Deep learning to predict short-term travel time on the highway

• Main Authors: YI-RU LIU, 劉易儒
• Other Authors: Chia-Yu Hsu
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/9s3t6f

碩士 === 元智大學 === 資訊管理學系 === 106 === On the highway, the passers-by can only judge the travel time through their own experience or the current road information. The mastery of the travel time may not accurately predict the actual road conditions, and the possible sources of the road conditions are numerous. Including speed, traffic flow, occupancy rate, accident, weather conditions, etc. The purpose of this study is to construct a short-term travel time prediction model between highway sections, based on traffic data such as traffic flow, speed, and travel time provided by the Electronic Toll Collection(ETC) system, through Deep Gated Recurrent Units(DGRU) neural network build a travel time prediction model to predict the travel time of each section of the short-term highway. Compared with the existing deep learning methods, the deep neural networks (DNN) model proposed by Vu et al. (2017) and the long short-term memory deep neural network (LSTM-DNN) proposed by Liu et al. (2017) and deep long-term memory (DLSTM) model similar to DGRU. The results show that the DGRU travel time prediction model has a predicted travel time value of 1 minute in most routes during the 5 minutes prediction interval. In addition, even if the DGRU prediction model predicts a time interval of 60 minutes, the average error value of the National Highway No. 1 South Kyung Lung-Hsinchu System (0.5S-104.5S) on the 104 km routes is about 1.5 minutes.