Agent-Based Modelling, a Quiet Revolution in Asset Management

Simon Bush (University of Auckland/ Opus), Theuns Henning, Jason Ingham

and Andrea Raith (University of Auckland)
Abstract

Over the past 20 years sophisticated technical models of the asset management process have
been created, but asset management is a socio-technical process, with the interaction between the
social and technical systems directly impacting strategy development and the long-term evolution
of the asset. To combine both the social and technical systems into one model a small number of
researchers have started to use agent-based modelling. By creating these models, stakeholders’
reactions to proposed policies can be explored prior to policy implementation. This ability to
explore stakeholder reactions means that, for the first time, asset management strategies can be
developed that meet stakeholder expectations, while ensuring the on-going functionality of the
asset.

This paper provides an introduction to this new modelling technique. This paper also describes
how agent-based models can be used to improve performance measurement and management,
thus creating a framework for improved decision making.

Keywords

Agent-Based Modelling (ABM), Asset Management, Decision Making, Simulation

Introduction the same as the service level setting exercise

that occurs in asset management
The asset management process is often (INGENIUM, 2006). In reality decisions are
depicted in a way which implies it is a simple not independent of each other and continue
cycle, with no or limited feedback and with to have repercussions that influence the
future actions dependent on past experience. system’s future operation (Sterman, 2001).
This cycle takes the same form as Sterman’s An alternative view, that more accurately
(2001) event orientated view of the world, reflects the decision making process found in
with goals and targets set at the start, asset management, is Sterman’s feedback
problems identified, decisions made and view of the world (Figure 2). In this view
results obtained (Figure 1). feedback loops are formed between the
decision environment, and the stakeholders
involved in the decision. This view is an
evolution of the event orientated view, as it
requires decision makers to actively assess
the impact of outside influences on the future
state of the system. The expectation that a
feedback view should be taken can be
inferred from recent audits (NZOAG, 2010,
Figure 1 – An Event Orientated View of the NZDIA, 2013), with auditors all finding that
World (Sterman, 2001) decision makers, while describing operational
activities in significant detail, are failing to
In this linear view the gap between the fully account for stakeholders in the decision
desired goal and the current state defines the making process and to clearly provide these
problem to be overcome (Sterman, 2001). stakeholders with choices. Osman (2012)
Once this problem is solved another is then also believes that to account for stakeholders
identified. This setting of goals and there should be a paradigm shift in asset
developing solutions to meet these goals is management modelling, because the current
asset centric viewpoint provides limited same environment decision makers can learn
understanding of what a stakeholder’s future about the possible future repercussions of
reaction to changes in service levels might their strategies.
be. Similarly, Bernhardt and McNeil (2008)
found that asset management decision What are the alternatives to agent-
support tools are predominantly asset based modelling?
focused and that these models have had
limited success in accurately simulating the A complex system has a large number of sub-
asset management decision making process. systems that are involved in many loosely
structured interactions, the outcome of which
is not predetermined (Jackson, 2003). This
interconnectedness results in behaviours that
cannot be quantified by simply summing the
outputs from each system component (Von
Bertalanffy, 1972). This greater than the sum
of the parts characteristic is known as
emergence (Hitchins, 2003). To help
describe a complex system, and eventually to
simulate the system, a number of techniques
can be used including systems-thinking,
systems dynamics, discrete event simulation,
and ABM (North & Macal, 2007), but not all of
them are appropriate for simulating a socio-
technical system. These approaches are
Figure 2 – The Feedback View of the discussed below.
World (Sterman, 2001)
Godau (1999), and Herder and Wijnia (2012)
To integrate stakeholders into asset suggested that a systems-thinking paradigm
management models Dijkema et al. (2013) should be used to model the asset
recommended that models should take a management decision making process.
socio-technical viewpoint, rather than the Systems-thinking was recommended as it is
more traditional, asset centric, technical commonly used to understand the operation
viewpoint. Herder and Wijnia (2012, p33) of complex systems. For example, Seddon
also found that because assets are and Brand (2008) and Skarzauskiene (2010)
embedded in society they are subject to have used systems-thinking to understand
societal changes as well, and as such asset how organisations function and to improve
management models should account for organisational performance. As such,
these societal influences. This new socio- systems-thinking provides a useful starting
technical paradigm requires asset point for modelling complex systems, as it
management models that are able to account provides a framework for system
for the interconnectedness between the visualisation. Nonetheless, to simulate the
decision and the stakeholders. If system mathematical models are required
stakeholders are to be included any new (Nikolic et al., 2013).
modelling approach should also be able to
simulate the behavioural attributes of each Systems-dynamics (Forrester, 1994) is one
individual stakeholder and the relationships method of creating mathematical models of
between these stakeholders. This paper the decision making process. In these
provides introduces Agent-Based Modelling models differential equations are used to
(ABM), a simulation method that has only simulate system components. For example,
recently started to be used to model the systems-dynamics has been used to model
complexities of asset management decision the impact of varying stocks and flows on
making. In ABM stakeholders and assets are delivery supply chains (North & Macal, 2007).
represented by computer simulations. By Systems-dynamics models have also been
modelling both stakeholders and assets in the used to simulate the impacts of a growing
demand for construction materials (Suryani et
al., 2010). Even so, systems-dynamics (Macal & North, 2010). As such, agent-based
models are not applicable for simulating modelling is the recommended method of
systems that have a strong spatial simulating the asset management decision
component (North & Macal, 2007), such as process.
the asset management process. Lewe (2005,
p104) also noted that for a systems-dynamics What does a model include?
model to be effective the whole system would
have to be described in detail, whereas an An agent-based model comprises three
agent-based model requires comparatively components: models representing
less data before it becomes a useful decision stakeholders, a model of stakeholders and
support tool. their relationships and a model of the
environment for the agents to interact with
Statistical modelling is technique that is (Macal & North, 2010) (Figure 3).
frequently used in asset management
decision making, but it is not recommended
for modelling socio-technical systems, as it
treats the decision process as a black-box.
By treating the decision process as black box
the relationship between the system’s inputs
and the system’s outputs becomes difficult to
understand (North & Macal, 2007). In
complex systems describing the connections
between system inputs and outputs is an
important part of understanding how the
system functions.

