Cheby-FENN: a Chebyshev-enhanced feature expansion neural network for electricity theft detection

Abstract Electricity theft is a major contributor to non-technical losses in power systems, causing substantial financial losses and potentially affecting grid reliability. This paper proposes a lightweight neural network framework that incorporates a third-degree Chebyshev polynomial feature-expansion layer at the input stage to enhance the representation of nonlinear electricity consumption patterns. The proposed framework is evaluated on two publicly available benchmark datasets, Open Energy Data Initiative (OEDI), and State Grid Corporation of China (SGCC), representing binary and multi-class electricity theft detection scenarios, respectively. A comprehensive ablation study examines the effect of Chebyshev polynomial degree, preprocessing components, and alternative nonlinear feature-expansion techniques. The results show that the third-degree Chebyshev configuration provides a favorable balance between predictive performance and computational complexity among the evaluated Chebyshev configurations. The proposed configuration achieves F1-scores of 96.8% on OEDI and 98.8% on SGCC, with corresponding accuracies of 99.3% and 99.6%, respectively. Five-seed five-fold cross-validation further demonstrates consistent performance across different stochastic realizations. Comparative experiments with standard polynomial, Legendre, Fourier, and radial basis function expansions indicate that some alternatives can achieve comparable or higher performance on individual metrics, while the proposed Chebyshev configuration maintains competitive detection performance with lightweight neural architecture. These findings demonstrate the potential of Chebyshev-based feature expansion as an efficient alternative for enhancing nonlinear feature representation in electricity theft detection.