A spatio-temporal pattern extension method for predicting traffic jams deviating from past traffic patterns

Abstract

We have researched a traffic flow prediction method using probe data to provide accurate, real-time traffic information for the purpose of reducing traffic jams and accidents. The percentile method, which statistically predicts traffic flow several hours in advance, has a problem with accuracy because it cannot predict traffic jams at certain times of the day or traffic jams extending beyond a certain length that deviated from the learned pattern. Therefore, we developed a time dilation method and a space-time dilation method for training data in response to changes in traffic density, and applied them to this method. As a result, we confirmed that it is possible to predict traffic jams and extended traffic jam lengths at times that have not been learned in the past, and achieved improved accuracy.