ArticlePDF Available

Discrete-event simulation is dead, long live agent-based simulation!



There has been much discussion about why agent-based simulation (ABS) is not as widely used as discrete-event simulation in Operational Research (OR) as it is in neighbouring disciplines such as Computer Science, the Social Sciences or Economics. To consider this issue, a plenary panel was organised at the UK Operational Research Society's Simulation Workshop 2010 (SW10). This paper captures the discussion that took place and addresses the key questions and opportunities regarding ABS that will face the OR community in the future.
Discrete-event simulation is dead, long live agent-based simulation!
P O Siebers
, C M Macal
, J Garnett
, D Buxton
and M Pidd
University of Nottingham, Nottingham, UK
Argonne National Laboratory, Argonne, IL, USA and The University of Chicago,
Chicago, IL, USA
University of the West of Scotland, Paisley, UK
David Buxton, dseConsulting LTD, Quorn, UK
Lancaster University, Lancaster, UK
Correspondence: P O Siebers, IMA Research Group, School of Computer Science,
University of Nottingham, Nottingham, NG8 1BB, UK. E-mail:
There has been much discussion about why agent-based simulation is not as widely
used as discrete-event simulation in Operational Research as it is in neighbouring
disciplines such as Computer Science, the Social Sciences or Economics. To consider
this issue, a plenary panel was organised at the UK Operational Research Society's
Simulation Workshop 2010. This paper captures the discussion that took place and
addresses the key questions and opportunities regarding agent-based simulation that
will face the Operational Research community in the future.
Keywords: discrete-event simulation, agent-based simulation, panel discussion, the
Discrete-Event Simulation (DES) has been the mainstay of the Operational Research
(OR) simulation community for over 40 years. The arrival of Agent-Based Simulation
(ABS) in the early 1990s promised to offer something novel, interesting and
potentially highly applicable to OR. However, there is relatively little evidence that
ABS is much used in the OR community, there being few publications relating to its
use in OR and OR-related simulation journal. This contrasts with the much greater
volume of ABS papers in journals from disciplines such as Computer Science, the
Social Sciences and Economics.
Why is this the case? Has the OR-related simulation community missed the ABS bus?
Are they monolithically stuck in a proverbial rut? Will they soon be outshone and
shown to be outdated by simulation work in other communities? Can we expect DES
to be completely redundant and incapable of solving the key problems in OR?
This paper is capturing the panel discussion on this topic which was initiated by Peer-
Olaf Siebers and has been held on the OR SW2010 in order to help answer these
questions and perhaps to develop a new image for ABS in OR. The panellists were
Charles Macal from the Center for Complex Adaptive Systems Simulation (USA),
Peer-Olaf Siebers from the University of Nottingham (UK), Jeremy Garnett from the
University of the West of Scotland (UK) and David Buxton from dseConsulting LTD
(UK). The panel was chaired by Michael Pidd from Lancaster University (UK).
In Section 2 we will discuss the questions raised above in a structured way. The
conclusions we have drawn from this discussions are presented in Section 3, which
also features a list of topics we thought would be important to discuss but could not
deal with in this panel discussion due to time constraints. We would like to encourage
the OR simulation community to discuss these topics at another panel.
The panel was organised in the following way. Before the panel took place four
questions were defined in order to structure the discussion and each panellist was
asked to prepare a five minute presentation to express his view on the question given.
Once the panel member had finished his presentation the question was passed on to
the other panel members to add their views. Finally, the question was opened up to the
audience to express their thoughts and discuss with the panellists. We used the same
structure for writing up the discussion of the four questions, which is presented in this
section. In addition we asked each panellist to write down their individual conclusions
from this panel discussion at the end of their sub-section.
Question 1, presented by Charles Macal: What kind of phenomena can ABS help
us to understand better?
Q1a: Summary of presentation
The short answer to this question goes like this. ABS will help us better understand
real-world systems in which the representation or modelling of many individuals is
important and for which the individuals have autonomous behaviours (i.e., actions are
not scripted but agents respond to the simulated environment).
The kinds of problems having these specific requirements arise in a surprisingly large
number of and variety of application areas. I’ve looked at the peer-reviewed literature
on Agent-Based Modelling (ABM) applications and found them strewn across many
disciplines and publication outlets. I looked at a sampling of 20 papers (which I
referenced in my keynote address) and distilled from them the reasons that the authors
contend for why they used ABM in their work. The single most given reason given
boils down to, and I am paraphrasing many papers here, essentially the same thing:
Agent-based models can explicitly model the complexity arising from individual
actions and interactions that arise in the real world. In other words, ABS allows
people to model their real-world systems of interest in ways that were either not
possible or not readily accommodated using traditional modelling techniques, such as
DES or System Dynamics (SD).
Although the OR community is perhaps lagging in applications of ABS, I attribute
this to the wide availability and experience of the OR community with DES software.
People use the modelling techniques that they are familiar with and attempt to fit
these to their problems, and go as long as they can. I see the situation as very likely
changing in the future as the number of people developing “agent-type” models grows,
for I see the OR community being called upon to address new kinds of problems that
have not been adequately addressed by DES, or SD approaches, and not readily
modelled using existing simulation software, toolkits, or development environments.
An example is epidemic or pandemic modelling. For a long time, differential equation
(and SD) models of disease states known as SIR models (for Susceptible-Infected-
Recovered population states), have been the primary modelling technique for
understanding the spread of infectious disease. These models have been very valuable
in providing information about tipping points and informing policy decisions. But it is
now recognised that these models are not adequate for modelling the human
behavioural aspects that are important in disease transmission and epidemic dynamics.
