Enhancing Electricity Theft Detection and Loss Prediction in Metro Cites Using Hybrid Machine Learning Model

The increasing level of urban power theft and distribution losses poses significant challenges to energy suppliers' income, grid stability, and environmental sustainability, especially in metropolitan areas characterized by high and varied consumption needs. This study presents a hybrid machine learning model with Neural Turing Machines that combines theft detection and loss prediction functionalities, specifically designed to address the problems of urban environments. The approach integrates classification methods with regression analysis and employs an external memory system to capture long-term usage patterns, providing great accuracy in detecting theft and forecasting losses. The findings indicate that the suggested model surpasses current methodologies, with a detection accuracy of 98.5%, an F1 score of 97.0%, and a root mean square error (RMSE) for loss prediction of 0.08. The findings proves that the model has potential to improve operational efficiency and reduce losses, becoming it a significant asset for energy suppliers in urban regions.