Adaptive Multi-Armed Bandit Reinforcement Learning-Based Energy Management and Real-Time Control System for Autonomous Vehicles

This paper proposes an intelligent energy management system (EMS) based on Multi-Armed Bandit (MAB) algorithm to enhance vehicle powertrain efficiency and reduce emissions in Conventional Autonomous Vehicles (CAVs) in a spark-ignition engine. The presented EMS includes Support Vector Machine (SVM) and a multi-objective MAB. The Multi-objective MAB algorithm aims to minimize fuel consumption and emissions. The MAB requires adaptive and online classification of data based on environment and vehicle specification. The SVM algorithm works inside the MAB algorithm to create adaptive online classification. The algorithm produces optimal torque for fuel consumption by choosing the best throttle angle based on online classification. The MAB algorithm chooses a suitable throttle angle to maintain a stoichiometric air/fuel ratio for the least emissions. The proposed controller adjusts engine torque to decrease fuel usage and CO and NOx emissions. The simulation results show that the proposed EMS can decrease vehicle fuel usage to 6.41 l/100km, 11.7% less than the vehicle without the controller. The designed EMS also decreases CO and NOx engine emissions by 4.5% and 4.4%, respectively.