Intelligent classification of smart city sectors from IoT-based energy consumption data using AI-based methods

Abstract

Efficient energy management is a critical component of smart city systems, yet most existing studies focus on demand forecasting rather than sector-level classification of energy consumption. This work addresses this gap by proposing a hybrid Artificial Intelligence (AI) framework for classifying urban sectors based on IoT-derived electricity usage patterns. Using hourly consumption data from the Tamil Nadu Electricity Board, nine supervised Machine Learning (ML) models were evaluated, with Gradient Boosting achieving the best baseline accuracy of 96.9%. Advanced Deep Learning (DL) architectures, including LSTM, CNN, and Transformer models, further improved performance. A Reinforcement Learning (RL) layer was integrated to adaptively tune hyperparameters and guide model selection during training based on performance feedback. The resulting RL-enhanced Transformer achieved a maximum classification accuracy of 99.2%. The findings demonstrate that combining deep representation learning with adaptive optimization improves classification accuracy and stability, offering practical value for sector-aware energy planning, load prioritization, and data-driven decision support in smart city energy management.