MIPM: A Multidimensional Information Perception Model for Estimating Time of Arrival on Real Road Networks

With the development of GPS-positioning technology and mobile Internet, the intelligent transportation system plays an important role in our daily lives. Thus, despite many existing works focusing on improving the efficiency and accuracy of the transportation system, however, few of them can handle multidimensional features on road networks. In this paper, we focus on a famous problem of the intelligent transportation system named estimated time of arrival (ETA). Specifically, we propose a novel multidimensional information perception model (MIPM) to address the ETA problem in a real-life environment. MIPM consists of the extract and recurrent modules. In the data processing phase, we generate the sparse and select features of the multidimensional features. Then, in the extract module, we compute Raw-ETA by regression method based on the select features and then extract the Refine-ETA from Raw-ETA. After that, we design a SPETAformer block to ensure that the extract module can capture the multidimensional correlation of the above features. The recurrent module further enhances the learning ability of the MIPM in the temporal domain by week, daily, recent, and weather four different periodic slices. By combining the extract and recurrent module, our MIPM has the capability to obtain an accurate ETA. Evaluation experiments on a large-scale real-world dataset are conducted to show the superiority of our proposed model.