See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/280035199

An Optimal Power Flow Formulation Including Risk of Cascading Events

Conference Paper · May 2014

CITATION READS

1 41

4 authors, including:

Line Roald Bing Li

ETH Zurich ETH Zurich
40 PUBLICATIONS   447 CITATIONS    10 PUBLICATIONS   19 CITATIONS   

SEE PROFILE SEE PROFILE

Göran Andersson
ETH Zurich
436 PUBLICATIONS   15,918 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

PlanGridEV View project

BPES - Optimal sizing and control of balancing power in the future EU power system considering transmission constraints View project

All content following this page was uploaded by Line Roald on 14 July 2015.

The user has requested enhancement of the downloaded file.

 XIII SIMPÓSIO DE ESPECIALISTAS EM PLANEJAMENTO DA
OPERAÇÃO E EXPANSÃO ELÉTRICA
XIII SEPOPE XIII SYMPOSIUM OF SPECIALISTS IN ELECTRIC OPERATIONAL
18 a 21 de Maio 2014
May – 18th to 21st – 2014 AND EXPANSION PLANNING
FOZ DO IGUAÇU (PR) -
BRAZIL

An Optimal Power Flow Formulation Including Risk of Cascading Events

Line ROALD, Bing LI, Frauke OLDEWURTEL, Göran ANDERSSON

Power Systems Laboratory, ETH Zurich
Switzerland
SUMMARY
Higher penetration of fluctuating renewable energy sources (RES) and market liberalization increase
both the demand for transmission capacity and the focus on cost efficient operation of the grid. To find
a better trade-off between system security and the cost of operation, we propose to use a risk-based
approach to system security. Risk-based security criteria might help to maintain system security while
decreasing operational cost by utilizing additional information about the security level in the system.
This paper proposes a new, risk-based security measure, which intends to reflect the risk of cascading
events in a risk-based optimal power flow (OPF) formulation. The proposed risk measure limits the
probability that a cascade is initiated, focussing on cascading through a sequence of line outages. It
accounts both for the probability of the initial N-1 contingency (primary event) and probability of
further, dependent outages (secondary trip). The probability of a secondary trip is assumed to be a
function of both the pre- and the post-contingency line flows.
The risk measure is implemented in a DC OPF formulation, forming a risk-based security constrained
OPF (RB-SCOPF) where so-called “risk-based N-1 constraints” are used to limit the risk of cascade
initiation. To understand how the proposed formulation influences not only the risk of initiating a
cascade, but also the impact of the full cascade in terms of expected load shed, the risk-based criterion
is implemented in a Monte Carlo framework. Within this framework, we simulate the full cascade and
compute the expected load shed based on the probability of the initial outage and the probability of the
secondary trip.
The proposed method is demonstrated through a case study for the IEEE 118 bus system. We show
how the RB-SCOPF can be used to control the risk of cascade initiation, and compare the RB-SCOPF
solution with the solution of a standard security constrained OPF (SCOPF). Further, we investigate
how the choice of the risk limit influences both the generation cost and the expected load shed. A
lower risk limit increases the cost of generation, and decrease the expected load shed in most cases.
However, in some cases a lower risk limit might also increase the expected load shed, since the RB-
SCOPF does not consider how the cascade evolves after it has been initiated.

KEYWORDS
Risk-based OPF, cascading events, security constrained OPF, Monte Carlo based security assessment

e-mail: [email protected]; address: Physikstrasse 3, CH-8092 Zurich; tel:+41 44 632 6577

