Urban Congestion Areas Prediction By Combining Knowledge Graph And Deep Spatio-Temporal Convolutional Neural Network

Traffic congestion is becoming more and more serious in large and medium-sized cities around the world. Urban traffic congestion can lead to travel time delays and heavy vehicle pollutant emissions. Therefore, it is of great significance to study the prediction methods of urban traffic congestion. This paper proposes an approach by combining knowledge graph and deep spatio-temporal convolutional neural network, called KG-ST-CNN, to collaboratively forecast the congestion area in a city. Specifically, we construct an urban knowledge graph (KG) that is organized by regions based on the multiple types of urban knowledge. The graph convolutional network (GCN) is introduced to extract features from the urban KG. These features are processed as the input of a spatio-temporal convolutional neural network. This method forecasts traffic congestion areas accurately by considering traffic network as images. Experiments on real data from Beijing show that our approach outperforms baseline methods.