a bidirectional LSTM model, combining a dynamic decision boundary optimized by a Q-learning algorithm, judging the abnormal type of the electric energy meter, and triggering XGBoost and the multi-attention fusion model to update parameters according to the judgment result of the abnormal type. The invention reduces the misjudgment rate and the missed judgment probability in complex scenes.

Inventors

LIU SUJIE
LI RUICHAO
FAN LI
HE PEIDONG
LI FANGSHUO
WANG JIAJU
TU YAXIN

Assignees

国网四川省电力公司营销服务中心

Dates

Publication Date: 20260505
Application Date: 20260120

Claims (10)

1. An electric energy meter abnormal state self-adaptive identification method is characterized by comprising the following steps: Acquiring basic data of an electric energy meter, wherein the basic data comprise multidimensional electrical characteristic data, environmental interference data and weather data; Training the basic data input XGBoost and a multi-attention fusion model to generate a real model, and generating a virtual model consistent with the real model time step based on the real model; calculating the relative deviation rate and trend difference degree of the electrical characteristic data in the real model and the virtual electrical characteristic data in the virtual model under the same time step, and taking the relative deviation rate as the characteristic difference degree; Determining average characteristic difference degree and maximum characteristic difference degree based on the characteristic difference degree, forming a three-dimensional evaluation index system by combining the average characteristic difference degree, the maximum characteristic difference degree and the trend difference degree, and marking time steps exceeding a threshold value as difference indexes according to the three-dimensional evaluation index system; Inputting the difference index into a bidirectional LSTM model, combining a dynamic decision boundary optimized by a Q-learning algorithm, judging the abnormal type of the electric energy meter, obtaining an abnormal type judgment result, and And triggering XGBoost the multi-attention fusion model to update parameters according to the abnormal type judgment result, so as to realize model self-adaptive optimization.
2. The method for adaptively identifying abnormal states of an electric energy meter according to claim 1, wherein the multidimensional electrical characteristic data comprises voltage, current, active power, reactive power, power factors, frequency and harmonic distortion rate; The environmental interference data includes temperature, humidity and electromagnetic radiation intensity; The weather data comprise local rainfall, wind speed, illumination intensity and atmospheric pressure; And acquiring all of the multidimensional electrical characteristic data, the environmental interference data and the weather data by adopting a uniform time interval.
3. The method for adaptively identifying abnormal states of an electric energy meter according to claim 2, further comprising: Preprocessing the basic data respectively to obtain preprocessed basic data, wherein the preprocessing comprises the following steps: Denoising the multidimensional electrical characteristic data by adopting a VMD and an empirical wavelet threshold, and generating an countermeasure network filling data loss by adopting a attention mechanism bidirectional time sequence to obtain preprocessed 7-dimensional electrical characteristic data; Removing abnormal values from the temperature and the humidity by adopting a3 sigma criterion, and obtaining preprocessed temperature and humidity data through normalization processing; for electromagnetic radiation, adopting sliding window normalization to obtain preprocessed electromagnetic radiation data, and finally obtaining preprocessed 3-dimensional environment interference data; And normalizing the weather data by adopting a sliding window to obtain preprocessed 4-dimensional weather data.
4. The method for adaptively identifying abnormal states of an electric energy meter according to claim 1, wherein training the basic data input XGBoost and the multi-attention fusion model to generate a real model comprises: Inputting XGBoost the base data into a multi-attention fusion model; Establishing training samples through Z-score standardization and timestamp hash matching, wherein 1 sample=n-dimensional characteristics of one timestamp and an anomaly label, and learning the association of the n-dimensional characteristics and the anomaly label through a tree integration model; Taking the importance score of the n-dimensional feature as an initial weight, and obtaining a weighted n-dimensional weight through three attention layers; Generating the reality model according to the association of the n-dimensional characteristics and the abnormal labels and the weighted n-dimensional weights; The three-layer attention layer comprises an electric characteristic attention layer, an environment attention layer and a weather attention layer, wherein the electric characteristic attention layer is used for calculating mutual information entropy of multi-dimensional electric characteristic data and various abnormal labels, the environment attention layer is used for calculating mutual information entropy of environment interference data and electric characteristic data, and the weather attention layer is used for calculating mutual information entropy of weather data and environment interference data; the abnormal labels comprise four types of abnormal labels of 0, 1, 2 and 3, wherein 0 represents that the electric energy meter is normal, 1 represents that the electric energy meter measures deviation, 2 represents that the environment is disturbed abnormally, and 3 represents that the weather is driven abnormally.
