Authors: Professor Adel Elgammal
Abstract: It is within this context of the growing popularity of electric vehicles (EVs) that the development of smart energy management, which can optimally manage the power consumption, increase the battery life, and enhance the vehicle efficiency in various driving patterns and conditions, has become essential. Conventional control strategies such as rule-based strategies and model predictive control can work well in controlled environments, but may be insufficiently resilient to the real-world complexity of changing traffic, gradients, and driver actions. In this work, a new real-time energy management strategy for EVs is developed by means of a RL-based optimal control framework, where DQN is adopted to dynamically optimize decisions about energy utilization. The proposed RL controller learns the optimal policies by exploring the real-time high-fidelity EV simulation environment, which accounts for vehicle dynamics, battery attributes, and external driving conditions. Unlike classical controllers, the RL-based solution does not require any predefined models or future prediction horizon to operate, as it continually learns from its own experience to decide in real-time on the power split between the electrical machine and auxiliary systems. The reward functions are designed to optimize for, for instance, energy efficiency, battery health, and driving performance features e.g. acceleration and driving smoothness. Simulation results show that the proposed RL-based controller can outperform benchmark strategies in various driving scenarios, obtaining up to 18% better energy efficiency and increased adaptability to changing situations. Moreover, the learned policy is robust in controlling battery temperature and state of charge (SOC) fluctuation which results in an increased battery life. This research reveals the capabilities of reinforcement learning as a promising scalable and self-adaptive technique for energy control in future EVs. For future works, we plan to further consider practical applications, multi-agent vehicle coordination, and integrating the proposed algorithm with V2I to realize cooperative energy optimization in smart transportation networks.
