SARSA-Based Voltage Regulation of PV-Buck Converter System Under Multi-Nature Load Conditions

DC microgrids are experiencing rapid evolution due to increasing renewable energy integration and diverse load requirements, necessitating advanced control strategies for stable operation. This paper presents a novel voltage regulation approach utilizing policy gradient reinforcement learning with SARSA algorithm for a PV-fed buck converter system. The proposed methodology addresses the complex dynamics of multiple load types through a hierarchical control framework that combines actor-critic networks for continuous control actions. The control strategy employs a comprehensive seven-parameter state space encompassing voltage error, current measurement, solar irradiance, duty cycle history, voltage dynamics, error integration, and average voltage deviation. The experimental validation employs a buck converter maintaining 48V DC output from a 100V PV source, tested under challenging scenarios including 100W constant power load, 50W pulsed power load, and variable linear load conditions. The results demonstrate significant advantages over traditional PI control: 83% reduction in steady-state error, 75% improvement in voltage stability, and 73% faster load transition response. The controller successfully maintains voltage variations within +-1V across diverse load combinations, effectively addressing the destabilizing effects of constant power loads and sudden pulsed load changes. The integration of policy-based learning with on-policy SARSA implementation enables real-time adaptation without prior system modeling requirements. This approach showcases exceptional performance in both steady-state maintenance and transient response, particularly in scenarios involving mixed load characteristics. The framework provides a robust solution for modern DC microgrid applications requiring stable voltage regulation under complex load dynamics.