Tags:Dynamic Vehicle Charging, Electric Vehicles, PPO, Reinforcement Learning and TD3
Abstract:
This paper works on implementation of Three-Phase Solar Powered Dynamic Vehicle Charging for Electric Vehicles (EVs) using Reinforcement Learning to train two agents to monitor and control the power from Photovoltaic (PV) arrays to supply power to an EV when it crosses motion sensors detected by a Sensor agent, trained with the PPO algorithm to send control signals to the Converter agent, trained with the TD3 algorithm to take measurements from the Solar Power system to control the Duty Cycle of a Buck-Boost converter to supply power to the EV, or the Solar Battery Energy Storage System based on the control signal from the Sensor agent. The results found when compared with a PID controller, the agents were able to charge a 400 V 100 kWh EV traveling 100 km/h with an average charging step size of 902*〖10〗^(-6)% to the State-of-Charge every 0.16 s, while the PID controller achieved 775*〖10〗^(-6)%.
Comparative Analysis of Photovoltaic MPPT P&O Algorithm and Reinforcement Learning Agents Utilizing Fuzzy Logic Reward System