Uncertainty-based prediction of work zone capacity using a Bayesian approach

Abstract

The uncertainty-based work zone capacity model is formulated as a mixed linear regression function of explanatory variables where capacity is a random variable following a lognormal distribution. Model parameters are calibrated using a Bayesian approach. Results show that the proposed work zone capacity distribution model can accurately predict the mean and prediction interval of work zone capacity at any given confidence level. It is found that work zones located in urban roads, with a larger number of open lanes or with long-term work duration have a larger mean work zone capacity. A short prediction interval (i.e. low uncertainty) for work zone capacity is found to be associated with the following situations: (a) a bigger number of open lanes; (b) rural work zone; (c) short-term work duration; (d) left lane closure; (e) daytime work and (f) a smaller percentage of heavy vehicles. The predicted interval length for the medium level of work intensity is increased, although the corresponding mean work zone capacity is reduced, as compared with the light level of work intensity. There will be a bigger measured work zone capacity on average if the measurement method is adopted in which work zone capacity is taken as the mean queue-discharge rate.