Incorporating multimodal context information into traffic speed forecasting through graph deep learning

Accurate traffic speed forecasting is a prerequisite for anticipating future status and increasing the resilience of intelligent transportation systems. However, most studies ignore involvement context information ubiquitously distributed over urban environment to boost prediction. The diversity complexity also hinder incorporating it into forecasting. Therefore, this study proposes multimodal context-based graph convolutional neural network (MCGCN) model fuse data prediction, including spatial temporal contexts. proposed comprises three modules, ie (a) hierarchical embedding learn representations by organizing contexts from different dimensions, (b) multivariate modeling capturing dependencies (c) attention-based fusion integrate with multi-step We conduct extensive experiments in Singapore. Compared baseline (spatial-temporal network, STGCN), our results demonstrate importance mean-absolute-error improvement 0.29 km/h, 0.45 km/h 0.89 30-min, 60-min 120-min respectively. explore how affect forecasting, providing references stakeholders understand relationship between