Self-Learning Charging Networks Using Reinforcement Learning Agents For Intelligent Electric Vehicle Charging Infrastructure

Authors: Vishu, Dr Raj kumar

Abstract: The rapid proliferation of electric vehicles (EVs) is imposing unprecedented stress on existing power grid infrastructure, necessitating intelligent, adaptive, and scalable charging management systems. Conventional rule-based and heuristic scheduling approaches fail to accommodate the stochastic nature of EV arrival patterns, fluctuating renewable energy availability, and real-time grid constraints. This paper proposes the Adaptive Multi-Agent Self-Learning Charging Network (AMSLCN), a novel multi-agent reinforcement learning (MARL) framework designed to optimize EV charging operations across distributed charging station networks. AMSLCN employs a decentralized execution with centralized training (DECT) paradigm, in which each charging station hosts an independent Deep Q-Network (DQN) agent that learns optimal scheduling policies through continuous interaction with a simulated smart grid environment. The proposed framework jointly optimizes charging efficiency, energy cost minimization, user waiting-time reduction, grid stability, and renewable energy utilization. The mathematical formulation of the problem is cast as a Markov Decision Process (MDP), with carefully designed state representations, action spaces, and a composite reward function that encodes multiple operational objectives. Extensive simulation experiments, conducted using real-world EV charging datasets from the ACN-Data repository and synthetic grid load profiles, demonstrate that AMSLCN achieves a 34.7% reduction in average energy cost, a 41.2% improvement in charging efficiency, a 38.5% decrease in user waiting time, and a 29.3% increase in renewable energy utilization compared to the best-performing baseline. AMSLCN significantly outperforms rule-based scheduling, genetic algorithm-based optimization, fuzzy logic controllers, standalone Q-Learning, and single-agent DQN across all evaluation metrics, establishing it as a compelling foundation for next-generation intelligent EV infrastructure.