Electrical Power and Energy Systems 43 (2012) 1106–1113
Contents lists available at SciVerse ScienceDirect
Electrical Power and Energy Systems journal homepage: www.elsevier.com/locate/ijepes
Interval arithmetic in current injection power flow analysis q L.E.S. Pereira 2, V.M. da Costa ⇑,1, A.L.S. Rosa 2 Department of Electrical Engineering, Federal University of Juiz de Fora, Campus Universitário, Bairro Martelos, 36036-330 Juiz de Fora, MG, Brazil
a r t i c l e
i n f o
Article history: Received 19 April 2011 Received in revised form 17 May 2012 Accepted 18 May 2012 Available online 15 July 2012 Keywords: Current injection power flow Data uncertainty Electrical power systems Interval arithmetic Krawczyk algorithm
a b s t r a c t This paper incorporates interval arithmetic into current injection method to solve power flow problem under both load and line data uncertainty. In this new methodology, the resulting interval nonlinear system of equations is solved using Krawczyk method. The proposed methodology is implemented in the Matlab environment using the Intlab toolbox. Results are compared with those obtainable by Monte Carlo simulations. IEEE test systems and a South – southeastern Brazilian network are used. Ó 2012 Elsevier Ltd. All rights reserved.
1. Introduction Power flow [1,2] is the most frequently performed study in electric power systems, and deals with the calculation of voltages and line flows, in a large sparse electrical network, for a given load and generation schedule. The conventional power flow solution comprises power equations expressed in terms of polar or rectangular voltage coordinates. Over the last years, the current injection power flow has been applied to different electric power system areas and remarkable results have been published in literature, as follows: Current-versions power flow in voltage rectangular coordinates are proposed in Refs. [3,4]. Based on [4], a sparse augmented formulation for solving a set of control devices in the current injection power flow problem is described in Refs. [5,6]. Many different flexible AC transmission system devices and controls are incorporated into the power flow problem by using this augmented methodology. Based on [4], a second order power flow methodology by using current injection equations expressed in voltage rectangular coordinates is presented in Ref. [7]. This proposed method leads to a substantially faster second order power flow solution, when com-
q This paper presents a new mathematical model regarding power flow solutions under load and line data uncertainties by using current injection equations expressed in rectangular voltage coordinates and the interval arithmetic. ⇑ Corresponding author. Tel.: +55 32 21023461; fax: +55 32 21023442. E-mail addresses:
[email protected] (L.E.S. Pereira), vander.costa@ ufjf.edu.br (V.M. da Costa),
[email protected] (A.L.S. Rosa). 1 Author has published other papers in the International Journal of Electrical and Energy Systems using the same kind of equations and coordinates. 2 Tel.: +55 32 32325352; fax: +55 32 21023442.
0142-0615/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.ijepes.2012.05.034
pared to the conventional method expressed in terms of power mismatches and written in voltage rectangular coordinates. The current injection rectangular power flow [4] is extended in [8] for solving an unbalanced distribution three-phase power flow problem. This methodology presents an expressive mathematical robustness and converges with a reduced number of iterations. A step size optimization factor to be used for solving unbalanced three phase distribution current injection is presented in Ref. [9]. On the other hand, regarding uncertainty in electric power systems, some comments become necessary. All loads are provided by measurement devices which are frequently inaccurate. This uncertainty in input data can be enlarged due to both round off error accumulation and truncating processes that occur in numerical computation. Additionally, electric power systems evolve through time, and it is reasonable to evaluate the range of all plausible conditions that might be encountered as a result of expected uncertainties [10]. Some technical publications can be pointed out as follows: The input data are subject to uncertainty. Interval analysis can be used for taking into account the effects of errors on numerical analysis. It is based on interval operations including interval arithmetic. The technique calculates the interval between the upper and lower bounds regarding variables under uncertainty. To yield solutions in a mathematical sense and to express the variable under uncertainty as an interval variable are the main advantages. On the other hand, the main limitation is sometimes to overestimate the interval between the upper and lower bounds. Thus, loads and other parameters can be characterized not by a single number but rather by a range of real values or a real interval [10–13]. In Ref. [10], the Newton operator is used to solve interval nonlinear
L.E.S. Pereira et al. / Electrical Power and Energy Systems 43 (2012) 1106–1113
equations. The Gauss–Seidel method is used to solve the interval linear equations. The input data for power flow problem can be described by random variables taking into account the probabilistic nature of loads, generation and networks. It is assumed that the parameter varies according to some probability distribution. The effects of uncertainty on the steady-state behavior of electric power systems can be assessed by a stochastic or probabilistic power flow [14–18]. Another approach reported in literature for dealing with uncertainties considers that the vague or imprecise information is represented by a fuzzy number. Loads and generations are represented using possibility distributions. As a result, power flow state and output variables have also possibility distributions [19–22]. There is a fundamental difference between probabilistic approaches and possibilistic ones including interval arithmetic. Both approaches are not competing but cooperative, and the type of information that each one provides is very different, because one is quantitative, though random, and the other is qualitative [10]. As can be noted, the current injection power flow formulation has been widely applied in several studies related to the power systems area and final results have been highly encouraging. In addition, when the input conditions are uncertain, numerous scenarios need to be analyzed and reliable solution algorithms that incorporate the effect of data uncertainties into the power flow analysis must be developed. Therefore, the main objective of this paper is to develop a new model of power flow under uncertainty by incorporating interval arithmetic into the current injection formulation. Besides, the computational performance of this new model in comparison with the interval power flow in polar coordinates is presented. The notations adopted in the paper are the conventional ones whenever possible. Matrices are shown in bold. The over scripts d and i refer to deterministic and interval quantities, respectively.
