Do We Need a New Architecture for Simulating Power Systems?

A Position Paper

Gary Howorth1, Ivana Kockar2

1Instituteof Energy and Environment, University of Strathclyde, 99 George St., Glasgow, UK
2Institute of Energy and Environment, University of Strathclyde, 204 George St., Glasgow, UK

{gary.howorth, ivana.kockar}@strath.ac.uk

Keywords: Agent Based Modelling, Multi Agent Systems, Multiscale, Multilayer, Power, Simulation, Smartgrid.

Abstract: The delivery of electric power and organisation of electric power systems is an incredibly complex system,
which will become ever more complex with an increased penetration of Distributed Energy Resources (DER),
including electric vehicles and renewables. Optimization of such a system using conventional techniques is
difficult and fraught with a myriad of issues, so simulation provides a more holistic approach to understanding
the evolving issues. We argue that although power simulation frameworks exist they may be inadequate for
simulating a more complex and evolving smart power grid infrastructure. A brief overview of research on
existing systems is provided and this paper argues for the development of a distributed multi-scale, multi
layered hybrid ABM/MAS system.

1 INTRODUCTION achieve. Essentially, our model of the power system

has to represent its physical and commercial aspects,
In his work, Watts (2003) characterised power including the grid, the market, the generators, the
systems as complex systems. Others have followed demand, DERs etc. The model also needs to include
this work but have focused on using graph based an ability to represent and study adapting participant
statistical measures to help understand their stability behaviours. Therefore, it is now recognized that a
e.g. cascades and other issues. However Holland more holistic approach is required, one that ultimately
(1999) suggests that a graph based statistical or requires a more sophisticated simulator. This short
mathematical approach gives us no insight as to how position paper presents initial scope/ideas for an
patterns e.g. price or power flows in our instance ABM/MAS system that meets our needs for an
arise. It tells us nothing about emergence in what is a electrical power simulator and can look at complex
Complex Adaptive System (CAS). This emergence in interactions in a future power system (Smart Grid). It
our context is an important element to understand and is therefore a concept paper and is not meant to solve
enable us to develop (i) policy rules for future system and give detailed architectural designs for our
operation and (ii) appropriate short-term technical framework at this stage.
control measures. Currently only a simulation The paper is organised as follows. In Section 2,
approach appears to provide us with a route to relevant examples of academic research on ABM and
understanding complex dynamics in these systems. MAS are presented. Section 3 focusses on
However, Agent Based (ABM) and Multi Agent ABM/MAS systems used in electrical engineering,
System (MAS) modelling could also provide us with while section 4 proposes the conceptual design of a
an important arsenal in discovering these patterns. new power system simulator. Finally, Section 5
Holland also suggests that as we add stochastic concludes this paper.
mobile agents to the system the potential for new
rules and patterns to emerge tends to grow rapidly, 2 OVERVIEW OF ABM/MAS
resulting in many more persistent states as a result of LITERATURE
system emergence. In our case, electric vehicles
(EV’s) are the stochastic mobile agents of Holland’s There have been many surveys on ABM and MAS
discussions. Understanding this complex system systems (Heath et al., 2009, Kantamneni et al., 2015,
behaviour seems to be an important ideal for us to
Leon et al., 2015, McArthur et al., 2007) plus many also use system dynamics. The agent behaviours can
dedicated webpages and tutorials outlining what they be modelled using JavaScript, but this is too limiting
can do in terms of Agent Communication Languages, for our purposes, as we require a fully-fledged
openness, programming language choice and so on. asynchronous Object Orientated Programming
Typically, the systems focus on a particular area of (OOP) language to model complex interactions. More
research i.e. they are specialised to analyse particular sophisticated agents using Java and Neural Nets
features or behaviour, or have other limitations. Only which are linked into AnyLogic using a Java Archive
a few specifically focus on power system file (JAR) have been developed in (Wallis and Paich,
applications. There are now over 70 ABM/MAS 2017). This method requires a greater degree of
systems in existence with a typical lifespan of 4-5 programmer intervention to link in the various
years. Only a few of these systems /designs (e.g. Jade components and is less flexible than our proposed
(Bellifemine et al., 2007) , Repast (Cardoso, 2009)) ) conceptual design.
have lifespans in excess of 10 years. EMLab has based their system on the Neo4J graph
Few researchers propose aggregation of systems database (Merkl Sasaki et al., 2017), rather than build
