Fishwick P.A. (editor) Handbook of Dynamic System Modeling

Подождите немного. Документ загружается.

30-12 Handbook of Dynamic System Modeling

NH4_plus

NH3

H_plus

OH_minus

Bu_minus

Pr_minus

CO3_2minus

CO2_total

HCO3_minus

FIGURE 30.8 Classic graph-based ontology editor showing simulation symbols and their relationships.

or with g/cm

. In the ontology, units are expressed not as simple strings but as a composition of other

units and operators (kg, /, m, power operator, 3) each of which are also symbols in the ontology. Such

composite units can be used to analyze the dimension of symbols appearing in equations. This facilitates

automatic unit conversions and helps determine the correctness of the equation and the dimension and

magnitude of the symbols.

30.4.3 SimulationEditor

SimulationEditor incorporates the EquationEditor described in the previous section and is used for

specifying overall model structure in the form of elements. SimulationEditor also contains facilities for

automatically generating and running simulations and generating reports. SimulationEditor is designed

to represent models of dynamic systems using graphic elements such as source, sink, storage, and ﬂow.

We adopt this concept from the compartmental model (Peart and Curry, 1998) and Forrester notation

(Forrester, 1971), which is widely used in agriculture and natural resource models.

Figure 30.9 shows a mathematical equation that has been converted to various different languages

including MathML, OpenMath, and Java. For program code generation, each equation in the model is

converted to a Java method returning the result of solving the equation. The code is compiled and executed

after it is generated. Independent threads are used for generating and compiling code. The threads check

and report the consistency of the equation and the correspondence of unit and value. Simulation results

are shown as graphs and tables created with a report generator that is integrated into the SimulationEditor.

30.5 Conclusions

This chapter explored several ways in which ontologies can be applied to simulation in the agriculture and

natural resource domains. As an example, an ontology-based simulation was developed in the bioprocess-

ing domain. Development involved seven steps including collection of relevant documents, deﬁning the

model in terms of elements, identifying classes, individuals, and properties, encoding equations, entering

initial values of state, constant, and other parameters, generating code, and executing the simulation.

We have developed an ontology database management system where the database management capabil-

ities are built entirely around an ontology language rather than a traditional relational or object database

Ontology-Based Simulation in Agriculture and Natural Resources 30-13

[ Source Equation]

A =B + C

[ generated documents]

<math display=

'block' xmlns ='

http://www.w3.org/1998/Math/MathML'>

<mrow>

<mi> B </mi><OMOBJ xmlns ="http://www.openmath.org/OpenMath"

version ="2.0" cbase ="http://www.openmath.org/cd">

</mrow><OMV name = "A" />

</mrow><OMA>

</math><OMS cd="arith1" name= "plus" />

<OMV name="B" /> public double A()

< /OMA> {

< /OMA

> "return B() +c()" ;

< /OMBJ> }

FIGURE 30.9 MathML, OpenMath, and Java code generated automatically from an equation.

model. Authoring tools for developing models are provide in this environment which have interfaces

similar to traditional tools such as conventional equation editors. They utilize an ontology as the back-end

for representing all concepts in the model.

The development of an ontology for simulation models explicitly exposes knowledge contained in mod-

els at a higher level. This knowledge can be further used for constructing conceptual models, simulations

of similar systems, and educational and training materials. The construction of an ontology will allow

better communication of knowledge about models, model elements, and data sources among different

modelers, enhance interoperability between different models, and increase the reusability and sharing of

model components.

References

Agriculture Research Service, The in-house research arm of United State Department of Agriculture, 2005,

http://www.ars.usda.gov/research/projects/projects.htm?ACCN_NO=407227.

Altman, R. B., Xiaoqian, M.B., Chai, J., Carillo, M. W., Chen, R. O., and Abernethy, N. F., RiboWeb: An

ontology-based system for collaborative molecular biology, IEEE Intell. Syst., 14(5), 68–76, 1999.

