Energy Optimization of Electric Vehicle's Acceleration Process Based on Reinforcement Learning

Abstract

During a typical driving mission in the urban envi ronment, vehicle's dynamic performance is limited due to traffic congestion. When the vehicle could not accelerate with its maximum acceleration, the economic performance of acceleration process will have a certain room to improve. So in this paper, a pedal control strategy for electric vehicles' acceleration process is proposed, which could reduce energy consumption under congested traffic environment. Firstly, a reinforcement learning framework for vehicle's acceleration process is built, then the framework is constantly optimized through iterative computation. When the iteration is finished, the final state-action matrix will be used to express our control strategy. In order to ensure that the control strategy can fully balance vehicles' dynamic and economic performances, a regulation coefficient is defined and introduced into the calculation of rewards to determine whether the control strategy pays more attention to vehicle's economic performance or dynamic performance. The test results show that the energy consumption can be reduced by at least 5% within the vehicle's velocity being lower than 50km/h. When the velocity is lower than 20km/h, which is the average maximum velocity under traffic congestion, the energy consumption will reduce by more than 10% with the acceleration time only increasing by 2s, which suggests that our control strategy is suitable for use under traffic congestion and can effectively reduce energy consumption of an acceleration process.