Scheduling and Revenue Management
Upcoming SlideShare
Loading in...5
×
 

Scheduling and Revenue Management

on

  • 881 views

Nabil Si Hammou, April 2012

Nabil Si Hammou, April 2012

Statistics

Views

Total Views
881
Views on SlideShare
875
Embed Views
6

Actions

Likes
1
Downloads
40
Comments
0

1 Embed 6

http://www.linkedin.com 6

Accessibility

Categories

Upload Details

Uploaded via as Microsoft PowerPoint

Usage Rights

© All Rights Reserved

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Processing…
Post Comment
Edit your comment

Scheduling and Revenue Management Scheduling and Revenue Management Presentation Transcript

  • Guideline on the use of Operations Research in the airline industry Nabil Si Hammou, Operations Research Analyst n.sihammou@gmail.comScheduling and Revenue Management April 2012
  • Abstract  As a member of AGIFORS (The Airline Group of the International Federation of Operational Research Societies ) and passionate on the operations research, I have established a summary of practices on the use of optimization methods for Scheduling and Revenue Management in the airline industry.  This summary comes as a result of 6 months of individual research on the optimization methods used by different airlines for Scheduling and Revenue Management. It’s based on various information sources (Air France seminar, AGIFORS symposium, AGIFORS presentations, specialized books in the airline industry, ….).  I would welcome the opportunity to discuss with you the potential for making a significant contribution in optimizing the scheduling and revenue management process. Feel free to call me at 00.212.6.18.98.38.61 or email me at n.sihammou@gmail.com.  Being an Operations Research Analyst, I am particularly interested in the positions: – Scheduling optimization specialist Nabil Si Hammou – Revenue management optimization specialist Optimization Specialist n.sihammou@gmail.com 00.212.6.18.98.38.61Nabil Si Hammou, April 2012 Best practices - Optimization process 2
  • Plan Outline ( page 3..6) Optimization Process Overview ( page 7..10) Scheduling (page 11..49) Revenue Management (Page 50..83) Conclusion : Robustness (page 84)Nabil Si Hammou, April 2012 Best practices - Optimization process 3
  • OutlineContext  The global airline industry consists of over 2000 airlines operating and more than 23 000 commercial aircraft, providing service to over 3700 airports . The world’s airlines flew more than 29 million scheduled flights and transported over 2.5 billion passengers (IATA, 2010).  Since the economic deregulation of airlines, cost management and productivity improvements has became central goals of airlines with the shift to market competition.  The airline schedule affects almost every operational decision, and on average 75% of the overall costs of an airline are directly related to the schedule. Given an airline schedule, a significant portion of costs and revenues is fixed  The management strategies and practices of airlines were fundamentally changed by increased competition within the industry.Nabil Si Hammou, April 2012 Best practices - Optimization process 4
  • OutlineContext  The main principle of airline management is to match supply and demand for its service in a way which is both efficient and profitable.  Airlines use numerous resources to provide transportation services for their passengers. It’s the planning and efficient management of these resources and sales that determine the survival or demise of an airline.  In practice, the objective of airline management is to maximize operating profit (increase sales and/or decrease costs) by defining the optimal resource scheduling and sale policy: Sales Investment Operations cost cost BenefitNabil Si Hammou, April 2012 Best practices - Optimization process 5
  • OutlineAirline management system  To maximize the operating profit, the airline management system takes into account various factors such as demands in various markets, available resources, airport facilities and regulation for achieving optimal solutions Airport operating Airport runway Airport charges Other regulations hours length Maintenance requirement Airport Facility constraints Passenger behavior Aircraft capacities connection time Airline Aircraft range Aircraft Demand limitation Decision Competitor Aircraft costs System schedules Passenger demand Operational costs Passenger Yield Minimum turn time Route Crew Managerial characteristic availability constraintNabil Si Hammou, April 2012 Best practices - Optimization process 6
  • Optimization processOptimization process  Currently, all airlines decompose the overall management problem into subproblems and solve them sequentially: sequential approach  Because of the reduced complexity generated by the decomposition, the sequential approach allows to solve decision problem more easily by using optimization algorithms.Nabil Si Hammou, April 2012 Best practices - Optimization process 7
  • Optimization processDecomposition  The decomposition is usually structured according on two dimensions: 1.Time horizon ( Strategic, Tactical and Operations) 2. Subject ( Aircraft, Crew, Ground and Sales)  Various decomposition used in the airline industry. Example of an optimization process used by one of the biggest airlines in Europe Best practices - Optimization process 8
  • Optimization processDecomposition  The subproblems which make up the overall airline decision system could be solved sequentially according to the below design.  In some cases, the sequence of these decisions is reversed, in that the identification of a profitable opportunity related to a subproblem might modify the decision related to the previous subproblem ( iterating system). 9
  • Optimization processScope  We focus in this presentation on the following subproblems : A. Scheduling: B. Revenue Management: 1. Fleet assignment 3. Crew pairing 5. Optimization 2. Maintenance routing 4. Crew assignment 6. ForecastingNabil Si Hammou, April 2012 Best practices - Optimization process 10
  • SchedulingNabil Si Hammou, April 2012 Best practices - Optimization process 11
