fast reoptimization algorithms(1,3) 08:15 (5,1) 2 a tour-variable model contains a variable for each...

99
Fast Reoptimization Algorithms for Multi-Elevator Systems Jörg Rambau Konrad-Zuse-Zentrum für Informationstechnik Berlin Problem Description: Elevator Group Control Efficiency of Reoptimization: Modeling, Algorithm, Result Effectiveness of Reoptimization: Simulation Results Conclusion Joint work with Philipp Friese Based on joint work with Benjamin Hiller, Sven O. Krumke, Luis M. Torres (ADAC team)

Upload: others

Post on 24-Sep-2020

0 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

Fast Reoptimization Algorithmsfor Multi-Elevator Systems

Jörg Rambau

Konrad-Zuse-Zentrum für Informationstechnik Berlin

. Problem Description: Elevator Group Control

. Efficiency of Reoptimization: Modeling, Algorithm, Result

. Effectiveness of Reoptimization: Simulation Results

. Conclusion

Joint work withPhilipp Friese

Based on joint work withBenjamin Hiller, Sven O. Krumke, Luis M. Torres(ADAC team)

Page 2: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

PROBLEM DESCRIPTION

THE PRACTITIONER’ S V IEW

Page 3: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

PROBLEM DESCRIPTION

THE PRACTITIONER’ S V IEW

Real System:

Task:

Online aspect:

Realtime aspect:

Currently:

Problem:

Page 4: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

PROBLEM DESCRIPTION

THE PRACTITIONER’ S V IEW

Real System:pallet transportation Herlitz PBS AG,

Falkensee near Berlin

Task:

Online aspect:

Realtime aspect:

Currently:

Problem:

Page 5: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

PROBLEM DESCRIPTION

THE PRACTITIONER’ S V IEW

Real System:pallet transportation Herlitz PBS AG,

Falkensee near Berlin

Task: requests→ elevators, scheduling

minimize flow times/empty moves

Online aspect:

Realtime aspect:

Currently:

Problem:

Page 6: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

PROBLEM DESCRIPTION

THE PRACTITIONER’ S V IEW

Real System:pallet transportation Herlitz PBS AG,

Falkensee near Berlin

Task: requests→ elevators, scheduling

minimize flow times/empty moves

Online aspect:future requests unknown

Realtime aspect:

Currently:

Problem:

Page 7: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

PROBLEM DESCRIPTION

THE PRACTITIONER’ S V IEW

Real System:pallet transportation Herlitz PBS AG,

Falkensee near Berlin

Task: requests→ elevators, scheduling

minimize flow times/empty moves

Online aspect:future requests unknown

Realtime aspect:response time≤ 1s

Currently:

Problem:

Page 8: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

PROBLEM DESCRIPTION

THE PRACTITIONER’ S V IEW

Real System:pallet transportation Herlitz PBS AG,

Falkensee near Berlin

Task: requests→ elevators, scheduling

minimize flow times/empty moves

Online aspect:future requests unknown

Realtime aspect:response time≤ 1s

Currently: choice between:

FIFO oderNEARESTNEIGHBOR

Problem:

Page 9: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

PROBLEM DESCRIPTION

THE PRACTITIONER’ S V IEW

Real System:pallet transportation Herlitz PBS AG,

Falkensee near Berlin

Task: requests→ elevators, scheduling

minimize flow times/empty moves

Online aspect:future requests unknown

Realtime aspect:response time≤ 1s

Currently: choice between:

FIFO oderNEARESTNEIGHBOR

Problem: FIFO:

inefficient

Page 10: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

PROBLEM DESCRIPTION

THE PRACTITIONER’ S V IEW

Real System:pallet transportation Herlitz PBS AG,

Falkensee near Berlin

Task: requests→ elevators, scheduling

minimize flow times/empty moves

Online aspect:future requests unknown

Realtime aspect:response time≤ 1s

Currently: choice between:

FIFO oderNEARESTNEIGHBOR

Problem: FIFO:

inefficient

NEARESTNEIGHBOR:

forgotten pallets

Page 11: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

PROBLEM DESCRIPTION

REOPTIMIZATION: USE OPTIMAL SOLUTION FOR SNAPSHOTPROBLEM

Page 12: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

PROBLEM DESCRIPTION

REOPTIMIZATION: USE OPTIMAL SOLUTION FOR SNAPSHOTPROBLEM

Reoptimization:

