Kalair, Kieran (2021) Data science perspectives on problems in intelligent transportation systems and mobility. PhD thesis, University of Warwick.

Preview

PDF WRAP_Theses_Kalair_2021.pdf - Submitted Version - Requires a PDF viewer. Download (47Mb) | Preview

Request Changes to record.

Abstract

This thesis is a body of work applying data science and mathematical modelling to problems in intelligent transportation systems. Utilising data collected from the M25 London orbital, four problems relevant to both industry and academia are considered. In chapter 4 we develop a novel methodology for anomaly detection on road networks. We determine a data-driven region of typical behaviour in the flow-density plane, tracking fluctuations from this to identify anomalies in real time. We find this offers generally comparable performance to existing methods, but is clearly superior when the distribution of speeds conditioned on time of week is bimodal.

In chapter 5, we quantify the prevalence of primary and secondary traffic incidents in our data using a novel self-exciting point process. The selfexcitation component suggests 6-7% of incidents are most likely secondary, occurring temporally and spatially in the wake of other incidents. Our modelling further identifies two spatial hotspots and captures commuting patterns in the UK. We are able to apply out-of-sample validation and show the model is statistically defensible.

Chapter 6 explores dynamic prediction of incident durations. We find non-parametric neural network models offer strong performance compared to a range of alternative candidates, achieving errors below current industry targets. By exploring feature importance, we find time series prove informative for predictions on short horizons whereas time of day and location do so at longer horizons.

We explore an emergent behaviour path planning model in the context of autonomous vehicles in chapter 7. This was developed in conjunction with engineers from Jaguar Land Rover and incorporates practical constraints realworld vehicles must satisfy. Formulating an optimization problem incorporating comfort, safety and progress, we show dynamically solving this results in emergent complex driving behaviours: vehicle following, passing and overtaking. Safety is based on a distributional prediction of drivers behaviours, with its variance indirectly defining properties of the emergent behaviours.
Our findings throughout this work offer models and methodologies that can be used to improve the management and better understand the behaviour of existing transportation infrastructure, as well as the development of future technologies.

Item Type:	Thesis (PhD)
Subjects:	H Social Sciences > HE Transportation and Communications Q Science > QA Mathematics T Technology > TL Motor vehicles. Aeronautics. Astronautics
Library of Congress Subject Headings (LCSH):	Transportation -- Mathematical models, Intelligent transportation systems, Traffic flow -- Data processing, Traffic flow -- Mathematical models., Automated vehicles -- Safety measures, Automated vehicles -- Mathematical models
Official Date:	March 2021
Dates:	Date Event March 2021 UNSPECIFIED
Institution:	University of Warwick
Theses Department:	Mathematics for Real-World Systems Centre for Doctoral Training
Thesis Type:	PhD
Publication Status:	Unpublished
Supervisor(s)/Advisor:	Connaughton, Colm
Sponsors:	Mathematics for Real-World Systems Centre for Doctoral Training ; Engineering and Physical Sciences Research Council
Format of File:	pdf
Extent:	xx, 276 leaves : illustrations
Language:	eng

Request changes or add full text files to a record

Repository staff actions (login required)

View Item

Downloads