5. The method of claim 4, wherein generating a virtual model consistent with the real model time step based on the real model, comprises: Obtaining real weather data based on the real model; generating virtual environment data through a linear regression and random forest fusion algorithm according to the real weather data; based on the virtual environment data, and combining the environment interference influence coefficient and the weather influence coefficient output by the reality model, obtaining virtual electrical characteristic data; Based on the virtual environment data and the virtual electrical characteristic data, a virtual model consistent with the real model time step is generated.
6. The method for adaptively identifying abnormal states of an electric energy meter according to claim 5, wherein the contribution weight of the multidimensional electrical characteristic data is an electrical characteristic weight weighted by an electrical characteristic attention layer; the environmental interference influence coefficient is an average value of environmental characteristic weights weighted by the environmental attention layer; the weather effect coefficient is an average value of weather feature weights weighted by the weather attention layer.
7. The method for adaptively identifying abnormal states of an electric energy meter according to claim 1, wherein the relative deviation rate is The calculation formula of (2) is as follows: ; In the formula, Is the electrical characteristic data of the real model, Is the electrical characteristic data of the virtual model.
8. The adaptive recognition method for abnormal states of an electric energy meter according to claim 1, wherein the calculation formula of the trend difference degree is as follows: ; In the formula, The pearson correlation coefficient representing the real data and the virtual data is Wherein Indicating the first in the sliding window Real-life electrical characteristic data of the individual time steps, Indicating the first in the sliding window Virtual electrical characteristic data for each time step, Representing the mean value of the sequence of electrical data, Representing the mean value of the virtual electrical data sequence, Representing the sum of covariance of the real and virtual electrical data sequences, Representing the standard deviation of the real electrical data sequence, Representing the standard deviation of the virtual data sequence, The value range is-1 to 1; The DTW distance is calculated as the formula Wherein Represent the first Personal reality data and the first The absolute deviation of the individual virtual data is, Representing the best matching path from the top left corner to the bottom right corner of the difference matrix, Representing the minimum cumulative distance of all absolute deviations on the optimal path, Indicating the number of time steps within the sliding window, Representing all of the difference matrices Is a maximum value of (a).
9. An electric energy meter abnormal state self-adaptation identification system, which is characterized in that the system comprises: The acquisition unit is used for acquiring basic data of the electric energy meter, including multidimensional electrical characteristic data, environment interference data and weather data; The three-mode association unit is used for inputting XGBoost the basic data and training the multi-attention fusion model to generate a reality model; A virtual model generating unit, configured to generate a virtual model consistent with the real model time step based on the real model; the computing unit is used for computing the relative deviation rate and the trend difference degree of the electric characteristic data in the real model and the virtual electric characteristic data in the virtual model under the same time step, and taking the relative deviation rate as the characteristic difference degree; The difference index forming unit is used for determining average characteristic difference degree and maximum characteristic difference degree based on the characteristic difference degree, forming a three-dimensional evaluation index system by combining the average characteristic difference degree, the maximum characteristic difference degree and the trend difference degree, and marking time steps exceeding a threshold value as difference indexes according to the three-dimensional evaluation index system; the anomaly judging unit is used for inputting the difference index into a bidirectional LSTM model, judging the anomaly type of the electric energy meter by combining a dynamic decision boundary optimized by a Q-learning algorithm, and obtaining an anomaly type judging result; And the self-adaptive updating unit is used for triggering XGBoost and the multi-attention fusion model to update parameters according to the abnormal type judging result so as to realize self-adaptive optimization of the model.
10. A computer-readable storage medium storing a computer program, wherein the computer program when executed by a processor implements a method for adaptively identifying abnormal states of an electric energy meter according to any one of claims 1 to 8.

