106
Int. J. Critical Infrastructures, Vol. 6, No. 2, 2010
Agent-based modelling of energy
infrastructure transitions
E.J.L. Chappin* and G.P.J. Dijkema
Faculty of Technology
Policy and Management
Delft University of Technology
P.O. Box 5015
2600 GA, Delft, The Netherlands
Fax: +31(0)152783422
E-mail: e.j.l.chappin@tudelft.nl
E-mail: g.p.j.dijkema@tudelft.nl
*Corresponding author
Abstract: Shaping energy transitions not only requires technical system
innovation and redesign but also new policies, regulations, Research and
Development (R&D) and investment strategies – a transition assemblage.
Transition management thus equates to designing and implementing such an
assemblage. Agent-Based Models (ABMs) may be used for ex-ante assessment
of transition assemblage alternatives. To help determine whether the design of a
particular model is fit for its purpose, we have developed a typology. Three
models were assessed:
1
a model on the impact of CO2 policy on the power production sector
2
a model on the transition of the global Liquefied Natural Gas (LNG)
infrastructure
3
a model on the imminent transition caused by the arrival of Light-Emitting
Diode (LED) lighting systems.
All three models can be used to compare transition assemblage alternatives and
could be adapted to assess regulatory adaptability.
Keywords: agent-based model; ABM; energy infrastructures; energy
modelling; transitions; transition management; power generation; carbon
policies; LNG market; consumer lighting.
Reference to this paper should be made as follows: Chappin, E.J.L. and
Dijkema, G.P.J. (2010) ‘Agent-based modelling of energy infrastructure
transitions’, Int. J. Critical Infrastructures, Vol. 6, No. 2, pp.106–130.
Biographical notes: Emile J.L. Chappin is a PhD Researcher at the Faculty of
Technology, Policy and Management of the Delft University of Technology,
the Netherlands. His research is on modelling and transition management of
energy infrastructures.
Gerard P.J. Dijkema is an Associate Professor of Energy and Industry at the
Faculty of Technology, Policy and Management, TU Delft, the Netherlands.
His specialisation is system innovation for sustainability.
Copyright © 2010 Inderscience Enterprises Ltd.
Agent-based modelling of energy infrastructure transitions
1
107
Introduction
Energy infrastructure systems are complex systems: they contain many interacting
physical and social components. Due to a myriad of decisions and interactions, system
structure, content and performance emerge over time. The power grid, for example, can
be considered a large-scale sociotechnical system (Chappin and Dijkema, 2008c) or
λ-system (Nikolic et al., 2009). Sustainability of our energy infrastructure is another
emergent property (Dijkema and Basson, 2009).
We define a system transition as “structural change in both technical and social
subsystems” (Chappin and Dijkema, 2008b). Transitions emerge over time as a
fundamental change of λ-systems. During transitions, the structure and the content of the
physical subsystem change. While these changes are caused by the social subsystem
which comprises actors and their interconnections, the body of rules and institutions that
govern actor behaviour and decision making, and the mutual dependence of physical and
social subsystems cause both to change in a complex web of interaction, feedback and
feedforward relations. The idea behind transition management is that if a transition is
wanted, actors could somehow manage its emergence in the system. What is lacking,
however, is a basic understanding of the sociotechnical design space available for
transitions and a recognition of the very complexity of many a sociotechnical systems
which may imply that we only have a certain chance of success to steer λ-system towards
some preferred state.
We postulate that policy design and implementation is part of the sociotechnical
design space. A policy is a transition instrument if policy makers implement it to cause
structural change; in other words, if it is intended to invoke a transition. The policy is
effective when it indeed initiates a transition and leads to some optimal end state while
additional requirements for the transition path often exist.
Transitions typically span decades wherein the combination of external influence,
actor behaviour and actor interaction is dynamic and complex. Consequently, elucidating
suitable design variables for shaping transitions is difficult and may even be impossible.
In transition and transition management literature, the focus is on descriptive case studies
on past transitions (Rotmans et al., 2000; Geels, 2002; Rotmans, 2003; Geels, 2004).
However, as we have argued elsewhere (Chappin and Dijkema, 2008a–b), where
description leads to understanding, transition management implies that transitions should
and can be steered and shaped. Transition managers design a coherent assemblage of
policies, regulations, Research and Development (R&D) strategies, financing and so on.
We define a transition assemblage as the all-inclusive set of transition instruments
applied. A transition assemblage design is, therefore, a unique selection of transition
instruments. The effectiveness of such a transition assemblage equates to the likelihood
of meeting the designers’ objectives.
During and after a transition, the components and interactions are different;
being-in-transition is one of the emergent system properties. Thus, observing a transition
is difficult and subjective, and complete understanding and management of energy
infrastructure transition may be impossible. However, Axelrod (1997b) already argued
that “the simulation of an agent-based model is often the only viable way to study
populations of agents who are adaptive rather than fully rational”. Also, it has been
demonstrated that physical subsystem models can be adequately incorporated in
108
E.J.L. Chappin and G.P.J. Dijkema
Agent-Based Models (ABMs) to yield models that increase our understanding of energy
infrastructures and industrial networks (Chappin et al., 2009b; Chappin and Dijkema,
2009; Davis et al., 2009; Nikolic, 2009).
Therefore, we conjectured that ABMs are suitable to assess transition assemblage
designs in the energy domain. While we do not claim that ABMs will produce perfect
predictions of these systems, we do believe, however, that it is possible to compile
valid ABMs that show transitions in energy systems. We deem such models to be valid if
they are ‘fit for purpose’. These models do not show what will happen, but what may
happen in a delineated part of society given a stringent set of assumptions and conditions.
With the results generated by such models, the modellers can obtain insights in
transitions in energy infrastructures and assess transition assemblage alternatives for
transition managers.
In this paper, we present a framework for the use of ABMs as transition models.
While we limit ourselves to ABMs, constructing these models still leaves many degrees
of freedom. The purpose of the framework is to assist modellers of ABMs on transitions.
The framework provides a cohesive overview of the building blocks available and
presents the options and restrictions that model developers have. Thus, it should allow a
balanced and sound model development. In addition, the framework may serve as a
typology of transition models; it characterises existing and new models in terms of
potential for elucidating transitions and transition management. We demonstrate the
usefulness of this framework by three applications.
The structure of this paper is as follows: First, we elaborate system transitions in
large-scale sociotechnical systems (λ-systems). Second, we introduce gent-based
modelling as a suitable building block for models intended for ex-ante assessment of
transition assemblage in energy infrastructure systems. Third, we present a typology for
categorising models used to evaluate transition management. Fourth, the models
reviewed using the framework are presented. These models include the following:
•
•
•
a model used to analyse CO2 policy and taxation for the power production sector and
to assess these transition instruments
a model used to elucidate the ongoing transition of the global Liquefied Natural Gas
(LNG) infrastructure
a model developed to assess the effect of a breakthrough technology, Light-Emitting
Diode (LED) lighting systems.
