Adaptive Duty Cycle Control for Optimal Battery Energy Storage System Charging by Reinforcement Learning

Abstract

This paper works on adaptive duty cycle control of a Solar power system using a Reinforcement Learning approach for optimizing the charging of a 12 V 30 Ah Battery Energy Storage System (BESS). The Twin-Delayed Deep Deterministic (TD3) algorithm is used to train an agent that adaptively controls the duty cycle of a Pulse-Width Modulation (PWM) signal to maintain the output voltage of the Photovoltaic (PV) system in the optimal range for charging the BESS. Results from MATLAB Simulink show that the TD3 algorithm optimizes the charging of the BESS when compared to other techniques, such as a PID controller, achieving an SOC of around 50.33% from an initial value of 50% with low noise for 10 seconds, whereas the PID controller achieved around 50.06% with high voltage spikes. This work opens avenues for expanding the system to include other renewable energy sources and their interactions for improving power distribution.