For including behaviour into such models, ABS is a natural approach.
On the more traditional operations front, agile manufacturing and dynamic supply
chains are natural application areas, if these applications require the modelling of
processes that are dynamic and must quickly adapt to changing requirements and
events on a real-time basis. For example, through ABS we can include descriptive
models of how people actually make decisions within a supply chain and see the
effects of all decision makers on the supply chain. In contrast, most operations models
take a normative approach, i.e., indicating what should be done rather than how the
system really works, such as EOQ (Economic-Order-Quantity) models for inventory
planning. In many knowledge domains, it can be straightforward to understand how
people make decisions, for example, how they forecast demand, how they decide
when to order to inventory, or how they select suppliers, etc. For that matter, one can
also develop some very interesting and useful queuing models in the traditional DES
applications domain, that incorporate general agent behaviours. For example, we can
model how people actually behave in the evacuation of a building or an area when
designing evacuation strategies.
In summary, I keep a running list of features for a problem to have that make it a good
candidate for an application of ABS, such as:
When the problem has a natural representation as agents when the goal is
modelling the behaviours of individuals in a diverse population
When agents have relationships with other agents, especially dynamic
relationships - agent relationships form and dissipate, e.g., structured contact,
social networks
When it is important that individual agents have spatial or geo-spatial aspects
to their behaviours (e.g., agents move over a landscape)
When it is important that agents learn or adapt, or populations adapt
When agents engage in strategic behaviour, and anticipate other agents’
reactions when making their decisions
When it is important to model agents that cooperate, collude, or form
When the past is not a predictor of the future (e.g., new markets that do not
currently exist)
When scale-up to arbitrary levels is important, i.e., extensibility
When process structural change needs to be a result of the model, rather than
an input to the model (e.g., agents decide what process to go to next)
Also, on a somewhat lighter note, I can say from experience that ABM tends to be a
fun way of doing simulation! Understanding how people behave and why, based on
interviews and observation of their behaviours, and then being able to implement the
behaviours in agent-based models to see their cumulative effects can be a very
interesting and educational experience.
Q1b: Summary of panel responses
There were many responses from the panel on the presentation which we summarise
here. Choosing between DES and ABS should be based on the problem requirements
rather than the application domain. There is a danger in considering the categories of
application rather than on the nature of the underlying research questions that drive
the applications. The discussion on the applicability of ABS could become one of a
tool looking for a problem, as in the adage “a hammer looking for a nail.” Good
modelling (and Management Science) practice dictates that you should identify the
research question, first, and then ask what methods would be most applicable in
solving it, second. So it would be valuable in moving the discussion forward to
enumerate the kinds of questions that ABS is being used to (uniquely) answer in
various application domains. One of the panellists stated that they had tried to
implement human behaviour models in a DES model, but did not succeed and are now
successfully using ABS for this problem. Another panellist stated that he has an SD
model that has become too complex to understand and is considering recasting the
problem as an ABS with the thought that the ABS would be easier to understand as
the model grows in complexity. In his view, the kinds of problems for which ABS is
applicable are similar to the kinds of problems for which SD is applicable. The
inherent lower level abstraction is easier to understand with ABS.
Q1c: Summary of audience comments and panel responses to these comments
A question came from the audience about for what kinds of problems I would
recommend to others using DES. I answered that DES is useful for problems that
consist of queuing simulations or complex network of queues, in which the processes
can be well defined and their emphasis is on representing uncertainty through
stochastic distributions. Many of these applications occur in manufacturing and
service industries as well as queuing situations. People tend to use whatever approach
that they are familiar with to model their problem and then switch into another
technique if and when their original approach turns out to be inadequate to fully
address the problem requirements. This multi-step approach to building models is
what we recommend in the book that I wrote with Michael North (North and Macal,
2007): Start out by using what you know and what you have, and build the model
through phases. For example, in the book we begin ABM by using SpreadSheets (SS).
Eventually the SS model will no longer scaled-up in terms of the number of agents,
complexity of agent behaviours that can be represented, etc., but serves an important
purpose in understanding the problem requirements for a full-scale model that were
not fully evident at the initiation of the project. A question came from the audience
concerning the possible applicability of ABS to sports, particularly team sports. Team
sports offers a rich set of modelling opportunities for ABS, for ABS could
hypothetically consider, in addition to individual players' abilities, the unique
interactions that occur between pairs of players.
Q1d: Concluding remarks
The OR community has plenty of experience with process modelling through DES.
New problems for the OR community require modelling of populations of diverse
individuals having a variety of behaviors and interactions. ABS has been developed to
address just this type of problem because existing simulation and DES software was
not adequate for these types of problems. The ABS paradigm, in terms of software,
naturally follows the object-oriented software paradigm. This paradigm allows
extensibility, for modelling more and more agents, more and more behaviours, and the
full range of diversity required of large-scale, real-world applications. A closer
relationship of the OR community to the Computer Science community would
facilitate the transfer of tools and software approaches to the OR community and be
beneficial for furthering the ABS modelling enterprise. New modelling procedures or
processes are needed to fully encompass the challenges of ABS. These include
procedures for encoding agent behaviours and developing agent model components.
In addition, validating agent behavioural model components is a new challenge.
Question 2, presented by Peer-Olaf Siebers: How can we define ABS in OR and
how does it differ from DES?
Q2a: Summary of presentation
Before starting the debate on this question panel members and audience were asked to
raise their hand if they have a clear idea of what ABS in an OR context actually
means. As expected only very few people in the audience and only half the panel
members raised there hands.