  • Scheduling Fleet assignmentNabil Si Hammou, April 2012 Best practices - Optimization process 12
  • SchedulingFleet assignment: Introduction  Given the fleet availability and flight schedule, the goal of fleet assignment is to find the best assignment of fleet type to flight legs that maximize the expected profit. 06h00 10h30 Airport A Which Which aircraft type ? aircraft type ? 07h30 08h30 09h00 10h100 Airport B Which Which aircraft type ? aircraft type ? Input Output 1.Schedule: set of flight legs with given departure and Assignment of fleet type to each flight leg of arrival times. the schedule with profit maximization 2. Fleet: aircraft owned by the company (number of aircraft (expected revenue – operation cost) or cost by type). minimization including spill cost 3.Profit : associated to the assignment of a fleet type to flight leg calculated throughout: – Cost: fuel…. – Revenue: usually substituted by (-) spill cost (rejected demand) 13 Best practices - Optimization process
  • SchedulingFleet assignment: Introduction Constraint Coverage: each flight leg is assigned to exactly one fleet type. Fleet availability : it limits the assigned aircraft of each fleet type to the number available. Balance: the total numbers of aircraft of each type arriving and departing each airport are equal. Additional restriction: technical restriction ( some aircrafts can’t cover some flight legs…), ….Nabil Si Hammou, April 2012 Best practices - Optimization process 14
  • SchedulingFleet assignment: Time-space network  For modeling the fleet assignment problem, we represent at first the flight schedule as time space network in order to facilitate the mathematical modeling of constraints. Time-space network Airport C Airport B Airport A Schedule cycle time (week, day..) : Flight arc: represents a flight leg with departure and arrival location : Arc’s origin node: represents a flight leg departure time : Arc’s destination node: represents a flight leg arrival time including turn time. : Ground arc: represents aircraft on the ground during the period spanned by the times associated with the arc’s end nodes : Count time : a point in time used specifically to count the number of aircraft needed to cover the aircraft rotations in a solutionNabil Si Hammou, April 2012 Best practices - Optimization process 15
  • SchedulingFleet assignment: Modeling Input Decision variables F : set of flight legs to be operated fik :1 if flight leg i is assigned to fleet type k, K: set of fleet types 0 otherwise. Mk number of aircraft available of type k. yak : number of aircraft of type k on the Lk: the set of flight legs could be covered by the fleet type k. ground arc a Nk : set of nodes (departure , arrival) could be served by the fleet type k Gk : set of ground arc could be covered by the fleet type k. O(k,n): set of flight legs Lk and originating by the node n I(k,n): the set of legs Lk and terminating at the node n N+: set of ground arc originating from node n Nk ( n- ground arc terminating at n Nk) CL(k) : the set of flight legs Nk and cross the count time. CG(k): the set of ground arc Gk and cross the count time Cik operating cost minus revenue of flying leg f with fleet type kNabil Si Hammou, April 2012 Best practices - Optimization process 16
  • SchedulingFleet assignment: Modeling Model Minimizing costs ( operation & spill) min Cik f i k k Ki F subject to Coverage constraint fi k 1 i F; k K Balance constraint k yn fi k k yn fi k k K n Nk; i O ( k ,n ) i I ( k ,n ) k Fleet availability constraint ya fi k Mk k K; a CG ( k ) i CL ( k ) fi k 0;1 k K i Nk Variable definition k ya 0 k K a Gk* Additional restriction constraints are expressed throughout parameter definitionNabil Si Hammou, April 2012 Best practices - Optimization process 17
  • SchedulingFleet assignment: Solving methods Solving Methods Exact Methods Approximate Methods Column Brunch and Meta-heuristic Generation & Specific Bound ( genetic Brunch and heuristic algorithm…) Bound Solution time Absolute optimum Implementing time flexibility 18
  • SchedulingFleet assignment: Solving methods  Airline companies and solution vendors use all methods presented in the previous diagram. However , exact methods tends to dominate the use of solving methods for the fleet assignment.  There is no rule that confirm that airline can get ( or not) a solution by using branch and bound in reasonable time given the size of the model. However, based on results of some airlines , we may guess that in case of 2.000 of flight legs and 10 fleet type, the use of branch and bound method is sufficient to solve the fleet assignment problem in reasonable time.  Besides, the biggest airlines use column generation method combined with branch and bound methods to solve the fleet assignment problem although the size problem complexity.Nabil Si Hammou, April 2012 Best practices - Optimization process 19
  • SchedulingFleet assignment: IT Development  Because of the size problem complexity, the program is usually developed with C++. The branch and bound method is already available as library provided by commercial solver software ( Cplex, Xpress,...) and other open source(GLPK).  The program is mainly made up of three parts : loading data, optimization algorithm, and report the fleet assignment. 1 2 3 Loading data Optimization Algorithm Report results Initialization Creating a Reduced Master Master model Problem RMP Fleet availability Call solver library for solving RMP Solver (brunch and bound method) Display the fleet Get the optimal assignment Flight schedule solution of RMP Introduction to No the best new C MP <=0 column Restriction Column generation diagram Optima solution found 20
  • SchedulingFleet assignment: Impact  Fleet assignment optimization, which has been applied widely in practice, is attributed with generating solutions that lead to significant improvements in operating profit: - USAir indicates annual savings of $15 million attributable to the use of a fleet assignment optimizer. - Fleet Assignment solution at American Airlines have led to a 1.4% improvement in operating margins.Nabil Si Hammou, April 2012 Best practices - Optimization process 21
