SG-GAN for Electric Vehicle Road Networks and Q-Learning in Energy Efficient Routing

This paper suggests a novel routing scheme for multiple electric vehicles (EVs) on a mesh-type road network, which is generated using a Spatial Graph-Generative Adversarial Network (SG-GAN). The scheme focuses on optimizing both the routing of EVs and their association with the respective charging stations(CSs) to reduce energy consumption. The challenge lies in effectively managing factors such as travel cost, the state of charge (SoC) of the battery, and congestion in the road network. To address this problem, we propose a Reinforcement Learning (RL) based approach that minimizes energy consumption and the overall travel time for EVs to reach their destinations. A Q-learning algorithm for RL is employed to solve the nonlinear constrained optimization problems. Through simulation experiments on a road network evolved through SG-GAN, which generates realistic and complex network topologies, we address the allocation of CSs and EVs optimal routing. The results demonstrate the effectiveness of our approach compared to the existing methods, achieving a 2.68% reduction in energy consumption along with a similarity measure, on an average Kullback–Leibler (KL) distance (divergence) 0.54 in the SG-GAN generated road network from the reference real one.