For each of these, the model setup and results are briefly presented and the model type is
reflected upon, notably the degrees of freedom that were addressed in each modelling
exercise. The paper ends with a conclusion and an outlook.
2
Designing system transitions in large-scale sociotechnical systems
2.1 Transitions of λ-systems
Energy infrastructure systems can be considered λ-systems that contain interdependent
subsystems or components which are also considered systems (Asbjørnsen, 1992). They
are evolutionary and they exhibit path dependency and lock-in. Current choices shape
options for system structure and content in the future and the options, in turn, were
Agent-based modelling of energy infrastructure transitions
109
shaped by the past. The λ-systems we observe today were not designed as such; they
evolved to their present state (Herder et al., 2008; Nikolic et al., 2009). Because the
needs of society change, technological components were not designed to meet the current
set of requirements (e.g., sustainability, reliability, flexibility and affordability). New or
refined needs, emerging from society, affect the need for policy and regulation (see
Figure 1). Only policy makers can design a policy that affects actor behaviour. A policy
influences the conditions of an actor’s decision making and, thus, it may change the
outcomes of his/her decisions: Must we invest in a power plant or a wind farm? Shall we
buy an LED or a conventional bulb? In other words, changes in the technical components
of λ-systems and their interaction only materialise when actors change their behaviour
and decide differently. Regulation will only affect the technical system content indirectly
through the actors. Actors change their preferences or perceptions by adopting new
strategies or by the introduction of a new policy. To add to complexity, changes may
materialise at a time when perceptions and preferences have changed again.
Figure 1
From policy design to improvement (see online version for colours)
Over time, this process may result in a fundamental change of system structure; we call
that process a system transition (see Figure 1). A system transition is, therefore, by
definition, an emergent property of a λ-system.
2.2 Policy design should incorporate system transitions
Policy makers face the challenge of designing effective policies in and for λ-systems
(Bijker et al., 1987; Hughes, 1987; Ottens et al., 2006). The energy infrastructure is
clearly a λ-system; competitive tasks (power and natural gas generation and retail
services) were unbundled from monopolistic tasks (grid and pipeline operators). The
sector is embedded in and strongly connected to several markets, i.e., fuel, emission
trading and spot markets. Transport of energy (electricity, natural gas, oil, LNG, etc.) is,
therefore, more and more separated from production and consumption. On the demand
side, users increasingly get more options to join in the game, for instance, by the
introduction of distributed generation and smart metering. The consumer is also more
involved because of the renewed introduction of consumer-side incentives to change their
consumptive behaviour.
110
E.J.L. Chappin and G.P.J. Dijkema
All the actors active in these markets have their own objectives and means to realise
them. The government optimises its objectives by setting the rules of the game, using
policy and regulation as main instruments. Government affects the behaviour of actors
who optimise their own objectives.
Incorporating the transition pathway and end state adds a new dimension to the
challenge of policy design.
2.3 Knowledge gaps in the design of system transitions
Based on literature on the design process (Maier and Rechting, 2002; Dym and Little,
2004), a number of knowledge gaps on the design of system transitions has been
identified. Since these have been described in detail elsewhere (Chappin and Dijkema,
2008b–c), we will only summarise them here. A design of a system transition
encompasses a transition assemblage that is estimated by a certain performance based on
transition tests.
First, there is a need for transition indicators which should unambiguously state the
performance of system transitions – their pathway and their end state – according to the
objectives and requirements of its designer.
Second, there is a need for a clear understanding of transition instruments, i.e., the
design alternatives, based on design variables, in order to come to the largest variety of
possible designs for system transitions.
Third, one needs to be able to check the different designs with suitable tests to assess
their performance, measured by performance indicators. Simulation models are tests for
different transition designs using indicators to measure their performance.
The framework, described in the next section, intends to fill the three gaps
mentioned above.
3
Framework for agent-based models of transitions in energy
infrastructure systems
We postulate that assessing the performance of system transitions is an extremely
difficult task. We believe that it is not sufficient to rely only on traditional assessment
methods such as top-down and economic optimisation models. Those models often
exhibit hidden assumptions (for example, being equipped for homogenous actors only).
For testing the performance of system transitions, one rather needs to assess the evolution
of λ-systems under different transition assemblage designs. Based on Axelrod (1997b),
we argue for ABMs to assess the performance of transition assemblage designs for
energy infrastructure systems. They fit the structure of λ-systems; the decision making of
relevant actors is translated to behavioural rules of agents; the technical subsystems are
modelled as physical networks of equipment and flows. When discussing transitions,
alternatives to ABMs lose relevance because they impose a fixed system structure to the
model preventing the observation of emergent transitions. The main necessities for
observing transitions include a dynamic system structure and the use of entities (discrete
and active). The main alternatives to ABMs are the other simulation schools. Computable
General Equilibrium (CGE) models are static equilibrium models (Jones, 1965; Leontief,
1998) based on linear equations. As such, they are continuous and do not allow for
111
Agent-based modelling of energy infrastructure transitions
discrete components. Also system dynamics (Forrester, 1958; 1969) only allows for
continuous variables. Dynamic systems (Rosenberg and Karnopp, 1983) are continuous
and are only applicable to physical systems and discrete event simulation (Gordon, 1978;
Boer et al., 2002; Boyson et al., 2003; Corsi et al., 2006) assumes passive components
not capable of acting.
We present a framework for building useful ABMs of transitions in energy
infrastructure systems (see Figure 2). The five main components of that framework are
as follows:
1
system representation
2
exogenous scenarios
3
design of transition assemblage
4
system evolution
5
impact assessment.
Let us have a closer look at each of these building blocks. We define different levels for
exogenous scenarios and for the design of transition assemblage. Selection of those levels
impact the whole model. Adopting higher levels means more requirements for other
model components. In return, higher levels allow for a more realistic type of model
dynamics, results and, in the end, better conclusions. While doing that, the framework
does not restrict the modelling paradigm; it promotes the intelligent use of the strengths
of different modelling paradigms for different parts of the system.
By introducing the levels of complexity for the transition assemblage, the framework
serves as the typology for transition models (see Table 1). Used as such, one may come to
the unfortunate conclusion that using the lowest levels is very common in the (small)
body of transition models in the literature and that, because of implicit assumptions, such
models cannot lead to insight into transitions.
Table 1
Typology of transition models
Ability of the model
Level 1
Level 2
Level 3
Level 4
x
x
x
x
x
x
x
x
x
Observe system evolution
Assessment of transition impact
Comparison of transition designs
Assessment regulatory adaptability
x
Let us now look at the five components.
3.1 System representation
Our framework intends to use ABMs as a modelling paradigm for building assessments
of transitions in energy infrastructure systems. This implies that the ABM represents the
energy infrastructure system. Therefore, we will define the ABMs and agents and provide
the steps to come to agent-based system representations for studying the design of system
transitions in λ-systems. In general, all subsystems or elements under relevant influence
by other subsystems or elements need to be included in the system representation.
