2021 年 9 巻 1 号 p. 8-19
In this study, we detect the detours of commercial vehicles during heavy rains in western Japan using machine learning technology and then analyze the cause of these detours. Due to heavy rains in 2018 in western Japan, road regulation was implemented over a wide area. GPS-generated probe trajectories revealed the detour routes taken. The necessity of taking detours is one of the traffic failures caused by disasters. To identify these detours, a road administrator must visually check and analyze the probe vehicle trajectory, which requires considerable labor. Therefore, in this study, we detected detours during a disaster by learning the probe vehicle trajectory under normal circumstances using a one-class support vector machine (OCSVM). Results of detour detection for Shikoku revealed that vehicles were using distant detour routes even when nearer detour routes were accessible. An analysis of the cause of these detours showed that the “risk” of the traffic failure was one factor.
