Академический Документы
Профессиональный Документы
Культура Документы
Renewable Energy
journal homepage: www.elsevier.com/locate/renene
a r t i c l e i n f o a b s t r a c t
Article history: Increasing penetration of wind power in power systems causes difficulties in system planning due to the
Received 23 June 2015 uncertainty and non dispatchability of the wind power. The important issue, in addition to uncertain
Received in revised form nature of the wind speed, is that the wind speeds in neighbor locations are not independent and are in
11 January 2016
contrast, highly correlated. For accurate planning, it is necessary to consider this correlation in optimi-
Accepted 19 February 2016
zation planning of the power system. With respect to this point, this paper presents a probabilistic multi-
Available online 21 March 2016
objective optimal power flow (MO-OPF) considering the correlation in wind speed and the load. This
paper utilizes a point estimate method (PEM) which uses Nataf transformation. In reality, the joint
Keywords:
Optimal power flow
probability density function (PDF) of wind speed related to different places is not available but marginal
Probabilistic multi-objective optimization PDF and the correlation matrix is available in most cases, which satisfy the service condition of Nataf
Correlated wind power transformation. In this paper biogeography based optimization (BBO) algorithm, which is a powerful
Nataf transformation optimization algorithm in solving problems including both continuous and discrete variables, is utilized
in order to solve probabilistic MO-OPF problem. In order to demonstrate performance of the method,
IEEE 30-bus standard test case with integration of two wind farms is examined. Then the obtained results
are compared with the Monte Carlo simulation (MCS) results. The comparison indicates high accuracy of
the proposed method.
© 2016 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.renene.2016.02.064
0960-1481/© 2016 Elsevier Ltd. All rights reserved.
S. Shargh et al. / Renewable Energy 94 (2016) 10e21 11
estimate methods (2m and 2m þ 1 scheme) are used in Ref. [22] to presents simulation results of examining proposed method on
solve the POPF. The correlated samples of wind power injections IEEE 30-bus test system and finally Section 6 concludes the remarks
and load points have been generated by Cholesky decomposition and results of the paper.
method, and the interior point method is used to solve the deter-
ministic optimal power flow calculation.
A modified PEM is proposed in Ref. [23] to solve POPF, which 2. Uncorrelated PEM for probabilistic multi-objective OPF
considers correlation among wind power, solar energy and demand
load. A review of the probabilistic techniques used for POPF studies 2.1. Probabilistic multi-objective optimal power flow formulation
and the application of unscented transformation (UT) method to
generate correlated samples is presented in Ref. [24]. Zhaos point The multi-objective OPF problem with uncorrelated input vari-
estimate method (PEM) combined with Nataf transformation applied ables can be formulated generally as below:
into correlated PLF calculation is proposed in Ref. [25]. The proposed
Minimize FðX; YÞ ¼ fF1 ðX; YÞ; F2 ðX; YÞg
method can deal with correlated input RVs with normal or non-
S:t: GðX; YÞ 0 (1)
normal probability distributions and instead of PDFs of multivariate
HðX; YÞ ¼ 0
RVs requires raw moments of the marginal distribution function of
each input RV and their correlation coefficients. In Ref. [26] traditional where F1(X,Y) and F2(X,Y) are objectives of probabilistic optimal
2m þ 1 PEM extended by using a transformation to handle correlated power flow problem (POPF). X is the vector of control variables and
RVs with arbitrary distributions in POPF problem. In [27], a method is Y is the vector of uncertain input variables of POPF problem. H(X,Y)
proposed for incorporati2m þ 1 ng the effects of spatially and and G(X,Y) are representing the equality and inequality constraints
temporally correlated input RVs within the point estimate method for of the multi-objective problem, respectively.
