You are on page 1of 5

A New Point Estimate Method for Probabilistic Load

Flow With Correlated Variables Including Wind Farms


1
Can Chen, 1Wenchuan Wu, 1Boming Zhang, 2Chanan Singh
1. Department of Electrical Engineering 2.Department of Electrical and Computer Engineering
Tsinghua University Texas A&M University
Beijing 100084, P.R.China College station,TX 77843 USA

Abstract—In this paper, a novel probabilistic load flow (PLF) to obtain PLF solutions[13]. Alternatively, approximate
method that combines Nataf transformation with Zhao’s point methods include point estimate method (PEM)[14][15] and first-
estimate method (PEM) is proposed. The new method can deal order second-moment method (FOSMM)[16]. These methods
with correlated non-normal input random variables (RVs) in approximate the statistical properties of the output variables.
PLF evaluation with improved accuracy and less computation Other methods in PLF research include the maximum entropy
time than existing methods. The proposed method is applied to a method with Gram-Charlier expansion, proposed for PLF
modified IEEE 118-bus test system with wind farms. A applications in 2013[17].
comparison with Monte Carlo Simulation (MCS) using
correlated variables is presented. In the end, the sensitivity Orthogonal transformation[18]-[20] is widely used to
analysis considering different correlation coefficients of wind transform RVs from correlated space to independent space,
speed is shown. which is easy to use. Recently, Nataf transformation[21] has
been used to sample correlated non-normal RVs. The Nataf
Key words—probabilistic load flow; Nataf transformation; point transformation has been applied into PLF evaluation,
estimate method; wind power combined with Latin hypercube sampling[22].
I. INTRODUCTION The main contribution of this paper is to combine Nataf
transformation with Zhao’s point estimate method[23] and to
The integration of wind power into power systems is one
apply it to correlated PLF evaluation. In contrast to Hong’s
of the most important applications of renewable energy. The
commonly used PEM formulation[27], Zhao’s PEM can
power output of wind turbines is fluctuating all the time, thus
inherently handle the correlated input RVs[25][26]. Zhao’s
bringing more uncertainties in power system operation and
original PEM with Rosenblatt transformation has been
planning. Besides, due to the geographical locations, the
applied to PLF evaluations with correlated output of wind
correlations of wind speeds in different wind farms should be
farms[27] with known joint probability density function (PDF).
taken into consideration.
Here, the Rosenblatt transformation is replaced by the Nataf
Probabilistic load flow (PLF) is an effective tool for transformation, which only requires the marginal distribution
power system analysis since it allows uncertain variables to function of each input RV and the correlation coefficients.
be taken into account. The concept of PLF was first The paper is organized as follows. Section 2 describes the
established by Borkowska[1]. After that, three kinds of detailed steps of combined Nataf transformation and Zhao’s
methods have been proposed: simulation methods, analytical point estimate method. Section 3 uses a modified IEEE118-
methods, and approximation methods. In simulation bus test system to demonstrate the accuracy and efficiency of
methods[2]-[4] , a large numbers of samples for uncertain input proposed method. Besides, a sensitivity analysis with different
variables are generated, and for each set of samples, a wind speed correlation coefficients is made and the average
deterministic power flow calculation is performed. relative errors compared with correlated Monte Carlo
Simulation methods, however, are typically time-consuming, Simulation (CMCS) are presented.
although achieving results with a high degree of accuracy. A
number of sampling techniques have been established to II. NOTATION AND DEFINITIONS
decrease the computational expense of simulation methods, L1 : independent standard normal space
including Latin hypercube sampling [5][6], Latin supercube
sampling[7], and importance sampling[8][9]. Analytical methods L2 : correlated standard normal space
are based on linearized [10] or multi-linearized[11] load flow
equations which include the cumulant method[12] and the L3 : desired sample space
convolution method[13]. The cumulant and Gram–Charlier
expansions have been applied to PLF evaluation and the ( z1 , z2 zn ) : random vector in L1
cumulative distribution curves of output RVs are given in
(w1 , w2 wn ) : random vector in L2
[12]. Discrete frequency domain convolution can also be used

This work was supported in part by the National Key Basic Research
Program of China (973 Program)( 2013CB228203) and National Science
Foundation of China (51177080) and New century excellent talents in
university(NCET-11-0281).
978-1-4799-6415-4/14/$31.00 ©2014 IEEE
Authorized licensed use limited to: Institute of Technology (Nirma University). Downloaded on November 10,2020 at 13:55:48 UTC from IEEE Xplore. Restrictions apply.
( y1 , y2 yn ) : random vector in L3 and CW are 1 and the non-diagonal elements are U0ij and Uij ,

CW :correlation coefficient matrix in L2 respectively for CY and CW , in the ith row and jth column.