Annop, N., Pullammanappallil, P. C., and Clarke, W. P., A dynamic mathematical model for sequential

leach bed anaerobic digestion of organic fraction of municipal solid waste, Biochem. Eng. J., 13,

21–33, 2003.

Beck, H.W., Albrigo, L. G., and Kim, S., DISC citrus planning and scheduling program. in Proceedings

of the Seventh International Symposium on Modelling in Fruit Research and Orchard Management,

P. Braun (Ed.,) pp. 25–32, 2004.

Beck, H.W., Papajorgji, P., Braga, R., Porter, C., and Jones, J.W., Object-oriented approach to crop

modeling: concepts and issues. Proc. of Congress on Computers in Agriculture and Natural Resource,

ASAE Publication Number 701P0301.

30-14 Handbook of Dynamic System Modeling

Cuske, C., Dickopp, T., and Seedorf, S., JONTORISK: an ontology-based platform for knowledge-based

simulation modeling in ﬁnancial risk management, Proc. of 2005 ESM Conference, 2005.

Fishwick, P.A., and Miller, J. A., Ontologies for modeling and simulation: issues and approaches. Proc. of

2004 Winter Simulation Conference, 2004.

Forrester, J.W., Principles of Systems, Wright Allen, Cambridge, MA, 1971.

Furmento, N., Mayer, A., McGough, S., Newhouse, S., Field, T., and Darlington, J., Optimization of

component-based applications within a grid environment. Proc. of the 2001 ACM/IEEE Conference,

2001.

Heidel, J., and Maloney, J., An analysis of a fractal Michaelis–Menten curve, J. Aust. Math. Soc. Ser. B, 41,

410–422, 2000.

Kiker, G.A. Development of JAVA based, object oriented modeling system for simulation of South African

hydrology, Proc. of Am. Soc. of Agricultural Engineers, ASAE Publication Number 01-2030, 2001.

Knutson, J., and Kreger, H., Web Services for J2EE, Version 1.0. Public Draft v0.3, IBM, 2002.

Lacy, L., and Gerber, W., Potential modeling and simulation applications of the web ontology language—

OWL, Proc. of 2004 Winter Simulation Conference, 2004.

Lai ,T.E. Rate Limiting Factors of the Anaerobic Digestion of Municipal Solid Waste in Bioreactor Landﬁlls,

PhD dissertation, Department of Chemical Engineering, The University of Queensland, Brisbane,

2001.

Lu, H.Y., Lu, C.T., Wei, M.L., and Chan, L.F., Comparison of different models for nondestructive leaf area

estimation, Taro. Agron. J., 96, 448–453, 2004.

Miled, Z. B., Webster, Y. W., Li, N., Bukhres, O., Nayar, A. K., Martin, J., and Oppelt, R., BAO, a biological

and chemical ontology for information integration, Online J. BioInformat., 1, 60–73, 2002.

Miller, J. A., Fishwick, P.A., Baramidze, G. T., and Sheth, A. P., Investigating ontologies for simulation

modeling, Proc. of 37th Annual Simulation Symposium (ANSS’04), 2004.

Mills-Tettey, G. A., Johnston, G., Wilson, L. F., Kimpel, J. M., and Xie, B., A security framework for the

agent-based environment for linking simulations (ABELS), Proceeding of the 2002 Winter Simulation

Conference, 2002.

Noy, N. F., Sintek, M., Decker, S., Crubezy, M., Fergerson, R. W., and Musen, M. A., Creating semantic

web contents with protege-2000, IEEE Intell. Syst., 16(2), 60–71, 2001.

OWL, Web Ontology Language Guide, http://www.w3.org/TR/owl-guide, 2005.

Park, M., and Fishwick, P. A., Ontology-Based Customizable 3D Modeling for Simulation, PhD disserta-

tion, Department of Computer Science Engineering, The University of Florida, Gainesville, 2005.