112
Figure 2
Note:
E.J.L. Chappin and G.P.J. Dijkema
Framework for assessing system transitions with ABMs (see online version
for colours)
The main components are system representation, exogenous scenarios,
transition assemblage design variables, system evolution and
impact assessment.
Agent-based modelling of energy infrastructure transitions
113
An agent-based simulation model is often defined as “a collection of heterogeneous,
intelligent, and interacting agents, which operate and exist in an environment, which in
turn is made up of agents” (Epstein and Axtell, 1996; Axelrod, 1997a). An ABM contains
a set of interacting ‘agents’ with certain properties acting based on a set of rules, reacting
upon factors coming from outside. An agent is defined as “an encapsulated computer
system that is situated in some environment and that is capable of flexible, autonomous
action in that environment in order to meet its design objectives” (Jennings, 2000).
Although in the literature, different sets of properties for agents are proposed (Bussmann
et al., 1998; Weiss, 2000), the core components of an agent are a set of goals, a working
memory, a social memory and a set of rules for social engagement. Physical elements do
not act themselves; they are passive. Therefore, properties and capabilities characterise
elements in the physical subsystem.
Many tools and methods exist for operationalising the system representation.
Developing a system representation is a process that combines the collection and
interpretation of knowledge about the system. The framework prescribes the structure for
translation of this knowledge into the representation of the system. This follows the
combination of using agents in ABMs and the chosen λ-systems perspective.
The model developer defines a conceptual model of the system containing all relevant
elements. Consecutively, implementation of those elements is formalising the identity
and decision rules of agents and the properties and capabilities of physical assets. In
addition, the definition of communication protocols for agent interaction allows for
creating social and physical networks.
Within this framework, there are still many system representations possible. Further,
operationalisation is a tailored design process, specific to the domain under study and the
researchers’ focus. Additional conventions or methodologies aid that process. For
instance, one can use the system decomposition method (Nikolic et al., 2006; 2009)
designed to capture tacit knowledge of actors in an ABM. That method prescribes the
systematic gathering of data from actors and domain experts. A formal computer model
contains a representation of the stakeholders’ knowledge. This knowledge can be
formalised and shared using ontologies (van Dam and Lukszo, 2009). Next, many
suggestions are formulated that increase the efficiency of a model development process
(Chappin, 2006, Chap. 9).
3.2 Exogenous scenarios
Useful models require strict delineation. Especially regarding the study of transitions,
deciding what should be included and excluded is difficult. Inherently, not all relevant
subsystems can be represented within the system. Therefore, assumptions need to be
made on the relationships between subsystems. Where possible, we define parts of the
system that are unaffected by other parts within the system; we exclude them from the
system. Everything outside the system boundary is, therefore, exogenous. Every thing
relevant but exogenous makes up the scenario space (Fahey and Randall, 1998; Enserink
et al., 2002). The scenario space can have one of the several levels of complexity. In all
those levels, relevant but unaffected components are modelled as exogenous parameters.
They can be static, be varied individually and be varied together.
114
E.J.L. Chappin and G.P.J. Dijkema
3.2.1 Static parameter values
The easiest way is to vary parameters between runs only. For instance, in each simulation
run, a particular value is assigned to the price of natural gas in the market, chosen from a
number of predefined values. If the number of possible values equals one, then that
implies a static value for all simulations which effectually excludes it from the scenario.
Most common is a range of values, sometimes with a nonuniform distribution. The
need for data is limited; for each parameter, the minimum, maximum and interval values
and possibly the distribution need to be determined. The range of available values reflects
the parameter’s uncertainty.
3.2.2 Varying trends
The modelling of scenario parameters as continuous trends is more difficult and
data-intensive. At this level, we require a representation of a price trend of, for instance, a
natural gas. One representation is a start value and a change pattern, possibly stochastic.
Modelling scenario parameters as trends has two consequences. First, this requires
additional parameters: a probability distribution and its properties. Although more
complicated to develop, this approach will enable the use of more realistic scenarios. The
variability in the trends characterise the uncertainty in the parameter. This uncertainty
can, with trends, be time-related (uncertainty can grow or decline over the simulated
time). Second, the model needs to adapt to changes of the value of this parameter. Since
parameters are not static within one simulation, there is a need for taking into account this
trend, for instance by forecasting agents. Therefore, the use of varying trends leads to
very different models.
3.2.3 Coupling with other models
Finally, one can develop or use existing models such as System Dynamics Models
(SDMs) or mathematical models to provide exogenous parameters. SDMs are a collection
of differential equations and are often considered an incompatible modelling paradigm to
ABMs since ABMs are discrete and SDMs are continuous (Schieritz and Milling, 2003;
Borshchev and Filippov, 2004). These models differ in the type of assumptions. That
makes it hard to adapt SDMs to the agent-based paradigm. We postulate, however, that
we should combine ABMs and SDMs into a hybrid to use the best of both worlds. A
single SDM may generate multiple scenario parameters. Again, this may be more
complicated than varying trends only as this approach not only leads to software
requirements but also requires more and different modelling skills. Using SDMs or
mathematical models to model exogenous scenario parameters should be considered if
multiple scenario parameters are strongly correlated especially when well-designed
SDMs are available. Mathematical models are often found in literature and can be used as
an external world to the agents in the model.
3.3 Design of transition assemblage
Similar to exogenous scenarios, different levels of complexity exist for modelling the
transition assemblage design. Table 2 presents an overview of those levels. They are
discussed separately.
Agent-based modelling of energy infrastructure transitions
Table 2
115
Transition assemblage design alternatives
Level
Description
Level of complexity
1
Implicitly modelled
2
Fixed system parameter
Model needs responsiveness.
3
Exogenous scenario parameter
Model requires flexibility.
4
Endogenous system parameter
Model requires regulatory adaptability.
We postulate that for adequate evaluation of transition assemblage designs, one should
aim at Level 3 or 4. It is possible to start at Level 2 and upgrade later. However, Level 1
should be avoided since reusability in a higher level model will prove impossible. One
should take notice that these levels are not exclusive and that different levels of policies
and regulations can be in one model simultaneously.
3.3.1 Implicitly modelled
In this case, the structure of the model accommodates a certain policy and regulation. The
assemblage is a fixed set of policy and regulation, the setting of which in the model is
implicit. Since the set is fixed, it may prove hard or impossible to adapt to changes.
System components do not have to be aware of the transition assemblage. As a dreadful
consequence, one can never assess the impact of the design assemblage. Therefore,
models using this level will not lead to useful models of transition management.
Consequently, we recommend not to apply Level 1 policy and regulation in transition
models. The very selection and design of policy and regulation is de facto a transition
assemblage design variable. If policy is not modelled as such, alteration of policy is
impossible without constructing a new model.
3.3.2 Fixed system parameter
When policy or regulation is a fixed system parameter, the model needs to be able to
respond to this parameter setting during the simulation.