1. Introduction
Market liberalization and increasing penetration of fluctuating renewable energy sources (RES) are
leading to a situation where power is not necessarily produced close to where it is consumed, but
rather where production is cheap, where the wind is blowing or the sun is shining. This trend leads to a
higher demand for transmission capacity and an increased pressure on efficient use of existing grid
infrastructure. In order to reduce cost, the transmission system operators (TSOs) are often forced to
operate the system closer to the limits, which increases operational risk. To cope with this new
situation, methods to find a better trade-off between system security and cost of operation are needed.
In [1], it is argued that additional information is needed to operate the system securely while
maintaining a reasonably low cost. As [1] suggests, this additional information could be provided in
terms of a risk-based measure of system security. In this paper, we propose a risk measure for use in
operational planning, by introducing a method to include the risk of cascading events in an optimal
power flow (OPF) formulation.
There exist two main approaches to model risk in power system operation. On the one hand risk, can
be modelled through overall reliability parameters such as Expected Energy Not Served (EENS). This
type of risk measure reflects the impact on the consumers in the system, and accounts for interruptions
of supply that occur due to cascading events, e.g., through cascading trips of transmission lines.
However, the evaluation of those risk measures requires extensive calculations (e.g., Monte Carlo
simulations), and they are thus better suited for analysis of the risk in a given operating condition
[1-2], as opposed to inclusion in an optimization problem.
On the other hand, risk can be modelled in terms of violation of technical limits, e.g., as a function of
the line flow or the voltage magnitude after an N-1 outage. This type of risk measure does not consider
how a potential cascade would develop further and thus does not directly reflect the impact of a
cascade on the consumers (in contrast to e.g. Monte Carlo-based risk measures). However, when risk
is modelled as a function of line flows or voltages, there is a direct relation to operational parameters,
and the risk is more easily understood and influenced by the system operators. This second type of risk
measure is also relatively easy to evaluate and have been proposed for inclusion in OPF formulations
by, e.g., [3-5].
Although several OPF formulations assume that a larger violation of an N-1 limit is more dangerous
than a small violation [3-6], there are, to the best of the authors’ knowledge, currently no OPF
formulations that explicitly account for the risk of cascading events within the optimization algorithm.
This paper aims at introducing a risk measure that better reflects the risk of cascading events in the
OPF formulation while maintaining computational tractability. To assess the effectiveness of the
proposed risk measure, the actual risk level measured in terms of expected load shed is evaluated
through a Monte Carlo simulation, and compared with the risk level achieved with a standard security
constrained OPF (SCOPF).
The contribution of this paper is threefold. First, we introduce a risk-based security criterion reflecting
the risk that one initial N-1 contingency (primary disturbance) can lead to further, dependent outages
(secondary trips) and thus initiate a cascading event. We model the probability of secondary trips as a
function of both the post-contingency line flow and the change in line flow from the pre- to the post-
contingency state, based on the formulation in [3]. Second, we formulate a risk-based security
constrained OPF (RB-SCOPF) where so-called “risk-based N-1 constraints” are used to limit the risk
of cascade initiation. Finally, we implement the probabilistic model for secondary trips in a Monte
Carlo simulation. The Monte Carlo framework allows us to assess the impact of the full cascade, and
to assess whether the RB-SCOPF can help to control the risk faced by consumers.
In a case study for the IEEE 118 bus system, the RB-SCOPF is compared with a standard SCOPF. We
compare the two OPF formulations in terms of the risk of cascade initiation (evaluated directly from
the OPF solution), the expected load shed (obtained from the Monte Carlo simulation), and the cost of
the generation dispatch.
The remainder of this paper is organized as follows: Section 2 introduces the risk model, Section 3
formulates the OPF problem and Section 4 describes the setup of the Monte Carlo simulation. The
case study is presented in Section 5, and Section 6 summarises and concludes the paper.
1
2. Risk modelling
The aim of this paper is to propose a risk measure that better reflects the risk of cascading events in an
OPF formulation. As mentioned above, it is hard to model the full length of a cascading event within
the OPF due to the high computational requirements. Instead, we consider the risk of cascade
initiation, i.e., the risk that an initial N-1 outage might lead to further outages, and compute the risk
based only on the operating condition after the N-1 outage. Formulating the risk in this way complies
with the N-1 security criterion, which requires TSOs to operate the grid such that no single
contingency leads to a major disturbance [7]. However, the risk measure represents two main
improvements compared with the deterministic N-1 criterion:
First, the risk measure accounts explicitly for the probability of the initial outage, while the N-1
criterion treats all outages as if they have the same probability. By accounting for the outage
probability, the risk-based formulation can accept lower security margins for outages that are unlikely
to happen, and increase the margins for contingencies that have a high probability of occurrence.
Second, the risk measure assumes that the probability of a secondary trip is a piecewise continuous
function of the pre- and post-contingency line flow, whereas the deterministic N-1 criterion assumes
that the probability of a secondary trip rises from 0 to 1 as the line flow reaches the line capacity limit.
By including additional information about the magnitude of the power flows and the extent of the N-1
violation, the risk measure allows us to model effects that can otherwise not be considered.

