Co-simulation of Smart Grids: Dynamically Changing Topologies

in Failure Scenarios

Lukáš Gryga and Bruno Rossi

Masaryk University, Faculty of Informatics, Brno, Czech Republic

Keywords:

Smart Grids, Smart Meters, Power Consumption, Co-simulations.

Abstract:

Smart Grids represent an important instance of cyberphysical systems for the energy sector. Due to the many

layers involved and the complexities of interrelations, co-simulations have emerged as a way to integrate

results from different simulators. In this paper, we propose a study of the possibilities of simulating node

failure scenarios with a modiﬁcation of the Mosaik co-simulation platform to allow for dynamic topologies

changes. We show how co-simulations can help in determining the impact of different failure patterns using a

sample scenario of households and PV units.

1 INTRODUCTION

Cyber-physical systems (CPS) represent collections

of interconnected computing devices linked to the

physical world by means of sensors and actua-

tors (Alur, 2015). As one of the most relevant critical

CPSs, a Smart Grid (SG) is the convergence of Infor-

mation and Communication Technology (ICT), sen-

sors, and power systems to supply electricity to con-

sumers via two-way digital communication with key

goals of reaching high levels of reliability, efﬁciency,

and resilience (Fang et al., 2011).

While SGs provide many beneﬁts, such as more

ﬂexible demand/response adaptations, or smart ser-

vices provided, the complexity and integration comes

at the expense of the higher importance of reliability

and security aspects (Farhangi, 2010; Lamba et al.,

2019). Complexity and dimensionality are two major

challenges to simulate cyber malfunctions and fail-

ures of Smart Grids. The issue of dimensionality de-

rives from the large number of components that inter-

act concurrently in the cyber and physical parts. The

issue of complexity arises from a wide range of cyber

elements and their interdependencies with the physi-

cal components (Lopez et al., 2018). One way to deal

with complexity and dimensionality issues is to de-

couple the system to smaller parts and test them in

isolation — however, missing in this way the holistic

run of the system (Chren et al., 2016; Schvarcbacher

et al., 2018).

https://orcid.org/0000-0002-8659-1520

For this reason, the coupling of the simulations

of different aspects (e.g., power and network domain

simulations) emerged as a way to study and predict

the reliability of SGs in complex scenarios. The cou-

pling of distinct simulators, each one running in their

own runtime environment is commonly deﬁned as co-

simulation. A co-simulation platform allows the inter-

connection of software simulators and hardware emu-

lators to study the behaviour of different components

under speciﬁc conditions and scenarios (Vogt et al.,

2018; Strasser et al., 2014).

In this paper, we deal with reliability scenarios

simulated by means of the Mosaik co-simulation plat-

form (Schütte et al., 2011) that we adapt for changes

to node topologies at runtime, allowing to study fail-

ure scenarios in which some power nodes are sub-

ject to failures. For example, a scenario could be a

massive storm causing part of the distribution grid to

shut down, propagating failures to other nodes of the

power network. Overall, we show how simulations

and co-simulations can be useful to study failure pat-

terns once the power network topology has been mod-

elled. We have the following contributions:

• The adaptation of the Mosaik co-simulation plat-

form for runtime dynamic topologies changes to

simulate node failures;

• An experimental evaluation simulating two differ-

ent types of failure patterns and showcasing how

co-simulation can help in evaluating failure pat-

terns;

The paper is structured as follows. In section 2 we

Gryga, L. and Rossi, B.

Co-simulation of Smart Grids: Dynamically Changing Topologies in Failure Scenarios.

DOI: 10.5220/0010448700630069

In Proceedings of the 6th International Conference on Complexity, Future Information Systems and Risk (COMPLEXIS 2021), pages 63-69

ISBN: 978-989-758-505-0

discuss the background about Smart Grids and their

reliability. In section 3 we discuss about adaptation to

the Mosaik framework (Schütte et al., 2011) for dy-

namic changes to topologies for SG failure modelling.