According to Ref. [10], interval addition and multiplication are associative and commutative. However, the distributive law does not always hold for interval arithmetic. The failure of the distribution law often causes overestimation. In some especial cases, the distributive law remains valid. In the Matlab toolbox Intlab [26,27] real intervals may be stored by either infimum and supremum or by midpoint and radius. Intlab enables basic interval operations to be performed on real and complex interval scalars, vectors and matrices. Interval functions such as trigonometric and exponential are also available. 3. Krawczyk method One of the most used approaches for solving a set of nonlinear equations is the Krawczyk method, which stems from Newton method. Let f be a nonlinear function such that
f ðxÞ ¼ 0
f ðyÞ ¼ f ðxÞ þ JðbÞðy xÞ
JðbÞðy xÞ ¼ f ðxÞ
JðXÞðX xÞ ¼ f ðxÞ
½I JðXÞðx yÞ ¼ f ðxÞ þ x y
diamðXÞ 2
ð3Þ
radðXÞ ¼
ð12Þ
I represents the identity matrix. Solving (12) for y
Kðx; XÞ ¼ x f ðxÞ þ ½I JðXÞðX xÞ
ð2Þ
ð11Þ
Adding term (x y) to both sides of Eq. (11)
Interval mathematics [23,24] considers a set of methods for handling intervals that approximate uncertain data. These methods are based on the definition of both interval arithmetic and optimal scalar product [25]. An interval number [x1, x2] is the set of real numbers x such that x1 6 x 6 x2. x1 is the infimum and x2 is the supremum. The interval X is defined by X ¼ ½x1 ; x2 ¼ f~ x2 Rjx1 6 ~ x 6 x2 g. The midpoint, the diameter and the radius of an interval X are given by
diamðXÞ ¼ x2 x1
ð10Þ
Defining the interval [x, y] e X then
Since [x, y] e Xy can be replaced with interval X
ð1Þ
ð9Þ
J represents the Jacobian matrix. Assuming f(y) = 0
y ¼ x f ðxÞ þ ½I JðXÞðy xÞ
1 ðx1 þ x2 Þ 2
ð8Þ
Let y be a incremental value from x and b be a value between x and y. By applying the mean value theorem to Eq. (8)
2. Interval arithmetic
midðXÞ ¼
1107
ð13Þ
ð14Þ
K(x, X) is called Krawczyk operator, and corresponds to interval solution in Eq. (13). The insertion of both a preconditioning matrix C and iteration h into Eq. (14) yields [11]
KðxðhÞ ; X ðhÞ Þ ¼ xðhÞ Cf ðxðhÞ Þ þ ½I CJðX ðhÞ ÞðX ðhÞ xðhÞ Þ
ð15Þ
X ðhþ1Þ ¼ X ðhÞ \ KðxðhÞ ; X ðhÞ Þ
ð16Þ
Eq. (16) means that the interval Krawczyk method provides the solution through the intersection of two interval sets. Besides, the main advantage of this operator is that no interval linear equations have to be solved at any iteration. C is a preconditioning matrix given by the midpoint inverse of J(X). 4. Interval current injection power flow – proposed method
Interval arithmetic operations are defined such that the interval encloses all possible real results [23] in order to guarantee the reliability of interval methods. The elementary operations, such as addition, subtraction, multiplication and division, are defined as follows
X þ Y ¼ ½x1 þ y1 ; x2 þ y2
ð4Þ
X Y ¼ ½x1 y2 ; x2 y1
ð5Þ
X Y ¼ ½minðx1 y1 ; x1 y2 ; x2 y1 ; x2 y2 Þ; maxðx1 y1 ; x1 y2 ; x2 y1 ; x2 y2 Þ
ð6Þ
X 1 1 if ¼ ½x1 ; x2 ; Y y2 y1
ð7Þ
0 R ½y1 ; y2
This paper proposes a new methodology to handle load and line data uncertainty in electric power systems by modeling the power flow problem through current injection equations written in rectangular voltage coordinates. Additionally, this paper proposes to calculate, in an interval manner, active and reactive generations, active and reactive power flows and losses. 4.1. Initialization of iterative process The interval current injection power flow (ICIPF) method is run after convergence of deterministic power flow. Its initialization is carried out based on deterministic voltage profile and on definition of load variations as follows:
1108
L.E.S. Pereira et al. / Electrical Power and Energy Systems 43 (2012) 1106–1113
Pidk ¼ ½Pddk ð1 aPk Þ;
P ddk ð1 þ aPk Þ
Q idk ¼ ½Q ddk ð1 aQ k Þ;
Q ddk ð1 þ aQ k Þ
ð19Þ
