Fishwick P.A. (editor) Handbook of Dynamic System Modeling
Подождите немного. Документ загружается.
31-8 Handbook of Dynamic System Modeling
There are various advantages of using VIS over observing decision makers in the real system:
•
The speed with which decision scenarios can be presented to the decision makers.
•
The ability to control the nature of the decision scenarios.
•
The ability to replicate the same decision scenarios with different decision makers.
•
The ability to easily record every attribute of the simulated system at each decision point.
Of course, certain problems with the use of VIS also exist:
•
The ability of a simulation to contain all the decision attributes of a decision scenario.
•
The extent to which a decision maker makes realistic decisions in a simulated environment.
•
The monotony of working in a simulated environment.
31.5.3 Stage 3: Determining the Decision Makers’ Decision Making
Strategies
AI methods are used to determine the decision making strategies (
ˆ
f
j
) of the different decision makers from
whom data have been collected. Various AI approaches can be used, for instance, rule-based expert systems,
case-based reasoning, and artificial neural networks, as well as statistical pattern recognition techniques
(e.g., logistic regression). The data collected in stage 2 are used to train an AI representation using, for
instance, the ID3 algorithm (Quinlan, 1986) for determining a decision tree in a rule-based system. Since
rule-based systems provide a decision tree they are favored. This is because they not only model a decision
making strategy, but they also have the greatest explanatory power. A separate AI representation (e.g.,
decision tree) should be created for each decision maker.
31.5.4 Stage 4: Determining the Consequences of the Decision Making
Strategies
In this step, the performance of the decision making strategies (
ˆ
f
j
) are determined by linking the AI
representation of each decision maker with the simulation model. Instead of requesting input from the
decision maker, the simulation now requests input from the AI models. This enables long simulation runs
to be performed so that the effect of different decision making strategies on the performance of the system
can be determined.
31.5.5 Stage 5: Seeking Improvements
Improved decision making strategies can be sought by combining the best strategies or by taking a good
strategy and discussing with the decision makers how improvements can be made. More formal heuristic
search methods could also be employed to look for better strategies. In each case, the alternative strategies
can be tested by running them with the VIS to determine their effectiveness.
31.6 A Case Study: Modeling Human Decision Making at
Ford Motor Company
The Ford engine assembly plant at Bridgend (South Wales) is one of the main production plants for
the Zetec petrol engine. The plant consists of three main assembly lines (Line A, Line B, and the Head
Subassembly Line) followed by a Hot Test area and an After Test Dress line for finishing operations. The
KBI methodology was applied to the problem of machine failures on the main assembly lines. When a
machine fails, the machine is first inspected and then the maintenance supervisors determine what courses
of action to take. The research project focused on the decision making strategies of the maintenance
supervisors. To reduce the scope of the work, only the maintenance activities in the first of the four areas
on the line were considered, representing about a quarter of the machines on the line.
