You are on page 1of 10

1808

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 27, NO. 4, NOVEMBER 2012

An Optimal Power Flow Algorithm to Achieve


Robust Operation Considering Load and Renewable
Generation Uncertainties
Han Yu, Member, IEEE, and W. D. Rosehart, Senior Member, IEEE

AbstractConsideration of uncertain injections in optimal


power flow (OPF) calculation is increasingly important because
more renewable generators, whose outputs are variable and
intermittent, are connected into modern power systems. Since
it is often difficult to predict the variations of both load and
renewable generator output accurately, this paper proposes an
OPF algorithm to make optimized results not only have a high
probability to achieve minimized generation cost, but also robust
to the uncertain operating states. In this paper, the objective of the
OPF is to minimize the generation cost of the scenario which has
the largest probability to appear in the future. In order to make
the OPF result be able to accommodate other possible scenarios,
the OPF constraints are modified. Considering the probabilistic
distributions of both load and renewable energy output, the
modified constraints are derived from Taguchis orthogonal array
testing and probabilistic power flow calculation. The effectiveness
of the proposed OPF method is demonstrated by the cases up to
the system with 2736 buses.
Index TermsOptimal power flow, probabilistic power flow, robust, Taguchis orthogonal array testing, uncertainties.

I. INTRODUCTION

PTIMAL power flow (OPF) is a powerful analyzing tool


in power system economic operation [1], [2]. The objective of OPF is to achieve a minimized power generation cost
with consideration of some specific power system constraints.
However, due to the existence of uncertain factors in power systems, such as load variation and intermittent output of renewable energy, the results of the traditional deterministic OPF are
not ensured to be the optimal control strategy for future power
system operating conditions. Therefore, considering uncertainties in the OPF formulation is of importance.
The most common way of considering uncertain factors in
OPF is probabilistic optimal power flow (P-OPF) [3]. In P-OPF,
both the considered uncertain factors and the final results are
modeled as probabilistic distributions. The computing methods
of P-OPF include Monte Carlo simulation [4], cumulant method
[5], [6], point estimate [7], first-order second-moment [8], etc.
Apart from P-OPF, another way of considering uncertain factors in OPF is fuzzy OPF [9], in which fuzzy logic is employed

Manuscript received May 12, 2011; revised October 06, 2011 and February
01, 2012; accepted March 30, 2012. Date of publication May 15, 2012; date of
current version October 17, 2012. Paper no. TPWRS-00438-2011.
The authors are with the University of Calgary, Calgary, AB T2N 1N4,
Canada (e-mail: ahan.yu@gmail.com; rosehart@ucalgary.ca).
Digital Object Identifier 10.1109/TPWRS.2012.2194517

in the modeling and problem solving. Compared with deterministic OPF, the computational burden of P-OPF and fuzzy OPF
methods are heavier. In the above works, the uncertain factors
considered are limited to load forecast uncertainty.
Currently, power system operators and researchers have recognized that it is important to consider uncertainties in power
system optimization to make the controlling strategy robust to
the uncertain variations. In many research works, the uncertainties of load and renewable energy are represented by selected
scenarios, which achieved by various scenario reduction techniques. Based on the well-known backward and forward scenario reduction method [10], [11], a series of specific scenario
reduction technique is applied in stochastic unit commitment
[12], power market trading [13][16]. Other methods, such as
particle swarm optimization [17] and interval linear programming [18] are also employed in scenario reduction. The worstcase selection [19] also can be regarded as a scenario reduction method. Apart from the power injection uncertainties, the
contingency of the power system devices can be recognized as
another kind of uncertainty. In order to achieve a controlling
strategy which adapt to the contingency cases, security constrained OPF (SCOPF) is developed as an important extension
of OPF since Alsac and Stott published their work [20] in the
1970s. In recent years, SCOPF algorithm with various emphasizes have been proposed, which include contingency filtering
technique [21], [22], SCOPF with consideration of FACTS devices [23], and tradeoff between SCOPF and real-time operation
[24], etc. Specifically, some researchers propose to achieve robust operation in other promising prospectives, such as risk-limiting dispatch [25] and day-ahead planning [26].
In this paper, considered uncertainties are focused on the load
and the output of the renewable generators. In power systems,
the output of the non-renewable generation is controllable,
while the load and the output of the renewable generation is
often uncontrollable. Based on the theory of robust design, the
controllable parameters can be adjusted to make the system
robust to the uncertain variations [27], so that it is inferred that
the controllable non-renewable generations can be employed
to make the power system robust to the uncertainties of load
and renewable energies. Therefore, this paper presents an OPF
method that aims to achieve an optimal generation schedule
which is not only cost-saving but also makes the power system
robust to the uncertain operating conditions caused by load and
renewable energy. This method is different from some former
works that incorporate the models of the uncertain factors into
OPF formulation directly, as this paper proposes a deterministic equivalent OPF formulation to consider the uncertainties.