In section 4 we showcase the modiﬁed co-simulation

platform with an experimental evaluation to compare

different patterns of failure emergence. Section 5 con-

cludes the paper.

2 BACKGROUND

A Smart Grid (SG) can be seen as a modern

power grid enabling two-way power ﬂow and at

the same time bi-directional communication between

power suppliers and consumers (Fang et al., 2011).

Controllers, sensors, computer systems, automation

equipment are integrated to provide efﬁcient trans-

mission of electricity, fast restoration in case of fail-

ures, and overall reduced costs for utilities (Goel

et al., 2015). SGs achieve lower power costs for

consumers, reduced peak demand, increased integra-

tion of large-scale renewable energy systems—among

other beneﬁts. Real-time monitoring and recovery of

power generation and distribution is another key char-

acteristic, as the actual state of the grid is monitored

and reported to the network, adapting the power out-

put to the real needs. SGs are also important to in-

crease the usage of renewable sources (e.g., solar en-

ergy), as excess energy generated can be sold, more-

over attempting to reach balance in demand response

programs (Siano, 2014).

Decentralization of the SG led to the introduction

of microgrids. A microgrid is an independent and

small network of electricity users (consumers / pro-

sumers) that can carry out operations independently

from the centralized grid and even isolate itself from

the rest of the power network in case of failure of

the grid, with the so-called islanding mode (Hebner,

2017). The decentralization attempts to avoid sin-

gle points of failure and "domino effects" of failures,

leading potentially to large blackouts.

Internet of Things (IoT) devices also play an im-

portant role in the context of SG, as they open the way

to so-called smart energy scenarios. For example, a

household using a solar-power system (with batter-

ies and sensors) can decide on the best moments for

recharging a Electric Vehicle (EV) (Hebner, 2017).

The many sensors, devices, automation equipment

and different layers pose many challenges both in

terms of security and reliability concerns, with the

smart grids expecting to provide self-monitoring and

self-healing capabilities.

Reliability of a software system is deﬁned as the

probability that the system will function as required

without failures and errors for a certain period of time

(O’Connor, 2012). From this deﬁnition, reliability

can be seen as associated to the concept of quality of

service. In traditional grids, reliability, and security of

cyber elements were not considered as critical for the

overall stability of power grids because of the less rel-

evant dependencies between cyber and physical parts.

This situation changed drastically with Smart Grids:

the incorporation of modern IT aspects moved SG

more towards cyber-physical systems, which brings

tighter constraints related to security and reliability.

In modern grids, the cyber part is essential for the

proper functioning of the whole power grid, as it pro-

cesses sensors’ data, monitors the grid, handles se-

curity, and makes power distribution decisions (Lei

et al., 2018). The physical part is thus strongly depen-

dent on the availability of the cyber layer. Power-grid

stability needs to take into account also possible cyber

failures. However, SGs can have a relative advantage

over traditional grids when examining fault-tolerance

and fault-recovery: Supervisory Control And Data

Acquisition (SCADA) systems employed in SGs can

communicate with a multitude of sensors in real-time.

In case of failures, SCADA systems can locate the

area subject to the failure and start self-healing and

notiﬁcation activities. Such large availability of data

can support a multitude of anomaly detection algo-

rithms and platforms (Rossi and Chren, 2020; Lip

cák

et al., 2019).

As every CPS, also SG can be modelled formally

as the interaction of reactive systems constrained by

temporal constraints. Deﬁning formally the compo-

nents can allow to ﬁnd conditions under which the

constraints are violated. Some parts can be even

solved by means of analytical equations. However,

the complexity of the interactions and the compu-

tational complexity required by the many solvers,

forces in many cases to rely on the usage of simu-

lations. However, due to the many system states, sim-

ulations can be only use to disprove the correctness

constraints deﬁned by invariants in the falsiﬁcation

process, that is simulations cannot cover all the possi-

ble run cases (Alur, 2015).

