Intelligent Highway Traffic Forecast Based on Deep Learning and Restructured Road Models

We propose a highway traffic forecasting system that informs the traffic condition of highways from a few in several months ahead. It can reflect the weather information of the regions of roads in the traffic data computation. We develop various road models to represent separate points of the highways based on traffic characteristics such as interchange, exit, endpoint, etc. Experimental results show our system outperforms a generic convolutional network model with 97.6% accuracy of travel-time prediction and the reduction by 30% of computing time for a moderate sized highway network.