0885-8950/$31.00 2012 IEEE

YU AND ROSEHART: OPTIMAL POWER FLOW ALGORITHM TO ACHIEVE ROBUST OPERATION

The OPF is modeled based on the most likely scenario, in


which the uncertain factors are assigned to the values which
have the largest probability to appear in the random factors
probabilistic distributions. Next, for the purpose of making the
OPF solution able to satisfy the operating constraints of other
possible operating scenarios, the constraints in the traditional
OPF formulation are modified with robust constraints, which
are deduced from probabilistic power flow (P-PF) formulation
and Taguchis orthogonal array testing (TOAT) [28].
TOAT is a method to select a minimum number of testing scenarios with good statistical information in the uncertain space. It
has been proven that TOAT is able to select optimal representative testing scenarios from the possible combinations in additive
and quadratic models [29]. Compared with Monte Carlo simulation, the number of testing scenarios of TOAT are much less,
therefore, the computational burden is alleviated. Additionally,
achieving scenarios with TOAT is much more simple than above
scenario reduction methods. In the power system field, TOAT
has been successfully applied in transmission network expansion planning [30].
This paper adopts DC OPF to show the method. The remainder of this paper is organized as follows: The theoretical
basis, which includes the probabilistic DC power flow calculation, the traditional OPF formulation, and TOAT method are
given in Section II. In Section III, the proposed robust OPF
method is explained. Next, in Section IV, three examples are
studied to prove the effectiveness of the proposed method. The
conclusions are given in the last section.

1809

where
is the array of
is the array of
is a matrix
whose non-diagonal elements are
, and diagonal
elements are
. In (5), the row and column
of the slack bus are deleted. Equation (5) can be further transformed into
(6)
Equation (4) is also written to be into matrix form as follows:
(7)
where
is the array of
and is the coefficient matrix.
Finally, the relation between the branch power flow and nodal
injected power is obtained by substituting (6) to (7):
(8)
Equations (6) and (8) are deterministic power flow formulations. These equations are changed into P-PF equations if
and
, which are arrays of deterministic variables, are replaced with arrays of random variables
and
, where
is a superscript to denote random variables. Thus the P-PF (9)
is achieved from (8). Using the mathematical convolution technique, the distributions of the active power
can be calculated
[32]:
(9)

II. THEORETICAL BASIS


A. Probabilistic DC Power Flow Calculation [31]

B. Traditional DC OPF Formulation

Power flow calculation studies the steady operating states of


the power system. The DC power flow is a linear model that is
approximated from AC active power flow formulations (1) and
(2):

For a power system with renewable energy generations, the


traditional DC OPF formulation for a specified operating scenario is as follows.

(1)

(10a)

(2)
where
is the active power injected to bus (except the slack
bus); and
is the voltage magnitude of bus and , respectively; is the number of nodes in the system;
and
are
the conductance and susceptance values between node and ,
respectively;
is the voltage angle difference between bus
and . Assume that the resistance is zero,
,
and
, (1) and (2) can be simplified to be (3) and (4)
respectively:
(3)
(4)
is the reactance between node and . Equation (5)
where
is achieved by writing (3) into matrix form:
(5)

(10b)
(10c)
(10d)
(10e)
where
is the number of non-renewable generators; , ,
and
are the cost coefficients of the th non-renewable generators power output
;
,
, and
are the arrays
of non-renewable generation power outputs, load values, and
renewable generation power outputs, respectively;
is the
angle of the slack bus;
is the capacity of line ;
and
are the arrays of the minimum and maximum constraints for
, respectively. The objective of this formulation is to minimize the total non-renewable generation cost of the system. The
equality constraint (10b) is the nodal active power injections;
the inequality constraint (10d) refers to the power flow limits.

1810

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 27, NO. 4, NOVEMBER 2012

C. Taguchis Orthogonal Array Testing (TOAT)


that is depicted by
, where
, are
controllable factors and
are
uncontrollable
uncertain factors. For the purpose of making
robust (less
sensitive) to the random variations of
, controllable
factors
are optimized. In this process, the uncertain
variations of
are represented by a series of
scenarios. Since it is impractical to consider all the possible
scenarios, only some representative scenarios are selected to
guide the optimization.
For each uncertain variable
, a total of
representative levels are selected. Hence, the full operating
states of
would be
combinations, which is still
computationally expensive when
is large. In order to further
reduce the number of testing, TOAT is employed to determine
the scenarios.
In TOAT, scenarios are decided by orthogonal arrays (OAs).
An OA is a matrix that represented by
, where and
are the number of rows and columns, respectively, and is
the number of the matrix element levels. As an example, an OA
is shown as follows:
Given

TABLE I
GENERATE SCENARIOS FOR SYSTEM
ON ORTHOGONAL ARRAY

system

For a given problem , the appropriate OA is determined from


OA libraries [33], [34] according to the following two considerations.
1) The appropriate OA level number : When forming
scenarios according to the selected OA, the number
of element levels
indicates the representative levels of random factors. Taguchi suggests that
employing OAs with different levels to make an appropriate coverage of the random space, i.e., determining
the number of the testing levels of each random factor
according to the feature of
[27]. If uncertain factor
has a linear effect on , then
should have
two testing levels. If
is symmetrically distributed,
and
should be
chosen, where
and
are the functions
to calculate the mean and standard deviation of
,
respectively. If
has a quadratic effect on , then
should have three testing levels. If
is symmetrically
distributed, then
,
and
should be chosen.
2) The number of OA columns : When forming the scenarios according to OA, the uncertain variables correspond to the OA columns. If there is not an existing OA
in the OA libraries whose number of columns
exactly equal to the number of random variables in the
problem, an OA
whose number of columns
is greater than the number of random variables
should be chosen. Then, a new OA is formed for TOAT
by ignoring the redundant columns in
because the resulting array is still an OA if some columns