CY :correlation coefficient matrix in L3 An empirical expression Uij TU0ij is proposed in [21].


Specially, for two-parameter Weibull distribution (wind
S Z :sample matrix in L1 speed), T has the following formulation

S W :sample matrix in L2
Vi V j V2 Vj
2

S Y :sample matrix in L3 T 1.063 - 0.004U0ij -0.2(  )- 0.001U02ij + 0.337( i2  2 )


Pi P j Pi P j
Vi V j V iV j
+ 0.007U0ij (  )- 0.007
III. METHODOLOGY Pi P j Pi P j
The set of load flow equation is expressed as X g (Y) . (1)
The goal of PLF evaluation is to obtain the statistical And for normal distribution (load power), T 1.
moments of output variables X . The input RVs in PLF
evaluation are load and wind turbines’ power output. In this C. Get sample matrix in space L3
paper, load is modeled as normal distributed as has been
The connection between random vector (w1 , w2 wn ) and
stated in [22][28] and wind speed is modeled as two-
parameter weibull distribution and a transformation from ( y1 , y2 yn ) is expressed as follows:
wind speed to power output is used[29]. Nataf transformation
is a novel method to achieve the transformation from wk )1 Fk ( yk ) , k 1, 2 n (2)
correlated non-normal space to independent standard normal
space, so it is proper to be used in PLF evaluation with Where ) is the cumulate density function (CDF) of
correlated wind power. standard normal distribution and Fk is the CDF of input
In this paper, Nataf transformation combined with Zhao’s variable yk . According to (2), the final sample matrix S Y can
point estimate method is proposed. Point estimate method be obtained.
uses a limited number of estimating points as well as their
weighting coefficients to represent random variables. The In this paper, the Nataf transformation is used to generate
procedure for the proposed method is described as follows. correlated non-normal estimating points in PEM and more
details about Nataf transformation can be found in [21].
A. Get sample matrix in space L1
D. Perform load flow calculation
In space L1 , for each standard normal variable, the
After obtaining the final sample matrix S Y , the
estimating points as well as their corresponding weight
deterministic load flow is then performed using each row of
coefficients can be obtained using Gaussian–Hermete
integration. The commonly used points and weight matrix S Y . For 3 point estimate method, there are altogether
coefficients are listed in [23]. In this paper, 3 points are used 3n computation times of deterministic load flow. Define
for each RV. So, the sample matrix S z has dimension of gi  m as the computation result when substitute the
3n u n where n is the number of input RVs. Each row (2i  m  2)th row into load flow equation.
represents a set of input variables. The (2i  m  2)th row can
E. Statistical analysis of output variables
be written as Zi  m (Pz1 , Pz2 zim Pzn ) where zim stands for
First of all, the load flow equation can be approximated
the mth estimating point of the i th variable and all the other using the following expressions.
variables are fixed at mean values and i 1, 2 n, m 1 1, 2,3 . n
g ( Y) | ¦ ( gi  g P )  g P (3)
B. Get sample matrix in space L2 i 1

Make Cholesky decomposition on correlation matrix CW , Where


T
and we have CW BB , where B is a lower triangular matrix.
gi g ( N 1 (Zi )) (4)
Then, the sample matrix in space L2 can be obtained using
SW SZ B . In the above expression, N 1 denotes the inverse Nataf
transformation and Z i is a vector in space L1 where the ith
In this step, the correlation matrix CW of correlated element is a RV and all the other elements are set to its mean
standard normal variables is unknown and should be value. Correspondingly, g P is the computation result with all
calculated from the matrix CY . The diagonal elements in CY the input RVs fixed at the mean values.