  • SchedulingFleet assignment: Improvements / Future  Some airlines add other constraints to the fleet assignment model such as time window that assumes departure time are not fixed and there is time window during which flight may depart.  Other companies integrate further parameters such as passenger spill decision in order to better estimate the spill costs ( Extended Fleet Assignment Problems)  In these above cases, the column generation method will be more useful to solve the fleet assignment problemNabil Si Hammou, April 2012 Best practices - Optimization process 22
  • Scheduling Maintenance routingNabil Si Hammou, April 2012 Best practices - Optimization process 23
  • SchedulingMaintenance routing: Introduction  Given the fleet assignment solution, the objective of maintenance routing is to identify the sequence of flight legs to be covered by the same aircraft within each fleet that satisfy operational and physical constraint.  The sequence of flight legs has to ensure that the aircraft is able to receive the required maintenance checks at the right time and at the right base. Maintenance Maintenan Airport base ce base Airport Airport 4 9 10 Hub1 Airport 6 Airport 11 Hub3 Airport Airport Hub2 Maintenance Airport 5 7 8 base  4 types of aircraft maintenance are required. The most frequent check is required every 30 hours ( 2- 3 days). This check can be performed overnight or during downtime during the flight day.Nabil Si Hammou, April 2012 Best practices - Optimization process 24
  • SchedulingMaintenance routing: Introduction Input Flight schedule with fleet assignment: set of flight legs with given departure and arrival times and fleet type assigned. 1 Routing generation 2 Routing evaluation 3 Solving optimization model OutputFor each fleet type, the best aircraft rotations that allows the aircrafts to undergoperiodic maintenance checks and satisfy other physical and operational constraints.Nabil Si Hammou, April 2012 Best practices - Optimization process 25
  • SchedulingMaintenance routing: Introduction Constraints 1.Flight coverage: each flight leg must be covered by only one aircraft. 2.Fleet availability: number of aircraft by fleet type must not exceed the number available 3.Feasible routing: The routing must incorporate the turn-around time. turn- around time is the minimum time needed for an aircraft from the time it lands until it is ready to depart again 4.Regular return (overnight) to the maintenance station has to be insured for each routing in order to provide the maintenance opportunity at least once in 3 days. 5.Optional constraints: 1.favor closed cycle: when an aircraft starts from a city, and at the end of the routing cycle, ends up at that same city to start another cycle. 2.Favor succession of flights with the same custom status ( Schengen to Schengen ..)Nabil Si Hammou, April 2012 Best practices - Optimization process 26
  • SchedulingMaintenance routing (1): Routing generation  At first, airlines should define its routing cycle. Many airlines set the routing cycle to 2 or 3 days.  We begin by generating all possible valid aircraft routings that satisfy physical and operational constraints routing: – The routing must incorporate the turn-around time. turn-around time is the minimum time needed for an aircraft from the time it lands until it is ready to depart again. – the routing must include at least one overnight stay at maintenance base in order to provide the first type of maintenance check. Overnight Day 1 day 1 Day 2 Overnight day2 05h00 13h30 15h05 16h05 17h10 18h10 6h20 7h20 14h25 15h25 17h00 21h30Routing 1 LAX JFK JFK ORD ORD JFK JFK JFK IAD IAD JFK JFK LAX LAX 06h15 07h45 09h00 12h00 13h10 15h40 09h10 12h00 13h10 15h40 17h00 18h30Routing 2 JFK BOS JFK JFK ATL ATL JFK JFK ATL ATL JFK JFK BOS BOS Nabil Si Hammou, April 2012 Best practices - Optimization process 27
  • SchedulingMaintenance routing (1): Routing generation  Automated systems are used extensively to generate and filter all these routes for the airlines in a relatively short time.  An overview of a methodology has been implanted for generating the rotations: 1 Creating all one day routing 2 Building routing by attaching one day routing 3 Examination of constraint satisfaction 4 Establishing a list of potential routing candidate  This generation could be enhanced by using constraint programming techniquesNabil Si Hammou, April 2012 Best practices - Optimization process 28
  • SchedulingMaintenance routing (2): Routing evaluation  The ultimate goal of the maintenance routing is to select the best flight legs sequences that contribute in the maximization of the airline profit.  In practice, airlines evaluate routings by various ways according to the structure adopted for the objective function of maintenance routing model : Objective function Maximizing Minimizing pseudo- Maximizing through maintenance cost values opportunitiesNabil Si Hammou, April 2012 Best practices - Optimization process 29
  • SchedulingMaintenance routing (3): Optimization model  After generating feasible routings that satisfy maintenance requirement, we should select from this list the optimal routings that satisfy the coverage flight constraint and the fleet availability limit.  Optional constraint are usually taken into account in the objective function in order to penalize some routings and/or favorite others.  The decision problem consists to chose routings from the long list of routing built that : - Satisfy constraints of coverage flight and fleet availability - Minimize cost (or Maximizing through values ..)Nabil Si Hammou, April 2012 Best practices - Optimization process 30
  • SchedulingMaintenance routing (3): Optimization model Input Decision variablesR: set of feasible routings 1. Xr :1 if routing r is chosen. 0 otherwiseL: set of flight legsN: number of aircrafts ( associated to the fleet type that is subject of the maintenance routing)Cr: cost of routing r&l,j: 1 if leg l is in routing r, 0 otherwiseNabil Si Hammou, April 2012 Best practices - Optimization process 31
