時空關聯之市區短期交通預測模式

Abstract

文獻中對交通流和旅行時間預測的研究多以時間序列或上下游偵測點的連結分別討論 時間與空間的關係。在市區道路上，一般駕駛者會因應道路的擁塞而改變路線，以減 少擁擠的時間，所以各空間位置上的交通量相互關係應為更加複雜。本研究的目的是 發展一個考慮時空關聯的模型框架，以分析在城市網路的短期交通預測。最近的一些 研究表明時空關聯在交通分析的重要性。然而，空間相關性一般都是外在給定而非模 型的內在變數因子，所以本研究提出使用狀態空間法去處理這個問題。模型將探討各 地點的交通狀態的分布與規則，並能分析及預測平日常的交通擠塞情況，以及那些因 突發事件所引起的擠塞對周邊的影響。我們將使用微觀模擬數據以及台北市網路資料 來分析。 本計劃為正在進行的國科會計劃的延伸 （市區道路之短期旅行時間預測模式，Project no. NSC 98-2221-E-009-108），而目標與方法論有很大的差異。

Previous studies on the traffic flow and travel time forecasting focused either on the temporal dimension of traffic flow as a univariate time series, or on the spatial dimension with the upstream-downstream relationship. With the fact that drivers change the route to less congested road in urban transportation network, the spatial relationship of locations should be more complex. The objective of this research is to develop a spatio-temporal modeling framework for short term traffic forecasting in urban network. Several recent researches showed the importance and potential of analyzing the traffic by the spatio-temporal techniques. However, the spatial correlations were given explicitly and it is worth investigation, and a state-space approach is proposed in this study to handle this issue. The developed model would explore the patterns of traffic states at different locations, and therefore the physical characteristics of the road network are learned implicitly. Therefore, it should be able to forecast the effect of everyday traffic congestion and also those due to unexpected incidents in the near future. Data from microscopic simulation as well as real-data for Taipei city network will be used in our analysis. This study is an extension of a previous on-going NSC research (Short-term travel time forecasting for urban network; Project no. NSC 98-2221-E-009-108), with substantial updating in the problem statement and methodology.