Modeling of an Agent System to Support the Management of

Cooperating and Rival Resources for Business Workﬂows

Agnes Werner-Stark, Tibor Dulai and Gyula

Abrah´am

Department of Electrical Engineering and Information Systems, University of Pannonia,

Egyetem str. 10., Veszpr´em, Hungary

Keywords:

Modeling, Agent, Workﬂow, Resource, Scheduling, Log.

Abstract:

Modeling and analysis of business workﬂows may be strategic on behalf of the optimal execution. This

paper proposes an innovative model-based approach, which can be used to resource scheduling of business

workﬂows. To this we deﬁned such functions that help the operation of processes, resource scheduling can

be described formally during the modeling. The system uses all the information during the scheduling, which

may be recorded in a log ﬁle in connection with the process execution. We can extract useful information

concerning allocation of resources by analysis of the historical data, which are used to assign the resources

to the implemented tasks. In the system the cooperation and contention of the resources as agents will play

important role. The scheduling can be tested in an agent simulation environment. By the aid of this approach

we can give decision proposal to the operator in real time to promote more optimal realization of the workﬂow.

1 INTRODUCTION

Increasingly, companies around the world are reengi-

neering their core business processes to be more prof-

itable and to improvecustomersatisfaction. Modeling

and analysis are two critical steps in any process re-

design effort. Bhaskar et al. (Bhaskar et al., 1994)dis-

cuss the need for simulation tools that can be used ef-

fectively to model, to document and to analysis busi-

ness processes. Tumay (Tumay, 1996) concentrates

also the simulation of business processes and how it

works. In (Ghosh, 2013), the authors deal with a sim-

ulation tool which makes the scheduling of service

tasks and the distributed human resources possible in

case of industrial process. The major pillars by the

side of which the model works and tries to minimize

the time costs of certain tasks: ﬂuctuation, short-term

service demands which depend on availability of the

resources, ﬂexible task performance, replacing of the

straggling behind tasks.

In our work the re-engineering and the simulation

are very important. Re-engineering can be followed

closely by the workﬂow process.

Scheduling is an important and widely used topic

of operations research. Besides of its theoretical

importance, industries can also beneﬁt from opti-

mal schedules and resource allocation. More and

more different algorithms are born for organizing pro-

cess elements on the time scale (Pinedo, 2012) and

(Brucker and Brucker, 2007), related to business pro-

cess (Barba and Del Valle, 2010) and (Xu et al.,

2010), or industrial processes (Sule, 2008). The re-

sults are usually represented on a Gantt chart that we

use also to illustrate the decision proposal.

Only some publications investigate the coopera-

tion possibilities that are enabled by the similar func-

tionality of resources in scheduling, see e.g., (Murthy

et al., 1997) or (Dulai and Werner-Stark, 2012).

Merdan et al. (Merdan et al., 2013) consider the

application of intelligent agent-based technology as

a promising tool to improve system performance in

complex and changeable environments. They exam-

ine it in the case of unforeseen events, e.g., machine

breakdowns that usually leads to a deviation from the

initial production schedule. A multi-agent approach

can be used to enhance system ﬂexibility and robust-

ness. In their work the approach to revise and re-

optimize the dynamic system schedule in response

to unexpected events is proposed. Our goal is also

the modeling of a special multi-agent system with the

help of which - detecting workﬂow changes and ana-

lyzing historical data - a new decision proposal can be

determined referring to a new allocation of resources

as the modiﬁcation of workﬂow.

So that, we can specify accurately the decision

proposal, we used a further method. As under the

407

Werner-Stark Á., Dulai T. and Ábrahám G..

Modeling of an Agent System to Support the Management of Cooperating and Rival Resources for Business Workﬂows.

DOI: 10.5220/0005001404070412

In Proceedings of the 4th International Conference on Simulation and Modeling Methodologies, Technologies and Applications (SIMULTECH-2014),

pages 407-412

ISBN: 978-989-758-038-3

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

current process of workﬂows the most important in-

formation is stored in a log ﬁle, therefore the current

state analysis can be carried out using these logs and

in case of unexpected events the multi-agent system

can react with a new resource scheduling proposal.

Van der Aalst et al. play an important role in area of

log analysis see e.g., (van der Aalst et al., 2003) and

(van der Aalst et al., 2007).

This paper builds on concepts from workﬂow

management, agent based technology, log analysis

and resources scheduling. We have modeled a new

solution to the resource scheduling of workﬂows in

random changing execution environment. This paper

describes the modeling phase of an agent oriented re-

source scheduler which is based on log analysis. In

our modeled method the role of cooperation and com-

petition of the resources as agents plays an important.

