Bus Dynamic Travel Time Prediction: Using a Deep Feature Extraction Framework Based on RNN and DNN

• DOI: 10.3390/electronics9111876
• Publication Date: 2020-11-09T00:03:37Z
• Authors: Yuan Yuan; Chunfu Shao; Zhichao Cao; Zhaocheng He; Changsheng Zhu; Yimin Wang; Vlon Jang
• Keywords: Leverage (statistics)

Travel time data is an important factor for evaluating the performance of a public transport system. In terms and space within nature uncertainty, bus travel dynamic flexible. Since change traffic status periodic, contagious or even sudden, changing mechanism that hidden mode. Therefore, prediction challenging problem in intelligent transportation system (ITS). Allowing large amount can be collected at present but lack precisely-conducting, it still worth exploring how to extract feature sets accurately predict from these data. Hence, extraction framework based on deep learning models were developed reflect state time. First, study introduced different historical stages signaling time, taxi speed, stop identity (ID) spatial characteristics, real-time possible arrival signified by fourteen spatiotemporal characteristic values. Then, embedding network proposed leverage wide structure mate temporal order meet dependence requirements, attention Recurrent Neural Network (RNN) was designed this research capture information. Finally, Deep Networks (DNN) implemented achieve prediction. Two case studies Guangzhou Shenzhen tested. The results showed algorithm more efficient than traditional machine-learning model promoted 4.82% compared neural applied initial space. Moreover, visualized weighted cost bus’s features during certain running state. demonstrated enabled understand transit with visualization.