Real-Time Optimal Eco-Driving for Hybrid-Electric Vehicles

摘要

This paper studies the eco-driving strategy for parallel hybrid-electric vehicles (HEVs). Its goal is to advice the driver with a fuel-optimal speed profile to follow and for this purpose two artificial neural networks (ANN) are designed to enable real-time implementation. To train the ANNs, an optimal control problem (OCP) is formulated, which is first solved using the dynamic programming (DP) technique. From the DP solutions obtained, several sequences of control modes are identified with the aid of semi-analytical solutions of the OCP. Then, a multi-class classification ANN is used to decide which control sequence to apply, and a regression ANN is further used to estimate the duration of each control mode in the control sequence. The ANN-reconstructed profiles are finally analyzed in comparison with the DP-computed speed profiles.