Scribd
Upload
58 views31 pages

Column Generation Methods For Disrupted Airline Schedules: Niklaus Eggenberg Dr. Matteo Salani and Prof. Michel Bierlaire

This document presents a column generation approach to solve the airline recovery problem (ARP) following disruptions. The ARP aims to determine a new flight schedule and recovery costs given the states of airplanes, initial schedule, maintenance requirements, and cancellation costs. The approach models the problem as a master problem solved using column generation, where the pricing problem of finding new columns is solved using recovery networks and resource constrained shortest path algorithms. The method was implemented and tested on real airline data instances involving up to 250 flights. Considering maintenance planning was found to be crucial for obtaining optimal solutions.

Uploaded by

Tarek Maamari
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
58 views31 pages

Column Generation Methods For Disrupted Airline Schedules: Niklaus Eggenberg Dr. Matteo Salani and Prof. Michel Bierlaire

This document presents a column generation approach to solve the airline recovery problem (ARP) following disruptions. The ARP aims to determine a new flight schedule and recovery costs given the states of airplanes, initial schedule, maintenance requirements, and cancellation costs. The approach models the problem as a master problem solved using column generation, where the pricing problem of finding new columns is solved using recovery networks and resource constrained shortest path algorithms. The method was implemented and tested on real airline data instances involving up to 250 flights. Considering maintenance planning was found to be crucial for obtaining optimal solutions.

Uploaded by

Tarek Maamari
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 31

1

Column Generation Methods for


Disrupted Airline Schedules

Niklaus Eggenberg
Dr. Matteo Salani and Prof. Michel Bierlaire

In collaboration with APM Technologies


Funded by CTI Switzerland
Index
2

Index
 Airline Scheduling
 The Airplane Recovery Problem (ARP)
 The Column Generation (CG) approach
 Solving the pricing problem with Recovery Networks
 Implementation and results
 Future work and conclusions
Airline Scheduling
3

Airline Scheduling Approach


1. Route Choice
2. Fleet Assignment
Technical Schedule
3. Tail Assignment
4. Crew Pairing
5. Crew Roistering
6. Passenger Routing (catering)
Airline Scheduling (2)
4

Maintenances
Maintenances are forced by RESOURCE consumption (eg. flown hours)

Resources are renewed during maintenance


Disruption and Recovery
5

Disrupted Schedule and Recovery


Initial Schedule Disrupted
Disruption

Back to normal
Recovery
Decisions
time
Survey: Kohl (2004)
The ARP
6

The Airplane Recovery Problem (ARP)


Input Output
• Planes’ States
• T
• Initial Schedule
• New schedule up to T
• Maintenances
• Recovery cost
• Cancelation Costs

• Delay Cost
The ARP (2)
7

Definitions:
PLANES:
Initial State : position, initial time, initial resource consumption

Final State: position, expected time, expected resource consumption


Feasible Flight Set: coverable flights

Feasible Final State Set: coverable final states

AIRPORTS:
Activity Slots: periods when take-off/landings are permitted

Maintenance Slots: periods when given plane type can perform maintenance
The ARP (3)
8

Definitions (2):
Flights:
Origin and Destination
Scheduled Departure Time (SDT)
Flight Duration
Flight Cost
Cancelation Cost
The ARP (4)
9

Determine a Final State:


GVA AMS BCN MIL BUD
The ARP (4)
10

Determine a Final State:


GVA AMS BCN MIL BUD

T
The ARP (4)
11

Determine a Final State:


GVA AMS BCN MIL BUD

T
The ARP (5)
12

Solution to the ARP:


A recovery scheme for each plane:

Initial State
Flights and Maintenances

Expected Final
State
Solving the ARP
13

Multi-objective optimization:
Minimize both T and recovery costs

Strategy: for fixed T find optimal recovery plan

Give several recovery plans for different values of T (decision aid)


Solving the ARP (2)
14

Column Generation Approach


Find out optimal solution by combining individual recovery schemes r R’
(master problem) on a subset R’ R of all feasible recovery schemes

Generate potentially improving recovery schemes r R-R’ dynamically for


each plane (pricing problem)
CG - Master Problem
15

Master Problem: MIP formulation


CG - Master Problem (2)
16

What is a column ?
• cost cr
• vector

Where
 = 1 if flight f is covered by column r
 = 1 if final state s is covered by r
 = 1 if column r is affected to plane p
CG – Pricing Problem
17

The Pricing Problem


Find new columns minimizing the reduced cost :
CG – Pricing Problem (2)
18

Recovery Networks (Argüello et al. 97)


1. Generate a recovery network for each plane
2. Update arc costs according to dual variables
3. Solve Resource Constrained Elementary Shortest Path
(RCESPP)
4. Add Columns to R’
5. Resolve restricted LP until optimality and branch
RN– Nodes and Arcs
19

Time – Space Network with


• source node n0 = [t, m, r]
• node n = [t, m, r]
• sink s = [t,m,r]

• flight arc [n, n’]


• maintenance arc [n, n’]
• termination arc [n,s]
• maintenance termination arc [n,s]
RN – Example
20

Recovery Network
GVA AMS BCN MIL BUD
CG – Solve Pricing with RCESPP
21

Updating arc costs


 flight arcs:

 maintenance arcs:

 termination arcs:

 maintenance term. arcs :

Solve RCESPP on networks returns column minimizing the reduced cost!


Righini & Salani (2006), which is an extension of Desrochers et al. (1988)
References
22

Some References
• Argüello et al. (1997): recovery without maintenance
up to 27 planes, 162 flights, 30 airports

• Desrosiers et al. (1997): daily scheduling NOT recovery


up to 91 planes, 383 flights, 33 airports; max delay of 30 minutes

• Clarke (1997): maintenances requirements but no decision on them


up to 177 planes, 612 flights, 37 airports; only 0 or 30 min delay

• Kohl et al. (2004): Descartes project, good survey of state of the art
no instance size mentioned for DAR

• Barnhart and Bratu (2006): passenger oriented recovery algorithm


up to 302 planes, 1032 flights, 74 airports
Implementation Issues
23

Implementation issues
 Implemented in C++ with COIN-OR BCP framework

 Used interior point methods to solve the LP

 Used linear time and logarithmical resource discretisation


 2 phase pricing:
• generation (keep also non optimal columns, heuristic pricing)
• proving optimality (optimal column only, exact pricing)
Implementation Issues (2)
24

Linear Time Discretization


1 2 3 4 5 6

Logarithmic Resource Discretization


1 2 3 4
Solved Instances
25

Real Instances
• Got real schedules from Thomas Cook Airlines (APM’s main customer)
• Solved original schedules up to 250 flights (algorithm validation)
• Generated disruption scenarios
 delayed planes (initial states)
 grounded planes (initial states)
 airport closures (activity slots)
 forced maintenances (initial resource consumption)
Solved Instances (2): Problem Sizes
26
Solved Instances (3): Added value of maintenances
27

Average results of 10 randomly generated instances

Considering maintenances is
crucial!!!
Solved Instances (4): Pareto Optimality
28

Pareto behavior for increasing T


Future Work
29

Future Work
• Benchmark solutions against practitioners

• Allow repositioning flights and early departures

• Extend Pricing Solver for acceleration

• Include in APM solutions


Conclusions
30

Conclusions
• Developed a flexible and fast algorithm

• Solutions are very promising

• Maintenance planning is an added value


31

THANKS for your attention!

Any Questions?

You might also like