FUEL-SAVING CONTROL STRATEGY FOR FUEL VEHICLES WITH
DEEP REINFORCEMENT LEARNING AND COMPUTER VISION |
Ling Han , Guopeng Liu , Hui Zhang , Ruoyu Fang , Changsheng Zhu |
School of Mechatronic Engineering, Changchun University of Technology |
|
|
|
|
ABSTRACT |
This study uses deep reinforcement learning (DRL) combined with computer vision technology to investigate
vehicle fuel economy. In a driving cycle with car-following and traffic light scenarios, the vehicle fuel-saving control strategy
based on DRL can realize the cooperative control of the engine and continuously variable transmission. The visual processing
method of the convolutional neural network is used to extract available visual information from an on-board camera, and
other types of information are obtained through the vehicle’s inherent sensor. The various detected types of information are
further used as the state of DRL, and the fuel-saving control strategy is built. A Carla–Simulink co-simulation model is
established to evaluate the proposed strategy. An urban road driving cycle and highway road driving cycle model with visual
information is built in Carla, and the vehicle power system is constructed in Simulink. Results show that the fuel-saving
control strategy based on DRL and computer vision achieves improved fuel economy. In addition, in the Carla–Simulink
co-simulation model, the fuel-saving control strategy based on DRL and computer vision consumes an average time of 17.55
ms to output control actions, indicating its potential for use in real-time applications. |
Key Words:
Deep reinforcement learning (DRL), Computer vision, Continuously variable transmission (CVT), Fuel
economy |
|