市區路段短期交通量預測

Abstract

近年來短期交通量預測的應用日益受到重視，許多先進交通旅行者資訊系統(ATIS)及先進交通管理系統(ATMS)的應用都需要估計及預測路網之交通狀況，其目的在於提供有用之旅行資訊給旅行者及提升整體路網的效率。透過道路上各種不同偵測器所蒐集到之交通狀況歷史資料，我們可以掌握道路上之即時資訊並用來估計目前的交通狀況及預測短期內可能發生的交通狀況。目前大部分交通流量預測的文獻都著重在高速公路流量的預測，事實上市區道路的交通流量預測由於還要考慮機車、紅綠燈的影響，且市區路網也較高速公路路網複雜許多，因此有必要深入研究。 本研究建立一個都市地區短期交通量的預測模式，所採用之方法為整合自我迴歸移動平均模式(ARIMA)及時空自我迴歸移動平均模式(STARMA)，其中時空自我迴歸移動平均模式為將時間序列之空間分布關係考慮進模式當中之自我迴歸移動平均模式。我們除了利用台北市24個偵測器所蒐集之交通量資料來做實例測試，也測試了可獲得即時資訊不同的情況下，時空自我迴歸移動平均模式預測能力的差異。 研究結果顯示：兩種模式之模式估計誤差與預測誤差都很低且非常接近，顯示兩者都適合用來預估市區路段的交通量。然而，整合自我迴歸移動平均模式每個偵測器最多要校估5個參數，而時空自我迴歸移動平均模式卻只需要6個參數，因此當路網中的偵測器數量增加時，較簡單之時空自我迴歸移動平均模式較適合用來預估整體路網之交通量。另外，交通量並非一個獨立的系統，而是會受到附近地區交通狀況之影響，因此時空自我迴歸移動平均模式將時間序列之空間關係考慮進模式中，的確會提升模式之預測能力。測試可獲得即時資訊不同的情況下模式之預測能力，結果顯示利用即時資訊來預測較利用歷史資料預測來得準確。

The interests and applications of short-term traffic forecasting have been growing in the recent years. Many of the applications in Advance Traveler Information System (ATIS) and Advance Traffic Management Systems (ATMS) , which aim at providing useful information to travelers and improving the overall efficiency of road network, require an estimation and forecasting of the traffic conditions of the network. With a historical database of past traffic data from various types of vehicle detectors, real-time traffic information is collected which will be used to estimate the current traffic conditions and predict the condition in near future. Whereas most of the literature focused on the traffic flow prediction on the freeways, modeling traffic flow in urban arterials is more challenging as there are disturbances such as motorcycles and traffic signals in urban area. In this study, traffic flow forecasting models for urban arterials are proposed. Seasonal autoregressive integrated moving average (ARIMA) and space-time autoregressive moving average (STARMA) model, which incorporates the spatial correlations of the time series, are investigated. A case study using the traffic data from 24 vehicle detectors in Taipei city, Taiwan are performed. The forecasting performance of STARMA model are also examined by static, 1-step ahead rolling and 2-step ahead rolling strategies when real-time information can be obtained. The findings of this thesis are as follows. The estimated results reveal that both ARIMA and STARMA model are suitable for traffic flows forecasting in urban area. However, in the ARIMA model, there are up to five parameters for each detector, whereas there are only 6 parameters in the STARMA model. With a large number of detector locations in the system to be forecasted, the STARMA model shows a relative simple structure as compared to the ARIMA model which is univariate in nature. Traffic flows of urban area are not an isolated system and will be influenced by the flows from other adjacent locations, consequently, STARMA model considering the spatial relationship between each time series can improve the forecasting accuracy. Finally, the results of forecasting performance tests prove that using real-time information to forecast is better than merely using historical information to forecast.