Electric Power Systems Research 136 (2016) 262–269 Contents lists available at ScienceDirect Electric Power Systems Research j o ur nal ho me page: www.elsevier.com/lo cate/epsr Nonlinear autoregressive neural network in an energy management strategy for battery/ultra-capacitor hybrid electrical vehicles Mona Ibrahim a,b,∗ , Samir Jemei b , Geneviève Wimmer a , Daniel Hissel b a Laboratory of Mathematics of Besancon, 16 route de Gray, 25030 Besancon, France b University of Franche-Comté, FEMTO-ST (UMR CNRS 6174)/FCLAB (FR CNRS 3539), Rue Thierry Mieg, F90010 Belfort, France a r t i c l e i n f o Article history: Received 23 October 2015 Received in revised form 25 January 2016 Accepted 3 March 2016 Keywords: Artificial neural networks Wavelet transform Hybrid vehicles Energy management Time series predictions a b s t r a c t Hybrid electric vehicles are one of the most promising solutions for reducing pollution and fuel con- sumption. However, their propulsion system comprises a number of different onboard power sources with different dynamic characteristics, meaning that some strategy is required for sharing power between them that takes their characteristics into account. In this paper, a new real time energy management strategy for battery/ultra-capacitor hybrid vehicles is proposed. This strategy is based on sharing the total power between the onboard power systems, namely the battery and the ultra-capacitors, using a Nonlinear Auto-Regressive Neural Network (NARNN) as a time series prediction model, and Discrete Wavelet Transform (DWT) as a time-frequency filter. The objective of this strategy is to lengthen the life of the battery. We simulated this new strategy using actual data from a military hybrid vehicle. The results were found to be promising and show the robustness of the proposed method. © 2016 Elsevier B.V. All rights reserved. 1. Introduction Electrical Vehicles (EV) [1], Hybrid Electrical Vehicles (HEV) [2] and Fuel Cell Electrical Vehicles (FCEV) [3] have become impor- tant topics for researchers, in view of a number of environmental and energy issues [4]. Compared to conventional vehicles they have substantial advantages: by combining different power sources, out- put power can be greater and more efficient [5] and the braking energy can be recovered [6]. However, the widespread develop- ment and adoption of these vehicles has been held back by issues relating in particular to lifetime, energy density and cost of pur- chase [7]. HEVs can incorporate multiple power sources including fuel cells, electric motors, batteries, and ultra-capacitors. Each type of onboard power source has its own particular dynamics. For instance, batteries and fuel cells are suitable for low, continuous dynamic loads, while ultra-capacitors can tolerate rapid transients. Longer lifetimes and higher energy densities of the sources may be obtained through the use of an Energy Management Strategy (EMS). An EMS is a mathematical algorithm, implemented in the ∗ Corresponding author at: University of Franche-Comté, FEMTO-ST (UMR CNRS 6174)/FCLAB (FR CNRS 3539), Rue Thierry Mieg, F90010 Belfort, France. Tel.: +33 616184205. E-mail addresses: mona.ibrahim@univ-fcomte.fr (M. Ibrahim), samir.jemei@univ-fcomte.fr (S. Jemei), gwimmer@univ-fcomte.fr (G. Wimmer), daniel.hissel@univ-fcomte.fr (D. Hissel). vehicle, and designed to share the requested power between dif- ferent onboard power sources, taking into account their different dynamic characteristics. A number of EMSs have been developed in the literature. Among these EMSs, Artificial Neural Networks (ANNs) have proved to be accurate methods adapted to real applications in energy manage- ment for hybrid electrical vehicles. Since 1982, ANNs have been applied successfully to a variety of electric power systems. An energy management strategy using wavelet-neural network combination was proposed in [8]; the purpose in this study was to control power distribution in a hybrid Fuel Cell/Ultra-Capacitor (FC/UC) vehicular system. The role of the ANN was to perform the charge sustaining of the UC, leading to a reduction in fuel consump- tion. An adaptive intelligent energy management for plug-in hybrid electric vehicles, based on a neuro-fuzzy inference system, was pro- posed in [9]. This method enables real time adjustments of power for different onboard power sources, taking into account different road geometries, wind and environmental thermal conditions. In [10], a Radial Basis Functional Neural Network (RBFNN) model was used to eliminate the effect of battery degradation on its State- Of-Charge (SOC). A 6 Ah Lithium Ion battery was used for this study, and the RBFNN gave a more accurate SOC estimation, as well as robustness in relation to aging cycles, temperature and loading profiles. http://dx.doi.org/10.1016/j.epsr.2016.03.005 0378-7796/© 2016 Elsevier B.V. All rights reserved.