Maintain dispatch to control server:

Page 13: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

PROBLEM DESCRIPTION

REOPTIMIZATION: USE OPTIMAL SOLUTION FOR SNAPSHOTPROBLEM

Reoptimization:

Maintain dispatch to control server:

new event

Page 14: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

PROBLEM DESCRIPTION

REOPTIMIZATION: USE OPTIMAL SOLUTION FOR SNAPSHOTPROBLEM

Reoptimization:

Maintain dispatch to control server:

new event↓snapshot of available data

Page 15: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

PROBLEM DESCRIPTION

REOPTIMIZATION: USE OPTIMAL SOLUTION FOR SNAPSHOTPROBLEM

Reoptimization:

Maintain dispatch to control server:

new event↓snapshot of available data↓

offline-optimization w.r.t. aux.-problem

Page 16: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

PROBLEM DESCRIPTION

REOPTIMIZATION: USE OPTIMAL SOLUTION FOR SNAPSHOTPROBLEM

Reoptimization:

Maintain dispatch to control server:

new event↓snapshot of available data↓

offline-optimization w.r.t. aux.-problem↓dispatch

Page 17: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

PROBLEM DESCRIPTION

HOW GOOD IS EXACT REOPTIMIZATION?

Page 18: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

PROBLEM DESCRIPTION

HOW GOOD IS EXACT REOPTIMIZATION?

QUESTIONS OF STUDY:

Page 19: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

PROBLEM DESCRIPTION

HOW GOOD IS EXACT REOPTIMIZATION?

QUESTIONS OF STUDY:

IS REOPTIMIZATION EFFICIENT?

(IS REOPTIMIZATION REALTIME -COMPLIANT?)

Page 20: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

PROBLEM DESCRIPTION

HOW GOOD IS EXACT REOPTIMIZATION?

QUESTIONS OF STUDY:

IS REOPTIMIZATION EFFICIENT?

(IS REOPTIMIZATION REALTIME -COMPLIANT?)

IS REOPTIMIZATION EFFECTIVE?

(REOPTIMIZATION OF OBJECTIVE =⇒ GOOD LONG-TERM VALUE OF OBJECTIVE?)

Page 21: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

A TOUR-BASED INTEGERL INEAR PROGRAM

Page 22: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

A TOUR-BASED INTEGERL INEAR PROGRAM

The transport graph for one elevator.

Page 23: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

A TOUR-BASED INTEGERL INEAR PROGRAM

Some requests.

Page 24: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

A TOUR-BASED INTEGERL INEAR PROGRAM

08:03

07:45

08:15 Their time stamps.

Page 25: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

A TOUR-BASED INTEGERL INEAR PROGRAM

08:03

07:45

08:15 The position of the elevator.

Page 26: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

A TOUR-BASED INTEGERL INEAR PROGRAM

08:03

08:15

07:45

(4,6)

Each request can be seen as a node.

Page 27: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

A TOUR-BASED INTEGERL INEAR PROGRAM

08:15

07:45

(4,6)

08:03

(1,3)

Each request can be seen as a node.

Page 28: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

A TOUR-BASED INTEGERL INEAR PROGRAM

07:45

(4,6)

08:03

(1,3)

08:15

(5,1)

Each request can be seen as a node.

Page 29: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

A TOUR-BASED INTEGERL INEAR PROGRAM

07:45

(4,6)

08:03

(1,3)

08:15

(5,1)

2

The elevator is a special node.

Page 30: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

A TOUR-BASED INTEGERL INEAR PROGRAM

07:45

(4,6)

08:03

(1,3)

08:15

(5,1)

2

The connecting moves between requests . . .

Page 31: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

A TOUR-BASED INTEGERL INEAR PROGRAM

07:45

(4,6)

08:03

(1,3)

08:15

(5,1)

2

. . . are arcs . . .

Page 32: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

A TOUR-BASED INTEGERL INEAR PROGRAM

07:45

(4,6)

08:03

(1,3)

08:15

(5,1)

2

. . . whose weights are empty moves.

