Machine Learning-Based Detection of Non-Technical Losses in Power Distribution Networks

This study focuses on the serious sustainability and reliability problem caused by non-technical losses (NTL) due to energy theft in electrical grid systems. In order to reduce these losses, we propose an artificial intelligence-based approach that utilizes deep learning architectures in the detection of different types of leakage (voltage leakage, current leakage and voltage-current leakage). Unlike the studies in the literature, the data set is converted into two-dimensional matrices and analyzed with today's popular approaches, Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) models; CNN surpassed LSTM's 64.17% accuracy rate with 97.50% accuracy rate. In addition, from the classical methods, 67.5 accuracy rate was obtained with the k-Nearest Neighbor (k-NN) method and 62.25 accuracy rate was obtained with the Support Vector Machines (SVM) method. Comparisons with such traditional methods have revealed the superiority of CNN in determining complex leakage patterns. The findings highlight the potential of CNN to be used as a reliable tool for real-time theft detection by integrating it into smart grid systems. Future research will aim to further increase the scalability and effectiveness of this solution by examining the integration of real-time data and hybrid model approaches.