本研究以模擬分析為基礎，構建固定式偵測器與移動式時空資料之動態轉換函數，並將兩種不同資料來源之時空一致化，進行旅行時間推估之資料融合運算。 本研究以一實際都市幹道為範圍，經模擬軟體(PARAMICS)模擬調查環境中固定式偵測器與移動式偵測器(即探針車)之行進軌跡並輸出模式運算所需交通資料與旅行時間真值。其中，固定式偵測器之資料運算，係以每時區(5分鐘)之佔有率所推算之路段密度為基礎經轉換函數運算得空間平均速度後以計算推估旅行時間；而以探針車資料為主之運算則以購建多輛個別之軌跡速率並加權之平均速率與真實平均旅時間之動態轉換。之後的融合運算則是運用經推導正規化之迴歸式後校估其對應之係數作為融合所需之加權係數。 本研究並以另加入不經過轉換函數處理之推估旅行時間基本模式與運用轉換函數推估旅行時間之模式進行比較。經以平均絕對誤差百分比(Mean Absolute Percent Error, MAPE)為指標之評量結果顯示，在移動式轉換函數模式部份，以探針車之軌跡點速度作為轉換函數之模式輸入項之旅行時間推估模式有良好的績效表現，其推估旅行時間校估之MAPE在1.96%到5.62%之間，模式驗證部分在2.77%到4.34%之間，整體平均為3.47%；而以固定式偵測器資料為基礎之轉換函數模式其旅行時間推估部分之MAPE在8.28%到26.98%之間; 模式驗證部分則在1.99%到26.94%之間，整體平均為16.45%；融合模式推估旅行時間校估部分在0.4%到10.43%之間，驗證部分在0.15%到9.27%之間，整體平均為4.52%。因此就本研究而言，以探針車資料為基礎之轉換函數模式具有相當優良之績效表現，而融合模式可有效平衡兩不同資料來源之旅行時間推估差異；此外相較於基本模式之MAPE值在6.77%到16.16%之間，運用轉換函數模式之旅行時間推估績效明顯較佳。 Dual-loop vehicle detectors and/or probe vehicle technique are two most commonly used devices to perform instantaneous (real time) traffic data collection for either traffic management or travel information. In particular, travel time information is of the major importance which can not be obtained directly but derived from various models/algorithms using detected traffic flow related data as input.
Transfer function methods with the capability to calibrate bivariate relations using time series data was recognized to capture the effect of dynamic nature real time data instead of the commonly used static or steady-state relation. It is therefore the focus of this thesis to develop an integrated procedure of the above-mentioned two difference data sources in the application of estimation of travel time along a specific road segment. Transfer functions were calibrated for traffic data from loop detector and probe vehicle data streams respectively for each unified space-time interval to recognize the inherent differences between them. Finally, data fusion technique was applied to obtain final estimate of travel time by integrating both estimates from fixed loop detector and probe vehicle. A basic model without applying transfer function technique was also constructed to serve as a benchmark, from which the credibility of transfer function was presented.
This study employed Paramics, a microscopic traffic simulation program, as primary research tool of data sources to evaluate the proposed issues. Care has been taken to calibrate simulation parameters using true field survey data to ensue the consistency.
Travel time estimations were performed and evaluated for four models: exclusive loop-detector data with transfer function, exclusive probe vehicle data with transfer function, fusing both data with transfer function, and fusing data without transfer function. Mean Absolute Percent Error (MAPE) values were calculated as evaluation criteria compared with simulated average over all true travel times traced by all vehicle trajectories within the studied space-time. The results showed that travel estimation by probe vehicle data with transfer function performs better than that estimated by vehicle detector data with transfer function. And the fusion model acted as balance role to bridge the deviation of travel times from both detector and probe vehicle models. Finally, comparing with base case, where no transfer functions applied, we demonstrated the quality of transfer function by significant improvement of MAPE values over those of the base model.
