A Kalman filter approach to traffic modeling and prediction

The objective of our work has been to develop and integrate prediction, control and optimization modules for use in highway traffic management. This is accomplished through the use of the Semantic Control paradigm, implementing a hybrid prediction/routing/control system, to model both macro-level (traffic control) as well as micro level (in-vehicle path planning and steering control). This paper addresses the design and operation of a Kalman filter(1) that processes traffic sensor data in order to model and predict highway traffic volume. This data was given in the form of hourly traffic now, and has been fit using a cubic spline method to allow observations at various time intervals. The filter is augmented via the Method of Sage and Husa(2) to identify the parameters of the system noise on-line, and to determine the dynamics of the traffic process iteratively to aid in the prediction of the future traffic. The results show a good ability to predict traffic flow at the sensors for several time periods in the future, as well as some noise rejection capabilities.