Towards real-time optimal energy management of HEV powertrains using stochastic dynamic programming

This paper is concerned with the control of hybrid electric vehicles. In the vast literature on the subject, focus is put on the minimization of the fuel consumption. To do so, the deterministic optimal control theory is intensively used. The major concern regarding this approach is the need for driving conditions to be known in advance. In this paper, we suggest to consider the driving cycle as a stochastic process, and make use of the stochastic dynamic programming (SDP) to design the power load sharing. We formulate a weighted criterion to minimize fuel consumption and allow for drivability conditions at the same time. The resulting control law may easily be implemented online as a static mapping function of the powertrain state. We provide simulation results based on a parallel HEV case study. We use statutory cycles to build a database of randomly generated cycles. This allows to assess the relevancy of this approach by comparing optimal fuel consumption (using deterministic programming), for each cycle, to the score obtained via the SDP framework. These preliminary results show that the SDP is a promising control tool to overcome the drawbacks inherent to the classical approaches.