| 초록 |
It is very important to predict the required kinetic energy of the vehicle for driving in the near future in order to improve the driving efficiency of the powertrain. If powertrain control is performed in consideration of future driving situations, energy efficiency and drivability, as well as gear-shifting performance of the transmission, can significantly improve because precise power requirements for future vehicle behavior can be used at the right time. One of the most important factors in estimating future vehicle behavior is predicting vehicle speed. However, it is quite difficult to predict the future speed of a vehicle accurately due to certain factors, including the driver’s habits, geometric information of the road ahead, and traffic flow. Due to trends concerning recently released vehicles, including advanced driving assistance systems(ADAS), such as high precision navigation with 3D map information and front radar and camera, it is now possible for a powertrain controller to utilize high-quality information that provides driving context to predict future vehicle behavior. In this study, we will be introducing a new, generative deep learning structure that continuously predicts future vehicle speed profiles in real time through a CNN-based, conditional VAE model that utilizes past driving data, current driver’s manipulation, and road traffic information. It is found that the prediction accuracy of the vehicle speed significantly improved with the new AI approach proposed in this paper. |
