기계 학습을 통한 구동 토크 예측 기반 속도 프로파일 최적화

Abstract

: In order to reduce fuel consumption and emissions caused by vehicles, a number of studies have been suggested that the optimal velocity profile for a given route is derived with dynamic programming (DP). In general, the vehicle dynamics model for accurately calculating energy consumption is required to solve the optimization problem. However, this model cannot exactly reflect the real characteristics of vehicle dynamics because of the nonlinearity of rolling resistance, air resistance, and gradient resistance. Therefore, this study proposes vehicle speed optimization algorithm with a machine learning network model that can predict the energy consumption from correctly calculating the traction torque. The performance of the proposed method is verified by simulation that driving environment corresponding to the real driving conditions is duplicated. The optimal speed profile is derived using dynamic programming with various constraints from the road information. The effectiveness of driving with the optimal speed profile is evaluated in comparison with driving with the conventional cruise control. As a result, the battery energy consumption improves to 8.4% by driving with the optimal speed profile for a given route of 27.3km.