Development of asphalt pavement temperature prediction models utilising multiple regression and artificial neural network approaches: a field study in North America

Abstract

ABSTRACT Weigh-in-motion technology plays a crucial role in efficiently collecting data for traffic monitoring and controlling freight weight limits, with accuracy influenced by pavement temperature. However, research in asphalt pavement temperature modeling faces challenges due to limited environmental data and a lack of comprehensive insight into prediction performance across various sensor installation depths. To address this, an investigation was conducted through a field study at MnROAD in Minnesota, USA, predicting asphalt pavement temperature using data from an on-site weather station. Nine weather factors were considered, and these weather factors were utilized to predict pavement temperature through four regression analysis methods, including linear regression, two types of polynomial regression, and an artificial neural network (ANN) model. To validate the predictions, real pavement temperature data was collected at various depths using embedded sensors. The results showed that the ANN model outperformed other statistical regression models, achieving an R-squared value of up to 0.96. The findings also suggested an optimal sensor installation depth of 2.25 inches, achieving an R-squared over 80% with the ANN model while only considering three highly correlated weather factors for pavement temperature. Additionally, the consistent results from a secondary weather station validate the developed pavement temperature models in this study..