Systems design for scheduling: Open Tools Patrick De Causmaecker, Peter Demeester, Greet Vanden Berghe and Bart Verbeke KaHo Sint-Lieven, Gent, Belgium.

Systems design for scheduling: Open Tools

Patrick De Causmaecker, Peter Demeester, Greet Vanden Berghe and

Bart Verbeke

KaHo Sint-Lieven, Gent, Belgium

http://ingenieur.kahosl.be/projecten/dingo




Vakgroep IT KaHo Sint-Lieven 2

Overview

Introduction: distributed personnel scheduling Local optimisation Agent model Web service Conclusion


Introduction: problem description Most companies have different departments When departments > critical size => isolated islands

Almost no communication with other departments Keep department information private

This leads to sub optimal behaviour Oversimplified example:

Hospital consisting of different departments One department is understaffed due to illness, holidays,… Some are overstaffed Exchange of nurses solves both their problems


Introduction: problem description

Sub optimality is solved at the moment: negotiations between different heads of

department Time consuming activity

Goal of this research: Automating this negotiation


Final solution

Negotiation

Department 1

Tabu Search

1st local solution

Department 2

Tabu Search

1st local solution

Department 3

Tabu Search

1st local solution

Final solution

Final solution

Initial assignment of personnel

Our approach


Local optimisation Since we are developing a general tool applicable

in different domains/sectors: Need of general evaluation method that can evaluate

different constraints => Numbering evaluation method1

Evaluation mechanism implemented in Java Constraints are expressed in XML

1 E.K. Burke, P. De Causmaecker, S. Petrovic, G. Vanden Berghe: Fitness Evaluation for Nurse Scheduling Problems, Proceedings of Congress on Evolutionary Computation, CEC2001, Seoul, IEEE Press, 2001, p. 1139-1146.


Numbering: example Planning period: 1 week

soft constraints: Maximum number of assignments 6 Minimum 2 Maximum number consecutive free days 8 Minimum 2 Maximum number consecutive assignments 4 Minimum 2 Maximum number of shifts per day 1 Maximum number of night shifts 3 Minimum number consecutive night shifts 2 Work full weekends

from until M Morning shift 06:45 14:45 L Late shift 14:30 22:00 N Night shift 22:00 07:00

N1

N2

N3


Numbering: example

* * * * * N1 0 0 0 1 1 1 2 2 2 3 3 3 4 4 4 5 5 5 6 6 6 N2 U U 0 U U 1 U U 2 U U 3 U U 4 U U 5 U U 6 N3 U U U U U U U U U U U U U U U 0 0 0 1 1 1

roster P1

constraints N1

max-total = 6

min-total = 2

max-between = 8

min-between = 2

max-consecutive = 4

min-consecutive = 2

max-pert = 1

counters

last

total

consecutive

constant

max

constraints N2

max-total = 3

min-consecutive = 2

constraints N3

min-consecutive = 2

Mon Tue Wed Thu Fri Sat Sun P1 M M L L N


Evaluation algorithm Initialisation: taking into account the previous planning

map numberings onto the previous planning period search values for counters: initialise

Intermediate evaluation: adjust counters and calculate costs if necessary

Final evaluation: calculate costs at the end of the planning period

per iteration the algorithm is executed 1 x per person in the planning independently of the number and the type of soft constraints


Constraints N1 in XML<constraint ID="1">

<numberingId>1</numberingId><variables>

<max_total weight="1">6</max_total><min_total weight="1">2</min_total><max_pert weight="1">1</max_pert><min_pert weight="1">-infinity</min_pert><max_between weight="1">8</max_between><min_between weight="1">2</min_between><max_consecutive weight="1">4</max_consecutive><min_consecutive weight="1">2</min_consecutive>

</variables></constraint>…


<day ID="4"><shift ID="1">3</shift><shift ID="2">3</shift><shift ID="3">3</shift>

</day><day ID="5">

<shift ID="1">4</shift><shift ID="2">4</shift><shift ID="3">4</shift>

</day><day ID="6">


</day><day ID="7">


</day></numbering>…

Constraints N1 in XML…<numbering ID="1">

<day ID="1"><shift ID="1">0</shift><shift ID="2">0</shift><shift ID="3">0</shift>

</day><day ID="2">


</day><day ID="3">


</day>


Evaluation algorithm

Every constraint that can be expressed in the proposed XML format can be evaluated!

