Research On The Feature Extraction Of Metering Data On The Power User Side And The Detection Of Electricity Theft

With the rapid development of the economy,China has entered the stage of fully building a smart grid.The problem of electricity theft is accompanied by the development of China's electric power industry,which has caused major losses to the national economy and also prevented the safe operation of the power grid.Stealing electricity from conventional stealing to high-tech stealing,showing diversification,concealment,and high-tech features.Manual inspection of electricity theft is not only time-consuming and labor-intensive,but also difficult to obtain evidence for high-tech means of stealing electricity.Aiming at this problem,the traditional stealing principle and high-tech stealing characteristics are analyzed,and the tamper detection model of "classification-feature extraction-abnormal detection-stealing detection" is constructed.Analyze the user's historical electricity consumption data,first classify the user's load curve,calculate the four indicators of the measurement load curve and obtain 15 characteristic variables;analyze the principal components of the extracted feature variables based on the classification(Principal Component Analysis,PCA)And reduce the dimension,calculate the local outlier factor(LOF)to filter the abnormal users of electricity;finally,use 8 tamper discriminant indicators and extract the principal components for the normal power users and power users,and use them as training.The sample input Extreme Learning Machine(ELM)stealing detection model.The main research contents of this paper are as follows:(1)The power consumption data of industrial power users in Yunnan Province and the measurement data related to the tampering behavior are adopted.According to the definition of stealing electricity,analyze the principle of stealing electrici ty and determine the characteristics of the stealing characteristics used.Analyze the load curve characteristics and load characteristics indicators to determine the characteristic indicators used for abnormal power consumption detection.Due to the regularity of the electricity consumption of industrial users,the average daily load curve of the same time point is calculated to obtain a typical daily load curve.The curve is clustered,and the user load characteristic curve composed of the central points is obtained,and the subsequent detection is performed on the basis of classification.(2)Anomaly detection model based on principal component analysis of feature indicators is constructed for screening abnormal users of power users for the problem of users including power stealing.For the abnormal users of electricity consumption,calculate the typical daily load curve of the user and the similarity,shape,trend and other indicators of the load characteristic curve to obtain 15 characteristic variables.On the basis of classification,analyze the principal component accumulation contribution rate of the obtained characteristic variables.The first two principal components with large contribution rate are used as coordinate axes,and the user maps the two-dimensional coordinate axes to the two-dimensional coordinate axes to realize the dimensionality reduction of the feature variables.Finally,the local outliers are calculated to realize the screening of the power users.(3)To detect tampering,a PCA-ELM tamper detection model is built on the basis of anomaly detection.Selecting some of the normal power users and the power users with abnormal power stealing indicators as the sample data,the sample data is extracted into the ELM model through the principal component extraction,and the suspected stealing users are detected.Experiments show that compared with the traditional BP neural network and ELM without principal component analysis optimization,PCA-ELM has improved in terms of running time and accuracy,and provides a new idea for discovering electricity stealing behavior.