
539c3c5a7375ef422983614c90dff140.ppt
- Количество слайдов: 48
Planning and Operating United Airlines: Business Model and Optimization Enablers Gregory Taylor Senior Vice President – Planning United Airlines R&D - United Operations
Operating Facts United Airlines flies 1, 700 daily flights Second largest United currently has 62, 000+ employees worldwide to carry customers safely, conveniently and efficiently airline in the world $11. 6 billion passenger revenue 58. 4 million domestic passengers $0. 6 billion cargo revenue United Express flies 1, 700 daily flights All numbers are for calendar year 2003 8. 7 million international passengers 2
Operating Facts United's customers enjoy access to more than 700 destinations around the world through Star Alliance, the leading global airline network 700+ destinations in 128 countries 109 destinations in 23 countries United's Mileage Plus® program, with almost 40 million enrolled members, regularly receives awards from leading business travel publications 3
Operating Fleet United currently uses 532 aircraft to support its worldwide operations Boeing 737 Airbus 319 United Express carriers currently use 200+ aircraft in their operations Jetstream 41 Airbus 320 EMB 120 Boeing 757 Beech craft 1900 Boeing 767 Canadair Boeing 777 BAE 146 Dornier 328 Boeing 747 United Airlines United Express 4
Large Hubs in Five Major Cities 5
United is the Largest International Carrier 6
United’s Route Network Model Air travel is dominated by thousands of small markets where total travel demand does not justify “point-to-point” non-stop flights Western United States Eastern United States Las Vegas (LAS) Boston (BOS) Seattle (SEA) Albany (ALB) Portland (PDX) Buffalo (BUF) LAS BOS SEA ALB PDX BUF 7
United’s Route Network Model United has chosen a “Hub-and-spoke” model that maximizes number of markets served with given aircraft assets LAS BOS ALB ORD SEA PDX BUF Hub-and-spoke • This model provides several additional connecting options to the customers through Chicago (ORD) • United is also able to carry local traffic between all six cities and ORD 8
United’s Route Network Model In addition to the 59 passengers from the original three markets, 91 more passengers from six new markets were accommodated In addition, United was able to carry 1600 passengers each-way between the six cities and its hub, ORD 9
The Chicago Hub Chicago 2003 Operating Statistics United and United Express Number of cities served 125 Number of markets 7800 Number of departures Total passengers 360, 377 15, 450, 424 Local passengers 8, 034, 220 (52%) Connecting passengers 7, 416, 204 (48%) 10
United’s Scheduling Strategy United’s scheduling strategy balances marketing goals and operating imperatives to meet financial goals Marketing goals • • Marketing strategy Maintain market share Competitive response Provide travel day and time flexibility to passengers • Market selection – Where should we fly? • Flight frequency/time – How often should we fly? Operating imperatives • • Safety/maintenance requirements Aircraft availability Crew availability Other operating restrictions Profitability Financial goals • • Maximize revenue Minimize cost – When should we depart/arrive? • Fleet selection – Which aircraft type should we use? 11
Low Price sensitive Willingness to commit in advance Passenger Segmentation Strategy Higher Ø Frequent schedules Ø Last minute availability Ø Full service Ø Global access Ø Recognition • Leisure travelers Ø Low fares And schedule flexibility • Business travelers F A R E S Ø Quality service High Lower 12
Capacity Control Problem: UA 881 on Sep 16 2004 334 0 187 7 14 Travel restrictions 3 Business 26 17 110 13 95 56 passengers paying an average fare of $238; total revenue $13, 328 17 Sale 7 79 24 Sale 14 60 Demand 69 passengers paying an average fare of $75; total revenue $5, 175 28 Fares 125 passengers paying an average fare of $148; total revenue $18, 503 Leisure No. of advance purchase days High 13
What is O&D Control ? SFO (1 Se at) (1 Seat) ORD LGA at) (1 Se LAX Itinerary Fare Demand LGA-ORD $100 5 ORD-LAX $100 2 ORD-SFO $100 1 LGA-ORD-LAX $150 5 LGA-ORD-SFO $225 1 14
O&D Control Yields Better Revenue SFO (1 Se at) (1 Seat) ORD LGA at) (1 Se LAX Itinerary Fare Demand Leg Based ORION LGA-ORD $100 5 1 0 ORD-LAX $100 2 1 1 ORD-SFO $100 1 1 0 LGA-ORD-LAX $150 5 0 0 LGA-ORD-SFO $225 1 0 $300 1 $325 15
Operations Research at United Airlines R&D - United Operations
