Improving Revenue by System Integration and Cooperative Optimization

1
Improving Revenue bySystem Integration
andCooperative Optimization

• Reservations Yield Management Study Group
• Annual Meeting Berlin 16 - 19 April
  2002

Georg Kliewer Sven Grothklags Klaus Weber
UNIVERSITY of PADERBORN
2
Motivation - Planning Processes
FleetAssignment
MarketModeling
Crew Pairing
OperationControl
NetworkDesign
AircraftRotation
RevenueManagement
Crew Rostering
3
Agenda

• Fleet Assignment Problem

• Business Process

• Mathematical Model

• Optimisation Algorithm

• OD Revenue Management

• Current System Integration

• System Integration Approaches

• Conclusions

4
Fleet Assignment - Business Process
SchedulingActivity
Development
OperationsControl
NetLineApplication
NetLine/Plan
NetLine/Sched
NetLine/Ops
weekly
fully-dated

• Find most Profitable Subfleet Assignment
• Use a Subfleet that is Feasible for each Leg
• Use the Existing Fleet and no extra Aircraft
• Obey Blocktime and Groundtime Constraints

Fleet Assignment
5
Fleet Assignment - Business Process (cont)
6
Strategic Profitability Evaluation Model

• Strategic-PEM is a forecasting model that
  measures the
• network profitability contribution of each
  schedule scenario

• The Strategic PEM forecasts are developed on an
  OD basis
• for a representative week for a schedule
  period

• Input Data
• average revenue for the ODs
• market demand for the ODs
• cost model
• fleeted schedule

• Output Data
• overall schedule profitability
• profit information for each leg-subfleet
  combination

7
Strategic Profitability Evaluation Model (cont)
Fleet Assignment
Profitability Evaluation Model
schedule av. revenues market demand
Connection Builder
itineraries
Market Share Model
assignment
unconstrained demand
prelim. capacity assignment
Spill Recapture
Simulated Annealing
constrained passenger flow
Revenue CostEstimation
objective function for fleet assignment
8
Fleet Assignment - Mathematical Model
PAD
leg l
xl,f1 ? leg l is assigned to subfleet f
FRA
yv,v aircraft flow on the ground
xl,f
yv,v
MUC
v-
v
v
ground arc (v-,v)
flight event
9
Fleet Assignment - Optimisation Algorithm
Fleet Assignment
Profitability Evaluation Model
finalassignment
fleet schedule restrictions
GenerateNeighbor Solution
Update Solution
Simulated Annealing
Accept/Reject
Objective Function
10
Fleet Assignment - Optimisation Algorithm

• Heuristic solution method based on local search

S initial solution T warming_up do do Snew
neighbor solution (S) ?cost cost(Snew) -
cost(S) if (?cost lt 0 or accept(?cost,T) ) S
Snew until equilibrium T update(T) until frozen
1
Equilibrium
P
Frozen
11
Fleet Assignment - Results
Algorithm Hill-Climber Simulated Annealing Mixed-Integer Program
?-solution quality 98,5 99,7 gt99,9
12
Agenda

• Fleet Assignment Problem

• OD Revenue Management

• Business Process

• System Overview

• OD Control Parameters

• Current System Integration

• System Integration Approaches

• Conclusions

13
OD Revenue Management Business Process
BusinessActivities
Development
Pricing
Revenue Management
OperationsControl
Application
NetLine/Plan
NetLine/Sched
NetLine/Price
ProfitLine
NetLine/Ops
14
OD Revenue Management System Overview
Forecasts based on either

• booking histories

• e.g. ODIF POSbooking limits

• or
• market model and price elasticity model

• e.g. fare buckets

• or
• both

15
OD Revenue Management OD Control Parameters
Parameters to beused by fleet assignment process
OD control parameters depend on type of optimizer

• EMSR varieties
• booking class limits per leg / compartment
• on various granularity levels, e.g. ODIF POS
• bid prices derived from EMSR curve or
  approximation
• fare bucket limits per leg / compartment
• revenue bucket limits per leg / compartment
• Dynamic programming
• bid prices per leg / compartment
• booking class limits derived from bid prices

• Reflect
• demand
• demand variability
• estimated fares

16
Agenda

• Fleet Assignment Problem

• OD Revenue Management

• Current System Integration

• Cooperative Approach

• System Integration Approaches

• Conclusions

17
Current System Integration - Cooperative Approach
Profitability Evaluation Model
Fleet Assignment
Connection Builder
itineraries
FinalAssignment
Market Share Model
1
unconstrained demand
4
Fleet Change on a Leg Sequence
Update Solution
capacity assignment
Simulated Annealing
Spill Recapture
5
3
constrained passenger flow
Accept/Reject
Revenue CostEstimation
6
2
Objective Function
18
Current System Integration - Cooperative Approach
(cont)
State

