A Deep Learning Framework for Traffic Flow Prediction Based on Weighted Graph Convolutional Long Short Term Memory Network

Traffic prediction is important for Intelligent Transport System, and it is also the basis of many practical applications. However, traffic data are often highly non-linear and have complex spatio-temporal relationships, making this task challenging. In this paper, a new deep learning framework, Weighted Graph Convolutional Long Short-term Memory Network (WGCN-LSTM), is proposed for urban traffic flow prediction. The model combines the advantages of graph convolutional networks (GCN) and Long short-term memory networks (LSTM) to capture both spatial and temporal features in traffic flow data. WGCN-LSTM can reflect the actual network topology more accurately by weighting the road length and traffic flow. The experimental results show that WGCN-LSTM has higher prediction accuracy than other baseline models on two real traffic flow data sets, which verifies the effectiveness and applicability of this method in complex traffic prediction tasks.