and

BASED

denote the two selected levels of

in
are ignored [35]. In the following paper,
for simplification,
is used to represent the OA
whose number of columns is equal to the number of
random factors in the problem.
After selecting an appropriate OA based on the above two criteria, the scenarios correspond to the rows of the selected OA.
For system , determined by
, a total of scenarios
are formed, and
is much smaller than
[33]. For example, assume that there are three uncontrollable random variables
and
in system , and two levels are selected in
each random variable for testing. Determined by the number of
variables and the number of variable levels, OA
is selected to form the testing scenarios. The way of forming four
testing scenarios according to
is shown in Table I. In this
case, a total of four testing scenarios are formed, which is less
than the number of full combinations . Therefore, the number
of testing is minimized.
The following features of an OA ensure that TOAT achieves
representative testing scenarios which are uniformly distributed
over the uncertain operating space [35], [36].
1) In each OA column, every level occurs
times. For
example, in
, 1 and 2 occur
times.
2) In any two columns, the level combinations appear the
same number of times. In
, 1 1, 1 2, 2 1,
and 2 2 occur once in any two columns.
III. OPF ALGORITHM TO ACHIEVE ROBUST
OPERATION CONSIDERING LOAD AND RENEWABLE
GENERATION UNCERTAINTIES
In the proposed OPF, the traditional non-renewable generations are regarded as controllable factors in the power system,
and the loads and renewable generations are considered to be
uncontrollable. Determined by the uncertainties of load and renewable energy output, the future operating states are uncertain.
In these large number of possible operating scenarios, the scenario with the largest probability to appear in the future is high
significant. Therefore, the objective of the proposed OPF is designed as minimizing the generation cost of the scenario which
has the largest probability to appear in the future. In order to
avoid the power flow becoming infeasible if other possible scenarios happen, TOAT and the probabilistic power flow formulation are employed to derive the modified OPF constraints. In this
study, the loads and renewable energies are modeled to be independent; and the loads are modeled to be normally distributed

YU AND ROSEHART: OPTIMAL POWER FLOW ALGORITHM TO ACHIEVE ROBUST OPERATION

[37]. The power output variation ranges of the intermittent renewable energies are between zero and the capacity. The details
of the proposed OPF are presented in this section.
A. Probabilistic OPF Constraints Considering Load and
Renewable Generation Uncertainties
In order to analyze the relation between the distribution of
the outputs and the inputs which include controllable generation
output, for each bus, the injected power sources are modeled as
follows:

1811

,
, and
are the submatrices of formed by
where
the columns that corresponding to the non-zero elements in
,
, and
; , , and
are the number of the transmission
lines, the number of loads, and the number of renewable generation, respectively;
is the expansion form of
;
is the array of nonzero elements in
;
and
is the array of nonzero elements in
.
According to (15), the constraint of the th line
is as (16):

(11)
Substituting (11) to (9), the equation for power flow distributions (12) is achieved:
(12)
For power system operation safety, it must be ensured that no
overload happens in the transmission branches, i.e., the power
flow distribution should within the capacity limits of the transmission lines as (13):
(13)
Substituting (12) to (13), the probabilistic constraints for the
controllable generations are achieved as follows:
(14)
Since both the loads and renewable energy outputs are uncertain variables, the above constraints are probabilistic. In order
to simplify these constraints, in the following section, (14) is
transformed into deterministic constraints by P-PF calculation
and TOAT.
B. Transform Probabilistic Constraints (14) to Deterministic
Constraints by TOAT
In (14),
,
, and
are sparse arrays whose lengths are
. The elements in
,
, and
are zeros if there are
no non-renewable generation, load, and renewable generation
connected to the corresponding buses, respectively. By deleting
the zero elements in
,
, and
, inequation (14) can be
written into (15):

..
.

..
.

(16)
,
, and
in which
are the th row elements in
,
, and
, respectively. The
steps of transforming the probabilistic constraints (16) into deterministic constraints with TOAT are as follows.
1) Aggregation of the deterministic and normal distributed
variables in (16).
Reducing the number of random variables is helpful
to alleviate the computational burden because more
random variables generally lead to more testing scenarios in TOAT. Here, the normal distributed random
variables and deterministic variables in (16) are aggregated using the following rules [38]: a) If a random
variable
is normally distributed with mean
and
variance , a linear transform
is still
a normal distribution with mean
and variance
. b) If
is also a normal uncertain variable
with mean
and variance , a linear aggregation of
is also a normal distribution with
mean
and variance
. Therefore,
a linear combination of the transmission line capacity
and load distributions in (16) can be calculated. Given
(17)
and
(18)
then the resulted
and
are also normal distributions, even if the load distributions are not identical.
After this step, inequation (16) is transformed into (19)

..
.
(19)

..
.

..
.

..
.

..
.

