- 學位論文
利用開放資料探討公共自行車營運服務水準與績效水準－以台中市YouBike 2.0為例
Assessing the Public Bike System Operational Level of Service and Level of Performance via Open Data: A Case Study of Taichung City YouBike 2.0
摘要
公共自行車具通勤與休閒的功能，騎乘中無噪音與空氣污染，更能達到環境永續，近年來在世界各地被廣泛建置。台灣地區以YouBike（微笑單車）為大宗，分布於12個縣市，其使用受到人口規模、建成環境、競合運具、天候與地理等因素影響，導致地區差異性大。過往文獻大多探討台北市或高雄市之公共自行車系統，較少見台中市相關之研究，而缺車/缺位、車輛週轉率等議題廣受各界關注之餘，尚缺乏一套整合的系統評估方法，作為主管機關、營運單位、使用者之間的溝通基礎。 有鑑於此，本研究以生活圈概念擇定台中市11個行政區、911個YouBike站點，蒐集2023年8月18日至2023年9月28日每5分鐘的各站車輛數與空閒車柱數（本研究稱站點即時資料，屬開放資料），以及使用者借還時間與起迄點交易紀錄141萬餘筆（本研究稱租借交易資料，透過台中市交通局申請獲取），創建適用於公共自行車的服務水準(Level of Service, LOS)及績效水準(Level of Performance, LOP)雙指標評估模型。服務水準反映使用者關切的借還便利性，沿續先前研究之結果，以站點即時資料計算各站乃至全系統之可靠度，亦即研究期間之缺車與缺位發生風險機率之對稱差集結果。然而，高可靠度恐存在低使用率的隱憂，故本研究首創績效水準指標，以租借交易資料計算日平均站點週轉率，亦即每站每柱日均借車與還車總數，來評量營運單位關切的微觀營運績效，並可輔助現行常見的巨觀（全系統）營運績效－車輛週轉率。 考量到YouBike 2.0系統站點密度高，鄰近站點有互相替代與支援的效果，與YouBike 1.0系統各站分散、獨立運作狀況不同，故每站之服務水準指標與績效水準指標係對應其方圓200公尺涵蓋「站點群組」的加權平均值，另依據使用者問卷調查結果設定可接受門檻值，將911個站點/站點群組分成(1)高服務水準且高績效水準、(2)低服務水準且高績效水準、(3)高服務水準且低績效水準、(4)低服務水準且低績效水準四種「站點組別」，各組透過地理資訊系統擷取站點周邊社會經濟特性、建成環境、土地使用類別等變數資料，以決策樹與隨機森林二種機器學習演算法探究各組之特性。二種演算法準確率皆超過80%，其分析結果相近，故皆採用之。具體研究發現如下： 一、透過時空分析發現，平日使用尖峰與一般通勤尖峰時段相符（上午7~9時、下午5~7時）、假日尖峰則出現在下午4~6時。租賃時間多為30分鐘內（中位數為9~12分鐘），對應之旅次距離中位數為1.2~1.7公里、90百分位數可達3.1~4.5公里。熱門旅次起迄短點多為高中大專院校周邊、中心商業區、行政辦公區以及交通場站。 二、營運績效方面，研究期間一般自行車車輛週轉率為3.49次/日/車，電輔車為4.25次/日/車，前20名站點之站點週轉率達10次/柱/日；可靠度方面，9月各級學校新學年開始後，可靠度與8月相比偏低，尖峰時段全系統可靠度甚至僅有70%，缺車狀況比缺位狀況嚴重約8倍。 三、服務水準與績效水準俱佳之站點占總量3.95%~7.68%（平日36站、假日70站），大多數附近有公車停靠站與路邊停車格，周圍人口密度小於約19,000人/平方公里，站點規模介於19~44柱，且站點所在路寬大於8公尺；服務水準與績效水準俱差之站點占總量3.51%~3.84%（平日35站、假日32站），大多數站點附近無公車停靠站，周邊站點數量小於1站，且位於公園附近；服務水準差但績效水準佳之站點占總量12.73%~5.26%（平日116站、假日48站），大多數站點周圍家戶數大約1,800戶，且站點所在路寬大於8公尺；服務水準佳但績效水準差之站點占總量79.47%~83.53%（平日724站、假日761站），此組周邊站點數量小於1站，周圍人口密度小於約19,000人/平方公里，且站點規模小於19柱。 本研究亦向主管機關與營運單位提出以下二點建議，作為提升公共自行車服務水準與績效水準以及後續研究之參考： 一、基於尖峰時段可靠度比全天可靠度低，營運單位應於尖峰時段加強調度運補，或者在系統可靠度低於80%情況下，發展尖峰定價模式；主管機關應檢討設站標準，站點周圍人口密度須達到標準（如19,000人/平方公里），或者是所在地路寬須大於8公尺方得設站。 二、後續研究可嘗試套用其他站點組別劃分方式，並對每個組別之特性進行探討；此外，建議主管機關應儘速完成圖資的建置，以及提升開放資料存取便利性。
並列摘要
Public bikes system have been considered an alternative mode of transport for first/last mile trips. It not only reduces air pollution but also protects the environment simultaneously and meets the goal of sustainable development. In Taiwan Region, the public bikes system known as "YouBike" holds the largest market share and has been widely implemented in recent years across every city and county. Its usage is affected by population size, built environment, competing transportation modes, weather, and geography, leading to significant regional variations. Previous literature takes Taipei City and Kaohsiung City as a case study largely, but Taichung City has hitherto not been explored in the literature. In addition, the issues about bikes and docks shortages attention by public, but an integrated evaluation approach has been lacking as a basis for interaction between administration, operational company, and users. Given that the reason above mention, this study selected 11 districts in Taichung City as case study, collected two different types of data, one is the number of bicycles and available docks at each station every 5 minutes from August 18, 2023, to September 28, 2023, referred to as "Station-based data". The other type of data includes records of user rental and return times and transaction data for origin-destination(O-D) pairs, referred to as "Trip-based data". We employed these data to establish a dual-indicator evaluation model applicable to public bikes system, composed of Level of Service(LOS) and Level of Performance(LOP). Building on previous studies, LOS reflected convince of rental and return which users concerned. The reliability indicator was existed, referring to symmetry complement between probability of bikes and docks unavailable or shortages. Nevertheless, high reliability concerns low usage. Therefore, we developed new LOP indicators using the station turnover rate calculated from trip-based data. Station turnover rate is equal to the total number of rentals and returns per dock per day at each station, evaluates the micro-level operational performance of concern to operators, and complements the macro-level daily average bike turnover rate of entire system. Considering the high station density of the