基於智慧票卡資料探索旅次及活動型態：土地利用與大眾運輸系統空間特性之觀點

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。由於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有關旅次的資料比如時間、起訖站點等等，因此相關紀錄成為交通分析上新興的資料型態。本研究擷取臺北市二零一九年五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活動型態及轉乘行為。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。本研究首先藉由初步分析確認資料的可用性，並經由分析發現捷運以及公車每天的旅次數會遵循固定的模式，而公共自行車每日之旅次數則會受到氣候的影響。 根據文獻回顧，多數智慧票卡票證資料的研究主要關注於目的推估、轉乘判斷以及活動模式的偵測及預測，針對活動模式與土地利用之間的關係之資料實證分析則相對有限。為釐清土地利用與大眾運輸旅次、活動及相關轉乘行為之關聯，並考量票證資料之特性，本研究捨棄了傳統以使用者為出發探討個人活動模式的方式，首先將旅次資料轉換為活動資料，並改採統計各個網格區域內活動的開始時間以及持續時長作為其特性；同時，本研究將乘客前後採用不同運具以及相同運具的活動分別統計以利轉乘活動與一般活動之判斷、分析。本研究將所彙整之數據轉換為二維的圖片，並利用卷積以及池化擷取圖片中的特性並同時達到降維的效果。爾後，再利用k平均算法將活動型態進行分群，分群模式將所有之網格區域依其相應之活動模式分為八群，此八個群集皆有不同的特性。其中主要活動包含了轉乘、過夜、上班及上學、以及短時長活動。本研究並以多項羅吉特模型進一步探討屬於不同群集之網格及其土地使用、大眾運輸系統和相關設施之間的關係。基於上述分析方法，本研究提供針對票證資料的分析架構，協助研究者了解區域的活動特性、土地使用與運輸系統之間的關係，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。

The smart card has become a vital paying method because of the development of the automated fare collection (AFC) system. Accordingly, trip information such as timestamps, origin and destination stations of a trip by transit can be collected automatically and continuously by the AFC system. Thereby, smart card data become an essential source for the analysis on travel-related research. This study applied one-month smart card data of May 2019, collected from three main public transportation systems, including the metro, bus, and public bike-sharing system, in Taipei, Taiwan. The preliminary analysis showed that numbers of trips each day by bus and metro followed regular patterns, while the weather influenced the pattern of the bike-sharing system significantly. This result could be reasonable and suggest the usability of the datasets. According to the literature review, previous studies focused on constructing complete trip information, detecting or predicting transfer and activity patterns. Relatively empirical studies have been conducted to investigate the relationship between the trip pattern and the land-use related data. This research circumvented the analysis method from the perspective of individual passengers in light of the characteristics of the dataset. The trip data were first transformed into activity data, and the study area was divided into grids. The activity pattern for each grid was characterized by the starting time and duration of activities. Furthermore, activities were categorized into two classes based on whether the different or same transportation modes were used before and after the activity. Then, the activity patterns were treated as figures, and the convolution and max pooling methods were used thereupon to reveal the characteristics of the figures and reduce the associated attributes. The k-means clustering method was further applied to distinguish the patterns. Eight clusters with different features were identified where the activity features considered primarily included transfer, staying overnight, working and study, and short-duration activities. A multinomial logit model was developed at last to explore the linkage between the cluster characteristics and other land-use related data. To sum up, this research proposed an analysis framework that can help researchers better understand the activity pattern and how it can be influenced by land-use and transit systems. The research insights derived from this study may be references for the government and relevant authorities in terms of planning transit systems or the development of urban areas.