Research On Feature Extraction And Anomaly Detection Of User-side Power Consumption Data

With the rapid development of national economic construction and power system,the data of user-side electricity consumption is also growing exponentially.Through data mining technology,these data are analyzed and studied,which is conducive to promoting the development of power grid intelligence.However,due to the aging of transmission lines,the failure of metering equipment and the fact that some users pay less or no electricity fees,the abnormal electricity consumption may occur in the actual power supply and utilization process.If these phenomena do not take timely measures will seriously affect the order of power supply power supply security and state economy will also cause serious threat.In order to solve the problems mentioned above,this paper proposes an anomaly detection model of electricity consumption data,which integrates isolated forest and local outlier factor algorithm,by analyzing the reason of generating abnormal data,the behavior of electricity consumption anomaly and the defects of the existing anomaly detection methods of power consumption data.The model takes full account of the possible global anomaly points and local anomaly points in the power consumption data.The main work contents of this article are arranged as follows:(1)Firstly,the user electricity data is classified by fuzzy clustering algorithm,that is,the power users with the same electricity consumption behavior are divided into the same class.It avoids the situation that a user with abnormal behavior of electricity consumption is hidden in another normal electricity consumption data,which will lead to leakage and wrong judgment of the abnormal detection model of electricity data.(2)By analyzing the characteristics of time series and the inherent laws of power consumption data,this paper extracts the characteristics of power consumption data from four power consumption indicators,namely,statistics,trend,variability and load patterns,and uses principal component analysis to reduce the dimensionality of the extracted characteristics of different types of users,eliminating the correlation between characteristic variables.Then the operation efficiency of the power consumption anomaly detection model is improved.(3)Taking advantage of the characteristics of the isolated forest algorithm that it does not need to calculate the relevant distance and density,has less adjustment parameters,runs fast,has low system overhead and the local outlier factor algorithm has high discrimination accuracy in local abnormal data detection,the power consumption data anomaly detection model based on the isolated forest algorithm and the local outlier factor algorithm is constructed.This model considers the global anomaly points and local anomaly points that may exist in power consumption data sets comprehensively and provides a new method for anomaly detection of power consumption data.(4)Finally,ROC curve,P-R and other related indicators are used to compare the anomaly detection model in this paper with other single detection models.The results show that the anomaly detection ability of the model which fuses multiple methods is significantly higher than the other single model,which proves the feasibility and applicability of the proposed method in electricity data anomaly detection.