Peart, R., and Curry, R. B., Agricultural Systems Modeling and Simulation, Marcel Dekker, New York, 1998.

Raubal, M., and Kuhn, W., Ontology-based task simulation, Spat. Cogn. Comput., 4(1), 15–37, 2004.

Reitsma, F., and Albrecht, J., Modeling with the semantic web in the geosciences, IEEE Intell. Syst., 20(2),

86–88, 2005.

TouchGraph Website, http://www.touchgraph.com, 2005.

Volz, R., Oberle, D., Staab, S., and Motik, B., KAON SERVER—a semantic web management system in

alternate track, Proc. of the Twelfth International World Wide Web Conference, 2003.

Yang, A., and Marquardt, W., COGents: applying software agent technology to support process modeling,

Presentation LPT–pre–2003–06, 2003.

Modeling Human

Interaction in

Organizational Systems

Stewart Robinson

University of Warwick

31.1 Introduction .................................................................31-1

31.2 Systems and Human Interaction .................................31-2

31.3 Why Model Human Interaction? ................................31-3

31.4 Modeling Human Interaction:

Research and Practice...................................................31-4

Modeling Human-to-System Interaction

•

Modeling

Human-to-Human Interaction

31.5 The KBI Methodology..................................................31-5

Stage 1: Understanding the Decision Making

Process

•

Stage 2: Data Collection through

Simulation

•

Stage 3: Determining the Decision Makers’

Decision Making Strategies

•

Stage 4: Determining

the Consequences of the Decision Making Strategies

•

Stage 5: Seeking Improvements

31.6 A Case Study: Modeling Human Decision

Making at Ford Motor Company ................................31-8

KBI Stage 1

•

KBI Stage 2

•

KBI Stage 3

•

KBI Stage 4

•

KBI Stage 5

•

Findings about KBI

31.7 Conclusion ...................................................................31-12

31.1 Introduction

Most systems involve some level of human interaction. Therefore, if we are to model such systems it

seems necessary to have at least some understanding of the nature of that interaction. It would also seem

necessary to be able to model the interaction of humans to obtain a proper understanding of the system

and its performance.

This chapter explores the idea of modeling the interaction of humans with organizational systems. First,

the concept of system and the way in which humans interact with systems, and each other, are discussed.

The motivation for modeling human interaction is then addressed, before reviewing some examples of

modeling human-to-system and human-to-human interaction. A key focus of the chapter is to outline

the knowledge-based improvement (KBI) methodology. This methodology focuses on modeling a speciﬁc

form of human interaction, namely human decision making. An example of the methodology in practice

at the Ford Motor Company is then described.

31-1

31-2 Handbook of Dynamic System Modeling

31.2 Systems and Human Interaction

Checkland (1981) identiﬁes four main types of system. Natural systems have their inception with the

origins of the universe and range from atomic systems, through living systems on earth, to galaxies.

Although many of these systems appear to arise from natural processes, there is evidence that some are

the result of (intelligent) design (Dembski, 1999). Designed physical systems are the result of conscious

design and exist to serve some human purpose. Examples are a wrench, a manufacturing plant, and a

car. Meanwhile, designed abstract systems, such as mathematics and literature, are similarly the product of

conscious design, but are abstract in form. Finally, there are human activity systems. These less tangible, but

observable, systems are more or less consciously ordered and exist for some purpose. Examples include a

political system, a health system, a charitable organization, and a football team.

All of these systems can, and often do, involve some level of human interaction. This is most obviously

the case for human activity systems, since they exist as systems of human interaction. Designed abstract

systems are developed and exist through human consciousness and interaction between human actors.

Designed physical systems are developed by humans and often involve human interaction in their use.

A car, for instance, exists as a designed physical system, but only becomes useful when a human starts

to interact with the system, that is, drives the car. Even natural systems involve some level of human

interaction. The climate is affected by human activity through emissions of CO

, and as I write, humans

are crashing a spaceship into a comet having some (small) effect on our galaxy.