or components or simple reuse. However, Cardoso a Relational Database Management System (RDMS).
(2015) in his paper on SAJas proposes the use of an This is an open source/commercial system used by
API to join Repast to Jade, with a justification that many to analyse Twitter feeds and relationships. It is
“multi-agent based system simulations (MABS) a very efficient and can be used to store knowledge
focus on applying MAS to model complex social maps, power networks and, most importantly, the
systems typically involving a large agent population. relationships between agents in different layers and
Several MAS frameworks exist, but they are often not between agents on the same layer. It can in the right
appropriate for MABS”. In a similar vein, Gormer et circumstances be a faster than a normal database
al. (2011) propose the JRep framework for simulating (RDMS). It can also be used to quickly analyse
an agent-based airport scenario, linking Repast networks and identify problem nodes for example.
(ABM) to Jade (MAS). As they note, “existing agent Due to its features which also fit nicely with the
frameworks focus on either the macro or the micro typical representations of power grids (i.e. they are
perspective”, but don’t combine the two. The node based) it may be a useful base for a future power
objective of the proposed combination of an ABM simulator framework.
and MAS frameworks will allow us to better Furthermore, the Mosaik platform (Rohjans et al.,
understand power system organisation/operation. 2013) is designed to allow reuse of components like
Matlab and other “simulators to create large-scale
3 POWER MODELS IN ABM/MAS Smart Grid scenarios”. It is written in Python,
provides an API for connecting these different
Although there are over 70 ABM/MAS systems in simulators including MATLAB, and uses JSON to
existence few have been developed to address power communicate between packages.
systems, with categorisation and analysis of MAS
applications in power provided in (Sujil et al., 2016). 4 AN IDEAL ABM/MAS POWER
Table 3 in that review summarizes papers that have SIMULATION SYSTEM
focused on lower level distributed simulation, while
Table 4 provides a useful breakdown of papers that An ideal ABM/MAS power simulator would provide
deal with specific power issues e.g. markets, the following functionality:
generators etc.  A design which allows us to understand how
For example, AMES (Repast) (Li and Tesfatsion, different behaviours of the various power
2009), ECMAS (Repast) (Conzelmann et al., 2005), system agents (generators aggregators,
EMLab (Agent Spring) (De Vries et al., 2013) are consumers and policy makers etc.) will affect
specific modelling environments that have power the system technically and commercially. For
system implementations, while “lower level example, how will power flows across the
modelling” of multi agent systems using Presage2 is system change? How will prices change at
presented in (Chen et al., 2016, Macbeth, 2015). various nodes in the system? Will they be too
Furthermore, Anylogic (AnyLogic, 2018), which is high? Will resulting power flows cause
not specifically designed for power systems, is a congestion in the system and require new
proprietary system that could be used and has an investment?
architecture design whose logic allows analysts to  Test out new policy rules or potential control
model not only agents, but also discrete events and algorithms;
 Try out different agent behaviour techniques temperature controlled loads, system operators,
e.g. policy agent rules, different agent learning regulators, and companies;
paradigms;  To reuse existing software components where
 Understand how voltage extremes (which ever possible;
would result in system performance
degradation or failure) might be generated by 4.1 Proposed Architecture
behaviours of participants in the system, which
in turn might be driven by policy makers rules; These requirements appear to drive us to a conceptual
The framework will also need to have the following design that would include the following elements
features: shown in Figure 1.
 Has both a market and network physical (i.e. We discuss the main elements of this proposed
power flow) layers that are incorporated into architecture in the sections below.
market clearing and bidding representations;
 Is extensible and is truly layer based allowing 4.1.1 Asynchronicity
us the ability to switch in and out different
layers and change agent behaviours as needed; A review of ABM surveys and systems shows that
 Easy to use; ABM (macro level) models are typically
 Could be solved in distributed manner so to synchronous, whereas micro level systems such as
allow analysis of large scale networks; Jade (MAS) are typically asynchronous. Youssefmir
 Plug in based while using a Component and Huberman (1997) have investigated the impact of
modularity architecture; asynchronicity on system performance. Figure 2
 Have models of agents representing various below shows such an impact from this paper. Note
actors such as generators, loads, electric that the system is usually stable and is punctuated
vehicles, aggregators, storage, atomic and with periods of instability.

