Neural network-driven multi-objective optimization for solid-state hydrogen sources dead-ended proton exchange membrane fuel cell power systems

To overcome the dual challenges of short endurance and poor environmental adaptability faced by portable mobile devices, this study proposes a dead-ended anode and cathode (DEAC), air-cooled proton exchange membrane fuel cell (PEMFC) power system based on online hydrolysis hydrogen generation. By integrating solid sodium borohydride hydrolysis hydrogen generation technology with a DEAC mode PEMFC, the power system constructs an internal "water-hydrogen-electricity" cycle, enabling the closed-loop utilization of reaction products. The cycle significantly enhances the system's energy density and liberates the system from dependence on external air. An artificial neural network-driven surrogate model is developed based on system simulation data. This model is coupled with a multi‑objective genetic algorithm to synergistically optimize key operating parameters: current density, temperature, purge duration, and purge interval. This multi-objective optimization framework is designed to simultaneously optimize three conflicting targets: electrochemical performance, water recovery, and oxygen utilization. Under the resulting optimal conditions, the proposed system outperforms traditional open‑cathode PEMFCs in dynamic voltage output, and its electrical efficiency is approximately 24.7% higher than that of traditional systems. Furthermore, in fixed‑endurance scenarios, the proposed system achieves a 66.55% higher gravimetric energy density than conventional high‑pressure hydrogen storage. This work provides theoretical and methodological support for developing next‑generation portable hydrogen power systems with high energy density and broad environmental adaptability.