A Risk-Based Methodology for Defining the Time of 
Intentional Controlled Islanding 
 
Pablo Fernández-Porras 
University of Costa Rica 
San Jose, Costa Rica 
jpablofp20@gmail.com 
 
Mathaios Panteli 
The University of Manchester 
Manchester, UK 
mathaios.panteli@manchester.ac.uk 
 
Jairo Quirós-Tortós 
The University of Manchester 
Manchester, UK 
jairoquirostortos@ieee.org 
 
Abstract--Power systems are operated close to their stability 
limits and this increases the probability of cascading outages 
leading to large-area blackouts. To mitigate these cascading 
outages, intentional controlled islanding (ICI) has been suggest-
ed as an effective corrective control strategy that splits the pow-
er system into suitable subsystems (islands). There are two pri-
mary aspects associated with ICI: i) where to island, and 
ii) when to island? This work proposes a risk-based methodolo-
gy that compares in a real-time fashion (i.e., quickly enough) the 
overall risk of the system without and with islanding (when an 
ICI scheme is activated) to define a suitable time for system 
splitting. Simulation results on the IEEE 9-bus system demon-
strate the effectiveness of the proposed methodology in identify-
ing the crossing point between the risks of the system without 
and with islanding, which in turns results in the most suitable 
time for the creation of islands. 
 
Index Terms--Blackout, intentional controlled islanding, risk 
assessment, system splitting. 
 
I. INTRODUCTION 
Interconnected power systems are prone to instabilities, as 
they are operated close to their stability limits [1]. This in-
creases the likelihood of cascading outages that typically lead 
to large-area power system blackouts [1], [2]. 
To mitigate these cascading outages, intentional controlled 
islanding (ICI) has been suggested as an effective corrective 
control action [2]. ICI – also referred as system islanding or 
system splitting – is an adaptive control strategy for power 
systems under emergency and in extremis states [2]-[4]. Is-
landing methods intentionally split the bulk power system 
into self-sustained islands (i.e., electrically separated subsys-
tems), and they are aimed to be used after instabilities have 
been detected, but before the system becomes uncontrollable. 
There are two primary aspects associated with ICI [5]-[7]: 
i) where to island, and ii) when to island. While the former 
aims to find the optimal set of transmission lines that must be 
disconnected to split the system into islands, the latter  seeks 
to define the optimal time (moment) for splitting the network. 
This work focuses on answering the question of “when to 
island”. To determine the set of lines that optimally splits the 
system in a real-time fashion (i.e., a few seconds), this work 
implements the approach presented in [4] (see section II-A). 
To address the question ‘when to island, decision trees and 
Prony-based methods have been previously proposed [7]-[8]. 
Although these methods can define a suitable time for split-
ting the electrical network based on system conditions rec-
orded in a database (i.e., they use previous information to 
learn how to predict future scenarios), unexpected system 
changes and unpredictable events may result in incorrect 
times for splitting, and thus large-area blackouts. 
In addition, given that ICI can be classified as System In-
tegrity Protection Schemes (SIPS) [9] and the interest for 
assessing their reliability and risk has attracted the interest of 
several researchers, such as [10]-[12], it is nowadays crucial 
to perform an adequate risk assessment of ICI schemes. 
This work proposes a risk-based methodology to define a 
suitable time to undertake system splitting, i.e., to help an-
swer the question ‘when to island’. The proposed methodolo-
gy assesses in a real-time fashion (i.e., quickly enough, 10ms 
in this work) the overall risk of the electrical power system 
(i.e., the impact of the event multiple by its probability) with-
out islanding and with islanding, thus avoiding delays. In 
contrast to other approaches, it uses data available in real-
time, thus avoiding the use of historical data. The suitable 
time for system splitting is defined as the crossing point be-
tween the two curves, i.e., when the risk without islanding is 
higher than the risk with islanding. The simplicity and scala-
bility of the proposed methodology is expected to help opera-
tors in the decision-making process of when to split the pow-
er system to mitigate cascading outages. 
The remainder of this extended abstract is organized as fol-
lows. Section II provides some background on ICI and risk 
assessment. Section III details the risk-based methodology. 
Section IV presents a set of simulation results that will be 
detailed and extended in the full paper. Finally, Section V 
concludes this paper and provides future venues for research. 
II. BACKGROUND INTETIONAL CONTROLLED ISLANDING 
AND RISK ASSESSMENT 
A.  Intentional Controlled Islanding 
ICI is an effective approach to mitigate cascading outages. 
This section summarizes the approach presented in [4] that is 
implemented in this work to determine the optimal set of 
lines to partition the bulk power system. The existing meth-
odology is used in a real-time fashion to evaluate the risk 
with islanding at every time sample (see section III). 
As thoroughly detailed in [4], this approach is based on 
spectral clustering [13], i.e., a computationally efficient graph 
theoretic technique that can partition systems using the eigen-
values and eigenvectors of a (Laplacian) matrix associated 
with the graph that represents the power system. To create 
electrically separated islands, the approach in [4] minimizes 
the power flow disruption while ensuring that each island 
contains only coherent generators (generators that oscillated 
similarly). This approach significantly enhances the transient 
stability of the islands, and it is implemented here given its 
efficiency and scalability to any power system size. The fol-
lowing steps are executed to determine the set of transmission 
lines that splits the electrical network into k islands [4]: 
1. Build a graph G that represents the power flow of the 
network (with n buses) at the moment of splitting. 
2. Compute the eigenvectors φ1, φ2, … , φk associated with 
the first k eigenvalues of LN (1) [13]: 
 
