市區道路之短期旅行時間預測模式

Abstract

旅行時間資訊是智慧型運輸系統(ITS)中非常重要的一環。如先進旅行者資訊系統 (ATIS) 的應用中，用路人需要知道當下或未來出發的旅行時間，以作個人行程規劃之 用。本研究的主要目的為預測市區道路之短期旅行時間，使用一般已有的偵測器資料， 建立出每天之交通狀態資料庫，再搭配收集之旅行時間，效估出車流量和交通狀態對 旅行時間之關系。以後只要有現今的偵測器資料，即可在資料庫找出類似的模式並預 測出短期旅行時間。交通事故、天氣或其他對交通有影響的事件也可考慮作模型的延 伸。 因市區道路的問題復雜程度比高速公路來的高，所以本研究的目的主要為衡量點到點 之間各路徑的優劣比較。道路旅行時間的預測可作方面的應用，如繞道建議，或從市 區到高速公路上應使用哪個交流道口可避開市區擁塞。

Travel time information is the fundamental component in Intelligent Transportation Systems without questions. It is essential in applications of Advanced Traveler Information System (ATIS) in which travelers would like to know the travel times in the past, present and the future for their trip planning. In this study, we investigate the travel time estimation and short-term forecasting for arterial roads in urban network. Making use of the currently available traffic data (such as Vehicle Detector), the day-to-day traffic status of the network, which varies at different times of the day, can be estimated and learned as a historical database of recurrent congestion. Such traffic status can be used to link to a travel time estimation model for arterial roads and to be calibrated with field collected travel time data. This approach, as reported in some studies, usually underestimates the travel time when the road becomes congested and overestimates when the traffic is reducing from the peak. This will be further extended for a real time forecasting, taking into account the externalities such as incidents, weather and events into the training of the historical database. In this sense, it provides more features when current traffic state pattern is being matched to a historical one. In contrast to prediction of freeway travel time, the problem for urban arterials subjects to many hysterical situations and very challenging. Given the complexity of the problem, it is emphasized that forecasting the travel time at “up to a second” accuracy is not our objective. However, our model can be used to predict travel time of major arterials, and thus evaluate a set of candidate paths between two-points in the network, suggesting a real-time shortest path. The application can be extended such as advising a path leading to the highway avoiding the urban congestion.