Near optimal control of hybrid fuel cell electric vehicle in real-time -

Abstract

Hybrid electric vehicles (HEV) improvements in fuel economy and emissions strongly depend on the energy management strategy used whose aim is to minimize the hydrogen and battery cost. In this work a new control strategy called single step dynamic programming optimization (SSDP) and an energy management system based on artificial neural network are presented. These real-time energy management systems for HEV are derived from a dynamic programming (DP) technique. The DP requires that a forecast of the car torque requirement over the whole trip or part of it is available. However, in real time the road and driving conditions are not known a priori. The methods presented in this work can easily lend themselves for real time implementation. The problem formulation accounts for the power balance at each stage, the power limits, the state-of-charge (SOC) limits, and the ramp rates constraints of the fuel cell and battery. The SSDP optimization technique differs from DP in that it is a forward-looking model in which the controller makes instantaneous decisions without the need for back-tracing; therefore, it is more realistic than backward-looking models. It requires that only the demand at the next period should be known a priori and not the whole road. The proposed ANN is trained based on DP results carried out off-line. It can be implemented in real time as it takes one step at a time. The results obtained using both methods show that the fuel economy that can be achieved is very close to optimal results. Moreover, the two proposed methods provide an easy mechanism to change from charge sustaining (CS) to charge depleting (CD) operation simply by changing the lower bound of the battery SOC. To solve the SSDP and the ANN methods the demand at the next step should be known a priori. This demand is obtained by applying a one step-ahead speed forecast. The model used is a first order linear model also known as persistence forecast; it does not require an adaptive mechanism to adapt to changes in road conditions in real