  • SchedulingMaintenance routing (3): Optimization model Model Minimizing costs min Cr * X r subject to Coverage constraint l ,r * Xr 1 l L r R Fleet availability constraint Xr N r R Variables definition Xr 0,1 r R* Maintenance requirement and feasibility routing constraint are satisfied by routing construction Nabil Si Hammou, April 2012 Best practices - Optimization process 32
  • SchedulingMaintenance routing (3): Optimization model Solving Methods Exact Methods Approximate Methods Column Branch and Meta-heuristic Generation & Specific Bound ( genetic Branch and heuristic algorithm…) Bound  The backbone of comparison analysis regarding exact and approximate method for the fleet assignment remains useful for the maintenance routing .  However, some airlines have expressed that the use of column generation for routing maintenance remains still a challenge because of non convergence issue.  Other airlines have implemented other approximate methods for solving the maintenance routing (formulated as asymmetric traveling salesman problem with side constraints ) by using Lagrangian relaxation and heuristicsNabil Si Hammou, April 2012 Best practices - Optimization process 33
  • SchedulingMaintenance routing (3): Optimization model  The maintenance routing problem as presented, is based on the flight schedule and the fleet availability. In reality , the flight schedule could be changed at the last minute because of disruptions.  The robustness of the maintenance routing solution becomes an essential criteria in order to keep the scheduling process feasible.  In addition to profit maximization, airlines could take into account robustness criteria (proxy) in different ways to define the best routingsNabil Si Hammou, April 2012 Best practices - Optimization process 34
  • Scheduling Crew scheduling: a. Crew pairing b. Crew assignmentNabil Si Hammou, April 2012 Best practices - Optimization process 35
  • SchedulingCrew scheduling: Introduction  After the flight schedule is developed and fleet are assigned to cover all the flight legs in the schedule, crew work schedules are started with the help of optimization techniques.  Crew scheduling involves the process of identifying sequences of flight legs and assigning both the cockpit ) and cabin crews to these sequences. Time Cockpit crews: charged with flying the aircraft Cabin crews: responsible for in-flight passenger safety and service. 36
  • SchedulingCrew scheduling: Introduction Cockpit Authorized for One fleet type The crew scheduling problem is solved VS separately for the Cabin cockpit crew and Able to work on Different cabin crew fleet type Cockpit Cockpit crew size depends on fleet type Scheduling trends to be Individual for VS cabin crew and per Cabin Number of team for cockpit Cabin crew size crew passengers depends on on board Best practices - Optimization process 37
  • SchedulingCrew scheduling: Introduction  Because of the complex structure of work-rules and crew costs, the crew scheduling problem is typically solved in a two-step process: Crew Generation of mini-schedules, called pairings Pairing typically spanning 1–5 days Assembling pairings into longer crew schedules Crew typically spanning about 30 days and assign it to Assignment crew members Crew pairing: the objective is to minimize the crew costs associated with covering all flight legs in the flight schedule, Crew assignment: The objective is mainly to assemble pairings into schedules that maximize the satisfaction levels of crews.Nabil Si Hammou, April 2012 Best practices - Optimization process 38
  • SchedulingCrew pairing: Introduction  A crew pairing is composed of a sequence of flight legs, with the flight legs comprising a set of daily work activities, called duty, separated by overnight rest periods.  The sequence of flight legs starts and ends at the same crew base(city in which the crew actually lives). The sequence may typically span from 1 to 5 days.  The objective of crew pairing is to find a set of pairings that covers all flights which: - satisfies various constraints such as union, government, and contractual regulations. - minimizes the total crew cost.Nabil Si Hammou, April 2012 Best practices - Optimization process 39
  • SchedulingCrew pairing: Constraints Constraints Feasibility others C.1 Flights in a pairing must be sequential in time and space; C.7 Flight covering C.2 The elapsed time between the arrival of a flight leg and the departure C.8 Fleet restriction of the subsequent flight leg in the pairing is bounded by a maximums and a minimums threshold: for cockpit crew a-connection time b-rest time C.3 Each duty should not exceed a maximum hours of flight time. C.4 The maximum number of hours worked in a day. C.5 The maximum time the crew may be away from their home base C.6 Pairings starts and ends at crew base Overnight C2.a Rest 9h30 12h00 13h10 15h40 16h10 19h10 9h10 12h10 12h30 14h00 15h00 16hh30 JFK ATL ATL JFK JFK MIA C2.b MIA JFK JFK BOS BOS JFKC6 C6 C1 Sign In : Sign out : C3 08h00 Duty Period 1 19h25 Sign In : Sign out : 08h10 Duty Period 2 16h40 C4 C5.Time Away From Base 40
  • SchedulingCrew pairing: Costs  The crew costs structure can vary widely by airline, with significant differences existing between airlines in different countries or regions. Example of a pairing cost structure in Europe Pairing cost Maximum of Minimum guaranteed Time away from base Sum of duty cost pairing pay cost Duty cost= Max of Total flying time cost Total duty time cost Minimum guaranteed per dayNabil Si Hammou, April 2012 41
  • SchedulingCrew pairing: Optimization model All possible feasible pairings are generated based on rules and regulations. Pairing generation Generators are normally equipped with filters to identify and select good potential pairings Pairing Select the best pairings that cover all the flight optimization and minimize the total crew costsNabil Si Hammou, April 2012 Best practices - Optimization process 42
  • SchedulingCrew pairing: Optimization model Input Decision variables F = Set of flights 1. Xp :1 if pairing p is chosen. 0 otherwise P = set of feasible pairings K = set of crew home-base cities al,j: 1 if flight i is covered by pairing j, 0 otherwise cj: crew cost in pairing j* For the cockpit crew pairing, the problem is solved by fleet family ( driving license)Nabil Si Hammou, April 2012 Best practices - Optimization process 43
