Energy management of fuel cell hybrid vehicle based on partially observable Markov decision process

Abstract

This paper presents a nonmyopic approach for controlling multiple energy flow in fuel cell hybrid vehicles. The control problem is solved by convex programming under a partially observable Markov decision process based framework. We propose an average-reward approximator to minimise a long- run average cost instead of utilizing a model to predict the specific future power request. Thus, the dependency between the closed- loop performance properties of the system and the accuracy of the model to predict the future power request is decoupled in the new energy management strategy for fuel cell hybrid vehicles. The designed energy management strategy for fuel cell hybrid vehicles includes a real-time self-learning system, an average-reward filter based on the Markov chain Monte Carlo sampling method, and an action selector system through rollout algorithm with a convex programming based policy. The performance achieved by the developed approach is shown in simulation via real-world driving experiments and compared to the ones obtained by other three benchmark schemes.