A Hybrid Method for Non-Technical Loss Detection in Smart Distribution Grids

Abstract

Non technical losses (NTLs) are a major problem faced by DSOs all over the world. Traditional meter inspection techniques are not effective in detecting NTL, while the wide adoption of smart meters and smart grid technologies provide new opportunities for solving the problem. The proposed NTL detection system utilizes voltage sensitivity analysis, power system optimization, and support vector machines (SVMs) for correctly detecting NTLs in the distribution grid under various conditions. Breakout detection is used for extracting features from consumption time series and training an SVM classifier. Generalized least squares is used for estimating the network voltage self-sensitivities. Finally, the NTL detection problem is formed as a non-linear non-convex optimization problem solved with semi-definite programming relaxation. The three modules can be operated either autonomously or combined. The proposed NTL detection system is demonstrated under different scenarios to test its effectiveness.