Prediction of Hybrid Electric Bus Speed for Energy Management Using Deep Learning 2020-01-1187
Nowadays, the automotive technology has changed rapidly in the world and hybrid electric vehicles (HEVs) have been widely using due to their significant effect in urban driving areas. For the study of such HEVs, many studies are conducted on the energy management strategy (EMS) using the simulation. In this case, the difference between simulation control and actual vehicle control is the presence or absence of a driving cycle. In the simulation, the driving cycle is known in advance, but not in real driving. Energy management in the HEVs is highly dependent on the prediction velocity. This means, knowing driving cycle in advance is actually important. In this paper, we propose a new method of the velocity prediction for the HEVs using deep learning method. Based on the velocity of the vehicle from the past to the present in a certain time, deep learning has shown great promises to predict vehicle velocity by using complex and multi-source field data. Moreover, we use a special kind of recurrent neural network (RNN), long short-term memory (LSTM) method, which solves the problem of RNN. LSTM is designed to avoid the long-term dependency problem and is suitable for driving cycles with a relatively long data range. This methodology can be applied with the real time optimization control by fast calculating. As a result, the speed prediction accuracy using deep learning have similar results with the actual speed data. In addition, the time to predict the velocity of the vehicle in deep learning is much faster than the dynamic programming. Using these results, we will be able to design a better HEVs energy management strategy and more efficient SOC management.
Giyeon Hwang, Sangyul Lee, Kyoungdoug Min, Jihwan Park, Seunghyup Shin, Jongmyung Kim, Huy Nguyen, Yeongha Hwang, Minjae Kim
Myongji University, Hoseo University, Seoul National University