Use this URL to cite or link to this record in EThOS:
Title: Bayesian dynamic graphical models for high-dimensional flow forecasting in road traffic networks
Author: Anacleto Junior, Osvaldo
Awarding Body: Open University
Current Institution: Open University
Date of Award: 2012
Availability of Full Text:
Access from EThOS:
Access from Institution:
Congestion on roads is a crucial problem which affects our lives in many ways: As a consequence, there is a strong effort to improve road networks in order to keep the traffic flowing. Flow forecasting models based on the large amount of traffic data, which are now available, can be a very useful tool to support decisions and actions when managing traffic networks. Although many forecasting models have been developed to this end, very few of them capture important features of high dimensional traffic data and, moreover, operating most of these models is a hard task when considering on-line traffic management environments. Dynamic graphical models can be a suitable choice to address the challenge of forecasting high-dimensional traffic flows in real-time. These models represent network flows by a graph, which not only is a useful pictorial representation of multivariate time series of traffic flow data, but it also ensures that model computation is always simple, even for very complex road networks. One example of such a model is the multiregression dynamic model (MDM). This thesis focuses on the development of two classes of dynamic graphical models to forecast traffic flows . Firstly, the linear multiregression dynamic model (LMDM), which is an MDM particular case, is extended to allow important traffic characteristics in its structure, such as the heterocedasticity of daily traffic flows, measurement errors due to malfunctions in data collection devices, and the use of extra traffic variables as predictors to forecast flows. Due to its graphical structure, the MDM assumes independence of flows at the entrances of a road network. This thesis therefore introduces a new class of dynamic graphical models where the correlation across road network entrances is accommodated, resulting in better forecasts when compared to the LMDM. All the methodology proposed in this thesis is illustrated using data collected at the intersection of three busy motorways near Manchester, UK.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral