運用電子票證資料推估大眾運輸旅次訖點之演算法構建與驗證

Applying Smart Card Data to Estimate and Validate the Destination of Individual Trips in Transit Systems

目前臺灣地區大眾運輸的刷卡系統分為一段式刷卡與兩段式刷卡（里程計費）兩種方式，但由於一段式刷卡電子票證之內容僅記錄乘客上車（或下車）單一資訊，無法統計出路網中各路線各站起訖點間之運量資料（O-D table），在相關應用分析上造成很多限制，另在過去學術研究中亦尚未看到利用電子票證資料結合外部異質資料的推估演算法。有鑑於此，本研究旨在利用電子票證一段次刷卡紀錄，並結合電子票證與地區土地分區之資料，建立一段式電子票證刷卡資料的3 階段旅次訖點推估演算法。為驗證本演算法的推估正確性，本研究選擇公路客運與市區公車各一條路線，以及北臺灣某縣市市區公車路網的電子票證資料作為測試對象，測試過程首先以完整記錄兩段式刷卡資料的市區公車與公路客運路線進行測試，並將電子票證資料訖點資訊隱藏後進行推估，測試結果兩條客運路線的訖點推估率均為100%外，其訖點正確率亦分別為81.4%與75.3%；之後再以北臺灣某縣市的市區公車一段式電子票證刷卡資料進行旅次訖點推估，除推估率為100%外，其平日與假日推估旅次與交通部所提出運輸需求模式推估量之差異分別僅有6.47%與12.78%，顯示本研究所提出一段式電子票證刷卡資料的3 階段旅次訖點之推估演算法，可作為後續路線檢討與班次修正的重要參考。

The smart card fare collection system currently used in Taiwan accordingto one or two sections, but the one-section system only record the boardingtransactions (“tap-in”) and not the alighting transactions (“tap-out”) in thesystem. This results in many restrictions in the analysis of related applications.Moreover, past studies have yet to show estimation algorithms that make use ofthe smart card fare information linked with external heterogeneous data. In thispaper, we make use of the one-section smart card fare tallying record, combinedwith the smart card records and regional zoning information, in order toestablish a three-stage trip algorithm to estimate the destinations. Tests areperformed on a highway bus route, an urban bus route and the whole city busroutes in a county in northern Taiwan. The data from a highway bus route andan urban bus route record the complete two-section ticket tallying information.Estimation was carried out after the “tap-out” data from the smart cardinformation had been hidden. Results indicated that the estimation rates for bothroutes were 100% and the accuracy rate was 81.4% and 75.3% respectively. Onthe other hand, the data from a county in northern Taiwan was then used for theestimation of destinations and shows the 100% estimation rate. Besides,compared with the predict-trip volume data (O-D table) from the MOTC inTaiwan, the gap of the trip number estimation for weekdays and weekends hadonly 6.47% and 12.78% respectively. Results show that the proposed three-stagealgorithm would be useful references for the review in timetable setting and busrouting/scheduling in the future.