out the interval power flow, and it is evaluated from the current injection Jacobian matrix at deterministic power flow solution. f(x) refers to interval current mismatches vector, that is, f ðxÞ ¼ ½DIir DIim t , and it is calculated only once. The state variable vector x corresponds to the real and imaginary voltage components which stem from the deterministic power flow, that is, x ¼ ½V dr V dm t . The vector X corresponds to the interval power flow solution. The over script t denotes transposed vector. After calculating the Krawczyk operator, a new interval voltage solution is obtained by using (16). To check convergence of the proposed method, the difference between radii at iteration (h + 1) and radii at iteration (h) is calculated. If the difference is greater than a specified tolerance, denoted by s, then Krawczyk method must be employed to calculate new interval voltages. Otherwise, the iterative process is stopped.
ð20Þ
4.4. Calculation of interval output variables
ð17Þ ð18Þ
where P dk þ jQ dk is the complex load power at bus k. aPk and aQ k are factors which denote active and reactive load variations at bus k. Variations of resistances , reactances and shunt susceptances of each branch are defined by equations similar to (17) and (18), with the exception that Pd and Qd are replaced by corresponding line data. Interval voltages are initialized by using the deterministic voltage profile as midpoint and the largest load or line data variation factor as radius of interval. Thus
V irk ¼ ½V drk ð1 amax Þ; V imk ¼ ½V dmk ð1 amax Þ;
V drk ð1 þ amax Þ V dmk ð1 þ amax Þ
where V rk þ jV mk are the real and imaginary voltage components at bus k. The strategy adopted is to consider amax as the largest factor between all aPk and aQ k in order to ensure a good initial condition for convergence of iterative process. 4.2. Calculation of interval current mismatches The real and imaginary components of current injection equations written in rectangular coordinates, regarding bus k, are given respectively by [7,9]
ð21Þ
X V m P k V rk Q k ðGki V mi þ Bki V ri Þ k 2 V rk þ V 2mk i2/k
ð22Þ
Pik V drk þ Q ik V dmk
DIimk ¼ Idmk
ðV dk Þ2 Pik V dmk Q ik V drk ðV dk Þ2
ð27Þ
The increment regarding g can be expressed as nonlinear functions of real and imaginary voltage components at buses k and m. As a consequence, it is possible to linearize Eq. (27) by using the Taylor series around the corresponding state variables calculated through deterministic power flow program. Therefore
@g @g @g @g DV r k þ DV m k þ DV r m þ DV mm @V rk @V mk @V rm @V mm
ð28Þ
where DV rk þ jDV mk is the complex voltage variation at bus k. The interval increment of g can be expressed as follows:
where Irk þ jImk is the complex injected current at bus k; Gki þ jBki is the k i element of bus admittance matrix; /k denotes the set of buses directly connected to bus k, including itself; Pk + jQk is the net complex injected power at bus k. The real and imaginary components of interval current mismatches derived from (21) and (22) are given by
DIirk ¼ Idrk
g ¼ gðV rk ; V mk ; V rm ; V mm Þ
Dg ¼
X V r P k þ V mk Q k Irk ¼ ðGki V ri Bki V mi Þ k 2 V rk þ V 2mk i2/k I mk ¼
Let g denote any output power flow variable such as power flow and losses, and let k m be the branch under analysis. Thus
ð23Þ
ð24Þ
Dg i ¼
Pik ¼ Pigk Pidk
ð25Þ
Q ik ¼ Q igk Q idk
ð26Þ
Vk is the voltage magnitude at bus k; DIrk þ jDImk is the complex current mismatch at bus k; Pgk þ jQ gk is the generated complex power at bus k. Interval powers are constant during the interval power flow problem. Therefore, current mismatches are calculated only once. 4.3. Iterative process Eq. (15) refers to the Krawczyk method. For the sake of clarity, some comments about this equation need to be presented. The interval Jacobian matrix is calculated using interval voltages and its structure is shown in Ref. [7]. The matrix C is constant through-
ð29Þ
Active and reactive power generations are calculated in the same way. Let k be the bus generation under analysis. The function g depends on the real and imaginary voltage components in all buses directly connected to k, including itself. Therefore, the number of partial derivatives presented in Eq. (29) depends on the number of adjacent buses. The interval voltages calculated at the end of the iterative process may be substituted in Eq. (29). However, new interval operations need to be carried out and this procedure may lead to a large and inaccurate diameter regarding Dig . In order to improve the accuracy of intervals, Eq. (29) should be written in terms of interval current mismatches which are calculated from the input data of power flow program. Therefore
2
where
@g @g @g @g DV irk þ DV imk þ DV irm þ DV imm @V rk @V mk @V rm @V mm
3
.. .