[Ascheuer 1996]

Page 33: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

A TOUR-BASED INTEGERL INEAR PROGRAM

1

3

122

0

1

5

07:45

(4,6)

08:03

(1,3)

08:15

(5,1)

2

A feasible dispatch

is atour through all nodes

starting at the elevator’s node

with precedence conditionson each floor.

Page 34: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

A TOUR-BASED INTEGERL INEAR PROGRAM

07:45

(4,6)

08:03

(1,3)

08:15

(5,1)

2

We do not show the arcs anymore

since they are implicitely given.

Page 35: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

A TOUR-BASED INTEGERL INEAR PROGRAM

5

07:45

(4,6)

08:03

(1,3)

08:15

(5,1)

2

If there is another elevator . . .

Page 36: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

A TOUR-BASED INTEGERL INEAR PROGRAM

5

07:45

(4,6)

08:03

(1,3)

08:15

(5,1)

2

. . . then a feasible solution is a

partitioning

of requests into tours

with precedence conditions

on each floor.

Page 37: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

A TOUR-BASED INTEGERL INEAR PROGRAM

5

x(R)=1

x(S)=1

07:45

(4,6)

08:03

(1,3)

08:15

(5,1)

2

A tour-variable model contains a variable for

each feasible tour of a server.

Page 38: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

A TOUR-BASED INTEGERL INEAR PROGRAM

5

x(R)=1

x(S)=1

07:45

(4,6)

08:03

(1,3)

08:15

(5,1)

2

A tour-variable model contains a variable for

each feasible tour of a server.

Problem:

astronomic number of variables

Page 39: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

A TOUR-BASED INTEGERL INEAR PROGRAM

5

x(R)=1

x(S)=1

07:45

(4,6)

08:03

(1,3)

08:15

(5,1)

2

A tour-variable model contains a variable for

each feasible tour of a server.

Problem:

astronomic number of variables

Solution:

dynamic column generation

Page 40: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

A TOUR-BASED INTEGERL INEAR PROGRAM

5

x(R)=1

x(S)=1

07:45

(4,6)

08:03

(1,3)

08:15

(5,1)

2

A tour-variable model contains a variable for

each feasible tour of a server.

Problem:

astronomic number of variables

Solution:

dynamic column generation

Precedence Conditions:

Good for us!

Page 41: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DYNAMIC COLUMN GENERATION

Page 42: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DYNAMIC COLUMN GENERATION

Maintain a feasiblereduced LP(RLP) with fewer variables. Then

Page 43: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DYNAMIC COLUMN GENERATION

Maintain a feasiblereduced LP(RLP) with fewer variables. Then

1. Solve (RLP)

Page 44: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DYNAMIC COLUMN GENERATION

Maintain a feasiblereduced LP(RLP) with fewer variables. Then

1. Solve (RLP)

2. Extractdual values

Page 45: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DYNAMIC COLUMN GENERATION

Maintain a feasiblereduced LP(RLP) with fewer variables. Then

1. Solve (RLP)

2. Extractdual values

3. Variables (columns) withnegative reduced costs?

Page 46: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DYNAMIC COLUMN GENERATION

Maintain a feasiblereduced LP(RLP) with fewer variables. Then

1. Solve (RLP)

2. Extractdual values

3. Variables (columns) withnegative reduced costs?

No: Return current solutionandstop.

Page 47: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DYNAMIC COLUMN GENERATION

Maintain a feasiblereduced LP(RLP) with fewer variables. Then

1. Solve (RLP)

2. Extractdual values

3. Variables (columns) withnegative reduced costs?

No: Return current solutionandstop.

Yes: Add all/some/one such variable(s) to (RLP) andrepeat

Page 48: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DYNAMIC COLUMN GENERATION

Maintain a feasiblereduced LP(RLP) with fewer variables. Then

1. Solve (RLP)

2. Extractdual values

3. Variables (columns) withnegative reduced costs?

No: Return current solutionandstop.

Yes: Add all/some/one such variable(s) to (RLP) andrepeat

I F USED “C OTS”:

Page 49: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DYNAMIC COLUMN GENERATION

Maintain a feasiblereduced LP(RLP) with fewer variables. Then

1. Solve (RLP)

2. Extractdual values

3. Variables (columns) withnegative reduced costs?

No: Return current solutionandstop.

