Multi-agent fuzzy Q-learning-based PEM fuel cell air-feed system control

In this study, a novel ultra-local model (ULM) control structure using multi-agent system fuzzy Q learning (MAS-FQL) is proposed for the air-feed system of a polymer electrolyte membrane fuel cell (PEMFC). The primary aim of the control goal is to optimize the net power output of the fuel cell while also preventing oxygen starvation. This is achieved by effectively managing the oxygen excess ratio to maintain it at its optimal value, particularly during rapid load fluctuations. In this study, a new advanced control structure for PEMFCs is first presented to effectively manage the oxygen excess rate in the PEMFC system. This work uses an ULM technique in conjunction with an extended state observer (ESO) to effectively manage the control-related concerns connected with the PEMFC. Furthermore, the inclusion of the MAS-FQL has been used to dynamically manage the gains of the ULM controller in an online adaptive manner. The analysis findings demonstrate that the controller exhibits robustness and has satisfactory performance when subjected to load fluctuations. Across all scenario assessments, the proposed controller consistently exhibits an improvement in oxygen excess ratio regulation of more than 31.32% compared to the proportional integral derivative (PID) controller, more than 17.51% compared to the model-free sliding mode control (SMC) controller, and more than 11.40% compared to the fuzzy PID controller across different performance criteria.