(15)

This transformation effectively shifts the random variable from the loads to a representation of the power
flow on transmission lines associated with the loads. The
number of new random variables is equal to 1, instead
of the number of transmission lines because (19) models
only a single transmission line for a particular value of
. Therefore, the
random variables in (16) are
reduced to
random variables in (19). Since this

1812

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 27, NO. 4, NOVEMBER 2012

step is a strict mathematical process, there is no difference between employing (16) and (19) in the followed
OPF solving. In the followed testing scenario selecting
process, each transmission line is considered effectively
separately with respect to flow limits and load/generation scenarios. In addition to the mathematical equivalency this is what allows for the process utilized in this
section.
2) Select the representative testing scenarios of (19) by
TOAT.
In (19), every random variable has a linear effect on
this inequation. Therefore, based on the statements in
Section II-C, two representative values are selected to
indicate each random variable. Accordingly, two-level
OAs are employed to guide TOAT. For normal distributed
and
,
and
are adopted as representative values, where indicates
or
. Since the outputs of the renewable energy
are always varied between zero and the capacity
,
, zero and capacity values are adopted
as representative values of the renewable generation
outputs. Then the scenarios are generated according to
the following steps.
i) Choose a two-level OA
which fulfills
from OA libraries.
ii) The representative values for random variables in
(19) are assigned as follows:
(20a)

After this step, the probabilistic constraint (19) are


changed into deterministic constraints (21), shown
at the bottom of the page, in which
,
and
are the values
,
, and
, respectively in
of
the th testing scenario determined by TOAT.
3) Determine the final deterministic upper and lower limits.
Inequation (21) can be simplified as follows. Since the
upper limits are
less than
a series of data, it is equivalent to
less than the minimum one. Similarly, the
are equivalower limits of
lent to more than the maximum value of the constraints.
Therefore, inequation (21) is simplified to (22)

..
.

(22)

..
.
Let

(20b)
(20c)
(20d)

..
.

(20e)
(20f)
In the lower limit of (19), random variables
and
are made corresponding to the
first
columns in
. Similarly, in
the upper limit of (19), random variables
and
are made corresponding to the first
columns in
. As indicated in
Table I, the representative values assigned by (20a)
to (20f) are substituted into
to generate
testing scenarios.

(23)
and

..
.
(24)
then inequation (22) is transformed into (25):
(25)

(21)

YU AND ROSEHART: OPTIMAL POWER FLOW ALGORITHM TO ACHIEVE ROBUST OPERATION

Considering all the lines of


, inequation (25)
can be rewritten into the following matrix form:

1813

TABLE II
COMPARISON OF THE NUMBER OF CONSTRAINTS AND THE NUMBER OF
CONSIDERED SCENARIOS BETWEEN TRADITIONAL AND ROBUST OPF

(26)
where

and
are the arrays of
and
, respectively. Inequations (25) and (26) are deterministic constraints. That is to say, after the above three steps,
the probabilistic constraints (16) is approximated into deterministic constraints (25), i.e., (14) and (15) are approximated
to (26).
It should be pointed out that in Step (2.b), the method of
forming testing scenarios from an OA is not single but multiple
because 1 and 2 in an OA column may, respectively, signify
the low and high representative values of a random variable, or
vice versa, i.e., each variable has two corresponding ways to get
testing scenarios. For
random variables in OPF problem,
the number of corresponding methods to get testing scenarios is
.
For very large power systems, if there is no available OA
table, Monte Carlo simulation can be employed to simulate the
possible scenarios. Then the probabilistic constraints (14) can
be transformed to deterministic constraints (26) accordingly.
C. Robust OPF Formulation
The proposed robust OPF formulation is shown in (27):

(27a)

(27b)
(27c)
(27d)
(27e)
is the array of load values with the largest
where
probability, and
is the array of the renewable energy
outputs with the largest probability. For normal distributed
loads, the values of the largest probability are the mean values.
The accurate
values should be achieved from a
forecast. For intermittent renewable energies, if accurate probabilistic models are unavailable, the intermediate values of the
power output variation ranges can be adopted as
.
The above OPF model is a deterministic OPF formulation
which can be solved easily. In this paper, interior point method
is adopted [39].
For the purpose of comparisons, the number of the constraints
and the considered scenarios of both traditional and robust OPF
is shown in Table II. It can be observed that the problem sizes
of the two OPF formulations are similar, but more scenarios are
considered in the robust OPF.
If
and
in the traditional OPF formulation (10) are assigned to be
and
, respectively, the only
difference between the traditional OPF (10) and the robust OPF
(27) is the power flow constraints (10d) and (27d). According

TNoC denotes total number of the constraints, and NoCS denotes the
number of the considered scenarios.

