Road Surface Crack Condition Forecasting Using Neural Network Models

摘要

This report summarizes the results obtained from a research project sponsored by211u001eFlorida Department of Transportation to develop a Backpropagation Neural Network 211u001e(BPNN) model for the forecasting of pavement crack condition of Florida's highway 211u001enetwork. The BPNN model, which is able to learn the hidden information from the 211u001ehistorical crack condition data, has the capability to forecast future crack 211u001econdition. In order to setup an effective model, the concept of BPNN was 211u001eintroduced along with its mathematical training algorithm. The neural network 211u001emodel was then trained and tested with field data collected from Florida's 211u001ehighway network. Further, the BPNN model was compared with a commonly used 211u001eautoregressive (AR) model. Finally, a validation step was performed to identify 211u001ethe forecasting errors on the 1998 data set. It was concluded that the BPNN model 211u001ewas more accurate than the AR model and could be applied to forecast pavement 211u001ecrack condition.