地理資訊系統及資料探勘技術在公共自行車設站之分析與研究 - 以臺北市公共自行車系統為例

近幾年公共自行車租賃系統已逐漸盛行，自行車具有便利性、低成本、無汙染的優勢，深受短程接駁通勤族喜愛，相較其它交通工具而言，自行車則為綠色出行的代表，隨著臺北市因建成環境越趨成熟，公共建設的臨立，商業活動發展以及人口的移入，臺北市在土地取用更顯不足，臺北市公共自行車系統於2008年開始發展，截至2020年為止已在臺北市12個行政區陸續建置了400個站點，其會員有效卡片數量達12,740,136，騎乘次數達164,690,616，相當於每年可創造近千萬次騎乘，以2019年10月所統計臺北市有效人口數為264.6萬，也就是說居住於臺北市的居民，多數已有公共自行車騎乘體驗，根據臺北市2017年公共自行車使用特性大數據分析報告，臺北市公共自行車Youbike系統被定位為公共運具中「第一哩」及「最後一哩」低碳轉乘具運之一，車輛平均周轉率為8~10次，這些數據足以顯示公共自行車對於有高密度人口的都市具有顯著性的需求。隨著城市基礎設施日益數字化，人類行為數據已無處不在。模式識別和機器學習（Machine Learning，簡稱ML）技術對於整理和分析大量現實世界的行為數據已日趨成熟，現行人工智能可透過對歷史資料的分析達到預測效果，其需求預測的相關應用也已落實在我們生活之中，如預估臺北市某個區域房價的漲跌，或是分析消費行為進行喜好推薦等，近年臺北市因商業性發展、公共建設興起以及外來人口移入造成通勤及居住增加，其城市樣貌已逐漸在改變，臺北市可運用於公共自行站點佈建之合適區域已朝向減少趨勢，如何在有限土地條件限制下，讓土地邊際效益最大化，持續增建公共自行車站點，對於現今租賃系統將會遭遇到困難，因此本研究期望能針對公共自行車設址問題，透過人工智能方式（科學選點）找出最適合設址的區域，讓公共自行車深入巷弄，貼近使用者需求。在文獻回顧部份，分成自行車介紹及自行車文獻相關研究二個主題，本研究之目的在於了解地理環境特徵中那些特徵對公共自行車使用有其影響性，故在資料來源取用上，著重在開放資料及POI（Point of Interest）等特徵上，再結合地理網格方式採以GIS圖層分析，在研究方法方面，則採用機器學習方式，以LightGBM（Light Gradient Boosting Machine）決策樹模型對數百組資料集進行抽樣訓練及測試，以找出最佳化預測模型，本研究以臺北市Youbike（微笑單車）公共自行車做為研究目標，在使用量與特徵值關係檢定，採用斯皮爾曼等級相關係數，而模型效度方面則以R2（r-squared）決定係數及Shap value做為模型解釋之方法。本研究採用預測方式找出每個地理網格各特徵表現，對於市政單位或營運商日後在佈建租賃站點可做為參考之用，讓資源可有效的配置在需要的區域上，也可讓站點建置成本予以降低，以達到都市永續發展之目的。

關鍵字

公共自行車系統；開放資料；機器學習； LightGBM決策樹

並列摘要

In recent years, the public bicycle rental system has gradually become popular. Bicycles have the advantages of convenience, low cost, and pollution-free. They are popular among commuters with short-distance connections. Compared with other modes of transportation, bicycles are a representative of green travel. Due to the maturity of the built environment in Taipei City, the immediate establishment of public construction, the development of commercial activities, and the migration of population, Taipei City’s land acquisition is even more insufficient. Taipei City’s public bicycle system began to develop in 2008 and has been completed by 2020. 400 stations have been built in 12 administrative districts of Taipei City. The number of valid cards for members reached 12,740,136 and the number of rides reached 164,690,616, which is equivalent to creating nearly 10 million rides per year. According to the statistics of Taipei City in October 2019, the effective population is 2.646 million, which means that most residents living in Taipei have already experienced public bicycle riding. According to the 2017 Taipei City Public Bicycle Use Characteristics Big Data Analysis Report, Taipei City Public Bicycle Youbike system is positioned as one of the "first mile" and "last mile" low-carbon transfers in the transportation equipment. The average vehicle turnover rate is 8-10 times per day. These data are sufficient to show that public bicycles are significant for cities with high-density populations demand. As urban infrastructure becomes increasingly digitized, human behavior data also become ubiquitous. Pattern recognition and machine learning (ML) technologies have become increasingly mature for collating and analyzing large amounts of real-world behavior data. The currecent artificial intelligence can produce predictive results through the analysis of historical data, and its demand prediction related applications have also been implemented in our lives, such as predicting the rise and fall of housing prices in a certain area of Taipei City, or analyzing consumer behavior to recommend preferences, etc. In recent years, Taipei City has increased commuting and housing due to commercial development, the rise of public construction, and the immigration of foreign populations. The appearance of the city has gradually changed. The suitable areas in Taipei City that can be used for the deployment of public self-service stations are been decreasing. How to maximize the marginal benefits of the land under limited land conditions and continue to build public bicycle stations. This research hopes to address the problem of public bicycle location, through artificial intelligence (scientific point selection) to find the most suitable area for the location of rental site. This study focused on open data and POI (Point of Interest) and other features, and then combined with geographical grid methods to extract GIS layer analysis, in the research method, We adopt machine learning method, LightGBM (Light Gradient Boosting Machine) . The Gradient Boosting Machine decision tree model sampled and tested hundreds of data sets to find optimized predictive models. This study used Youbike (Smile Bike) public bicycles in Taipei City as research objectives, and used Spearman's rank correlation coefficient to examine usage and characteristic values, and model validity was interpreted with R2 (r-squared) decision factor and Shap value. This study uses model prediction to find out the characteristics of each geographic grid. It can be used as a reference for municipal units or operators in the construction of rental sites in the future, so that resources can be effectively allocated in the required areas. The cost of site construction can be reduced to achieve the goal of sustainable urban development.