Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete...
Transcript of Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete...
![Page 1: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/1.jpg)
Integer ProgrammingGlobal Impact
George NemhauserGeorgia Institute of TechnologyAtlanta, GA, USAEURO, INFORMSRome, ItalyJuly 2013
![Page 2: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/2.jpg)
Integer ProgrammingOptimization models with
integer variablesmostly binary
![Page 3: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/3.jpg)
Binary variables used to modelDecisionsLogical relations
NonlinearitiesNonconvexities
Integer Programming
![Page 4: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/4.jpg)
Agenda for today: Past, present and futureReal applications
Supply chainEnergy Finance
TransportationHealthSports
![Page 5: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/5.jpg)
A bit of history
![Page 6: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/6.jpg)
CUTTING PLANES
Dantzig, Fulkerson, Johnson (1954)
49-city TSP solved by linear programmingand cutting planes
Gomory (1958)
General IP, finite cutting plane algorithmImplementation in FORTRAN
![Page 7: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/7.jpg)
BRANCH-AND-BOUND
Land and Doig (1960)
LP + enumeration
Little et al. (1963)
Coined the term, successful computation
for general MIP
Assignment problem+ enumeration for TSP
![Page 8: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/8.jpg)
HEURISTICSReiter and Sherman (1965) Local search
1
2
3
4
![Page 9: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/9.jpg)
FORMULATIONSDantzig (1957)Many proposed models, but solutions ???
Scheduling problem, more cutting planes thanenumerating all possibilities
![Page 10: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/10.jpg)
Realapplications
![Page 11: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/11.jpg)
Almost a decade to go from methodsand models to solving industrial scaleproblems
Documentation of early applicationsvery difficult to findMostly personal communication
![Page 12: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/12.jpg)
● The petro-chemical industry provided some of the first motivation(Land and Doig were supported by BPto work on maritime routing)
● The first code to successfully "solve" real MIPs was CEIR's LP 90/94 (late 60's)Branch-and-bound code by Martin Beale(together with Forrest, Shaw, Small and Tomlin)
![Page 13: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/13.jpg)
First applications (from Max Shaw)
● Philips Electronics - Location of factories in Spain.
● British Petroleum - Oil refineries and transport.
First published application● UK Military - Allocation of ships and
airplanes (1968)
![Page 14: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/14.jpg)
1970's ● Transition to more powerful MIP codes
● More industrial applications
UMPIRE
SICONIC
MPS X
MPS X 370
![Page 15: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/15.jpg)
Political
(with R. Garfinkel)Districting
Late1960's
Mystartin IP
![Page 16: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/16.jpg)
Political districting Divide a state into districtsRoughly equal population (one man, one vote)
Contiguity, compactness, natural boundariesSafe districts + other politicalconsiderations
![Page 17: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/17.jpg)
Set partitioning problem Generate all feasible districts and their respective "costs".
Choose districts such that each population unit is in exactly one of them and cost is minimized.
![Page 18: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/18.jpg)
Our algorithmImplicit enumeration (no LP) with fathoming rules similar to Constraint Programming
With very careful assembly language programming we could solve problems with
50 populations units, 5-10 districts
![Page 19: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/19.jpg)
Current practiceHuge problems solved with branch-and-price
One man one vote is achieved, but results can significantly favor the political party in power
![Page 20: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/20.jpg)
Transition period1970's - 1980's
![Page 21: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/21.jpg)
Lots of theory developedComplexity theory
Polynomial time LP algorithms
Practical barrier method to compete with Simplex
Specialized cuts for solving MIPsKnapsack cover, flow cover, clique, mixed-integer rounding
![Page 22: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/22.jpg)
But still...Basic LP-based branch-and-bound
Application of MIP very limited
![Page 23: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/23.jpg)
An exception:
Hundreds of thousands of variables
Hundreds of constraints
Cooperation between airlines and academia
Airline crew scheduling
![Page 24: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/24.jpg)
Modern IntegerProgramming
1990's to present
![Page 25: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/25.jpg)
Ability to solve real problemsSpeedups:CPLEX 1.2 (1991)
CPLEX 11 (2007)
29530x
Gurobi 1.0 (2009)
Gurobi 5.5 (2013)
20.5x
![Page 26: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/26.jpg)
Speedups together:1991
2013
256000x
![Page 27: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/27.jpg)
Assuming (a modest) 1000x machine speedup
Speedups together:
It took It takes>4 months (early 90's)
>7 years (early 90's)
1 second (2007)
1 second (now)
![Page 28: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/28.jpg)
What made the incredibleimprovements possible?
Cuts
Heuristics
Branch-and-bound
Better LP
Preprocessing
![Page 29: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/29.jpg)
What made the incredibleimprovements possible?1. Steepest edge dual Simplex for LP2. PreprocessingFixing variables, eliminating constraints, reducing coefficients
3. Cutting planesGomory cuts and many others to tighten bounds
4. HeuristicsSolving sub MIPs to get better solutions
5. Disjunction selection for branching (still open)
![Page 30: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/30.jpg)
Modern IntegerProgramming Codes
CommercialNon-commercial
(open source)
CPLEX
GurobiXPRESS-MP
MINTO
SCIPCOIN-OR
![Page 31: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/31.jpg)
Impact of modernInteger Programming
![Page 32: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/32.jpg)
How do we measurethis impact?Franz Edelman prize finalists (6 per year)
Since 2000
53% of them used MIP or somekind of discrete optimization
![Page 33: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/33.jpg)
Improving Performance and Flexibility at JeppesenJeppesen Sanderson, Inc.
Optimized Crew SchedulingAir New Zealand
Optimizing Customer Mail StreamsFingerhut
Rightsizing and Management of Prototype Vehicle TestingFord Motor Company
Pricing AnalysisMerrill Lynch
Crew recoveryContinental Airlines
Combinatorial and Quantity-Discount Procurement AuctionsMars - IBM
![Page 34: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/34.jpg)
Optimizing Periodic Maintenance OperationsSchindler Elevator Corporation
Transforming Operations to Develop Operating PlansCanadian Pacific Railway
Optimizing Air NetworkUPS
Optimizing Network RoutingMenlo Worldwide Forwarding
Accelerating the Profitability of Supply ChainsHewlett-Packard
Reinventing the Supplier Negotiation ProcessMotorola
Routing OptimizationWaste Management
![Page 35: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/35.jpg)
Improving Fractional Aircraft Ownership OperationsBombardier Flexjet
Gaining Elastic Capacity Using a Decision-Support SystemHonkong international Terminals
An OR/MS Approach to Managing an Educational ComplexNanzan Gakuen
Expressive Competition Enabled by OptimizationProcter & Gamble
Optimizing Supply Chain for Delivering Calcium Carb. SlurryOmya Hustadmarmor
Becoming a Travel RetailerTravelocity
Operations Research Advances Cancer TherapeuticsMemorial Sloan-Kettering Cancer Center
![Page 36: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/36.jpg)
The New Dutch TimetableNetherlands Railways
Scheduling of home care to assist elderly and disabledCity of Stockholm
Optimizing Natural Gas Production and TransportGassco / StatoilHydro
Reduce Contamination Risks in Drinking WaterUS Environmental Protection Agency
Cashing In on Optimized Equipment DistributionCSX Railway
Transforming Product Portfolio ManagementHP
Improving Global ProfitabilityNorske Skog
![Page 37: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/37.jpg)
New Operating and Settlement SystemINDEVAL
Inventory OptimizationProcter & Gamble
Achieving Transportation Asset Management ExcellenceNew Brunswick Department of Transportation
A Strategic Empty Container Logistics OptimizationCSAV
Branch Reconfiguration PracticeIndustrial and Commercial Bank of China Limited
Retail Price OptimizationIHG
Application of OR for Energy and Ancillary Services MarketsMidwest Independent Transmission System Operator
![Page 38: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/38.jpg)
Supply Chain–Wide OptimizationTNT Express
Advancing Public Health and Medical PreparednessUS Centers for Disease Control and Prevention
Reengineering Global Distribution ProcessZara
![Page 39: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/39.jpg)
Some impact areasSupply chainEnergyFinanceHealth
TransportationNatural resourcesSports
![Page 40: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/40.jpg)
Transportation
![Page 41: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/41.jpg)
Airline optimization Pioneers in really using optimization in practice
A huge impact on development of integer programming methodology
![Page 42: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/42.jpg)
Network planningFleet assignmentCrew planning and rosteringGate assignmentRobust planning and operations recovery
Airline optimization
![Page 43: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/43.jpg)
Crew schedulingHow to partition a set of flights by crews
Set partitioning problemwith constraints for every flight and 0-1 variables for every subset of flights that a specific crew could fly over a duty of 4-5 days.
![Page 44: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/44.jpg)
Crew scheduling1960's: Airlines explore IP to solve the problem1980's: Sub MIPs were solved, precursor of the primal methods used today1980's: Special branching rules (Led to branching needed to solve IPs with exponential number of variables by Branch-and-Price)
1990's: Column generation to try to deal with billions of variables
![Page 45: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/45.jpg)
Crew schedulingToday: LPs with 1012 - 1014 variables solved by column generation to produce IPs with about 20,000 - 30,000 variables and 1,000 constraints
![Page 46: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/46.jpg)
Fleet assignmentWhat capacity should be assigned to each flight to maximize revenue
![Page 47: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/47.jpg)
Fleet assignmentMIP introduced as tool in early 1990's (only basic models could be solved)Weekly models can be solved with 5,000 - 6,000 daily flights and 12-15 subfleetsModels include crew, maintenance, airport operations and flight retiming constraintsChallenges in moving from leg to itinerary based fleet assignment
![Page 48: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/48.jpg)
Robust scheduling:Integrated planning and operations recoveryNewer areas that point to the demand to deal withuncertainty and producing online solutions
Much academic work over the last decade
Implementation has just begun and proceeds slowly (limited data, unclear objectives and huge models)
Network planning (markets, frequency, code sharing)
![Page 49: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/49.jpg)
Supply chain
![Page 50: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/50.jpg)
Maritime inventory routingCombines inventory and supply at demand portswith routing of vessels that move the inventory
![Page 51: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/51.jpg)
Maritime transportationSectors servedOil & gasAgricultureMany others
Ships travel the globeTankersBulk carriersContainer shipsLiquefied gas carriersRoll-on roll-off ships for cars
![Page 52: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/52.jpg)
Modal splitby million metric tons
![Page 53: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/53.jpg)
Modal transport costper ton-mile
![Page 54: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/54.jpg)
Maritime transportationleverages the power of MIP
MIP used for strategic and tactical planning
Split pickup & split delivery inventory routing models
Large-scale instances involve routing many ships over a long planning horizon
MIPModel
2M variables500kconstraints
![Page 55: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/55.jpg)
Current practice for a mega oil company
Strategic planning problemTime horizon: 365 periods (days)
Ports: 5Ships: 15
Time to 5% gap (vs. optimal): 1 - 5 hoursTime to 1% gap (relative to bound): days
Need to solve bigger problems, reduce gap and improve times (enormous costs. e.g., demurrage)
![Page 56: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/56.jpg)
Energy
![Page 57: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/57.jpg)
The unit commitment problem Solving the economic dispatch of power
more efficiently and reliably
World gross production (2009)
20,000 Terawatt hrat 100 megawatt hr: Cost $2,000 billion/year
1% savings = $20 billion per year
![Page 58: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/58.jpg)
US System Operated by about 10 regional organizations
Coordinate, control and monitor electricity transmission by nodal pricing
Electricity cannot be stored
![Page 59: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/59.jpg)
US System Real-time for efficient dispatchDay-ahead for efficient unit scheduling
Auctions
Electric network optimization106 nodes106 transmission constraints105 binary variables
![Page 60: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/60.jpg)
MIP paradigm change In the last 15 years optimization switched from Lagrangian to MIPPre-1999
- MIP too slow- LR - simpler models, no commercial solvers
Now MIP is prefered- Ease of development and maintenance- Exact models of complex functionality- Ability to specify solution accuracy- Continuous improvements from multiple vendors
![Page 61: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/61.jpg)
MIP paradigm change 2011 - MIP creates savings > 500 million annually in the US
2015 - Savings predicted to be > 1 billion annually and 10-30% more savings possible
![Page 62: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/62.jpg)
Sports scheduling
Photo: Don Davis
![Page 63: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/63.jpg)
Is this really an impact area?Sports: $300 billion annually
2x
7x
automobile industry
movie industry
TV scheduling constraints are key to revenue
![Page 64: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/64.jpg)
Major league baseball
in the US
Sports Scheduling GroupGN, Mike Trick and Kelly Easton
![Page 65: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/65.jpg)
Why is baseball scheduling hard?