to the steps in Section III-B, it can be inferred that the solution


space of (27d) is the common solution spaces of (10d) for different scenarios. Therefore, the solution space of robust OPF
(27) is the common solution space of the series of traditional
OPF (10) based on different scenarios with
and
. The robust OPF (27) is feasible if the
solution space is not empty. However, if the probabilistic constraints (14) is restrictive, i.e., the considered variation ranges of
uncertain variables are large, it is possible that the scenarios generated from (14) make the above-mentioned traditional OPFs
common solution space empty, i.e., the robust OPF (27) is infeasible. This implies that there is no rescheduling that can be
used to find a feasible solution in this operating mode. It is important to understand if the variation causing the infeasible solution is realistic, in which case this implies that the transmission
system is not adequate for the uncertain variations. However, if
the infeasible solution is due to unrealistic models of variations,
which may be caused by long-term forecasts, then utilization of
more appropriate forecasts is required. One approach to regain
a feasible solution to (27) is to reduce the variation considered,
and the drawback to this approach is the robustness of the solution will be reduced. If some variations are reduced, it is important to ensure such reductions are based on an understanding of
the system and reasonable models of the uncertainty. When infeasible cases are encountered, then these particular cases need
to be studied from both an operational perspective, as in this
paper, but also a planning perspective to ensure the system is
reinforced for potentially realistic operating conditions.
D. Confirmatory Experiment
After robust OPF formulation (27) is solved, the proposed
OPF results are tested to verify feasibility with uncertain loads
and renewable energy outputs. In this confirmatory experiment,
the non-renewable generators output is assigned to be the optimized results; meanwhile, a series of deterministic testing scenarios are generated by Monte Carlo simulation based on the
distributions of load and renewable energy. For each scenario,
a deterministic power flow calculation is performed. If no overload happens, the testing scenario is defined as feasible, otherwise, it is defined to be infeasible. The feasibility of the optimized schedules is defined as follows:
(28)
where is the total number of testing, and
the feasible scenarios.

is the number of

1814

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 27, NO. 4, NOVEMBER 2012

TABLE III
GENERATOR PARAMETERS OF IEEE 14-BUS SYSTEM

TABLE IV
TRANSMISSION LINE CAPACITIES OF IEEE 14-BUS SYSTEM

E. Comments on Employing TOAT in AC OPF


Theoretically, TOAT is applicable for both linear and nonlinear problems. However, there are some difficulties in dealing
with nonlinear large-scale AC OPF: different from DC model,
the uncertain active and reactive power injections of the load
demands in nonlinear AC model cannot be aggregated as the
way in Section III-B. For large-scale AC OPF with large number
of random variables, if TOAT is employed for generating scenarios, according to Section II-C, a large size orthogonal array
will lead to large number of proxy constraints. Those constraints
cannot be aggregated and will cause heavy computational cost
in an AC OPF.
IV. EXAMPLE STUDIES
The proposed robust OPF formulation is tested using the
IEEE 14-bus, IEEE 118-bus, and a 2736-bus system. In these
three systems, the mean of the load distribution are assumed to
be the values of the base case, and the load standard deviations
are assumed to be 5% of the mean. The renewable energy is
assumed to be wind power. The distribution of wind speed is
modeled as Weibull, and the scale parameter and shape parameter are assigned to be 11.0086 and 1.9622 m/s, respectively.
The cut-in speed , cut-out speed
, and rated wind speeds
of wind power generators are assigned to be 4, 25, and
13.61 m/s [40]. After the wind speeds are simulated, the wind
power output is calculated as follows:

otherwise

(29)
where
is the wind power rated capacity. In this paper,
the number of the testing scenarios of Monte Carlo simulation
is
. Under the above assumptions, the values of
are the wind farm capacity values.
The programs are developed with Matlab R2009a on an iMac
computer with Intel Core i7 processor and 8 GB RAM. The
proposed robust OPF program is developed based on DC OPF
program in MATPOWER [39].
A. IEEE 14-Bus System
The IEEE 14-bus system consists of 14 buses, 5 generations,
and 20 lines. The network data of this system can be found in
[41]. The generation power output limit and cost curve data are
shown in Table III. The transmission line capacities are assigned
as Table IV. It is assumed that two wind farms are connected

TABLE V
TEST RESULTS OF IEEE 14-BUS SYSTEM

to bus-2 and bus-3, respectively, and the capacity of each wind


farm is 40 MW.
In the robust OPF calculation, the 11 uncertain loads are aggregated to be one uncertain variable according to (17) and (18).
Adding the two renewable generations, the number of uncertain
variables to be dealt with TOAT in this system is three. Therefore, a two-level OA
is selected to generate the testing
scenarios. Then the robust constraints (27d) are determined according to these testing scenarios.
As discussed in Section III-B, there are
corresponding
ways to generate the testing scenarios according to
. In
order to check the efficiency of the proposed algorithm, all the
corresponding ways are tested and both the minimum and
maximum costs and degrees of robust are given in Table V.
Additionally, for the purpose of comparisons, a traditional deterministic OPF is also calculated, in which wind power outputs
and load values are assigned to the same values as the robust
OPF, i.e., wind power capacities and the mean values of the distribution, respectively. The cost and degree of feasibility of traditional deterministic OPF are also shown in Table V.
B. IEEE 118-Bus System
The network data of the IEEE 118-bus system can be found
in [41]. The OPF data are same as OPF settings of IEEE 118-bus
system in Matpower [39]. Also the transmission line capacities are assigned as Table VI. Five wind farms are connected
to bus-16, bus-37, bus-48, bus-75, and bus-83 separately. Also
the capacity of each wind farm is 200 MW.
After the load is aggregated to one random variable according
to (17) and (18), the number of random variables to be treated
with TOAT is reduced to
. Therefore, a two level
OA
is selected to generate the operating scenarios as
follows:

