Scribd
Upload
39 views

08 Integer Programming 1 Print

1) The document discusses integer programming and modeling problems with integer variables. It provides examples of using binary variables to represent yes/no decisions and integer variables to model quantities. 2) Branch and bound is introduced as a technique for solving integer programs, where the problem space is partitioned into branches that are evaluated and bounded to find optimal solutions. 3) Set covering models are discussed as a class of integer programs for selecting subsets to cover requirements while minimizing costs. These have applications in areas like hub location and set partitioning.

Uploaded by

eviroyer
Copyright
© Attribution Non-Commercial (BY-NC)
Available Formats
Download as PDF, TXT or read online on Scribd
39 views

08 Integer Programming 1 Print

1) The document discusses integer programming and modeling problems with integer variables. It provides examples of using binary variables to represent yes/no decisions and integer variables to model quantities. 2) Branch and bound is introduced as a technique for solving integer programs, where the problem space is partitioned into branches that are evaluated and bounded to find optimal solutions. 3) Set covering models are discussed as a class of integer programs for selecting subsets to cover requirements while minimizing costs. These have applications in areas like hub location and set partitioning.

Uploaded by

eviroyer
Copyright
© Attribution Non-Commercial (BY-NC)
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 35

Integer Programming I

Modeling with Integer Variables How the Solver Works Complexity

15.057 Spring 03 Vande Vate

A Simple Example
Steel Cable Company
Integer Programming Example Market Limit (Tons)

Products Profit/Ton Bands $50,000 Coils $60,000

Hours/Ton 3 3 2 4 Total Hours 12


2

15.057 Spring 03 Vande Vate

An Integer Program
3

Coils Limit
2

Production Capacity
1

Bands Limit Coils


1 2 3 4

Bands
3

15.057 Spring 03 Vande Vate

Branch & Bound


3

Node 0: The linear program


Coils Limit
2

$50,000*3 + $60,000*0.7 = $192,000


Production Capacity

X
Coils
1 2 3

Bands Limit Bands


4

15.057 Spring 03 Vande Vate

Branching
3

Upper Branch Coils 1

Coils

Lower Branch Coils 0


1 2 3 4
15.057 Spring 03 Vande Vate

Bands
5

Bounding
3

$50,000*3= $150,000
Upper Branch Coils 1

Coils

Node 1: Lower Branch Coils 0


1 2

X 3

Bands
6

15.057 Spring 03 Vande Vate

Bounding
3

The best answer yields at least $50,000*3= $150,000


2

Node 2: Upper Branch Coils 1

Yields $188,571

Coils

Lower Branch Coils 0


1 2 3 4
15.057 Spring 03 Vande Vate

Bands
7

Further Nodes
3

Node 3: Lower Branch Bands 2


1

X Yields $184,000

Coils
1 2 3 4

Bands
8

15.057 Spring 03 Vande Vate

Further Nodes
3

Upper Branch Coils 2


2

Node 4: Lower Branch Coils 1 Coils


1 2

X Yields $160,000

Bands
9

15.057 Spring 03 Vande Vate

Further Nodes
3

X Yields $177,143
Node 5: Upper Branch Coils 2

Lower Branch Coils 1 Coils


1 2 3 4

Bands
10

15.057 Spring 03 Vande Vate

The Best Answer


3

The best answer yields $60,000*2 + $50,000*1= $170,000


Node 6: Lower Branch Bands 1

Yields $170,000

Coils
1 2 3 4

Bands
11

15.057 Spring 03 Vande Vate

Rounding Fails
3

IP Solution Coils Limit


2

X
Production Capacity LP Solution

X
Coils
1 2 3

Bands Limit Bands


12

15.057 Spring 03 Vande Vate

Branch & Bound


Implicit enumeration An Integer Program with 30 binary variables can require over 1 billion nodes! Comes with guarantees The answer is no worse than And no better than

Typically finds a good answer quickly Spends a long time guaranteeing it


15.057 Spring 03 Vande Vate 13

A Solver Model
Steel Cable Company
Integer Programming Example Products Hours/Ton Profit/Ton Production Plan Products Tons Produced Maximum Sales Bands Coils 3 4 $ 50,000 $ 60,000

Bands 0 3

Coils 0 2 Hours Available 12 Total $ -

Production Constraint Hours Used 0 <= Profit Summary Bands $ Coils $ -

15.057 Spring 03 Vande Vate

14

Integer Variables
Integer: -2, -1, 0, 1, 2, Binary: 0 or 1 Binary: Yes or No, On or Off

15.057 Spring 03 Vande Vate

15

Single Sourcing
Assign each customer to only one DC Simplifies service network

15.057 Spring 03 Vande Vate

16

Old Ramp Allocation Southern US

Dealers sourced by multiple ramps

New Ramp Allocation Southern US

Dealers sourced by single ramps

Lower Cost?
Does Single-Sourcing Reduce/Increase distribution cost?