Figure 1: Proposed high-level architecture

 Figure 2: Asynchronicity impacts on system dynamics.

Youssefmir and Huberman's paper considers asynchronous investments on price patterns.

multiple agents who take decisions and act on the Although written nearly 15 years ago, few authors
system simultaneously in an asynchronous manner to and simulator designers have taken this approach,
improve their utility. Our problem domain has exactly which as they argue is more realistic. All of the power
these characteristics and will incorporate learning and system simulators that we have investigated assume
adaptive agent behaviour. Initial experiments on a synchronous investments. We know from our own
simple power system model show similar patterns of domain experience, that in the real world, power
punctuated bursts of activity followed by “stability”. investors do not act synchronously. In our context,
Cornforth et al. (2005) discusses MAS agent earlier large capital investments in infrastructure may
update strategies using a cellular automata (CA) “lockout” later investments, so timing is crucial.
framework as a case study. They examine updating Modelling asynchronous behaviour in our context
strategies associated with some real life systems is therefore extremely important, so we would argue
where the agents behave with different asynchronous that models that ignore asynchronicity would find it
update schemes and compare this with synchronous difficult to predict future system states accurately.
updating. The paper provides results on the dynamics Our domain has elements of all the examples we cite
of the CA system, under the different update schemes above. We therefore propose that any future power
and shows that the outputs can be significantly simulation environment provide a mechanism to
different. Again initial experiments on a power switch between modes of synchronization.
system show similar behaviours.
In a future smart grid system, peer to peer (P2P) 4.1.2 Multi-Scale (Equation Free Modelling)
bargaining/interactions will have an important impact
both on local conditions and further afield in the From a CAS perspective, emergence occurs when
wider power system. By their nature these events in one scale (micro) are propagated to another
transactions are asynchronous, but other parts of the scale (macro) and vice versa. Capturing those effects
system will have synchronous interactions e.g. like (Holland, 1999), is key to identifying and
market clearing. Designing and testing the understanding emergent behaviour in systems. In the
interactions between these types of system e.g. P2P context of the power domain, we suggest that it is
and the system operator, will allow us to understand important that system modellers investigate, these
how the system might perform in the future. A phenomena, so that they can design appropriate
simulator with the ability to try out different mitigation strategies. The multi-scale architecture
synchronous/asynchronous protocols does not allows us to model these propagation effects. It also
currently exist in the power domain and would be a fits well with the idea that we need to combine ABM
useful addition to the power engineers toolkit. (macro) and MAS (micro) architectures.
In finance (Jacobs et al., 2004) developed the However, developing models that can simulate a
“JLM stock market simulator” to look at the effect of combination of events that occur at both the second
(for generators, EV’s) and the years’ timescale (for new conceptual design allow users to easily add,
investments in infrastructure), are typically define or remove layers, to allow experimentation
computationally inefficient. We require some kind of with different designs. This would be easier using a
glue or bridge to join these timescales. graph database structure as the links in the database
There have been many papers on multi-scale would define the layers and their interconnections.
simulations, in recent years, and this provides a
potential solution for our specific problem area.
However, as we discussed above, our systems
typically are “stable” for large periods and are
punctuated with bursts of activity. Equation free
modelling (DeAngelis and Yurek, 2015, Kevrekidis
et al., 2004, Kevrekidis et al., 2003, Kevrekidis and
Samaey, 2009, Kevrekidis, 2004, Le Maître and
Mathelin, 2010) provides a promising
viewpoint/solution for this particular aspect and we
believe warrants further investigation, particularly in
the methodology to trigger the micro level simulation.
In this regard, there have been far less papers focused
Figure 3: Multi-layer concept.
on this specific aspect, especially in recent years. This
approach has not been implemented in the area of 4.1.4 Distributed and Scale
power systems simulation and therefore
corresponding techniques need to be developed. Many of the models we have investigated have been
developed using smaller problems and therefore may
4.1.3 Multi-Level be difficult to scale. To provide a realistic simulation,
our problem space requires that a potential model can
Although there has been growing interest in simulate tens of thousands of power nodes with
developing models on multi levels and multi time thousands of power consuming and producing
scales, there still only a few concrete examples devices attached to each of those nodes. This drives
(Sarjoughian et al., 2001, Ferreira et al., 2015). The us to a potential distributed architecture.
layered approach is discussed in many papers, but
typically as a conceptual model, rather than used as a 4.1.5 Power Simulation Specifics
programming paradigm. This layer or multi-level
model also fits well with the conceptual model A power system can be represented as a cause and
presented by SGAM (Santodomingo et al., 2014) for effect diagram, as shown in Figure 4, which indicates
Smart Grid interactions in power. flows, interactions and relationships between
We would propose that any new conceptual actors/agents e.g. generators. Note that that this
design adopts a multilayer structure so that it can diagram will evolve through time, e.g. 5-10 years ago
capture different views of the system represented as EV’s would not have existed on this diagram.
layers in a model (see Figure 3), such as a physical
layer (devices power nodes, flows, congestion),
market layer (prices) etc. We also propose that any
 Figure 4: Representation of a power system.