The first part of the question focuses on defining some vocabulary. It is important to
have a definition for the terminology we use (e.g. ABS, ABM, agent) to which we as
the OR community can relate. Here we give some broad definitions that we think are
useful in finding definitions that we can all agree on. In order to define ABS we adapt
the definition of what constitutes a simulation by Shannon (1975): ABS is the process
of designing an ABM of a real system and conducting experiments with this model
for the purpose of understanding the behaviour of the system and/or evaluating
various strategies for the operation of the system. In ABMs a complex system is
represented by a collection of agents that are programmed to follow some (often very
simple) behaviour rules. System properties emerge from its constituent agent
interactions (Bonabeau, 2002). Agents are “objects with attitudes” (Bradshaw, 1997).
They are discrete entities that are designed to mimic the behaviour of their real world
counterparts. Agents have each their own set of goals and behaviours and their own
thread of control. Unlike objects, agents are capable of making autonomous decisions
(i.e. they are able to take flexible action in reaction their environment) and agents are
capable of showing proactive behaviour (i.e. actions depend on motivations generated
from their internal state).
The second part of the question focuses on comparing the two different model types,
typical DES and ABS models. Table 1 presents a summary of the attributes that allow
classifying a model to be of either one type or the other.
DES models
ABS models
Process oriented (top down modelling approach);
focus is on modelling the system in detail, not the
Individual based (bottom up modelling approach);
focus is on modelling the entities and interactions
between them
Top down modelling approach
Bottom up modelling approach
One thread of control (centralised) Each agent has its own thread of control
Passive entities, i.e. something is done to the
entities while they move trough the system;
intelligence (e.g. decision making) is modelled as
Active entities, i.e. the entities themselves can
take on the initiative to do something; intelligence
is represented within each individual entity
Queues are a key element
No concept of queues
Flow of entities through a system; macro
behaviour is modelled
No concept of flows; macro behaviour is not
modelled, it emerges from the micro decisions of
the individual agents
Input distributions are often based on
collect/measured (objective) data
Input distributions are often based on theories or
subjective data
Table 1: Attributes that define the model type
Looking at these definitions we come to the conclusion that true ABS models in OR
do not exist. Instead, in OR we have combined DES/ABS models where we represent
the process flow as a DES model and then add some active entities (to replace the
passive DES entities) that are autonomous and can display proactive behaviour.
Q2b: Summary of panel responses
There was a general agreement by the other panel members regarding the conclusions.
However, some additional comments were made. One panel member stated that most
of the discussions around the topic only focus on the content of the model while other
important topics like methodologies and practice are often neglected. Another panel
member commented that it is high time that academics decide on names, definitions,
and frameworks quickly, so that we can all get on with our business. He reminded
everyone that the choice of method should be problem driven. Finally a panel member
referred to ongoing discussions he is having with Averill Law (Averill M. Law &
Associates, Inc.), who produced a DES model with active entities and concluded that
it was unclear what ABS had to offer beyond what DES already offers. However, the
fact is that currently very few DES models do take an approach in which the entities
are really the centre of focus and in which they exhibit active behaviour. If we say it is
impossible for DES to include active entities we could clearly differentiate DES and
ABS (in OR).
Q2c: Summary of audience comments and panel responses to these comments
Some people in the audience realised from the definitions given above that they must
have used some form of ABS but were not aware of it. In addition a healthcare
example was given that challenged the above statement that ABS in OR does not exist.
This model which dates back to the early 1990s looked at HIV infection (Brailsford et
al, 1992). It was used for testing different strategies for controlling the disease. Some
of the interventions tested were behavioural interventions. In this manually coded
model the people’s own behaviour determined what happened to them next, which
today seems like a typical ABS approach. However, a DES structure (process) was
forced on top which was feeling unnatural at the time but was the standard in OR. The
question was raised if this is really an OR application or in fact a Social Science
application of behavioural change which lead to a short discussion if Social Science is
part of OR. It was agreed that this question cannot be solved in this panel. Finally, it
was discussed when a model should be called agent-based. The general opinion was
that this is often decided at the conceptual and not on the implementation level. Some
indicators for agent-based models are that they are object oriented, that the entities
become the focus of interest (they are driving the system) and that unique decision
making internal to the entity is modelled. The perspective you take (agent perspective
or process perspective) defines how you implement your model.
Q2d: Concluding remarks
The application of combined DES/ABS seems to be the way forward to tackle the
problems in what becomes more an investigation into behavioural OR due to the
recent shift of attention from manufacturing to service industry. Software vendors
need to respond to this demand and develop easy to use software that allows adding
intelligence to the entities themselves, rather than having all the intelligence stored in
the process flow definition. Academia has now started to add ABS to their research
toolbox while the industry is still struggling, having issues regarding the credibility of
the simulation results.
Question 3, presented by Jeremy Garnett: Is there a difference between ABS as
used in disciplines such as Computer Science, Social Science and Economics and
ABS as used in OR?
Q3a: Summary of presentation
Historically, DES has always been closely linked to OR. DES software developers
have always been closely aligned with the ideas and objectives of OR. Most of the
major developments concerning methodology and practice have come from within the
OR community. Not so with ABS, here it has been quite a different story. ABS is
associated with a large number of different disciplines but not particularly with OR.
The development and use of ABS has been driven by computer scientists, economists,
biologists and sociologists each with their own particular goals and objectives. A
discussion about the use of ABS within OR therefore needs to review how other
disciplines use it, and decide what we can learn from those disciplines. Do different
disciplines use ABS in the same way as we do in OR? What can OR learn from other
disciplines? What should be OR's main contribution to the application of ABS?
