Abstract
In this paper, an auto-regression method is applied to pavement performance modelling to improve the predictive accuracy of predictions when there are only limited or incomplete data available. Using age and past measured conditions as independent variables, the average trend within a pavement group is captured by a ‘global’ function shared by all pavements, while any pavement-specific effects are reflected through the past pavement conditions involved in the model. In a case study, different auto-regression models with varying lags are developed based on measured pavement condition data and the predictive accuracy of the models is studied. The models are also compared with the traditional regression models, such as the shifted family curve and individual regression curves. The results show that the predictive accuracy of the auto-regression models generally increases when more lags or past conditions are used. The auto-regression models are able to provide more accurate predictions than other models. It is recommended that the auto-regression model be used in pavement management systems by highway agencies.