15.057 Spring 03 Vande Vate

19

Modeling Single-Sourcing
Vanilla Approach
set DCS; set CUSTOMERS; param Capacity{DCS}; param Demand{CUSTOMERS}; param Cost{DCS, CUSTOMERS}; var Assign{DCS, CUSTOMERS} binary; minimize TotalCost:
sum{dc in DCS, cust in CUSTOMERS} Cost[dc, cust]*Assign[dc, cust] + Other Costs
15.057 Spring 03 Vande Vate 20

DC gets all the Demand


s.t. SingleSource{cust in CUSTOMERS}:
sum{dc in DCS} Assign[dc, cust] = 1;

s.t. ObserveCapacity{dc in DCS}:


sum{cust in CUSTOMERS} Demand[cust]*Assign[dc,cust] <= Capacity[dc];

15.057 Spring 03 Vande Vate

21

Or Part of a Larger Model


s.t. MeetDemand{dc in DCS}:
sum of flows into the dc = sum{cust in CUSTOMERS} Demand[cust]*Assign[dc,cust];

15.057 Spring 03 Vande Vate

22

Whats the Problem?

Size!
Thousands of customers Scores of DCs Hundreds of thousands of Integer Variables Most useless! Assign a Customer in ME to a DC in CA?
15.057 Spring 03 Vande Vate 23

Practical Solution
Dont naively include all possible assignments Only include those:
Within a specified distance Among the N closest .

15.057 Spring 03 Vande Vate

24

Fixed Costs Revisited


Steco's Warehouse Location Model
Unit Costs Warehouse A B C Lease Unit Cost/Truck to Sales District 2 3 4 ($) 1 $ 7,750 $ 170 $ 40 $ 70 $ 160 $ 4,000 $ 150 $ 195 $ 100 $ 10 $ 5,500 $ 100 $ 240 $ 140 $ 60 Monthly Trucks From/To Decisions Yes/No Lease A 0 Lease B 0 Lease C 0 Total TrucksTo Demand (Trucks/Mo) Lease Cost $ $ $ $ 1 0 0 0 0 100 2 0 0 0 0 90 3 0 0 0 0 110 4 0 0 0 0 60 Total 0 0 0 Eff. Cap. 0 0 0 Cap. 200 250 300

A B C Totals

To 1 $$$$-

To 2 $$$$-

To 3 $$$ 0 $ 0

Truck To 4 $ $$ $ (0) $ (0) $$ 0 $ (0) $ 0

Total Cost $ $ (0) $ 0 $ 0

25

Binary Switches
Variables
Yes/No: Lease warehouse or not (Binary) Shipments from Warehouses to Sales Districts

Constraints:
Meet Demand Dont exceed Leased Capacity
Yes/No leased Warehouse *Capacity of Warehouse At Warehouse B: Effective Capacity = 0*250 At Warehouse A: Effective Capacity = 1*200

Yes/No must be binary


15.057 Spring 03 Vande Vate 26

Objective
Variable Cost for Trucks
$100/truck from Warehouse C to District 1

Fixed Costs for Leasing


$5,500 if we lease Warehouse C $5,500*Yes/No Lease Warehouse C

Careful about combining operational costs and capital costs


15.057 Spring 03 Vande Vate 27

An AMPL Model
set WAREHOUSES; param Capacity{WAREHOUSES}; param Lease{WAREHOUSES}; set DISTRICTS; param Demand{DISTRICTS}; param TruckCost{WAREHOUSES, DISTRICTS};

15.057 Spring 03 Vande Vate

28

var Open{WAREHOUSES} binary; var Ship{WAREHOUSES, DISTRICTS} >= 0; minimize TotalCost: sum{w in WAREHOUSES} Lease[w]*Open[w] + sum{w in WAREHOUSES, d in DISTRICTS} TruckCost[w,d]*Ship[w,d]; s.t. MeetDemand{d in DISTRICTS}: sum{w in WAREHOUSES} Ship[w,d] >= Demand[d]; s.t. ObserveEffectiveCapacity{w in WAREHOUSES}: sum{d in DISTRICTS} Ship[w,d] <= Capacity[w]*Open[w];