To one degree or another, the nature and degree of human activity determines the performance of these

systems. Place a hesitant driver in a Formula 1 car and its performance is going to be very different from that

expected when a professional is behind the wheel. The performance of a regional health system is hugely

dependent on the human activitythatdeﬁnes that system. Indeed,as we move fromnatural systems through

to human activity systems, human interaction has an increasing impact on the performance of the system.

In this chapter, the focus is on organizational systems primarily as they exist in business or public sector

organizations. These can primarily be classiﬁed as designed physical systems or human activity systems.

Indeed, rather than seeing these as bipolar extremes, there exists a continuum from systems that are largely

designed physical systems through to systems that mainly exist as human activity. This is illustrated in

Figure 31.1 with some examples of organizational systems. On the far left are designed physical systems

that involve little or no human interaction in their daily operation. Although an automated warehouse

requires human activity for its initial design and build, there is little or no human activity in its daily

operation with the exception of occasional manual overrides and maintenance.

To the far right are systems that exist purely as human activity. In between these two extremes are a

whole range of systems that are in part designed physical systems and in part human activity systems. A

retail bank, for instance, is designed with a layout and facilities. It also acts as a human activity system

with customers interacting with staff in some purposeful activity. The performance of the bank is in part

determined by the design and in part by the human activity.

Automated

warehouse

Designed physical

systems

Human activity

systems

Call center

Manufacturing

plant

Systems design by humans

No human interaction in day-

to-day operation

Systems of human activity

Exist purely as human

interaction

Retail bank

Regulated

market

Group decision

making

Health

system

Social

network

FIGURE 31.1 Systems and human interaction: a continuum.

Modeling Human Interaction in Organizational Systems 31-3

Within these organizational systems there are two types of human interaction. The ﬁrst involves an

interaction between the human and the designed physical system (human-to-system interaction). Taking

the example of the bank, a customer interacts with the layout of the bank in choosing a route to a service

counter. The customer may also interact with an automated teller. A member of staff interacts with the

equipment at the service counter.

Secondly, there is human-to-humaninteraction. In the bank customers interact with staff when receiving

service. Customers also interact with one another (e.g., collision avoidance) as do staff. As we move further

to the right in Figure 31.1, human-to-human interaction gradually predominates over human-to-system

interaction.

31.3 Why Model Human Interaction?

Put simply, it is important to model human interaction because systems involve human activity and

because that activity helps determine the performance of a system. Balanced against this is the difﬁculty

of modeling human interaction, not least because humans are complex. A particular difﬁculty is being

able to obtain data on how humans behave. Customer behavior in a bank, for instance, is complex. What

determines their choice of service? How do customers choose which queue to join? When do customers

balk, jockey, or leave a queue? How do customers decide how to move from location to location in the bank?

What are the determinants of the time it takes to serve customers? Added to the difﬁculty of answering

these questions is the recognition that every customer behaves differently.

While human behavior is complex and important in determining system performance, expert modelers

advise that models should be kept as simple as possible (Robinson, 1994; Pidd, 1999). There are a number

of advantages to simpler models (Innis and Rexstad, 1983; Ward, 1989; Salt, 1993; Chwif et al., 2000;

Thomas and Charpentier, 2005):

•

Faster development

•

More ﬂexible

•

Less onerous data requirements

•

Faster execution speed

•

Easier interpretation of the results

These advantages are lost as the complexity of a model increases. This does not mean that complex models

should not be developed. However, it does mean that the aim should be to develop the simplest model

possible to achieve the objectives of a modeling study.

Robinson (2004) illustrates the need for simple models with the diagram in Figure 31.2. As complexity

(the scope and level of detail of the model) increases, the accuracy of the model also increases, but with

diminishing returns. Eventually a point is reached where the accuracy of the model may diminish with

Human interaction

Scope and level of detail (complexity)

100%

Model accuracy

FIGURE 31.2 Model accuracy and increasing complexity. (Adapted with permission from Robinson, S., Simulation:

The Practice of Model Development and Use, p. 68, 2004.)