For many people, their first awareness of and interest in ABS was prompted by
reading about Complexity theory and the setting up of the Santa Fe Institute in the
1990’s (Waldrop, 1994). The Santa Fe Institute was a deliberate attempt to bring
experts together from different disciplines and for them to share and exchange
problems and ideas. This resulted in various models which demonstrated how simple
rules of interaction can lead to complex behaviours. There are many well-known
models from this time, such as the ‘game of life’, ‘flocking’, slime mould and
Schelling segregation. These examples are of great academic interest; in particular,
they seem to point towards universal, fundamental theories of nature. However, they
are very theoretical, and none of them are based on actual implementations. Therefore,
they are of limited relevance to the practice of OR; OR is not generally concerned
with uncovering fundamental theories of nature. There are a few examples of general
theories in the OR text books, such as Little’s Law for queuing systems. But theory, in
OR terms generally concerns problem-solving methodologies. In the 1990's ABS
seemed more relevant to academic research in the pure sciences than to an applied
discipline such as OR.
Q3b: Summary of panel responses
The panel discussed how ABS has developed and evolved rapidly over the last ten
years. Whilst many of those early examples were concerned with general theories,
there is now an increasing body of literature documenting successful applications of
ABS. These examples concern traffic and transportation, financial markets, supply
chains, energy usage, health and social policies (see Macal and North, 2007). These
applications show that ABS is of much more relevance to OR than originally thought.
On the other hand, many of these projects are conducted by people who would see
themselves primarily as economists, biologists or sociologists, not as Operational
Q3c: Summary of audience comments and panel responses to these comments
The audience raised a number of points on the topic of different disciplines: OR with
its strong problem-solving focus is much closer to engineering than any pure science.
The focus of OR is on decision support for problem owners, and its theories concern
the use of models and problem-solving methodologies. OR is also more concerned
than other disciplines with the practice of modelling. It is in these areas where OR can
make useful contributions to the application of ABS. For example, the collection and
use of good data from real systems is a problem common to most simulation projects.
OR can also provide guidance in the relationship between modellers and stakeholders,
such as improving the understanding of those stakeholders. One audience member
stated that it is important to consider the purpose of any simulation project. Objectives
such as greater system understanding, predicting future behaviour and providing
decision support are all objectives familiar to OR. Using an ABS application to test an
underlying theory is not a common OR objective.
Panel and audience agreed that OR practitioners can usefully conduct ABS projects
within a range of different problem domains. These projects generally require a range
of skills. An ABS model may be based on underlying theories, and require detailed
knowledge about a particular domain, such as how an infection is transmitted, or how
a particular network structure shapes communication between customers. ABS might
be considered as a tool to develop the relationship between the pure and applied
branches of a particular discipline, in much the same way as experimental and
theoretical physics have always been closely linked. Behavioural economics, for
example, is one discipline that has attracted much recent interest (Levitt and Dubner,
2005). Potentially, it will provide a useful alternative theory to rational choice theory.
However, there remains much to do in developing and testing that theory, quite
probably using ABS.
Q3d: Concluding remarks
As we learn to successfully apply ABS within OR, it is important to consider the use
of ABS by other disciplines. It is important to learn from and work with those
disciplines. We can spot opportunities for applications to business problems which
might not be tackled by other disciplines. Where other disciplines are already using
ABS, we can provide guidance for how best to use it.
Question 4, presented by David Buxton: DES is well known and used by both
practitioners and academics. What barriers exist for ABS to achieve such universal
Q4a: Summary of presentation
The emergence of ABS as a technique in OR is timely. Globalised business is a highly
complex management process, and making decisions in this environment is not well
supported by the current set of tools, including DES (North and Macal, 2007). ABS is
not a new technique; as far back as the year 2000 the SD practicioner John Sterman
noted that the technique presented a huge opportunity to progress simulation methods
and enhanced simulation applications (Sterman, 2000). However, today, 10 years later,
the adoption of the technique has not yet filtered into the mainstream, either within the
academic community, although evidence suggests that this is increasing, and certainly
not within industry.
The modern history of DES is dominated by the software and computers. During the
1970s early general simulation languages such as SIMAN, latterly Arena and
ExtendSim provided software for simulating complex queuing theory and resource
allocations problems. The increasing computer power and evolving user interfaces led
the software and hence the DES technique to progressively move towards “drag and
drop”. Languages, such as the Simul8, then emerged to make the process accessible
and cost effective for all business sizes. The technique’s use in industry did not truly
mature until the early 1990s and the acceptance of DES as a management tool (not
just a researcher tool) is clearly linked to the development of software application
(Kelton et al, 2003).
This presents the first barrier for ABS. Although there are a number of excellent
academically developed tools (Repast, NetLogo), the commercially available software
is limited to AnyLogic, and all of these products expect knowledge of object oriented
programming techniques and the modeller needs to be comfortable with Java. These
are not traits that the average manager has focussed on developing during his career.
For this reason, ABS remains the domain of a relatively few skilled experts and
academic researchers.
The first challenge is therefore for the software development community, working in
collaboration with current users to establish how and where software can simplify the
more technical aspects of ABS and reduce this barrier to entry. Reducing the amount
of java code to be written is a must.
Evidence of the second challenge can be seen in the title of this debate “DES is dead –
long live ABS”. I do not consider ABS to be a replacement for all the alternative
mature techniques. It is not a panacea that will answer all questions, quicker, better
and more robustly than either SD or DES. It is, however, a highly flexible modelling
approach that can answer a range of questions better than anything else I have come
across in my career. In considering that ABS can and should the tool to be used to
answer all things, then the debate continues to be centred on methodology rather than
problems and opportunities. Resolve the debate and acceptance of ABS will follow.
