Leveraging Graph Neural Network with LSTM For Traffic Speed Prediction

Abstract

Accurate traffic forecasting plays an important role in the smart city and is of great significance for urban traffic planning, management, and traffic control. However, road speed prediction is a challenge task due to the complex topological structure of road networks and stochastic traffic patterns. For capturing spatial and temporal dependencies simultaneously, in this paper we propose a novel graph neural network based traffic speed forecasting model, the graph Long short term Memory (GLSTM) model which consists Graph neural network (GNN) and Long short term Memory. To the best of our knowledge, this is the first time to combine LSTM and GNN to feed graph-structured data as input for train models in traffic speed prediction. More specifically, at first we construct a unweighted directed graph from road network and feed LSTM cell with graph-structured data. After that, we train the whole model based on encoder-decoder architecture and message-passing mechanism of GNN. Experiments show that our proposed method is able to utilize road structure to capture spatial-temporal dependencies based on GNN while capture long-term dependencies based on LSTM. The result of real world dataset shows that proposed method outperform state-of-the-art baseline methods.