Considering this, one of the necessary features of a

new proposed architecture is that it can easily allow
changes of the relationships and flows.
We take as given that in the power domain that
any model would also need to represent power flows
and be able to “clear” the market on a large scale. This
would necessitate that any framework have a
methodology and a database structure/design suitable
for power and particularly for designs associated with
the evolving smart grid area and its new participants.
Links to existing power system simulators e.g.
MATPOWER (Murillo-Sánchez et al., 2013,
Zimmerman et al., 2011) should be considered.

4.2 The Suitability of Existing Systems

Figure 5: Navajo blanket of an ideal ABM/MAS power
Using a “traffic light” ratings approach for gap and
simulator.
needs analysis we have created a “Navajo blanket” of
our power research problem to help us understand
Each row represents a potential need or
how four systems stack up against the requirements
requirement for our ideal simulator. The four columns
for an “ideal” power system simulator (see Figure 5).
represent the four currently available systems that we
For brevity we only show one of the “Navajo”
are comparing. Scores from 1 – 10 have been given
blankets” analysed. Extensive experimentation with
to each cell, with 10 representing that the system
the various systems discussed above has also been
meets that current need. This is the equivalent of dark
performed and forms the basis of the scores presented
green in the figure. Zero represents that the system
in Figure 5.
does not currently have that functionality. Colours are
provided automatically by conditional formatting in
Excel 2016 using a graduated green - yellow – red using an agent orientated design methodology.
colour scale. Although models that are circled in Investigations into multiscale methodologies
Figure 5 can be regarded as most comprehensive in (particularly equation free modelling) are ongoing.
class, gaps for our requirements remain. Therefore,
the above analysis indicates that there is a need for a
hybrid ABM/MAS simulator to model electricity ACKNOWLEDGEMENTS
systems that fills these gaps.
It is clear from the proceeding sections that there
This research is funded by the UK Engineering and
are many useful ideas and components in the existing
Physical Science Research Council (EPSRC) and the
systems that can be reused, and therefore we would
International Strategic Partner (ISP) Research
not advocate the complete redesign of a simulation Studentship.
system, but the reuse of large parts of existing
simulators (e.g. EMLab, AMES).

