Enhancing Hybrid Microgrid Dynamics Using an Agent-Based Reinforcement Learning (RL) Framework

Abstract

Hybrid microgrids, integrating renewable, and conventional energy sources are critical for sustainable and resilient power systems. Their dynamic performance is affected by uncertainties in load demand, generation variability, and control strategies. This paper investigates the performance of a grid-connected inverter in a hybrid microgrid and compares different controllers, including Artificial Neural Network (ANN), Adaptive Neuro-Fuzzy Inference System (ANFIS), and a Reinforcement Learning (RL) agent. The proposed system integrates solar panels and wind turbines with traditional sources such as batteries and fuel cell stacks, with maximum power extraction achieved using a hill-climb MPPT technique. Four converters regulate the microgrid DC link voltage, and the RL agent's performance is evaluated under both static and dynamic conditions. Simulation results, validated in MATLAB/Simulink, demonstrate that the RL agent outperforms ANN and ANFIS controllers in terms of stability, power quality, and dynamic response.