Theft detection dataset for benchmarking and machine learning based classification in a smart grid environment

Abstract

Smart meters are key elements of a smart grid. These data from Smart Meters can help us analyze energy consumption behaviour. The machine learning and deep learning approaches can be used for mining the hidden theft detection information in the smart meter data. However, it needs effective data extraction. This research presents a theft detection dataset (TDD2022) and a machine learning-based solution for automated theft identification in a smart grid environment. An effective theft generator is modelled and used for obtaining a multi-class theft detection dataset from publicly available consumer energy consumption data, owned by the “Open Energy Data Initiative” (OEDI) platform. This is an important and interesting phase to explore in the smart grid field. The proposed dataset can be used for benchmarking and comparative studies. We evaluated the proposed dataset using five different machine learning techniques: k-nearest neighbours (KNN), decision trees (DT), random forest (RF), bagging ensemble (BE), and artificial neural networks (ANN) with different evaluation alternatives (mechanisms). Overall, our best empirical results have been recorded to the theft detection-based RF model scoring an improvement in the performance metrics by 10% or more over the other developed models.