Using a data-driven approach to improve intersection modelling

Accurate traffic models are essential to test the effectiveness of road and infrastructure designs. In the absence of site-specific data, traffic modellers often use default parameters or apply rules of thumb. As a result, model predictions often deviate from reality and subsequent costly project reworks are needed.

This PhD project investigates the use of big data and advanced mathematical techniques to better model the traffic flow at intersections. Based on high-quality trajectory data extracted with modern video content analytic techniques, it aims to improve parameters estimation for existing commercial modelling packages and to develop a novel data-driven model.

It also looks to obtain deeper insights about the complex traffic dynamics at intersections through a comparison study between the different models.

Participants

• University of Western Australia
  

Project background

Road designers rely on traffic models to test the performance of their designs, especially at intersections where many complicated vehicle interactions occur. Accurate modelling of traffic flow at intersection is crucial for success and this often requires good parameters estimation. The model calibration process to find the optimal set of parameters is difficult as there is often a large number of parameters that influence model results. Furthermore, calibration can only be performed for existing sites and not for greenfield sites yet to be developed.

In the absence of more site-specific data, traffic modellers often rely on default parameters or apply rules of thumbs. As a result, the model prediction often deviates from reality and project reworks are needed which places social and economic burdens on the community.

Furthermore, some commonly used traffic models are originated from the distant past and calibrated using small datasets with limited features. They do not always capture Western Australia’s traffic adequately, such as the influence of heavy vehicles.

This PhD project will investigate the use of big data and a data-driven approach to address the aforementioned modelling issues. The aim is to develop a novel data-driven model free of underlying assumptions that can better describe the traffic dynamics at intersections. The effectiveness of such model will be compared with existing analytical and simulation-based modelling packages.

This project plans to advance the state of data-driven modelling in traffic flow and implements state-of-the-art statistical and mathematical techniques to draw deeper insights from high quality trajectory data.

Project objectives

The first objective of this project is to develop a novel, data-driven model to better capture the traffic dynamics at intersections based on the available high-quality trajectory dataset. Such a data-driven model is free of conventional assumptions of the traffic dynamics and should have the minimal complexity needed to describe the available data.

The second objective is to improve and extend existing commercial traffic modelling packages. One such aspect is to improve the parameter estimation for these packages using advanced statistical techniques and mathematical techniques such as the shadowing filters.

The third objective is to conduct a comparison study between our data-driven model and the modelling packages to test the effectiveness of our model and to draw deeper insights from the trajectory data regarding the traffic dynamics at intersections.

Profile of researcher, Samson Ting

Find out more about this project’s researcher, including what questions the research will answer, the obstacles Samson has overcome, what comes next after his PhD, and also hear from his supervisor, Dr Chao Sun. Read the profile at: Samson Ting – iMOVE PhD student.