“20th PSCC 2018 papers submission and review platform

Full Program »

Unsupervised Classification for Non-Technical Loss Detection

Electricity theft has been a major issue for DSOs for years. However, despite efforts to detect it and the application of legal deterrents, the phenomenon insists. NTL detection methods can be categorized as data oriented, network oriented and hybrids. Data oriented methods are divided to supervised and unsupervised methods. In this paper, unsupervised NTL detection techniques are tested on a smart meter data set provided by the Greek DSO, HEDNO (about 500 commercial consumers) and a publically available smart meter data set (about 5000 residential and commercial consumers). Frauds are simulated and the Twitter breakout detection toolbox is used for extracting features. The algorithm uses energy statistics to detect shifts in mean which will be combined with typical features already found in literature for detecting frauds. After fraud is simulated traditional clustering techniques (k-means) are used to group consumers according to their yearly consumption profile. Then, unsupervised classifiers such as rule systems, local outlier factor (LOF) and one-class SVM are demonstrated. Finally, a number of metrics are calculated for each data set and classifier such as accuracy, detection rate, precision, false positive rate and F1 score. In addition, the amount of stolen energy detected is proposed as a metric for NTL detection system studies.

Author(s):

George Messinis    
National Technical University of Athens
Greece

Nikos Hatziargyriou    
National Technical University of Athens
Greece

 

Powered by OpenConf®
Copyright ©2002-2014 Zakon Group LLC