Long Term Prediction of Traffic Flow

Abstract Prediction of traffic flow is now becoming a basic problem in designing variable information boards or route guidance (route telling) systems in road networks. Although, time series analysis is widely applied to this problem, it is known that the error of prediction increases rapidly as the prediction lead time increases and it is difficult to decrease the delay of the predictor as long as the methods based on time series analysis are used. In this paper, a new method defined by the combination of autoregressive model fitting and pattern matching, is proposed. It uses the features extracted from historical data of traffic flow to compensate the delay caused by the predictor itself and also real time data are combined in the form of autoregression to trace the sudden changes of traffic flow.