[𝑳𝑵]𝑖𝑗 = {
1 𝑖𝑓 𝑖 = 𝑗
−𝑤𝑖𝑗 √𝑑𝑖𝑑𝑗⁄ 𝑖 ≠ 𝑗 𝑎𝑛𝑑 𝑖𝑗 is a branch
0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒.
 (1) 
 
where wij is the power flow between buses i and j and 
𝑑𝑖 = ∑ 𝑤𝑖𝑗
𝑛
𝑗=1  [13]. 
3. Place the eigenvectors 𝝋1, 𝝋2, … , 𝝋𝑘 as columns to create 
𝐗 = [𝝋1 𝝋2 … 𝝋𝑘]. 
4. Normalize each row of X to obtained Y. 
5. Group every load-bus with the nearest generation-bus 
based on the proximity between the vectors in Y that were 
mapped in the Euclidean space in point 4. 
6. Group the clusters that contain the coherent generators. 
The steps mentioned above (detailed in [4]) are imple-
mented every time sample (defined in this work as 10ms) in 
the proposed risk-based methodology to determine the impact 
of splitting the power system (the risk of the system with 
islanding), as further explained in section III. 
B. Risk Assessment 
The risk of an electrical event is given by the product of 
the probability of occurrence of the event and its impact on 
the network (e.g., the total load shedding), as follows: 
 
𝑅𝑖𝑠𝑘 = 𝑃𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 × 𝐼𝑚𝑝𝑎𝑐𝑡 (2) 
 
When considering ICI schemes as an available control ac-
tion to the system operators, then the risk introduced to the 
network by an undesirable operation of these schemes needs 
to be taken into account in the risk assessment procedure 
(similar to the risk assessment of other SIPS). The failure 
modes of the ICI schemes mainly include the failure to oper-
ate following the occurrence of the electrical event and the 
incorrect/unnecessary operation when there is no event in the 
network. An electrical event comprises here a large disturb-
ance and its subsequent consequences that can threaten the 
network stability (loss of synchronism, etc.). The impact of 
these failure modes varies and depends on the prevailing 
system conditions. For instance, if the ICI scheme splits the 
system into islands when not required during normal electri-
cal conditions, then it may not have a high impact on the 
system stability. If, on the other hand, it splits it during 
stressed conditions, then it may trigger a series of cascading 
outages that may compromise the network stability and lead 
to customer interruptions, as it happened in the Irish disturb-
ance of 2005 [14]. 
In order to evaluate the risk of ICI schemes, the probabili-
ties of failure on demand (PFD) and fail-safe (PFS, i.e., the 
probability of ICI unnecessary operations) are required. Then, 
the impact of these failure modes on network stability needs 
to be assessed. In this study, the amount of load shedding as a 
result of the ICI failure modes is used as an impact index. 
Finally, the risk introduced by having the ICI schemes in 
operation is given by the product of the probability of the ICI 
failure modes (PFD and PFS, respectively) and the impact of 
these undesirable events. 
Fault tree analysis (FTA) is used here for estimating PFD 
and PFS. FTA is a systematic method for identifying the 
events or combination of events that can lead to the top event 
of the fault tree [15], i.e., ICI undesirable operation in this 
case. The reliability data of the individual components of the 
ICI scheme, e.g., logic processor, circuit breakers, communi-
cation links etc., are assumed for illustration purposes. Partic-
ularly, the PFD and PFS of the individual ICI components are 
required, which are then inserted in the fault trees for estimat-
ing the overall PFD and PFS of the ICI scheme. 
III. PROPOSED RISK-BASED APPROACH FOR DEFINING THE 
TIME OF ICI 
Fig. 1 shows the proposed risk-based methodology for de-
termining the suitable time for applying the ICI scheme. It 
should be noted that the assessment of the risk of the system 
without islanding and with islanding is undertaken in parallel, 
considering that the ICI scheme has been switched in opera-
tion by the system operator and it is ready to be activated 
when required. If the ICI scheme is not implemented, i.e., w/o 
ICI, then the methodology estimates the impact of the electri-
cal event using the amount of load shedding as an impact 
index, as discussed earlier. If the ICI scheme is in operation, 
i.e., with ICI, then the PFD and PFS using FTA, along with 
the impact of the ICI failure modes, are estimated. The prob-
ability of the occurrence of the electrical event threatening the 
network stability is also taken into account by the proposed 
methodology, which is fed as input to the risk assessment for 
both w/o the ICI scheme and with the ICI scheme. 
Following this, the risk without (Rw/o) and with (Rw) the ICI 
scheme is estimated and compared. This procedure is repeat-
ed every time sample (10ms in this work) to determine the 
suitable time to implement the ICI scheme based on the over 
all system risk, i.e., the risk introduced by the electrical event 
Electrical Event
Risk 
Assessment
Rw/o > Rw ?
Time to Apply ICI
Yes
No
Probability of 
electrical event
Impact 
assessment of 
electrical event
w/o ICI
Estimation of 
PFD and PFS
Impact 
assessment of 
ICI failure modes
with ICI
 
