本研究之目的在發展一套有效的方法來預測即時短期的高速公路交通狀態,在研究中所指之交通狀態為車流密度,定義為單位長度路段上的車輛數。本研究將車流密度分解為三個部分:規則密度型態、結構的差異以及隨機變異。由於密度規則型態可以利用已知歷史資料之中位數來表示,因此,預測車流密度的問題也等價於預測車流密度與規則型態間之差異。本研究所發展的結構狀態空間模型包括狀態方程式與量測方程式;在狀態方程式方面,本研究提出採用m階之多項式趨勢模型來描述密度之結構差異。接著本研究設計適應性卡爾曼濾波器演算法,利用遞迴方式求解狀態方程式與量測方程式,以獲得即時之交通狀態結構差異。在實際測試時,本研究以2010年2月到5月平常日國道五號主線上偵測器的密度資料為歷史資料,求得各時階的密度中位數,預測2010年6月平常日的車流密度,並以多個常用指標來衡量預測結果。研究結果顯示本研究所發展之模型能夠有效預測高速公路即時交通狀態。
This study aims to develop an effective approach for predicting real-time short-term traffic states on the freeway. The traffic state of particular interest is the density of a traffic stream, defined as number of vehicles in a unit length of road segment. We assume the actual traffic density as the combination of the regular density pattern (i.e., historical trend), structural deviation from the regular pattern (i.e., the variation in travel time), and random fluctuation. Since the regular density pattern is represented as the median of historical densities, predicting traffic density is equivalent to predicting the structural deviation of traffic density from the regular pattern. The proposed structural space model consists of state and measurement equations. In the state equation, an m-order polynomial trend model is adopted to describe the structural deviation of density. Then, an adaptive Kalman Filter algorithm was developed to solve recursively the two equations and to obtain predicted real-time densities. The traffic density data collected by the loop detectors on freeway number 5 from February to May, 2010, were used to determine the regular pattern. The proposed approach was then used to predict traffic densities on freeway number 5 in June, 2010. The results show that the approach is effective in predicting real-time freeway traffic states and superior to a commonly-used method in the literature.