Local search algorithm: tabu search Use OpenTS Framework for implementation http://www.coin-or.org/OpenTS/

http://www.coin-or.org/OpenTS/


Non-distributed case

Local searchcomponent

WebserviceDepartment

webservice


Webservice

To guarantee openness: we provide a webservice that allows communication with the evaluation component

Webservice: implemented in Apache SOAP (http://ws.apache.org/soap/)

As a proof of concept: We developed a GUI that uses webservice to

generate solutions

http://ws.apache.org/soap/




Agents

Final solution

Negotiation

Task 1

Tabu Search

1st local solution

Task 2

Tabu Search

1st local solution

Task 3

Tabu Search

1st local solution

Final solution

Final solution

Initial assignment of personnel


Agent model Every department is represented by an agent Every member of personnel is represented by an

agent Note:

agents are not used in local search algorithm However they are used to (improve) negotiate about the

solution

Question: how organise the exchange of personnel among the departments?

Need of a coordination model


Coordination model

Contract Net Protocol

Announce a task

manager

contractors

Submit a bid

manager

contractors

manager

contractors

Grant contract


Agent Architecture (CNP)OmbudsAgent

DepartmentAgent

Moz

art/

Oz

Java

DepartmentAgent

DepartmentAgent

DepartmentAgent

Local search algorithm

XM

L-R

PC


Agent Architecture ( CNP)DepartmentAgent

Moz

art/

Oz

Java

XM

L-R

PC

EmployeeAgent

EmployeeAgent

Local search algorithm


Coordination Model

Blackboard: Demand/supply for personnel are written on a

black board Token Ring:

there is no boss, every department can ask a question per round


Department 1Department 2

Department 3Department 4

Negotiationserver

Distributed case


Webservice

evaluationcomponent


Webservice

evaluationcomponent


Webservice

evaluationcomponent


Webservice

evaluationcomponent


Communication Local search algorithm implemented in Java Agent negotiation mechanism implemented in

Mozart/Oz Communication:

Socket? Webservice?

Opt to use webservice: SOAP? Alternative: XML-RPC?


Conclusion Platform independent implementation of

personnel planning application that is useful for many domains

Open: Every constraint that can be expressed in

proposed XML format can be evaluated Methods can be called via web service

Tools will be available for download in November 2005 from http://ingenieur.kahosl.be/projecten/DINGO/

http://ingenieur.kahosl.be/projecten/DINGO/




Questions?

Thank you for your attention!


M L N

Solution structure

Example0|1|0||1|0|0||0|0|0||0|0|1||0|1|0||0|0|1||1|0|0|

0|0|1||1|0|0||1|0|0||1|0|0||0|0|1||0|1|0||0|0|1|

1|0|0||1|0|0||1|0|0||0|0|0||1|0|0||0|0|0||0|1|0|

1|0|0||1|0|0||0|0|0||1|0|0||0|0|0||0|1|0||0|1|0|

0|1|0||0|1|0||0|1|0||1|0|0||1|0|0||0|1|0||0|1|0|

1|0|0||0|0|0||0|0|0||0|1|0||1|0|0||0|1|0||1|0|0|


Moves

Flip move: 0 => 1 or 1 => 0 Swap move: swap 2 elements Row and column swap move


Ombuds Agent Department Agenti

CFP Every one sends most expensive cost + time

slot. Result of local search algorithm

(Cost, T)i

Evaluate every proposed change and

generate corresponding cost

ACCEPT PROPOSAL

REJECT PROPOSAL

Involved Department Agents exchange

personnel and adapt department timetable

for that shift

CNP

CFP

(Cost, T)i

Only these agents that have done changes

send their costs

CNP

Employee Agentj

send timetable to every personnel

member

CFP: Qmax, Tmax

Cost

ACCEPT PROPOSAL

Every involved agent evaluates

its own constraints and

generates a cost

REJECT PROPOSAL

Take highest cost

Send others a REJECT

Costi

ACCEPT PROPOSAL

REJECT PROPOSAL

CFP: Qmax, Tmax

CNP

If Costi < threshold, then change is

accepted

Otherwise not

Sends agents that have done changes

a new CFP

Systems design for scheduling: Open Tools Patrick De Causmaecker, Peter Demeester, Greet Vanden Berghe and Bart Verbeke KaHo Sint-Lieven, Gent, Belgium.

Documents