An equivalent consumption minimization strategy based on 1-step look-ahead stochastic dynamic programming

M. Fleuren, T.C.J. Romijn, M.C.F. Donkers

Research output: Contribution to journalConference articlepeer-review

6 Citations (Scopus)
598 Downloads (Pure)


In this paper, a systematic procedure to determine the equivalence factor in the equivalent consumption minimisation strategy (ECMS) is proposed. This is relevant when ECMS is not only used for controlling the power split between the internal combustion engine and the electric machine of a hybrid electric vehicle (HEV), but also for controlling several auxiliary systems. In this case, the number of controlled components and energy buffers increases, which causes the number of tunable equivalence factors to increase. The procedure to determine the equivalence factors proposed in this paper is based on the observation that ECMS can be considered as a time-invariant feedback policy and dynamic programming (DP) also yields a time-invariant feedback policy when the time horizon of the control problem approaches infinity and the disturbances are constant or absent. As the drive cycle can be considered as a (stationary stochastic) disturbance, we propose to formulate ECMS as the solution of a 1-step look-ahead stochastic dynamic program (1slSDP). This strategy results in an energy management strategy that performs close to optimal and yields similar fuel consumption results, when compared to a well-tuned ECMS. The absence of any parameters to be tuned and the fact that fuel consumption is similar to a well-tuned ECMS makes 1slSDP a useful strategy for energy management of HEVs.
Original languageEnglish
Pages (from-to)72-77
Issue number15
Publication statusPublished - 2015
Event4th IFAC Workshop on Engine and Powertrain Control, Simulation and Modeling (E-COSM '15), August 23-26, 2015, Columbus, OH, USA - Ohio State University, Columbus, United States
Duration: 23 Aug 201526 Aug 2015


Dive into the research topics of 'An equivalent consumption minimization strategy based on 1-step look-ahead stochastic dynamic programming'. Together they form a unique fingerprint.

Cite this