Translated to ABMs, this implies that agents base their decisions on this policy
setting and assume (or are uncertain about the) stability of this policy setting. Since the
policy is unrelated to other system properties, it is exogenous to the model.
With Level 2, it is still impossible to assess the effect of a transition assemblage.
The only advantage of using this level over making it implicit is that the model is
upgradeable to the Level 3. Upgrading implies adding agents’ responsiveness to other
policy values while the model structure remains intact. Hence, we recommend to start at
least at Level 2.
3.3.3 Exogenous scenario parameter
A policy can be a (set of) scenario parameter that is exogenous to the system in transition.
In this setup, policy is one of the three levels of scenario parameters:
1
varying parameter values between runs
2
varying trends between runs
116
3
E.J.L. Chappin and G.P.J. Dijkema
based on SDMs – all with their advantages and disadvantages (see the previous
section on exogenous scenarios).
Only at this level and upwards it is possible to vary the modelled policy or regulation in
order to derive and test different transition assemblage design alternatives. Therefore, this
is the lowest level that a modeller should aim for when modelling policy in this
framework. However, as stated above, one may start with fixed system parameters
(Level 2) as this will not limit model expansion.
3.3.4 Endogenous system parameter
At this level, policy development is endogenous. This implies that the government is an
actor included in the system representation who decides during a simulation run on their
policy and regulation. Government’s actions are the outcome of its decision rules and the
state of the system, i.e., past, current and expected values on system parameters. Since
the systems state depends on the agent’s reaction to government’s policy, the government
behaviour would be a result of its own behaviour in the past in relation to the behaviour
of other agents. As a consequence, the policy setting is an emerging property of
the system.
Modelling this level transition assemblage is modelling a form of adaptive
governance which is considered a keystone of transition management (Rotmans et al.,
2000; 2001; Loorbach, 2007). We, therefore, conjecture that only this level can really
assess transition management as referred to in the literature.
However, modelling policy and regulation as an endogenous system parameter that
leads to tough requirements for the other model components. One needs all relevant
interdependencies with other parameters in the model. It may prove to be difficult to
validate data on government’s responsiveness under different conditions. A solution is to
model uncertainty on that data as (Level 3) exogenous scenario parameters. Adoption of
that solution leads to models in which the adopted policy is the outcome of the interplay
of the system’s state, the decision rules of the government and the parameters of
exogenous scenarios. This will lead to robust evaluations of designs since it
acknowledges the evolution of policy as well as uncertainty in government’s actions
and responses.
3.4 System evolution
By the actions of agents, the system will evolve over time. They act as part of the system
by reacting on exogenous scenarios and endogenous parts of the system.
Since agents are interdependent, system level properties and system behaviour
are emergent. As stated above, policy can also be emergent when modelled
endogenously. Variety in parameter settings in input will provide differences in outcomes
of simulation runs. Therefore, the evolution of the system in one simulation is not a
prediction of the future of that system. One needs an impact assessment by using
different system evolutions at different locations in the parameter space in order to come
to sound conclusions.
Agent-based modelling of energy infrastructure transitions
117
3.5 Impact assessment
Together, the above notions are the necessary ingredients for the impact assessment of
the design alternatives. How do we decide which transition assemblage design is to be
preferred? The impact assessment has to encompass a well-designed set of experiments
and a solid analysis of their results.
3.5.1 Parameter sweep: experimental design
In order to assess and compare the performance of different transition assemblage design
alternatives, one can use the literature on design of experiments (e.g., Kim and Kalb,
1996; Box et al., 2005; Goupy and Creighton, 2007). An experimental design is the way
in which different factors of the model are varied between different model runs.
Classical methods include factorial designs in which the factors are varied
independently (Iman et al., 1981). Within the class of factorial designs, the main design is
full factorial, a design in which the experiments take on all possible combinations of the
levels of the factors. Usually, each of the factors has only two different values.
If the number of factors is too high to be executed within a reasonable amount of
time, given the available computational power, a fractional factorial design may be
adopted. An efficient form of a fractional factorial design is obtained by a technique
called Latin Hypercube Sampling (LHS) (McKay et al., 1979). This technique allows
selecting any preferred number of experiments where the resulting set has a uniform
distribution over the multidimensional parameter space. Thus, the number of experiments
can be set depending on time and computing resources available.
The use of environment scenarios (Fahey and Randall, 1998; Enserink et al., 2002)
leads to a different setup although the experimental design can be seen as a different class
of fractional factorial designs. Each scenario is a combination of values on a set of factors
that were modelled separately in the full and fractional factorial designs. In other words,
parameters are grouped by their variation which leads to a smaller number of possible
combinations. To arrive at a suitable variation of the values of factors in a scenario, one
may again use one of the experimental designs described. For example, a scenario may
have three groups of factors that are varied with a full factorial design. In that design, you
have eight distinct scenarios (the corners of a cube). Altogether, this is a fractional
factorial design that is fundamentally different to LHS because preselected groups of
factors are varied in concert. As a consequence, the use of environment scenarios is based
on the assumptions that the factors within each scenario are interdependent and that each
factor is independent from the factors in other groups.
3.5.2 Analysis of the results: assessment methods
The raw simulation result is a full record of the state of the evolving system during all
experiments in the parameter sweep. The recorded parameters should include not only the
selected performance indicators but also the input variables in order to allow testing
for correlations.
Since the parameter space is large and modern computational power allows large sets
of runs to be completed in reasonable time, this full record is often a huge amount of
data. One can use visualisation methods to grasp some specifics hidden in the data but
118
E.J.L. Chappin and G.P.J. Dijkema
this does not lead to real assessments. Instead, statistical methods for data analysis must
be used for assessing and comparing the performance of different transition assemblage
designs. However, statistical methods generally are of a static nature and are not capable
of adequately analysing the results. There is a need for adapting and building statistical
methods to assess and compare different designs by their variety and uncertainty in
evolving performance. One example is by making a series of student T-tests over time to
assess differences in means (Chappin and Dijkema, 2008c).
3.6 Conclusion
In order to underpin transitions in energy infrastructure systems and the potential of
transition design and transition management, we need to develop models that assess
transition designs. We formulated, based on this line of thinking, a typology of transition
models containing four levels. These levels have different requirements for the following
five components of the model:
1
system representation
2
transition assemblage
3
exogenous scenarios
4
evolution
5
assessment.
The model typology is summarised in Table 1. On Level 1, only system evolution is
observed in which transition assemblage is implicitly modelled and assessing transitions
is impossible. Level 2 allows for transition assessment. On Level 3, the comparison of
different transition designs also becomes possible, but keystone transition management
principles can only be evaluated on Level 4. As a consequence, only Levels 2, 3 and 4
models are useful. Therefore, the most innovative results are to be expected from
Levels 3 and 4 models.