Perhaps we should therefore be concentrating research on establishing a clear and
accepted problem driven general framework for the answer to Q1, where the benefits
and drawbacks of all of the simulation approaches are highlighted.
The third and final challenge I identify is the current set of managers! The dominant
tool is use in industry is Excel, anecdotally, I find awareness of other quantitative
decision making tools low, including for DES simulation. Excel alone does not
represent good preparation for the current range global business challenges and we
need to ensure that the new generation of managers in production at Universities
understands all the available tools, where they can be applied and what benefits can be
delivered. ABS should be core to this. Taught undergraduate, post-graduate or
executive courses in ABS are few and far between, reinforcing low awareness and
low adoption. Given the literature about the potential benefits of ABS (North and
Macal, 2007) surely we have not done enough yet to equip the students fully.
Q4b: Summary of panel responses
The response of the panel reinforced the need to increase awareness of ABS in the
field. It was noted that there was huge demand for ABS learning but very little supply.
The panellists also highlighted that a key development in DES awareness, currently
lacking for ABS, was the publication of tool centred training material, such as the
textbook on the simulation software Arena from Kelton et al (2003).
One of the panellists raised the complex issue of validation, suggesting that this may
be responsible for low adoption by DES practitioners. DES has established rules for
validation which cannot be directly transferred to ABS.
It was felt that one of the major obstructions to ABS adoption and clearly linked with
the low software maturity, was the current model development time. Simulation has
benefitted from the ability to do quick and dirty models that get 80% of the answer in
a short time frame. Moving to ABS would represent a step backwards for many
simulation researchers and practitioners used to the benefits of a mature tool. This
view was supported by one of the other panellist, who suggested that maybe the best
technique to apply regardless of the problem is the one the modeller knows best.
Q4c: Summary of audience comments and panel responses to these comments
There was a general consensus that software tools were a fundamental problem.
Without visual approaches to ABS then it will always be too technical for mass
adoption and difficult to integrate into teaching where skills and interests differ
amongst students; not all OR students will be interested in mastering Java. One of the
panellist responded by suggesting it may be possible to create a series of blueprints or
templates for typical agent actions and behaviours. However, substantial further work
in this area will be needed.
A member of the audience commented that validation should not be a barrier. Raising
the example of SD, it was noted that this techniques also suffer from the same
validation questions but this has not proved a substantial barrier. SD has addressed
problems by focussing on validating the inputs and the emergence.
A final comment came from the chairman who said that DES is so popular due to the
availability of good (easy to use) software. It is the issue in ABS; software supporting
the implementation of models is not so widely available. A way around this might be
if DES vendors would think about offering integrated development environments for
their current products that would better support the implementation of ABS concepts,
e.g. allowing to define the model using a state chart approach.
Q4d: Concluding remarks
For a practitioner ABS is a more natural way to simulation and achieves better buy in
from customers. It allows simulation to be applied to a new range of problems more
robustly without many of the workarounds that need to be applied when using SD or
DES. The highly technical nature of the current software is a significant barrier to
acceptance by the wider community. More fundamentally, the OR community needs
to accept and define the roles for each of the simulation techniques.
In conclusion we can say that the discussion has been very lively and the panel has
sparked a lot of interest in the topic discussed during and after the panel. We would
like to thank everyone who contributed to it. It has been a great success and we hope
that we will have some follow-up activities from this discussion.
The key finding of the panel has been that there is lots of interest in using ABS in
academia and industry but most people don't know how to apply it. There are no
established frameworks or methodologies to guide researchers and analysts through
the agent-based modelling and simulation process, there is no specific guidance on
ABS output analysis, there are no easy to use drag-and-drop agent-based modelling
and simulation tools, and there are no text books focussing on practitioner needs. All
of this leads to ABS not getting a foot in the door in OR.
But changes are within reach. We are currently in the process of setting up a Special
Interest group for ABS in the OR Society that will further investigate the issues raised
here and promote the application of ABS and combined DES/ABS in OR. Also, the
software vendors in the panel audience showed some interest in the discussion around
extending their existing software packages to accompany some ABS ideas inside their
DES software, e.g. developing some agent templates for drag-and-drop or defining
entities through state charts.
In Warwick, we had an OR-only panel and audience for our plenary panel. It would
be interesting to hear and compare the views of members of other communities, e.g.
Computer Science, Social Sciences, and Economics on the topics discussed. However,
we hope this will be the topic of another paper.
Bonabeau E (2002). Agent-based modeling: Methods and techniques for
simulating human systems. In: Proceedings of the National Academy of
Science of the USA. 99:7280-7287.
Bradshaw J (1997). Software agents, MIT Press, Cambridge, MA. (cited in
Odell J (2002). Objects and agents compared, Journal of Object Technology,
Brailsford SC, Shahani AK, Basu Roy R, and Sivapalan S (1992). Simulation
modelling for HIV infection and AIDS. International Journal of Biomedical
Computing, 31:83-88.
Kelton WD, Sadowski RP, Sturrock DT (2003). Simulation with Arena, 3
Ed., McGraw-Hill: New York
Levitt S and Dubner S (2005) Freakonomics: A rogue economist explores the
hidden side of everything, Allen Lane: London.
Macal, CM and North MJ (2007). Agent-based modeling and simulation:
Desktop ABMS. In: Proceedings of the 2007 Winter Simulation Conference.
Eds: Henderson SG, Biller B, Hsieh MH, Shortle J, Tew JD and Barton RR,
Washington DC, Dec. 9-12.
