Abstract

This paper presents a nonmyopic energy management strategy (EMS) for controlling multiple energy flow in fuel cell hybrid vehicles. The control problem is solved by convex programing under a partially observable Markov decision process-based framework. We propose an average-reward approximator to estimate a long-term average cost instead of using a model to predict future power demand. Thus, the dependence between the system closed-loop performance and the model accuracy for predicting the future power demand is decoupled in the energy management design for fuel cell hybrid vehicles. The energy management scheme consists of a real-time self-learning system, an average-reward filter based on the Markov chain Monte Carlo sampling, and an action selector system through the rollout algorithm with a convex programing-based policy. The performance evaluation of the EMS is conducted via simulation studies using the data obtained from real-world driving experiments and its performance is compared with three benchmark schemes.