## Application of a Hybrid Genetic Algorithm to Airline Crew Scheduling (1996)

Venue: | Computers & Operations Research |

Citations: | 21 - 0 self |

### BibTeX

@ARTICLE{Levine96applicationof,

author = {David Levine},

title = {Application of a Hybrid Genetic Algorithm to Airline Crew Scheduling},

journal = {Computers & Operations Research},

year = {1996},

volume = {23},

pages = {547--558}

}

### Years of Citing Articles

### OpenURL

### Abstract

This paper discusses the development and application of a hybrid genetic algorithm to airline crew scheduling problems. The hybrid algorithm consists of a steady-state genetic algorithm and a local search heuristic. The hybrid algorithm was tested on a set of forty real-world problems. It found the optimal solution for half the problems, and good solutions for nine others. The results were compared to those obtained with branch-and-cut and branchand -bound algorithms. The branch-and-cut algorithm was significantly more successful than the hybrid algorithm, and the branch-and-bound algorithm slightly better. 1 Introduction Genetic algorithms (GAs) are search algorithms that were developed by John Holland [17]. They are based on an analogy with natural selection and population genetics. One common application of GAs is for finding approximate solutions to difficult optimization problems. In this paper we describe the application of a hybrid GA (a genetic algorithm combined with a local s...

### Citations

8314 |
Genetic Algorithms
- Goldberg
- 1989
(Show Context)
Citation Context ...opulation each generation by their offspring and is the traditional genetic algorithm defined by Holland [17]. The hope is that the offspring of the best strings carry the important "building blo=-=cks" [12]-=- from the best strings forward to the next generation. The GRGA, however, allows the possibility that the best strings in the population do not survive to the next generation. Also, as pointed out by ... |

3037 | H.: Adaption in Natural and Artificial Systems - Holland - 1975 |

1314 | Handbook of Genetic Algorithms
- DAVIS
- 1991
(Show Context)
Citation Context ...the best strings forward to the next generation. The GRGA, however, allows the possibility that the best strings in the population do not survive to the next generation. Also, as pointed out by Davis =-=[6]-=-, some of the best strings may not be allocated any reproductive trials. It is also possible that mutation or crossover destroy or alter important bit values so that they are not propagated into the n... |

1140 |
Integer and Combinatorial Optimization
- Nemhauser, Wolsey
- 1988
(Show Context)
Citation Context ...beyre et al. [2], "The knowledge of a family of good solutions is far more important than obtaining an isolated optimum." Fourth, the NP-completeness of finding feasible solutions in the gen=-=eral case [23]-=- and the enormous size of problems of current industrial interest make the SPP a good problem on which to test the effectiveness of GAs. 2 The Hybrid Genetic Algorithm Genetic algorithms work with a p... |

558 |
Uniform Crossover in Genetic Algorithms
- Syswerda
- 1989
(Show Context)
Citation Context ...the next generation by the parent's offspring. The steady-state genetic algorithm (SSGA) is an alternative to the GRGA that replaces only a few individuals at a time, rather than an entire generation =-=[26, 27]-=-. In practice, the number of new strings created each generation is usually one or two. The new string(s) replace the worst-ranked string(s) in the population. In this way the SSGA allows both parents... |

423 | A Comparitive analysis of selection schemes used in genetic algorithms, Foundation of genetic algorithms
- Goldberg, Deb
- 1989
(Show Context)
Citation Context ...ible solutions more attractive than feasible ones. 2.4 GA Operators The primary GA operators are selection, crossover, and mutation. We choose strings for reproduction via binary tournament selection =-=[12, 13]-=-. Two strings were chosen randomly from the population, and the fitter string was allocated a reproductive trial. To produce an offspring we held two binary tournaments, each of which produced one par... |

143 |
Some guidelines for genetic algorithms with penalty functions
- Richardson, Palmer, et al.
- 1989
(Show Context)
Citation Context ...that carry useful information but lie outside the feasible region will be ignored and their information lost. If the penalty term is not strong enough, the GA may search only among infeasible strings =-=[25]-=-. As part of the work in [20] we investigated several different penalty terms without any conclusive results. For the results reported in this paper we used the linear penalty term m X i=1si n X j=1 j... |

