Traffic speed prediction using ensemble kalman filter and differential evolution

Abstract

Importance of traffic state prediction steadily increases with growing volume of traffic. Ability to predict traffic speed in short to medium horizon (i.e. up to one hour) is one of the main tasks of every newly developed Intelligent Transportation System. There are two possible approaches to this prediction. The first is to utilize physical properties of the traffic flow to construct an exact or approximate numerical model. This approach is, however, almost impossible to implement on a larger scale given the difficulty to obtain enough traffic data to describe the starting and boundary conditions of the model. The other option is to use historical traffic data and relate information and patterns they contain to the current traffic state by application of some form of statistical or machine learning approach. We propose to use combination of Ensemble Kalman filter and Cell Transmission Model for this task. These models combine properties of physical model with ability to incorporate uncertainty of the traffic data.