Due to the complexity for analytical solutions,

many models were proposed over time for failure

propagation in power grids, each one covering differ-

ent aspects (Guo et al., 2017; Cai et al., 2016; Xiao

and Yeh, 2011): topological models (based on net-

work analysis), stochastic simulation models (proba-

bilistic simulations), statistical models, dynamic sim-

ulation, and interdependent models (studying cou-

pling of interdependent networks cyber and physical).

Since traditional power grids were designed and

COMPLEXIS 2021 - 6th International Conference on Complexity, Future Information Systems and Risk

built more than a century ago, there are many mod-

els that examine the reliability of the traditional grids

that have been adapted to Smart Grids. For example,

Barabási-Albert Network Model is based on a simple

failure propagation model to calculate a reliability in-

dex (Chassin and Posse, 2005). Another model that

evaluates the reliability of large photovoltaic power

systems connected to a power grid is proposed in

(Zhang et al., 2012). Co-simulations of IT and power

networks have also been widely used for studying SG

reliability and security concerns (e.g., (Chromik et al.,

2017)).

Simulation in the context of SG has often involved

covering both the power and the network aspects: on

one side, either the simulation of power generation

(e.g., PyPower) or real power emulation devices (RT-

Lab), on the other side network simulators (e.g., OM-

NET++ or NS-3), or speciﬁc SG ad-hoc solutions

(e.g., GridLab-D, a whole power distribution system

simulator and analysis platform).

In this paper, we utilize the Mosaik co-simulation

framework to showcase the usage for the deﬁnition

of failure scenarios under the condition of changing

network topologies.

3 MOSAIK FRAMEWORK

ADAPTATION

Mosaik is an open-source, discrete-event co-

simulation framework written in Python (Schütte

et al., 2011). It allows interconnecting independent

SG simulators, handling the dataﬂow between them.

Mosaik also ensures time synchronization between

the different simulators and the simulation can run

with some timing constraints or in real-time mode.

Mosaik is based on SimPy, which is a more general

process-based simulation framework. Although

Mosaik handles data exchange between entities of

different simulators, interconnection of data among

entities in one simulator is beyond the scope of

Mosaik.

Mosaik scenario API provides a way to create

the simulation scenarios, create entities, and establish

dataﬂow between them. When the scenario is built, it

can be run. Mosaik wraps simulators to SimPy pro-

cesses deﬁned in the scheduler module and gives con-

trol to the SimPy framework. For the goal of this pa-

per, to simulate an attack or a malfunction of a power

node we needed a way to change dynamically the

dataﬂow (topology) at runtime. For this reason, we

extended the Mosaik framework to support dynamic

topology changes.

In the adopted Mosaik version (2.5.2) the Sce-

nario module allows to deﬁne complex topologies

with hundreds of entities and large number of con-

nections. However, dynamic topology changes at run-

time are not possible. Once a scenario is deﬁned and

run, Mosaik passes control to the Mosaik scheduler

that starts the complex chain of processes and events.

When the simulation ﬁnishes, the scheduler returns

control to Mosaik. There is no way to change the

topology once the simulation has been started. Fur-

thermore, many Mosaik modules do not support dy-

namic changes in topology. For example, mosaik-web

for visualization of simulation results cannot update

the initial topology once the simulation has started.

3.1 Changes in Scenario Module

The Mosaik scenario module provides an API for

starting simulators, instantiating models of the simu-

lators, and connecting entities (models’ instances) of

different simulators to establish a dataﬂow.

As a ﬁrst step for dynamic topologies support, we

extended the module of the disconnect method to re-

move edges from the dataﬂow graph and the entity

graph. A simple way to simulate node failure, shut-

down, or malfunction is to disconnect the node from

the grid, but the scenario module does not provide

ways to disconnect. We added the possibility to track

time and connection plus disconnection of nodes and

update the scenario’s attribute that stores the last time

the topology was changed. Simulators can request

this information from Mosaik and check whether the

topology has changed since the last step.

