利用貝氏理論於旅行時間推估之研究

Abstract

旅行時間預測在先進旅行者資訊系統中是一項重要且基本的資訊，能協助用路人選擇合適的路徑，避開交通擁擠路段，提高運輸效率，有效解決交通壅塞問題。大多數旅行時間預測的方法，都必須仰賴可靠的歷史資料，採用不可靠的歷史資料，將會造成預計結果顯著的誤差。 本研究提出一套以貝氏更理論為基礎的歷史旅行時間推估方法。本方法首先比對即時資料與歷史資料是否存在顯著差異，亦即判斷即時資料是否可靠。若不存在顯著差異，則採用歷史資料；若存在顯著差異，則利用貝氏更新法修正即時資料，並進而推估路徑旅行時間。 本研究以高速公路的實際資料進行測試，同時比較以標準差為權重的貝氏更新法、以變異數為權重的貝氏更新法，以及直接採用即時資料的方法。測試結果顯示這三種模式在不同的路段長度情境下都有令人滿意的結果，尤其以標準差為權重的貝氏更新法在長度為10公里左右的路段表現最佳。

Travel-time prediction is a very important and foundational function in ATIS. Base on this information, travelers can choose appropriate route to avoid traffic jam. It helps to increase highway capacities and tackle the traffic jams. Most travel time prediction methods rely on an accurate historical database. Inaccurate database usually results in significant impacts for travel time prediction. We proposed a Bayesian based method for generating historical travel time. This method first compares the real time data with those of historical database. If there exists no significant difference, historical data is used. Otherwise, a Bayesian Updating method is invoked to compromise the real time data. We developed three variations of Bayesian Updating method, which are standard deviation weighted, variance weighted, and no Bayesian Updating, respectively. We applied these three models on the highways using real world data. The testing results show that all these three models yield satisfactory results. The standard weighted Bayesian Updating method performed best when the length of road segment is less than 10 km.