4
Case on production: design for a model of transition in power generation
A quantitative ABM was developed to simulate the evolution of the structure and
performance of a hypothetical electricity market in the next 50 years using insights from
microeconomics, market design, agent theory, process systems engineering and complex
systems theory (Chappin and Dijkema, 2008b–c). The main objective is to get insights on
the potential long-term impact of policy interventions such as a carbon tax or emissions
cap on the power sector. A detailed analysis of this case and its results has been the
subject of publications (Chappin et al., 2009a–b). A schematic overview of how
the ABM is set up is presented in Figure 4. This model can be called a Level 3 model.
The model allows for evaluation and comparison of different transition assemblages. We
are developing an upgrade to Level 4 which is discussed below.
Agent-based modelling of energy infrastructure transitions
119
4.1 System representation
The model reflects the real world situation of six independent electricity producers who
have different generation portfolios and who make different decisions regarding the
operation of their generators, investment and decommissioning. As in the framework, the
model contains subsystems for agents and installations. The agents in the model have an
operational behaviour, i.e., power producers need to negotiate contracts for feedstock,
sales of electricity and, in the case of emissions trading, emission rights. They also
exhibit a strategic behaviour, i.e., in the longer term the agents need to choose when to
invest, how much capacity to build and what type of power generation technology to
select. Agents interact through negotiated contracts and organised exchanges and the
physical flows and their constraints and characteristics are modelled.
Figure 3
Case on production: the framework applied to carbon policies and power generation
(see online version for colours)
120
Figure 4
E.J.L. Chappin and G.P.J. Dijkema
Case on trade: conceptual model for transition in LNG markets (see online version
for colours)
Markets for CO2 rights, power and fuels are modelled as exchanges in which 100% of the
product is traded every time step. The time step of the model is one year and the
simulations span a horizon of 50 years. A consumer agent is modelled to consume all
electricity. To allow for correct mass and energy balances, an environment agent reflects
all uptakes and emissions. The government agent implements policy interventions.
4.2 Exogenous scenarios
A range of scenario parameters is in Level 1. It is specific to the Dutch market. In
addition, the electricity demand profile consists of ten steps per year that reflect a typical
load duration curve and has a rising trend. The demand rises as a Level 2 trend. Fuel
prices are modelled as a variety of Level 2 trends as well.
Agent-based modelling of energy infrastructure transitions
121
4.3 Design of transition assemblage
The main options for emission reduction for government are called carbon policies.
Therefore, they are selected as main design variables for system transitions. The two
main candidates are Emission Trading System (ETS) which is implemented in the EU
and Carbon Taxation (CT) which has been implemented on a smaller scale in Norway.
Next to those two options, no intervention is chosen as a base reference. All policy
interventions and implementations are modelled in the government agent.
The main policy variable of the ETS is the emissions cap. In the model, the cap is
set to reflect the likely design of the Phase 3 of the EU ETS in which the CO2 cap is
reduced every five years by 3 Mtonne for a market with the size of the Netherlands.
With an initial cap of 50 Mtonne, a 50% reduction is achieved in a little more than 40
years. Another important policy variable is how many emission rights can be obtained
through the Clean Development Mechanism (CDM).1 This is set to 5 Mtonne per year
over the entire simulated time period. The main CT policy variable is the tax level. To
allow a fair comparison between ETS and CT, the tax level in our model has been
calibrated to the average CO2 price that emerges in the simulated emission market. The
initial tax level equates to 20 per tonne which reflects the current CO2 price under ETS.
Over time, tax level increases to 80 per tonne. These values were estimated based on the
runs under ETS.
The transition assemblage is, therefore, modelled at Level 3 using exogenous
parameters, leading to strong requirements, i.e., the agents need to be able to act under
ETS and CT policies, for the other model components. We have plans to upgrade this
model to Level 4 by advancing the role of the government agent. It could enhance the
strength of the implemented policy based on past successes.
4.4 System evolution
The characteristics of the modelled system are emergent; the generation portfolio and
merit order, fuel choice, abatement options as well as electricity and CO2 prices and
emissions emerge as a result of the decisions of the agents. In the model, the following
schedule of actions is implemented which will be repeated yearly:
•
•
•
Purchase emission rights in the annual auction. The auction bids are based on the
‘willingness to pay’ per installation which is determined as the expected electricity
price less the marginal costs of each unit divided by the CO2 intensity. The bid
volume equals the expected electricity sales volume times the CO2 intensity of the
power plants that are expected to be in merit.
Offer electricity to the market (which is modelled as a power pool). Each plant’s
capacity is offered at a variable generation cost (fuel cost, variable operating and
maintenance cost and CO2 cost). The CO2 costs of a generator equal the CO2 price
times its CO2 intensity. In case insufficient CO2 rights have been obtained, CO2 cost
equals to the penalty for noncompliance.2
Acquire the required amounts of fuel from the world market which are calculated
from the actual production and fuel usage.
122
•
E.J.L. Chappin and G.P.J. Dijkema
Pay the penalty in case there is a shortage of CO2 rights. Surpluses and shortages are
calculated from the actual production levels and the volume of emission rights
owned by the agent.
4.5 Impact assessment
Simulations have been done for the three transition designs: no-carbon policy, ETS and
CT. Impact assessment was made possible by making the pressure of the intervention of
ETS and CT comparable (calibrating the average price). Many runs were done and plots
were made of emission levels, emission intensities and power portfolios. Some included
stochastic information. It was found that all three transition designs performed
differently. CT outperformed ETS in the chosen scenario.
5
Case on trade: modelling transitions in the LNG market
The LNG market is subject to considerable changes, the most principal of which is the
departure from long-term project specific sales and purchase agreements towards a truly
global LNG market with flexible spot trading governed by master sales agreements. This
is exemplified by the rise of the LNG spot market which was virtually nonexistent in the
early 1990s (e.g., 1.2% of the total trade in 1992) and represented 16% of the total LNG
trade by 2005 (Morikawa, 2008). It is widely expected in the industry that this share will
further increase to 30% within the next decade (Aissaoui, 2006; NGI, 2007) or less (IEA,
2004). To explore this imminent transition of this market, a sociotechnical system
perspective (based on the ideas of Hughes, 1987) is combined with principles of
transition management, institutional economics and the ABM paradigm. This model is
being developed at the moment. We aim at a Level 3 model by implementing a variety of
strategies for the core LNG agents in the model that may lead to a system transition.
5.1 System representation
In order to acquaint the reader with the complexity of the LNG trade and create a
common understanding of the LNG market model at hand, a system representation is
given in Figure 4. The central idea is that the LNG market contains LNG agents who
invest in LNG projects and who operate them. In order to create value and to realise a
return on investment, each LNG agent who owns an LNG project, for instance, a
liquefaction plant, needs to negotiate a contract with another LNG agent who owns the
complementing LNG project (a regasification terminal) to create a functional value chain.
The result of this contract negotiation process determines the credit level of the LNG
agent and its performance. This model seeks for quantification of the relation between the
key market drivers and the inhibitors of the market on the one hand and the strategic
behaviour of the LNG agents on the other. In this case, the focus is even stronger on the
strategic behaviour than the first case discussed in this paper.