  • SchedulingCrew pairing: Solving methods Solving Methods Exact Methods Approximate Methods Column Branch and Meta-heuristic Generation & Specific Bound ( genetic Branch and heuristic algorithm…) Bound  The comparison analysis regarding exact and approximate method for the fleet assignment remains useful for the maintenance routing .  The use of column generation combined with branch and bound algorithm is highly recommended for solving the problem exactly. The pricing problem included in the column generation procedure could be treated as a shortest path problem. In this case , a column is equivalent to a pairing  Other airlines have implemented approximate methods for solving the crew pairing problem by using mainly genetic algorithm. 44
  • SchedulingCrew assignment: Introduction  Once the crew pairing problem is solved, the second phase is crew assignment. It’s the process of assembling the pairings into longer schedule (usually on a monthly basis) and assigning individual crew members to this schedule.  The schedule assigned take into account vacation time, training and rest.  The crew assignment problem is usually solved by using either bidline or rostering approach: Or Bidline Rostering 1.Generic schedules are built from pairing. 1.Specific schedules are constructed trying to satisfy certain crew bids with priority based on 2.Crew members bid on theses schedules seniority. 3.Assignment based on seniorityNabil Si Hammou, April 2012 Best practices - Optimization process 45
  • SchedulingCrew assignment: Rostering model Input Decision variables P :set of dated pairings Xs,k: 1 if the schedule s is chosen for employee k, K : set of crew members of given type 0 otherwise F : set of flights Sk:set of schedules for employee k in K Np: number of selected schedules that must contain p Cs,k : cost of schedule s if it’s assigned to employee k ( represent the choices and the priority) ap,s : 1 if pairing p is in the schedule s,0 otherwise* For the cockpit crew rostering, the problem is solved by fleet family ( driven license) and for each crew type separatelyNabil Si Hammou, April 2012 Best practices - Optimization process 46
  • SchedulingCrew assignment: Solving methods Solving Methods Exact Methods Approximate Methods Column Meta-heuristic Branch and Generation & Specific ( genetic Bound Branch and heuristic algorithm…) Bound  Basically, the approach used for solving crew pairing could be used for crew assignment. However many airlines still use heuristics to optimize the crew assignment.Nabil Si Hammou, April 2012 Best practices - Optimization process 47
  • SchedulingCrew scheduling: Impact  For large airlines, the improvement in solution quality related to crew scheduling (pairing & assignment), translates to savings on the order of $50 million annually.  Beyond the economic benefits, crew scheduling optimization tools can be used in contract negotiations to quantify the effects of proposed changes in work rules and compensation plans.Nabil Si Hammou, April 2012 Best practices - Optimization process 48
  • SchedulingScheduling: challenges & opportunities Integrated schedule Maintenance Schedule design Fleet assignment Fleet assignment routing Fleet assignment Crew pairing Maintenance Crew pairing routing Crew pairing Crew assignment Schedule Fleet Maintenance design assignment routing Fleet Maintenance Crew pairing assignment routingNabil Si Hammou, April 2012 Best practices - Optimization process 49
  • Revenue ManagementNabil Si Hammou, April 2012 Best practices - Optimization process 50
  • Revenue ManagementPlan Outline Optimization Network revenue Fare class mix management Demand forecasting ImplementationNabil Si Hammou, April 2012 Best practices - Optimization process 51
  • Revenue ManagementOutline  For maximizing the income revenue given the scheduled flight and capacities, the airline should sell the right seats to the right customers at the right prices and at the right time  The revenue maximization process is mainly made up of two components: - Pricing ( or differential pricing) - Revenue Management ( or Yield Management) Pricing Revenue Management Customer Product Price Capacity allocation segmentation design decision  For most airlines, revenue management is the primarily tactical decision in the revenue maximization process. However, for low-costs, pricing tends to be the primarily tactical decisionNabil Si Hammou, April 2012 Best practices - Optimization process 52
  • Revenue ManagementOutline: Pricing  The airline offer various product called “fare product or fare class” for each future flight departure. The traditional fare product structure is mainly defined by following restrictions : Advance Number of days required The option of refundability (or between booking and flight not ) purchase departure (7,14,21…) Fare Non- refundability Change fee product The requirement to stay at Saturday night Penalties of changes in itinerary Saturday night after purchase  Service amenities could been added into others characteristics for each product.  For each product, the airline associates a price allowing to : - attract the right costumer by the right product. - maximize the wiliness to pay for each consumerNabil Si Hammou, April 2012 Best practices - Optimization process 53
  • Revenue ManagementOutline: Revenue Management  Given the fare classes and the price associated to each fare class, the revenue management is the subsequent process of determining how many seats to make available at each fare level for maximizing the revenue  Revenue management system is mainly made up of two components (1)Optimization and (2)Demand forecasting.Nabil Si Hammou, April 2012 Best practices - Optimization process 54
  • Revenue ManagementOptimization  The correct RM strategy is to manage the seat inventory of each flight departure to maximize total flight revenues generated by all the network.  In practice the airlines attempt to achieve this goal by implementing either of these approaches: Fare Class mix Network Revenue Management Maximization of the revenue Maximization of the revenue generated by each single flight Vs generated by the network Max Revenue i Max RevenueO-D i: single flight O-D: itinerary origin destinationNabil Si Hammou, April 2012 55