Modeling Human Interaction in Organizational Systems 31-9
31.6.1 KBI Stage 1
Initial investigations in stage 1 of the methodology were carried out by observing maintenance engineers
at work and through discussion with the engineers and plant manager. These revealed that there were
three decision variables surrounding the failure of a machine (decision options are shown in brackets):
•
What action should be taken? (repair immediately [RI],“standby”—operate manually [SB], switch-
off the machine [SO])
•
Who should act? (group leaders [L1, L2], second skilled engineer [L3], semiskilled engineer [L4],
unskilled engineer [L5], ask the production manager [APM])
•
Should planned maintenance be performed on the machine? (perform planned repair[PR],schedule
planned repair [SPR])
The decision vector in this context is therefore expressed as a set of binary variables:
ˆ
D
i,j
= [RI SB SO L1 L2 L3 L4 L5 APM PR SPR] (31.5)
Following these investigations (and some use of the VIS in stage 2 of the methodology), the decision
attributes were found to be:
•
Type of fault (TF)
•
Estimated repair time (ERT)
•
Machine number (MN)
•
Time of day (TD)
•
Number of engines produced so far this shift (NEP)
•
Engines waiting on the conveyor before the machine (EWC)
•
Number of heads in the buffer (NHB)
•
Number of breakdowns on this machine today (NBMT)
•
Number of breakdowns on this machine this month (NBMM)
•
Number of breakdowns of this type on this machine today (NBTMT)
•
Number of breakdowns of this type on this machine this month (NBTMM)
These attributes could each take on various levels. The vector of decision attributes is expressed as follows:
ˆ
A
i
= [TF ERT MN TD NEP EWC NHB NBMT NBMM NBTMT NBTMM] (31.6)
31.6.2 KBI Stage 2
A VIS model of the assembly line already existed, but this did not contain any complex decision making
logic for maintenance decisions; it defaulted to a repair immediately decision. The model was adapted to
enable decisions to be entered by an expert user as machines failed and an interface was devised for data
collection (Figure 31.4 and Figure 31.5). The interface, written in Visual Basic
®
,
1
reported the status of
the simulation (the decision attributes) and asked the user to enter a decision (the decision variables).
Data collection sessions were run with three decision makers; the maintenance supervisor for each of
three shifts on the first area of the line. Each was presented with the same set of 63 scenarios (
ˆ
A) via the
simulation. These scenarios were selected from a historic trace of data on machine failures obtained from
the plant. The trace was adjusted, removing repetitive and simple cases, to ensure that there was a wide
range of decision scenarios. Data collection sessions took around 1 h with each decision maker. This was
found to be both as long as the maintenance supervisors could spare for this exercise and as long as their
concentration when using the model could last.
1
Registered trademark of Microsoft.
31-10 Handbook of Dynamic System Modeling
FIGURE 31.4 Visual Basic interface for VIS model: reporting decision attributes.
FIGURE 31.5 Visual Basic interface for VIS model: entering decision variables.
31.6.3 KBI Stage 3
Based on the data obtained (
ˆ
D
j
), three approaches were used to learn the decision making strategies
(
ˆ
f
j
) of the three decision makers: artificial neural networks, logistic regression, and rule-based expert
systems. MATLAB
®
,
2
SPSS
®
,
3
and the XpertRule
®4
software were used, respectively, for implementing
these methods. All provided a reasonable fit to the original decisions. As stated previously, the rule-based
expert system was preferred due to its explanatory power in that it provides a decision tree showing each
decision making strategy.
An example extract of a decision tree generated using this approach is shown in Figure 31.6. If machine
Op1025 breaks down, the decision maker will repair immediately if the estimated repair time is less than
20 min, otherwise standby operation will be used. For Op1060 a different repair time threshold applies.
2
Registered trademark of The Math Works Inc., Natick, MA.
3
Registered trademark of SPSS Inc., Chicago, IL.
4
Registered trademark of XpertRule Software, Leigh, UK.
Modeling Human Interaction in Organizational Systems 31-11
Machine
Op1025
Repair time 20
Repair time 20
Op1060
Op1120
Repair immediately
Standby
Repair time 36
Repair time 36
Repair immediately
Repair immediately
Heads in buffer 36
Heads in buffer 36
Repair immediately
Standby
FIGURE 31.6 Extract from the decision tree derived for one of the decision makers. (From Robinson et al. 2005.
Journal of the Operational Research Society 56: 918. With permission.)
TABLE 31.1 Performance of Decision Making Strategies in
Terms of Assembly Line Daily Throughput
DM1 DM2 DM3
Mean daily throughput 325.53 325.15 318.87
Standard deviation 36.28 28.54 34.34
Further to this, the decision maker takes account of the number of engine heads that are available in the
buffer. For Op1120 the decision is always to repair immediately.
Notice that only a few of the decision attributes are required to determine what decisions would be
taken. The same is true for all the decision makers, although some may take account of different attributes.
