Abrupt user load change detection based on multiple features and LOF algorithm

Abstract

The sudden load changes impact power grids by frequency and power oscillations. In order to distinguish the complex and massive abnormal user load data， this paper proposes a method combining multiple features and LOF （local outlier factor） algorithm. Firstly， the statistical characteristic mean value， standard deviation， waveform characteristic dispersion coefficient， kurtosis， waveform factor and pulse factor of load data are extracted， and the effective features are obtained through dimensionality reduction of PCA （principal component analysis）. Furthermore， the LOF algorithm is used to detect abnormal user load data every day. This detection algorithm is used in the Zhejiang power data center based on Alibaba cloud. The results show that it can detect users with abrupt load changes in massive measured data at fixed times of every day and realizes online detection with high accuracy.

Keywords

• machine learning
• lof algorithm
• abrupt load change
• big data