基於GPS之行動通訊網路訊號交通資訊估測方法之改善

Abstract

隨著市區道路使用者大幅增加，提供道路用者即時的交通資訊，可以幫助他們避開壅塞的地區節省旅行時間。在交通資訊收集的部分，透過追踪手機的位置來推測道路交通資訊的方法被稱為蜂窩浮動車數據(Cellular Floating Vehicle Data, CFVD)，因手機普及率高且不需要額外的偵測裝置，回報資料數量較多且更具效益。羅嵐茵提出一種新穎的CFVD演算法來估計市區主要幹道的交通資訊，然而由於沒有實際路況資料，導致無法驗證成效。 本研究利用 GPS 軌跡資料擷取出實際路況資料，提出了一些的CFVD演算法的分析來評估。我們首先利用實際路況資料驗證市區主要幹道的交通資訊估計是否準確，另外我們嘗試在其他平面道路使用該方法，並分析其結果是否有效，最後我們藉由改善自由流指標與旅行時間的對應表改善旅行時間估計的誤差。實驗數據顯示，該方法在光復路上的確可提供準確且穩定的即時交通資訊。但是在其他行動通訊網路訊號數量不足的路段無法提供準確的交通情況。另外，我們確實可以藉由改進對應表提高旅行時間估計與實際路況資料的相關係數並降低絕對誤差平均值。

With the increasing number of road users, it is important to provide real-time traffic information to road users so that they can avoid congested areas and save travel time. The approach that tracks mobile phones’ locations to estimate road traffic information has been referred to as Cellular Floating Vehicle Data (CFVD). CFVD can be very cost-effective because it uses the control signals of cellular networks to collect real-time traffic information; no additional device is needed to be deployed or maintained. Lo proposed a novel CFDV algorithm to estimate the travel time of urban arterial roads. However, she did not verify the experiment results because of lacking the ground truth. In this thesis, we obtain the ground truth from GPS-equipped probe vehicles and analyze the performance of Lo’s CFVD algorithm. First, we verify the accuracy of Lo’s CFVD algorithm on Lo’s experiment urban arterial road. Second, we apply the algorithm to other urban arterial roads and check if it is still effective or not. Finally, we improve this algorithm by using the ground truth to adjust the correspondence table of the free flow index and travel time. The experimental results indicate the traffic condition on Guang-Fu Road provided by Lo’s CFVD algorithm is accurate; the MAPE with respect to the ground truth is less than 20% and the correlation coefficient is greater than 0.6. However, the results of other experimental urban roads indicate that this algorithm may not detect traffic congestions when CFVD are insufficient. Finally, we reduce the MAPE between the CFVD travel time and the ground truth up to 20.9% (6.42% in average for three experiment road segments) by adjusting the correspondence table using hill climbing algorithm.