Model predictive control with BPNN-based prediction for electric drive thermal management system of electric vehicle
Abstract
Under varying driving conditions, the electric drive thermal management system (EDTMS) of electric vehicle is prone to poor thermal stability, temperature fluctuation, and unnecessary auxiliary energy consumption. To improve the thermal control performance of the EDTMS, an improved EDTMS architecture and a backpropagation neural network (BPNN)-assisted model predictive control (MPC) strategy are investigated. In the improved EDTMS, the motor is cooled by insulating oil, while the motor controller is cooled by a water-cooling circuit. A liquid–liquid heat exchanger and a three-way valve are introduced to realize thermal coupling between the motor oil-cooling circuit and the motor controller water-cooling circuit, as well as switching between the water-cooling short loop and the radiator main loop. For the control strategy, t-distributed stochastic neighbor embedding (t-SNE)-assisted sensitivity analysis is used to select key input variables that characterize the thermal dynamic behavior of the EDTMS. The BPNN prediction model then provides one-step-ahead predictions of the motor oil outlet temperature and the motor controller coolant outlet temperature for MPC. Based on these predictions, MPC coordinates the oil pump speed, water pump speed, and three-way valve opening under actuator and thermal-safety constraints. The EDTMS model (EDTMSM) and BPNN prediction model are validated using experimental data, and the control performance of BPNN-MPC is compared with proportional-integral-derivative (PID) control under three driving cycles. Results indicate that the average error rates of the EDTMSM are below 3.5%, and the BPNN prediction model achieves a MAPE below 0.35% and an absolute error below 0.18°C. Compared with PID control, BPNN-MPC reduces the peak motor oil outlet temperature by 3.79°C, 1.78°C, and 3.71°C, and the peak motor controller coolant outlet temperature by 3.87°C, 1.66°C, and 4.64°C under WLTC, CLTC, and NEDC conditions, respectively. The total actuator energy consumption is reduced by 2.17%, 5.07%, and 4.35%, respectively. These results indicate that BPNN-MPC improves the thermal control performance and energy-saving operation of the EDTMS under multiple driving cycles.