  • Revenue ManagementOptimization  Because of its relative simplicity, the fare class mix is the most common approach used in the airline industry.  Some biggest airlines have recently implemented the network revenue management in order to increase the revenue by taking into account the interdependence between flights. Fare Class mix Network Revenue ManagementInterdependence of flights Absolute optimum Implementing timeNabil Si Hammou, April 2012 Best practices - Optimization process 56
  • Revenue ManagementOptimization: Fare class mix Definition  Fare class mix (called also leg-based Revenue Management) consists to allocate optimally the capacity of each single flight leg to different fare classes.Nabil Si Hammou, April 2012 Best practices - Optimization process 57
  • Revenue ManagementOptimization: Fare class mix Control types  The capacity allocation control could be implemented within the reservation system under one of these decision forms : Booking limits Bid price Partitioned Nested Remained flight capacity  Booking limits are controls that limit the  Bid-price control sets a threshold amount of capacity that can be sold to any price such that a request is particular class at a given point in time. accepted if and only if its revenue exceeds the threshold priceNabil Si Hammou, April 2012 58
  • Revenue ManagementOptimization: Fare class mix Modeling: Input Output Deterministic Random Optimal policy of selling the flight seats at eachJ :set of fare class time given the remaining flight capacity ( best Dj,t : demand of farePi : price associated to fare allocation of flight capacity on fare classes) class j at period t<=T class I (Pi > Pi+1)C : flight capacityT : flight date Assumptions Or Static Model Dynamic Model (Non overlapping demand) (Overlapping Non overlapping) 59
  • Revenue ManagementOptimization: Fare class mix Static model:  The static model is mainly based on the assumption of Non overlapping demand : - demand for the n classes arrives in n stages, one for each class, with classes arriving in increasing order of their revenue values. Non overlapping demand Static model Input Decision policy (Control policy) Deterministic Random U(j,x): Quantity of demand to accept given remaining flight capacity. xJ :set of fare class Dj: demand of farePi : price associated to class j Or fare class i (Pi > Pi+1) Booking limit controls Bid price controlsC : flight capacity limitj (x) : maximum Bid Price (x,j): price number of demand of threshold for accepting fare class j..1 to accept the demand during the given remaining capacity stage j given the at the start of stage j remaining capacity x 60Nabil Si Hammou, April 2012
  • Revenue ManagementOptimization: Fare class mix Static model: method solving  The optimal policy related to the revenue management model could be found by using either dynamic programming or heuristics. Solving Methods Exact Methods Approximate Methods Dynamic Heuristics Programming ( EMSR…) Solving time Absolute optimum Implementing timeNabil Si Hammou, April 2012 Best practices - Optimization process 61
  • Revenue ManagementOptimization: Fare class mix Static model: method solving Dynamic Programming (EMSR Expected marginal seat revenue…) Model Model EMSR-a : version a EMSR-b : version b j k2 Yk k1 k2 1 S j Dk k2 Pk 2 Pk1 * P rob(D 1 k Yk ) k 1 1 j and pk * E[ Dk ] k2 Pk 2 Pk1 * P rob(D 1 k Yk 1 1) p* j k 1 j j E[ Dk ] Yj Ykj 1 k 1 k 1 Pj 1 p* * P rob(Sj j Y jj 1 ) Pj 1 p* * P rob(Sj j Y jj 1 1) Optimal policy Optimal policy Optimal policy Booking limitj (x) Bid Price (x,j): (x,j) Bid Price Booking limitj (x) Booking limitj (x)  Even though the higher solution quality provided by the dynamic programming and its simplicity, many airlines still use approximate methods : EMSRNabil Si Hammou, April 2012 Best practices - Optimization process 62
  • Revenue ManagementOptimization: Fare class mix Dynamic model  Unlike static model, dynamic model allows for an arbitrary order of arrival with the possibility of interspersed arrivals of several classes. (overlapping demand). Overlapping demand  In addition to other assumptions retained by the static model, the dynamic model requires assumption markovien arrivals Dynamic model Dynamic ProgrammingNabil Si Hammou, April 2012 Best practices - Optimization process 63
  • Revenue ManagementOptimization: Fare class mix Static model Vs Dynamic model  The choice of dynamic model versus static models depends mainly on which set of approximations is more acceptable and what data is available Assumptions Data availability Non overlapping Vs Markovien arrivals demand Or Static Model Dynamic ModelNabil Si Hammou, April 2012 64
  • Revenue ManagementOptimization: Fare class mix Impact  Effective use of fare class mix combined with other technique of RM (overbooking) have been estimated to generate 4%-6% incremental increase in revenue.  The fare class mix (leg-based RM approach ) is used to maximize revenues on each flight leg. For connecting itinerary demand, the lack of availability of any one flight leg seat in the itinerary limits sales. Interdependence between flights Revenue resulted from leg- based RM approach is not necessarily the maximum of the total revenues on the airline’s network  Revenue maximization over a network of connecting flights requires to jointly manage the capacity controls on all flights Latest version of Network Revenue Management revenue management system65Nabil Si Hammou, April 2012
  • Revenue ManagementOptimization: Network revenue management Definition  Network revenue management (called also Origin–Destination Control) is to manage the seat inventory by the revenue value of the passenger’s O-D itinerary on the airline’s network  O-D control represents a major step beyond the fare class mix capabilities of most third-generation RM systems, and is currently being pursued by the largest and more advanced airlines in the world.Nabil Si Hammou, April 2012 Best practices - Optimization process 66
