Deep reinforcement learning based hierarchical eco-driving strategy for a series-parallel hybrid electric truck

It has been well recognized that the development of information technology and the construction of intelligence infrastructure are becoming better solution to energy consumption and pollutant emissions. In this paper, deep deterministic policy gradient (DDPG) algorithm is embedded into a hierarchical eco-driving framework which is designed to improve the driving performance. The structure is divided into two layers, namely, the upper adaptive cruise control (ACC) and the lower energy management strategy (EMS). Because of the complex power model, pre-optimized progress is embedded into the DDPG loop. The experimental results show that compared with the Green Light Optimal Speed Advice (GLOSA) strategy with similar travel time, the proposed EMS can be reduced by about 8% and the fuel economy can reach 87.2% to 90.7% of dynamic programming (DP) based EMS. Overall, the proposed framework shows better energy saving advantages.