15.057 Spring 03 Vande Vate

29

Challenge: Challenge 2 Revisited


At least 5 funds At most 10 funds At least 10% if any
Fund Ratings Fund Name
Fidelity Adv Equity Fidelity Advisor Gro Fidelity Equity-Income Fidelity Equity Income-II Fidelity Growth/Income Fidelity Ins Cash Po Fidelity Investment Fidelity Intermediat Fidelity Limited Ter Fidelity Mortgage Se Fidelity Retirement Fidelity Short-Term Fidelity Value Fund Fidelity Worldwide F Totals 0.00 0.00 0.00 0.00 0.00 0.43 0.16 0.00 0.00 0.00 0.25 0.00 0.00 0.16 1.00 T-Bill 7 0 0 0 2 100 0 13 5 53 0 44 0 0 43.00 43 43.00 Large Value 71 48 60 66 47 0 0 0 18 0 8 0 50 27 6.24 3 3.24 3.00 3.24 Large Growth 2 5 5 4 0 0 0 0 0 0 35 0 5 0 8.75 3 5.75 3.00 5.75 Small Value 6 26 20 16 17 0 2 0 0 0 24 0 31 14 8.52 5 3.52 5.00 3.52 Small Growth 7 7 0 0 11 0 0 0 0 0 16 0 1 0 4.00 4 4.00 Japan 2 0 3 2 3 0 0 0 0 2 1 0 4 11 1.98 10 8.02 10.00 8.02 Pacific 0 0 0 1 0 0 0 0 4 1 0 0 0 0 2 2.00 2.00 2.00 Europe 0 11 0 0 5 0 0 0 0 3 3 0 8 37 6.56 5 1.56 5.00 1.56 Emerging Markets 5 2 3 6 2 0 4 0 0 0 11 6 2 11 5.13 10 4.87 10.00 4.87 Government 0 0 0 0 0 0 92 83 45 34 0 25 0 0 15.00 15 15.00 High Yield 0 0 9 5 12 0 1 0 28 7 0 23 0 0 0.16 0 0.16 (0.00) 0.16 International Bonds 0 0 0 0 0 0 0 3 0 0 0 3 0 0 0 0.00 (0.00) 0.00

Targets
Excess ShortFall Adjusted Deviation

29.29

Summary
var Invest{FUNDS} binary; var Fraction{FUNDS} >= 0;

s.t. MaxHoldings;
sum{f in FUNDS} Invest[f] <= 10;

s.t. MinHoldings;
sum{f in FUNDS} Invest[f] >= 5; s.t. MinimumPercentage{f in FUNDS}: Fraction[f] >= 0.10*Invest[f];

...

15.057 Spring 03 Vande Vate

31

Dont Forget
S.t. DefineInvest{f in FUNDS}:
Invest[f] >= Fraction[f];

Cant put money in fund unless you admit to being invested in it.

15.057 Spring 03 Vande Vate

32

Set Covering Models


F Seattle F

F F F F F Denver F F Los Angeles F F F F F F FF Houston F New Orleans F F F F Atlanta F F F F Chicago F F F F F F F F Pittsburgh F F F F New Y ork F

Boston

San FranciscoF F F

4
F

F F

WesternAir
Western Airlines Hub Selection Mile 1000 CH 674 1,005 1,008 1,067 2,054 912 802 452 1,390 2,142 2,013 DE 1,398 1,949 1,008 1,019 1,059 1,273 1,771 1,411 504 1,235 1,307 HO 789 1,804 1,067 1,019 1,538 356 1,608 1,313 1,438 1,912 2,274 LA 2,182 2,979 2,054 1,059 1,538 1,883 2,786 2,426 715 379 1,113 NO 479 1,507 912 1,273 356 1,883 1,311 1,070 1,738 2,249 2,574 NY 841 222 802 1,771 1,608 2,786 1,311 368 2,182 2,934 2,815 PI 687 574 452 1,411 1,313 2,426 1,070 368 1,826 2,578 2,465 SL 1,878 2,343 1,390 504 1,438 715 1,738 2,182 1,826 752 836 SF 2,496 3,095 2,142 1,235 1,912 379 2,249 2,934 2,578 752 808 SE 2,618 2,976 2,013 1,307 2,274 1,113 2,574 2,815 2,465 836 808 Distance Matrix AT BO 1,037 AT BO 1,037 CH 674 1,005 DE 1,398 1,949 HO 789 1,804 LA 2,182 2,979 NO 479 1,507 NY 841 222 PI 687 574 SL 1,878 2,343 SF 2,496 3,095 SE 2,618 2,976

15.057 Spring 03 Vande Vate

34

Summary
Mixed Integer Programming Models
Mostly about Binary Variables (Logic) Significantly Harder to solve

Significantly More Modeling Power


Fixed Costs If-Then Constraints Cardinality Constraints Set Covering Models .

15.057 Spring 03 Vande Vate

35

You might also like