31-4 Handbook of Dynamic System Modeling

increasing complexity. This is because the data required to support such models are not available. The

modeling of human interaction in organizational systems certainly involves increasing the complexity of

a model, and possibly to the point where the accuracy begins to diminish.

There is something of a paradox when it comes to modeling human interaction in organizational

systems. At one level it is important to include human factors because they affect the performance of a

system, but at another, the implied complexity suggests that human factors should be excluded from any

model of a system. The paradox is resolved to an extent by noting that the requirement is to design the

model with respect to the objectives of the modeling study (Robinson, 2004). The desire to model human

interaction should, therefore, be driven by a speciﬁc requirement to study human interaction. Among the

many reasons for this might be:

•

Understanding the effects of human interaction on the performance of the system

•

Improving the performance of human actors in the system

•

Improving system performance

None of these necessarily require highly detailed or complex models. High model ﬁdelity is not always

required to make a model useful (Hodges, 1991; Robinson, 2001). Instead, it is argued here that the key

aim is to provide a learning environment, in which the model is used to better understand the interaction

between human actors and the system. Through this improved understanding, actions can be taken to

improve the performance of the human actors and/or to improve the performance of the system.

31.4 Modeling Human Interaction: Research and Practice

Before describing a speciﬁc example of modeling human interaction, it is worth pausing to obtain an

understanding of the range of research and practical work that is being carried out in this area. That said,

it should be noted that the modeling of human interaction is a very underdeveloped ﬁeld. Much of the

work remains at the level of research on examples, with only some practical applications. It is also a diverse

ﬁeld with very little cohesion. As a result, the aim here is not to give a complete review of the ﬁeld, but to

provide some examples of the type of work that has been, and is being, performed. This work is described

under two headings: modeling human-to-system interaction and modeling human-to-human interaction.

31.4.1 Modeling Human-to-System Interaction

Perhaps the ﬁrst place to start in discussing the modeling of human-to-system interaction is to recognize

that simulation games incorporate human interaction with a system. A business game, for instance,

involves human actors entering policies and decisions on a periodic basis, which then determine the

performance of the organization being modeled. Simulators also incorporate human interaction with a

system. The best known example is probably a ﬂight simulator where a human interacts with a simulation

as a means of practicing ﬂight procedures. Neither of these strictly model human interaction since they

rely on interaction between real people and the game or simulator rather than on a model of the human

actors. What they do provide is a means of assessing the impact of human actors on a system.

In terms of speciﬁcally modeling human interaction, one area with a reasonably long history of research

has been the modeling of human decision making. Flitman and Hurrion (1987),Williams (1996), and Lyu

and Gunasekaran (1997) all describe examples of linking simulation models with artiﬁcial intelligence (AI)

representations of human decision makers. Flitman and Hurrion link a discrete-event simulation model of

a coal yard with a rule-based expert system, enabling human actors to input control decisions. The expert

system learns from the control actions taken. Williams uses a simulation to elicit knowledge from experts

concerning the scheduling of replenishment at sea by the Royal Navy. The knowledge is used to train a

rule-based expert system, which is then linked back to the simulation model to test the performance of the

scheduling strategy. Lyu and Gunasekaran represent managers’decisions concerning unloading operations

at a wharf using a rule-based expert system, and then test the performance of decision strategies using a

Modeling Human Interaction in Organizational Systems 31-5

simulation. Similar ideas are discussed by Robinson et al. (1998) and Standridge and Steward (2000). In

all these cases, rule-based expert systems are used for representing the decision making strategies.

As an alternative to the use of rule-based expert systems for representing human decision making

strategies, both Liang et al. (1992) and Curram (1997) use neural networks. Meanwhile, Moffat (2000)

discusses the use of Bayesian inference for modeling command and control in simulation models of

military combat.

