用于交通需求预测的时空地理加权回归模型及应用 周禹佳,沈辛夷,金盛 (漸江大学建筑工程学院,浙江杭州310058) 摘要:交通需求预测对交通规划具有重要作用,但当对交通需求进行预测,需同 时考虑时空因素对其造成的影响时则具有较大挑战性,本文选取时空地理加权回归 模型(GT瞅R模型〕进行预测,该模型能较好的捕捉交通需求的时空特性以及交通需 求与建成环境之间的关系,运用2016年杭州市车牌数据验证该模型的精确性,本 文运用一般最小二乘回归模型(0LS模型)、地理加权回归模型(GR模型)以及 GTwR模型分别对各个交通小区的交通需求进行预测,结果可以发现:GTwR模型在 需同时考虑时空非平稳性方面有较大优势,且就拟合度方面来说:0LS模型、GWR 模型以及θTR模型的精度分别为12.90%,51.04%以及91.85%,即GTR模型在描 述交通需求预测方面优于传统模型 关键词 交通需求预测;LS模型;GR模型;GTwR模型;时空非平稳性 中图分类号:U238 Geographically and Temporally weighted regression Model for Traffic demand Forecasting and Its application Zhouru-jia, Shen Xin-yi, Jin Sheng College of Civil Enginee ing and Architecture, Zhejiang University, Hangzhou 310058) Abstract: Traffic demand estimation is of great importance to transportation planning. Both the spatial and temporal dependences need to be considered simultaneously, which makes traffic demand estimation challenging. This paper introduces a Geographically and Temporally Weighted Regression(GTWR) model to capture the spatiotemporal characteristics of traffic demand and correlations between traffic demand and built environment. Experimental data, the license plate data in Hangzhou, 2016, is utilized to evaluate the accuracy of the GTwr model. This paper uses the global Ordinary Least Squares(OLS)model, Geographically ighted Regression(GWR)model and GTwR model to predict the traffic demand of each raffic Analysis Zone (TAZ) respectively. The results show that the GTwr model has substantial benefits in modeling both spatial and temporal non-stationarity simultaneously. In the test sample, in terms of goodness-of-fit, 12.90% of the variation in the traffic demand can be explained by the OLs model, while 51.04% by the GwR model, 91.85% by the Gtwr 通讯作者:金盛(1982—),男,浙江温州人,副教授,博士生 (c)1994-2019ChinaAcademicJournalElectronicPublishingHouse.Allrightsreservedhttp://www.cnki.net用于交通需求预测的时空地理加权回归模型及应用1 周禹佳,沈辛夷,金盛 (浙江大学建筑工程学院，浙江 杭州 310058） 摘 要： 交通需求预测对交通规划具有重要作用，但当对交通需求进行预测，需同 时考虑时空因素对其造成的影响时则具有较大挑战性，本文选取时空地理加权回归 模型（GTWR 模型）进行预测，该模型能较好的捕捉交通需求的时空特性以及交通需 求与建成环境之间的关系，运用 2016 年杭州市车牌数据验证该模型的精确性，本 文运用一般最小二乘回归模型（OLS 模型）、地理加权回归模型（GWR 模型）以及 GTWR 模型分别对各个交通小区的交通需求进行预测，结果可以发现：GTWR 模型在 需同时考虑时空非平稳性方面有较大优势，且就拟合度方面来说：OLS 模型、GWR 模型以及 GTWR 模型的精度分别为 12.90%，51.04%以及 91.85%，即 GTWR 模型在描 述交通需求预测方面优于传统模型。 关键词： 交通需求预测；OLS 模型；GWR 模型；GTWR 模型；时空非平稳性 中图分类号：U238 Geographically and Temporally Weighted Regression Model for Traffic Demand Forecasting and Its Application ZhouYu-jia, ShenXin-yi, JinSheng ( College of Civil Engineering and Architecture,Zhejiang University, Hangzhou 310058） Abstract: Traffic demand estimation is of great importance to transportation planning. Both the spatial and temporal dependences need to be considered simultaneously, which makes traffic demand estimation challenging. This paper introduces a Geographically and Temporally Weighted Regression (GTWR) model to capture the spatiotemporal characteristics of traffic demand and correlations between traffic demand and built environment. Experimental data, the license plate data in Hangzhou, 2016, is utilized to evaluate the accuracy of the GTWR model. This paper uses the global Ordinary Least Squares (OLS) model, Geographically Weighted Regression (GWR) model and GTWR model to predict the traffic demand of each Traffic Analysis Zone (TAZ) respectively. The results show that the GTWR model has substantial benefits in modeling both spatial and temporal non-stationarity simultaneously. In the test sample, in terms of goodness-of-fit, 12.90% of the variation in the traffic demand can be explained by the OLS model, while 51.04% by the GWR model, 91.85% by the GTWR                                                              通讯作者：金盛（1982—），男，浙江温州人，副教授，博士生