By the aid of this approach we can give decision pro-

posal in real time to the operator, promoting more op-

timal realization of the workﬂow.

2 BASIC NOTIONS

2.1 Relationship between Workﬂow and

Agent

Business processes are modularized and these mod-

ules can be reorganized by the workﬂowsto form new

business processes, so as to react quickly to unex-

pected changing business needs and conditions. The

user in the workﬂowsystem does not haveto deal with

the structure of the business process, but only con-

centrates on the task itself. The current generation of

workﬂow management systems has some shortcom-

ings, e.g.:

• Lack of Automation: workﬂow systems only de-

termine the process logic, but most of the activi-

ties are still fulﬁlled by human. The workﬂowsys-

tem can’t even start a workﬂow without human’s

intervention.

• Lack of Reactivity: the workﬂow systems require

a pre-deﬁned representation of a business process

and all potential deviations from that process.

• Lack of Resources Management: the workﬂow

systems do not control the resources of a business

process.

We want to solve these problems by using a new

agent-based approach. Agents can represent re-

sources. Task assignments and resource allocations

are done through negotiation among these agents.

Agents react to changing circumstances and have the

ability to generate alternative execution paths. The

agents are independent to each other and each of them

is responsible for process execution.

2.2 The Function of the Agent

Technology in the Scheduling

The beneﬁts of applying agent technology to business

process management include:

• Automation: The inherent autonomy of software

agents can fulﬁll activities as human substitution.

• Resource Management: Agents can represent re-

sources. Task assignments and resource alloca-

tions are done through negotiation among these

agents.

• Reactivity: Agents react to changing circum-

stances and have the ability to generate alternative

execution paths.

• Intelligent Decision-making: We can use the

learning ability of agents that is helpful in the

workﬂow scheduling.

Our goal is that we integrate these beneﬁts in special

solving to a support system to organize cooperating

and rival resources.

2.3 Log Analysis

Process mining techniques have emerged in the area

of business process modeling (van der Aalst et al.,

2004), theoretical computer science and artiﬁcial in-

telligence. Taking the data set of real process execu-

tions, the event logs, these techniques can be used for

process discovery, i.e., to construct a discrete event

model from them, or for conformance checking. An

event log is a set of ﬁnite event sequences, whereas

each event sequence corresponds to a particular mate-

rialization of the process. The execution of each event

needs some resources which execute the activity in

optimal time.

2.4 The Structure of the System

Figure 1 shows the structure of the multi-agent sys-

tem for scheduling business workﬂows. To this we

modeled a method for supporting of the user deci-

sions, the decision support (DS) part of it we detail

in this paper. The task of the agent simulation system

is to simulate the remainder of the process based on

the current state of the process and to give recommen-

dations for situations where it needs to decide on the

resource assignment. Thus, the agent system speciﬁes

that in the course of simulation which choice results

SIMULTECH2014-4thInternationalConferenceonSimulationandModelingMethodologies,Technologiesand

Applications

408

Figure 1: The full system architecture.

the best execution. We can use this information for

determining the possibility of the cooperation among

agents on behalf of reduce costs and execution time.

We would like to use a DS system which gets a work-

ﬂow from the main system (e.g., in xpdl format ) and

the DS system needs to analyze this workﬂow using

event logs. Given the deadline of the process, the sys-

tem examines from a given point whether it can give a

new scheduling to the resources for reaching optimal

performance.

3 DEVELOPED MODELS

For the above mentioned problem set, we created a

method to carry out the scheduling. The work intends

to be the basis of several different tasks with different

goals (in some cases the total cost of the processes is

the critical parameter, while in other cases the opera-

tion time should be minimal). In this section we intro-

duce our approach to which some models are attached

and these models represent the function to contribute

to the decision making for the resource scheduling.

The model and its parameters are designed in such

a way, that it should be applied for different kinds of

scheduling problems.

For this reason we collected the necessary param-

eters that make possible the development of a general

framework which uses different kinds of resources

in different processes and is able to handle different

needs.

3.1 Agent Model

This model speciﬁes the characteristics of the agents

and is a reference point to the other models. The agent

identiﬁcation is based on a previous conceptualization

phase where actors and use cases were modeled.

R = {r

,...,r

}, is the set of resources;

A resource can be anything what does an opera-

tion on a product type in an operation mode in a given

time.

O(r

) = {o

,...,o

}, is the set of the operation

modes of resource i;

The same operation of the same resource on the

same product type can be carried out in different o-

peration modes, which may differ e.g., in their cost or

the time they required.