Authorized licensed use limited to: Institute of Technology (Nirma University). Downloaded on November 10,2020 at 13:55:48 UTC from IEEE Xplore. Restrictions apply.
Based on the approximation load flow equation, the mean In order to demonstrate the effectiveness of proposed
and kth order statistical moments of the output variables in method, correlated MCS with 5000 iterations is used as
PLF evaluation can be calculated as benchmark. Generally speaking, 5000 sampling is large
n
enough to get stable and accurate results in PLF evaluation. In
PX ¦ (P
i 1
xi  gP )  gP (5) this paper, only mean value and standard deviation of bus
voltage and line power are calculated and compared.
n
Four accuracy indices ( H PV ǃ H VV ǃ H PP ǃ H VP ) are
M kX ¦M
i 1
kxi ,k 2,3 (6)
defined.
Where
3
H PV :average relative error of bus voltages’ mean value
Px ¦p m
1
g ( N (Zi  m )) (7)
H VV
i
m 1 :average relative error of bus voltages’ standard
3 deviation
M kxi ¦p m ( g ( N 1 (Zi  m ))  P xi ) k , k 2,3 (8)
m 1
H PP :average relative error of line power’s mean value
In (7) and (8), pm is the weight coefficient of the mth
estimating point for each standard normal distribution. H VP :average relative error of line power’s standard
deviation
The Nataf transformation employs normal Copula
function[30] to build joint PDF of the input RVs with the In order to demonstrate the accuracy of Nataf
information of marginal distribution and correlation transformation, the orthogonal transformation is also used in
coefficients and has the advantage of being adaptable while Zhao’ point estimate method. The results are compared with
retaining a high degree of accuracy. The Zhao’s PEM CMCS and the four indices using the two transformation
combined with Nataf transformation can be easily used in methods are listed in table2 and table3.
correlated PLF evaluation integrated with wind farms.
TABLE 2 AVERAGE RELATIVE ERRORS FOR BUS VOLTAGES
IV. CASE STUDIES USING TWO TRANSFORMATION METHODS
Average relative error for bus voltages (%)
To demonstrate the performance of the proposed method, a Transformation
modified IEEE 118-bus test system is used. The parameters of Method H PV H VV
the IEEE 118-bus test system can be found in [31]. Each of Nataf 0.0237 2.4338
the buses (bus 81, 82, 95 and 96) is added into a wind farm via Orthogonal 0.0238 2.5921
a transmission line with parameter 0.01 + j0.01 p.u. There are
40 wind turbines in each wind farm and the parameters for TABLE 3 AVERAGE RELATIVE ERRORS FOR LINE POWER
wind turbines are listed in Table 1. USING TWO TRANSFORMATION METHODS
Transformation Average Relative Error For Active Power (%)
TABLE 1 WIND TURBINE PARAMETERS
Parameter Value Method H PP H VP
Rating capacity (MW) 2.5 Nataf 1.3709 2.4101
Cut-in speed (m/s) 3 Orthogonal 1.3710 2.5527
Cut-out speed (m/s) 25
Rating speed (m/s) 12.5
From table2 and table3, it can be seen that compared with
The wind speed data is from a real wind farm in Jilin orthogonal transformation method, Nataf transformation can
province, China. After maximum likelihood estimation, the achieve somewhat better results both in bus voltages and line
shape parameter is 2.01 and scale parameter is 7.28 in Weibull power. For the Nataf transformation takes the changes of
distribution. The total load in the test system is 4242MW, so correlation coefficient of wind speed in different spaces, the
the wind power penetration is 9.43%. The correlation computation results are more accurate.
coefficient for load is 0.1, and for wind speed it is 0.4. The The proposed method is implemented using Matpower 4.1,
correlations between wind powers and load demands are and simulations are carried out on a computer with an Intel i5-
neglected. 2410 CPU with a clock speed of 2.30 GHz and 4.0-GB of
The active power of load is normal distributed and the RAM. The time required for the simulations of the IEEE 118-
reactive power changes with active power according to a bus test system using the PEM are listed in tables 4, where the
constant power factor. The mean value for normal distribution time required using the CMCS is also shown for comparison.
can be found in [31] and the standard deviation is set 5% of TABLE 4 SIMULATION TIME COMPARISON
the mean value. This fluctuation is feasible for probabilistic Method PEM CMCS
load flow calculation. Normal distribution is commonly used Time (s) 14.58 220.71
for load power and the proposed method can be applied to From table4, it can be seen that the proposed method is
other kinds of distribution functions of load power. The wind computationally efficient compared with simulation method.
farms are modeled as PQ bus with power factor of 0.85 The computation time listed in table4 has three main parts: the
lagging. time for obtaining sample matrix; the time to perform

Authorized licensed use limited to: Institute of Technology (Nirma University). Downloaded on November 10,2020 at 13:55:48 UTC from IEEE Xplore. Restrictions apply.
deterministic load flow calculation and the time to make From Fig.1-Fig.4, it can be seen that with the correlation
statistical analysis about output variables. The most time- coefficient of wind speed increasing, H PV and H PP keep almost
consuming part is deterministic load flow calculation.
unchanged, with slight increase. However, H V and H V
V P
Sensitivity Analysis
become larger with correlation coefficient. It can also be
In order to check out the computation performance when concluded that the proposed method can deal with different
different wind speed correlations are considered, the following degrees of correlations in wind speed and has good
tests are performed. The correlation coefficient of load power computation accuracy.
is fixed 0.1 while that of wind speed varies from 0.3 to 0.6.
The results are shown in Fig.1-Fig.4. V. CONCLUSIONS
The method combining Nataf transformation with Zhao’s
0.03
PEM is proposed for PLF evaluation with correlated variables
average relative error(%)

