Particle Swarm Optimization-Based Neural Network Control of PEM Fuel Cell Air Supply System

To boost the net power output of proton exchange membrane (PEM) fuel cell systems under variable operating conditions, this study proposes an adaptive neural network (NN) control strategy that integrates parameter optimization. The air supply subsystem is the primary focus, as its performance is crucial to the overall net power. First, a comprehensive model of the air supply subsystem is developed, along with a detailed analysis of cathode pressure, oxygen excess ratio (OER), and net power output. Then, a two-dimensional particle swarm optimization (TDPSO) algorithm is used to optimize the reference signals for cathode pressure and OER, thereby maximizing net power. By applying input–output linearization techniques, the originally coupled nonlinear multi-input multi-output (MIMO) system is decoupled and transformed into a canonical form. Based on this transformation, an adaptive NN controller is designed to regulate the pressure valve and compressor. A series of hardware-in-loop (HIL) tests confirm that the proposed control strategy effectively optimizes net power across diverse operating scenarios. Quantitative results show that the proposed method achieves a net power output of 28.6 kW to 42.1 kW over the tested current range of 100–300 A. Meanwhile, the comparisons show that the proposed controller achieves OER tracking with root mean square error (RMSE) of 0.1221 and cathode pressure with RMSE of 0.0033. In comparison, the fuzzy logic controller (FLC) achieves OER with RMSE of 0.1453 and pressure with RMSE of 0.0044, while proportional–integral–derivative (PID) controller achieves OER with RMSE of 0.2133 and pressure with RMSE of 0.0109.