2.1. Definition of the risk measure

The risk measure reflects both the probability of the initial outage and the severity of the resulting
operating condition. Here, is the probability of contingency , and is the severity of the
operating condition on line after outage . Since we consider the risk of cascade initiation, severity is
defined as the probability that outage will lead to a trip of line , denoted as the probability of
secondary trip . Thus, the risk related to line after contingency is expressed as

. (1)
We assume that the probability of each initial contingency can be estimated based on historical data
(e.g., from [8]), and be updated according to seasonal variations and weather effects. In particular, we
assume that it is independent of operational variables such as the generation dispatch or the line flows.

2.2. Probability of Secondary Trip

It is well known that errors in the protection system can lead to secondary trips. Statistics for outages
in Germany [8] show that about 0.5% of all outages in the transmission grid (220 and 380 kV) lead to
tripping of equipment which should have remained in service. It is relatively common to assume that
higher line flows lead to higher probability of failure, due to increased conductor sag and due to
increased probability of hidden failures in the protection system (e.g., [6, 9]). However, investigations
of the power flow on transmission lines during three major events involving cascading outages in the
US, presented in [10], show that the tripped lines experienced not only high post-contingency flows,
but also a large change in power flow from the pre- to the post-contingency state. Based on this data,
the authors of [10] suggest that the probability of failure should be expressed as a function of both the
post-contingency power flow and the power flow change and fit a logistic model to the obtained data.
The argumentation in [3] follows along the same lines, assuming that 1) a higher post-contingency line
flow increases the probability of a trip due to increased conductor sag and 2) a larger change in the
line flow from the pre- to the post-contingency state reflects higher transients and thus higher
probability of inadvertent tripping. The model for secondary trips used here is based on the model
proposed in [3], though extended to a more general form for inclusion in the OPF.
We consider a system with lines and contingencies on the N-1 contingency list.
denotes the line flows in the operating condition, with being normal operation and
being the operating condition after the contingency. The line flow of line is given by the
entry, denoted by . Based on the formulation proposed in [3], we model the probability of

2
secondary trip as a function of the power flow on line in normal operation, the post-
contingency power flow on line and the threshold , which is the power flow leading to a
certain trip of line . Assuming that the line flow jumps to a higher value, is expressed as [3]
(2)
.

By construction, (2) fulfils the assumptions on the probability of secondary trip, as increases
when either the post-contingency flow or the power flow change increase. Eq. (2)
can be interpreted as the ratio between the change in power flow, , and the initial
distance to the threshold of certain trip, , for a positive change in line flow.
However, the line flow can also “jump” in negative direction. In this case, the initial distance to the
threshold is given by | | as illustrated in Fig. 1. To account for both cases, we
formulate as
(3)
( )
⏟ ⏟

Finally, we assume that if is below 90% of the capacity limit , and set
if exceeds . The final expression for is thus given by

| |

( ) | | (4)

{ | |

Fig. 2 illustrates for positive post-contingency line flows and different values of .

1 1
Probability of secondary trip Prl,k

Probability of secondary trip Prl,k

negative flow change positive flow change P0L(1)=0.6 Pmax

0.8 0.8 L(1)

P0L(1)=0.7 Pmax
0.6 0.6 L(1)

P0L(1)=0.8 Pmax
L(1)
0.4 0.4
P0L(1)=0.9 Pmax
L(1)

0.2 0.2
P0L(1)=Pmax
L(1)
distance to lower limit distance to upper limit 0
0 0.9 Pmax Ptrip
0 0
0.9 Pmax
L L Ptrip
L L
k
Post-contingency
Post-contingency powerpower
flow flow
PkL(l) P[% of P] max
[% max
L(l)of P L(l)
]
L(l)
Figure 1: The power flow can change in either positive or Figure 2: Probability of secondary trip as a
negative direction. Dependent on the the power flow change, we function of the post-contingency flow , plotted for
consider either the distance to or as the line flow different values of .
change in the denominator of (3).

3
3. Formulation of the risk-based optimal power flow
In this section, the risk measure introduced above is incorporated in an OPF formulation to form a
risk-based security constrained OPF (RB-SCOPF). Starting from a standard SCOPF, we replace the
standard security constraints limiting the post-contingency line flow with risk-based constraints that
account for the risk of cascade initiation. The reformulation of the RB-SCOPF to a tractable
optimization problem is explained, and a method to define reasonable risk limits is proposed.