POPF simulation. This method is used to solve multi-period POPF In this paper, fuel cost and total emission of the system are
including correlation among time intervals of wind power generation. considered as objectives of multi-objective POPF problem, which
Most of the papers studying effect of the correlation of uncertain can be formulated as follows [29]:
variables on POPF problem presented different methods to solve Fuel cost function:
the problem but few of them have studied effect of variation of
correlation on control and output variables. For instance, a new X
NG
2
method is presented in Ref. [26] in order to observe effect of cor- F1 ðX; YÞ ¼ ai þ bi Pgi þ ci Pgi (2)
relation on uncertain variables following different distributions but i¼1
do not investigate how the variables will affect with variation of
Emission cost function:
correlation. In other case, a new method for solving POPF with
correlated variables is presented in Ref. [23] and effect of correla- NG
X
2
tion on accuracy of the presented method is studied, but the impact F2 ðX; YÞ ¼ ai þ bi Pgi þ gi Pgi þ εi exp li Pgi (3)
of variation of the correlation on control variables was not studied. i¼1
It should be noted that, OPF problem in accurate model, should
Where F1(X,Y) is the total fuel cost of generation units in $/h, NG
consider many objectives simultaneously, as in reality is required
is total number of generation units, Pgi (MW) is active power output
[28]. So it is important to investigate effect of RVs and their cor-
of the ith generation unit and ai, bi, ci are the fuel cost coefficients of
relation in multi objective POPF which is more closer to reality and
ith generation unit. F2(X,Y) is total emission released from ith
can be more useful in planning of system to reach different targets,
generation unit (ton/h), Pgi(pu) is active power output of ith
simultaneously.
generator and ai, bi, gi, εi and li are their emission coefficients.
In this paper, Biogeography Based Optimization algorithm with
This paper uses weighted sum method to solve multi-objective
weighted sum method is used to solve probabilistic multi-objective
optimal power flow:
optimal power flow problem and traditional PEM based Nataf
transformation is utilized to handle correlation of wind sources and F1 ðX; YÞ F ðX; YÞ
load demands. This paper studies the effect of correlation on con- F ¼ k1 þ k2 2 (4)
F1min F2min
trol and output variables of MO-POPF problem. It is noteworthy
that, the effect of correlation on the control and output variables of where, k1 and k2 are the weighting coefficient of fuel cost and
multi-objective POPF problem is studied for the first time in this emission cost objectives, simultaneously, and F1min and F2min are
paper. The standard IEEE 30-bus test system with two extended minimum amount of them which can be obtained solving single
wind farms is used to examine the proposed method. objective OPF. In this approach, for balanced weighting of objective
Contributions of the paper are as follows: functions k1 and k2 are considered to be equal to 1.
Control variables:
1. Using Hongs point estimate method and Nataf transformation to
solve POPF problem with correlated input variables. X ¼ Pg ; Vg ; T; Qc 1N (5)
2. Solving probabilistic multi-objective optimal power flow
considering wind power and load points correlation. h i
3. Using biogeography based optimization method as the optimi- Pg ¼ Pg1 ; Pg2 ; …; PgðNg 1Þ (6)
1ðNg 1Þ
zation method.
4. Observing effect of correlation on the control variables and both h i
objectives of the probabilistic multi-objective optimal power flow. Vg ¼ Vg1 ; Vg2 ; …; VgNg (7)
1Ng
where, NL and Npq are the number of lines and PQ buses, Sij is Z ¼ FðX; YÞ (22)
apparent power flowing in the line connecting bus i to bus j, and VLi
In this model X and Y are control variables and uncertain input
is the voltage magnitude of PQ buses. It should be mentioned that
variables, respectively.