  • Revenue ManagementOptimization: Network revenue management Control types  The capacity allocation control could be implemented in the reservation system by the extension of controls defined for the fare class mix. A product in this case is an origin-destination itinerary fare class combination.Partitioned Booking limits Virtual Nesting Bid price Maximum of seats on each Mapping to virtual class of single flight for each itinerary single flight and use nesting control of single flight Used only for computations Complexity of mapping Simpler, popular Not used for controlNabil Si Hammou, April 2012 Best practices - Optimization process 67
  • Revenue ManagementOptimization: Network revenue management Modeling: Input Deterministic Random M :set of single flight Dj(t) :1 if the product j is realized in N : set of product (itinerary O-D with fare class). the period t. 0 otherwise ai,j : 1 if the single flight i used by the product j. Xi : reaming capacity of single flight) t: time ( running from1 to T).; pj: price of product j Decision policy Uj(t):1 if we accept a request for product j in period t 0 otherwise. Dynamic Programming Complexity of dynamic programming for network Approximation revenue management 68
  • Revenue ManagementOptimization: Network revenue management Modeling:  One of the most popular approximation used in the practice is based on the aggregation of the expected future demand substitute the future demand by its expected value. Deterministic linear model Input Decision variable M :set of single flight Yj maximum number of demand : N : set of product (itinerary O-D with fare class). for product j ( ODIF itinerary with ai,j : 1 if the single flight i used by the product j. fare class ) to accept. Xi : remaining capacity of single flight i “partitioned booking limits” pj: price of product j E[Dj ]:expected value of the future demand of the product jNabil Si Hammou, April 2012 Best practices - Optimization process 69
  • Revenue ManagementOptimization: Network revenue management Modeling: deterministic linear model Model Maximizing total revenues max Pj * Y j j N subject to Single flight capacity constraint ai , j * Y j Xj i M j N Itinerary demand limit constraint 0 Yj E[ D j ] j NNabil Si Hammou, April 2012 Best practices - Optimization process 70
  • Revenue ManagementOptimization: Network revenue management Modeling: deterministic linear model Solving Methods Exact Methods Approximate Methods Column Branch and Meta-heuristic Generation & Specific Bound ( genetic Branch and heuristic algorithm…) Bound  The comparison analysis regarding exact and approximate method for the fleet assignment remains useful for the network revenue management.  The use of column generation combined with branch and bound algorithm has already demonstrated its powerful for some airlines to solve the deterministic linear model of network revenue management.Nabil Si Hammou, April 2012 Best practices - Optimization process 71
  • Revenue ManagementOptimization: Network revenue management Modeling: deterministic linear model Primal solution Dual solution Definition Definition of primal Definition of bid Definition of partitioned booking solution price dual solution limits Partitioned booking limits =Primal solution Bid price= Dual solution limitj=Xj BidePricei= i for each product j ( itinerary with fare class) for each single flight capacity constraint i Primal solution size > Dual solution size Bid price control the most useful controlNabil Si Hammou, April 2012 72
  • Revenue ManagementOptimization: Network revenue management Modeling: deterministic linear model  By using bid price control, the decision policy becomes: Accept thedemandof product j if p j i single flight i itinerrary j Rejectotherwise with : Pj : price of product j i : bid price of flight leg i  Some airlines have also used these values of bid price for the fleet assignment and/or fleet planning ( demand-driven dispatch). The bid price value associated to a single flight represent the marginal value of revenue would be generated in case of increasing the flight capacity by one seat.Nabil Si Hammou, April 2012 Best practices - Optimization process 73
  • Revenue ManagementOptimization: Network revenue management Modeling: deterministic linear model improved  The deterministic linear model makes one particularly hard assumption: demand is deterministic.  In order to incorporate the stochastic information into the deterministic linear model, airlines could replace the expected value of demand in the mathematical model by simulating many times the randomized demand.  The bid price become the average of bide prices related to each sample. This approach is called the randomized linear programming model 74
  • Revenue ManagementOptimization: Network revenue management Impact  Simulation studies of airline hub-and-spoke networks have demonstrated notable revenue benefits from using network revenue management over leg-based revenue management (fare class mix).  While the potential benefit may be high, network RM poses significant implementation and methodological challenges such as volume of data, organizational challenges.. . Best practices - Optimization process 75
  • Revenue ManagementOptimization Other Alternatives  In addition to the incremental revenue generated by optimization models either fare class mix or network revenue management, the airline could also enhance its incomes by : - Taking into account the cancellation and non-show passenger in the process of the capacity allocation control ( overbooking) - Improving the quality of optimization model inputs ( forecasting)  A 10% improvement in forecast accuracy can translate into 0.5% incremental increase in revenue generated from the RM system.Nabil Si Hammou, April 2012 Best practices - Optimization process 76
