Optimal sizing of the PV-BESS energy system for off-grid electric vehicle charging station using deep reinforcement learning techniques

Abstract

This research applies Deep Reinforcement Learning (DRL) techniques to determine the optimal sizing of PV systems and BESS for off-grid EV charging stations by comparing the performance of the PPO, A2C, and DQN algorithms. The study found that the PPO technique yielded the best results, reducing total costs and maximizing energy efficiency, thereby reducing electricity costs by 48.79%. In terms of economics, the project is investment-worthy with an NPV of 9.7 million baht, an IRR of 20.89%, a BCR of 1.804, and a payback period of 5 years. Environmentally, the system can reduce CO₂ emissions by up to 1,383 tons over the 20-year project lifespan. The developed off-grid system helps reduce dependence on fossil fuels, enhances energy security, and promotes sustainable energy use. Therefore, the PPO technique is the most suitable approach for sizing energy production systems and evaluating the viability of off-grid EV charging stations.