Discrete event simulation (DES) is a process

that is used to model complex systems (North
& Macal, 2007). DES can also be used to
model the feedback view of the world, as it is
able to simulate the impact of successive
decisions on future scheduled events (North
& Macal, 2007). Even though DES is able to Figure 3 – An agent-based asset
simulate these impacts it is not appropriate management model
for modelling socio-technical systems,
because it does not focus on agents and their To simulate stakeholders their behaviours are
interactions (North & Macal, 2007), which is described by rules, which define how they
required in a model of a socio-technical react to situations and how they interact with
system. each other (Van Dyke Parunak et al., 1998).
In an asset management context, the
ABM is an approach that can be used to modeller will have to understand who the key
model the behaviours of stakeholders and stakeholders are and their preferences, how
decision makers, how asset performance these stakeholders influence each other and
changes with time, and the effects of how they interact with the asset. For
changing asset performance on stakeholder example, how is the asset manager
holder satisfactions levels. In fact, ABM has influenced by asset risks, corporate strategic
been used to model many types of human- objectives, the asset user, and peers within
environment systems including coupled the working environment? Figure 4 shows a
human and natural systems (An & López- simple framework of how these rules could be
Carr, 2012) and how human-infrastructure structured.
systems evolve with time (Nikolic & Dijkema,
2010). ABM was used, because it is the best To gather data for the model, stakeholder
method for modelling systems that have a surveys are still used, but in a much more
strong geospatial component and where focused way. At the moment surveys are
stakeholder behaviours influence outcomes used either to understand people’s
preferences for a particular project or whether created separately from the system. By
they are satisfied with the service being running the model it is also possible to
delivered. In ABM this data is still required, understand whether defined targets are
but new data is also needed. For example, achievable, finally making it possible to
in-order to model agent behaviour the correlate maintenance and improvement
stakeholder survey would be used when the costs with stakeholder satisfaction.
level of satisfaction falls below an acceptable
limit and the factors that influence these What do I need to develop an Agent-
changes, and how stakeholders interact with based model?
each other. To develop these surveys new
members will be required in the asset New skills are required to understand
management team including behavioural stakeholder behaviour and to collect the
scientists, economists and operational appropriate data. Further to these skills,
researchers. This change in team dynamics programmers will be required to create the
is a reflection that asset management model. The environment the model is
decision making is a complex process that created in will range from purpose built
can no longer be managed by through to a specific agent-based modelling
individuals(Frangopol & Bocchini, 2012). packages. Based on comments by North and
Macal (2007) rapid pro-typing and simple to
intermediate level models are best developed
in a specific agent-based modelling package,
with more complex models being purpose
built for a specific situation. To date asset
management models have been created in
agent-based modelling packages, but there
are in excess of 70 software packages
(Wikipedia, 2014) to choose from, which
creates the problem of which one to use. The
package to use can however be narrowed
down by whether it has a geospatial
component. Other considerations for using a
particular software platform are whether the
level of documentation is adequate and the
level of activity on the user groups. These
are important considerations, as some
systems have limited documentation, which
leads to difficulty understanding how the
system works and this difficulty can be
compounded if there is no user group.