2 3 .. 7 6 . 7 6 i 6 6 DV r 7 6 7 7 k 7 6 X 7 6 7 6 7 6 DV irm 7 6 7" 6 # Y 7 6 7 DI i 6 . 7 6 6 m 6 . 7 ¼ 6 .. 7 6 . 7 6 . 7 7 DI i 7 6 r 7 6 DV i 7 6 6 Z 7 6 mk 7 6 7 7 6 7 6 DV i 7 6 4W 5 6 mm 7 .. 5 4 .. . .
ð30Þ
where X, Y, Z and W are the rows of inverse current injection Jacobian matrix evaluated after convergence of deterministic power flow program. By substituting Eq. (30) in (29)
Dg i ¼
@g @g @g @g Xþ Zþ Yþ W @V rk @V mk @V rm @V mm
"
DIim DIir
# ð31Þ
1109
L.E.S. Pereira et al. / Electrical Power and Energy Systems 43 (2012) 1106–1113
The term between brackets is solved through simple algebraic operations. The corresponding interval of g is computed as follows:
g i ¼ g d þ Dg i :
ð32Þ
Table 1 Voltage magnitudes – IEEE 14 bus. Bus
Method
Lower voltage (pu)
Upper voltage (pu)
Deterministic voltage (pu)
4
ICIPF MCS
1.01724 1.01718
1.01827 1.01816
1.01775
9
ICIPF MCS
1.05536 1.05514
1.05666 1.05673
1.05601
10
ICIPF MCS
1.05044 1.05021
1.05172 1.05176
1.05108
13
ICIPF MCS
1.04988 1.04996
1.05107 1.05079
1.05048
14
ICIPF MCS
1.03505 1.03457
1.03635 1.03644
1.03570
4.5. Solution methodology The proposed algorithm can be summarized in the following steps: Step 1: Run a deterministic power flow program. Step 2: Calculate both load and line data variations using (17) and (18). Step 3: Initialize the interval voltage profile using (19) and (20). Step 4: Calculate the current mismatches using (23) and (24). Step 5: Apply the Krawczyk operator according to Eq. (15). Step 6: Update the interval voltages vector using Eqs. (33) and (34).