  • Revenue ManagementDemand forecasting Introduction  Optimization models use stochastic models of demand and hence require an estimate of the complete probability distribution or at least parameter estimates (e.g., means and variances) for an assumed distribution.. Forecasting Optimization Inventory system Control  The outputs of the forecasting module are fed to the optimization module for producing RM controls such as booking limits, bid prices...Nabil Si Hammou, April 2012 Best practices - Optimization process 77
  • Revenue ManagementDemand forecasting Forecasting  For RM, airlines are mostly interested in forecasting demand at various levels of aggregation (flight leg fare class vs. origin-destination fare class; fare class vs. booking class).  Usually, airline needs also to forecast other quantities such as, cancellation and no-show rates ….  The input requirements of the optimization module drive RM forecasting requirementsNabil Si Hammou, April 2012 78
  • Revenue ManagementDemand forecasting Forecasting methods :  Forecasts may be made by using different types of models and each technique may be used to forecast a variety of behaviors.  In terms of forecasting methods, the emphasis in RM systems is on speed, simplicity, robustness and accuracy, as a large number of forecasts have to be made and the time available for making them is limited.Nabil Si Hammou, April 2012 Best practices - Optimization process 79
  • Revenue ManagementDemand forecasting Forecasting methods : practices  Because of its relative simplicity, exponential smoothing tends to be the most common methods used for demand forecasting in the airline industry. Exponential smoothing  Some vendors have combined the exponential smoothing with other methods such as Kalman filter or linear regression to improve the demand forecasting quality. Exponential Kalman filter smoothing Weighted combined forecast  For modeling passenger choice behavior, some vendors have regressed this behavior as multinomial logit model that contains following variables: Outbound displacement Elapsed time Number of Logit Model Origin point connections presence Fare “logarithm(fare))” 80Nabil Si Hammou, April 2012
  • Revenue ManagementImplementation Change assessment  Before implementing a new optimization or forecasting system, the airline should analyze the potential revenue impact of changing to new RM system: revenue- opportunity assessment. Revenue-opportunity assessment. Investment Implementation cost Benefit Preimplementation phase Post implementation phase Current RM system V0 New RM system V1 Leg based control Network control Booking limit control Bide price control … … . . … . Exponential smoothing Kalman filter & Exponential smoothing  Simulation methodology is the most common method used in practice for revenue-opportunity assessment 81
  • Revenue ManagementImplementation Revenue-opportunity assessment : Simulation  By modeling the current control processes , the planned control processes and customer behavior, a reasonably estimation of revenue benefits of changing to a new revenue management system can be obtained via simulation. VS 82
  • Revenue ManagementChallenges / Future Choice-based revenue Airline alliances managementNabil Si Hammou, April 2012 Best practices - Optimization process 83
  • ConclusionNabil Si Hammou, April 2012 Best practices - Optimization process 84
  • ConclusionRobustness  Substantial progress in optimization techniques and computing power has allowed significant progress to be made in the optimization of : - aircraft and crew scheduling - revenue management.  The schedule planning and optimization processes at airlines produce plans that are rarely executed exactly as planned on a daily basis because of disruptions.  To respond to the disruptions, airlines must replan and create feasible and cost- effective recovery plans in a short period of time. Two approaches are possible: Schedule recovery Vs Robust schedule 1.Develop a new schedule in case of 1.Integrate the expected recovery irregular operations to reassign costs in the objective of the usual resources and adjust the flight schedule process. schedule . 2.The usual schedule becomes more 2.Keep the usual schedule process resilient to disruptions and easier to invariable repair when replanning is necessary.Nabil Si Hammou, April 2012 Best practices - Optimization process 85
  • Thanks for your interestNabil Si Hammou, April 2012 Best practices - Optimization process 86
  • CV: Nabil Si Hammou  Being an Optimization Specialist with strong background in the use of operations research and forecasting methods in the airline industry, I am particularly interested in the positions: – Scheduling optimization specialist – Revenue management optimization specialist  During my professional career, I have developed optimization programs to support decision making system in different industries. – Crew scheduling within Royal Air Maroc : reduction of operating cost by 250.000€ annually. – Transportation scheduling within LOreal France : reduction of transportation cost by 8% – ….  I would welcome the opportunity to discuss with you the potential for making a significant contribution in optimizing the scheduling and revenue management process. Feel free to call me at 00.212.6.18.98.38.61 or email me at n.sihammou@gmail.com. Nabil Si Hammou Optimization Specialist n.sihammou@gmail.com 00.212.6.18.98.38.61 87
  • Information sourcesSeminars & references Seminar organized by Air France Seminar organized by AGIFORS Operations Research within Air Advancement of Operations France Research in the airline industry The global Airline industry A Unified Column Generation Revenue Management Optimization Mr P. Belobaba, Mrs C. Approach for Crew Pairing and at Air Canada Barnahart crew restoring at Lufthansa Mr J.Pagé Mr N.Howak Airline Operations and Scheduling Operations research and scheduling at Revenue Management O-D control Mr M. Bazargan American airlines at KLM Mr T.Carvalho Mr A.Westerhof Demand Forecasting Computational Intelligence in The Theory and Practice of Revenue at United Airlines Integrated Airline Scheduling Management Mr K.Usman Mr T. Groshe Mr K. Talluri, Mr G.Ryzin 88
  • Best practices on the optimization process in the airline industry Nabil Si Hammou, Optimization Specialist n.sihammou@gmail.comScheduling and Revenue Management April 2012