標題: 以交通狀態為基礎之旅行時間預測
A Traffic State-Based Model for Freeway Travel Time Prediction
作者: 邱孟佑
Chiu, Meng-Yu
汪進財
Wong, Jinn-Tsai
運輸與物流管理學系
關鍵字: 旅行時間預測;集群分析;補值策略;Travel Time Prediction;Cluster Analysis;Imputation Strategy
公開日期: 2010
摘要: 傳統上,交控中心可以藉由線圈偵測器或影像偵測器蒐集交通流量資料與交通狀況,以對旅行時間進行預測與推估,但由於線圈偵測器無法識別所偵測之車輛,且過去的預測模式往往並未詳細考慮車流續進與延滯之特性,預估之旅行時間仍為瞬間旅行時間。為了考慮車流續進的過程以及彰顯車輛旅行時間與車流狀態間之關係並簡化預測模式之處理,本文藉由資料採集技術與迴歸分析設計出一套能預測高速公路交流道間旅行時間的預測模式,首先以集群分析方法對每個線圈偵測器之歷史資料作交通狀態分類處理,再經由迴歸分析構建不同交通狀態類別對旅行時間影響之旅行時間預測模式。接著,以ETC (Electronic Toll Collection) 車輛通行於收費站間之通行資料所計算出之旅行時間,作為旅行時間預測模式之校估依據,模式結果顯示其不但有相當良好之預測能力,校估之係數值亦可提供系統管理者豐富的訊息,以更了解各路段之幾何與交通特性對擁擠交通狀態所造成旅行時間增加之原由,進而研提有效的管理策略。然而,任何一個即時的交通資料預測系統在實際運作時,遺漏值的處理是無可避免,當面臨遺漏值現象時,過去的補值策略往往並未詳細考慮車流續進與延滯之特性,僅以偵測器本身的歷史均值或是以移動帄均方式填補遺漏值,本文另一研究重點為提出一混合補值模式,利用CART (Classification And Regression Tree) 演算法建構各偵測點與其相鄰偵測器及路段ETC旅行時間所關聯之分類決策樹,當某偵測點發生遺漏值時,則以該點對應之CART決策樹作為補值之預測依據,經過以實際有效樣本資料驗證補值結果顯示,透過交通狀態分類後之CART演算法可以有效提供長時窗遺漏值情況下的補值作業,另外,本文也發現在不同遺漏時窗數情境下,應以不同的補值策略進行補值,才能符合即時多變的偵測器遺漏值補正之需。最後,針對越來越趨成熟的探測車資料蒐集技術,本文設計一VVD(Virtual Vehicle Detector)虛擬偵測器機制,以改善傳統探測車技術之缺點:高通訊傳輸量、資料過濾與圖資媒合比對等複雜程序,經實例驗證與模擬系統取樣之研究結果都足以顯示VVD機制確實可行。
Traditionally, travel time estimation and prediction in a Traffic Management Center is mostly based on the data obtained from loop and/or image detectors. A prediction model solely based on these data, however, is difficult to consider the dynamic transformation and delay of traffic flow. To partially resolve this issue, this paper proposes a novel travel time prediction framework with the capability to predict inter-ramp travel time at a satisfactory level of prediction performance. First, historical traffic data collected by each loop detector were classified into different traffic states. For each state, regression techniques were then applied to build up a travel time prediction model. And then the travel time of vehicles passing Electronic Toll Collect (ETC) booths was considered to adjust the predicted traffic states and link travel time. The results showed satisfactory performance of the proposed models. More importantly, the estimated traffic parameters could provide system managers with fruitful information about how travel time is increased by different road geometry and traffic characteristics. Consequently, effective control strategies could be devised. Therefore, missing values is an inevitable issue in actual operations. Mean and moving average values based on historical data are common choices to replace missing values in past studies, which does not consider the features of vehicle flow continuation and lagging. The other purpose of this study is proposes a hybrid imputation strategy which, based on data mining techniques, a decision tree was then established using Classification And Regression Tree (CART) to connect each detect point to the adjacent detectors and the ETC travel time on the associated road section. When missing data were imputed based on the developed CART model. The empirical study showed that CART imputation method based on traffic state works effectively to impute data with missing values, especially under the circumstance of long-period data missing. Moreover, it was found that hybrid imputation strategies varied in different missing time-windows circumstances fit better into real time and various traffic conditions. Finally, due to more sophisticated probe car data collection technology, which have the disadvantages: High transmission capacity, data filtering, map data matching complex procedures and so on. To resolve this problem, a Virtual Vehicle Detector (VVD) system with a method to set up VVD on road network is proposed. The VVD mechanism that verified by the field testing and simulation results is feasible.
URI: http://140.113.39.130/cdrfb3/record/nctu/#GT079136806
http://hdl.handle.net/11536/40339
Appears in Collections:Thesis


Files in This Item:

  1. 680601.pdf