變動k值之k-NN法應用於高速公路旅行時間預測之研究

Abstract

提供準確的旅行時間給用路人能使其進行更妥善的旅次規劃，決定出發時間與路線選擇，進而達到分散車流、紓解交通擁擠之目的。本研究提出一套以 k-NN演算法為基礎進行改良之旅行時間預測模組，並應用於高速公路此類無號誌路段上。本研究之變動 k 值旅行時間預測模組先以一般 k-NN 法依照施測之歷史旅行時間資料庫進行校估，得到預測使用之 k 值，再以現況車流狀態進行 k 值之變動調整，相較於一般 k-NN 方法所使用之 k 值為一定值，本研究預測使用之變動 k值能依各時階之車流狀態作更適合之配適。本研究利用國道高速公路之車輛偵測器（Vehicle Detector,VD）作為資料來源，並對高速公路尖峰時段進行實測，根據實測結果，使用本研究所提出之變動 k 值方法相較於單純使用 k-NN 方法進行預測其預測結果更為準確。

Providing accurate travel time to travelers could not only make them do proper trip plannings but the decisions of depart time and route choices, achieving the goal of distributing quantity of vehicles and releasing traffic congestions. This research provide a travel time prediction model which improved from k-NN method and applied it on the segment of freeway which without signalize intersections. This proposed dynamic k-NN travel time prediction model get it’s primal k from k-NN method, and adjust the value k base on the situation of current traffic characteristic. Compare with k-NN method, the prediction model this research proposed can make the value k be more suitable for each time segment. This research using the vehicle detectors on the freeway as data source , and test the model with the peak hours of the freeway. Based on the testing results, the performance of the prediction model this research proposed is better than that of using k-NN method only.