Enterprise Optimization - Overview Mission. Provide thought leadership and ground breaking research capabilities that challenge the status quo ; partner with business units and delivery groups to create value through excellence in modeling and research. The Group Experts in optimization and forecasting techniques dedicated to solving complex business problems § Approximately 45 people § Advanced degrees in Mathematics, OR, Statistics, Transportation Science, Industrial Engineering, and related fields § 19 Ph. Ds § Mix of employees from academia, the airline industry, and management consulting § Partnerships with universities The Activities Solve complex business problems using math modeling, forecasting, stochastic modeling, heuristic optimization, statistical modeling, game theory modeling, artificial intelligence, data mining, and other numerical techniques § Review business processes in highleverage areas § Rapidly develop model prototypes to validate theories and provide quick returns § Partner with IT professionals to build full blown, robust production systems 17
Enterprise Optimization – Business Areas Aircraft Scheduling § Profitability forecasting to make long term business plan decisions including market selection and frequency of operations. § Fleet Assignment models for fleet planning and profit maximization. § Aircraft Routing models to operationally route aircraft § Codeshare Optimization to effectively manage the growing revenue opportunity through partner airline relationships. Crew Planning § Crew Scheduling Models to efficiently plan trips and monthly schedules for pilots and flight attendants. § Crew Manpower Planning Models for pilots and flight attendants to manage complex decisions including staffing levels, training levels, vacation allocations and distribution of crew among geographically dispersed bases. Revenue Management § Revenue Optimization models focused on inventory, pricing, and yield. § O&D Demand forecasting to feed decision making in revenue optimization models. § Next Generation Revenue Management model to more effectively compete with growing airline segment of Low Cost Carriers. Supply Chain Management § Models to balance reduction in inventory costs while maintaining and improving the reliability of our operation. Day of Operations • Models to respond and recover from irregular operations. 18
Overview of United’s Network Planning Automation Suite - Zeus 19
ZEUS Enables All Stages of Planning and Scheduling Strategic Planning Schedule Optimization Process Strategic Planning Long Term Planning Mid Term Planning Operational Planning Time* > 180 days 180 -108 days 108 -80 days 80 -52 days Activities Key Models • Hub Planning • Fleet Plan • Acquisitions • Schedule Structure • Profitability Forecast (PFM) • Joint UA-UAX Fleet Planning • Codeshare Optimizer • Markets • Frequencies • International Slots • Fleeting • Crew Interactions • Reliability • Maintenance • Operability • Aircraft Flows • De-peaking • Reliability • Flight Number Integrity • Weekends, Transition • PFM • Joint UA-UAX Fleet Assignment • UA Fleet Assignment • Re-Fleeting • Routing • Through Assignment / Routing • Flight Number Continuity • Exception Scheduling • De-peaking Suite *Time = days from schedule start date 20
The Zeus Suite International Flouting O&D Fleeting Slot Administrator SIMON Data Query & Analysis Airline Simulation Profitability Forecast AIRFLITE Schedule Database/Editor Weekend Cancellation Fleet Assignment Level of Operations (LOOPS) Re-fleeting Models Through Assignment Neighborhood Search Maintenance Routing Dissemination IDEAS 1 PLAN Web Portal 21
Profitability Forecast Model (PFM) Objective PFM is United’s strategic network-planning tool. PFM incorporates historical cost and fare data with itinerary-level passenger forecasts to determine schedule profitability Inputs Methodology and Key Capabilities Outputs Competitive Schedules (OAG) PFM employs advanced econometric techniques (Multinomial Logit (MNL) methodology) • Passenger preference factors for itinerary attributes (# of stops, departure time, equipment, codeshare, etc. ) are simultaneously estimated using MNL techniques • Consistent with passenger utility-maximizing choice behavior Passengers (total, local) Industry Demands Cost model Industry fares Fares (local, OD) Revenue (local, OD) PFM aids strategic decisions such as: • Merger and acquisition scenarios • Codeshare scenarios • Equipment preference studies • Hub location/buildup studies Profitability of future schedule MAPD – Mean Absolute Percent Deviation 22