There is something of a disconnect then between the attributes that a decision maker suggests he/she uses
in forming decisions and those that are actually used. This helps confirm the view that simply asking
decision makers what decisions they take and how is unlikely to yield accurate information. Very often
experts simply do not know how they reach decisions or even what decisions they might take, since much
knowledge is tacit.
31.6.4 KBI Stage 4
The rule-based expert system was then linked to the VIS to perform a long simulation run of over 100
days. As decision points were reached during the model run, the expert system provided decisions in place
of the human decision maker. The model then continued its run simulating the consequences of each
decision.
This simulation runs showed some variations in performance between the decision makers as measured
by the throughput of the assembly line (Table 31.1). Note that a multiplier has been applied to the figures
for reasons of confidentiality. Decision makers 1 and 2 achieve a similar level of throughput, but decision
maker 2 achieves a lower variation on a day-to-day basis. Decision maker 3 achieves a mean throughput
of between 6 and 7 less engines a day.
31-12 Handbook of Dynamic System Modeling
As previously stated, these decision makers only supervise maintenance activities on the first area of
the assembly line. There are three further areas. In the simulations, it was assumed that for the rest of
the facility the decision is always to repair immediately. As a result, the effect of the decision makers
studied on overall assembly line performance is limited. To understand the full effect of maintenance
decisions on assembly line performance, a further study of the other three line areas would have to be
carried out.
31.6.5 KBI Stage 5
In the Ford project stage 5 was not performed explicitly. However, it is notable that at one point in the
project a new supervisor used the model as a training device. When he was uncertain about which decision
to take for a specific scenario, he sought advice from a colleague. In this way the methodology acted as a
catalyst for improving the supervisor’s decision making strategy.
31.6.6 Findings about KBI
This project demonstrated the validity of the KBI approach, showing that a VIS could be used to elicit
knowledge from decision makers and an AI tool could then learn their (approximate) decision making
strategies. Further to this, the VIS and AI models could be linked to determine the performance of the
manufacturing system.
Various issues concerning the methodology also emerged:
•
Stage 1: the use of formal problem structuring methods might aid this stage, e.g., cognitive mapping
(Eden and Ackermann, 2001) or soft systems methodology (Checkland, 1981).
•
Stage 2: the use of VIS needs further investigation, particularly the design of the human–computer
interface, e.g., level of visual representation, reporting of decision attributes, and selection of
decision scenarios to obtain sufficient valid data rapidly.
•
Stage 3: further exploration of AI approaches is required to identify those with greatest learning and
explanatory power.
•
Stage 4: validation of the AI representations is problematic and requires methodological develop-
ments. How can we be sure that an AI representation of a decision making strategy is close to the
actual strategy used in the real system?
•
Stage 5: use of informal and formal methods for improving decision making strategies need to be
explored.
A second project is focusing specifically on the issues surrounding stage 2 (Lee et al. 2005).
31.7 Conclusion
The KBI methodology provides an approach for obtaining data on, and representing, human decision
making strategies. It appears to do this with some success, but also a number of limitations. This suggests
the need for further development to the methodology.
Meanwhile, human decision making is just one form of human interaction witha system. There are many
other forms of interaction between humans and systems (e.g., speed of work, requests for information, and
environmental effects on humans) and between humans within systems (e.g., communication and physical
interaction). Since the performance of many organizational systems is in part determined by the interaction
of humans, it would be useful to better understand the nature of that interaction and its effects on system
performance. The modeling of systems and human interaction provides a means for achieving this.
At present examples of modeling systems and human interaction are quite limited and the field requires
further development with emphasis on two key areas. First, there is the need for modeling frameworks
to represent human interaction. KBI provides one such framework. Other approaches might be found
in fields such as AI, agent-based methods, game theory, evolutionary algorithms, and Bayesian belief
Modeling Human Interaction in Organizational Systems 31-13
networks. A second requirement is for data on human behavior and human interaction to be able to
populate the modeling frameworks. This is a complex task. Further development of the field surely
requiresa multidisciplinary approach with input from modeling experts, computer scientists, and cognitive
psychologists.