0.028 including wind farms. The effectiveness of the method is


0.026 demonstrated using a modified IEEE 118-bus test system.
0.024
0.022
The major sources of error derive from the following
sources. First, the point estimate method is an approximate
0.02
0.3 0.4 0.5 0.6
method, since it uses a limited number of estimation points
along with weighting coefficients to represent random
correlation coefficient
variables. This inherent error in the method cannot be
Fig.1 Average relative error of bus voltages’ mean value under different
overcome. Second, errors from the Nataf transformation also
correlation coefficients arise. An empirical expression is used to obtain Uij from
Uij TU0ij .
3.2
average relative error(%)

3 The main features of the proposed method are as follows.


2.8 Nataf transformation is more accurate in dealing with
2.6
correlated non-normal RVs’ sampling, so the proposed
2.4
method can achieve computation results with higher accuracy.
2.2
2
When the correlation coefficients of the wind speed increase,
0.3 0.4 0.5 0.6 the average relative errors for both bus voltages and line
correlation coefficient
power’s mean values remain almost constant. However, the
average relative errors of the standard deviations of these
quantities increase slightly. Compared with the CMCS method
Fig.2 Average relative error of bus voltages’ standard deviation under with 5000 iterations, the computational time is significantly
different correlation coefficients reduced. In conclusion, the method reported here is found to
be effective for power system PLF calculations with correlated
1.5 wind farms.
average relative error(%)

1.3 ACKNOWLEDGMENT
1.1
0.9 The author also thanks China Scholarship Council(CSC)
0.7 that provides financial support of studying in the Texas A&M
0.5 University.
0.3 0.4 0.5 0.6
VI. REFERENCES
correlation coefficient
[1] B. Borkowska, “Probabilistic load flow,” IEEE Trans. Power App.
Syst.,vol.PAS-93, no.3, pp. 752–759, Apr. 1974.
Fig.3 Average relative error of line power’ mean value under different
[2] R. Y. Rubinstein and D. P. Kroese, Simulation and the Monte Carlo
correlation coefficients
Method. New York: Wiley, 2011, pp. 5-20.
[3] A. B. Rodrigues and M. G. Da Silva, “Probabilistic assessment of
3.5 available transfer capability based on Monte Carlo method with
average relative error(%)

3 sequential simulation,” IEEE Trans. Power Syst., vol. 22, no.1, pp. 484–
492, Feb. 2007.
2.5
[4] P. Jorgensen, J.S. Christensen, and J.O. Tande, “Probabilistic load flow
2 calculation using Monte Carlo techniques for distribution network with
1.5 wind turbine,” in Proc.1998 Int. Conf. Harmonics and Quality of Power,
pp.1146–1151, 1998.
1 [5] M. Stein, “Large sample properties of simulations using Latin hypercube
0.3 0.4 0.5 0.6 sampling,” Technometrics, vol. 29, no. 2, pp. 143–151,1987.
correlation coefficient [6] J. C. Helton and F. J. Davis, “Latin hypercube sampling and the
propagation of uncertainty in analyses of complex systems,” Reliab. Eng.
Fig.4 Average relative error of line power’ standard deviation under different Syst. Safe., vol. 81, no. 1, pp. 23–69, 2003.
correlation coefficients