Outside of modeling human decision making in systems, Harmon et al. (2006) provide a general

discussion on human behavior representation in simulation with a focus on the validation of such repre-

sentations. A speciﬁc example of modeling worker behavior in manufacturing simulation models is given

by Baines and Kay (2002). They develop a pilot model, which includes the effects of worker stereotype (e.g.,

action oriented or thinking person) and the environment (e.g., noise level, temperature, and cleanliness)

on worker performance. The model utilizes a neural network to evaluate the relationship between these

factors and performance. In a similar vein, Brailsford and Schimdt (2003) model patients’ attendance

behavior for diabetic retinopathy. Using Schmidt’s PECS framework (Schmidt, 2000), the model takes

into account the patients’ physical state, emotions, cognitions, and social status. Both of these projects

incorporate ideas taken from cognitive psychology.

31.4.2 Modeling Human-to-Human Interaction

Agent-based modeling provides concepts and constructs for modeling human-to-human interaction.

Unlike traditional simulation approaches which aim to model a system as a deﬁned set of relations between

system elements (top-down approach), agent-based models adopt a bottom-up approach. Agents with

given behaviors interact with other agents, leading to a system with emergent properties. The prisoners’

dilemma illustrates how complexity can arise from what appears to be a simple two-person human-to-

human interaction (Axelrod, 1997). Silverman et al. (2006) discuss the state-of-the-art in human behavior

modeling and identify the key challenges in improving models of socially intelligent agents.

Two speciﬁc examples of agent-based simulation in an organizational context are given by Baxter et al.

(2003) and Schelhorn et al. (1999). In the former, an agent-based model is developed to understand how

word of mouth inﬂuences the adoption of products and services by customers. Meanwhile, Schelhorn

et al. simulate pedestrian movements in an urban area based on the inﬂuence of spatial conﬁguration,

predeﬁned activity schedules and different combinations of land use.

Another branch of work in modeling human-to-human interaction is in evacuation simulations. These

models not only involve interactions between evacuees, but also interactions between people and the

system (e.g., a building). A number of examples of this type of simulation exist, for instance, Kim et al.

(2004) describe progress toward a simulation for evacuation in marine accidents, while Learmount (2005)

summarizes work on simulating evacuation from aircraft.

Social network analysis (SNA) provides a means for modeling and understanding human-to-human

interaction in organizations. Cross and Parker (2004) discuss the use of SNA in business organizations,

while Carley (2003) focuses on their use for intelligence work in identifying and understanding terrorist

networks.

What is apparent in this work is that a key challenge in modeling both human-to-system and human-to-

human interaction is developing valid models of human behavior. Further to this, it is difﬁcult to obtain

data to populate these models. What now follows is a description of a methodology that aims to address

both these needs within the context of modeling human decision making. The methodology is illustrated

by a case study example in a Ford manufacturing plant.

31.5 The KBI Methodology

KBI uses visual interactive simulation (VIS) (Pidd, 2004; Robinson, 2004) and AI to develop an under-

standing of human decision making within organizational systems, primarily manufacturing systems. VIS

is used to elicit knowledge from decision makers and AI is used to learn and represent different decision

31-6 Handbook of Dynamic System Modeling

Stage 1

Stage 2

Stage 4

Stage 3

Stage 5

Investigation of the

operations system

Understanding the process

and the decision

making required

Generate decision

making scenarios

Predict performance of

decision making strategies

Provide input to the

VIS in place of

the decision-makers

Decisions taken

under scenarios

(datasets held for

each decision maker)

Elicit

knowledge

Trains

Represent the decision making

strategy of each decision-maker

Seek improvements

VIS model

AI model

Data

sets

FIGURE 31.3 The Knowledge Based Improvement methodology. (From Robinson et al. 2005. Journal of the

Operational Research Society 56: 913. With permission.)

making strategies. The VIS and AI are then linked to predict the performance of the system under different

decision making strategies.