Fleet Assignment Models Objective The O&D models are used to obtain the optimal fleet assignment for a flight schedule based on itinerary based demands and market share Inputs Methodology and Key Capabilities Outputs UA Schedule Itinerary Level demand fare forecasts Aircraft Inventory By Type Aircraft Characteristics, Cost, Operational, other constraints The model uses advanced Operations Research techniques to solve the entire network to determine the optimal fleet assignment. Uses a Mixed Integer Linear Program. Maximizes UA’s profitability subject to various operational and other constraints. Fully fleeted schedule Time Windows capability creates opportunity for further improve profitability by making small changes to departure/arrival times 23
Codeshare Optimizer Objective Codeshare Optimizer is a strategic decision-making tool to determine the best set of flights to code share based on market share and prorate agreements. Inputs Methodology and Key Capabilities OAG Schedule Codeshare Optimizer uses a Dynamic Program-like approach to model incremental code share opportunities and PFM’s itinerary building algorithms and LOGIT methodology Market List The objective is to maximize incremental revenue while satisfying the flight number and other marketing constraints Outputs Airport-pair passenger forecasts Marketing Constraints Ability to support several scenarios: • Evaluate new codeshare or expand existing codeshare • Optimize flight number usage when there is a shortage of flight numbers • Make tactical market/flight changes during major schedule change List of flights with best Codeshare Revenue 24
Exception Scheduling Model Objective Optimize exceptions on weekends to improve profitability while adhering to operational constraints Inputs Methodology and Key Capabilities UA Schedule Outputs The model uses a Mixed Integer Linear Program to model the weekend schedule and maximize the profitability subject to operational and other constraints Demand Fare Forecasts Operational Constraints Associated business process changes have resulted in independent construction of optimal weekday and weekend schedules Fully Fleeted Weekend Schedule The model ensures that the weekend schedule meshes seamlessly with the surrounding weekday schedules The model recaptures demands from canceled flights and moves the demand to neighboring flights in the market 25
Hub De-peaking Suite Objective Fine-tune United’s schedule to meet airport capacity requirements with minimal revenue impact Methodology and Key Capabilities Outputs An Integer-Programming optimizer determines the flight retimings from the baseline schedule Inputs De-peaked Schedule UA Schedule PFM Demand Forecasts Objective is to minimize revenue loss while satisfying depeaking and gating constraints De-peaking and Gating Restrictions 26
Schedule Improver (Simon) Objective Simon determines the optimal schedule to fly from a given base schedule and a large superset of potential flight opportunities. Inputs Methodology and Key Capabilities Mandatory and optional flights Given an aircraft inventory and a list of potential flights to fly, SIMON selects flight legs and assigns fleet types to flight legs in order to maximize contribution. O&D level demand Simon honors a host of operational constraints including those related to maintenance, noise, and crew availability. In addition, users can specify schedule structure constraints. Outputs O&D level fares Optimal Schedule By varying the amount of the schedule that is considered mandatory, users can control the amount of changes to an existing schedule in an incremental manner. Cost model Simon can intelligently determine the best pattern of flights to retain in any market 27
Revenue Management Automation Suite R&D - United Operations
This Section Will Focus on Yield (Inventory) Management Schedules Objective: Develop optimal schedule network based on market forces, estimated demand/fares, available capacity, operational imperatives, etc. Pricing Objective: Set the fares to maximize revenue across customer segments and to effectively compete in the market place Yield Management Objective: Given a schedule and estimated demand/fares, optimally allocate the seat inventory on each flight to ensure revenue-maximizing passenger mix 29