Agent-based modelling of energy infrastructure transitions
123
5.2 Exogenous scenarios
The demand profile for natural gas from LNG is derived from Global Insight (2007).
Demand is set exogenously for the period 2005–2025. In addition, the availability of
innovative technologies is exogenous and refers to potential innovations such as floating
storage and regasification units and the construction of an LNG trading hub that enables
LNG suppliers and buyers to store and trade LNG over an extended period of time.
5.3 Design of transition assemblage
Next to the rapid expansion of the LNG trade, there are other imminent market drivers
that may facilitate a transition towards a spot market for LNG. These include, but are not
limited to, decreasing capital costs, more flexible demand requirements, the rise of an
uncommitted tanker fleet and the initiation of new liquefaction plants without the
complete contractual coverage of the produced volumes. Examples of this trend are
Malaysia LNG Tiga, Australia’s Northwest Shelf (NWS) Train 5 (Tusiani and Shearer,
2007), and Sakhalin II Phase 2 Project (Ball et al., 2004) for which the go-ahead for
construction was given despite a significant volume of uncommitted production capacity.
The momentum of the LNG spot market can act as a self-reinforcing loop wherein
expectations about the future development influence the decisions on whether or not to
become active on the LNG spot market. Brito and Hartley (2007) stated that “while
exogenous changes in costs or demand are critical to promoting a change in market
structure, there is also a substantial endogenous component. Expectations about the
evolution of the market influence investments and trading decisions and can make the
change in market structure much faster and more abrupt.”
5.4 System evolution
The proposed LNG market simulation seeks to uncover the emergent behaviour of the
LNG market as a whole by looking at the investment decisions and contract negotiation
process of the individual and autonomous LNG agents. Accordingly, it interprets a
change in the market structure as a departure from the traditional LNG market towards a
global LNG market that actively pursues more flexible spot trading models. The model
seeks to implement the following schedule of actions:
•
•
investment decision
•
contract negotiation
•
timing of investment
•
project realisation
trade LNG.
5.5 Impact assessment
It is our belief that the interplay of exogenous forces and endogenous expectations can
move the LNG market along the pathway of transition towards the breakthrough phase in
which visible structural changes are the forerunners for a new market equilibrium. We
expect to underpin this idea by implementing and operationalising this model and
executing and analysing a vast parameter sweep.
124
6
E.J.L. Chappin and G.P.J. Dijkema
Case on consumption: transitions in the consumer lighting sector
In consumer lighting, changes are forthcoming. The EU’s phase out of incandescent
lighting is a clear strategy that will change the sector by removing from stores the
cheapest forms of inefficient household lighting (CEC, 2009). Although implied, it is
uncertain whether or not the lighting sector will be efficient overnight; either consumers
may switch to forms of inefficient lighting that are exempted from the phase out or
consumers’ behaviour will change. The precise dynamics within the sector are unknown.
A conceptual model is presented from which an ABM will be developed that is
expected to lead to improved understanding of the dynamics of transitions that may occur
in this sector (see Figure 5). Policy makers and other players in the sector can use the
simulation model to test their strategies, learn to understand patterns that appear and
design feasible transition assemblage designs. This model is still a conceptual design. We
aim at a Level 4 model (including regulatory adaptability) where the government and
lighting companies engage in adaptive strategy formulation.
Figure 5
Case on consumption: conceptual model for simulating transitions in the consumer
lighting sector (see online version for colours)
6.1 System representation
The consumer lighting sector is a true sociotechnical system; the social part of the system
contains a network of a variety of actors having strategic and operational behaviours.
Consumers visit stores, buy luminaires and lamps, dismantle lamps when they break and
communicate with other consumers through word of mouth. Stores sell new luminaires
and replacement lamps; manufacturers innovate and supply the stores. The consumer’s
Agent-based modelling of energy infrastructure transitions
125
decision making regarding purchases of lamps and luminaires will be operationalised
based on heterogeneous preferences for different characteristics and aspects of lamps like
price, lifetime cost, colour and environmental aspects such as luminary efficiency.
The technical part of the system consists of the lamps people have in their homes. In
the model, a consumer owns a number of luminaires. Attached to these are a number of
light bulbs that match the socket type and wattage. The light output is controlled using
either a switch or a dimmer.
6.2 Exogenous scenarios
Key parameters, such as prices of light bulb technologies, will be supplied by scenario
trends. Technological improvements may be modelled endogenously, i.e., relating
innovation to adoption.
6.3 Design variables for system transitions
Government regulation is initially included as fixed parameters in the model. Later, it
may be transformed to an endogenous system parameter so the model can be used to
test several adaptive governmental strategies. Potential strategies to be tested are
eco-labelling, forced phase out of all low efficiency alternatives, phase in of an eco-tax,
etc. All transition assemblages can first be modelled as Level 3 exogenous scenario
parameters, directly determining the strategies of the actors. One could also engage in
interactive strategy formulation, modelling on Level 4 where the effect of the timing of
intervention of parties can be studied and where the effect of the interaction of the
transition strategies of different actors can be evaluated. This does require a more
advanced model though.
6.4 System evolution
When a lamp fails, a consumer goes to buy a new one that fits the luminaire, but a
consumer may also change his/her luminaire itself if he/she prefers to. The decision is
based on the alternatives available and individual preferences. In a consumer’s decisions,
experiences with lamps of a certain kind are recorded in memory and influence the
decision. Consumers also influence each other by communicating their experiences.
Consumers are also influenced by marketing from stores or through other media.
Depending on the needs of consumers, manufacturers will respond by increasing the
pace of innovation. If demand is higher for specific products, manufacturers will be
tempted to supply these products. The interaction of the level of luminaries and bulbs on
the one hand and innovation and purchase of consumers on the other hand is complex and
is expected to lead to interesting dynamics.
6.5 Impact assessment
The model allows for testing governmental interventions. The model is set up in a
modular way, allowing the introduction of new policies next to the ones formulated
previously. On the other hand, marketing strategies of manufacturers can also be tested.
First, different individual runs will be analysed. Later on, parameter sweeps will be
executed to test the robustness of the strategies of the agents in the model.
126
7
E.J.L. Chappin and G.P.J. Dijkema
Conclusion and outlook
Since transitions in energy infrastructure systems are to be ‘managed’, we developed a
framework to use ABMs as transition simulation models for the assessment of transition
assemblage design alternatives. To reflect the main components and interactions of
energy infrastructure systems, this framework adopts a large-scale sociotechnical systems
perspective and expands it to allow modelling transitions of and in such systems.
The proposed framework consists of a number of parts. First, the system
representation is based on agent-based modelling and systems thinking. Second,
exogenous scenarios use scenario analysis and/or system dynamics. Third, design
variables for system transitions are based on policy making and transition literature.
Fourth, system evolution reflects complex systems thinking. Fifth, the impact assessment
can be a combination of experimental design, scenario analysis, statistics and data
mining.