30 teams, 2 leagues of 15 each, 3 divisions of 5 each per leagueEach play 162 games over 180 daysMany hard and soft constraints
Huge size
![Page 66: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/66.jpg)
Why is baseball scheduling hard?
Balanced home and away gamesTravel is limitedStadium constraintsTime between 2 teams playing each otherCan't be away for several consecutive weekendsTelevision
Huge size
![Page 67: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/67.jpg)
Sports scheduling is mainly a
feasibility problem overconstrained and with many soft constraints put into the objective with weights adjusted over several iterations.
![Page 68: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/68.jpg)
The full problem is too big100,000200,000
binary variablesconstraints
broken into subproblems with
15,00020,000
binary variablesconstraints
pieced together heuristically
Process takes 3 - 6 months
![Page 69: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/69.jpg)
Is there any hopeof optimizing withone MIP?
![Page 70: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/70.jpg)
Health
![Page 71: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/71.jpg)
Treatment of prostate cancer with brachytherapy (placement of radioactive seeds inside a tumor)
Integer programmingin the operating room
Over 500,000 new cases per yearOver 30% mortality rate
![Page 72: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/72.jpg)
More cells are killed as seed gets closer3-D location problem
Objective: maximize dose to tumor, avoiding (insofar as possible) exposure to normal tissue
Kill rate
Cell distance from seed Tumor Seeds
![Page 73: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/73.jpg)
Technical challenge:Very dense constraint matrices
ResultsSafer and more reliable treatments (45-60%) reduction in complications20-30% reduction in number of seeds15% reduction in needlesAverage treatment cost in the US reduced by $5,000Similar application to radiation beam treatment of other types of cancer
![Page 74: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/74.jpg)
FinancePhoto: http://www.flickr.com/photos/68751915@N05/
![Page 75: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/75.jpg)
$800 billion to $1 trillion of active fundsmanaged with optimization using estimate of expected returns
Probably another $8 trillion use optimization for passive managementAdd assets to a portfolio to improve its risk characteristics
Managing funds
![Page 76: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/76.jpg)
Min risk,Basic models
constraint on return
Max return,constraint on risk
orInputs: Investment universe, expected returns,
covariance of returns, budget, max acceptable risk or min acceptable return
Output: Portfolio weights
Need quadratic constraints or objective
![Page 77: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/77.jpg)
Fixed transaction costsThreshold holdings ( or )If-then constraintsTax considerations
Binary variables are needed for
![Page 78: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/78.jpg)
up toTypical model size:
10,000 assets1 to 4 0-1 variables per asset1 or 2 continuous variables per assetMany constraints with various types of exposure limits
Optimizing over multiperiods and multiple portfolios still too large to do with MIP
![Page 79: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/79.jpg)
Social networks
![Page 80: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/80.jpg)
Social networks
● Terrorist networks● Intelligent data● Privacy● Data mining
Optimization on graphs
![Page 81: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/81.jpg)
Thefuture
![Page 82: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/82.jpg)
Almost in every area discussed there is a need for:
Dealing with uncertainty
Bigger models (multiple periods, combined systems)
Solving much faster (real-time applications)
![Page 83: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/83.jpg)
Technology innovations
Parallel
Learning
What else?
Execution
![Page 84: Global Impact Integer Programming€¦ · Since 2000 53% of them used MIP or some kind of discrete optimization . Improving Performance and Flexibility at Jeppesen Jeppesen Sanderson,](https://reader034.fdocuments.us/reader034/viewer/2022042122/5e9cb7b441778513ec35f98c/html5/thumbnails/84.jpg)
Thanks to:Max Shaw (retired) and John Tomlin (Yahoo) - Early computational systems and applications
Bob Bixby and Zhongao Gu (Gurobi) - CPLEX and Gurobi
Sergey Shebalov (Sabre) - Airline optimization
Kelly Easton (Sports Scheduling Group) - Major league baseball scheduling
Dimitri Papageorgiou (ExxonMobil) - Maritime inventory routing
Dick O'Neill (US Department of Energy) - Energy optimization
Pam Vance (Axioma) - Finance
Rodolfo Carvajal (Georgia Tech) - General help