Optimal tuning and neural emulation of MPC for power management in fuel cell hybrid electric vehicles

Power management in a fuel cell hybrid electric vehicle (FCHEV) consists of splitting efficiently the power generated by the battery and the fuel cell (FC), ensuring that the net delivered power meets the total power requested by the vehicle. In this work, we investigate the use of Model Predictive Control (MPC) to perform such power management task. The proposed MPC scheme features a novel approach for the optimal choice of the cost function weights, based on Particle Swarm Optimization (PSO), in order to achieve multiple control objectives, such as requested power tracking and minimum supplies consumption. Also, the MPC controller employs a neural data-driven approximation of the real plant as internal prediction model; this makes the controller employable in real-case scenarios, where only input-output plant measurements are available. We also present an alternative controller, based on feedforward neural networks (NNs), which emulates the whole MPC-based optimal control law. The NN-MPC controller is able to reliably reproduce the control action of the original controller, with a significantly lower computation time, making it suitable for real-time implementation on low-end control units.