• Strategic PEM Fleet Assignment NetLine/Plan
• Tactical PEM Fleet Assignment NetLine/Sched
• fully-dated schedule
• Basic OD Fleet Assignment
• Approximative revenue estimation
• Fast evaluation loop essential for Simulated
  Annealing
• even faster with internal feedback by passenger
  flow model

19
Agenda

• Fleet Assignment Problem

• OD Revenue Management

• Current System Integration

• System Integration Approaches

• Internal Feedback

• External Feedback by OD Revenue Management
  System

• System View

• Conclusions

20
System Integration Internal Feedback
Profitability Evaluation Model
Fleet Assignment
Connection Builder
FinalAssignment
itineraries
Market Share Model
1
3
unconstrained demand
Fleet Change on a Leg Sequence
Update Solution
4
capacity assignment
Simulated Annealing
Spill Recapture
Passenger Flow LP-Model
passenger flow
Accept/Reject
Revenue CostEstimation
Dual Variables for Legs
2
5
Objective Function
21
System Integration External Feedback by OD RMS
Profitability Evaluation Model
Fleet Assignment
OD Revenue Management
FinalAssignment
Connection Builder
itineraries
ODForecast Engine
Market Share Model
Update Solution
OD Forecasts
unconstrained demand
Fleet Change on a Leg Sequence
Simulated Annealing
Capacity Assignment
capacity assignment
OD Optimiser
Spill Recapture
OD Control Parameters
passenger flow
RevenueEstimation
Accept/Reject
CostEstimation
Objective Function
22
System Integration System Details
OD Revenue Management
Iterative Proration (ProBP)
ODIF POS buckets
OD Control Parameters
23
System Integration System Details (cont)
OD Revenue Management
ODForecast Engine
OD Forecasts
Capacity Assignment
OD Optimiser
OD Control Parameters
RevenueEstimation
Estimated revenue sum of EMSR values
24
System Integration System Details (cont)
Dynamic Bid Price (DynBP) models revenue
management as multistage decision process
OD Revenue Management
OD Control Parameters
Bid price matrix per compartment
25
System Integration System Details (cont)
Bid price matrix per compartment
OD Revenue Management
ODForecast Engine
OD Forecasts
Capacity Assignment
OD Optimiser
OD Control Parameters
RevenueEstimation
Estimated revenue sum of bid prices
26
System Integration Discussion
State
Challenges

• two loop system
• revenue estimation loop
• fleet assignment loop
• significant runtime differences
• coordination of loops
• capacity dependency of revenue management
  optimization
• runtime! Approximation?
• partly overlapping system functions (market
  model, OD-Forecast)
• redundancy, requires further integration

• forecast uncertainty and actual fares taken into
  account
• simple interfaces between fleet assigner and
  revenue management system
• more accurate revenue estimation

27
Agenda

• Fleet Assignment Problem

• OD Revenue Management System

• Current System Integration

• System Integration Approaches

• Conclusions

28
Conclusions - Fleet Assignment in Practice
Simulated Annealing-based Fleet Assignment
Optimization

• substantial part of commercialFleet Assignment
  tool

Cooperative Approach

• applied in Fleet Assignment Systems of ...

Integrated Approaches

• prototype

29
Conclusions - Summary

• Integration of Fleet Assignment and Revenue
  Management can improve overall revenue
• Facets of the fleet assignment problem
  introducedbusiness, mathematics, simulated
  annealing
• Substantial part Profitability Evaluation Model
  (PEM)
• Revenue management system introducedparticularly
  important for integration control parameters
• Current state of integration Cooperative
  Approach with PEM (without integration of RMS) -
  successfully used by three airlines
• System integration prototype
• based on internal passenger flow model
  (deterministic LP)
• based on external link to OD RMS EMSR-based,
  dynamic programming

good
better
almost perfect
30
Conclusions - Ongoing Work, Outlook

• Ongoing tests of system integration prototype
• Market Model used in FA is also used in Pricing
• further Integration possible
• Integration of the Passenger Flow Model in Fleet
  Assignment
• optimization approach for enhanced OD Fleet
  Assignment
• Usage in short-term ? pre-implementation phase
• booking dependent (demand driven) fleet
  assignment

31
Thank you for your attention!Any questions?
UNIVERSITY of PADERBORN
sven_at_uni-paderborn.de georg.kliewer_at_upb.de klaus.w
eber_at_lhsystems.com