United has been the Leader in Adopting Cutting Edge Yield (Inventory) Management Technologies Major Airlines Overbooking systems 1980 s Leg based Inventory Management systems with fare class control reservation systems AA, SAS implemented O&D systems in the 1990 s. CO, LH started using O&D controls in the mid 1990 s 1990 - 1995 1996 - 2000 Overbooking systems Static O&D system with O&D control Orion Development 2001 - 2003 Orion implementation included path based forecast, network optimization and dynamic passenger valuation Enhancements to systems to compete with Low Cost Carriers 2004 and Beyond Strategic research to compete with Low Cost Carriers 30
United’s Yield Management System - Orion Pricing and Accounting Systems tickets, data published fares rules Orion Passenger Valuation Base Fares adjustments RM Planners PV parameters controls Optimization Path level demand & no-show forecast adjustments Aircraft Scheduling Demand Forecasting AU Levels Displacement Costs Inventory System (Apollo) bookings cancellations schedule change departure data Travel Agents United Res. Online Agencies schedule 31
High-Level Orion Statistics • Flight Network Q Orion optimizes revenue on approximately 3, 600 UA and UAX daily departures Q About 27, 000 unique paths are flown each day by United’s customers • Forecast and Optimization Statistics Q Orion produces 13 million forecasts for all 336 future departure dates Q All future departure dates are optimized every day Q Orion produces flight level controls for nearly 1. 1 million flights in the future Q Options exist for analysts to load changes into Apollo throughout the day Q Passenger valuation produces new base fares every two weeks • Hardware infrastructure Q A dedicated IBM supercomputer complex is utilized to run the forecasting and optimization algorithms 32
Demand Forecasting System Objective Q Estimate future bookings at the path, fare class, point of sale level for all future departure dates; Estimate the cancellation rates of existing and future bookings Inputs UA schedule Methodology and Key Capabilities Model Technology • Exponential smoothing based forecasting method utilizes most relevant historical data Path level booking and cancel data Special events calendar User adjustments Outputs Types of Forecast Models • Rejected Demand • Seasonality • Special events – Used for targeted periods • Groups • No-shows • Future path class point of sale booking forecasts • Cancellation rates of current and future bookings 33
Passenger Valuation System Objective Q Forecast the expected value of future passenger demand Inputs Current fares for future travel periods Methodology and Key Capabilities Outputs • Establish the fare value proxy for O&D using • Weighted average of historical usage • Current selling fares for future travel periods • User adjustments Historical usage of fare products • O&D fare forecasts • Fares are updated every two weeks, to reflect accurate information on future fares User Adjustments • Fares can be established based on • Day of week • Connection type • Departure date range • Point of sale 34
Optimization System Objective Q Determine optimal space planning levels based on no-show, cancellation forecasts and upgrade potential; Estimate the displacement costs of each future flight leg Q Use displacement costs and other parameters to optimally allocate seats to buckets on each flight leg Inputs UA schedule Path level demand, cancel forecasts No-show forecasts O&D fare forecasts Methodology and Key Capabilities Optimization Model - Displacement Adjusted Virtual Nesting (DAVN) • Space planning • Overbooking model • Upgrade potential • LP based network optimization to determine displacement costs • Capacity control • EMSR(b) method to optimally allocate seats Outputs • Flight bucket authorization levels • Displacement costs Key Capabilities • Space planning model distinguishes between true noshows and revenue standbys • Overbooking dials to throttle bookings 35
Availability Processing Objective: Q Evaluate availability requests based on path value and bucket availability Inputs Flight bucket level authorizations Displacement costs for all future flights Methodology and Key Capabilities • Each booking request is broken up as one-way paths • Each path is assigned a value based on the fare class, point of sale and other information • Fare Class-to-Bucket mapping is determined using the fare value and displacement cost of the legs traversed by the path • Bucket availability on each leg of path is used to accept or reject booking • Virtual nesting leads to dynamic mapping of paths to buckets UA schedule Outputs • O&D availability of inventory • Accept/reject decisions of booking requests
