Reinforcement Learning-Based IoT System for Adaptive EV Charging and Pollution Reduction in Smart Urban Environments

Abstract

The growing trend toward electric vehicle adoption in urban situations is a constant, but the same cannot be said for charging infrastructure that has not been able to keep up with its traditionally volatile nature and relationship to environmental conditions. Current systems are predominantly based on inelastic schedules or single-objective optimization, resulting into peak load stress, wasted energy and unintended rise to localized pollution. The mentioned gap is addressed in this work by proposing a real-time adaptable Multiple Decision Making-based IoT framework for Electric Vehicles charging decisions, considering both grid conditions and air quality signals with Reinforcement Learning. It incorporates charging station data, traffic flow and pollution sensors into the system wrap a model of environment to be as a decision process with a continuous adjustment of charging action by PPO-based learning on that. The proposed strategy is to trade-off between energy-efficient and emission control by jointly optimizing them instead of separately optimizing. The system smooths the demand for charging by learning from temporal patterns and feedback from its surroundings, thus avoiding high-impact zones when pollution spikes. Experimental evaluation demonstrates strong performance—offering a 97.2% accuracy with low prediction errors (RMSE: 0.018, MAE: 0.012). It also decreases makespan by 28%, increases energy efficiency by 31% and decrease operational cost by 26%. It is also relevant to the environment as it lowers pollution by 22%. The proposed system is more stable because it adopts a clustering, supervised learning strategy compared to traditional rule-based, linear programming and even regular deep learning models; faster decision-making in terms of this procedure; less overfitting or adaptiveness which results in consistent multi-objective optimization. Such results suggest that adaptive, learning-based control may open an achievable pathway to smarter, cleaner urban charging systems.