基于城市区域多模态融合的人群流量预测

Exploring the Multimodal Fusion of Urban Regions for Crowd Flow Prediction

提出了一种基于多模态融合的人群流量预测算法(multimodal fusion for crowd flow prediction,MFCFP)。首先使用图卷积算子来探索区域之间的相关性以建立多模态,不同的模态可以捕捉不同的影响因素。然后进行多模态融合,并将带有注意力机制的基于图卷积神经网络应用于本文模型,以更好地建立相关区域关联。在真实数据集的实验证明了所提模型可以准确地预测人群活动流量。

A multimodal fusion based crowd flow prediction algorithm called multimodal fusion for crowd flow prediction(MFCFP)was proposed.Firstly,the graph convolution operator was used to explore the correlation between regions to establish multi-modality.Different modalities could capture different influencing factors.Then,multi-modal fusion was performed,and the graph-based convolutional neural network with attention mechanism was applied to the model to better establish the correlation of related regions.Finally,the experiments on the real Shanghai data set proved that the proposed model could accurately predict the flow of crowd activities.