Have agent-based models been used

in asset management before?

ABM is only a recent addition to asset

management simulations and at this point in
time these models describe two types of
Figure 4 – An agent modelling framework system. For example, Nikolic and Dijkema
(2010) used agent-based models to
An added benefit of modelling stakeholders understand the evolution of complex asset
and the asset in one simulation environment systems. These can be classified as large
is improved performance management. This scale agent-based models. The second
occurs because stakeholder and asset model type, created by Osman (2012), Moore
attributes can be extracted from the model et al. (2008), and Bernhardt and McNeil
and used as performance measures, unlike (2008) has fewer agents and is used to
current methods, whereby measures are develop effective strategies for maintaining
locally managed infrastructure assets. These asset performance. The act of creating these
can be categorised as small-scale models, as new models advances asset management
they have limited numbers of agents. thinking, because it encourages modellers to
change from an outdated linear decision
In the Osman (2012) example, a small-scale approach to the newer feedback view of the
model was created to highlight how ABM world, which is more appropriate for decision
could be used in road asset management. In making in a complex environment. To create
Osman’s model a statistical deterioration agent-based models is only incremental step,
model is used to simulate the change in the but one which unifies many of the processes
performance of a pavement asset and simple that are already being used into one
behavioural models are used to simulate integrated decision support model. For these
stakeholders. In the simulation process the reasons agent-based modelling presents a
road condition deteriorates and because of major advance on current modelling
this deterioration users are exposed to an approaches and as such will become much
increasing number of sections that have poor more prevalent in asset management
ride quality. As the users encounter more decision making.
and more poor quality road sections the count
of low quality service interactions rises. As a References
result of this rising count of service
interactions more and more complaints are An, L. & López-Carr, D. (2012).
issued to the asset manager and to "Understanding Human Decisions in
government officials. Once the number of Coupled Natural and Human Systems".
complaints rises above a tolerance threshold Ecological Modelling, 229, 1-4.
budgets are increased to improve Bernhardt, K. L. S. & McNeil, S. (2008).
performance and improve satisfaction levels. "Agent-Based Modeling: Approach for
If no complaints are encountered then Improving Infrastructure Management".
budgets are lowered. In the simulation the Journal of Infrastructure Systems, 14, 253-
overall satisfaction is defined by their level of 261.
tolerance to poor road quality. In total five Dijkema, G. P. J., Lukszo, Z. & Weijnen, M. P.
user groups are defined with the tolerance of C. (2013). "Chapter 1: Introduction". In:
these groups ranging from highly tolerant to Dam, K. H., Nikolic, I. & Lukszo, Z. (eds.)
intolerant. This simple model represents the Agent-Based Modelling of Socio-Technical
ability, for the first time, to directly link asset Systems. Springer Netherlands.
performance and stakeholder satisfaction. Forrester, J. W. (1994). "System Dynamics,
This ability is a major improvement over Systems Thinking, and Soft Or". System
event orientated asset management models, Dynamics Review, 10, 245-256.
as asset managers can now assess Frangopol, D. M. & Bocchini, P. (2012).
stakeholder’s future reactions prior to a new "Optimisation under Uncertainty:
strategy being implemented. Accomplishments and Challenges".
Structure and Infrastructure Engineering:
The Osman model only focused on road Maintenance, Management, Life-Cycle