superscripts min and max show lower and upper bounds of the
corresponding parameter. X ¼ ½PG ; VG ; T; QG ; Y ¼ ½Pw ; Pd (23)
In probabilistic MO-OPF model, formulated in this paper, PEM is
2.2. Uncertainty modeling used to determine 2m þ 1 points on standard normal distribution for
each of the uncertain variables. Then samples of wind speeds and
One of the most important aspects of the probabilistic planning loads following specific distributions can be obtained by transforming
is to propose appropriate statistical model for the uncertain vari- the samples from standard normal distribution. After running pro-
ables in system. In this paper, load and wind generation are posed model 2m þ 1 times, raw moments of output variables
modeled as uncertain variables which are developed as following. including total fuel cost, released emission and control variables are
calculated. According to the 2m þ 1 PEM, each random variable is
clarified with three locations and weights which are determined by:
2.2.1. Load modeling
Load uncertainty is commonly modeled with normal distribu- yik ¼ mi þ xik si ði ¼ 1; 2; …; m; k ¼ 1; 2; 3Þ (24)
tion [30,19]. In this paper, active load demand is modeled as normal
distributed random variable and load power factor is assumed to be where, mi and si are the mean and standard deviation of yi and xik is
constant [30]. The corresponding PDF can be formulate as following the coefficient representing the kth location on its probability dis-
equation: tribution and m is the number of random input variables. Standard
S. Shargh et al. / Renewable Energy 94 (2016) 10e21 13
Fig. 1. Output power variation for a typical wind turbine according to wind speed.
locations xik are calculated using (25). 3. Biogeography based optimization for solving MO-OPF
problem
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
li3 3
xik ¼ þ ð 1Þ3k li4 l2i3 ðk ¼ 1; 2Þ xi3 ¼ 0 (25) This method uses the concept of how a good habitat is found in
2 4
process of migration from one habitat to another. In this method, an
where, li3 and li4 are the third and fourth standardized central optimization solution vector is known as habitat, the variables of
moments of yi known as skewness and kurtosis. Weighting factors every solution vector are called suitability index variables (SIV s)
related to each location yik are calculated by: and fitness of the solution vector is known as quality for living in a
habitat, called habitat suitability index (HSI). When some species
ð1Þ3k migrate from one habitat to other habitats, the process is called
wik ¼ ðk ¼ 1; 2Þ (26) emigration and when some species migrate into one habitat from
xik ðxi1 xi2 Þ
other habitats, the process is called immigration.
The following probabilistic mathematical model represents the
1 1
wi3 ¼ (27) concept of the emigration and the immigration process in BBO
m li4 l2i3 algorithm:
Assume that at time t the probability of existence of exactly s
After calculation of concentrations for all uncertain variables,
species in a habitat is Ps(t) then at time t þ Dt the probability is:
2m þ 1 vectors of input variables Y are constructed as follows:
Ps ðt þ DtÞ ¼ Ps ðtÞð1 xs Dt hs DtÞ þ Ps1 xs1 Dt þ Psþ1 hsþ1 Dt
Yj ði; kÞ ¼ ½m1 ; m2 ; …; mi1 ; yik ; miþ1 ; …; mm ðk ¼ 1; 2; j
(33)
¼ 1; 2; …; 2mÞ (28)
where, hs and xs are the emigration and immigration rates at time t.
Y2mþ1 ¼ ½m1 ; m2 ; …; mm (29) Emigration rate hk and immigration rate xk for k number of the
species can be calculated using the following equations:
Then deterministic MO-OPF problem is solved for every Yj vec-
tor, so MO-OPF is solved 2m þ 1 times and 2m þ 1 compromise Ek
hk ¼ (34)
solution vector of Z is obtained. Finally statistical moments of n
output variables of Z are calculated using weighting factors, by the
following expressions: k
xk ¼ I 1 (35)
n
X
m X
3
mn ¼ E½Zn ði; kÞ ¼ wik ðZn ði; kÞÞ (30) where, n is the total number of the species in the habitat, E and I are
i¼1 k¼1 the maximum value of emigration and immigration rates,
respectively.
h i Xm X
3 BBO algorithm uses migration and mutation mechanisms to
E Zn ði; kÞ2 ¼ wik Zn ði; kÞ2 (31) search in the feasible solutions.
i¼1 k¼1
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
h i rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi 3.1. Migration
sn ¼ E Zn ði; kÞ2 ðEðZn ÞÞ2 ¼ E Z 2n m2n (32)
In the migration mechanism, emigration and immigration rates
In these expressions, n is number of output variables. mn and sn are determine whether a solution will modify or not, which depend on
mean and standard deviation of nth output variable, respectively. other solutions. When a solution is selected to be modified, the
14 S. Shargh et al. / Renewable Energy 94 (2016) 10e21
3.2. Mutation
1. Define system data such as bus data, line data, load data, cost
and emission coefficient of generators and algorithm pa-
rameters such as number of iteration and habitats and so on.
