Modelling of traffic flow using modern stochastic approaches

Abstract

The importance of traffic state prediction steadily increases together with growing volume of traffic. The ability to predict traffic speed and density in short to medium horizon is one of the main tasks of every Intelligent Transportation System. This prediction can be used to manage the traffic both to prevent the traffic congestions and to minimize their impact. This information is also useful for route planning. Traffic state prediction is not an easy task given that the traffic flow is very difficult to describe by numerical equations. Other possible approach to traffic state prediction is to use historical data about the traffic and relate them to the current state by application of some form of statistical approach. This task is, however, complicated by complex nature of the traffic data, which can, due to various reasons, be quite inaccurate. This thesis is focused on finding the algorithms that can exploit valuable information contained in traffic data from Czech Republic highways to make a short-term traffic speed predictions. My proposed algorithms are based on modern stochastic approaches like hidden Markov models, dynamic Bayesian networks, ensemble Kalman filters, Monte Carlo simulation and Markov chains. These models are naturally able to capture all complexities in the traffic and incorporate uncertainty of the traffic data