Recently,deep learning based city flow prediction has been extensively used in the establishment of smartcities.These methods are data-hungry,making them unscalable to areas lacking data.Although transfer learningcan ...Recently,deep learning based city flow prediction has been extensively used in the establishment of smartcities.These methods are data-hungry,making them unscalable to areas lacking data.Although transfer learningcan use data-rich source domains to assist target domain cities in city flow prediction,the performance of existingmethods cannot meet the needs of actual use,because the long-distance road network connectivity is ignored.Tosolve this problem,we propose a transfer learning method based on spatiotemporal graph convolution,in which weconstruct a co-occurrence space between the source and target domains,and then align the mapping of the sourceand target domains’data in this space,to achieve the transfer learning of the source city flow prediction modelon the target domain.Specifically,a dynamic spatiotemporal graph convolution module along with a temporalencoder is devised to simultaneously capture the concurrent spatiotemporal features,which implies the inherentrelationship among the road network structures,human travel habits,and city bike flow.Then,these concurrentfeatures are leveraged as cross-city invariant representations and nonlinearly spanned to a co-occurrence space.Thetarget domain features are thereby aligned with the source domain features in the co-occurrence space by using aMahalanobis distance loss,to achieve cross-city bike flow prediction.The proposed method is evaluated on the publicbike flow datasets in Chicago,New York,and Washington in 2015,and significantly outperforms state-of-the-arttechniques.展开更多
基金supported by the National Natural Science Foundation of China(Nos.62103124 and 62033012)the Major Special Science and Technology Project of Anhui Province,China(No.202003a07020009)the Open Project Program of Key Laboratory of Ministry of Education of System Control and Information Processing,China(No.SCIP20230109)。
文摘Recently,deep learning based city flow prediction has been extensively used in the establishment of smartcities.These methods are data-hungry,making them unscalable to areas lacking data.Although transfer learningcan use data-rich source domains to assist target domain cities in city flow prediction,the performance of existingmethods cannot meet the needs of actual use,because the long-distance road network connectivity is ignored.Tosolve this problem,we propose a transfer learning method based on spatiotemporal graph convolution,in which weconstruct a co-occurrence space between the source and target domains,and then align the mapping of the sourceand target domains’data in this space,to achieve the transfer learning of the source city flow prediction modelon the target domain.Specifically,a dynamic spatiotemporal graph convolution module along with a temporalencoder is devised to simultaneously capture the concurrent spatiotemporal features,which implies the inherentrelationship among the road network structures,human travel habits,and city bike flow.Then,these concurrentfeatures are leveraged as cross-city invariant representations and nonlinearly spanned to a co-occurrence space.Thetarget domain features are thereby aligned with the source domain features in the co-occurrence space by using aMahalanobis distance loss,to achieve cross-city bike flow prediction.The proposed method is evaluated on the publicbike flow datasets in Chicago,New York,and Washington in 2015,and significantly outperforms state-of-the-arttechniques.
