An Optimal Vehicle Speed Planning Algorithm for Regenerative Braking at Traffic Lights Intersections based on Reinforcement Learning

For electric vehicle or hybrid electric vehicles, the regenerative braking is one of the important means to realize energy saving, for which braking ahead of a traffic light intersection is a representative scenario. The uncertainty in driver behavior and future traffic flow, however, make it challenging to achieve optimal dynamic energy recovery through conventional braking operation by drivers. Therefore, in this paper, an energy recovery optimization-oriented vehicle speed planning algorithm ahead of traffic lights intersection is proposed, for autonomous vehicle or driving assistance system. First, the reward function is designed, taking the energy recovery amount, traffic efficiency and driving smoothness into consideration. Then, the information of traffic lights at intersections is obtained in advance through V2I (vehicle to infrastructure) communication. Finally, the q-table and neural network are trained in the framework of reinforcement learning, deriving optimal vehicle speed profile. Simulation results on a high-fidelity model show that the amount of recovered electrical energy using q-learning algorithm is 45.08% higher than that of uniform deceleration. The amount of electrical energy using DQN (Deep Q-network) algorithm is 2.24% higher than q-learning, showing to be a better candidate in terms of comprehensive optimality than q-learning.