The topology builder module is implemented as

a Mosaik simulator. This approach is based on the

idea that the scheduler already manages synchroniza-

tion with the simulator. The scheduler never advances

a successor simulator that is further in the dataﬂow

chain, before a predecessor simulator stepped further

in time than the successor simulator. In other words,

the topology builder module is inserted at the begin-

ning of the dataﬂow chain, so it can perform topology

changes on time when needed. The topology builder

simulator communicates with the scenario module ex-

clusively via high-level API as the low-level API does

not support ‘connect/disconnect‘ methods. Therefore,

the simulator has to be started via in-process mode.

The topology builder implements a single model,

and we can create a single entity (topology-change) of

the model. The topology-change entity has to be con-

nected to all simulators whose entities are connected

or disconnected by this entity. It is enough to connect

the topology-change entity to a single arbitrary entity

of each simulator to ensure synchronization. When

we create the topology-change entity, we pass a dic-

Co-simulation of Smart Grids: Dynamically Changing Topologies in Failure Scenarios

Figure 1: Modiﬁed topology from mosaik-demo (adapted from (Ofﬁs, 2012)).

tionary that describes all changes in topology during

the simulation as a parameter of the simulators’ create

method.

4 EXPERIMENTAL RESULTS

Since the modiﬁed Mosaik can disconnect nodes dy-

namically, it can aid in simulating new categories of

scenarios where nodes are subject to failures due to

device faults or due to cyberattacks. Further details

such as the data from the experimental runs can be

found in (Gryga, 2020).

4.1 Testing Topology

To showcase a failure scenario, we use the topology

from the mosaik-demo repository (Fig. 1). The sce-

nario simulates power distribution in a grid organized

in ﬁve main branches.

Mosaik-pypower is a model of power distribu-

tion grid that consists of nodes that are connected via

power lines. There are two types of nodes. PQ node

(PQbus in Fig. 1) takes a real power P and reactive

power Q as input and calculates voltage magnitudes

and angle as outputs. The reference node (RefBus in

Fig. 1) has a constant voltage magnitude and angle.

There is exactly one in the grid and computes reac-

tive power as output. We can also specify parameters

of lines connecting the nodes, such as resistance per

kilometer and maximum current (Ofﬁs, 2019). All PQ

nodes are connected to the grid and organized into

branches. A branch is a sequence of PQ nodes that

starts at the central node, and every other node in the

branch is further from the central node. The branch

can fork into subbranches.

Mosaik-household-sim consists of a single house

model. This model represents households that con-

sume power from the grid. It loads a consumption

proﬁle for the speciﬁc entities from the ﬁlesystem. A

proﬁle in mosaik-demo tries to capture realistic house-

hold consumption patterns, like lower consumption

during the night and higher during the day time.

Mosaik-csv consists of a simple model that loads

data in time:parameter:value format and provides this

data as output. This model is used to simulate PV

units (PV in Fig. 1) that supply the power to the grid

during daytime. It generates the most energy at noon

and lower amount in the morning and evening.

The Monitor is a custom simulator developed for

the purpose of the showcase (red failure nodes at the

end of branches in Fig. 1). It takes voltage as input

and calculates an average voltage per each hour.

4.2 Failures Scenario

To demonstrate the topology builder module, we

modiﬁed the original demo simulation and analyzed

scenarios where PV units are subject to failures and

how they affect the voltage at the end of branches

(Fig. 1). We add monitors at the end of the branches.

For easy referencing, we mark nodes at the end of

the branches/subbranches as branch 1-5 (Fig. 1). The

length of the branch is the number of PQ nodes from

RefBus to the last node in the branch. For example,

COMPLEXIS 2021 - 6th International Conference on Complexity, Future Information Systems and Risk

the length of branch four is 9, and the length of branch

ﬁve is 11.

