[1M3-GS-10] AI application

[1M3-GS-10-01] Switching methods of predicting traffic congestion according to the characteristics of traffic flow

〇takahiro suzuki¹, kengo okano¹, ryoma nakamura¹, masaki matsudaira² (1. Oki Electric Industry Co., Ltd, 2. Oki Consulting Solutions Co., Ltd)

Keywords:probe data, traffic flow prediction

We have studied methods of predicting traffic flow using probe data to reduce traffic congestion and traffic accident. In this paper, we propose two types of prediction methods and an algorithm for switching between the two methods depending on traffic flows. For traffic flows that are not affected by sudden events such as traffic accidents, we apply the first prediction method that learns the traffic density transition in the stored data and predict statistically according to the traffic density transition on the observed day. For traffic flows that are affected by sudden events, we apply the second prediction method that learns the traffic density in the stored data as a pattern independent of location and time and make predictions by comparing patterns in observed data with learned patterns. We evaluated our algorithm on the days of sudden event and results showed that the prediction is close to actual traffic flows.