A Predictive PHEV Energy Management Strategy Based on GRU and Enhanced Dynamic Programming

This thesis puts forward a predictive energy management strategy (P-EMS) for a parallel shaft input-power-split hybrid electric SUV in order to improve the energy economy and reduce the frequency of engine start-up. The EMS administers a hierarchical control framework which covers a GRU-based predictor, a DP-based receding horizon controller and a bottom controller. GRU is responsible for using historical driving information to predict vehicle speed in the next 10s. By vectorizing state and control variables and optimizing the number of discrete grids, an enhanced dynamic programming algorithm is developed as the optimization algorithm of receding horizon controller. And the bottom controller based on PID with fast sample time is built to harmonize the engine and the two electro-motors. To test the effectiveness of the predictive EMS, besides UDDS cycle, an extra Urban_cycle is generated from the real-world driving dataset. Simulation results deduce that compared with the rule-based EMS, the predictive EMS can save energy consumption by 11.7% and 9.6%, respectively. While the number of engine start-up time is also reduced.