3.1. Standard SCOPF

The OPF is based on a DC power flow formulation. We consider a system with buses, lines and
generators. The vector describes the active power injections of the generators and the
loads at each bus in the operating condition , and the vectors and describes the
load and voltage angles at each bus. The post-contingency generator in-feeds and the post-contingency
line flows are modelled using linear sensitivity factors. The generation at generator after outage of
generator is given by

, (5)

where the matrix models how the lost generation is balanced by the remaining
generators. Assuming that all generators contribute based on their nominal maximum output, we set
⁄∑ for and . Further, we choose . The
post-contingency line flow on line after the outage of a generator or a line are given by (6) and (7).
(6) (7)
The Generalized Generation Distribution Factors describe how a change in
generation at generator changes the line flow on line [11], given that the power mismatch is
balanced according to . The Line Outage Distribution Factors describe how the
outage of line influences the flow on line [12]. The resulting SCOPF problem considering
outages is given by
(8)

subject to
, (9) , (11)
, (10) , (12)
for , (13)
for , (14)
, for . (15)
The cost function is given by (8), with being the cost coefficients of the generators. Eq. (9) is
the nodal power balance constraint, with being the nodal admittance matrix and
a matrix relating the generators to the buses. Eq. (10) sets the angle of the slack bus to
zero. The remaining constraints refer to the limits on line flows and generators in normal operation
(Eq. (11) and (12)) and in post-contingency conditions. Eq. (13) and (14) limits the post-contingency
line flow after line and generator outages, and (15) limits the post-contingency generator output.

3.2. Risk-based SCOPF

For the RB-SCOPF, we consider risk-based constraints for the post-contingency line flows following
either a generator outage or a line outage. The generator constraints (15) are not changed, since we do
not consider generator trips as part of the cascading mechanism, but rather as a hard limit. To obtain
the RB-SCOPF, we replace (13) and (14) by constraints that limit the risk as formulated in (1),

4
 , (16)
where is the largest acceptable risk. Assuming that the outage probability of the initial
outage is known a-priori, (16) can be reformulated as a constraint on the probability of secondary trip
, (17)
where ⁄ is a contingency specific limit on . To include (17) in the OPF problem, we
need to reformulate (17) as a limit on the post-contingency flow . Since is a piecewise
affine function of (as given by (4) and shown in Fig. 2) and is not known a-priori, the
reformulation is a quite difficult task. We therefore use an iterative approach which is explained in
Section 3.3. However, before we explain the iterative process, we will analyse how constraint (17) is
expressed for the three different cases in (4), i.e., for post-contingency flows in the range | |
, | | or | |.
We first formulate constraint (17) for post-contingency line flows | | . For this
case, is given as the point-wise maximum of two functions in (4). Since (17) represents an upper
bound on , we can avoid the max function by including both constraints in the optimization
problem. For a line outage , we express the power flow change on line as
, and replace (17) by (18) and (19).

(18) (19)

With some manipulations, (18) and (19) can be expressed as the risk-based N-1 constraint

. (20)
For a generator outage , we express the line flow change on line as ,
and get a similar risk-based constraint

. (21)
Comparing (20) to (13) and (21) to (14) reveals a very similar structure. However, when the
probability of secondary trip is , the power flow change has a higher weight compared
with the SCOPF formulation. On the other hand, since we assume that in most cases,
the risk-based constraints (20), (21) will often be less tight than (13), (14).
For the case with high or low post-contingency line flows (i.e., | | or | | ),
the probability of secondary trip is constant, and do not depend on the line flow. For the case of
a high post-contingency line flow | | , . This means that the outage will lead to
a trip of line independent of how much exceeds . If we choose , we allow
and thus allow to become arbitrarily large, i.e., | | . Thus, we can just delete constraint
(17) for contingencies with . For the case with low post-contingency line flow
| | , independent of the value of . This means that we do not need to
constrain the post-contingency flow to a value lower than , and thus the tightest possible
constraint on the line flow of line after outage is given by
(22)
(23)
To determine whether the constraints (20), (21) or the constraints (22), (23) should be enforced, we
use the iterative optimization approach described below.
5
 3.3. Iterative optimization
In a first step, we ensure that all constraints (17) with are deleted, since the post-contingency
flow is unbounded in this case. With for all contingencies , the iterative optimization can
start. The starting point is the OPF problem given by (8) - (15), with (13), (14) replaced by (20), (21).
We run the following iterative optimization:
1. Formulate the optimization with the risk-based constraints (20), (21).
2. Run the OPF
3. Check if any of the active constraints result in post-contingency line flows that are lower than
.
4. If yes: Relax the particular constraint by replacing it with (22) or (23) → move to step 1.
5. If no: The optimal solution is found.
Since the optimization problem is a linear program, a constraint will never need to be changed back if
it has been replaced once. The maximum number of iterations is thus finite and limited by the number
of constraints which can be replaced. In practice, only a few iterations are needed. For small values of
, more iterations are needed.
If is sufficiently small, the initial OPF problem becomes infeasible. In this case, all the risk-based
constraints (20), (21) are replaced by (22), (23), and the OPF is rerun. We then check whether any of
active constraints can be relaxed if (22), (23) is replaced by (20) or (21). If yes, the optimization is run
again, and if no, the optimal solution is found.