V rðhþ1Þ ¼ V rðhÞ \ Kðx; XÞ ðhþ1Þ ðhÞ ¼ Vm \ Kðx; XÞ Vm
ð33Þ ð34Þ
Step 7: Check convergence through s. The value of s adopted in this paper is 104. If convergence is not reached, then go back to Step 5. Otherwise go to Step 8. Step 8: Calculate power flows, losses and power generations. 5. Results
Table 2 Phase angles – IEEE 14 bus. Bus
Method
Lower angle (°)
Upper angle (°)
Deterministic angle (°)
4
ICIPF MCS
10.32885 10.51332
10.21161 10.11254
10.27021
9
ICIPF MCS
14.94749 15.21633
14.82033 14.66093
14.88390
10
ICIPF MCS
15.10555 15.37550
14.97857 14.81613
15.04204
13
ICIPF MCS
15.16411 15.44432
15.03773 14.87507
15.10091
14
ICIPF MCS
16.03751 16.33681
15.91053 15.73436
15.97401
5.1. Initial considerations In order to perform this study some simulations were accomplished by using IEEE 14 and 300 bus systems and a practical Brazilian network composed of 1768 buses, 2527 branches and 119 generation buses. For the three test systems, the total amount of buses, branches and generation buses is 2082, 2958 and 171, respectively. The tolerance adopted for convergence of the iterative process, related to both deterministic and interval power flow methods, is 104 pu. The Monte Carlo simulation (MCS) method validates the proposed methodology. One million of Monte Carlo simulations were performed for IEEE test systems, and three hundred thousand for 1768 bus system. For each simulation, different values of power injections and line data, within the intervals previously defined, are selected and conventional power flow solutions are performed. Interval solutions are obtained by monitoring the largest and the smallest values of both state and output variables calculated during all simulations. For the lack of space, results are displayed for only buses and branches that present the five largest relative errors (e) in comparison with MCS method. These errors are expressed in terms of absolute values. The radius of an interval associated with any input and output variable is defined around its respective deterministic value. In this paper, the radius of 5% is considered for series resistances and reactances and shunt susceptances of all IEEE 14 bus branches. Radii of 3% and 2% are considered for all active and reactive loads of 300 bus and 1768 bus systems. In addition, radii of 3% and 2% are considered for only 20% of the branches of IEEE 300 bus and 1768 bus, respectively. These branches were randomly chosen. The radius of 1% is assumed for all active power generations.
Table 3 Voltage magnitudes – IEEE 300 bus. Bus
Method
Lower voltage (pu)
Upper voltage (pu)
Deterministic voltage (pu)
17
ICIPF MCS
1.03297 1.06133
1.09686 1.06835
1.06492
120
ICIPF MCS
0.92968 0.95378
0.98719 0.96300
0.95844
139
ICIPF MCS
0.98136 1.01065
1.04206 1.01279
1.01171
192
ICIPF MCS
0.90933 0.93249
0.96558 0.94131
0.93746
234
ICIPF MCS
1.00754 1.03765
1.06987 1.03987
1.03871
Table 4 Phase angles – IEEE 300 bus. Bus
Method
Lower angle (°)
Upper angle (°)
Deterministic angle (°)
17
ICIPF MCS
13.81965 13.52391
12.30700 11.87730
13.04295
120
ICIPF MCS
9.23880 10.20123
8.20891 7.46745
8.70912
139
ICIPF MCS
3.72098 3.96595
3.30109 3.14304
3.50478
5.2. Calculation of state variables
192
ICIPF MCS
11.59565 11.95755
10.31392 10.57828
10.93694
Tables 1–4 present the state variables yielded by ICIPF and MCS methods for IEEE 14 and 300 bus systems. For example, the ICIPF
234
ICIPF MCS
21.81043 21.27168
19.54066 19.52708
20.64997
1110
L.E.S. Pereira et al. / Electrical Power and Energy Systems 43 (2012) 1106–1113
Fig. 1. Voltage magnitude error – IEEE 14.
method calculates that the voltage magnitude at bus 10, IEEE 14 bus, is included in the interval [1.05044, 1.05172] pu. Therefore, the voltage magnitude at bus 10 is not smaller than 1.05044 pu and not larger than 1.05172 pu. On the other hand, the phase angle at bus 10 is not smaller than 15.10555° and not larger than 14.97857°. The same reasoning can be extended to all output variables presented throughout the paper. The largest possible voltage variation around its deterministic value is 2.99% at bus 139 (IEEE 300), and the smallest possible voltage variation is 0.05% at bus 8 (IEEE 14). Regarding phase angles, the largest and the smallest variations are 6.08% at bus 120 (IEEE 300) and 0.19% at bus 11 (IEEE 14), respectively. By comparing with MCS, Figs. 1–4 depict the largest relative errors of voltage magnitudes and phase angles for some buses including those ones presented in Tables 1–4. NSV denotes the amount of each state variable under consideration. The proposed formulation calculates 4164 voltage magnitudes and the same amount of phase angles taking into account the lower and upper bounds. It can be observed that 98.82% of the voltage magnitudes yielded by ICIPF method present an error less than 1% and only 1.18% present an error greater than 1%. Regarding phase angles, 96.47% present an error less than 1% and only 0.47% an error greater than 5%.
5.3. Calculation of output variables
Fig. 2. Phase angle error – IEEE 14.