130 |
Solving airline crew-scheduling problems by branch-and-cut
- Hoffman, Padberg
- 1993
(Show Context)
Citation Context ...ticular flight leg is a cost, c j . The matrix elements a ij are defined by a ij = ( 1 if flight leg i is on rotation j 0 otherwise. (4) Airline crew scheduling is an economically significant problem =-=[1, 3, 11, 16]-=- and often a difficult one to solve. One approximate approach (as well as the starting point for most exact approaches) is to solve the linear programming (LP) relaxation of the SPP. A number of autho... |

99 |
GENITOR: A different Genetic Algorithm
- Whitley, Kauth
- 1988
(Show Context)
Citation Context ...the next generation by the parent's offspring. The steady-state genetic algorithm (SSGA) is an alternative to the GRGA that replaces only a few individuals at a time, rather than an entire generation =-=[26, 27]-=-. In practice, the number of new strings created each generation is usually one or two. The new string(s) replace the worst-ranked string(s) in the population. In this way the SSGA allows both parents... |

92 | Genetic algorithms are not function optimizers
- Jong
- 1993
(Show Context)
Citation Context ...that genetic algorithms are outperformed by traditional operations research algorithms. GAs are general-purpose tools that will usually be outperformed when specialized algorithms for a problem exist =-=[6, 7]-=-. However, several points are worth noting. First, the nw models are relatively easy to solve with little branching and may not by indicative of current SPP problems airlines would like to solve [15].... |

69 | A Parallel Genetic Algorithm for the Set Partitioning Problem
- Levine
- 1994
(Show Context)
Citation Context ...thm with a local search heuristic is beneficial. It combines the GA's ability to widely sample a search space with a local search heuristic's hill-climbing ability. Our early experience with the GRGA =-=[19]-=-, as well as subsequent experience with the SSGA [20], was that both methods had trouble finding optimal (often even feasible) solutions. This led us to develop a local search heuristic to hybridize w... |

44 |
Optimal solution of set covering/partitioning problems using dual heuristics, Management Science 36
- Fisher, Kedia
- 1990
(Show Context)
Citation Context ...d tree search may not be effective [1, 3, 11]. Exact approaches are usually based on branch-and-bound, with bounding strategies such as linear programming and Lagrangian relaxation. Fischer and Kedia =-=[10]-=- used continuous analogs of the greedy and 3 \Gamma opt methods to provide improved lower bounds. Eckstein developed a general-purpose mixed-integer programming system for use on the CM-5 parallel com... |

40 | Parallel genetic algorithms in combinatorial optimization
- Mühlenbein
- 1992
(Show Context)
Citation Context ...hether it duplicates a string already in the population. If it does, it undergoes (possibly additional) mutation until it is unique. 2.5 Local Search Heuristic There is mounting experimental evidence =-=[6, 18, 22]-=- that hybridizing a genetic algorithm with a local search heuristic is beneficial. It combines the GA's ability to widely sample a search space with a local search heuristic's hill-climbing ability. O... |

37 |
Parallel Branch-and-Bound Algorithms for General Mixed Integer Programming on the CM-5
- Eckstein
- 1994
(Show Context)
Citation Context ...to provide improved lower bounds. Eckstein developed a general-purpose mixed-integer programming system for use on the CM-5 parallel computer and applied it to, among other problems, set partitioning =-=[9]-=-. Desrosiers et al. developed an algorithm that uses a combination of Dantzig-Wolfe decomposition with restricted column generation [8]. Hoffman and Padberg report optimal solutions when they use bran... |

30 | Exact solution of crew scheduling problems using the set partitioning model: recent successful applications. Networks 11:165–177 - Marsten, Shepardson - 1981 |

24 |
The airline crew scheduling problem: a survey, Transport
- Arabeyre, Fearnley, et al.
- 1969
(Show Context)
Citation Context ...l methods may have trouble accommodating the addition of new constraints as easily. Third, at any iteration, genetic algorithms contain a population of possible solutions. As noted by Arabeyre et al. =-=[2], "Th-=-e knowledge of a family of good solutions is far more important than obtaining an isolated optimum." Fourth, the NP-completeness of finding feasible solutions in the general case [23] and the eno... |