3.4. Choosing the risk limit

Including the risk-based constraints (20), (21) in the optimization problem allows us to control the risk
level in the system according to our preferences, but also requires us to define the risk limit in a
reasonable way. We propose to use the solution of a deterministic SCOPF as a benchmark, and define
the risk limit relative to the risk observed from the SCOPF solution. To achieve this, we evaluate the
probability of secondary trip and the corresponding risk of cascade initiation for each
outage and each line based on the SCOPF solution. The risk limit is defined relative to the
maximum risk from the SCOPF as given by (24). Since the outage probabilities can vary by
orders of magnitude, we ensure that the probability limit does not exceed , the maximum
from the SCOPF solution, given by (25). The resulting limit is given by (26) where the tuning
parameters decide how conservative the limit on should be relative to the SCOPF solution.

( ) (24) ( ) (25)

( ) for each contingency (26)

4. Monte Carlo-based cascade simulation

To understand how a cascade would evolve after the cascade initiation (i.e., if a secondary outage
takes place) and what the effect would be in terms of shed load, a Monte Carlo-based risk assessment
including a full cascade simulation is implemented. For each cascade simulation, we assume that we
have an initial outage and then generate the secondary outages randomly based on the secondary
outage probability . The further development of the cascade is determined by the line flows: If a
line flow exceeds the capacity limit, the line is tripped. This process continues until the cascade
terminates or the entire system collapses. If the system breaks up into islands, we assume that all
generators participate in balancing the load/generation mismatch and that they can ramp up/down to
their maximum/minimum output. If there is not enough available generation capacity, part of the load
at each bus is shed. We choose to use a relatively simple cascade simulation for demonstration
purposes, but the simulation could be made more sophisticated by considering, e.g., hidden failures in
the protection system, frequency deviations or operator actions like redispatch [1, 9]. Although these
extensions are not included here, they could be incorporated without any major changes to the method.
6
For each initial outage , we simulate the cascading process times. After each cascade simulation
, the load shed is recorded and a new simulation is started. When cascade
simulations have been run, the expected load shed ̅ and the sample variance is calculated by
(25) and (26).
̅ ∑ (25) ∑ ( ̅ ) (26)
The number of samples needed to guarantee that the actual expected load shed lies in the range
̅ ̅ with a confidence level of 99.8% is given by [13]

. (27)

If , we run another simulations before is calculated again.

When all contingencies have been simulated, the total risk of cascading in the system, measured in
terms of the total expected load shed, is calculated as the sum of the expected load shed per
contingency, weighted by the contingency probability:

] ∑ ̅ ) (28)
(

Below, the Monte Carlo approach is described step by step:

1. Initialization for each contingency , :
1.1 Load initial system data from the OPF solution (topology, load/generation profile).
1.2 Take element out of operation.
1.3 Restore load/generation balance.
1.4 Run DC power flow.
1.5 Calculate for every line .
2. Monte-Carlo simulation for :
2.1 Draw a uniformly distributed random number for each line.
2.2 Check for outages: If , outage line .
Compute the losses for this situation:
2.3 Check for island in the system and restore load/generation balance.
2.4 Record shed load .
2.5 Run DC power flow.
2.6 Remove lines with . Are there new outages? → move to step 2.3.
2.7 Update . If → move to 2.1.
Check if samples is enough to achieve desired accuracy:
2.8 Calculate the mean load shed ̅ and the sample variance .
2.9 Calculate number of samples needed to achieve the desired accuracy.
If , set → move to 2.1.
If , save ̅ .
3. If , set → move to 1.1
4. If , compute total expected load shed ].

