利用資料驅動方法優化共享單車站點之容量規劃策略

Abstract

公共共享單車系統經常面臨租借與歸還效率不彰的問題，導致整體營運效能低落並影響服務品質。此問題主要來自兩種極端情況：空站（無可租車輛）與滿站（無可歸還空位）。本研究提出一套離散時間的建模框架，將「租借」與「歸還」視為兩個獨立的隨機過程，並據以建構站點的無邊界車輛庫存變化過程，使得後續可對各站在不同容量情境下進行服務效能模擬與損失量化分析。 本研究的核心價值，在於發展出一套創新的模組化隨機過程：加註伽瑪抽薄複合過程（Marked Compound Gamma-Thinned Process），並結合自歷史資料中學習出的非齊次隨機強度函數（Non-Homogeneous Stochastic Intensity Function），用以刻劃單車事件發生的時間與數量之雙重不確定性。該模型兼具靈活性與可解釋性，能有效模擬現實中的租還行為變化。進一步地，本研究設定兩項優化目標：最小化系統停機時間與最小化流量損失，並透過蒙地卡羅模擬探索最佳容量配置，以減少服務中斷與未滿足需求，提升整體營運績效。透過臺北市兩個 YouBike 站點的實證案例驗證，本研究提出之方法可顯著降低總體服務損失，展現出良好的實用性與應用潛力。

Public bike-sharing systems often encounter operational inefficiencies, primarily due to two conditions: empty stations (no bikes available for rental) and full stations (no available docks for returns). These disruptions negatively impact service quality and overall system performance. This study introduces a discrete-time modeling framework that treats rental and return events as two separate stochastic processes, enabling the construction of an unbounded inventory trajectory for each station. This formulation allows for comprehensive performance analysis under varying capacity configurations. The key contribution of this work lies in the development of a novel stochastic modeling approach—the Marked Compound Gamma-Thinned Process—which is integrated with a non-homogeneous stochastic intensity function estimated from historical bike usage data. This compound process captures both the timing and magnitude uncertainty of rental and return events, offering a flexible and interpretable framework for simulating realistic user behavior. Two optimization objectives are considered: minimizing system downtime (periods when rentals or returns are unavailable) and minimizing flow loss (unfulfilled demand due to capacity limits). Monte Carlo simulations are employed to identify the optimal station-wise capacity allocation that reduces service interruptions and unmet demand. The proposed method is validated using empirical data from two YouBike stations in Taipei City, provided by the open platform of Transport Data eXchange (TDX), and demonstrates significant improvements in reducing total service loss and enhancing operational efficiency.