A = {a

,...,a

, pause(t)}, is the set of basic ac-

tivities, where pause(t) is an empty activity with

length of t hour;

Each basic activity is an operation which can be

carried out on a product type by a resource in a given

operation mode. Each basic activity has a length

and a cost which depends on the previously men-

tioned parameters, too (product type, resource, oper-

ation mode).

ra : R× A → {0,1}, is a function for determining

whether a resource is able to perform an activity;

More basic activities could be carried out by the

same resource and the same activities could be per-

formed by different resources.

rap : R× A× P → {0,1}, is a function for deter-

mining whether a special activity of a resource can be

applied for a product type;

ModelingofanAgentSystemtoSupporttheManagementof

CooperatingandRivalResourcesforBusinessWorkflows

409

This function is needed because it may happen,

that a special resource while carries out a basic ac-

tivity is not able to be used for a given product type

(e.g., the resource doesn’t have the physical dimen-

sions which is needed to put the given product type

into it).

t : R× A× P× O → N, provides the suggested op-

eration time of a resource in a given operation mode

performing a given activity on a given product type;

E.g., a test cabin may do the same test in 80 min-

utes with 70

◦

C temperature, while the same test takes

only 50 minutes on the same product type with the

same resource in case of 90

◦

C. The different temper-

atures represent different operation modes of the test

cabin.

c : R×A× P×O×N → N, provides the cost of an

activity of a resource in an operation mode on a given

product type with a given operation time;

This function works similar to function t().

In the model 4 storages are distinguished: Set of

resources, Set of activities, Set of products, Set of op-

eration modes.

3.2 Organization Model

This model represents the humming replace ability of

agents among one another.

We used the cooperation to scheduling because we

introduced a special vector, i.e., cooperation vector.

coop(r

, p

) = [h

,...,h

], is the coopera-

tion vector, where m is the number of resources, h

is a natural numbers for all 1 ≤ n ≤ m, r

is the i

resource, a

is a basic activity, p

is a product type,

is an operation mode of resource i, h

denotes

how much percent of the productivity of resource i is

needed in a given activity in a given operation mode

on a given product type to substitute totally resource

n supposing unchained operation time.

E.g., h

= 30 means that resource i is able to sub-

stitute resource j, moreover, resource i is able to sub-

stitute 3 pieces of resource j in the given operation

mode in the given activity on the given product type.

= 200 means that resource i is physically able to

substitute resource j, however, two pieces of resource

i is needed to substitute one piece of resource j in

the given operation mode in the given activity on the

given product type. h

= 0 means that resource i is

unable to substitute resource j in the given operation

mode in the given activity on the given product type.

3.3 Knowledge Model

We introduced some special functions:

q : R× A× P× O → [0,100], provides capacity in-

formation: how many percent of a resource capacity

is occupied by one piece of a given product type in a

given operation mode of a resource while performing

a given activity;

The occupation information is given in percent of

the resource’s total capacity because this parameter

can be used more universal in this way. E.g., a gas

tester during a normal gas test in normal operation re-

lated to a special pressure sensor as product type has

capacity 20%: it means that the gas tester can handle

5 pieces of the same product type at the same time;

however, next to 2 pieces of the special pressure sen-

sor the gas tester could handle one other type of prod-

uct, too, which needs at most 60% of the tester’s ca-

pacity.

e : R× A×P× O → [0, 1], provides the probability

of resource failure during performing a given activity

on a given product type in a given operation mode;

s : R×A×P×O → {0,1}, results in a binary deci-

sion: whether a resource activity in a given operation

mode on a given product type can be suspended with-

out restarting it from its beginning;

Assume a basic activity of a resource, which has 5

days operation time. Suppose, that after 3 days from

the start, the resource fails. The impact of the failure

depends on whether the basic activity of the resource

in the given operation mode on the given product type

was suspendable. When it was, the continuation of

the activity needs only 2 days. If not, it needs another

5 days. This parameter has importance not only in

case of device failure, but in normal operation, too,

e.g., in case of shared usage of a resource between

time-shifted activities.

rreq : R × A × P × O × DateTime → P(R,N)

(power set on pairs of a resource and a natural num-

ber), provides the additional resource need of a re-

source’s given activity on a given product type in a

given operation mode in a given hour;

Some resources require other resources, too, while

perform an activity. This additional need may depend

on the phase of the operation, too. E.g., a tempera-

ture shock test requires not only the tester device (the

tester has to be used during the whole duration of the

test), but a human helper, too. However, the human

operator is needed only in the ﬁrst and the last hours

of the 5 days long test.

