Adaboost-based electricity theft detection

Abstract

Traditionally, electricity theft detection is estimated by field crews and verified by on site visit. It is rather difficult to enhance the precision to locate electricity theft. Extensive application of smart meters facilitates data-driven based electricity theft detection. However, electricity theft detection approaches based on the premise that electricity usage decline due to electricity theft could bring about higher false positive rate and cannot be put in use for industrial applications. AdaBoost based electricity theft identification algorithm is proposed in this paper for low false positive rate. The cross validation is employed to optimize hyper-parameter of Adaboost to low the false positive rate. Numerical simulation with metering data of marketing system and metering system of power utilities indicates that the proposed approach outperform SVM and BP based approach in either false positive or precision.