This proposal deals with the development of flexible models for the representation of traffic dynamics in ways that allow the practical use of rich and diverse data sources and provide insight into the traffic flow problem. The fundamental traffic flow theory relationships are a classic way of modelling traffic dynamics. In this project, an alternative paradigm for traffic dynamics models, appropriate for traffic simulation models, will be developed, based on machine learning approaches such as clustering, classification and local regression techniques. While these models may not directly provide as much insight into traffic flow theory, they allow for easy incorporation of additional explanatory variables, and hence, may be more appropriate for use in traffic estimation and prediction models, especially simulation based.
The objectives of this research are three-fold:
- Develop flexible alternatives to the classical traffic dynamics models, especially in the context of traffic simulation models, that integrate the state-of-the-art in diverse research fields (such as machine learning) and exploit emerging data collection techniques (such as richer data, increasingly becoming available from probe vehicles and Automatic Vehicle Identification Systems, AVI)
- Compare how the presented alternative models perform with respect to their traditional counterparts, considering several different criteria, such as overall predictive accuracy, computational performance and robustness, and
- Develop insight into their operation.
Preliminary empirical results indicate that the use of such flexible traffic models for speed estimation can considerably improve the estimation accuracy, providing up to 40% improvement over the base case of using models based on conventional traffic flow theory.
Fields of science
Call for proposal
See other projects for this call