融合多模态信息的Prophet-DCRNN交通预测方法

Prophet-DCRNN Traffic Forecasting Method With Multi-modal Information

交通预测问题具有明显时空特性。相比传统数据挖掘方法,深度学习因为其解决复杂问题的能力越来越受到研究人员青睐。同样在交通预测中,深度学习借助其被精心设计的网络结构很好的捕捉了时空特征。然而,实际城市交通状态还受天气、节假日等因素影响,往往在恶劣天气、节假日更倾向于出现交通拥堵。为解决上述问题,文中提出了融合多模态信息的Prophet-DCRNN路口交通预测方法,利用Prophet时序预测算法捕获节假日效应,采用DCRNN捕获交通时空特性,此外采用类stacking技术,融合Prophet算法、DCRNN算法及天气信息,得到最终融合多模态信息的混合模型。最后通过实验验证了Prophet-DCRNN混合模型在节假日、不同天气状况等场景下交通预测的准确性。

Traffic forecasting has obvious temporal and spatial characteristics. Compared with traditional data mining methods,deep learning is more and more favored by researchers because of its ability to solve complex problems. Also in traffic forecasting,deep learning uses its carefully designed network structure to capture spatio-temporal characteristics. However,the actual traffic is also affected by factors such as weather conditions and holidays. It is more prone to traffic congestion during severe weather and holidays. To address these challenges,we proposes a Prophet-DCRNN intersection traffic forecasting method that integrates multi-modal information. The holiday effect is captured by the Prophet time series prediction algorithm,and the temporal and spatial characteristics are captured by DCRNN. In addition,a hybrid model with multi-modal information is obtained by using similar stacking technology,combining Prophet algorithm,DCRNN algorithm and weather information. Finally,the experiment verified the accuracy of the Prophet-DCRNN hybrid model in traffic forecasting in scenarios such as holidays and terrible weather.