利用集群分析方法產製站牌班表之研究

Abstract

近年來隨著私有運具持有數的持續成長，許多交通問題因而產生，對此，政府冀望透過建置先進公共運輸系統的方式將公車資訊服務提供予使用者、縮短乘客候車時間，常見的作法為透過智慧型站牌提供公車預估到站時間，但礙於成本因素而不易普及。過去的研究認為在交通狀況較為穩定的地區，公車行駛時間波動幅度較小，可透過產製張貼於沿線公車站牌上的靜態班表提供公車預計到站時間、減少乘客候車時間。 本研究將公車車載機所蒐集之到離站資料，先透過密度集群演算法去除離群值後、再使用下緣線基準法產製沿線各站的公車站牌班表，其特色為納入了多數司機員之駕駛習慣所產製而成。本研究利用公車到離站資料進行驗證，結果顯示本研究提出的方法所產製之站牌班表於離峰情境時確實能增加準點的資料筆數、降低了表定到站時間與實際到站時間的落差，提升了公車站牌班表的可信度。

In recent years, with the continued growth of private holdings, many traffic problems have arisen. In this regard, the Government has made it possible to provide bus information services to users through the construction of advanced public transportation systems(APTS) and to shorten passenger waiting time. It is common practice to provide bus time through intelligent bus stops, but it is not easy to be popular due to cost factors. Previous studies suggest that the region which traffic conditions are relatively stable makes the bus travel time fluctuation relatively minor, therefore we can post the static bus arrival schedule on the bus stop along the bus route to reduce passenger waiting time. In this study, we will use a systematic way to generate the bus arrival schedule, with the data collected by the bus-onboard-machine and then removed the outliers through the DBSCAN algorithm, to incorporate the driving habits of the majority of drivers. By using previous bus data to verify, the results show that the method proposed by this study can increase the amount of data on time, reducing the time drop between bus arrival schedule and actual arrival time, enhancing the credibility of the bus arrival schedule.