Acknowledgement
Sections of this chapter are based on the paper: Robinson, S., J.S. Edwards and E.P.K. Lee. 2005. Simulation
based knowledge elicitation. In International Conference on Human–Computer Interface: Advances for
Simulation Modeling, pp. 18–24. San Diego, CA: Society for Modeling and Computer Simulation.
References
Axelrod, R. 1997. The Complexity of Cooperation: Agent-Based Models of Competition and Collaboration.
Princeton, NJ: Princeton University Press.
Baines, T.S. and J.M. Kay. 2002. Human performance modelling as an aid in the process of manufacturing
system design: a pilot study. International Journal of Production Research 40(10): 2321–2334.
Baxter, N., D. Collings and I. Adjali. 2003. Agent-based modelling—intelligent customer relationship
management. BT Technology Journal 21(2): 126–132.
Brailsford, S. and B. Schmidt. 2003. Towards incorporating human behaviour in models of health care
systems: an approach using discrete event simulation. European Journal of Operational Research 150:
19–31.
Brooks, R.J. and A.M. Tobias. 1996. Choosing the best model: level of detail, complexity and model
performance. Mathematical and Computer Modelling 24(4): 1–14.
Carley, K.M. 2003. Estimating vulnerabilities in large covertnetworks. In Dynamic Social NetworkModeling
and Analysis: Workshop Summary and Papers. Washington, DC: The National Academies Press.
Checkland, P.B. 1981. Systems Thinking, Systems Practice. Chichester, UK: Wiley.
Chwif, L., M.R.P. Barretto and R.J. Paul. 2000. On simulation model complexity. In Proceedings of the 2000
Winter Simulation Conference, pp. 449–455. Piscataway, NJ: IEEE.
Cross, R. and A. Parker. 2004. The Hidden Power of Social Networks: Understanding How Work Really Gets
Done in Organizations. Boston, MA: Harvard University Press.
Curram, S. 1997. Representing Intelligent Decision Making in Discrete Event Simulation: A Stochastic Neural
Network Approach. PhD Thesis, Coventry, UK: University of Warwick.
Dembski, W.A. 1999. Intelligent Design: The Bridge between Science and Theology. Downers Grove, IL:
InterVarsity Press.
Eden, C. and F. Ackermann. 2001. SODA—The Principles. In Rational Analysis for a Problematic World
Revisited, pp. 21–42. Chichester, UK: Wiley.
Flitman, A.M. and R.D. Hurrion. 1987. Linking discrete-event simulation models with expert systems.
Journal of the Operational Research Society 38(8): 723–734.
Harmon, S.Y., C.W.D. Hoffman,A.J. Gonzalez, R. Knauf and V.B. Barr. 2006. Validation of human behavior
representations. www.dmso.mil/public/library/projects/vva/found_02/sess_papers/b3_harmon.
pdf (accessed June 2006).
Hodges, J.S. 1991. Six (or so) things you can do with a bad model. Operations Research 39: 355–365.
Innis, G. and E. Rexstad. 1983. Simulation model simplification techniques. Simulation 41(1): 7–15.
Kim, H., J.-H. Park, D. Lee and Y.-S. Yang. 2004. Establishing the methodologies for human evacuation
simulation in marine accidents. Computers and Industrial Engineering 46(4): 725–740.
Learmount, D. 2005. Evacuation tool given green light. Flight International 167(4982): 12.
Lee, E.P.K., S. Robinson and J.S. Edwards. 2005. Improving the use of simulation-assisted AI in elic-
iting operations knowledge. In Third European Conference on Intelligent Management Systems in
Operations, pp. 105–115. Birmingham, UK: The Operational Research Society.
