JADE-Based Feature Selection for Non-technical Losses Detection

Nowadays, non-technical losses, usually caused by thefts and cheats in the energy system distribution, are among the most significant problems an electric power company has to face. Several actions are employed striving to contain or reduce the implications of the conducts mentioned above, especially using automatic identification techniques. However, selecting a proper set of features in a large dataset is essential for successful detection rate, though it does not represent a straightforward task. This paper proposes a modification of JADE, an efficient adaptive differential evolution algorithm, for selecting the most representative features concerning the task of computer-assisted non-technical losses detection. Experiments on general-purpose datasets also evidence the robustness of the proposed approach.