Abstract
We proposed a forecast model that uses a neural network to predict what the winter road surface tem- perature will be after three hours in order to increase the efficiency of anti-icing programs. In order to establish the forecast model, learning models of the road surface temperature using three sets of input var- iables at a specific site were created. These learning models were then applied to other sites along an ex- pressway, and the correct classification of the road surface temperature was examined. The results of the present study revealed that the road surface temperature along an expressway can be predicted three hours in advance more than 76% of the time by applying a learning model to the time series variation of the road surface temperature and other variables, such as the hourly traffic volume and the effect of spreading anti- icing chemicals.