5. Case study
We demonstrate the proposed method on the IEEE 118 bus system [14]. The system has
transmission lines and generators. The load level is set to 110% of the base case load, and we
assume that the generator capacity is 125% higher than the values listed in [14]. The line outage
probabilities were calculated based on the line length (estimated from the line reactance) and the
average outage probability for transmission lines in Germany [8]. Due to lack of reliable data for the
generator outage probabilities, we only consider line outages in the calculation of the risk in this paper.
In the RB-SCOPF, we thus consider the optimization problem (8)-(15), with (13) replaced by (20). For
the Monte Carlo-based risk assessment, we compute the total expected load shed based on the
expected load shed for the line outages .
7
 5.1. Comparison of and for the SCOPF and the RB-SCOPF
We first compare and for the standard SCOPF and the RB-SCOPF. For the SCOPF
solution, and are evaluated a-posteriori based on (1) and (4). The risk limits for the
RB-SCOPF are calculated from (26), based on the SCOPF solution. We run the RB-SCOPF with two
different risk limits, computed with parameters and , and evaluate all
and for both solutions. We then compare the values for and obtained
with the SCOPF and the RB-SCOPF.
Fig. 3 shows all non-zero for the SCOPF and the RB-SCOPF with and . The
highest probability of secondary trip in the SCOPF solution is (denoting the
probability of secondary trip of line 37 following an outage of line 8). Comparing the results, we see
that the RB-SCOPF increases in some cases (e.g., for ), and decreases in other
cases (e.g., for the RB-SCOPF with ).
Fig. 4 shows the risk of cascade initiation , i. e., weighted by . Comparing the
with , we see that some of the outages leading to a high have a low risk .
This is because is low. The highest in the SCOPF is . For the
RB-SCOPF with , is the same as for the SCOPF. For the RB-SCOPF with ,
and the highest risk is .

From Fig. 3 and 4, we see that there are very few outages that can lead to secondary trips, i.e., only 16
out of totally 186 possible line outages. Further, we recognize which of the lines that are influenced by
each other. We see that there are several outages that might lead to an outage of line 129 (connecting
bus 82 and 83), helping us to recognize line 129 as a vulnerable point in the system.

159 Pr(l,k) 159 Riskl,k

132
Affected Line (l)

132
Affected Line (l)

-5
131 1 131 x 10
129 129
103 103
78 0.9 78
54 54 3
41 41
37 SCOPF SCOPF
0.8 37
8 33 38 51 97 130 131 132 135 144 145 146147 150 153 158 8 33 38 51 97 130 131132 135144 145 146147 150 153158
Outaged Line (k) Outaged Line (k) 2.5
0.7
159 159
132 132
Affected Line (l)

Affected Line (l)

131 0.6 131

129 129 2
103 103
78 0.5 78
54 54
41 RB-SCOPF  =1.0 41 1.5
RB-SCOPF  =1.0
37 0.4 37
8 33 38 51 97 130 131 132135 144 145 146 147150 153 158 8 33 38 51 97 130131 132135 144145 146 147150 153158
Outaged Line (k) Outaged Line (k)
0.3 1
159 159
132 132
Affected Line (l)
Affected Line (l)

131 0.2 131

129 129
103 103 0.5
78 78
0.1
54 54
41 RB-SCOPF  =0.8 41 RB-SCOPF  =0.8
37 37
0 0
8 33 38 51 97 130131 132 135 144145 146 147 150153 158 8 33 38 51 97 130 131132 135144 145 146147 150 153158
Outaged Line (k) Outaged Line (k)

Figure 3: All outages that lead to a non-zero probability of Figure 4: All outages that lead to a non-zero probability of
secondary trip are shown on the horizontal axis, and secondary trip are shown on the horizontal axis, and
the affected lines are listed on the vertical axis. The color of the affected lines are listed on the vertical axis. The color of
the corresponding dots encodes the value of . Only 16 the corresponding dots encodes the value of . Only
outages that are listed, because all other outages have 16 outages that are listed, because all other outages have
From the top: SCOPF, RB-SCOPF with From the top: SCOPF, RB-SCOPF with
and RB-SCOPF with . and RB-SCOPF with .

