A DATA-DRIVEN POWER LOSS PATTERN RECOGNITION FOR SUSTAINABLE ENERGY MANAGEMENT

Abstract

Power loss remains a major challenge in smart grids (SGs), making accurate prediction models vital for sustainable energy management.This study employs deep neural networks (DNNs) to predict power loss using key attributes such as temperature, grid load, and environmental factors.Principal component analysis (PCA) identified grid temperature and load as the most influential variables, contributing 40.6% and 15.93% of the total variance, respectively, with selected features accounting for 76.28% overall.Comparative evaluation of DNN architectures showed that the 6layer model outperformed configurations with fewer layers, achieving an R² of 94.5%, the lowest MSE (1.00E-03), RMSE (3.40E-02), and MAPE (4.83%).Although it required slightly longer processing time, its superior predictive accuracy justified its selection.Pearson correlation analysis revealed weak positive relationships between temperature, voltage, and power loss, while regional analysis demonstrated that rising temperatures increase consumption and losses.Overall, the results demonstrate that the proposed DNN-based approach provides a robust and data-driven solution for power loss prediction, supporting improved grid efficiency and sustainability, with future work focusing on realtime data integration and additional environmental factors.