Control of a Hybrid Electric Vehicle using Control Vector Parameterization and Reinforcement Learning

A hybrid electric vehicle (HEV) uses an electric motor (EM) driven by a battery alongside the conventional internal combustion engine (ICE) to provide the required energy to run the vehicle. Distribution of the required power demand between the ICE and the EM is one of the crucial tasks of a vehicle energy management system (EMS). Finding the optimal torque distribution between the two sources of energy while minimizing fuel consumption of the vehicle and maintaining the state of charge (SoC) of the battery is an optimal control problem. Offline control and real-time control of HEV are studied. For offline control, the optimal control problem is converted into a nonlinear programming (NLP) problem through control vector parameterization (CVP) and solved using sequential quadratic programming (SQP) for a given predetermined drive cycle. For real-time control, a model-free reinforcement learning (RL) framework is proposed to handle external unknown disturbances such as acceleration / deceleration during a drive cycle. The results obtained with the RL-framework for urban and rural drive cycles show the efficacy of the RL-framework for real-time optimization of fuel consumption in a HEV under uncertain conditions.