在消費者選擇行為下的航空網路營收問題

Abstract

在1979年美國解除管制後，航空公司根據市場區隔與差別定價的觀念，發展出費率艙等的機制(fare class mechanism)與機位存貨的控管(seat inventory control)。基於現今航空軸輻網路的營運型態，營收管理(Revenue Management)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。同時，過去的營收管理模式，大多假設各個需求獨立，少有研究考量消費者的選擇行為及衍生的需求相關性。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，均存在著限制，如進行控管時有必須重新求解靜態模式的議題存在。此外，在考量消費者選擇行為後，艙等的開放不再依費率呈巢式的配置；尤其是在網路的架構下，產品的組合相當多，找出最佳的機位控管決策是一項極具挑戰性的研究課題。本研究以參數式(parameterized)的期望收益近似函數為基礎，發展一個動態控管機位的演算法，不需要重覆求解，且可以輕易地產生產品開放的決策。並且透過數值測試的結果，該演算法的控管效果非常接近動態規劃的最佳控管決策，實務上應有應用的可行性。

After the De-regulation in the United States, airlines implemented the fare class mechanism and the seat inventory control based on the concepts of market segmentation and price discrimination. Due to the current hub-and-spoke operation, the focus of the Revenue Management (RM) research has shifted from the traditional single-leg problem to the network-type version. Meanwhile, 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. For the two mainstream approaches of the network RM problem: the bid price control and the virtual nesting control, both in general suffer from the procedural complexity and the heavy computational load. In addition, there are considerable methodological and practical limitations for these approaches. For example, there exists the issue of re-solving the static model when implementing the actual dynamic booking control. In particular, 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 products, making the seat control decision under the network context is a challenging task. Based on the parameterized function for approximating the expected revenue, this study develops a dynamic seat control algorithm, in which no-resolving is necessary and the approximate control decision can be generated easily. Based on the numerical experiment, the developed algorithm can achieve the result fairly close that of the optimal control from the dynamic program model and its applicability to the real-world problem is very promising.