由於經濟成長及政府長期以來以公路交通為導向之城鄉發展政策,近年來民眾為追求便利性而導致小客車之持有率逐年攀高,與之相應也產生了停車空位不足的問題,在人口密集之都市地區,此情況顯得更為嚴重。在面對交通工具的選擇時,價格的考量常常是影響選擇的主要因素之一,因此近年來為抑制小客車的數量成長,道路收費與停車場之費率調整為一種有效且直接的方式,而其中停車場費率調整執行難易門檻較低,因此更受重視。另一方面,現有之停車場費率採用先到先停,固定費率之形式,此一成不變的商業模式除了難以因應未來之彈性停車場管理,同時也限制了停車場業者的收益。 為解決上述問題,本研究希望結合收益管理之理念,透過了解尖離峰時段顧客數量的差異,訂定相應之合適價格。在尖峰時段空位稀少時調高停車費率,使停車場業者賺取價差,同時也透過高價抑制該時段之顧客數量;而在離峰時段空位供過於求時,透過降低價格吸引更多顧客前來,消除剩餘空位。考量到停車場業者要實行動態價格可能有執行上的困難,例如公告價格的方式等,本研究將以預約系統的形式落實動態費率。 本研究首先假設停車價格具有彈性,即消費者之多寡可以價格控制。研究內容包含未來顧客數量預測與動態定價模型兩個主要部分。在未來顧客數量預測方面,首先將資料進行前處理,濾除漏失值和異常資料,之後透過ARIMA模型和離散時間傅立葉轉換法預測未來顧客數量變化情形。在動態定價模型部分,為防止特殊情況發生時當日顧客數量變化情形與預測不符,本研究首先建立基於卡曼濾波法和k最近鄰法的需求更新模組。該模組將透過所蒐集之即時資料更新預測,之後動態定價模型便可以此預測結果為基礎,產出定價結果。
Under considerations of economy growth and the car-oriented transport policy, the number of vehicles increases rapidly in recent years. This leads to the problem of lacking parking space, which becomes more and more serious. In large cities, this problem is even worse, where it causes inconvenience to the public. Since the price is important for the transport model choice, parking operators can control the demand of vehicle parking efficiently by increasing parking fee. On the other hand, because most of pricing policies related to parking are inflexible, the revenue of parking lot is capped. According to these reasons, the pricing schemes of parking should be improved. In order to solve the above problems, a dynamical scheme was proposed for the parking, that is based on the concept of revenue management. During rush hours, the price is higher. The parking lot operators can make more profits, although the number of customers is suppressed. During off-peak hour, the price is lower in order to attract customers. The parking lot operators may encounter lots of problems while implementing the dynamical pricing policy, such as the troubles in notifying customer prices. Therefore, we construct a dynamical pricing model which is based on reservation system. This thesis assumes that number of customers can be controlled by adjusting parking price. The research is mainly made of two parts, which are the dynamical pricing model and the forecast model. A data preprocessing module was used to remove the outliers and perform missing data interpolation. The forecast model based on ARIMA and Fourier transform is capable of predicting the number of customers. In order to predict the number closer to the actual situation, Kalman filter and k-NN algorithm are employed when updating prediction. The preferred price can then be set by the dynamical pricing model according to the prediction.