YU AND ROSEHART: OPTIMAL POWER FLOW ALGORITHM TO ACHIEVE ROBUST OPERATION

TABLE VI
TRANSMISSION LINE CAPACITIES OF IEEE 118-BUS SYSTEM

TABLE VII
TEST RESULTS OF IEEE 118-BUS SYSTEM

TABLE VIII
TEST RESULTS OF 2736-BUS SYSTEM

The robust constraints of the proposed OPF is determined according to these testing scenarios. The test results of both robust
OPF and traditional deterministic OPF in which wind power
outputs and load are assigned to the capacity values and mean
values are shown in Table VII.
C. 2736-Bus System
The Polish 2736-bus testing case is obtained from MATPOWER. Four wind farms are connected to bus-16, bus-157,
bus-975, and bus-1675 separately. The capacity of each wind
farm is 200 MW. OA
is selected to generate the operating scenarios. The testing results are shown in Table VIII.
D. Results Analysis
The simulation results in Tables V, VII, and VIII show that although the traditional deterministic OPF achieves the minimum
costs, for systems with large renewable generation capacities,

1815

the low feasibilities imply that traditional deterministic OPF


cannot adapt to the uncertain operating conditions. This manifests the significance of considering the uncertainties of both
load and renewable energy output in OPF problem.
In the proposed robust OPF, the uncertainties of the load and
renewable generations are discretized based on the probabilistic
distributions. By directly considering the uncertainties in the
OPF solution, the probability of actually encountering infeasibility in operation would be reduced. Although the costs of robust OPF are higher than the cost of the deterministic OPF, it is
obvious that the high degree of feasibility makes the proposed
OPF more practical to guide the power system operation. It can
be summarized that the multiple scenarios in robust OPF have
a better coverage of the operating area than the single scenario
in traditional OPF, which guarantees the robustness; meanwhile
the expense of achieving the robustness of an OPF is that the
generation cost is increased.
Generally, in the traditional OPF calculation, in order to
achieve the minimal generation cost, the lower the cost a
generator has, the more output it is dispatched. However,
the generators with relatively low generation costs do not
always operate at their full capacity because of power flow
constraints. Therefore, it can be inferred that, in traditional
OPF, the generations are adjusted to a state that the power
flows on some transmission lines are as large as the capacity
values. As for the transmission system, the deterministic OPF
result can be regarded as an extreme operating state. If this
generation schedule obtained from deterministic OPF is applied in the cases that the load and renewable generations are
uncertain, the variations of the load and renewable energy
make the transmission system have a large probability to be
overload. In contrast, the variations of the load and renewable
generations are considered in the transmission constraints of
robust OPF, which means that some transmission capability
is reserved for the variations. So that compared with traditional OPF, the generation cost in robust OPF is increased
because generators are dispatched in a different way of the
traditional OPF.
For the IEEE 14-bus system, the computing time is too short
to compare. And the time consuming of IEEE 118-bus and
2736-bus system is shown in Tables VII and VIII, respectively.
It can be seen that the computing time of robust OPF is longer
than traditional OPF. This happens because of the different
sparsities of (10d) and (27d) when they are represented in matrix form. Constraint (10d) has a very good sparsity while
in (27d) is a relatively full matrix. Therefore the computational
burden of robust OPF is heavier than the traditional one in
each iteration. Additionally, it was observed that the robust
OPF often needed more iterations than the traditional OPF to
achieve convergence.
It can also observed from the results that, in the process of
generating testing scenarios, different corresponding ways results in different costs and feasibilities of the proposed OPF.
This happens because TOAT does not provide a full coverage
of the entire random operating space. Although this difference
is unwelcome, it is an inevitable feature of scenario generation techniques because the limited number of scenarios always

