Online-learning type of traveling time prediction model in expressway

Abstract

There are many requirements for predicting traveling times on express ways. It is difficult to predict traveling time with high precision when traffic flows change dynamically, such as at the beginning and end of traffic jams. It is therefore important to develop a traffic simulation model using time-series data using traffic counters. The model must cope with secular changes of traffic-flow characteristics such as construction of new expressways or environment condition changes. This paper proposes a new travel time prediction system which has a model learning function using time-series data processing. The new system is a mixed structure type neural network based travel time prediction system. It has a modeling function and a model learning function using field data processing on a time series basis. The proposed system has already been tested on an actual expressway, and satisfactory results were achieved.