8
 5.2. Comparison of the generation cost and the expected load shed for the SCOPF
and the RB-SCOPF
Above, we demonstrated how and can be controlled with the RB-SCOPF. Now, we
investigate how the cost of the generation dispatch is influenced by the choice of the risk limit
parameter , and compare the obtained generation cost with the expected load shed for each of the
solutions. We then analyse the outages that lead to a non-zero load shed in more detail to understand
how the cascade evolves. Again, we use the standard SCOPF as a benchmark, and consider risk limits
given by and for the RB-SCOPF.
The generation cost decreases with increasing , as shown in Fig. 5. This is because a higher value for
leads to higher values of , which leads to a relaxation of (20). For high risk limits, the cost
decrease saturates. In this case, the of all the active constraints are bounded by the max. probability
of secondary trip from the SCOPF, . For (i.e., when the risk limit in the RB-
SCOPF is equal to the maximum risk of the SCOPF), the cost of the RB-SCOPF is lower than the cost
of the SCOPF. Fig. 6 shows the generation cost plotted against the expected load shed of the SCOPF
and each of the RB-SCOPF solutions. In most cases, a higher generation cost (meaning a lower value
of ) leads to a lower load shed. This decrease is however not monotonous. There is one case with an
unexpected high load shed, the case where . To explain this high load shed, we investigate
how the cascade evolves in more detail.
Fig. 7 shows the expected load shed for the line outages with non-zero . For some of the outages,
the cascade stops before it leads to any load shed. For outages of the lines 97 and 144-150, the
expected load shed is relatively similar, and the differences are due to different values for . For
outage 153, the differences in expected load shed are however quite large, and the expected load shed
is high particularly for low values of , e.g., for . When analysing the situation after the outage
of line 153 in more detail, we see that there are two lines with and thus four possible outage
combinations that can occur.
480 480 480 480
 =0.5  =0.5
 =0.6  =0.6
479.5 479.5 479.5 479.5
Generation Cost [$/h]

Generation Cost [$/h]

Generation Cost [$/h]

Generation Cost [$/h]

479 479 479 479

SC OPFSC OPF  =0.7  =0.7
SC OPFSC OPF
478.5 478.5 478.5 478.5

478 478 478 478

 =0.8  =0.8
477.5 477.5 477.5 477.5

477 477 477 477  =0.9  =0.9

 =1-1.6 =1-1.6
476.5 476.5 476.5 476.5
0.6 0.8
0.6 0.8
1 1.2
1 1.4
1.2 1.6
1.4 1.6 1 1.5
1 1.5
2 12 1.5
1 1.5
2 12 1
Risk limit
Risk[]limit [] Expected
Expected
Load Shed
Load[MW 10-3] x 10-3]
Shedx [MW

Figure 5: Generation cost from the RB-SCOPF plotted Figure 6: Generation cost for the RB-SCOPF plotted
against the value of the risk limit parameter .The red star against the expected load shed (evaluated by Monte Carlo
denotes the cost of the SCOPF solution. simulation). The red star denotes the cost and expected load
shed of the SCOPF solution.
-4
x 10
8 RB-SCOPF:
=0.6
E(Ltot)=PrkLk [MW]

6 =0.8
=1.0
4
=1.2
2 =1.4

0 SCOPF
8 33 38 51 97 130 131 132 135 144 145 146 147 150 153 158
Outaged Line No. [-]

Figure 7: Expected load shed per contingency for the SCOPF and the RB-SCOPF with different values of .

9
The four combinations and their corresponding load shed are listed in Table I. Notably, case II with
secondary trip of line 129 leads to 37.4 MW load shed, while case IV with a secondary trip of both
129 and 159 leads to 0 MW of shed load. Assuming that the secondary outage of line 129 and line 159
are independent events, the probability of case II is given by ( ).
A high value for decreases , and thus decreases the expected load shed of the
initial outage 153. From Fig. 3, we see that when . The high expected load
shed is due to a high probability of case II.
This analysis illustrates that it is not possible to control the expected amount of load shed only by
controlling the risk of cascade initiation. A lower risk limit might increase the expected load shed in
some cases, in particular when there are several possible secondary outages involved.
Table 1 Possible developments of the cascade after the outage of line 153.
Case Secondary outage Development of cascade Load shed
I - No further outages 0 MW
II 129 Cascading outage, system breaks into islands 37.4 MW
III 159 No further outages 0 MW
IV 129, 159 Loss of large generator at bus 99, no further outages 0 MW

