Literature Study
Introduction
GENERAL PART ABOUT AIRLINE DISRUPTION MANAGEMENT……………..
General research is airline disruption management
Due to disruptions in the operations of the day airlines are forced to reschedule flights, crew and passengers. A small disruption can cause many problems for the rest of the operation. Daily flight schedulers are busy to reschedule flights after a disruption. The main aim is to have less cancellations and delays. They strive to return to the old schedule as soon as possible. First flights are rescheduled and after that a crew is assigned to the flight. If it is not possible to assign a crew to a flight the flight has to be rescheduled again. Passenger connections are not always taken into account with the decision making.
Nowadays flight schedulers are anticipating on disruptions and with their knowledge and insight they make decisions about what is best for the airline. Computerized models would definitely help the flight schedulers to make better decisions about which flights have to be rescheduled and which have to be cancelled. Many factors are involved in the decision making and that makes it a complex problem. Much research during the past years has been done to develop models to assist in the decision making of flight schedulers. The challenge is to develop a model that reschedules flights which takes crew assignment and passenger connection into account. Since, this is a very complex problem it is divided in three parts: aircraft assignment, crew assignment and passenger connection.
This literature study is organised as follows: First an overview of the literature about the aircraft assignment is described. Thereafter, an overview of the literature about crew assignment is described followed by an overview of the literature about rescheduling passengers. Finally, models and solution techniques are chosen for the dynamic crew pairing recovery problem.
Aircraft recovery problem
When a disruption occur flight schedulers are responsible for reschudeling flights, aircraft, crew and passengers. First the flight schedule is rescheduled after which an aircraft is assigned to the specific flights. Thereafter, a crew is assigned to the specific flights. When it is not possible to find feasible crew the flight schedule will be rescheduled again. Finally, passenger connections will be taken into account.
Within disruption management many research is done to the aircraft recovery. Probably this is because the rules for aircraft rescheduling are less complex than the rules of crew recovery. This paragraph gives a chronological overview of the research done in the past about the aircraft recovery problem.
Teodorovic & Guberinic (1984)
Teodorovic & Guberinic are the first researchers that came up with a model for the aircraft recovery model from an Operations Research perspective. In 1984 they introduced a model that minimizes the total passenger delays by swapping and delaying flights. They considered a situation where one or more aircraft are taken out of service. They solved the problem with a branch-and-bound method. A very simple example presents some results of the model. The solution does not consider maintenance constraints and uses a single fleet type.
Teodorovic & Stojkovic (1990)
In 1990 Teodorovic & Stojkovic continued with the model of Teodorovic & Gubernic. They included flight cancellations and station curfews. The objective is to minimize the total number of cancelled flights. When there are several schedules with an equal number of cancelled flights, the schedule with the minimum total passenger delay is chosen. They used a heuristic algorithm based on a sequential approach to solve the problem. The authors used a single fleet type and preflights are not allowed in the model.
Jarrah et al. (1993)
In the paper of Jarrah et al. two models are presented. The first model is only based on delaying flights, where the second model is only based on the cancellation of flights. The objective of both models is to minimum costs by swapping, ferrying or using spare aircraft. The problem is solved with a Busacker-Gowen's dual algorithm (SSPM???). The limitations of the models are that it does not take different aircraft types into account and the two models cannot consider cancellations and delays simultaneously.
Yan & Yang (1996)
Yan & Yang developed a model in which flight cancellation, delays and ferry flights were included. This was the first model that had all these parameters included in a single model and the objective of the model is to minimize the total costs. They developed four models in which the first two models are pure network flow problems and the latter two are network flow problems with side constraints. The former are solved with the simplex method and the latter are solved with the Lagrangian relaxation with subgradient methods. Yan & Yang considered a situation in which one aircraft is broken down in a single fleet type with non-stop flights. Yan & Lin improved the model by including station closures and Yan & Tu improved the model by including multiple fleets. Yan and Young added multi-stop flights to the method. Furthermore, these papers are identical only some constraints are added which influence the outcoming results.
Cao & Kanafani (1997)
Until now there are no models that are successful in combining cancellation and delaying in a real-time decision support tool. Cao & Kanafani are the first who introduced a model that combines cancellation and delaying while solving realistic problems. The model builds on the work done by Jarrah et al. The objective is to maximize the total profit by reducing the swapping and delaying costs. However, this model contains several errors as showed in Love and Sorenson (2001).
Argüello (1998)
The models of Jarrah et al. and Yan & Yang served as a foundation for the work of Argüello. Argüello developed a model that uses a greedy randomized adaptive search procedure (GRASP) to reconstruct aircraft routings in response to disruptions in the operation. A heuristic is presented that quickly generates cost-effective aircraft routings when one or several aircraft are grounded. The objective is to minimize the cost of reassigning aircraft to flights and the flight cancellation costs. The method is only made for a single fleet recovery problem and maintenance is not considered.
Thengvall et al (2000)
Thengvall extended the models of Yan & Yang and Yan & Tu by including penalties for deviations from the original schedule. The objective is to maximize passenger revenues and the model does not take crew and maintenance issues into account. In 2001 and 2003, Thengvall et al. continued on their work by including airport closures and multiple fleets.
Love et al (2001)
Love et al. developed a model that uses weights in the objective function. By changing the weights different characteristics in terms of the number of swaps, cancellations and delays are possible.