North MJ and Macal CM (2007). Managing business complexity: Discovering
strategic solutions with agent-based modeling and simulation, Oxford
University Press: New York.
Shannon RE (1975). Systems simulation: The art and science. Prentice-Hall:
Englewood Cliffs, NJ.
Sterman J (2000). Business Dynamics: Systems thinking and modeling for a
complex world, Irwin McGraw-Hill: New York.
Waldrop MM (1994). Complexity: The emerging science at the edge of order
and chaos, Penguin Books: Harmondsworth, England.
Agent-Based Simulation (ABS) is a novel technique that is rapidly gaining an
international 'following'. However, in fields like OR/MS it is still in its infants. This
paper captures the discussion that took place at the UK Operational Research
Society's Simulation Workshop 2010 and addresses the key questions and
opportunities regarding ABS that will face the OR community in the future.
... Data-driven models can provide more accurate forecasts at the expense of explicit modeling of propagation mechanisms. Methods such as agent-based simulation (ABS) [21] and machine learning (ML) methods have been employed for infectious disease outbreak analysis and disease prediction. ...
... After visual inspection and evaluation of the characteristics of many sets of identified clusters, we choose the HC clusters with k = 9 as a good balance to support the objectives of this study, i.e., to identify clusters of reasonable size and similarity that also reflect a level of regionally specific diversity that can be leveraged to support public health decision-making. (44) days-UnhealthyMental (30) NHIA (28) under18 (21) health-insurance (39) under18 (30) gender-ratio (26) dem-rep-ratio (21) pct-4yr-degree+ (33) pct-Smokers (24) median-income (22) PC2-wx (20) pct-Smokers (29) over65 (24) pop-density (18) health-insurance (19) pct-woHSdiploma (23) ...
COVID-19 is a global pandemic threatening the lives and livelihood of millions of people across the world. Due to its novelty and quick spread, scientists have had difficulty in creating accurate forecasts for this disease. In part, this is due to variation in human behavior and environmental factors that impact disease propagation. This is especially true for regionally specific predictive models due to either limited case histories or other unique factors characterizing the region. This paper employs both supervised and unsupervised methods to identify the critical county-level demographic, mobility, weather, medical capacity, and health related county-level factors for studying COVID-19 propagation prior to the widespread availability of a vaccine. We use this feature subspace to aggregate counties into meaningful clusters to support more refined disease analysis efforts.
... Agent-based simulation (ABS), on the other hand, is a rather up-and-coming approach focusing on self-governing individual agents (Maidstone 2012). ABS is deemed highly suitable to model the specific behaviour of and interactions between agents, whereas DES is more top-down and drawn to modelling the system in detail (Siebers et al. 2010). The ultimate choice between these two types is, however, entirely up to the researcher. ...
... The historical development and use of simulation in OR modelling may lie at the base of this underrepresentation. Namely, discrete-event simulation has been the most prominent simulation type until recently (Siebers et al. 2010) (see Table 4). As discussed earlier, this simulation type is oriented towards systemic process modelling, rather than a detailed examination of agents within the system. ...
At the core of every high-performing warehouse is an efficient order picking (OP) system. To attain such a system, policy choices should be carefully aligned with subjects responsible for the actual picking within the established system. Despite recent advancements in automating the picking process due to Industry 4.0, human operators will continue to play a crucial role in the future of warehousing. However, unlike robots, human operators have specific skills, conduct, and perceptions, which are only partly accounted for in current planning models. This review adopts a multimethod approach to identify and analyse how these phenomena are currently integrated into OP planning problems. In addition, we assess the relevance and adequacy of human factors modelling in academic literature with practice-based insights gathered via semi-structured interviews. This leads to five major human factors integration constructs and dedicated recommendations on how to refine them. We then take the analysis one step further and make suggestions on how to integrate these constructs with leading research methodologies in the context of Industry 5.0. The results highlight the prevalent need to increasingly account for psychosocial phenomena and their impact on operational performance. Future research opportunities provide a substantiated foundation to assist in human-centric work design.
... In the category of people as agents, the construction site is an environment of increasing interest for ABMS applications. As dynamic environments with many uncertainties, the movement of people on construction sites is notoriously hard to model and therefore a domain well suited for ABMS [126]. Thus, the applications of crowd simulations are extended from static environments, e.g., the evacuation of completed buildings, to dynamic environments such as construction sites. ...
Over the last two decades, the use of agent-based models (ABMs) to model and simulate the dynamics of complex systems has increased significantly among various scientific fields, including architecture. Based on a systematic literature review, this paper presents a classification for agent-based modeling and simulation (ABMS) in architecture based on the individual entities being modeled as agents. The classification is based on a reproducible search method capable of incorporating findings from different domain-specific databases to systematically retrieve relevant literature for ABMS in architecture. Subsequently, in each of the ABMs encountered in the selected literature, we identify what entity an agent in the model represents. Based on this identification, a comprehensive classification for ABMs in architecture is achieved. By describing each of the resulting categories, we provide new insights into the field of ABMS in architecture. Finally, we discuss limitations, as well as future trends and possibilities for ABMS in architecture.
... Simulation provides a tool for evaluating various strategies for the system's operation. Discrete-event simulations (DES) are well suited for modeling systems with complex queuing theory and resource allocation problems (32). This study uses DES to examine the impacts of various design and exogenous parameters on UAT fleet performance. ...
