Cascade feedforward neural network based deep greedy search: An efficient MPPT strategy for proton exchange membrane fuel cell system

The operational performance of proton exchange membrane fuel cells (PEMFCs) is highly sensitive to variations in operating conditions such as temperature and fuel pressure, which severely limit their power generation efficiency. Conventional maximum power point tracking (MPPT) methods, including metaheuristic algorithms (MhAs) and mathematical approaches, operate within an unknown solution space, leading to low efficiency and excessive power fluctuations. To address these challenges, this study develops a data-driven MPPT approach for PEMFC systems that integrates neural networks with an efficient search algorithm to achieve stable and rapid optimization under dynamically changing conditions. Specifically, a cascaded feedforward neural network (CFNN) is employed to model the nonlinear mapping between the input duty cycle and output power, while a deep greedy search (DGS) algorithm adaptively narrows the search range to locate the global maximum power point (GMPP) and provides high-quality incremental training samples for the CFNN. Compared with seven advanced MhAs, the proposed CFNN-DGS method effectively mitigates power fluctuations and slow convergence caused by random initialization during global search, demonstrating superior adaptability to rapidly varying environments. Furthermore, validation across six representative scenarios—including start-up, step changes, long-term stochastic variations, sensitivity analysis, stability testing, and economic evaluation—confirms that CFNN-DGS achieves excellent optimization performance, enhances production efficiency, reduces power oscillations, and lowers economic cost. Finally, the engineering feasibility of the proposed method is verified through hardware-in-the-loop (HIL) experiments.