1816

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 27, NO. 4, NOVEMBER 2012

cannot cover the whole uncertain space. Therefore, the results


are dependent on the considered scenarios.
V. CONCLUSION
This paper proposes an OPF model that takes into account
the uncertainties of both load and renewable energy. Based on
the idea that optimizing the control parameters is able to make
the system robust to variations of the uncertain factors, this
paper suggests to optimize the outputs of the controllable generations not only to save the generation cost but also to make
power system robust to the uncertain load and renewable energy
output. Firstly, the objective of the proposed OPF is changed
into achieving the minimum cost for the most possible scenario,
in which the uncertain factors are assigned to the values which
have the largest probability to appear in the random factors
probabilistic distributions. Secondly, for the purpose of making
the OPF solution able to satisfy the operating constraints of
other possible operating scenarios, the constraints in the traditional OPF formulation are modified with robust constraints,
which are deduced from P-PF formulation and TOAT. Therefore, the probabilistic OPF problem is transformed into a deterministic OPF problem. The effectiveness and robustness of the
proposed robust OPF method is applied to IEEE 14-bus, IEEE
118-bus, and Polish 2736-bus system. The studied results have
confirmed the effectiveness of the proposed OPF algorithm.
It should be noted that the proposed robust constraints is not
only limited in OPF. It can also be applied in other OPF-based
problems, such as security constraint OPF, dynamic OPF, OPF
in market environment, optimal dispatch, and power system
planning, to achieve robust solutions. Moreover, the method
presented in the paper can be used for an
analysis without
additional difficulty. This can be incorporated as an additional
uncertainty or using a scenario based approach. However, since
this is not the focus of the paper, we have not included further
simulations.
REFERENCES
[1] J. Carpentier, Contribution a letude du dispatching economique, Bulletin de la Societe Francaise des Electriciens, vol. 3, pp. 431447, Aug.
1962.
[2] H. W. Dommel and W. F. Tinney, Optimal power flow solutions,
IEEE Trans. Power App. Syst., vol. PAS-87, pp. 18661876, Oct. 1965.
[3] M. Madrigal, K. Ponnambalam, and V. H. Quintana, Probabilistic optimal power flow, in Proc. 1998 IEEE Can. Conf. Electrical and Computer Engineering, Waterloo, ON, Canada, May 1998, pp. 385388.
[4] H. Zhang and P. Li, Probabilistic analysis for optimal power flow
under uncertainty, IET Gen., Transm., Distrib., vol. 4, no. 5, pp.
553561, 2010.
[5] A. Schellenberg, W. Rosehart, and J. Aguado, Cumulant-based probabilistic optimal power flow (P-OPF) with Gaussian and gamma distributions, IEEE Trans. Power Syst., vol. 20, no. 2, pp. 773781, May
2005.
[6] A. Tamtum, A. Schellenberg, and W. Rosehart, Enhancements to the
cumulant method for probabilistic optimal power flow studies, IEEE
Trans. Power Syst., vol. 24, no. 4, pp. 17391746, Nov. 2009.
[7] G. Verbi and C. A. Caizares, Probabilistic optimal power flow in
electricity markets based on a two-point estimate method, IEEE Trans.
Power Syst., vol. 21, no. 4, pp. 18831893, Nov. 2006.
[8] X. Li, Y. Li, and S. Zhang, Analysis of probabilistic optimal power
flow taking account of the variation of load power, IEEE Trans. Power
Syst., vol. 23, no. 3, pp. 992999, Aug. 2008.
[9] V. Miranda and J. T. Saraiva, Fuzzy modelling of power system optimal load flow, IEEE Trans. Power Syst., vol. 7, no. 2, pp. 843849,
May 1992.

[10] H. Heitsch and W. Rmisch, Scenario reduction algorithms in stochastic programming, Comput. Optim. Appl., vol. 24, pp. 187206,
2003.
[11] N. Grwe-Kuska, H. Heitsch, and W. Rmisch, Scenario reduction and
scenario tree construction for power management problems, in Proc.
IEEE Power Tech Conf., Bologna, Italy, Jun. 2003.
[12] L. Wu, M. Shahidehpour, and T. Li, Stochastic security-constrained
unit commitment, IEEE Trans. Power Syst., vol. 22, no. 2, pp.
800811, May 2007.
[13] J. M. Morales, S. Pineda, A. J. Conejo, and M. Carrin, Scenario reduction for futures market trading in electricity markets, IEEE Trans.
Power Syst., vol. 24, no. 2, pp. 878888, May 2009.
[14] A. J. Conejo, R. Garca-Bertrand, M. Carrin, A. Caballero, and A.
de Andrs, Optimal involvement in futures markets of a power producer, IEEE Trans. Power Syst., vol. 23, no. 2, pp. 703711, May
2008.
[15] F. J. Heredia, M. J. Rider, and C. Corchero, Optimal bidding strategies
for thermal and generic programming units in the day-ahead electricity
market, IEEE Trans. Power Syst., vol. 25, no. 3, pp. 15041518, Aug.
2010.
[16] T. Li, M. Shahidehpour, and Z. Li, Risk-constrained bidding strategy
with stochastic unit commitment, IEEE Trans. Power Syst., vol. 22,
no. 1, pp. 449458, Feb. 2007.
[17] V. S. Pappala, I. Erlich, K. Rohrig, and J. Dobschinski, A stochastic
model for the optimal operation of a wind-thermal power system,
IEEE Trans. Power Syst., vol. 24, no. 2, pp. 940950, May 2009.
[18] Y. Wang, Q. Xia, and C. Kang, Unit commitment with volatile node
injections by using interval optimization, IEEE Trans. Power Syst.,
vol. 26, no. 3, pp. 17051713, Aug. 2011.
[19] A. T. Sari and A. M. Stankovi, An application of interval analysis
and optimization to electric energy markets, IEEE Trans. Power Syst.,
vol. 21, no. 2, pp. 515523, May 2006.
[20] O. Alsac and B. Stott, Optimal load flow with steady-state security,
IEEE Trans. Power App. Syst., vol. PAS-93, no. 3, pp. 745751, May
1974.
[21] F. Capitanescu, M. Glavic, D. Ernst, and L. Wehenkel, Contingency
filtering techniques for preventive security-constrained optimal power
flow, IEEE Trans. Power Syst., vol. 22, no. 4, pp. 16901697, Nov.
2007.
[22] F. Capitanescu, S. Fliscounakis, P. Panciatici, and L. Wehenkel, Dayahead security assessment under uncertainty relying on the combination of preventive and corrective controls to face worst-case scenarios,
in Proc. PSCC2011, Stockholm, Sweden, Aug. 2226, 2011.
[23] C. Lehmkster, Security constrained optimal power flow for an economical operation of FACTS-devices in liberalized energy markets,
IEEE Trans. Power Del., vol. 17, no. 2, pp. 603608, Apr. 2002.
[24] F. Xiao and J. D. McCalley, Risk-based security and economy tradeoff
analysis for real-time operation, IEEE Trans. Power Syst., vol. 22, no.
4, pp. 22872288, Nov. 2007.
[25] P. P. Varaiya, F. F. Wu, and J. W. Bialek, Smart operation of smart
grid: Risk-limiting dispatch, Proc. IEEE, vol. 99, no. 1, pp. 4057,
Jan. 2011.
[26] P. Panciatici, Y. Hassaine, S. Fliscounakis, L. Platbrood, M. OrtegaVazquez, J. L. Martinez-Ramos, and L. Wehenkel, Security management under uncertainty: From day-ahead planning to intraday operation, in Proc. IREP Symp., Buzios, Brazil, 2010.
[27] R. N. Kackar, Off-line quality control, parameter design, and the
Taguchi method, J. Quality Technol., vol. 17, no. 4, pp. 176188,
1985.
[28] G. S. Peace, Taguchi Methods: A Hand on Approach. Reading, MA:
Addison Wesley, 1993.
[29] Q. Wu, On the optimality of orthogonal experimental design, Acta
Mathematicae Applagatae Sinica, vol. 1, no. 4, pp. 283299, Nov.
1978.
[30] H. Yu, C. Y. Chung, and K. P. Wong, Robust transmission network
expansion planning method with Taguchis orthogonal array testing,
IEEE Trans. Power Syst., vol. 26, no. 3, pp. 15731580, Aug. 2011.
[31] B. Borkowska, Probabilistic load flow, IEEE Trans. Power App.
Syst., vol. PAS-93, pp. 752759, Apr. 1974.
[32] R. N. Allan, C. H. Grigg, and M. R. G. AL-Shakarchi, Numerical
techniques in probabilistic load flow problems, Int. J. Numer. Meth.
Eng., vol. 10, pp. 853860, 1976.
[33] M. S. Phadke, Quality Engineering Using Robust Design. Englewood Cliffs, NJ: Prentice-Hall, 1989.
[34] Orthogonal Arrays (Taguchi Designs). [Online]. Available:
http://www.york.ac.uk/depts/maths/tables/orthogonal.htm.