2. Initialize habitats within the feasible solutions.
3. Check for equality constraints by running NR load flow pro-
gram for every habitat.
4. Calculate fitness of solutions (fuel cost, emission) known as
HSI.
5. Calculate emigration rate (h) and immigration rate (x) for
habitats.
6. Create new solutions using migration and mutation
strategies.
7. Verify feasibility of the solutions and run NR load flow
program.
8. Calculate HSI for new solutions.
9. Select the best solutions of the modified set of solutions and
the old set of solutions according to the keep rate for creating
the new set of habitats for the next iteration.
10. Go to step 5 and repeat following steps until max iteration
reach.
Table 1
Benchmark functiones [40,41].
where, r0ij and rij are the non-diagonal elements of the Cy and Cw, OPF problem Dou to nonlinear interaction among parameters
respectively. Note that diagonal elements of correlation matrix are may have nonlinear object function and constraints and may have
1. g is the coefficient of variation, that is g ¼ s/m. In Refs. [38,39], the non-convex feasible region. Dou to nonlinearity and non-convexity
authors provided 49 empirical formulas for calculating T among 10 of the OPF problem global optimal cannot be guaranteed. In order
different probability distributions. to investigate capability of the optimization algorithms benchmark
For the two parameter Weibull distribution (wind speed), T has functions are used. In this paper, In order to test the efficiency of the
the following formulation [38]: proposed BBO algorithm, five benchmarks which are given in
Table 1 are employed. Mean and standard deviation of 100 times
T g; r0ij ¼ 1:063 0:004r0ij 0:200 gi þ gj 0:001r20ij run for each of the benchmarks are presented in Table 2. These
þ 0:337 g2i þ g2j þ 0:007r0ij gi þ gj 0:007gi gj results are compared with results of genetic algorithm (GA) [40],
particle swarm optimization (PSO) [40] and group search optimizer
(40) (GSO) [40]. According to this table BBO converged to better result
with less standard deviation.
It should be mentioned that for normal distribution (load po-
The standard IEEE 30-bus test case is used to illustrate the
wer) T is 1. After calculating the correlation matrix Cw, Cholesky
proposed algorithm. This system has 30 buses, 41 transmission
decomposition is used to factor the Cw as follows:
lines, 6 generation units, 4 tap changing transformers and 9 shunt
capacitors. Bus and line data of the test case system can be found in
Cw ¼ L0 LT0 (41)
Ref. [42]. Coefficients used in the objective function of the emission
released from generation units can be found in Ref. [29].
Table 2
Generators cost and emission coefficients.
F. Method
Table 4
Wind turbine parameters.
Parameter Rating capacity (MW) Cut in speed (m/s) Cut out speed (m/s) Rating speed (m/s)
Table 5
Comparison of mean value for different wind speed correlation coefficients using PEM.
Table 6
Comparison of standard deviation for different wind speed correlation coefficients using PEM.
Matpower4.1 is used for load flow calculations. Voltage limit and tap 5.2. Probabilistic MO-OPF and correlation
of transformers' limits are considered in the range of [0.95,1.1] p.u.
and [0.9,1.1], respectively. Reactive power injected by shunt ca- In this paper, the active load demand is modeled as normal
pacitors is limited to 0.05 pu with the step size of 0.01 pu. distribution with mean value of nominal load at load points of
system and standard deviation equal to 5% of mean values [25]. The
power factor assumed to be 0.85 in load points and reactive power
5.1. Assessment of BBO algorithm in solving deterministic OPF demand varies corresponding to active power demand according to
problem the power factor [25]. Two wind farms including 4 wind turbines
were added to this system and connected to buses 29 and 30 both
In this section, the performance of BBO is evaluated in solving via a transmission line with the impedance of 0.01 þ j0.01 p.u.