Authorized licensed use limited to: Institute of Technology (Nirma University). Downloaded on November 10,2020 at 13:55:48 UTC from IEEE Xplore. Restrictions apply.
[7] A. B. Owen, “Latin supercube sampling for very high-dimensional
simulations,” ACM Trans. Modeling and Computer Simulation
(TOMACS), vol. 8, no. 1, pp. 71–102, Jan. 1998.
[8] P. W. Glynn and D.L. Iglehart, “Importance sampling for stochastic
simulations,” Management Science, vol. 35, no. 11, pp. 1367–1392, 1989.
[9] R. E. Melchers, “Importance sampling in structural systems,” Structural
Safety, vol. 6, no. 1, pp. 3–10, Jul. 1989.
[10]R. N. Allan and M. R. G. Al-Shakarchi, “Probabilistic a.c. load flow,”
Proc. Inst. Elect. Eng. vol. 123, no. 6, pp. 531–536, 1976.
[11]A. P. S. Meliopoulos, G. J. Cokkinides, and X. Y. Chao, “A new
probabilistic power flow analysis method,” IEEE Trans. Power Syst., vol.
5, no. 1, pp. 182–190, Feb. 1990.
[12]P. Zhang and S. T. Lee, “Probabilistic load flow computation using the
method of combined cumulants and Gram–Charlier expansion,” IEEE
Trans. Power Syst., vol. 19, no. 1, pp. 676–682, Feb. 2004.
[13]R. N. Allan, A. M. Leite Da Silva, and R. C. Burchett, “Evaluation
methods and accuracy in probabilistic load flow solutions,” IEEE Trans.
Power App. Syst., vol. PAS-100, no. 5, pp. 2539–2546, May. 1981.
[14]S. Chun-Lien, “Probabilistic load-flow computation using point estimate
method,” IEEE Trans. Power Syst., vol. 20, no. 4, pp. 1843–1851, Nov.
2005.
[15]J. M. Morales and J. Perez-Ruiz, “Point estimate schemes to solve the
probabilistic power flow,” IEEE Trans. Power Syst., vol. 22, no. 4, pp.
1594–1601, Nov. 2007.
[16]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. 385–388.
[17]C. C. Trevor Williams, “Probabilistic load flow modeling comparing
maximum entropy and Gram-Charlier probability density function
reconstructions,” IEEE Trans. Power Syst., vol. 28, no. 1, pp. 272–280,
Feb. 2013.
[18]A. Tamtum, A. Schellenberg, and W.D. Rosehart, “Enhancements to the
cumulant method for probabilistic optimal power flow studies,” IEEE
Trans. Power Syst., vol. 24, no. 4, pp. 1739–1746, Nov. 2009.
[19]M. E. Harr, “Probabilistic estimates for multivariate analyses,” Applied
Mathematical Modelling, vol. 13, no. 5, pp. 313–318, May. 1989.
[20]J. M. Morales, L. Baringo, A. J. Conejo, and R. Minguez, “Probabilistic
power flow with correlated wind sources,” IET Gen., Transm., Distrib.,
vol. 4, no. 5, pp. 641–651, May. 2010.
[21]P. L. Liu and A. Der Kiureghian, “Multivariate distribution models with
prescribed marginals and covariances,” Probabilistic Engineering
Mechanics, vol. 1, no. 2, pp. 105–112, Jun. 1986.
[22]Y. Chen, J. Wen, and S. Cheng, “Probabilistic load flow method based
on Nataf transformation and Latin hypercube sampling,” IEEE Trans.
Sustainable Enery., vol. 4, no. 2, pp. 294–301, Apr. 2013.
[23]H. Li, Z. Liu, and X. Yuan, “Nataf transformation based point estimate
method,” Chinese Science Bulletin, vol. 53, no. 6, pp. 627–632, Sep.
2008.
[24]H. P. Hong, “An efficient point estimate method for probabilistic
analysis,” Reliab. Eng. Syst. Safe., vol. 59, no. 3, pp. 261–267, Mar.
1998.
[25]Y. G. Zhao and T. Ono, “New point estimates for probability moments,”
J. Eng. Mech., vol. 126, no. 4, pp. 433–436, Apr. 2000.
[26]Y. G. Zhao and T. Ono T, “Moment methods for structural reliability,”
Structural Safety, vol. 23, no. 1, pp. 47–75, 2001.
[27]M. Mohammadi, A. Shayegani, and H. Adaminejad, “A new approach of
point estimate method for probabilistic load flow,” Int. J. Elect. Energy
Syst., vol. 51, no. 54–60, Oct. 2013.
[28]D. Villanueva, J. L. Pazos and A. Feijoo, “Probabilistic load flow
including wind power generation,” IEEE Trans. Power Syst., vol. 26, no.
3, pp. 1659-1667, Aug. 2011.
[29]H. Kim, C. Singh, and A. Sprintson, “Simulation and estimation of
reliability in a wind farm considering the wake effect,” IEEE Trans.
Sustainable Enery., vol. 3, no. 2, pp. 274-282, Apr. 2012.
[30]R. B. Nelsen, An Introduction to Copulas. New York: Springer,1999, pp.
5-15.
[31]R. Christie, “Power systems test case archive,” Electrical Engineering
dept., University of Washington, Apr. 2000. [online], Available:
http://www.ee.washington.edu/research/pstca.

Authorized licensed use limited to: Institute of Technology (Nirma University). Downloaded on November 10,2020 at 13:55:48 UTC from IEEE Xplore. Restrictions apply.

You might also like