bound(r

, p

) = o

, provides the impact

of circumstances based on the experiences, where r

the i

resource, a

is a basic activity, p

is a product

type, o

is an operation mode of resource i, and c

a regular expression on time and basic activity infor-

mation;

It is written in the form, that the ”normal” opera-

SIMULTECH2014-4thInternationalConferenceonSimulationandModelingMethodologies,Technologiesand

Applications

410

tion mode can be transformed into a ”modiﬁed” op-

eration mode because of the circumstances which de-

pend on the time and the preceding activities. E.g.,

a human may perform a work slower just after lunch

time or after a preceding heavy task. This factor inﬂu-

ences the most the human resources. Process mining

may serve the necessary information for determining

this function based on the experiences from past be-

haviors.

map : R × DateTime → N, provides the informa-

tion on accessibility of resources: how many resource

of a given type is accessible in a given hour;

F : R → N, provides the expected number of hours

how long a given resource is unavailable in case of its

failure;

This function may reﬂect the average time needed

to repair a resource type in case of technical failure.

This parameter can be used when schedules are

investigated from the viewpoint of failure-awareness.

3.4 Coordination Model

We can represent the communication and cooperation

of the software agents operating in the system by this

model. In the model three main agents can be seen:

AMS, Controllers and Task agents.

The Agent Management System (AMS) is an

agent, which operates built into the system. The AMS

realizes system services, any agents can make use of

which. Such services are, e.g., starting a new store,

starting a new agent, migration of agents between

stores, etc.

Controllers: In this group there are two agents.

The ﬁrst agent is the sentinel agent, the task of which

is, the communication with the external system, i.e.

taking over of data coming from there or sending data

from the decision support system. The second agent

will be responsible for the efﬁcient allocation of the

tasks – in the arriving process model – among re-

sources.

The resources may apply for each task in deﬁnite

time interval, then, when the timer expires; the winner

agent gets the right to perform this task.

Task agents: these agents represent the resources.

The task of these agents is the execution of the activi-

ties of the process model.

Each agent is connected to all other agents in the

system. The resources can send requests and answers

to one another or can receive from another. In the

following we will look at the operation of the more

importantcommunicationchannels by some sequence

diagrams.

The further coordination model describes the dy-

namic interactions between software agents by using

sequence diagrams to represent the conversations and

interventions carried out between them.

3.5 The Process

During the developmentphase of DS system, it is nec-

essary to simulate the system feasibility before the

formal implementation. Because of it, we develop a

process model to represent the behaviour of the DS

system and then to simulate the model performance.

Process = {proc

,..., proc

}, is the set of pro-

cesses;

Contains all the processes; the elements can be,

e.g., a test sequence of a company or a production

process.

P = {p

,..., p

}, is the set of product types;

We need product type because a resource can

handle the different product types in different ways

(for example needs different working time); however,

products belong to the same product type are handled

in the same way in a production process.

start

: Process → DateTime, shows the start time

of a process;

This parameter describes the time when the ﬁrst

basic activity of a process begins.

maxend

: Process → DateTime, shows the maxi-

mum ﬁnish time of a process;

This parameter is to describe the latest time when

a process has to be ﬁnished. It means a constraint in

scheduling: basic activities of a process can not be

pushed forward in time with arbitrary time units.

prev : Process × A × N

→ {A,

0}, provides the

prior basic activity of the n

occurrence of a given

basic activity in a given process;

This function is to describe the sequences of ba-

sic activities in a process. In our model processes are

built up by sequences of basic activities. No other

structure of them are allowed.

next : Process × A × N

→ {A,

0}, provides the

following basic activity of the n

occurrence of a

given basic activity in a given process;

This function is the opposite function of prev().

maxdelay : Process× A × N

→ N, provides the

maximal duration of time out, which is tolerated by

the n

occurrence of a given basic activity in a given

process just after its execution;

There may exist neighboring basic activities in a

process, which do not tolerate arbitrarily huge time

gap between them. E.g., during cooking tea, after

boiling the water we may not to wait too much be-

fore putting into the water the tea ﬁlter (anyway we

should repeat the previous process - the boiling).

dur : Process × A × N

→ Q, provides the time

scale, which number the default operation time of the

ModelingofanAgentSystemtoSupporttheManagementof

CooperatingandRivalResourcesforBusinessWorkflows

411

of the n

occurrence of a given basic activity in a

given process has to be multiplied with, for getting

the real operation time of the activity.

Users of the scheduling software may modify the

default schedule of different activities. E.g., a man-

ager - based on the special requirement of a customer

- chooses that the temperature shock test should last

7 days instead of the default 5 days. In this case the

given test process’s chosen temperature shock test has

time scale with value 7/5.

