Abstract

Rapid EV growth requires efficient charging infrastructure and routing for urban sustainability, vital for reducing congestion, wait times, and energy use while promoting renewables. This paper presents a two-stage Deep Q-Network (DQN) framework for optimal EV station placement and routing. A DQN placement model minimizes future route energy by selecting station locations; a second DQN guides vehicles by approximating cost-to-go functions. Our DQN approach outperforms baselines (Q-Learning, NN, DGNN, Random Walk) in energy consumption, travel time, and route success. DQN placement achieved 1.2387 kWh average energy and 6.34 km average distance. DQN routing delivered 1.2587 kWh average energy, 1.98 average hops, and 96.67% success, demonstrating our DRL's effectiveness for scalable, reliable, and energy-efficient EV infrastructure planning.