Improving power theft detection using efficient clustering and ensemble classification

One of the main concerns power generation systems around world is theft. This research proposes a framework that merges clustering and classification together in order to theft detection. Due fact most datasets do not have abnormal samples or are few, we added original using artificial attacks create balance increase correct detection rate. We improved crow search algorithm (CSA) used weight feature Crows improve performance phase. Also, between diversification intensification, calculated awareness probability parameter (AP) dynamically at iterations algorithm. To evaluate performance, cross validation technique stacking its training The results extensive experiments on three reference showed high detect evaluation if data collected correctly sufficiently, this can effectively any actual grid. for new attacks, their patterns be detected from data, it easily possible implement these types attacks.