Comparative Evaluation of Exploratory Data Analysis Techniques for Power Theft Detection in Residential Distribution Grids

This paper presents a comparative evaluation of the performance of various Exploratory Data Analysis (EDA) techniques for detecting power theft in residential distribution networks. The techniques include Pearson Correlation Coefficient (PCC), Fuzzy C-Means Clustering (FCM) and Density-Based Spatial Clustering of Applications with Noise (DBSCAN). They are assessed using well-known classification assessment measures such as Accuracy, Geometric Truth Rate, Youden’s Index and Area under Receiver operating characteristics curve. In the present work, power theft scenarios are divided into four different categories. Suspicious consumers are then identified using the concepts of feature extraction, correlation, and clustering. The information needed to detect the corrupt customers comes from a monitoring system at the distribution transformer for a block and electricity load data from smart meters installed at customer homes in the same block. The analysis is tested on actual dataset from UK power networks and it is found that FCM outperforms the other techniques. Sensitivity analysis is further performed to determine how the technique performs significantly w.r.t. various parameters.