Reinforcement Learning and Metaheuristic Optimization Approach to PID with Derivative Filter Controller Design for PEMFC-Driven SEPIC Converter

Abstract The increasing interest in renewable energy sources has sped up research on power electronics-based control systems for these sources. In this study, control of Single Ended Primary Inductor Converter (SEPIC) converter connected to the output of a Proton Exchange Membrane Fuel Cell (PEMFC) is addressed using a PID with derivative filter controller. In the first stage, SEPIC converter is modeled, after that, its parameters are determined with respect to maximum power point (MPP) of PEMFC. A PID controller with derivative filter, which has not been previously applied to this system in the literature, is designed using recent optimization algorithms such as Artificial Protozoa Optimization (APO), African Vulture Optimization (AVOA), Mountain Gazelle Optimizer (MGO), Geometric Mean Optimizer (GMO), Crayfish Optimization Algorithm (COA), and Artificial Hummingbird Algorithm (AHA). The performance of each controller was compared statistically for Integral Absolute Error (IAE), Integral Square Error (ISE), Integral Time-Weighted Absolute Error (ITAE), Integral Time-Weighted Square Error (ITSE) with constraint and then Friedman test was performed for significance. Among them, the most up-to-date APO-based controller is further enhanced using a Reinforcement Learning (RL) approach. In this context, state, action, and reward structures are defined for RL. The action space consists of weighting the $$K_p$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:msub> <mml:mi>K</mml:mi> <mml:mi>p</mml:mi> </mml:msub> </mml:math> and $$K_i$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:msub> <mml:mi>K</mml:mi> <mml:mi>i</mml:mi> </mml:msub> </mml:math> parameters of PID with derivative filter controller, and reward function is designed to be maximized based on the system states. Simulation results show that APO-optimized PID with derivative filter controller supported by RL successfully tracks MPP and exhibits superior control performance in terms of the defined performance criteria.