A deep-learning framework considering multiple motifs for traffic travel time prediction

Abstract

How to accurately predict short-term traffic travel time (TTT) is an important problem in intelligent transportation systems. Traffic data usually exhibit high non-linearities and complex patterns, and predicting TTT is a challenge. Most previous studies have used the topological adjacency of road networks to explore spatial correlations. However, a real road network contains higher-order connectivity patterns that have different statistical significance. The topology adjacency cannot reflect these higher-order connectivity patterns. To obtain topological adjacency and higher-order connection pattern information, a novel deep-learning framework (multi-motif graph convolutional recurrent neural networks) for TTT prediction is proposed. The accuracy of TTT prediction was improved with this model. There are two blocks in each unit of the model: (a) spatial blocks, which capture spatial pattern information by the multi-motif graph convolution network and motif graph embedding, and (b) temporal blocks, which capture temporal pattern information by the combination of a long short-term memory network and a fully connected layer. To prove the effectiveness and accuracy of the prediction model, experiments were conducted on real-world TTT data sets.