REFERENCES
5 CONCLUSIONS AnyLogic. 2018. Any Logic Platfrom:Simulation Modeling
Software Tools & Solutions for Business [Online].
This paper introduced current research and advocates Available: https://www.anylogic.com/ [Accessed].
for the development of a hybrid ABM/MAS system Bellifemine, F. L., Caire, G. & Greenwood, D. 2007.
for power simulation. It presented an initial scope and Developing multi-agent systems with JADE, John
ideas for this potential simulator. The addition of a Wiley & Sons.
multitude of electric vehicles and DER’s to the power Cardoso, H. L. 2009. Introduction to RepastJ. University of
grid will exacerbate the complexity of the system and Portugal NIAD&R – Distributed Artificial Intelligence
and Robotics Group.
suggests that we should develop a distributed multi-
Cardoso, H. L. 2015. SAJaS: enabling JADE-based
scale, multi- layer architecture that will make use of simulations. Transactions on Computational Collective
reusable components. We are not advocating that we Intelligence XX. Springer.
should build a completely new system, but see a Chen, M., Syed, M., Sansano, E. G., McArthur, S. D., Burt,
solution in reusing and linking existing simulators, G. M. & Kockar, I. Distributed negotiation in future
while some changes to their designs to accommodate power networks: Rapid prototyping using multi-agent
better asynchronous modelling capabilities will be system. PES Innovative Smart Grid Technologies
required. We have looked at a number of systems Conference Europe (ISGT-Europe), 2016 IEEE, 2016.
applicable to our problem domain and found that they IEEE, 1-6.
Conzelmann, G., Boyd, G., Koritarov, V. & Veselka, T.
lacked the following:
Multi-agent power market simulation using EMCAS.
 An easy way to address asynchronicity and its Power Engineering Society General Meeting, 2005.
effects on power systems especially at scale i.e. IEEE, 2005. IEEE, 2829-2834.
millions of agents; Cornforth, D., Green, D. G. & Newth, D. 2005. Ordered
 A system that models both the micro levels asynchronous processes in multi-agent systems.
interactions and communications that are likely Physica D: Nonlinear Phenomena, 204, 70-82.
to occur between Consumers, EV’s and system De Vries, L. J., Chappin, É. J. L. & Richstein, J. C. 2013.
mangers in the longer term without the need to EMLab-Generation An experimentation environment
simulate every second for every agent; for electricity policy analysis.
DeAngelis, D. L. & Yurek, S. 2015. Equation-free
 Systems that adequately link OPF to the
modeling unravels the behavior of complex ecological
ABM/MAS environment at scale; systems. Proceedings of the National Academy of
Sciences, 112, 3856-3857.
Our proposal therefore adds the following main Ferreira, A., Leitao, P. & Vrba, P. 2015. Simulating smart
features to existing ABM/MAS power simulators in grid using a two-layer multiagent framework. 2982-
the power domain; An asynchronous/synchronous 2987.
base applied to an existing ABM and MAS simulator; Gormer, J., Homoceanu, G., Mumme, C., Huhn, M. &
the use of an equation free modelling technique or Muller, J. P. Jrep: Extending repast simphony for jade
some other alternative to simulate at the micro scale agent behavior components. Proceedings of the 2011
IEEE/WIC/ACM International Conferences on Web
when required, and an existing large scale OPF
Intelligence and Intelligent Agent Technology-Volume
model. We are currently looking to design and build 02, 2011. IEEE Computer Society, 149-154.
this simulator and are developing agent Heath, B., Hill, R. & Ciarallo, F. 2009. A survey of agent-
methodologies and detailed interaction protocols based modeling practices (January 1998 to July 2008).