We showcase two different patterns of failures,

that we called random and deterministic, based on

the way PV units are subject to failures. We follow

a similar deﬁnition of these two patterns as deﬁned in

the context of SG Intrusion Detection Systems (IDS)

(Chromik et al., 2017). A random strategy shuts down

nodes all over the grid. Such strategy could be more

similar to random occurrences of failures. The deter-

ministic strategy chooses PV units based on the loca-

tion of previously failing units, mimicking more cas-

cading failures or intentional cyberattacks. The goal

is to compare the impact of both strategies on the sta-

bility of the grid. We also discuss how failures in one

branch affect the stability of other branches.

4.3 Voltage Drops and Undervoltage

According to the European standard EN 50160 (Start,

1995), a voltage drop is a sudden lowering of the ef-

fective voltage value to a value of between 90% and

1% of the stipulated nominal value, followed by the

“immediate” recovering of this voltage. The most

common causes of voltage drop are starting currents

(for example, inrush current for a capacitor) and short

circuits. Undervoltage, unlike voltage drop, has a

longer-lasting character and is also deﬁned as lower-

ing of the effective voltage value to a value between

90% and 1% of the stipulated nominal value. Un-

dervoltage occurs when facilities ask for more power

than the power grid can deliver. It can be caused either

when facilities suddenly increase their power con-

sumption or when power generators decrease the sup-

ply of power to the grid. Both voltage drop and under-

voltage can lead to huge problems, such as the drop-

out of production processes, product quality prob-

lems. IT systems are also susceptible to voltage drops,

which can cause damages.

4.4 Simulation

In our scenario, we primarily focus on undervoltage,

since it is caused by the insufﬁcient power supply to

the grid, and we simulate failures of power generators.

The showcase has three phases. We make 100 runs

of the simulation in each phase and then process the

results. PV units are randomly connected to the grid

for each of the runs, although the distribution of PV

units over the runs is the same for each phase.

In the ﬁrst phase, we measure the average volt-

age magnitude (V m) for each branch in the stable grid

(Fig. 1). First we run simulations with a random dis-

tribution of PV units. In the end, we calculate the av-

erage from all runs. The voltage magnitude is around

230V . It is higher at noon, because of the increased

power production of PV units.

In the next two phases, we compare the impact of

the two strategies (random, deterministic). We mea-

sure the negative impact as a difference between av-

erage voltage per each hour measured at monitored

nodes (red nodes in Fig. 1) in the stable grid and av-

erage voltage measured at the same nodes in the grid

where PV units fail. Figure 2 shows the summarized

results from all runs in the form of boxplots for each

measured branch and both strategies. The boxplots

marked as random or deterministic in Fig. 2 show a

deviation of V m from the average V m of the stable

grid (as in calculated in phase one) for each branch

from 1:00 PM to 2:00 PM.

In the second phase, we simulate a failure of PV

units and how it affects voltage in the grid. We chose

the bad-case scenario when the PV units fail at the

same time. This phase employs a random strategy

for picking PV units to shut down. Five randomly

picked PV units are shut down in each run for one

hour at 1:00 PM, time when PV units are the most

active. We observe a slight drop in voltage in most of

the branches. This slight drop is not alarming since

the worst result is voltage drop by 9V in branch 5,

and the undervoltage is deﬁned as 10% lower voltage

(around 23V in our case).

In the third phase, we use a deterministic strategy

for picking PV units in the grid. We again chose the

bad-case scenario when the PV units are shut down at

1:00 PM for one hour. However, now we shut down

ﬁve PV units connected to branch 5 if there are at

least ﬁve PV units connected to it. If there are not

ﬁve PV units connected to branch 5, we shut down all

that are connected. It can simulate real-world scenar-

ios where, for example, a storm affects devices in the

same area, or attackers performing a more sophisti-

cated attack.