22 |
A Global Approach to CrewPairing Optimization
- Anbil, Johnson, et al.
- 1992
(Show Context)
Citation Context ...ticular flight leg is a cost, c j . The matrix elements a ij are defined by a ij = ( 1 if flight leg i is on rotation j 0 otherwise. (4) Airline crew scheduling is an economically significant problem =-=[1, 3, 11, 16]-=- and often a difficult one to solve. One approximate approach (as well as the starting point for most exact approaches) is to solve the linear programming (LP) relaxation of the SPP. A number of autho... |

21 |
Optimizing flight crew schedules
- Gershkoff
- 1989
(Show Context)
Citation Context ...ticular flight leg is a cost, c j . The matrix elements a ij are defined by a ij = ( 1 if flight leg i is on rotation j 0 otherwise. (4) Airline crew scheduling is an economically significant problem =-=[1, 3, 11, 16]-=- and often a difficult one to solve. One approximate approach (as well as the starting point for most exact approaches) is to solve the linear programming (LP) relaxation of the SPP. A number of autho... |

17 | A Genetic Algorithm for the Set Partitioning Problem
- Chu, Beasley
- 1995
(Show Context)
Citation Context ...ve mutation rate or simulated-annealing-like move in the ROW heuristic to maintain diversity in the population in order to sustain the search warrants further investigation. Recently, Chu and Beasley =-=[5]-=- have shown that preprocessing the constraint matrix to reduce its size, as well as modifications to the fitness definition, parent selection method, and population replacement scheme lead to improved... |

13 |
A hybrid search for genetic algorithms: Combining genetic algorithms, tabu search and simulated annealing
- Kido, Kitano, et al.
- 1993
(Show Context)
Citation Context ...hether it duplicates a string already in the population. If it does, it undergoes (possibly additional) mutation until it is unique. 2.5 Local Search Heuristic There is mounting experimental evidence =-=[6, 18, 22]-=- that hybridizing a genetic algorithm with a local search heuristic is beneficial. It combines the GA's ability to widely sample a search space with a local search heuristic's hill-climbing ability. O... |

11 |
Airline crew scheduling - supercomputers and algorithms
- Barutt, Hall
- 1990
(Show Context)
Citation Context |

10 |
lp solve
- Berkelaar, D, et al.
- 2007
(Show Context)
Citation Context ...ize of the problem. However, we did not have access to such a capability. To try to gain some insight into the difficulty of the test problems, we solved them using the public-domain lp solve program =-=[4]-=-. This program solves linear programming problems by using the simplex method and solves integer programming (IP) problems by using the branchand -bound algorithm. The columns in Table 1 are the name ... |

8 |
private communication
- Gregory
(Show Context)
Citation Context ...ple many areas of the search space and let the GA discover the most promising ones. We developed a modified random initialization scheme. Block random initialization, based on a suggestion of Gregory =-=[14]-=-, uses information about the expected structure of an SPP solution. A solution to the SPP typically contains only a few ones and is mostly zeros. We can use this knowledge by randomly setting to one a... |

6 |
Application of combinatorial programming to a class of all-zero-one integer programming problems, Manayement Sci
- Pierce
- 1968
(Show Context)
Citation Context ... natural. A bit in a GA string is associated with each column j. The bit is one if column j is included in the solution, and zero otherwise. We also ordered the SPP matrix into block "staircase&q=-=uot; form [24]-=-. Block B i is the set of columns that have their first one in row i. B i is defined for all rows but may be empty for some. Within B i the columns are sorted in order of increasing c j . Ordering the... |

1 |
The Airline Crew Pairing Problem
- Desrosiers, Dumas, et al.
- 1993
(Show Context)
Citation Context ...mputer and applied it to, among other problems, set partitioning [9]. Desrosiers et al. developed an algorithm that uses a combination of Dantzig-Wolfe decomposition with restricted column generation =-=[8]-=-. Hoffman and Padberg report optimal solutions when they use branch-and-cut for a large set of real-world SPP problems [16]. Several motivations for applying genetic algorithms to the set partitioning... |