Inland river vessel traffic flow prediction based on environment knowledge embedding and spatial-temporal information extraction

Accurate and reliable vessel traffic flow prediction technology is of increasing concern to maritime regulatory authorities due to the growing demand for inland waterway shipping and the rapid increase of vessel traffic. Vessel navigation is not only influenced by nearby vessels, but also vulnerable to environmental factors. The existing inland vessel traffic flow prediction (IVTFP) methods often ignore the spatial correlation within the vessel traffic flow and the correlation from the external environment, leading to the accuracy of IVTFP not meeting the actual needs. Therefore, this paper proposes an IVTFP model based on environmental knowledge embedding and spatial-temporal graph attention networks. The environmental factors of traffic flow are described by knowledge graph and transformed into the form of matrix by knowledge representation learning algorithm. After that, the knowledge matrix is fused with the traffic flow data using the knowledge embedding module, and it is inputted to the spatial-temporal graph attention network. The experimental results on real experimental dataset show that the model improves the prediction accuracy of IVTFP, and gets better prediction results in upstream and downstream prediction, long-term prediction, and has good robustness.