Fig. 1 Proposed risk-based methodology for defining the time to apply ICI 
and the risk by the ICI undesirable operation. If Rw/o < Rw, 
then the ICI solution (found using the spectral clustering-
based methodology) is not applied. However, if Rw < Rw/o 
then the corresponding time is defined as the suitable time to 
undertake islanding actions, as the overall system risk with 
the ICI scheme becomes smaller than the risk without the ICI 
scheme. This leads to a risk-based approach for deciding 
when to apply the islanding in order to minimize the risk of 
the electrical disturbance. Further details on the simulation 
procedure will be provided in the full paper. 
IV. SMULATION RESULTS 
This section presents the simulation results. The dynamic 
model of the IEEE 9-bus test systems [16] is used here to 
demonstrate, with time-domain simulations, the effectiveness 
of the risk-based methodology in defining the suitable time to 
split the bulk power system. 
As detailed in section II-A, the spectral clustering-based 
methodology [4] is used every time sample (i.e., 10ms) to 
determine the optimal set of lines that splits the power system 
into self-sustained islands that also satisfy the generator co-
herency constraint. The actual implementation of the spectral 
clustering-based methodology is not presented here as this is 
out of the scope of the paper (see [4] for more details). 
All time-domain simulations are performed in DIgSILENT 
PowerFactory [17], and the proposed methodology has been 
implemented in MATLAB [18]. 
Test case description: It is considered that at time t = 0.1s, 
a three phase-to-ground solid fault occurs near bus 5 at line 5-
7, and is cleared after local relays open the faulty line at 
t = 0.25s. If no control action is undertaken, it can be noted in 
Fig. 2 that the system loses synchronism (two coherent 
groups of generators can be clearly seen in Fig. 2). Indeed, it 
 
Fig. 2. Results for the IEEE 9-bus system without islanding. 
 
Fig. 3.  Risk assessment without and with islanding 
can be noted in Fig. 2 that the system collapses given that no 
control action was undertaken. 
Fig. 3 shows the risk of the system without islanding from 
the moment that the fault is cleared (t = 0.25s). It can be seen 
that the risk without islanding increases with time, given that 
no control action is taken and the system is collapsing. 
In parallel, the risk with islanding (given a solution found 
by the ICI methodology) is calculated. Fig. 3 shows the risk 
with islanding. As it can be observed, when t = 0.45s, the risk 
with islanding becomes smaller than the risk without island-
ing. Therefore, according to a risk-based methodology (Fig. 
1), this time is considered to be the most suitable time to split 
the power system into two islands (given that two coherent 
groups of generators were created after the disturbance). 
Additional details on the coherent generator groups and the 
islands created will be presented in the full paper. 
Fig. 4 shows the results of implementing the islanding so-
lution found by the spectral clustering-based methodology [4] 
0 1 2 3 4 5
0
0.5
1
1.5
V
o
lt
a
g
e
 (
p
u
)
0 1 2 3 4 5
0.9
1
1.1
S
p
e
e
d
 (
p
u
)
0 1 2 3 4 5
-200
0
200
Time(s)
R
o
to
r 
A
n
g
le
 (
d
e
g
)
0.25 0.35 0.45 0.55 0.65 0.75 0.85 0.95
0
0.5
1
1.5
2
2.5
3
3.5
4
Time (s)
R
is
k
 (
k
£
 /
 h
)
 
