Deep reinforcement learning for electric vehicle routing: A proximal policy optimization approach with a token-based reward system

The widespread adoption of electric vehicles (EVs) faces significant challenges due to limited driving range and insufficient charging infrastructure, complicating route planning and increasing travel costs. Although various optimization algorithms have been proposed for the Electric Vehicle Routing Problem (EVRP), critical issues such as trip duration, travel cost, and charging logistics remain. This research presents a novel approach based on Proximal Policy Optimization (PPO) to optimize EV routing, considering factors such as battery capacity, charging time, and travel duration. To enhance system efficiency and sustainability, the framework integrates Distributed Generation (DG) modeling to reduce grid dependency and generation costs. Furthermore, a token-based reward system is introduced to incentivize timely charging, improving charging station utilization and promoting demand-side flexibility. The proposed PPO framework is evaluated against state-of-the-art algorithms on multiple benchmark scenarios, demonstrating up to a 35% improvement in routing efficiency while Distributed Generation–Demand Side Management (DG-DSM) coordination reduces marginal charging prices by approximately 14% and improves the load factor from 0.536 to 0.579. A Monte–Carlo sensitivity analysis using load variations from an Indian Utility 17-bus system, combined with probability density function to assess robustness under stochastic traffic and grid fluctuations. The results underscore its scalability and potential as a practical solution for sustainable EV route planning.