基于扩张因果卷积的城市客流量预测算法

Urban crowd flow prediction algorithm based on dilated casual convolution

人群的迁移行为可以通过时空相关轨迹和用户上网行为进行记录。通过分析用户的上网行为分布情况发现,用户在不同场景下的浏览内容具有一定的偏好性,据此构建了融合用户上网行为及迁移行为异构信息网络表征城市人群的转移行为。基于该异构信息网络,提出了一种基于扩张因果卷积的城市客流量预测模型,采用扩张因果卷积模块捕捉客流量分布特征和用户上网行为特征,并构建了异构信息融合模型来融合客流量分布特征与用户上网行为特征。客流量分布特征提取是通过不同时间尺度下时间序列提取客流量时间依赖关系,用户上网行为特征提取是根据2种场景下的用户上网内容。特征提取采用扩张因果卷积减少了模型层数,提高了模型效率。异构信息融合模型融合了多维特征信息,提高了模型在预测有突发事件时的即时客流量的准确率。

The migration behavior of the crowd can be recorded through the spatio-temporal correlation trajectories and users′ online behavior. By analyzing the distribution of users′ online behavior, it is found that users have a certain preference for browsing contents in different scenarios. Based on this, a heterogeneous information network that integrates users′ online behavior and migration is constructed to characterize the transfer behavior of urban people. Based on the heterogeneous information network, a city crowd flow prediction model based on dilated causal convolution is proposed, the dilated causal convolution module is used to capture the characteristics of crowd flow distribution and characteristics of users′ online behavior. A heterogeneous information fusion model is constructed to integrate the characteristics of crowd flow distribution and users′ online behavior characteristics. The crowd flow distribution characteristics is extracted through the time dependence of the crowd flow by time series at different time scales, and the users′ online behavior characteristics is extracted from the users′ online content under two scenarios. Feature extraction uses dilated causal convolution to reduce the number of model layers and improve model efficiency. The heterogeneous information fusion model integrates multi-dimensional feature information, and improves the accuracy of the model′s instant crowd flow prediction when emergencies occur.