Paper Contents
Abstract
The detection of electricity theft (ETD) plays a crucial role in ensuring cost-efficiency within smart grids. However, current ETD methods struggle to handle the massive volume of data available today, facing challenges such as missing values, high variance, and non-linearity. Additionally, an integrated infrastructure is necessary to synchronize various processes in electricity theft classification. To address these issues, a novel ETD framework is proposed, incorporating three distinct modules. The first module addresses missing values, outliers, and unstandardized electricity consumption data. The second module introduces a newly proposed hybrid class balancing approach to tackle the issue of highly imbalanced datasets. The third module employs an enhanced artificial neural network (iANN)-based classification engine, which predicts electricity theft cases accurately and efficiently. To improve the performance of standard ANNs in handling more complex classification tasks using smart meter (SM) data, we propose three unique mechanisms: hyper-parameter tuning, regularization, and skip connections. Additionally, different iANN structures are explored to enhance the generalization and function fitting capabilities of the final classification model. Numerical results from real-world energy consumption datasets demonstrate that the proposed ETD model outperforms existing machine learning and deep learning methods, offering effective solutions for industrial applications.
Copyright
Copyright © 2025 MALLARAPU POOJITHA. This is an open access article distributed under the Creative Commons Attribution License.
