Abstract: As a key transportation hub, the traffic efficiency of ports is of great strategic importance to regional economic development. However, accurate prediction of port traffic flow is a challenging task due to its stochastic uncertainty and time-unsteady characteristics. In order to improve the accuracy of port traffic flow prediction, this paper proposes a port traffic flow prediction model based on knowledge graph and spatiotemporal diffusion graph convolution network, taking into account the external disturbances such as meteorological conditions and the opening and closing status of the port-adjacent highway. The knowledge graph represents the factors related to the port traffic network, the knowledge representation method learns the semantic information of each external factor from the port knowledge graph, and the diffusion graph convolutional network (DGCN) and gated recurrent unit (GRU) can effectively extract the spatiotemporal dependent features of the port traffic flow. Detailed comparative experimental results show that the proposed KG-DGCN-GRU algorithm has good prediction accuracy, and the RMSE of the KG-DGCN-GRU model is reduced by (4.54-17.33)% and the MAE is reduced by (5.8-13.63)% in comparison with the classical prediction algorithm.

摘要： 港口作为关键交通枢纽，其交通效率对区域经济发展具有十分重要的战略意义。然而，由于港口交通流量具有随机不确定性、时间不平稳特征，因此港口交通流量的精准预测是一项具有挑战性任务。为了提高港口交通流量预测精度，本文考虑气象条件和港口相邻高速公路开闭状态等外部干扰因素，提出了一种基于知识图谱和时空扩散图卷积网络的港口交通流量预测模型(KG-DGCN-GRU)。知识图谱表示港口交通网络相关因素，知识表示方法从港口知识图谱中学习各外部因素的语义信息，扩散图卷积网络(DGCN)和门控循环单元(GRU)能有效挖掘港口交通流量的时空依赖特征。详细的对比实验结果表明所提出的KG-DGCN-GRU算法具有良好的预测精度，与经典预测算法相KG-DGCN-GRU模型的RMSE降低了(4.54-17.33)%，MAE降低了(5.8-13.63)%。

关键词: 港口交通流量预测, 知识图谱, 时空依赖, 门控循环单元, 图卷积网络