Yes: Add all/some/one such variable(s) to (RLP) andrepeat

I F USED “C OTS”:

SLOW CONVERGENCEIN THE BEGINNING =⇒ NOT REAL-TIME COMPLIANT

Page 50: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DYNAMIC PRICING CONTROL (ADAC-PROJECT)

Page 51: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DYNAMIC PRICING CONTROL (ADAC-PROJECT)

Tours can be searched in aprefix tree.

Page 52: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DYNAMIC PRICING CONTROL (ADAC-PROJECT)

The root corresponds to the

“don’t-move-tour”.

Page 53: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DYNAMIC PRICING CONTROL (ADAC-PROJECT)

The next level corresponds to

all tours visiting one request.

Page 54: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DYNAMIC PRICING CONTROL (ADAC-PROJECT)

The next level corresponds to

all tours visiting two requests.

Page 55: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DYNAMIC PRICING CONTROL (ADAC-PROJECT)

And so forth . . .

Page 56: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DYNAMIC PRICING CONTROL (ADAC-PROJECT)

. . . until all request sequences

have been enumerated.

Page 57: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DYNAMIC PRICING CONTROL (ADAC-PROJECT)

Task:

find (all) paths in prefix tree

that have negative/minimum reduced

cost.

Problem:

in the beginning,

input data (dual prices) far off.

Page 58: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DYNAMIC PRICING CONTROL (ADAC-PROJECT)

Ingredients of

Dynamic Pricing Control:

Dynamic Size Of Search Space

Dynamic Variation Of Search Space

Dynamic Threshold For Reduced Cost

Many Iterations Of (RLP)-Runs

Page 59: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DYNAMIC PRICING CONTROL (ADAC-PROJECT)

Restriction ofdegreeanddepth:

degree1, depth1.

If columns found, reoptimize (RLP).

Page 60: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DYNAMIC PRICING CONTROL (ADAC-PROJECT)

Restriction ofdegreeanddepth:

degree1, depth2.

If columns found, reoptimize (RLP).

Page 61: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DYNAMIC PRICING CONTROL (ADAC-PROJECT)

Restriction ofdegreeanddepth:

degree1, depth3.

If columns found, reoptimize (RLP).

Page 62: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DYNAMIC PRICING CONTROL (ADAC-PROJECT)

Restriction ofdegreeanddepth:

degree1, depth4.

If columns found, reoptimize (RLP).

Page 63: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DYNAMIC PRICING CONTROL (ADAC-PROJECT)

Restriction ofdegreeanddepth:

degree2, depth1.

We greedily select the

2 best next requests.

If columns found, reoptimize (RLP).

Page 64: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DYNAMIC PRICING CONTROL (ADAC-PROJECT)

Restriction ofdegreeanddepth:

degree2, depth2.

By changing the2-best-next criterion

we obtain variations in search space.

If columns found, reoptimize (RLP).

Page 65: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DYNAMIC PRICING CONTROL (ADAC-PROJECT)

Restriction ofdegreeanddepth:

degree2, depth3.

By changing the2-best-next criterion

we obtain variations in search space.

If columns found, reoptimize (RLP).

Page 66: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DYNAMIC PRICING CONTROL (ADAC-PROJECT)

Restriction ofdegreeanddepth:

degree2, depth4.

Usually, we stop increasing depth

when no new columns are found.

If columns found, reoptimize (RLP).

Page 67: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DYNAMIC PRICING CONTROL (ADAC-PROJECT)

Restriction ofdegreeanddepth:

degree2, depth4.

Usually, we stop increasing depth

when no new columns are found.

If columns found, reoptimize (RLP).

