Abstract:

A hierarchical control method was proposed for adapting the working condition in the driving process for hybrid-electric-vehicle (HEV) platoon based on a Kullback-Leibler (KL) divergence working condition recognition algorithm. The upper layer controller utilized vehicle-vehicle communication technology, obtained state information of the leading vehicle in the platoon. It adopted the Model Predictive Control (MPC) algorithm, achieved longitudinal control of the platoon, and calculated the optimal speed for the following vehicle. The lower layer controller, initially based on typical working conditions, solved the transition probability matrix of demanded power offline. Trained the optimal Q-table offline, embedded it in the complete vehicle model by the Q-Learning algorithm. During driving, the transition probability matrix was updated online at regular intervals, and KL divergence was used to recognition the working conditions. According to the identified working conditions types, combined the current moment vehicle speed, the demanded power, and the battery state of charge (SOC), the torque allocation were achieved through an online lookup table. The results show that the fuel consumption of this strategy is reduced by 8.6% compared with the strategy without considering the condition identification, and increased by 4.8% compared with the dynamic programming (DP) as the benchmark. Under the premise of maintaining the same fuel consumption as DP, the off-line simulation time is reduced by 21%, which can not only be applied online, but also adapt to the change of working conditions in real time.

Key words: hybrid electric vehicle (HEV), vehicle platoon, working condition recognition, model predictive control (MPC) algorithm, Q-Learning algorithm, KL (Kullback-Leibler) divergence