Applications of Constraint Programming and Mathematical Programming Techniques to Solve Large-Scale Cabin Crew Scheduling Problems

碩士 === 國立交通大學 === 運輸科技與管理學系 === 90 === Because crew cost is only second to fuel in airline operations, the crew scheduling problem has been widely studied in the literature. Most existing literature on this problem is based on mathematical programming (MP) models. This research takes a new approach...

Full description

Bibliographic Details
Main Authors: Kuo-Chen Wang, 王國琛
Other Authors: Anthony Fu-Wha Han
Format: Others
Language:zh-TW
Published: 2002
Online Access:http://ndltd.ncl.edu.tw/handle/13287679441306777875
Description
Summary:碩士 === 國立交通大學 === 運輸科技與管理學系 === 90 === Because crew cost is only second to fuel in airline operations, the crew scheduling problem has been widely studied in the literature. Most existing literature on this problem is based on mathematical programming (MP) models. This research takes a new approach using constraint programming (CP) to complement MP in solving large-scale crew scheduling problems. Essentially, a crew scheduling problem can be decomposed into a pairing generation problem, and a minimum-cost set partitioning problem (SPP). In this research, we took the pairing generation problem as a constraint satisfaction problem (CSP) and developed a CSP model to generate feasible pairings. Since the number of feasible pairings tends to be enormous, we developed a CSP-CP model to generate feasible constrained pairings (CP) by adding some expert-knowledge rules to CSP model. Considering most crew scheduling problems in practice are large-scale, we combine the CSP-CP model and SPP model to develop a CP-based column generation method. To evaluate the performance of this method, we adopted two large-scale problems in literature as our test problems, each has a problem size of 1.7*106 and 6*107 respectively. Computational results showed that CP-based column generation yields an improvement of 3% and 2.7% over the published best-known solutions of the two test problems respectively. Such results imply that the CSP-CP model can be used to complement the MP in column generation effectively. Because the CSP-CP model can effectively generate feasible pairings, we also applied the CSP-CP model to solve the crew scheduling problem in the “generate-and-optimize” framework. The generated CP includes a set of basic pairings (BP), each of which contains no deadheads, and a set of extended pairings, each of which contains exactly one deadhead. We found that the CP-based generate-and-optimize method yielded the same optimal costs as we obtained in the CP-based column generation method. Such results imply that the CSP-CP model may effectively reduce the problem size without losing solution quality in solving large-scale cabin crew scheduling problems.