基於智慧票卡資料探索旅次行為及旅行模式

Abstract

近年來由於自動收費系統的發展，智慧票卡逐漸成為大眾運輸系統重要的付費工具。相較於傳統調查方式，智慧票卡所記錄的票證資料可以被自動且持續性地儲存所有旅次的相關資料，譬如交通工具、時間及起訖站點等，因此票卡所儲交易紀錄成為交通分析上新興的資料型態。藉由了解大眾運輸使用者的行為既可以提供更好的服務也可以幫助交通部門的規劃與建設。本研究使用北市交通局提供之二零二零年三月的智慧票卡之票證資料，透過其中所載旅次紀錄之時空特性，探索其所可能隱含之旅次行為與旅行模式。資料包含了臺北市主要的三種大眾運輸工具（捷運、公車、以及公共自行車）的票證交易紀錄。本研究的目的是利用上述資料，對臺北的大眾運輸用戶的旅行模式進行分類和深入的分析。藉由探討時間和空間變化、活動模式、社會人口特徵和運具選擇有關的特徵來分類出具有類似性質的群體。 過往多數智慧票卡票證資料的研究主要關注於旅次目的推估、下車地點與主要活動地點之估計以及活動模式的偵測及預測，針對週間旅次行為與旅行模式的多屬性分析則相對有限。本研究建立25個用於描述旅客行為的變數，透過因素分析消除共線性，再利用k -prototype分群法將活動型態進行分群。其結果顯示，分群模式將所有大眾運輸使用者依其旅次行為分為八群，此八個群集皆有不同的特性。四個群組是由上班族和學生組成的定期用戶，他們主要在週間進行通勤旅行，其中一些人亦在週末進行休閒旅行。其餘四個群組是非經常用戶，由當地的休閒旅行者和為旅遊和商務目的而旅行的遊客組成。接著根據地點訪問頻率、活動時間及空間距離等條件建立了預估使用者主要活動地點及家庭位置的模型。藉由對家庭位置的確認進一步利用內政部資料開放平臺之村里收入中位數及人口組成資料，補完智慧票卡票證資料所缺失的持卡人資訊。基於上述分析方法，本研究提供針對票證資料的分析架構，協助研究者了解旅客的旅行模式，分析成果並可提供給政府和相關單位在規劃大眾運輸或者城市發展時參考。

In recent years, smart cards have gradually become a critical payment tool in the mass transportation system due to the development of Automatic Fare Collection (AFC) technology. Compared with the traditional survey method, trip information such as timestamps, origin and destination stations of a trip by transit can be collected automatically and constantly by the AFC system. Hence, smart card data become an essential source for transportation planning and travel-related research. Understanding the behavior of mass transit users may help the planning and construction of the transportation sector and thereby enable better service. This study uses the ticketing data of Easy Card recorded during March 2020 provided by Taipei Transportation Bureau and explores the possible hidden trip behaviors and travel patterns through the temporal and spatial characteristics of the trip records contained therein. The data includes the ticket transaction records of the three main public transportation modes (MRT, bus, and bike-sharing system) in Taipei City. This research seeks to leverage the data to classify and analyze the travel patterns of mass transit users in Taipei. Groups with similar properties are identified by exploring temporal and spatial variations of activity patterns. According to literature review, most of existing studies focused on constructing complete trip information, estimating the trip purpose, and detecting or predicting transfer and activity patterns. However, the weekly travel behavior and multi-attribute analysis of travel patterns are relatively limited. This study adopts 25 variables to describe passenger behaviors while eliminating collinearity through factor analysis and then uses the k-prototype to cluster the activity patterns. The results show that the clustering model divides all mass transit users into eight groups characterized by different trip behaviors. Four clusters are frequent users of commuters (including students) who travel for commuting on weekdays and some leisure trips on weekends. The remaining four groups are casual users, consisting of business travelers, tourists, or local leisure travelers. Thereupon, a model is established to estimate the user’s main activity location and home location according to location visit frequency, activity time and spatial distance. By confirming the residential location, the income level and population composition data of the villages are further used to fill in the cardholder information that is unavailable in the smart card data. The research insights derived from this study may be references for the government and relevant authorities in planning transit systems or urban development.