31-14 Handbook of Dynamic System Modeling
Liang, T.-P., H. Moskowitz and Y. Yuehwern. 1992. Integrating neural networks and semi-Markovprocesses
for automated knowledge acquisition: an application to real-time scheduling. Decision Sciences
23(6): 1297–1314.
Lyu, J. and A. Gunasekaran. 1997. An intelligent simulation model to evaluate scheduling strategies in a
steel company. International Journal of Systems Science 28(6): 611–616.
Moffat, J. 2000. Representing the command and control process in simulation models of conflict. Journal
of the Operational Research Society 51(4): 431–439.
Nance, R.E. 1994. The conical methodology and the evolution of simulation model development. Annals
of Operations Research 53: 1–45.
Pace, D.K. 2000. Simulation conceptual model development. In Proceedings of the 2000 Spring Simulation
Interoperability Workshop. www.sisostds.org. (accessed July 2005).
Pidd, M. 1999. Just modeling through: a rough guide to modeling. Interfaces 29(2): 118–132.
Pidd, M. 2004. Computer Simulation in Management Science, 5th ed. Chichester, UK: Wiley.
Powell, S.G. 1995. Six key modeling heuristics. Interfaces 25(4): 114–125.
Quinlan, J.R. 1986. Induction of decision trees. Machine Learning 1(1): 81–106.
Robinson, S. 1994. Simulation projects: building the right conceptual model. Industrial Engineering 26(9):
34–36.
Robinson, S. 2001. Soft with a hard centre: discrete-event simulation in facilitation. Journal of the
Operational Research Society 52(8): 905–915.
Robinson, S. 2004. Simulation: The Practice of Model Development and Use. Chichester, UK: Wiley.
Robinson, S., T. Alifantis, J.S. Edwards, J. Ladbrook and T. Waller. 2005. Knowledge based improvement:
simulation and artificial intelligence for identifying and improving human decision making in an
operations system. Journal of the Operational Research Society 56(8): 912–921.
Robinson, S., J.S. Edwards and W. Yongfa. 1998. An expert systems approach to simulating the human
decision maker. In Proceedings of the Winter Simulation Conference 1998, pp. 1541–1545. Piscataway,
NJ: IEEE.
Salt, J. 1993. Simulation should be easy and fun. In Proceedings of the 1993 Winter Simulation Conference,
pp. 1–5. Piscataway, NJ: IEEE.
Schelhorn, T., D. O’Sullivan, M. Hakley and M. Thurstain-Goodwin. 1999. Streets: an agent-based pedes-
trian model. Working Paper Series: Paper 9. University College London: Centre for Advanced Spatial
Analysis.
Schmidt, B. 2000. The Modelling of Human Behaviour. Erlangen, Germany: SCS Publications.
Silverman, B.G., M. Johns, J. Cornwell, and K. O’Brien. 2006. Human behavior models for agents in
simulators and games. Part 1: Enabling science with PMFserv. Presence 15(2): 139–162.
Standridge, C.R. and D. Steward. 2000. Using expert systems for simulation modeling of patient
scheduling. Simulation 75(3): 148–156.
Thomas, A. and P. Charpentier. 2005. Reducing simulation models for scheduling manufacturing facilities.
European Journal of Operational Research 161(1): 111–125.
Ward, S.C. 1989. Arguments for constructively simple models. Journal of the Operational Research Society
40(2): 141–153.
Williams, T. 1996. Simulating the man-in-the-loop. OR Insight 9(4): 17–21.
Zeigler, B.P. 1976. Theory of Modelling and Simulation. New York: Wiley.