Crew recovery problem
GENERAL PART ABOUT CREW RECOVERY PROBLEM…………
Crew management is usually the bottleneck in the recovery process. Due to the complicated crew schedules, restrictive crew legalities and the hub-and-spoke network adapts by major airlines the recovery of crew is very complex. Crew management can be divided in two different parts: Crew pairing and crew assignment.
A crew pairing consists of a series of flight segments or flight legs. The crew starts from a home base of the carrier and ends there as well. A pairing consists usually out of three to four days. Several options are possible when a disruption occurs: deadheading, domicile, crew swaps, sit crews and reserve crews. The crew pairings must conform to the crew legalities.
A crew assignment ………………
This paragraph gives a chronological overview of the research done in the past about the aircraft recovery problem.
Wei (1997)
Wei et al. developed a system-wide multi-commodity integer network flow model for the crew recovery problem. A heuristic branch-and-bound search algorithm is used to solve the problem. The objective is to return to its original schedule as soon as possible, preferably in a cost-effective way. In the paper the assumption is that the flight schedule has been fixed and thus is given. The model is efficient enough for practical applications, since it was tested on various problems of realistic sizes.
Stojkovic (1998)
In 1998, Stojkovic developed a set partitioning problem for the crew recovery. A column generation method is implemented in a branch-and-bound search algorithm to solve the problem. The objective is to cover all flights from a given time period at minimal costs while minimizing the disturbance of crew members. The solution time is reasonable and good results of the objective function are achieved. The model uses a fixed flight schedule.
Nissen & Haase (2006)
Nissen & Haase developed a duty-period-based formulation of the crew recovery problem. The model is based on European airlines that use fixed crew salaries. A branch-and-price algorithm is used to solve the problem. The algorithm is a hybrid technique between column generation and branch-and-bound algorithm. The objective is to minimize the impact on the original schedule when new crew assignments are determined. The model is able to provide solutions in a short time period.
Medard (2007)
Beetje vreemd artikel….. Even kijken in andere verslagen hoe die beschreven staat. Ook weinig literatuur die in andere papers voorkomen…..
Gershkoff 1989 = classic crew scheduling!!!!!!!!
Kohl overview of crew rostering!!!!
Passenger recovery problem
GENERAL PART ABOUT PASSENGER RECOVERY PROBLEM…………
The most researchers mentioned before do not consider passengers in their objective. When a disruption occurs and the flights schedule has to be rescheduled passengers have to be rebooked as well, since they miss their connection. Passengers are not modelled explicitly and hence, passenger delay costs are only approximated. This paragraph gives a chronological overview of the papers that consider passengers in the objective.
Teodorovic & Guberinic (1984)
The objective of the model of Teodorovic & Guberinic is to minimize the passenger delays as mentioned before. However, they assume that passengers only fly a single flight leg. The model is a simplified version of reality and does not generate realistic solutions due to the less number of constraints.
Bratu & Barnhart (2006)
The first authors who developed a model which is more focused on passenger recovery is the model of Bratu & Barnhart. They describe two different models which are focussed on passenger recovery considering the rules of surrounding aircraft and crew. The objective is to find the optimal trade-off between airline operating costs and passenger delay costs. The first model is the so called DPM model (Disruption Passenger Metric), in which the sum of operating and disrupted passenger costs is minimized. This model uses an approximation of the passenger costs. The second model is the so called PDM (Passenger Delay Metric), in which the sum of passenger delay costs and operating costs are minimized. PDM cannot be solved in real-time, whereas DPM is fast enough to be used by operations controllers in real time. The models use the approximation of reserve crews to deal with the crew recovery.
Zhang and Hansen (2008)??
Bisaillon (2011)
Bissaillon developed a model that minimizes a weighted sum of three types of costs: operating costs, passenger inconvenience costs, and inconsistency costs. A large neighbourhood search heuristic alternates between construction, repair and improvement phases. It iteratively destroys and repairs parts of the solution. The first two phases produce an initial solution that satisfies the constraints. The third phase attempts to identify an improved solution. The solution method is depicted in Figure …….
The algorithm is fast enough to use in real time operations; however, the model assumes that all information is known at the beginning of the recovery period.
Jozefowiez (2013)
Jozefowiez developed a heuristic which is divided in three phases, called the new connection flight heuristic method (NCF). The first phase of the heuristic integrates the disruptions into the initial plan. The set of disruptions are considered sequentially and the phase is composed of the following steps: (i) cancelled flights; (ii) delayed flights; (iii) aircraft breakdown; (iv) airport capacity drops; (v) airport capacity overflow repair; (vi) connectivity repair. Each step has its own algorithm. The goal of the second phase is to reassign as many passenger groups as possible to the existing set of rotations to reach their destination. This is done with the shortest path method. The third phase tries to extend the aircraft rotations for the remaining passenger groups who are not accommodated in phase two. For a detailed description of all the algorithms and results is referred to Jozefowiez et al (2013). An advantage of the model is the computational time. In most cases the model finishes within 1 minute and the remaining runs do never pass the 4 minutes.
Sinclair (2014)
In 2014, Sinclair et al. presented a research which suggests improvements to the model developed by Bisaillon. They introduce a number of refinements in each phase in order to improve the outcomes. The improvements are depicted in Figure …..
The model shows that the modifications significantly improve the solution costs. In 17 out of 22 instances the model obtains the best known solution within five minutes of computing time and in 21 out of 22 instances it obtains a solution within 10 minutes of computing time. In case of high numbers of cancelled itineraries it showed to be advantageous to run the algorithm for longer than 10 minutes.