Tables 5 and 6 present the active and reactive losses yielded by ICIPF and MCS methods for IEEE 14 bus. The largest possible active loss variation around its deterministic value is 15.86% at branch 13–14 (IEEE 14), and the smallest possible loss variation is 0.53% at branch 6–11 (IEEE 14). Regarding reactive losses, the largest
Table 5 Active losses – IEEE 14 bus. Branch
Method
Lower active loss (MW)
Upper active loss (MW)
Deterministic active loss (MW)
4–5
ICIPF MCS
0.48756 0.49086
0.54246 0.53876
0.51230
1–5
ICIPF MCS
2.64202 2.64390
2.88102 2.88714
2.76284
9–10
ICIPF MCS
0.01097 0.01138
0.01478 0.01437
0.01283
9–14
ICIPF MCS
0.10455 0.10717
0.12782 0.12533
0.11563
13–14
ICIPF MCS
0.04590 0.04791
0.06232 0.06040
0.05379
Fig. 3. Voltage magnitude error – IEEE 300.
Table 6 Reactive losses – IEEE 14 bus.
Fig. 4. Phase angle error – IEEE 300.
Branch
Method
Lower reactive loss (MVAr)
Upper reactive loss (MVAr)
Deterministic reactive loss (MVAr)
4–5
ICIPF MCS
1.53793 1.54833
1.71107 1.69942
1.61594
7–9
ICIPF MCS
0.75321 0.76604
0.85208 0.83898
0.80001
9–10
ICIPF MCS
0.02915 0.03024
0.03925 0.03817
0.03408
9–14
ICIPF MCS
0.22240 0.22797
0.27189 0.26659
0.24597
13–14
ICIPF MCS
0.09345 0.09755
0.12688 0.12299
0.10953
1111
L.E.S. Pereira et al. / Electrical Power and Energy Systems 43 (2012) 1106–1113 Table 8 Reactive power flow – IEEE 300 bus. Deterministic reactive power flow (MVAr)
71.67347 71.81591
195.43647 194.07225
132.23509
ICIPF
120.57547
5.67565
62.45981
MCS
120.33121
5.15433
ICIPF
5.57547
18.87626
Method
15–14
ICIPF MCS
120– 116 172– 139 225– 192
Fig. 5. Active loss error – IEEE 14.
Upper reactive power flow (MVAr)
Branch
234– 228
Lower reactive power flow (MVAr)
MCS
5.81241
17.48281
ICIPF
380.75735
589.13778
MCS
381.33703
587.10320
ICIPF
261.55547
150.87546
MCS
260.06179
153.95327
7.21049
88.08206
57.57514
Tables 7 and 8 present the active and reactive power flows yielded by ICIPF and MCS methods for IEEE 300 bus. The largest possible active power flow variation around its deterministic value is 295.44% at branch 225–192 (IEEE 300), and the smallest possible active power flow variation is 10.35% at branch 90–92 (IEEE 300). Regarding reactive losses, the largest and the smallest variations are 568.85% at branch 225–192 (IEEE 300) and 16.28% at branch 86–102 (IEEE 300), respectively. By comparing with MCS, Figs. 7 and 8 depict the largest relative errors of power flows for some branches including those ones presented in Tables 7 and 8. The proposed formulation calculates 11,832 active power flows and the same amount of reactive power flows taking into account the lower and upper bounds. It can be
Fig. 6. Reactive loss error – IEEE 14.
and the smallest variations are 15.20% at branch 9–10 (IEEE 14) and 1.62% at branch 10–11 (IEEE 14), respectively. By comparing with MCS, Figs. 5 and 6 depict the largest relative error of losses for some branches including those ones presented in Tables 5 and 6. NOV denotes the amount of output variable under consideration. The proposed formulation calculates 5916 active losses and the same amount of reactive losses taking into account the lower and upper bounds. It can be observed that 98.93% of the active losses yielded by ICIPF method present an error less than 1% and only 0.22% an error greater than 5%. Regarding reactive losses, 98.02% present an error less than 1% and 0.46% an error greater than 5%.
Table 7 Active power flow – IEEE 300 bus.
Fig. 7. Active power flow error – IEEE 300. Deterministic active power flow (MW)
Branch
Method
Lower active power flow (MW)
Upper active power flow (MW)