32
Military Modeling
Roger Smith
Modelbenders LLC
32.1 Introduction .................................................................32-1
32.2 Applications ..................................................................32-1
32.3 Representation .............................................................32-2
Engineering
•
Virtual
•
Constructive
•
Live
•
Environment
32.4 Dynamics ......................................................................32-4
Movement
•
Perception
•
Exchange
•
Engagement
•
Reasoning
•
Dynamic Environment
32.5 Modeling Approach .....................................................32-8
Physics
•
Mathematic
•
Stochastic
•
Logical Process
•
Artificial Intelligence
32.6 Military Simulation Systems........................................32-11
32.7 Conclusion ...................................................................32-12
32.1 Introduction
The military has always been a very heavy user and innovative developer of modeling techniques and
technologies. The nature of military missions requires that they rehearse missions to better understand
their complex interactions and to estimate outcomes. This need has led them to apply modeling and
simulation to a number of different activities over the last 300 years. In this chapter, we will explore the
major applications of military modeling and will discuss the most common forms of dynamic modeling.
32.2 Applications
The U.S. military has made its own unique definitions of the terms “modeling” and “simulation.” For their
purposes, modeling is often defined as, “a descriptive, functional, or physical representation of a system”
(National Simulation Center, 2000). These representations may take the form of a mathematic equation, a
logical algorithm, a three-dimensional digital image, or a partial physical mock-up of the system. Models
are applied so widely that the variety of systems of interest is almost without bounds. In these systems,
military weapons systems are usually very prominently represented, to include land, air, and sea vehicles;
communications and radar equipment; handheld weapons; and individual soldiers. But models also rep-
resent the decision-making process and automated information processing that occurs inside the human
brain and within battlefield computers. They extend to representations of the environment that is made
up of terrain, vegetation, cultural features, the atmosphere, ocean, and radio frequency (RF) environment.
Different combinations of all of these are needed to accurately represent potential military situations.
One military definition of simulation is, “a system or model that represents activities and interactions
over time. A simulation may be fully automated, or it may be interactive or interruptible” (NSC, 2000).
32-1
32-2 Handbook of Dynamic System Modeling
This definition attempts to encompass human-in-the-loop simulators for training as well as systems that
serve as analytical tools for computing outcomes without the aid of a human participant.
The official categorization of the use of models and simulation within the military is to divide them
into three large application groups.
The first is for use in “research, development, and acquisition.” In these applications, models are used
to provide insight into the cost and performance of military equipment, processes, or missions that are
planned for the future. These use scientific inquiry to discover or revise facts and theories of phenomena,
followed by transformation of these discoveries into physical representations.
The second category is in exploring “advanced concepts and requirements.” These models present
military systems and situations in a form that allows the military to conduct concept exploration and
trade studies into alternatives. These trade studies often explore multiple variations on a new weapon or
tactic and attempt to measure the effectiveness of each of them. The result is a general appreciation for
the different options available and some rough measure for ranking them. The models may be used to
understand physical weapons or equipment, but they may also explore different processes for organizing
and executing a mission. These require an understanding of processes and the interactions that occur
between the different steps in the processes. The models assist the military in creating a doctrine of
operations, constructing an internal organization, and selecting materials for acquisition.
The third category is in “training, education, and military operations.” Models that are embedded
in a simulation system are used to stimulate individuals and groups of personnel with specific military
scenarios. The goal is to determine the degree to which they have learned to execute the doctrines they
have been taught. It also gives them the opportunity to experiment with new ideas and to determine how
useful these might be in a real warfare situation. All of this can be done in a controlled environment that
is free of the life-threatening situations that are part of real combat operations.
Finally, it should be noted that military modeling and simulation has always been the basis for a large
segment of entertainment products. Many of the modeling concepts behind paper board-wargaming in
the 1950s were developed simultaneously by the RAND Corporation for serious military training and by
Charles Roberts at the Avalon Hill game company for popular entertainment (Perla, 1990). This trend
has continued for over 50 years and can be seen today in comparing realistic three-dimensional military
training systems and the product of the very popular computer gaming industry. Systems like America’s
Army provide an environment for experimentation and training in the military, a device to enhance Army
recruitment and education about the military lifestyle, and a game for use by anyone looking for a little
excitement in their free time (America’s Army, 2006).