Crucial:

Extremely fast Reoptimization of

(RLP)

CPLEX 8.x

Page 68: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DUMMY CONTRACTOR

Page 69: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DUMMY CONTRACTOR

Problem: Too many requests/server=⇒ long tours in optimum=⇒ DPC insufficient

Page 70: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DUMMY CONTRACTOR

Problem: Too many requests/server=⇒ long tours in optimum=⇒ DPC insufficient

Rescue:Dummy contractor: service after fixed delay

Page 71: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DUMMY CONTRACTOR

Problem: Too many requests/server=⇒ long tours in optimum=⇒ DPC insufficient

Rescue:Dummy contractor: service after fixed delay

Effect: Cut-off of tail requests in dispatch

Page 72: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DUMMY CONTRACTOR

Problem: Too many requests/server=⇒ long tours in optimum=⇒ DPC insufficient

Rescue:Dummy contractor: service after fixed delay

Effect: Cut-off of tail requests in dispatch

Benefit: Tour lengths in optimum predictable

Page 73: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

DUMMY CONTRACTOR

Problem: Too many requests/server=⇒ long tours in optimum=⇒ DPC insufficient

Rescue:Dummy contractor: service after fixed delay

Effect: Cut-off of tail requests in dispatch

Benefit: Tour lengths in optimum predictable

RESULT:

EFFICIENT FOR ALL MODELS WITH EFFECTIVE TIME WINDOWS FOR REQUESTS.

Page 74: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

FEATURES OFIMPLEMENTATION

Page 75: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

FEATURES OFIMPLEMENTATION

. Configurable runtime bound

Page 76: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

FEATURES OFIMPLEMENTATION

. Configurable runtime bound

. Parametrized objective function consisting of terms for

Page 77: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

FEATURES OFIMPLEMENTATION

. Configurable runtime bound

. Parametrized objective function consisting of terms for

. time window violations for requests

Page 78: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

FEATURES OFIMPLEMENTATION

. Configurable runtime bound

. Parametrized objective function consisting of terms for

. time window violations for requests

. time window violations for servers

Page 79: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

FEATURES OFIMPLEMENTATION

. Configurable runtime bound

. Parametrized objective function consisting of terms for

. time window violations for requests

. time window violations for servers

. server movements

Page 80: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

FEATURES OFIMPLEMENTATION

. Configurable runtime bound

. Parametrized objective function consisting of terms for

. time window violations for requests

. time window violations for servers

. server movements

. service

Page 81: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

FEATURES OFIMPLEMENTATION

. Configurable runtime bound

. Parametrized objective function consisting of terms for

. time window violations for requests

. time window violations for servers

. server movements

. service

. Various reoptimization strategies includingREPLAN andIGNORE

Page 82: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

FEATURES OFIMPLEMENTATION

. Configurable runtime bound

. Parametrized objective function consisting of terms for

. time window violations for requests

. time window violations for servers

. server movements

. service

. Various reoptimization strategies includingREPLAN andIGNORE

. Start heuristics (BESTINSERTION, 2OPT)

Page 83: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

FEATURES OFIMPLEMENTATION

. Configurable runtime bound

. Parametrized objective function consisting of terms for

. time window violations for requests

. time window violations for servers

. server movements

. service

. Various reoptimization strategies includingREPLAN andIGNORE

. Start heuristics (BESTINSERTION, 2OPT)

. Embedded in simulation environment

Page 84: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

RESULT

Page 85: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFICIENCY OF REOPTIMIZATION: MODELING, ALGORITHM, RESULT

RESULT

SOUND BITES :

OPTIMAL SOLUTIONS WITH PROOF IN 100MS FOR UP TO 7 REQUESTSPER SERVER

REASONABLE SOLUTION AFTER 5MS

(EFFECTIVE TIME WINDOWS NECESSARY)

Page 86: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFECTIVENESSOF REOPTIMIZATION: SIMULATION RESULTS

SIMULATION ENVIRONMENT

Page 87: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFECTIVENESSOF REOPTIMIZATION: SIMULATION RESULTS

SIMULATION ENVIRONMENT

AMSEL Simulation Library

[Ascheuer]

Elevator Simulation

[Rambau]

Herlitz System

[Hauptmeier, Müller, Rambau]

Multi-Capacity Elevators

[Rambau, Weider]

Multiple Elevators

[Friese, Rambau]

Page 88: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFECTIVENESSOF REOPTIMIZATION: SIMULATION RESULTS

INVESTIGATED ONLINE-ALGORITHMS

Whenever a new requests occurs, dispatch it so as to . . .