15–14
ICIPF MCS
239.76552 238.18281
8.35758 8.04065
123.39609
120– 116
ICIPF
22.23347
144.65247
65.69656
MCS
22.52047
143.59711
172– 139
ICIPF
21.65324
15.76347
MCS
21.66069
15.51364
225– 192
ICIPF
36.86554
19.32554
MCS
38.78419
19.98552
ICIPF
55.43678
52.13547
MCS
56.25572
51.82317
234– 228
19.46229
9.88830
2.55217 Fig. 8. Reactive power flow error– IEEE 300.
1112
L.E.S. Pereira et al. / Electrical Power and Energy Systems 43 (2012) 1106–1113
Table 9 Reactive power generation – 1768 bus. Bus
Method
Table 11 Computation times relationships – 1768 bus. Upper reactive power generation (MVAr)
Lower reactive power generation (MVAr)
Deterministic reactive power generation (MVAr)
10
ICIPF MCS
803.11232 802.96909
566.24378 567.50238
143.9106
12
ICIPF MCS
1596.57575 1597.05173
667.65247 667.21511
433.0160
14
ICIPF MCS
84.13478 84.29319
130.03576 129.55984
25.3061
16
ICIPF MCS
1069.34268 1069.05045
226.13537 225.03091
422.6598
18
ICIPF MCS
507.75434 503.58183
1007.87645 1008.67623
222.4069
Tasks
IPPF
ICIPF
C f(x) J(X) K(x, X) Total time
1.14947 1.04351 2.83300 1 2.80120
1 1 1 1 1
that 96.82% of the generations yielded by ICIPF method present an error less than 1% and only 1.53% an error greater than 5%. 5.4. Computational Performance
Fig. 9. Reactive power generation error – 1768 bus.
observed that 98.62% of the active power flows yielded by ICIPF method present an error less than 1% and only 0.19% an error greater than 5%. Regarding reactive power flow, 98.20% present an error less than 1% and 0.49% present an error greater than 5%. The 1768 bus system is used to calculate the active and reactive power generations at slack bus. The interval active generation yielded by ICIPF method is [451.252, 470.487] MW and by MCS is [451.304, 469.704] MW. On the other hand, the reactive generations yielded by both methods are [1615.244, 1607.195] MVAr and [1616.652, 1608.332] MVAr, respectively. Therefore, a good agreement between results can be observed. The largest relative error of 0.166% refers to the upper bound of active power generation. Table 9 presents the reactive power generations yielded by ICIPF and MCS methods. The largest possible reactive generation variation around its deterministic value is 458.06% at bus 10 (1768 bus), and the smallest possible reactive generation variation is 113.09% at bus 55 (1768 bus). By comparing with MCS, Fig. 9 depicts the largest relative errors for some generation buses including those ones presented in Table 9. The proposed formulation calculates 342 power generation taking into account the lower and upper bounds. It can be observed
Table 10 Computation times – 1768 bus.
The ICIPF method achieves convergence with 3, 2 and 2 iterations for IEEE 14, IEEE 300 and 1768 bus, respectively. Table 10 displays computation times per iteration, in seconds, required by ICIPF to calculate the power flow solution regarding the 1768 bus system. In addition, there is a column for interval polar power flow (IPPF) with the purpose of comparing the performance of both methodologies. Computation times were obtained when using a personal computer AMD Athlon II X4 630 processor and 4 GB RAM. Table 11 displays the relationships between computation times required by both voltage coordinates for each one of the tasks related to interval power flow calculation. The main tasks considered are the assembling of preconditioning matrix C and the calculation of f(x), J(X) and K(x, X). ICIPF times are taken as reference. According to Table 10, the largest time is associated with the assembling of both interval Jacobian matrices. The times for the remaining tasks can be neglected. The assembling of interval current injection Jacobian matrix is faster than the polar version because the majority of its elements are constant, whereas the polar Jacobian matrix has all elements calculated through trigonometric functions. The ICIPF method presents a computational gain around 180% in comparison with the IPPF approach. Therefore, the use of current injection formulation requires a smaller computation time and, therefore, leads to a substantially faster power flow analysis subjected to uncertainty. 6. Conclusion This paper presents a new methodology for handling uncertainty in electrical power systems by using the interval arithmetic incorporated into the current injection power flow formulation. If input data vary within relatively small ranges, interval arithmetic yields good results that include all possible solutions. This methodology is very simple and reliable, and converges with a few iterations. Convergence is not an issue even in large systems. In general, ICIPF method not only presents good results in comparison with Monte Carlo simulations, but it also propitiates a remarkable computational gain in comparison with polar power flow coordinates. Since electric system data are uncertain, this paper demonstrates that the proposed method can be regarded as a powerful tool in power flow analysis under uncertainty and it can be indeed recommended for general use. References
Tasks
IPPF
ICIPF
C f(x) J(X) K(x, X) Total time
10.46428 1.01703 1790.26936 0.00001 1802.83546
9.10356 0.97462 631.93411 0.00001 643.59364
[1] Kulworawanichpong T. Simplified Newton–Raphson power flow solution method. Int J Electr Power Energy Syst 2010;32(6):551–8. [2] Mallick S, Rajan DV, Thakur SS, Acharjee P, Ghoshal SP. Development of a new algorithm for power flow analysis. Int J Electr Power Energy Syst 2011;33(8):1479–88. [3] Exposito AG, Ramos ER. Augmented rectangular load flow model. IEEE Trans Power Syst 2002;17(2):271–6.
