Assessment of hydrogen vehicle fuel economy using MRAC based on deep learning

Many controllers have been developed to control the motors of electric vehicles, but in the case of hydrogen vehicles, more delicate control is required, because power is provided through hybridization with batteries. The use of deep learning techniques enables optimization across various ranges based on extensive data to be more feasible. This study investigates the performance of three control strategies-PI control, Model Reference Adaptive Control, and Deep Learning-based MRAC-when applied to the motor of a hydrogen vehicle. The system of a hydrogen vehicle, including fuel cell systems, batteries, DC-DC converters, 3-phase inverters, and electric motors, was constructed and integrated to form a complete vehicle system. The performance of MRAC and DL-MRAC was compared based on step inputs of vehicle speed references, and the performance of PI control, MRAC, and DL-MRAC was evaluated under disturbances. Evaluations included considerations of the battery's state of charge for fuel economy. Results indicate DL-MRAC exhibited the best fuel economy characteristics. While DL-MRAC's current fuel economy advantage is slight, it is anticipated to improve with additional training.