A Bayesian Approach to Modelling Uncertainty in Transport Infrastructure Project Forecasts
K Cheung, Arup, UK; J Polak, Imperial College London, UK
The increased participation of the private sector in the delivery of transport infrastructure projects has increased the emphasis on understanding the accuracy and uncertainty of traffic demand forecasts. Transport models which provide these forecasts rely on simplified assumptions usually involving a combination of physical, socioeconomic, environmental and individual factors for a modelled base and future time period. Uncertainty in the value of input parameters and their conditional relationships results in uncertainty in the outturn forecasts. The accuracy of model predictions is normally tested through a number of quantitative and statistical methods. This paper presents a summary of the approaches used to model uncertainty in practice including scenario testing, sensitivity testing and statistical risk analysis using Monte-Carlo methods. However, other techniques are now available, and may offer superior insight into the structure of the underlying problem. In this paper, Bayesian belief networks, together with Monte Carlo Markov Chain techniques, are applied as an alternative method for modelling uncertainty in transport modelling. We illustrate the technique on a simplified toll road case study, based on a motorway in São Paulo, Brazil, in which we compute equilibrium solutions for traffic flow, travel time and cost for fixed demand and elastic demand problem formulations. The paper concludes on a comparison between the Bayesian belief network and a more conventional sensitivity analysis and discusses the relative merits of each approach.
Association for European Transport