Urban air taxi (UAT) is envisioned as a point-to-point, (nearly) on-demand, and per-seat operation of passenger-carrying urban air mobility (UAM) in its mature state. A high flight load factor has been identified as one of the influential components in the successful operation of UAT. However, the uncertainties in demand, aircraft technology, and concept of operations have raised doubts about the viability of UAT. This study examines the impacts of exogenous parameters, such as demand intensity, demand spread, and ground speed, in addition to design parameters, including aerial speed, maximum acceptable delay, and reservations on average load factor and rate of rejected requests. The dynamic and stochastic problem of UAT fleet operation is studied by implementing a dynamic framework that aims to provide a solution to the problem via a discrete-event simulation. The results highlight the significance of demand spread, ground speed, and maximum acceptable delay in demand consolidation. Therefore, to ensure a high aircraft load factor, the UAT operator should specify the maximum acceptable delay and reservation time window given the demand pattern and ground-based transportation in the network.
... Recently, increased attention has been given to decentralized approaches, specifically agent-based simulation (ABS) [18]. According to the ABS approach, the general behavior of the system emerges from the properties of the agent, its rules, and the interaction with other agents, which in turn influence their behaviors [19,20]. This paper is the second part of a study that uses a stochastic ABS model in a cow-calf operation to evaluate the performance of reproductive strategies. ...
In a companion paper, Ojeda-Rojas et al. (2021) [1] describe a stochastic agent-based simulation (ABS) model of a cow-calf operation on a commercial farm in São Paulo, Brazil. The model's parameterization was based on data collected from two sources: a real beef cattle herd and related scientific literature. Based on the mentioned simulation model, this study aims to assess the economic outcome of 10 different reproductive scenarios: Natural mating only (ONM); one timed artificial insemination (TAI) plus natural mating (NM) (1TAI + NM); two TAI plus NM, with 24, 32, and 40 days between TAI (2TAI/24 + NM, 2TAI/32 + NM, and 2TAI/40 + NM, respectively); three TAI without NM, with 24, 32, and 40 days between TAI (3TAI/24, 3TAI/32, and 3TAI/40, respectively); and three TAI plus NM, with an interval between TAI of 24 (3TAI/24 + NM) and 32 days (3TAI/32 + NM). The simulation was performed on an animal-by-animal basis over a time horizon of 5000 days. Each scenario had 32 farms, and each farm kept up to 400 adult females. According to the scenario, a bull population was composed of 0, 7, or 15 individuals. The outcomes, represented as means ± standard deviations, were assessed after reaching a steady-state (1825 days). The model outcomes showed that the 3TAI/24 + NM scenario resulted in higher incomes (US$ 96,479.19 ± 709.81), whereas the ONM scenario had the lowest incomes (US$ 79,753.37 ± 741.87). The 3TAI/24 + NM (US$ 101.720.6 ± 79.21) and ONM (US$ 90.898.58 ± 59.17) scenarios presented the highest and lowest total operating costs (TOC), respectively. However, when TOC was evaluated per kg of the weaned calf, the highest and lowest costs were associated with the ONM (US$ 2.8 ± 0.03/kg) and 2TAI/24 + NM (US$ 2.17 ± 0, 04/kg) scenarios, respectively. Our model suggests that reproductive strategies that use TAI have a better economic performance than those under NM. However, when performing three TAI with an interval of 40 days, the benefit was lower; in some cases, it was even worse than the ONM. Combining TAI with early pregnancy diagnosis resulted in better economic performance than other TAI programs and NM. The 2TAI/24 + NM scenario outperformed the others due to the contrast between its high income and moderate costs. Beef cattle production is a highly complex system. Simulations models, specifically ABS models, could make the decision-making process on complex systems straightforward and effective. Furthermore, ABS models can overcome the limitations of conventional research approaches, such as high costs and long experimentation periods.
... Such tools are constantly evolving as vendors try to respond to demand: most SD packages can now employ probabilistic sampling, and some DES tools can incorporate continuous flows. Following a lively debate about the future of DES at the 2010 UK OR Society Simulation Workshop, Siebers et al. (2010) suggested that DES software could be improved by including statecharts. It seems the vendors of Simul8 were listening: the developers subsequently incorporated statechart functionality into the Professional version of the package. ...
Full-text available
Discrete-event simulation (DES) has been recognised for many years as a powerful tool to support the commissioning and resourcing of health and social care services, due to its ability to capture real-world variability. However, the complex interactions between two distinct but clearly related processes, disease progression and care provision, can lead to such models being cumbersome and lacking in transparency. Representing disease progression as a series of queues and activities is not always intuitive to a non-modeller. This paper presents a novel hybrid simulation approach in which health status is modelled using statecharts, thus combining DES with agent-based simulation. This hybrid approach allows disease progression to be modelled in a more natural way, keeping the overall model structure relatively simple. The approach is illustrated by a case study that evaluates the impact of telecare services for supporting people with dementia.
Optimisation of factories, a cornerstone of production engineering for the past half century, relies on formulating the challenges with limited degrees of freedom. In this paper, technological advances are reviewed to propose a “daydreaming” framework for factories that use their cognitive capacity for looking into the future or “foresighting”. Assessing and learning from the possible eventualities enable breakthroughs with many degrees of freedom and make daydreaming factories antifragile. In these factories with augmented and reciprocal learning and foresighting processes, revolutionary reactions to external and internal stimuli are unnecessary and industrial co-evolution of people, processes and products will replace industrial revolutions.