Description

Electric energy meter abnormal state self-adaptive identification method, system and medium Technical Field The invention relates to the technical field of electric energy meters, in particular to a method, a system and a medium for self-adaptive identification of abnormal states of an electric energy meter. Background Along with the rapid development of smart power grids, an electric energy meter is used as core equipment for electric power metering and state monitoring, and the stability of the running state of the electric energy meter directly influences electric power transaction fairness and power grid safety. The traditional electric energy meter is characterized in that the abnormality identification is mostly dependent on basic electric characteristics such as voltage and current, the ageing of components in the electric energy meter is omitted, the metering deviation is caused, the electric network line impedance is affected by heavy rain, strong wind and other weather, the electric parameters of the electric energy meter are indirectly caused to fluctuate, and the correlation model of the electric characteristics, the environment and the weather is not established in the prior art, so that the abnormal misjudgment rate is higher. In view of this, the present application has been made. Disclosure of Invention The technical problem to be solved by the invention is that the existing electric energy meter anomaly identification method omits the fact that the high-temperature and high-humidity environment is easy to cause ageing of components in the electric energy meter, metering deviation is caused, and weather such as heavy rain, strong wind and the like can indirectly cause fluctuation of electric parameters of the electric energy meter by influencing the impedance of a power grid line, but the prior art does not establish a correlation model of electric characteristics, environment and weather, so that the anomaly misjudgment rate is higher. The invention aims to provide a self-adaptive identification method, a self-adaptive identification system and a self-adaptive identification medium for an abnormal state of an electric energy meter, which are used for carrying out self-adaptive identification on the abnormal state of the electric energy meter by integrating multidimensional feature screening and deep learning. The method comprises the steps of constructing a real model and a virtual model, providing real electric characteristic data, environment interference data and weather data for the virtual model by the real model, generating theoretical normal multidimensional electric data in a non-abnormal state by the virtual model according to a mapping rule from weather to environment and then to electric characteristics of the received real data, calculating electric data characteristic differences of the real model and the virtual model by relative deviation rates, calculating trend difference degrees of the multidimensional electric data of the real model and the virtual model by combining pearson correlation coefficients and DTW distances, and accurately capturing double problems of data deviation and trend abnormality by constructing a three-dimensional evaluation index system of average characteristic difference degrees, maximum characteristic difference degrees and trend difference degrees, thereby reducing misjudgment rates and missed judgment probabilities in complex scenes. The invention is realized by the following technical scheme: In a first aspect, the present invention provides a method for adaptively identifying an abnormal state of an electric energy meter, where the method includes: acquiring basic data of the electric energy meter, wherein the basic data comprises multidimensional electrical characteristic data, environmental interference data and weather data corresponding to an installation position; inputting XGBoost basic data and training a multi-attention fusion model to generate a real model; Calculating the relative deviation rate and trend difference degree of the electrical characteristic data in the real model and the virtual electrical characteristic data in the virtual model at the same time step, and taking the relative deviation rate as the characteristic difference degree; Determining average characteristic difference degree and maximum characteristic difference degree based on the characteristic difference degree, forming a three-dimensional evaluation index system by combining the average characteristic difference degree, the maximum characteristic difference degree and the trend difference degree, and marking the time step exceeding a threshold value as a difference index according to the three-dimensional evaluation index system; Inputting the difference index into a bidirectional LSTM model, combining a dynamic decision boundary optimized by a Q-learning algorithm, judging the abnormal type of the electric energy meter, obtaining an abnormal type judgment re