Advanced Availability Processing Challenges and Opportunities • Consumers are price conscious and conditioned to shop for travel • Availability of internet outlets is increasing shopping activity • Most airlines are experiencing higher look to book ratios, stretching computing capability • Opportunity to further tailor product offering to passenger segments Advanced Availability Processing • Increased inventory control capabilities Q Improved channel control Q Customer centric RM • Distribution capabilities • Manages dramatic growth of availability requests and reduces processing costs • Maintains revenue integrity through realtime application of inventory controls • Open system architecture for faster development 37
Day of Operations Automation Suite R&D - United Operations
Airport Manpower Assignment Models How many employees do we need at the airport for daily Operations? Customer Service Gate Agents Baggage Handlers Airport Employees Passengers Input Demand & Schedule Output Overestimating Need Costly, Idle employees Underestimating Need Long lines, dissatisfied customers How many employees? Their respective assignments OR-Based Assignment Model Considerations Multiple start times Overtime/Parttime Employees call in sick IRROPS (Bad Weather) 39
Block Time Forecasting Model How many minutes should United take to fly between a City Pair? Initial Response to the Question above: Why doesn’t United fly the most fuel efficient route and use that time? Let’s Use JFK-LAX as an example The range used for a 767 is anywhere between 5: 10 & 5: 30 Output Input Demand Fuel cost Crew Cost # minutes to fly Block Time Forecasting Going Too Fast: Higher fuel cost Going Too Slow: Higher crew costs Missed connections Complications: Enroute Air traffic delays FAA re-routes Weather Statistical Forecasting Techniques 40
Real-time IRROPS Management Models Q: When things go “wrong” on the day-of-operations, what is the best way to “Respond and Recover” ? What can go wrong? 1. Bad Weather (60 days out of 360 days) 2. Aircraft needs maintenance 3. Crew shortage 4. Airport Congestion What are the choices? 1. Cancel the flight(s) 2. Delay a flight 3. Get a Spare Aircraft 4. Get Reserve Pilots/Flight attendants Challenges: All of this has to be done in close to “real time” All Resources have to be “repositioned” so that the next day Operations can run smoothly United has built a whole host of math-based Applications to assist in these decisions 41
Irregular Operations Management at United A “Bad” Day at ORD GDP Issued for ORD Operations Data Store FAA Real-time Information Sky. Path Analyze the Impact of Proposed Re-ordering Dyna. Block Operations Data Warehouse ODS Arrival Sequencing Optimized Re-sequencing of Arrivals at ORD Feedback to Planning Resource Recovery Analyze the Impact of Proposed Cancellations & Recovery Delay Vs Cancels Aircraft Reassignment Pilot Apps Flight Attendant Recovery Optimized set of Cancellations Passenger Recovery All these tools work interactively to provide the overall solution 42
The Future for Operations The Operations Holy Grail: Can there be one Global application that can make ALL these decisions? R&D - United Operations
Irregular Operations Management at united A “Bad” Day at ORD GDP Issued for ORD Operations Data Store FAA Real-time Information Sky. Path Analyze the Impact of Proposed Re-ordering Dyna. Block Operations Data Warehouse ODS Arrival Sequencing Optimized Re-sequencing of Arrivals at ORD Feedback to Planning Resource Recovery Analyze the Impact of Proposed Cancellations & Recovery Delay Vs Cancels Aircraft Reassignment Pilot Apps Flight Attendant Recovery Optimized set of Cancellations Passenger Recovery 44
Irregular Operations Management at united A “Bad” Day at ORD GDP Issued for ORD Operations Data Store FAA Feedback to Planning Real-time Information Sky. Path Analyze the Impact of Proposed Re-ordering Dyna. Block Operations Data Warehouse ODS Arrival Sequencing Ops Global Solver Optimized Re-sequencing of Arrivals at ORD 45
46
Next Frontiers – A Sample • Game theoretic models to predict and respond to competitor actions • Multiple Criteria Decision Making • Modeling trade-offs between key decision variables • Data Mining 47
Summary • The airline industry presents many high-value opportunities for Operations Research systems • United has historically invested, and continues to heavily invest in state-of-the-art tools • United has also consistently partnered with academia to develop cutting edge models • Increasing computing power at lower cost many high value opportunities remain 48
539c3c5a7375ef422983614c90dff140.ppt