Combining artificial neural networks with traditional statistical tools to predict rut depth on Swedish road pavement

摘要

Deterioration models are of great value for pavement engineers because they provide them with some insight into current/future pavement condition. Within Pavement Management System in Sweden, a lot of attention is paid on predicting of the pavement rut depth. Rutting represents a deformation in the transverse profile of the pavement. It is one of the major types of deterioration experienced in Sweden, because of the common use of studded wires during wintertime and the damaging influence of slow moving traffic. A modeling approach, combining linear and nonlinear models, is investigated for predicting the three-year-ahead rut depth. It is implemented in four main stages: (a) predicting the rut-depth using a linear regression model, (b) computing the residuals, (c) modeling the residuals using a multi-layer percep-tron, (d) combining the outputs from both models. Simulations performed on the basis of data collected on the Swedish road network demonstrate the effectiveness of the suggested prediction approach.