6. Conclusion
This paper proposes a new, risk-based security measure, which aims at reflecting the risk of cascading
events in a risk-based optimal power flow (OPF) formulation. The risk-based security criterion is
based on the probability of cascade initiation, i.e., the risk that one initial N-1 contingency can lead to
secondary outages. The probability of secondary trips was modelled as a function of both the post-
contingency line flow and the change in line flow from the pre- to the post-contingency state. Based
on the proposed criterion, the RB-SCOPF was formulated, which allows us to control the risk of
cascade initiation through the use of risk-based N-1 constraints. Further, the risk-based security
criterion was implemented in a Monte Carlo-based simulation framework, which allows us to compute
the expected load shed for a given generation dispatch.
The RB-SCOPF and the Monte Carlo based assessment is applied to a case study for the IEEE 118 bus
system, and compared with the outcome of a standard SCOPF. We show that the RB-SCOPF can
effectively reduce the risk of cascade initiation, but that the reduction in risk comes at an increased
cost. Based on the Monte Carlo simulation, we see that we in most cases are able to reduce the
expected load shed by reducing the risk limit in the RB-SCOPF. However, dependent on how the
cascade actually evolves, a lower risk limit might in some cases increase the total expected load shed,
illustrating the fact that it is not possible to control the expected amount of load shed by only
controlling the risk of cascade initiation.
ACKNOWLEDGEMENT
This research work described in this paper has been carried out within the scope of the project ”Innovative tools
for future coordinated and stable operation of the pan-European electricity transmission system (UMBRELLA)”,
supported under the 7th Framework Programme of the European Union, grant agreement 282775.

BIBLIOGRAPHY
[1] D. Kirschen and D. Jayaweera, “Comparison of risk-based and deterministic security
assessments”, IET Generation, Transmission & Distribution, vol. 1, no. 4, pp. 527–533, 2007.
[2] K. Uhlen, G. H. Kjølle, G. G. Løvås and Ø. Breidablik, “A Probabilistic Security Criterion for
Determination of Power Transfer Limits in a Deregulated Environment”, 2000 Cigre Session,
Paris, France, 2000.
[3] F. Xiao and J. McCalley, “Power System Risk Assessment and Control in a Multiobjective
Framework”, IEEE Transactions on Power Systems, vol. 24, no. 1, pp. 78–85, 2009.
[4] W. Fu and J. D. McCalley, “Risk-based optimal power flow,” IEEE PowerTech Conference,
Porto, Portugal, 2001.

10
 [5] G. Hug, “Generation cost and system risk trade-off with corrective power flow control”, 50th
Annual Allerton Conference on Communication, Control and Computing, Allerton, USA, 2012.
[6] M. Zima and G. Andersson, “On security criteria in power systems operation,” IEEE PES
General Meeting, San Francisco, USA, 2005.
[7] ENTSO-E, "Continental Europe Operation Handbook," European Network of Transmission
System Operators for Electricity, 2009.
[8] M. Obergünner et al, “Using the VDN Statistic on Incidents to Derive Component Reliability
Data for Probabilistic Reliability Analyses”, Verband der Netzbetreiber – VDN – e.V. beim
VDEW, Berlin, 2004.
[9] O. Alizadeh Mousavi, R. Cherkaoui, and M. Bozorg, “Blackouts risk evaluation by Monte Carlo
Simulation regarding cascading outages and system frequency deviation,” Electric Power
Systems Research, vol. 89, no. 0, pp. 157–164, 2012.
[10] Z. Jianfeng and F. Alvarado, “A heuristic model of cascading line trips,” Probabilistic Methods
Applied to Power Systems (PMAPS), Ames, Iowa, 2004.
[11] W. Ng, “Generalized generation distribution factors for power system security evaluations,” IEEE
Transactions on Power Apparatus and Systems, vol. PAS-100, no. 3, pp. 1001 –1005, 1981.
[12] R. Christie, B. F. Wollenberg, and I. Wangensteen, “Transmission management in the
deregulated environment,” Proceedings of the IEEE, Vol. 88, No. 2, pp. 170–195, 2000.
[13] C. Graham and D. Talay, “Stochastic Simulation and Monte Carlo Methods. Mathematical
Foundations of Stochastic Simulation ”, Springer Verlag, Berlin, 2013, p.41-47.
[14] Illinois Institute of Technology, ”IEEE 118-bus, 54-unit, 24-hour system: Unit and Network
Data”, Available: http://motor.ece.iit.edu/data/JEAS_IEEE118.doc, Date: 23.08.2013.

11

View publication stats