市區路段動態旅行時間預測之研究

Abstract

先進用路人資訊系統中，行前路徑導引系統提供的路徑旅行時間預測值和路線選擇資訊極為重要，重要性在於旅行者可彈性選擇合適的出發時間、運輸工具和路線規劃以避開擁塞路段減少時間成本的浪費。以旅行者需求觀點來說，旅行者使用系統的時機大致包括已在車上、正準備前往開車和未來規劃要開車。路段旅行時間預測值為路徑旅行時間預測值演算的基礎。由於旅行者有不同使用時機的需求，可能於不同時間點進入路段，因此必須提供這些進入路段的路段旅行時間預測值作為行前路徑導引系統演算的基礎。 有鑒於目前國內、外設計符合旅行者需求之旅行時間預測系統尚不普遍，因此，本研究將嘗試利用a-b-r濾波器、遺忘因子之遞迴最小平方和離散傅立葉變換等預測方法，同時預測未來即時、短期和長期等不同預測時階數的市區道路路段旅行時間，評估模式適用的預測時階數，藉此設立模式適用的門檻值。本研究的價值在於可針對不同預測時階數選擇不同的旅行時間預測模式，有效提升市區道路路段旅行時間預測的可靠度。 本研究以市區道路實際偵測器資料依不同工作日類型（星期一和星期五、星期二至星期四）進行範例分析，評估依門檻值劃分方式的預測績效與其它a-b-r濾波器、遺忘因子之遞迴最小平方和離散傅立葉變換等單一模式的預測績效，研究結果顯示，a-b-r濾波器模式和離散傅立葉模式可以穩健預測不同預測時階數的路段旅行時間值，但遺忘因子之遞迴最小平方則隨著預測時階數的減少，預測的結果較不穩健；預測模式和其門檻值會隨著不同工作日類型而變動且依門檻值劃分的預測方式可有效降低旅行時間預測誤差。

In recent years, there have been increased attention beging given to dynamic route guidance systems in the related literature. Dynamic path travel time information and route planning are increasingly important for dynamic route guidance systems because travelers are able to decide the departure time, transportations, and route planning.Travelears may use the route guidance systems when they are already on the cars, plan to drive cars, or plans to drive cars in the future.A number of studies have noted that the links travel time prediction information are the algorithmic basis for route guidance systems.Therefore, this study may be critically important in providing the links travel time prediction information on any occations for travelers . However, there have been less researches on the dynamic links travel time information systems which satisfied travel needs of travels. The purpose of this study attempts is to develop a dynamic links travel time information system which satisfied travelers travel needs. In this research, we use three different methods, which are a-b-r filter algorithem, recursive least square algorithem and descrete Fourier transform algorithem.We use these algorithems to predict urban links travel time of different time horizon. On the other hand, in order to provide travelers with reliable travel time information, we construct suitable thresholds of algorithems based on the evaluation of predictive effects. This research finds that the a-b-r filter algorithem and descrete Fourier transform algorithem is able to predict steady-state.Besides, the predictive algorithem and the threshold of algorithems are changeable with different week.