YU AND ROSEHART: OPTIMAL POWER FLOW ALGORITHM TO ACHIEVE ROBUST OPERATION

[35] Y. W. Leung and Y. Wang, An orthogonal genetic algorithm with


quantization for global numerical optimization, IEEE Trans. Evol.
Comput., vol. 5, no. 1, pp. 4153, Feb. 2001.
[36] S. H. Park and J. Antony, Robust Design for Quality Engineering and
Six Sigma. Singapore: World Scientific, 2008.
[37] A. C. Saramourtsis, A. G. Bakirtzis, P. S. Dokopoulos, and E. S. Gavanidou, Probabilistic evaluation of the performance of wind-diesel
energy systems, IEEE Trans. Energy Convers., vol. 9, no. 4, pp.
743752, Dec. 1994.
[38] T. T. Soong, Fundamentals of Probability and Statistics for Engineers. West Sussex, U.K.: Wiley, 2004.
[39] R. D. Zimmerman, C. E. Murillo-Snchez, and R. J. Thomas, MATPOWER: Steady-state operations, planning and analysis tools for
power systems research and education, IEEE Trans. Power Syst., vol.
26, no. 1, pp. 1219, Feb. 2011.
[40] T.-H. Yeh and L. Wang, A study on generator capacity for wind
turbines under various tower heights and rated wind speeds using
Weibull distribution, IEEE Trans. Energy Convers., vol. 23, no. 2,
pp. 592602, Jun. 2008.
[41] Power Systems Test Case Archive. [Online]. Available: http://www.ee.
washington.edu/research/pstca.

1817

Han Yu (M11) received the B.S., M.S., and Ph.D. degrees in electrical engineering from North China Electric Power University, Beijing, China, in 1995,
1998, and 2009, respectively.
Currently, she is a Postdoctoral Researcher at the University of Calgary, Calgary, AB, Canada. From 1998 to 2010, she worked as a Lecturer in North China
Electric Power University. She also worked in The Hong Kong Polytechnic University as a Research Assistant from February 2007 to October 2008, and as a
Research Associate from March 2009 to June 2010. Her research interests include power system planning and operation as well as power system reliability.

W. D. Rosehart (M01SM06) received the B.Sc, M.Sc., and Ph.D. degrees in


electrical engineering from the University of Waterloo, Waterloo, ON, Canada,
in 1996, 1997, and 2001, respectively.
Currently, he is a Professor and the Head of the Department of Electrical and
Computer Engineering at the University of Calgary, Calgary, AB, Canada. His
main research interests are in the areas of numerical optimization techniques,
power system stability, and modeling power systems in a deregulated environment.

You might also like