This framework functions as a typology for existing and new transition models by
classifying the way in which the transition designs are modelled. Level 1 is modelling
transition designs implicitly which is not useful for assessing the effect of the transition
assemblage although it is often used. Level 2, using fixed system parameters, is better,
but only in the sense that it easily facilitates upgrading to Level 3 which is using
exogenous system parameters. Only Level 4 truly allows for assessing the merits of
transition management by introducing regulatory adaptability and assessing its potential.
The framework brings together many research domains which fits the
multidisciplinary approach needed to elucidate transitions and underpin transition
management. Its intention is not to limit the researcher, but rather to structure and
explicate his choices and build models that are not only useful but also reusable
and modular.
To explore its applicability, the framework is currently being tested and elaborated
upon in several case studies that will be the subject of future publications. In conceptual
terms, three cases have been described in this paper. The framework allows for
developing models of energy infrastructure transitions that focus on production, trade
and consumption.
Acknowledgements
This work has been partially funded by the Next Generation Infrastructures Foundation.
Agent-based modelling of energy infrastructure transitions
127
References
Aissaoui, A. (2006) ‘Market risks in a changing LNG world: exploring alternative mitigation
strategies for MENA projects’, Middle East Economic Survey, Vol. XLIX, No. 44,
http://www.mees.com/postedarticles/oped/v49n44-5OD01.htm.
Asbjørnsen, O. (1992) System Engineering Principles, Skarpodd, Houston.
Axelrod, R. (1997a) ‘Advancing the art of simulation in the social sciences’, in R. Conte,
R. Hegselmann and P. Terna (Eds.) Simulating Social Phenomena, Berlin/New York:
Springer, pp.21–40.
Axelrod, R. (1997b) The Complexity of Cooperation: Agent-Based Models of Competition and
Collaboration, Princeton, NJ: Princeton University Press.
Ball, A., Schneider, K., Fairhead, L. and Short, C. (2004) ‘The Asia Pacific LNG market: issues
and outlook’, Technical report, Australian Bureau of Agricultural and Resource Economics,
Canberra.
Bijker, W., Hughes, T. and Pinch, T. (1987) The Social Construction of Technological Systems:
New Directions in the Sociology and History of Technology, Cambridge, MA: MIT Press.
Boer, C.A., Verbraeck, A. and Veeke, H.P.M. (2002) Distributed Simulation of Complex Systems:
Application in Container Handling.
Borshchev, A. and Filippov, A. (2004) ‘From system dynamics and discrete event to practical
agent-based modeling: reasons, techniques, tools’, The 22nd International Conference of the
System Dynamics Society, Oxford, England, http://www.xjtek.com/file/142.
Box, G., Hunter, W. and Hunter, J. (2005) Statistics for Experimenters: Design, Innovation, and
Discovery, Hoboken, NJ: John Wiley & Sons, Inc.
Boyson, S., Corsi, T. and Verbraeck, A. (2003) ‘The e-supply chain portal: a core business model’,
Transportation Research Part E, Vol. 39, No. 2, pp.175–192.
Brito, D. and Hartley, P. (2007) ‘Expectations and the evolving world gas market’, The Energy
Journal, Vol. 28, No. 1.
Bussmann, S., Jennings, N.R. and Wooldridge, M. (1998) Multiagent Systems for Manufacturing
Control, Berlin: Springer.
CEC (2009) ‘Commission regulation (ec) no. 244/2009 of 18 March 2009 implementing directive
2005/32/ec of the European parliament and of the council with regard to ecodesign
requirements for non-directional household lamps’, Official Journal of the European Union,
Vol. 52, No. L076, pp.3–16.
Chappin, E.J.L. and Dijkema, G.P.J. (2008a) ‘Agent-based modeling of energy infrastructure
transitions’, International Conference on Infrastructure Systems – Building Networks for a
Brighter Future, NGInfra, Rotterdam, The Netherlands.
Chappin, E.J.L. and Dijkema, G.P.J. (2008b) ‘On the design of system transitions – is transition
management in the energy domain feasible?’, IEEE IEMC: International Engineering
Management Conference, IEEE, Estoril, Portugal.
Chappin, E.J.L. and Dijkema, G.P.J. (2008c) ‘Towards the assessment of policy impacts on system
transitions in energy’, 31st IAEE International Conference, Bridging Energy Supply and
Demand: Logistics, Competition and Environment, IAEE, Istanbul, Turkey.
Chappin, E.J.L. and Dijkema, G.P.J. (2009) ‘On the impact of CO2 emission trading on power
generation emissions’, Technological Forecasting & Social Change, Vol. 76, No. 3,
pp.358–370.
Chappin, E.J.L., Dijkema, G.P.J. and Vries, L.J.d. (2009a) ‘Agent-based simulation of carbon
policies and power generation’, 32st IAEE International Conference, Energy, Economy,
Environment: The Global View, IAEE, San Francisco, USA.
Chappin, E.J.L. (2006) Carbon Dioxide Emission Trade Impact on Power Generation Portfolio,
Agent-based Modelling to Elucidate Influences of Emission Trading on Investments in Dutch
Electricity Generation, Delft University of Technology, Delft.
128
E.J.L. Chappin and G.P.J. Dijkema
Chappin, E.J.L., Dijkema, G.P.J. and Vries, L.J.d. (2009b) ‘Carbon policies: do they deliver in the
long run?’, in P. Sioshansi (Ed.) Carbon Constrained: Future of Electricity, Global Energy
Policy and Economic Series, Elsevier, in press.
Corsi, T.M., Boyson, S., Verbraeck, A., Van Houten, S., Han, C. and Macdonald, J.R. (2006) ‘The
real-time global supply chain game: new educational tool for developing supply chain
management professionals’, Transportation Journal, Vol. 45, No. 3, p.61.
Davis, C., Nikolic, I. and Dijkema, G. (2009) ‘Integration of life cycle assessment into agent-based
modeling: toward informed decisions on evolving infrastructure systems’, Journal of
Industrial Ecology, Vol. 13, No. 2, pp.306–325.
Dijkema, G. and Basson, L. (2009) ‘Complexity and industrial ecology: foundations for a
transformation from analysis to action’, Journal of Industrial Ecology, Vol. 13, No. 2,
pp.157–164.
Dym, C. and Little, P. (2004) Engineering Design: A Project-Based Introduction, USA: John
Wiley & Sons, Inc.
Enserink, B., Koppenjan, J. and Thissen, W. (2002) Analyse van Complexe Omgevingen, leerboek,
TBM TU Delft, Delft.
Epstein, J. and Axtell, R. (1996) ‘Growing artificial societies: social science from the bottom up’,
Complex Adaptive Systems, Washington, DC: Brookings Institution Press; MIT Press.
Fahey, L. and Randall, R.M. (1998) Learning from the Future, Hoboken, NJ: John Wiley & Sons,
Inc.
Forrester, J. (1958) ‘Industrial dynamics: a major breakthrough for decision makers’, Harvard
Business Review, Vol. 117, No. 4, pp.37–66.