4 CONCLUSION AND FUTURE

WORK

In this work we introduced a novel approach, which

is based upon some models for resource scheduling of

workﬂow-elements. Resources of the workﬂow are

represented by cooperating and rival agents, whose

capabilities are carrying out activities on products.

We intended to model these agents and to work out

their coordination, management in such a way, that

the whole system realizes a well-scheduled workﬂow

which carries out the desired operation in a robust,

fault tolerant way making possible the substitution of

the system-elements by each other. A system which is

created based on the presented models is applicable in

wide spectrum of practical use for solving scheduling

problems related to business workﬂows. In current

development phase the system is suitable for time-

based or cost-based scheduling. The system is able

to compare the unscheduled and scheduled workﬂows

to each other and display the result. Moreover, it is

well applicable for comparing the results of different

scheduler algorithms.

In our future work we would like to create a com-

plete workﬂow management system which integrates

this method to schedule workﬂows of real applica-

tions. The set of system modules will be expanded

with a log analysis module to make the scheduling

more efﬁcient using historical data from log ﬁles. The

system is created also for analyzing the efﬁciency of

different log analyzer methods - similar to the case of

different scheduling algorithms. Furthermore we plan

to add a mixed scheduling factor which includes the

time and the cost with any weight at the same time.

ACKNOWLEDGEMENTS

This publication has been supported by the European

Union and Hungary and co-ﬁnanced by the European

Social Fund through the project TAMOP-4.2.2.C-

11/1/KONV-2012-0004 - National Research Center

for Development and Market Introduction of Ad-

vanced Information and Communication Technolo-

gies.

REFERENCES

Barba, I. and Del Valle, C. (2010). A job-shop schedul-

ing model of software development planning for

constraint-based local search. International Journal

of Software Engineering & Its Applications, 4(4).

Bhaskar, R., Lee, H. S., Levas, A., P´etrakian, R., Tsai, F.,

and Tulskie, B. (1994). Analyzing and re-engineering

business processes using simulation. In Simulation

Conference Proceedings, 1994. Winter, pages 1206–

1213. IEEE.

Brucker, P. and Brucker, P. (2007). Scheduling algorithms,

volume 3. Springer.

Dulai, T. and Werner-Stark, A. (2012). Immediate event-

aware routing based on cooperative agents. In Pro-

ceeding of Factory Automation, pages 144–148.

Ghosh, J. (2013). Distributed task scheduling and human

resource distribution in industrial service solution pro-

duction: a simulation application. International Jour-

nal of Production Research, 51(10):2900–2914.

Merdan, M., Moser, T., Sunindyo, W., Bifﬂ, S., and Vrba,

P. (2013). Workﬂow scheduling using multi-agent sys-

tems in a dynamically changing environment. Journal

of Simulation, 7(3):144–158.

Murthy, S., Akkiraju, R., Rachlin, J., and Wu, F. (1997).

Agent-based cooperative scheduling. In Proceedings

of AAAI Workshop on Constraints and Agents, pages

112–117.

Pinedo, M. (2012). Scheduling: theory, algorithms, and

systems. Springer.

Sule, D. R. (2008). Production Planning and Industrial

Scheduling: Examples, case studies, and applications.

CRC press.

Tumay, K. (1996). Business process simulation. In Pro-

ceedings of the 28th conference on Winter simulation,

pages 93–98. IEEE Computer Society.

van der Aalst, W. M., Reijers, H. A., Weijters, A. J., van

Dongen, B. F., Alves de Medeiros, A., Song, M., and

Verbeek, H. (2007). Business process mining: An in-

dustrial application. Information Systems, 32(5):713–

732.

van der Aalst, W. M., van Dongen, B. F., Herbst, J.,

Maruster, L., Schimm, G., and Weijters, A. (2003).

Workﬂow mining: A survey of issues and approaches.

Data & knowledge engineering, 47(2):237–267.

van der Aalst, W. M., Weijters, T., and Maruster, L. (2004).

Workﬂow mining: Discovering process models from

event logs. Knowledge and Data Engineering, IEEE

Transactions, 16(9):1128–1142.

Xu, J., Liu, C., Zhao, X., and Yongchareon, S. (2010).

Business process scheduling with resource availabil-

ity constraints. In On the Move to Meaningful Internet

Systems: OTM 2010, pages 419–427. Springer.

SIMULTECH2014-4thInternationalConferenceonSimulationandModelingMethodologies,Technologiesand

Applications

412