 Journal of Artificial Societies and Social Simulation, Conference (ENERGYCON), 2014 IEEE International,
12, 9. 2014. IEEE, 751-758.
Holland, J. H. 1999. Emergence: From Chaos to Order, Sarjoughian, H. S., Zeigler, B. P. & Hall, S. B. 2001. A
Perseus Publishing. layered modeling and simulation architecture for agent-
Jacobs, B. I., Levy, K. N. & Markowitz, H. M. 2004. based system development. Proceedings of the IEEE,
Financial market simulation. Journal of Portfolio 89, 201-213.
Management, 142-152. Sujil, A., Verma, J. & Kumar, R. 2016. Multi agent system:
Kantamneni, A., Brown, L. E., Parker, G. & Weaver, W. concepts, platforms and applications in power systems.
W. 2015. Survey of multi-agent systems for microgrid Artificial Intelligence Review, 49, 153-182.
control. Engineering Applications of Artificial Wallis, L. & Paich, M. Integrating artifical intelligence with
Intelligence, 45, 192-203. anylogic simulation. Simulation Conference (WSC),
Kevrekidis, I. G., Gear, C. W. & Hummer, G. 2004. 2017 Winter, 2017. IEEE, 4449-4449.
Equation‐free: The computer‐aided analysis of complex Watts, D. J. 2003. Small Worlds: The Dynamics of
multiscale systems. AIChE Journal, 50, 1346-1355. Networks between Order and Randomness Princeton
Kevrekidis, I. G., Gear, C. W., Hyman, J. M., Kevrekidid, University Press
P. G., Runborg, O. & Theodoropoulos, C. 2003. Youssefmir, M. & Huberman, B. A. 1997. Clustered
Equation-free, coarse-grained multiscale computation: volatility in multiagent dynamics. Journal of Economic
Enabling mocroscopic simulators to perform system- Behavior & Organization, 32, 101-118.
level analysis. Communications in Mathematical Zimmerman, R. D., Murillo-Sánchez, C. E. & Thomas, R.
Sciences, 1, 715-762. J. 2011. MATPOWER: Steady-state operations,
Kevrekidis, I. G. & Samaey, G. 2009. Equation-free planning, and analysis tools for power systems research
multiscale computation: Algorithms and applications. and education. IEEE Transactions on power systems,
Annual review of physical chemistry, 60, 321-344. 26, 12-19.
Kevrekidis, Y. G. Equation-Free Modeling For Complex
Systems USFOE-Symposia, 2004.
Le Maître, O. & Mathelin, L. 2010. Equation‐free model
reduction for complex dynamical systems. International
Journal for numerical methods in fluids, 63, 163-184.
Leon, F., Paprzycki, M. & Ganzha, M. 2015. A review of
agent platforms. Technical Report, ICT COST Action
IC1404, Multi-Paradigm Modelling for Cyber-Physical
Systems (MPM4CPS).
Li, H. & Tesfatsion, L. 2009. Development of open source
software for power market research: The AMES test
bed. The Journal of Energy Markets, 2, 111.
Macbeth, S. 2015. Presage2 Software [Online]. Available:
https://github.com/Presage/Presage2 [Accessed].
McArthur, S. D., Davidson, E. M., Catterson, V. M.,
Dimeas, A. L., Hatziargyriou, N. D., Ponci, F. &
Funabashi, T. 2007. Multi-agent systems for power
engineering applications—Part I: Concepts,
approaches, and technical challenges. IEEE
Transactions on Power systems, 22, 1743-1752.
Merkl Sasaki, B., Chao, Joy & Howard, R. 2017. Graph
Databases for Beginners Neo4J. Neo4J.
Murillo-Sánchez, C. E., Zimmerman, R. D., Anderson, C.
L. & Thomas, R. J. 2013. Secure planning and
operations of systems with stochastic sources, energy
storage, and active demand. IEEE Transactions on
Smart Grid, 4, 2220-2229.
Rohjans, S., Lehnhoff, S., Schütte, S., Scherfke, S. &
Hussain, S. mosaik-A modular platform for the
evaluation of agent-based Smart Grid control.
Innovative Smart Grid Technologies Europe (ISGT
EUROPE), 2013 4th IEEE/PES, 2013. IEEE, 1-5.
Santodomingo, R., Uslar, M., Goring, A., Gottschalk, M.,
Nordstrom, L., Saleem, A. & Chenine, M. SGAM-
based methodology to analyse Smart Grid solutions in
DISCERN European research project. Energy