4.5 Outcome of the Simulation

We can see a drop in voltage, in comparison to the

voltage in the stable grid, in almost all branches in

the case of PV unit failures using both strategies. The

only exception is branch 2 that shows slight increase

of voltage in most of the simulation runs.

When we take a closer look at the random pattern,

we can see that the grid is capable of compensating for

the sudden failures of PV units without a signiﬁcant

drop of voltage in none of its branches. The highest

drop occurs in Branch 5, but the distribution of the

deviation of V m still lies far from the undervolage ac-

cording to the deﬁnition in section 4.3. Since the av-

Co-simulation of Smart Grids: Dynamically Changing Topologies in Failure Scenarios

Figure 2: Voltage drop during PV unit failures (13:00 PM).

erage voltage in the grid is around 230V, the voltage

lowering would have to be at least 23V to be catego-

rized as the undervoltage.

The deterministic pattern, which shuts down PV

units only from Branch 5, leads to more interesting

results. We can see that the strategy affects Branch 5

more signiﬁcantly than the random pattern, while it

does not cause drops in voltage in other branches. All

three quartiles of Branch 5 boxplot lie in the low val-

ues. Such a drop in voltage can cause problems to

speciﬁc devices connected to the grid even though it is

still not undervoltage according to the deﬁnition (sec-

tion 4.3). To note that this lowering of voltage has a

long-lasting character (one hour in our case).

Overall, the random strategy affects the voltage

stability of the grid slightly while deterministic strat-

egy cause more signiﬁcant destabilization in the part

of the grid — meaning that failures in a single branch

remain local. On the other hand, the grid could,

for example, reroute power from branches where the

voltage is high enough (Branch 2 in our scenario) to

compensate for the deﬁciency in the power supply in

Branch 5.

5 CONCLUSION

In this paper, we used the open source co-simulation

framework Mosaik, adapting it to allow run-time

changes of topology and availability to simulate dif-

ferent smart grid failure scenarios. We provided the

implementations details and then set-up an experi-

ment to test the newly implemented functionality of

dynamic topology changes at runtime. We examined

two different patterns, random and deterministic, to

simulate possible failure cases that can occur in the

grid due to device issues or cyberattacks and evalu-

ated the stability of the grid. These scenarios were

used to showcase how co-simulations can be used to

study also more complex scenarios of cascading fail-

ures in the grid, taking into account and modelling

nodes disconnections at runtime.

ACKNOWLEDGMENT

The research was supported from ERDF/ESF "Cy-

berSecurity, CyberCrime and Critical Informa-

tion Infrastructures Center of Excellence" (No.

CZ.02.1.01/0.0/0.0/16_019/0000822).

REFERENCES

Alur, R. (2015). Principles of cyber-physical systems. MIT

Press.

Cai, Y., Cao, Y., Li, Y., Huang, T., and Zhou, B. (2016).

Cascading failure analysis considering interaction be-

tween power grids and communication networks.

IEEE Transactions on Smart Grid, 7(1):530–538.

Chassin, D. and Posse, C. (2005). Evaluating north amer-

ican electric grid reliability using the barabasi-albert

network model. Physica A: Statistical Mechanics and

its Applications, 355:667–677.

Chren, S., Rossi, B., and Pitner, T. (2016). Smart grids

deployments within eu projects: The role of smart

meters. In Smart Cities Symposium Prague (SCSP),

2016, pages 1–5. IEEE.

Chromik, J. J., Pilch, C., Brackmann, P., Duhme, C., Ever-

inghoff, F., Giberlein, A., Teodorowicz, T., Wieland,

J., Haverkort, B. R., and Remke, A. (2017). Context-

COMPLEXIS 2021 - 6th International Conference on Complexity, Future Information Systems and Risk

aware local intrusion detection in scada systems: A

testbed and two showcases. In 2017 IEEE Int. Con-

ference on Smart Grid Communications (SmartGrid-

Comm), pages 467–472.