32.3 Representation
Models, by their very nature are an abstraction or simplification of the real world. Therefore, it is possible
to create an almost infinite number of variations on the representation of objects, actions, and events in
a simulation. Over the past several decades, a number of different types of models have been developed
for representing a military system or mission. These have gradually converged into commonly recognized
categories of representation. These categories have significantly improved the ability of military modelers
to communicate with each other and to exchange models with a better understanding of the differences
between the products being created.
32.3.1 Engineering
Engineering models focus on the details of what a system does. These capture the physical properties
of materials, liquids, aerodynamics, servomechanisms, and computer control of specific systems. They
also include interactions between two physical objects or between an object and its environment. An
engineering model attempts to understand the physical capabilities of the system at a level that is accurate
enough to be used to design the system. Historically, physical prototypes were used to conduct these
Military Modeling 32-3
experiments. However, advanced computer technologies and modeling techniques have allowed us to
create digital models of systems that are nearly as predictive as are live physical tests. These models offer
many advantages over their physical counterparts. They are almost infinitely malleable so that experiments
can be conducted on many thousands of variations rather than just a few physical prototypes. They are
nearly infinitely instrumentable. It is possible to collect data from all points in space and time around
the event of interest. When using physical prototypes we are often limited by our ability to place sensor,
communication, and recording equipment at the precise place and time of interest. Engineering models
become a more prominent part of creating or studying military systems because of the accessibility of the
required computers and more mature methods for representing real systems.
32.3.2 Virtual
A “virtual model” often refers to a three-dimensional representation of a system that is operating in a
digital three-dimensional environment. The focus is usually on the visual appearance of the object and the
environment, more than on the properties of physics that are the focus of engineering models. Because of
its visual focus, the objects most often represented are military vehicles and humans that would appear on
a battlefield. This category is closely aligned with the more popularly recognized term “virtual reality.”
A virtual model and environment are usually constructed to simulate individual soldiers who are
immersed in a system that generates visual, aural, and tactile stimuli. The goal is usually to train, test, or
measure the ability of the human to respond in a desirable manner to the stimuli. Flight simulators are
the most popularly recognized form of these models and systems.
32.3.3 Constructive
A “constructive model” represents objects that are separate from the human user or player, but which are
under the control of this person. The user sends commands to these objects, but is not immersed in the
middle of the battle as he or she would be in a virtual environment. Historically, constructive models have
often been aggregated as well. Rather than representing individual vehicles or people, the model represents
groups of these in an attempt to reduce the number of details about each and to make it possible for the
computer and the human to control many more of them. More recently, constructive simulations that
represent individual objects have become very popular and very powerful. These are often referred to as
semiautomated forces (SAF) systems because of the way that control of the objects is shared between a
human user of the system and intelligent models of the human behavior embedded in the software. A
human may provide the overall mission and direction, but the SAF will supplement this with detailed
control of activities like movement and engagement.
A constructive model may represent a flight of four aircraft as a single item in the simulation or it may
represent each aircraft individually. What separates the constructive from the virtual is usually the method
of human interaction, the lack of a three-dimensional representation of the object, and the number of
objects that are controlled by a single user. A constructivemay also group several hundred vehicles, humans,
and equipment into a single object model. This model must then represent the aggregated behaviors of
its many different constituent parts. There are a number of motivations for this type of modeling. First, it
allows the simulation system developers to capture the operations of a much broader battlefield in a form
that can be run on a reasonable computer suite. Second, in many cases the behavior of groups of objects
are not understood at the engineering or virtual level, but can be represented as a higher-level aggregate.
Third, this type of model mimics the organization, representation, and information that are used in the
real military organizational hierarchy.
Very basic constructive models of military operations can be seen in many board and computer games,
such as Chess, Stratego, and Risk. Constructive simulation systems differ from virtual systems in that the
human operator or player is often positioned outside of the battle. Engagements are not usually targeted
at the human player, so they are in a position to think more strategically about the situation and are not
required to react to individual events that appear to threaten them personally, as would occur in a virtual
system.