Optimal predictive power management strategy for fuel cell electric vehicles using neural networks in real-time

Abstract

Optimal predictive power management strategies (PMSs) for hybrid electric vehicles have a significant potentiality to achieve near-optimal solutions in real-time. Providing a priori prediction for power demand and implementing simplified, yet accurate, driveline models, to yield an optimal control strategy online are key challenges for predictive power management algorithms. Finding suitable solutions to resolve these challenges contributes to the ability of real-time PMSs to define efficient power handling strategies and hence promote better energy efficiency in electrified powertrains. This paper presents a neural networks-based predictive PMSs for fuel cell vehicles. The proposed method implements two types of networks, time-delay and nonlinear autoregressive network with exogenous inputs, to generate the required predictive models for the PMS. The online control module investigates an optimal power split strategy over the predicted horizon, considering minimal energy consumption and on-board charge retention. For comparative evaluation, rule-based method and the global optimal solution for a test driving cycle are considered. Results analysis revealed the ability of proposed method to yield an improvement of 20.71 % in energy efficiency without mitigating the state-of-charge on energy storage systems.