Research on Energy Management Strategy of a Hybrid Commercial Vehicle Based on Deep Reinforcement Learning

Given the influence of the randomness of driving conditions on the energy management strategy of vehicles, deep reinforcement learning considering driving conditions prediction was proposed. A working condition prediction model based on the BP neural network was established, and the correction coefficient of vehicle demand torque was determined according to the working condition prediction results. An energy management strategy and deep reinforcement learning were integrated to build an energy management strategy with deep reinforcement learning based on driving condition prediction. Simulation experiments were conducted according to the actual collected working condition data. The experimental results show that the energy management strategy, i.e., deep reinforcement learning considering working condition prediction, has faster convergence speed and more vital self-learning ability, and the equivalent fuel consumption per 100 km under different driving conditions is 6.411 L/100 km, 6.327 L/100 km, and 6.388 L/100 km, respectively. Compared with the unimproved strategy, the fuel economy can be improved by 3.18%, 3.08%, and 2.83%. The research shows that the energy management strategy, the deep reinforcement learning based on driving condition prediction, is effective and adaptive.