 
With Islanding
Without Islanding
 Fig. 4 Results for the IEEE 9-bus system with islanding. 
at the time found by the risk-based methodology presented in 
this paper i.e., at t = 0.45s. As it can be observed, the system 
recovers and two stable islands are created. Further details of 
the spectral clustering-based methodology will be detailed in 
the full paper.  
V. CONCLUSION AND FUTURE WORK 
This extended abstract proposes a risk-based methodology 
to define the most suitable time to undertake system splitting. 
This addresses the “when to island” aspect in the intentional 
controlled islanding procedure, which benefits the power 
system operators in the decision making when to undertake 
islanding actions.  
The proposed risk-based methodology has been tested us-
ing the dynamic model of the IEEE 9-bus system. Time-
domain simulations have been carried out to demonstrate the 
effectiveness of the approach, which seeks to identify the 
crossing point between the risk of the system without and 
with islanding (if islanding is required). 
It has been shown that the most suitable time to split the 
power system to mitigate cascading outages is when the risk 
of the system with islanding becomes smaller than the risk of 
the system without islanding. 
The full paper will further detail the use of the spectral clus-
tering-based approach to find the optimal set of lines to split 
the system. Moreover, it will further discuss the risk assess-
ment undertaken without and with islanding that is eventually 
used in the risk-based methodology. Finally, it will fully dis-
cuss the implementation of the proposed risk-based approach 
on the IEEE 9-bus system and will provide further venues of 
research. 
REFERENCES 
[1] U.G. Knight, Power System in Emergencies. London, 2001. 
[2] M. Vaiman, et al., "Risk assessment of cascading outages: Methodolo-
gies and challenges," IEEE Trans. Power Syst., vol. 27, no. 2, pp. 631-
641, May 2012. 
[3] J. Quirós-Tortós, M. Panteli, V. Terzija, and P. Crossley, "On evaluat-
ing the performance of intentional controlled islanding schemes", 
IEEE PES General Meeting, pp. 1-5, 2013. 
[4] J. Quirós-Tortós, R. Sánchez-García, J. Brodzki, J. Bialek and, V. 
Terzija, "Constrained spectral clustering-based methodology for inten-
tional controlled islanding of large-scale power systems," IET Gen., 
Trans. & Dist., vol. 9, no. 1, pp. 31-42, Jan. 2015. 
[5] N. Senroy and G.T. Heydt, "A conceptual framework for the con-
trolled islanding of interconnected power systems," IEEE Trans. Pow-
er Syst., vol. 2, no. 2, pp. 1005-1006, May 2006. 
[6] M.M. Adibi, R.J. Kafka, S. Maram, and L.M. Mili, "On power system 
controlled separation," IEEE Trans. Power Syst., vol. 21, no. 4, pp. 
1894-1902, Nov. 2006. 
[7] N. Senroy, G.T. Heydt, and V. Vittal, "Decision tree assisted con-
trolled islanding," IEEE Trans. Power Syst., vol. 21, no. 4, pp. 1790-
1797, Nov. 2006. 
[8] N. Senroy and G. Heydt, "Timing of a Controlled Islanding Strategy," 
IEEE PES Trans. and Distr. Conference and Exhibition, May 2006.  
[9] J.D. McCalley and W. Fu, "Reliability of Special Protection 
Schemes," IEEE Trans. Power Syst., vol. 14, no. 4, pp. 1400-1406, 
Nov. 1999. 
[10] W. Fu, S. Zhao, J.D. McCalley, V. Vittal and N. Abi-Samra "Risk 
assessment for special protection systems", IEEE Trans. Power Syst., 
vol. 17, no.1, pp. 63-72, Feb. 2002. 
[11] H. Tsun-Yu and L. Chan-Nan, "Risk Informed Design Refinement of 
a Power System Protection Scheme", IEEE Trans. on Reliability, vol. 
57, no. 2, pp. 311-321, Jun. 2008. 
[12] M. Panteli, P.A. Crossley, and J. Fitch, "Quantifying the reliability 
level of system integrity protection schemes," IET Generation, Trans-
mission & Distribution, vol. 8, no. 4, pp. 753-764, Apr. 2014. 
[13] U. Von Luxburg, "A tutorial on spectral clustering", Statistics and 
Computing, vol. 17, no. 4, pp. 395-416, 2007.  
[14] ESB and National Grid, "Report on investigation into System Disturb-
ance of August 5th 2005," Dec. 2005. 
[15] W. E. Vesely, et al.: "Fault Tree Handbook", 1981. 
[16] P.M. Anderson and A. A. Fouad, "Power System Control and Stabil-
ity", 2nd ed. New York: IEEE Press, 2003.  
[17] DIgSILENT PowerFactory, ed. Heinrich-Hertz-Straße, Germany. 
[18] MATLAB R2010a: Natick, Massachusetts: The MathWorks Inc., 
2010.
 
0 1 2 3 4 5
0
0.5
1
1.5
V
o
lt
a
g
e
 (
p
u
)
0 1 2 3 4 5
0.9
1
1.1
S
p
e
e
d
 (
p
u
)
0 1 2 3 4 5
-200
0
200
Time(s)
R
o
to
r 
A
n
g
le
 (
d
e
g
)