Kenta HARIHARA

Predicting the Risk of Getting Stuck on Winter Highways Using a State Space Model

Kazushi SANO

In this study, we proposed a risk indicator for the occurrence of getting stuck, based on the background of frequent traffic disruptions caused by extremely heavy snowfalls in recent years.
To construct the risk indicator, a state-space model was constructed to estimate the average speed of vehicles in a section, because the occurrence of stuck vehicles is caused by a decrease in driving speed. The model structure that takes into account the road surface condition, which is considered to have a significant impact, was examined from a simple aggregate analysis. The formulated model was applied to several past cases of stuck vehicles.
As a result of the application, it was confirmed that in some cases, the regression coefficients were estimated with extremely negative divergence just before the occurrence of a stack. In these cases, it is considered that the occurrence of an abnormal event caused a significant reduction in driving performance, which led to a situation in which a stack was likely to occur. Therefore, it was considered effective to focus on such characteristics, and a quantitative risk index SRI for the occurrence of stuck vehicles was developed based on a probability density function. We also verified the validity of the constructed risk index. As a result, it was confirmed that the risk indicator continuously showed dangerous values from several hours before the stacks occurred in the section where the stacks actually occurred and the surrounding area. In the sections far from the point where the stacks occurred, the SRI values were calculated to be on the safe side.
In this study, we developed an index SRI that can identify sections where the risk of getting stuck is increasing. In fact, it was confirmed that the SRI was able to adequately detect the occurrence of stacks in past cases. On the other hand, in order to make practical use of the research results, it is necessary to obtain more detailed information on the abnormal events described above. This will contribute to the early detection of the increased risk of stacks and the establishment of a more rapid countermeasure system.