L.E.S. Pereira et al. / Electrical Power and Energy Systems 43 (2012) 1106–1113 [4] Da Costa VM, Martins N, Pereira JLR. Developments in the Newton–Raphson power flow formulation based on current injections. IEEE Trans Power Syst 1999;14(4):1320–6. [5] Da Costa VM, Pereira JLR, Martins N. An augmented Newton–Raphson power flow formulation based on current injections. Int J Electr Power Energy Syst 2001;23(4):305–12. [6] Variz AM, Da Costa VM, Pereira JLR, Martins N. Improved representation of control adjustments into the Newton–Raphson power flow. Int J Electr Power Energy Syst 2003;25(7):501–13. [7] Ferreira CA, Da Costa VM. A second order power flow based on current injection equations. Int J Electr Power Energy Syst 2005;27(2):254–63. [8] Garcia PAN, Pereira JLR, Da Costa VM, Martins N. Three phase power flow calculations using the current injection method. IEEE Trans Power Syst 2000;15(2):508–14. [9] Da Costa VM, Oliveira ML, Guedes MR. Developments in the analysis of unbalanced three-phase power flow solutions. Int J Electr Power Energy Syst 2007;29(2):501–13. [10] Wang Z, Alvarado FL. Interval arithmetic in power flow analysis. IEEE Trans Power Syst 1992;7(3):1341–9. [11] Barboza L, Dimuro G, Reiser R. Towards interval analysis of the load uncertainty in power electric systems. In: 8th international conference on probabilistic methods applied to power systems. Ames, Iowa; 2004. [12] Vaccaro A, Canizares CA, Villacci D. An affine arithmetic-based methodology for reliable power flow analysis in the presence of data uncertainty. IEEE Trans Power Syst 2010;25(2):624–32. [13] Zhang P, Li W, Wang S. Reliability-oriented distribution network reconfiguration considering uncertainties of data by interval analysis. Int J Electr Power Energy Syst 2012;34(1):138–44. [14] Dimitrovski A, Tomsovic K. Boundary load flow solutions. IEEE Trans Power Syst 2004;19(1):348–55.
1113
[15] Stefopoulos GK, Meliopoulos AP, Cokkinides GJ. Probabilistic power flow with non-conforming electric loads. Int J Electr Power Energy Syst 2005;27(4):627–34. [16] Su CL. Probabilistic load-flow computation using point estimate method. IEEE Trans Power Syst 2005;20(4):1843–51. [17] Hu Z, Wang X. A probabilistic load flow method considering branch outages. IEEE Trans Power Syst 2006;21(2):507–14. [18] Usaola J. Probabilistic load flow with wind production uncertainty using cumulants and Cornish–Fisher expansion. Int J Electr Power Energy Syst 2009;31(3):474–81. [19] Ramaswamy M, Nayar KR. On-line estimation of bus voltages based on fuzzy logic. Int J Electr Power Energy Syst 2004;26(1):681–4. [20] Bijwe PR, Hanmandlu M, Pande VN. Fuzzy power flow solutions with reactive limits and multiple uncertainties. Electr Power Syst Res 2005;76(2):145–52. [21] Cortés-Carmona M, Palma-Behnke R, Jiménez-Estévez G. Fuzzy arithmetic for the DC load flow. IEEE Trans Power Syst 2010;25(1):206–14. [22] Kalesar BM, Seifi AR. Fuzzy load flow in balanced and unbalanced radial distribution systems incorporating composite load model. Int J Electr Power Energy Syst 2010;32(1):17–23. [23] Moore RE. Methods and applications of interval analysis. Philadelphia: SIAM; 1979. [24] Keafort RB, Kreinovich V. Applications of interval computations. Boston: Kluwer; 1996. [25] Kulish UW. Advanced arithmetic for the digital computer design of arithmetic units. Electron Notes Theor Comput Sci 1999;24. [26] Rump SM. Intlab-interval laboratory in developments in reliable computing. In: Csendes T. editor. Dordrecht. Kluwer; 1999. p. 77–104. [27] Hargreaves GI. Interval analysis in Matlab, Manchester Centre for Comp Math, Manchester, Numerical Analysis Report N 416; 2002.