Real-Time Prediction of Traffic Flows Using Dynamic Generalized Linear Models

Abstract

In previous real-time flow prediction studies, the emphasis was placed on the prediction accuracy of the model. The accuracy of the prediction bounds (or limits), on the other hand, was largely ignored. Prediction bounds are, however, important input parameters in such applications as real-time stochastic traffic control, incident detection, and route guidance in the context of dynamic traffic assignment. The objectives of this study are to explore the statistical nature of traffic flows when aggregated at short time intervals and to examine the potential of using the generalized linear model in the dynamic setting to predict traffic flows and provide prediction bounds. Specifically, this study derives recursive algorithms based on the quasi-likelihood principle and performs on-line, multiple-step-ahead predictions of short-term arrival flows for signalized intersections. Preliminary results are presented using a simulated data set from CORSIM and a real data set collected from signalized intersections.