以動態規劃演算法求解考量選擇行為之網路營收管理問題

Abstract

美國航空公司(American Airlines)在解除管制(The Deregulation)之後，根據市場區隔與差 別定價的觀念，透過費率艙等的機制(fare class mechanism)與機位存貨的控管(seat inventory control)，成功地擊敗低價競爭的新進航空公司。從此，營收管理(revenue management, RM)在航空運輸業逐漸成為一項廣泛運用的技術。基於現今航空軸輻網路 的營運型態，營收管理研究的重心亦由當初單一航段之問題轉向網路型態之問題。此 外，過去的營收管理模式，大多假設各個費率艙等的需求為獨立，少有研究考量消費者 的選擇行為及衍生的需求相關性。因此，本研究以考量消費者選擇行為的網路營收管理 問題為研究的主要課題。目前求解網路營收管理問題，最常見的兩類方法為競價法(bid price control)和虛擬巢式法(virtual nesting control)，兩者各有優劣，但基本上運算都相當 繁瑣，且不管在方法理論上及實務運用上，均存在著相當的限制。另外，在考量消費者 選擇行為後，艙等的開放不再依費率呈槽式的配置，基於可能的艙等組合相當多，因此 要在網路的架構下，找出最佳的機位控管決策是一項極具挑戰性的研究課題。本研究計 畫以參數化(parameterized)的期望收益近似函數為基礎，發展可以兼顧求解品質與運算 效能的動態演算法。最後，本研究將以能反映實際運作情形的例題進行數值測試，以驗 證所發展演算法的可用性。

In the post-deregulation era, American Airlines implemented the fare class mechanism and the seat inventory control based on the concepts of market segmentation and price discrimination and eventually defeated the new airlines with low-price strategies. Since then, revenue management (RM) has become a common technique in the airline industry. Due to the current hub-and-spoke operation, the focus of the RM research has shifted from the traditional single-leg problem to the network-type problem. In addition, most RM models assume that the demands for the various fare classes are independent and do not consider consumer choice behavior and demand inter-dependency. Therefore, this research project chooses the network RM problem that considers the customer choice behavior as the key research topic. There are two mainstream approaches for the network RM problem: the bid price control and the virtual nesting control, which are both in general ill with complicated procedure and heavy computational load. In addition, there are considerable limitations for these approaches in terms of methodological theory and practical application. Moreover, after taking choice behavior into account, the optimal control is no longer nested in terms of fares. As there are numerous possible sub-sets of fare classes, finding the seat control decision under the network context is a challenging task. Based on a well-chosen parameterized function to approximate the expected revenue, this study plans to develop a dynamic algorithm with a good trade-off between solution quality and computational load. Finally, a numerical experiment that can reflect to real operation in the field will be performed to validate the applicability of the developed solution algorithm.