condition, but the method can applied to any Design and Performance, 8, 341-356.
other public asset, and to performance Godau, R. I. (1999). "The Changing Face of
measures other than condition. All that is Infrastructure Management". Systems
required is knowledge of stakeholder’s Engineering, 2, 226-236.
reactions to changing asset performance Herder, P. M. & Wijnia, Y. (2012). "A Systems
levels and asset specific technical models. View on Infrastructure Asset
Management". In: Van der Lei, T., Herder,
Conclusion P. & Wijnia, Y. (eds.) Asset Management.
Netherlands: Springer.
ABM has only started to be used as an asset Hitchins, D. K. (2003). Advanced Systems
management decision making support tool in Thinking, Engineering, and Management,
the last six years, but these early models Norwood, MA, United States, Artech
have shown that it is possible, for the first House Inc.
time, to link stakeholder expectation and
INGENIUM. 2006. "Developing Levels of Zealand: New Zealand Office of the
Service and Performance Measures". Auditor General.
Jackson, M. C. (2003). Systems Thinking: Osman, H. (2012). "Agent-Based Simulation
Creative Holism for Managers, Chichester, of Urban Infrastructure Asset Management
United Kingdom, John Wiley & Sons, Ltd. Activities". Automation in Construction, 28,
Lewe, J. (2005). An Integrated Decision- 45-57.
Making Framework for Transportation Seddon, J. & Brand, C. (2008). "Debate:
Architectures: Application to Aviation Systems Thinking and Public Sector
Systems Design. Doctor of Philosophy, Performance". Public Money &
Georgia Institute of Technology. Management, 28, 7-9.
Macal, C. M. & North, M. J. (2010). "Tutorial Skarzauskiene, A. (2010). "Managing
on Agent-Based Modelling and Complexity: Systems Thinking as a
Simulation". Journal of Simulation, 4, 151- Catalyst of the Organization Performance".
162. Measuring Business Excellence, 14, 49-
Moore, C., Tjioe, M., Manzella, A., Bernhardt, 64.
K. S. & McNeil, S. Asset Management Sterman, J. D. (2001). "System Dynamics
Insights Using Agent Models. Proceedings Modeling: Tools for Learning in a Complex
of 2007 International Workshop on World". California Management Review,
Computing in Civil Engineering, American 43, 8-25.
Society of Civil Engineers, , July 2008 Suryani, E., Chou, S.-Y., Hartono, R. & Chen,
2008 Carnegie-Mellon University. 176-183. C.-H. (2010). "Demand Scenario Analysis
Nikolic, I., Dam, K. H. & Kasmire, J. (2013). and Planned Capacity Expansion: A
"Chapter 3: Practice". In: Dam, K. H., System Dynamics Framework". Simulation
Nikolic, I. & Lukszo, Z. (eds.) Agent-Based Modelling Practice and Theory, 18, 732-
Modelling of Socio-Technical Systems. 751.
Springer Netherlands. Van Dyke Parunak, H., Savit, R. & Riolo, R.
Nikolic, I. & Dijkema, G. P. J. (2010). "On the L. (1998). "Agent-Based Modeling Vs.
Development of Agent-Based Models for Equation-Based Modeling: A Case Study
Infrastructure Evolution". International and Users’ Guide". In: Sichman, J., Conte,
Journal of Critical Infrastructures, 6, 148- R. & Gilbert, N. (eds.) Multi-Agent
167. Systems and Agent-Based Simulation.
North, M. J. & Macal, C. M. (2007). Managing Springer Berlin Heidelberg.
Business Complexity, Discovering Von Bertalanffy, L. (1972). "The History and
Strategic Solutions with Agent-Based Status of General Systems Theory". The
Modeling and Simulation, New York, Academy of Management Journal, 15,
United States, Oxford University Press. 407-426.
NZDIA (2013). "Report of the Local Wikipedia (2014). Comparison of Agent-
Government Infrastructure Efficiency Based Modeling Software [Online].
Expert Advisory Group". Wellington, New Available:
Zealand: Department of Internal Affairs. http://en.wikipedia.org/wiki/Comparison_of
NZOAG (2010). "New Zealand Transport _agent-based_modeling_software
Agency: Information and Planning for Accessed 2014].
Maintaining and Renewing the State
Highway Network". Wellington, New