As discussed above, the prime motivation of KBI is not to develop a more accurate simulation, but to

create a learning environment in which good practice can be identiﬁed and shared among decision makers.

KBI consists of ﬁve stages (Figure 31.3):

•

Understanding the decision making process

•

Data collection through simulation

•

Determining the decision makers’ decision making strategies

•

Determining the consequences of the decision making strategies

•

Seeking improvements

Each stage is summarized below, with a more complete discussion to be found in Robinson et al. (2005).

In the KBI methodology the problem is conceived as follows. A decision making process consists of two

vectors:

i, j

= [d

, d

, ...]

= [a

, a

, ...] (31.1)

i,j

is a set of decisions made by decision maker j at decision point i. Each decision made is represented

by decision variables (d

, d

, ...). The values that these decision variables take are known as the decision

options.VectorA

lists the decision attributes (a

, a

, ...) at decision point i, described as the decision

scenario. The decision attributes can take on various attribute levels.

Modeling Human Interaction in Organizational Systems 31-7

The aim of the KBI methodology is to identify and improve the function f

, which represents the decision

making strategy of decision maker j:

i, j

= f

) (31.2)

Function f

is itself a model, in this case of a human decision making strategy. As such, the discussion

above concerning models as simpliﬁcations applies. Indeed, KBI does not set out with the notion that a

complete model of a decision making strategy can be obtained, but instead that only a partial explanation is

possible. This is similar to Zeigler’s (1976) view that the base model (a complete explanation of a system) is

unknowable, and that instead a lumped model (a partial and appropriate explanation) should be utilized.

The aim of KBI, therefore, is to identify a partial model of a decision making strategy based on a partial

understanding of the decision variables and decision attributes. As a consequence, the decision making

strategy as identiﬁed by KBI can be expressed as

i, j

(

) (31.3)

denoting that the two vectors and the function

are all approximate.

This does leave the question of what is the appropriate level of approximation for the function

.It

is supposed that a relation similar to that set out in Figure 31.2 applies. Similarly, the determination of

how far to move along the curve should be driven by the objectives of the study. If the aim is to devise

an AI system to largely replace a human decision maker in a real system, then a much greater level of

accuracy is required than if the aim is to gain a better understanding of the effects of human decision

making on system performance. The selection of an appropriate simpliﬁcation in modeling is itself a ﬁeld

for further research. For discussions on this topic see Zeigler (1976), Ward (1989), Nance (1994), Powell

(1995), Brooks and Tobias (1996), Pidd (1999) and Pace (2000).

31.5.1 Stage 1: Understanding the Decision Making Process

In the ﬁrst stage of KBI, the aim is to identify the decision variables, decision options, decision attributes,

and attribute levels. Although interviews and discussion with the decision makers can reveal some infor-

mation about the decision making process, usually a decision maker cannot explicitly identify and list all

the decision making components. Various other approaches can be used to develop a better understanding

of the decision variables and attributes:

•

Observing the decision makers as they take decisions.

•

Hypothesizing about likely decision variables and attributes.

•

Interaction with a simulation model (stage 2 of the methodology).

31.5.2 Stage 2: Data Collection through Simulation

This step involves collecting example decisions from the decision makers to obtain the decision matrices:

⎡

⎢

⎣

1,1

1,2

i,1

i,2

I,1

⎤

⎥

⎦

A =

⎡

⎢

⎣

1,1

1,2

i,1

i,2

I,1

⎤

⎥

⎦

(31.4)

where

is the series of decisions taken by the decision maker j under the decision attributes represented

A. A total of I scenarios are presented to each decision maker. These data are collected through a VIS

model. The decision maker interacts with a VIS of the system in question entering his/her decisions to

the model. The model records the value of each decision variable and decision attribute to a data ﬁle. As

a result, a set of values for the matrices

and

A are collected. Separate

matrices can be generated for

each decision maker by presenting them with the same set of scenarios,