基于时空门控多图卷积网络的网约车需求预测

Research on Demand Prediction of Online Car Reservation Based on Spatio-Temporal Gated Multi-Graph Convolutional Network

动态的交通需求对于开发有效的实时交通管理和控制策略、算法至关重要。准确预测网约车需求具有一定的挑战性,但对智能交通系统的发展具有重要作用,有助于协调区域内的网约车供需,提高车辆的利用率,为乘客减少等待时间。文章提出一种时空门控多图卷积网络(Spatial-Temporal Gated Multi-Graph Convolutional Network,STGMGCN)模型,使用门控循环单元挖掘时间特征,并研究了三种不同的图卷积网络挖掘空间上的相关性。文章首先使用门控循环(GRU)单元提取研究区域的网约车的需求的时间相关性,之后构建三种不同图结构提取空间特征包括邻近关系图、功能相似性图、交互关系图并对输出结果进行融合,得到最终的预测结果;最后将该研究模型在真实网约车数据集上与基准模型进行对比实验,实验结果表明该模型的预测性能优于其他模型。

Dynamic traffic demand is crucial for developing effective real-time traffic management and control strategies and algorithms.Accurate ride-hailing demand forecasting is challenging but valuable for the development of intelligent transportation systems,which can help coordinate the supply and demand of online vehicles in a region,improve the utilization of vehicles,and reduce the waiting time for passengers.In this paper,we propose a Spatio-Temporal Gated Multi-Graph Convolutional Network(STGMGCN)model that uses gated recurrent units to mine temporal features and three different graph convolutional networks to mine spatial correlations.After that,three different graph structures are constructed to extract spatial features including proximity graph,functional similarity graph,and interaction graph,and the output results are fused to obtain the final prediction results.Finally,the model is compared with the benchmark model on a real online taxi dataset,and the experimental results show that the model outperforms other models.