Forrester, J.W. (1969) Principles of Systems: Text and Workbook, Chaps 1–10, 2nd prelim. ed.,
Cambridge, MA: Wright-Allen Press.
Geels, F. (2002) ‘Understanding the dynamics of technological transitions’, PhD thesis, University
of Twente, Enschede.
Geels, F.W. (2004) ‘From sectoral systems of innovation to socio-technical systems insights about
dynamics and change from sociology and institutional theory’, Research Policy, Vol. 33,
pp.897–920.
Global Insight (2007) ‘The LNG market – globally now to 2025 and the implications for the UK’,
Technical report, Global Insight, London.
Gordon, G. (1978) ‘The development of the General Purpose Simulation System (GPSS)’, History
of Programming Languages I, Table of Contents, pp.403–426.
Goupy, J. and Creighton, L. (2007) Introduction to Design of Experiments with JMP Examples,
3rd ed., Cary, NC: SAS Press.
Herder, P.M., Bouwmans, I., Dijkema, G.P., Stikkelman, R.M. and Weijnen, M.P. (2008)
‘Designing infrastructures using a complex systems perspective’, Journal of Design Research,
Vol. 7, No. 1, pp.17–34.
Hughes, T. (1987) ‘The evolution of large technological systems’, in W. Bijker, T. Hughes and
T. Pinch (Eds.) The Social Construction of Technological Systems: New Directions in the
Sociology and History of Technology, Cambridge, MA: MIT Press, pp.51–82.
IEA (2004) ‘Security of gas supply in open markets, LNG and power at a turning point’, Technical
report, International Energy Agency, Paris, France.
Iman, R., Helton, J. and Campbell, J. (1981) ‘An approach to sensitivity analysis of computer
models, part 1. introduction, input variable selection and preliminary variable assessment’,
Journal of Quality Technology, Vol. 13, No. 3, pp.174–183.
Jennings, N.R. (2000) ‘On agent-based software engineering’, Artificial Intelligence, Vol. 117,
No. 2, pp.277–296.
Jones, R.W. (1965) ‘The structure of simple general equilibrium models’, The Journal of Political
Economy, Vol. 73, No. 6, p.557.
Agent-based modelling of energy infrastructure transitions
129
Kim, J.S. and Kalb, J.W. (1996) ‘Design of experiments: an overview and application example’,
Medical Device & Diagnostic Industry Magazine, Vol. 1996, No. 3, p.78.
Leontief, W. (1998) ‘Environmental repercussions and the economic structure: an input-output
approach’, International Library of Critical Writings in Economics, Vol. 92, pp.24–33.
Loorbach, D. (2007) Transition Management – New Mode of Governance, Utrecht, The
Netherlands: International Books.
Maier, M. and Rechting, E. (2002) The Art of Systems Architecting, Boca Raton, FL: CRC Press.
McKay, M., Conover, W. and Beckman, R. (1979) ‘A comparison of three methods for selecting
values of input variables in the analysis of output from a computer code’, Technometrics,
Vol. 21, pp.239–245.
Ministry of VROM and SenterNovem (2005) ‘Allocation plan for CO2 emission allowances
2005 – 2007, Dutch national allocation plan regarding the allocation of greenhouse gas
emission allowances to companies’, Technical report, SenterNovem, http://www.senternovem
.nl/mmfiles/Dutch%20allocation%20plan040820 tcm24-110316.pdf.
Morikawa, T. (2008) ‘Natural gas and LNG supply/demand trends in Asia Pacific and Atlantic
markets*’, Technical report, The Institute of Electrical Engineers of Japan.
NGI (2007) ‘Dubai bids for LNG center stage with storage hub’, http://www.businesswire.com/
news/google/20070621005898/en.
Nikolic, I. (2009) ‘Co-evolutionary process for modelling large-scale socio-technical systems
evolution’, PhD thesis, Delft University of Technology.
Nikolic, I., Dijkema, G., Dam, K.v. and Lukszo, Z. (2006) ‘General methodology for
action-oriented industrial ecology’, IEEE International Conference On Networking, Sensing
and Control, Ft. Lauderdale, Florida, USA.
Nikolic, I., Dijkema, G.P. and van Dam, K.H. (2009) ‘Understanding and shaping the evolution of
sustainable large-scale socio-technical systems – towards a framework for action-oriented
industrial ecology’, in M. Ruth and B. Davidsdottir (Eds.) The Dynamics of Regions and
Networks in Industrial Ecosystems, Edward Elgar, ISBN: 978-1-84720-742-5.
Ottens, M., Franssen, M., Kroes, P. and Van De Poel, I. (2006) ‘Modelling infrastructures as
socio-technical systems’, Int. J. Critical Infrastructures, Vol. 2, Nos. 2–3, pp.133–145.
Rosenberg, R.R. and Karnopp, D.C. (1983) Introduction to Physical System Dynamics, New York,
NY: McGraw-Hill, Inc.
Rotmans, J. (2003) Transitiemanagement: Sleutel voor een duurzame samenleving, Van Gorcum,
Assen.
Rotmans, J., Kemp, R., Asselt, M.v., Geels, F., Verbong, G. and Molendijk, K. (2000) ‘Transities
& transitiemanegement: De casus van emissiearme en-ergivoorziening’, Technical report,
International Centre for Integrated Studies (ICIS) and Maastricht Economic Research Institute
on Innovation and Technology (MERIT), Maastricht.
Rotmans, J., Kemp, R. and Van Asselt, M. (2001) ‘More evolution than revolution: transition
management in public policy’, Foresight, Vol. 3, No. 1, pp.15–31.
Schieritz, N. and Milling, P. (2003) ‘Modeling the forest or modeling the trees – a comparison of
system dynamics and agent-based simulation’, 21st International Conference of the System
Dynamics Society, New York.
Tusiani, M. and Shearer, G. (2007) LNG: A Nontechnical Guide, Tulsa, OK: PennWell
Corporation.
Van Dam, K.H. and Lukszo, Z. (2009) ‘Model factory for socio-technical infrastructure systems’,
in R. Negenborn, Z. Lukszo and J. Hellendoorn (Eds.) Intelligent Infrastructures, Intelligent
Systems, Control and Automation: Science and Engineering, Springer.
Weiss, G. (2000) Multiagent Systems, A Modern Approach to Distributed Artificial Intelligence,
Cambridge, MA: MIT Press.
130
E.J.L. Chappin and G.P.J. Dijkema
Notes
1
2
Under the pressure of the industry, the Dutch government acquires additional emission rights
through the CDM. In the Dutch ETS allocation plan, it was announced that government
reserved 600 million euros for this purpose, the equivalent of 20 Mtonne CO2 rights (Ministry
of VROM and SenterNovem, 2005).
When the CO2 price exceeds the penalty level, agents will rationally choose to pay the penalty
rather than purchase more CO2 credits. Consequently, this penalty level functions as a price
cap for the CO2 market.