GREEDYDRIVE greedily minimize empty moves (no reopimization)

REOPTDRIVE minimize empty moves plusε-lateness (complete reoptimization)

XLOADDRIVE dto. with squared lateness and Dummy Contractor (complete reoptimization)

GREEDYLATE greedily minimize lateness (no reoptimization)

REOPTLATE minimize lateness plusε-empty-moves (complete reoptimization)

XLOADLATE dto. with squared lateneess and Dummy Contractor (complete reoptimization)

REOPTMIX minimize lateness plus empty-moves (complete reoptimization)

XLOADMIX dto. with squared lateness and Dummy Contractor (complete reoptimization)

Page 89: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFECTIVENESSOF REOPTIMIZATION: SIMULATION RESULTS

SIMULATION RESULTS (VARIOUS LOADS, 10 SAMPLES À 24H EACH)

Page 90: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFECTIVENESSOF REOPTIMIZATION: SIMULATION RESULTS

SIMULATION RESULTS (VARIOUS LOADS, 10 SAMPLES À 24H EACH)

SOUND BITES :

Page 91: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFECTIVENESSOF REOPTIMIZATION: SIMULATION RESULTS

SIMULATION RESULTS (VARIOUS LOADS, 10 SAMPLES À 24H EACH)

SOUND BITES :

GREEDILY M INIMIZING EMPTY MOVES IS UNACCEPTABLE

Page 92: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFECTIVENESSOF REOPTIMIZATION: SIMULATION RESULTS

SIMULATION RESULTS (VARIOUS LOADS, 10 SAMPLES À 24H EACH)

SOUND BITES :

GREEDILY M INIMIZING EMPTY MOVES IS UNACCEPTABLE

GREEDILY M INIMIZING LATENESSIS GOOD

Page 93: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFECTIVENESSOF REOPTIMIZATION: SIMULATION RESULTS

SIMULATION RESULTS (VARIOUS LOADS, 10 SAMPLES À 24H EACH)

SOUND BITES :

GREEDILY M INIMIZING EMPTY MOVES IS UNACCEPTABLE

GREEDILY M INIMIZING LATENESSIS GOOD

REOPTIMIZATION DELIVERS BEST CORR. LONG-TERM OBJECTIVE

Page 94: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

EFFECTIVENESSOF REOPTIMIZATION: SIMULATION RESULTS

SIMULATION RESULTS (VARIOUS LOADS, 10 SAMPLES À 24H EACH)

SOUND BITES :

GREEDILY M INIMIZING EMPTY MOVES IS UNACCEPTABLE

GREEDILY M INIMIZING LATENESSIS GOOD

REOPTIMIZATION DELIVERS BEST CORR. LONG-TERM OBJECTIVE

M INIMIZING SQUARED LATENESSDELIVERS QOS-STABILIZATION

Page 95: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

CONCLUSION

EXPERIMENTAL EVIDENCE :

Page 96: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

CONCLUSION

EXPERIMENTAL EVIDENCE :

REOPTIMIZATION IS EFFICIENT AND EFFECTIVE

FOR ONLINE GROUPCONTROL OF TRANSPORTELEVATORS

Page 97: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

CONCLUSION

EXPERIMENTAL EVIDENCE :

REOPTIMIZATION IS EFFICIENT AND EFFECTIVE

FOR ONLINE GROUPCONTROL OF TRANSPORTELEVATORS

STILL M ISSING:

Page 98: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

CONCLUSION

EXPERIMENTAL EVIDENCE :

REOPTIMIZATION IS EFFICIENT AND EFFECTIVE

FOR ONLINE GROUPCONTROL OF TRANSPORTELEVATORS

STILL M ISSING:

COMPUTABLE MATHEMATICAL PERFORMANCEGUARANTEES

=⇒ RESEARCHIN PROGRESS(DFG CENTER C6)

Page 99: Fast Reoptimization Algorithms(1,3) 08:15 (5,1) 2 A tour-variable model contains a variable for each feasible tour of a server. Problem: astronomic number of variables Solution: dynamic

JÖRG RAMBAU

CONCLUSION

EXPERIMENTAL EVIDENCE :

REOPTIMIZATION IS EFFICIENT AND EFFECTIVE

FOR ONLINE GROUPCONTROL OF TRANSPORTELEVATORS

STILL M ISSING:

COMPUTABLE MATHEMATICAL PERFORMANCEGUARANTEES

=⇒ RESEARCHIN PROGRESS(DFG CENTER C6)

THANK YOU FOR YOUR ATTENTION !