Statistical Learning applied to the energy management in a Fuel Cell Electric Vehicle M. Cavalletti * J. Piovesan ** C. T. Abdallah ** S. Longhi * P. Dorato ** G. Ippoliti * * Universit` a Politecnica delle Marche, DIIGA - Via Brecce Bianche, 60131 Ancona, Italy. cavalletti@diiga.univpm.it {g.ippoliti, sauro.longhi}@univpm.it ** Universityof New Mexico, Department of Electrical and Computer Engineering, Albuquerque NM 87131, USA {jlpiovesan,chaouki,peter}@ece.unm.edu Abstract: The paper considers a high efficiency energy management control strategy for a hybrid fuel cell vehicle using neural networks and Statistical Learning theory. Hybrid Electric Vehicles may potentially improve fuel economy, reduce emission gases, and achieve performance similar to conventional cars. The use of different power sources and the presence of different constraints makes the power management problem highly nonlinear. Probabilistic and statistical learning methods are used to design the weights of a neural networks to minimize the fuel consumption during a given path. Numerical results are obtained using the model of a real hybrid car, “Smile ” developed by FAAM, using a stack of fuel cells as the primary power source in addition to ultracapacitors. The results are satisfactory in terms of fuel consuming and efficiency of ultracapacitors and batteries. Keywords: Fuel Cell Electric Vehicle; Statistical Learning; Neural Network Control. 1. INTRODUCTION The history of electric vehicles started with the invention of the battery by A. Volta and the discovery of elec- tromagnetic induction by M. Faraday. This culminated, in 1873, with the invention of the first electric vehicle (Westbrook [2005]). Even though the first vehicles were actually electric, gasoline and diesel cars overtook electric vehicles since the 19th century, thanks to their better energy-weight ratio. In recent years, the increase in the size and weight of passenger cars have made gasoline and diesel vehicles more pollutant and less efficient. In addition, the ever increasing cost of fuel as well as pollution problems are motivating car companies to look for new solutions to minimize fuel consumption and the production of polluting gases (O.Fuji [2002]). Hybrid Electric Vehicles (HEVs) actually combine the efficiency of electric cars with the high autonomy of con- ventional vehicles and are considered a potential solution to such problems. The combination of electric motors with various storage elements (i.e. fuel cell, thermal engine, ul- tracapacitors, etc..) brought about more complex systems, as well as different control strategies to manage the vehicle powertrain (Maggetto and Mierlo [April 2000]). Hybrid vehicle controllers are based on a supervisor that chooses, in the presence of different constraints, the best power path to satisfy the power demands of the drive line, while minimizing the fuel consumption and the produc- tion of the pollution gases. Various solutions were devel- oped in the literature in order to achieve different perfor- mances: Dynamic Programming and Quadratic Program- ming are used to minimize the fuel consumption over all paths (G Rizzoni [December 2003], Sciarretta and Guzzella [April 2007], Koot [2006]). Heuristic controllers, based on Boolean of fuzzy logic rules, are used to minimize the fuel consumption using different vehicular variables such as torque demand or car speed (N. Jalil and Salman [1997], Sciarretta and Guzzella [April 2007]). Artificial neural net- works have also been used to achieve various performance objectives during different driving cycles (J. Moreno and Dixon [2006],N. Jalil and Salman [1997]). An alternative solution to analytical optimization ap- proaches is provided by statistical learning methods (Koltchinskii et al. [2001]). In such an approach, a perfor- mance index is minimized empirically while guaranteeing that the difference between the empirical solution and the optimal one is arbitrarily small with high probability. In this paper, a neural network controller is proposed and Statistical Learning theory is used to choose the networks’ weights in order to reduce the fuel consumption during a given path. The controller is applied to a Fuel Cell Electric Vehicle (FCEV) called “Smile ” and produced by FAAM S.p.A. (Italy). The vehicle has fuel cell stacks, that convert hydrogen to electric power, using hydrogen as primary power source. A buffer of energy in the powertrain is provided by the lead battery pack and ultracapacitors. The performance of the proposed controller is evaluated via numerical simulation. The paper is organized as follows. In Section 2 the powertrain and main power devices are described. The details of the neural network controller are discussed in Section 3 and Statistical Learning theory is presented in Section 4. The results of the numerical Proceedings of the 17th World Congress The International Federation of Automatic Control Seoul, Korea, July 6-11, 2008 978-1-1234-7890-2/08/$20.00 © 2008 IFAC 4659 10.3182/20080706-5-KR-1001.3053