YouBike 2.0 system in urban areas, neighboring stations can support each other, unlike the YouBike 1.0 system where each station operates independently and is more dispersed. Therefore, LOS and LOP for each station correspond to the weighted average values of the "Station Group" within a 200-meter radius. Based on objective classification criteria, stations/station groups are divided into 4 "Station Class": (1) high LOS and high LOP, (2) low LOS and high LOP, (3) high LOS and low LOP, and (4) low LOS and low LOP. Subsequently, we utilized Geographic Information Systems (GIS) to identify various characteristics, including but not limited to the socioeconomic attributes of each station. Finally, we investigated the characteristics of stations within each class using decision tree and random forest approaches. We adopted the result generated by both models based on their accuracy rate of more than 80%, and their analysis results are similar. The significant findings include the following main items: A. We found that peak usage times on weekdays are similar to typical commuting peak hours (7~9 am. and 5~7 pm.), whereas weekend peaks occur between 4~6 pm. by spatio-temporal analysis. Most rental durations are within 30 minutes (with a median of 9~12 minutes). Correspondingly, the median trip distance is 1.2~1.7 kilometers, with the 90th percentile reaching 3.1~4.5 kilometers. Popular trip origins and destinations are primarily around high schools, universities, central business districts (CBD), administrative areas, and transportation stations. B. In terms of operational performance, the bike turnover rate of normal bikes is 3.49 times/day/bike, and electric-assisted bikes are 4.25 times/day/bike in whole research period. In terms of reliability, in September was lower compared to August, coinciding with the start of the new semester for schools. During peak hours, the overall system reliability dropped to as low as 70%, with bike shortages being approximately 8 times more severe than dock shortages. C. Stations with both high LOS and LOP accounted for 3.95% to 7.68% of the total (36 stations on weekdays, 70 stations on weekends). Most of these stations were located near bus stops and curb parking spaces, with surrounding population densities less than approximately 19,000 people per square kilometer, station scale ranges from 19 to 44 docks as well as located at roads which wider than 8 meters. Stations with both low LOS and LOP accounted for 3.51% to 3.84% of the total (35 stations on weekdays, 32 stations on weekends), most of these stations have not nearby bus stops, the number of surrounding stations is less than one, and they are located near parks. Stations with low LOS but high LOP accounted for 12.73% to 5.26% of the total (116 stations on weekdays, 48 stations on weekends), most of stations have approximately 1,800 households in their surrounding areas as well as located at roads which wider than 8 meters. Stations with high LOS but low LOP accounted for 79.47% to 83.53% of the total (724 stations on weekdays, 761 stations on weekends). Stations in this class with surrounding population densities less than approximately 19,000 people per square kilometer, the number of surrounding YouBike stations is less than one, as well as station scale less than 19 docks. We provide strategy suggestions for administration, and operational company as below, as a reference for enhancing the LOS and LOP of public bicycle systems and for future research: A. Given that peak hour reliability is lower than overall daily reliability, operational company should enhance dispatching and replenishment during peak hours. Therefore, we suggest strengthening dispatch during peak hours or developing a peak pricing policy when the reliability indicator falls below 80%. The administration should review station settings standards, ensuring that the surrounding population density meets the criteria(e.g., 19,000 people per square kilometer) or that the road width at the station location exceeds 8 meters. B. Future research could explore other methods of station class classification and investigate the characteristics of each class. Additionally, it is recommended that administration expedite the completion of mapping data and improve the accessibility of open data.