We present a hybrid simulation methodology designed to support freight rail operations in the mining industry. We aim to bridge the gap between hybrid simulation modelling research and practice. Through discussion of a case study, we contribute to the hybrid simulation methodological literature, explaining why at a conceptual level the hybrid model design and adopted modelling frame are well suited to the problem at hand. The methodology we present can be used in mining freight rail operations planning to determine train destinations across a network and generate a feasible timetable that satisfies operational needs. The method combines discrete-event simulation and agent-based modelling with heuristics to govern train movements destination selection, incorporating an ensemble of simulation runs. We demonstrate the capability of our method to produce a train timetable that satisfies the requirements of the mining operation. Choosing optimal destinations from many options for a large fleet of trains in a vast network is a significant computational challenge (NP-hard in the general case). The method presented significantly reduces the parameter space for which full enumeration of all options would not be computationally tractable.
Conference Paper
Full-text available
Agent-based modeling and simulation (ABMS) is a new approach to modeling systems comprised of autonomous, interacting agents. ABMS promises to have far-reaching effects on the way that businesses use computers to support decision-making and researchers use electronic laboratories to support their research. Some have gone so far as to contend that ABMS "is a third way of doing science," in addition to traditional deductive and inductive reasoning (Axelrod 1997b). Computational advances have made possible a growing number of agent-based models across a variety of application domains. Applications range from modeling agent behavior in the stock market, supply chains, and consumer markets, to predicting the spread of epidemics, the threat of bio-warfare, and the factors responsible for the fall of ancient civilizations. This tutorial describes the theoretical and practical foundations of ABMS, identifies toolkits and methods for developing agent models, and illustrates the development of a simple agent-based model of shopper behavior using spreadsheets.
Agent-based modeling and simulation (ABMS) - a way to simulate a large number of choices by individual actors - is one of the most exciting practical developments in business and government modeling since the invention of relational databases. It represents a new way to understand data and generate information that has never been available before - a way for businesses and governments to view the future and to understand and anticipate the likely effects of their decisions on their markets, industries, and territories. It thus promises to have far-reaching effects on the way that businesses and governments in many areas use computers to support practical decision-making. This book has three purposes: first, to teach readers how to think about ABMS, that is, about agents and their interactions; second, to teach readers how to explain the features and advantages of ABMS to other people; and third, to teach readers how to actually implement ABMS by building agent-based simulations. It aims to be a complete ABMS resource and also provides a complete collection of ABMS business and government applications resources.
381 p., ref. bib. : 2 p.1/2 In a rented convent in Santa Fe, a revolution has been brewing. The activists are not anarchists, but rather Nobel Laureates in physics and economics such as Murray Gell-Mann and Kenneth Arrow, and pony-tailed graduate students, mathematicians, and computer scientists down from Los Alamos. They've formed an iconoclastic think tank called the Santa Fe Institute, and their radical idea is to create a new science called complexity. These mavericks from academe share a deep impatience with the kind of linear, reductionist thinking that has dominated science since the time of Newton. Instead, they are gathering novel ideas about interconnectedness, coevolution, chaos, structure, and order―and they're forging them into an entirely new, unified way of thinking about nature, human social behavior, life, and the universe itself. They want to know how a primordial soup of simple molecules managed to turn itself into the first living cell-and what the origin of life some four billion years ago can tell us about the process of technological innovation today. They want to know why ancient ecosystems often remained stable for millions of years, only to vanish in a geological instant―and what such events have to do with the sudden collapse of Soviet communism in the late 1980s. They want to know why the economy can behave in unpredictable ways that economists can't explain-and how the random process of Darwinian natural selection managed to produce such wonderfully intricate structures as the eye and the kidney. Above all, they want to know how the universe manages to bring forth complex structures such as galaxies, stars, planets, bacteria, plants, animals, and brains. There are common threads in all of these queries, and these Santa Fe scientists seek to understand them. Complexity is their story: the messy, funny, human story of how science really happens. Here is the tale of Brian Arthur, the Belfast-born economist who stubbornly pushed his theories of economic change in the face of hostile orthodoxy. Here, too, are the stories of Stuart Kauffman, the physician-turned-theorist whose most passionate desire has been to find the principles of evolutionary order and organization that Darwin never knew about; John Holland, the affable computer scientist who developed profoundly original theories of evolution and learning as he labored in obscurity for thirty years; Chris Langton, the one-time hippie whose close brush with death in a hang-glider accident inspired him to create the new field of artificial life; and Santa Fe Institute founder George Cowan, who worked a lifetime in the Los Alamos bomb laboratory, until-at age sixty―three―he set out to start a scientific revolution. Most of all, however, Complexity is the story of how these scientists and their colleagues have tried to forge what they like to call "the sciences of the twenty-first century.".
In this paper we describe a computer model developed jointly by mathematicians and medical consultants. The aim of the model is to provide practical help to people involved with caring for HIV-infected patients. The program is easy to use and can provide a wide variety of output, ranging from resource requirements and costs to detailed clinical information. The model uses the technique of computer simulation to study the progression of HIV-related diseases in a set of patients. The model can help planners concerned with the broad issues of health care provision, as well as clinical users whose main interest is the management and treatment of individual patients. The model is currently being tested in the Department of Genito-Urinary Medicine in the Royal Victoria Hospital, Bournemouth. The potential capability of the model is illustrated by some results from program runs using data from a set of Bournemouth patients. We argue that the power and flexibility of computer simulation as a technique for dealing with uncertainty and variability is especially appropriate in the case of HIV and AIDS.
Agent-based modeling is a powerful simulation modeling technique that has seen a number of applications in the last few years, including applications to real-world business problems. After the basic principles of agent-based simulation are briefly introduced, its four areas of application are discussed by using real-world applications: flow simulation, organizational simulation, market simulation, and diffusion simulation. For each category, one or several business applications are described and analyzed.