Fang, X., Misra, S., Xue, G., and Yang, D. (2011).

Smart grid—the new and improved power grid: A

survey. IEEE communications surveys & tutorials,

14(4):944–980.

Farhangi, H. (2010). The path of the smart grid. IEEE

power and energy magazine, 8(1).

Goel, S., Hong, Y., Papakonstantinou, V., and Kloza, D.

(2015). Smart grid security. Springer.

Gryga, L. (2020). Mosaik framework for co-simulations

of smart grids reliability. Bachelor thesis, Masaryk

University, Brno.

Guo, H., Zheng, C., Iu, H. H.-C., and Fernando, T. (2017).

A critical review of cascading failure analysis and

modeling of power system. Renewable and Sustain-

able Energy Reviews, 80:9 – 22.

Hebner, R. (2017). Nanogrids, microgrids, and big data:

The future of the power grid. IEEE Spectrum Maga-

zine, page 23.

Lamba, V., Šimková, N., and Rossi, B. (2019). Recom-

mendations for smart grid security risk management.

Cyber-Physical Systems, 5(2):92–118.

Lei, H., Chen, B., Butler-Purry, K. L., and Singh, C. (2018).

Security and reliability perspectives in cyber-physical

smart grids. In 2018 IEEE Innovative Smart Grid

Technologies - Asia (ISGT Asia), pages 42–47.

Lip

cák, P., Macak, M., and Rossi, B. (2019). Big data plat-

form for smart grids power consumption anomaly de-

tection. In 2019 Federated Conference on Computer

Science and Information Systems (FedCSIS), pages

771–780.

Lopez, J., Rubio, J. E., and Alcaraz, C. (2018). A resilient

architecture for the smart grid. IEEE Transactions on

Industrial Informatics.

O’Connor, Patrick, K. A. (2012). Practical Reliability En-

gineering. John Wiley & Sons.

Ofﬁs (2012). Mosaik quickstart. Available from https://

mosaik.readthedocs.io/en/latest/quickstart.html.

Ofﬁs (2019). mosaik-pypower. Available from https://

bitbucket.org/mosaik/mosaik-pypower/src/master/.

Rossi, B. and Chren, S. (2020). Smart grids data analysis:

A systematic mapping study. IEEE Transactions on

Industrial Informatics, 16(6):3619–3639.

Schütte, S., Scherfke, S., and Tröschel, M. (2011). Mo-

saik: A framework for modular simulation of active

components in smart grids. In 2011 IEEE First Inter-

national Workshop on Smart Grid Modeling and Sim-

ulation (SGMS), pages 55–60.

Schvarcbacher, M., Hrabovská, K., Rossi, B., and Pitner,

T. (2018). Smart grid testing management platform

(sgtmp). Applied Sciences, 8(11).

Siano, P. (2014). Demand response and smart grids—a

survey. Renewable and sustainable energy reviews,

30:461–478.

Start, D. (1995). A review of the new cenelec standard en

50160. IET Conference Proceedings, pages 4–4(1).

Strasser, T., Stifter, M., Andrén, F., and Palensky, P. (2014).

Co-simulation training platform for smart grids. IEEE

Transactions on Power Systems, 29(4):1989–1997.

Vogt, M., Marten, F., and Braun, M. (2018). A survey and

statistical analysis of smart grid co-simulations. Ap-

plied Energy, 222:67–78.

Xiao, H. and Yeh, E. M. (2011). Cascading link failure

in the power grid: A percolation-based analysis. In

2011 IEEE International Conference on Communica-

tions Workshops (ICC), pages 1–6.

Zhang, P., Wang, Y., Xiao, W., and Li, W. (2012). Relia-

bility evaluation of grid-connected photovoltaic power

systems. IEEE Transactions on Sustainable Energy,

3(3):379–389.

Co-simulation of Smart Grids: Dynamically Changing Topologies in Failure Scenarios