deterministic OPF in introduced test system. In this case no uncer- Information of wind turbines are given in Table 4.
tainty and correlation is considered for sake of comparison of the In this paper the wind speed is approximated with two
proposed solution method with the available methods. In this case parameter Weibull distribution, the scale parameter is 7.28 and
fuel cost and released emission of generation units of system is shape parameter is 2.01 [25].
considered as objectives. Algorithm parameters which are used in In order to demonstrate effect of correlation of wind speed and
this study case consist of 20 habitats, 500 iteration, keeping rate of load, separately, on control variables, fuel cost and emission, two
0.4 and mutation limit 2% of variable limitation. These results are cases are defined.
tabulated in Table 3 in comparison of many other research results. In the first case, effect of wind speed on output variables is
This comparison demonstrates effectiveness of BBO algorithm in observed. In this case correlation among load points is considered
solving OPF problem.
Table 8
Table 7 Comparison of standard deviation for different load correlation coefficients using
Comparison of mean value for different load correlation coefficients using PEM. PEM.
PG1 (MW) 118.8492 118.5745 118.7102 PG1 (MW) 3.3998 5.1627 6.0256
PG2 (MW) 56.5677 56.8417 56.9682 PG2 (MW) 1.6317 2.4826 2.9083
PG5 (MW) 26.0050 26.9290 26.0275 PG5 (MW) 0.6526 0.9914 1.1499
PG8 (MW) 35.0000 35.0000 34.6831 PG8 (MW) 6.08e-8 3.23e-8 0.7086
PG11 (MW) 23.0233 23.5634 23.4872 PG11 (MW) 1.3955 2.1489 2.5233
PG13 (MW) 23.0233 22.9042 23.1068 PG13 (MW) 1.3087 2.0221 2.3728
Fuel cost ($/h) 804.7959 805.2633 805.4185 Fuel cost ($/h) 29.9023 45.7929 55.4595
Emission (ton/h) 0.251123 0.251077 0.251326 Emission (ton/h) 0.00315 0.00482 0.00568
S. Shargh et al. / Renewable Energy 94 (2016) 10e21 17
Fig. 3. Variation of normalized mean output value and control variables with corre-
Fig. 5. The convergence of Monte Carlo results for mean values.
lation change.
Table 9
Average error of moments for different correlation coefficients with 2000 runs of MCS.
6. Conclusion
Then errors for mean and standard deviation of generated power Appendix
of generation units, fuel cost and emission are calculated using these
equations. Table 9 presents the errors of POPF results for different The one line diagram of the IEEE 30 bus test system is shown in
amounts of correlation coefficients and also with no correlation. In Fig. 7. Bus data, branch data, generator data, tap setting data, shunt
this table, average error for generation units is presented. capacitor data and cost and emission coefficients of this network
Convergence characteristics of both mean and standard are presented in the following tables.
1 2 5 7 8
3 4 6 28
13 12 10 9 11 27
14 15 16 17 22 29 30
23 18 19 20 21 24 25
26
Bus No Bus type Real power (MW) Reactive power (MVar) Voltage magnitude (p.u.) Voltage angle (degree) Base voltage (KV)
1 3 0 0 1.0600 0 132.00
2 2 21.7000 12.7000 1.0430 5.4800 132.00
3 1 2.4000 1.2000 1.0210 7.9600 132.00
4 1 7.6000 1.6000 1.0120 9.6200 132.00
5 2 94.2000 19.0000 1.0100 14.3700 132.00
6 1 0 0 1.0100 11.3400 132.00
7 1 22.8000 10.9000 1.0020 13.1200 132.00
8 2 30.0000 30.0000 1.0100 12.1000 132.00
9 1 0 0 1.0510 14.3800 1.00
10 1 5.8000 2.0000 1.0450 15.9700 33.00
11 2 0 0 1.0820 14.3900 11.00
12 1 11.2000 7.5000 1.0570 15.2400 33.00
13 2 0 0 1.0710 15.2400 11.00
14 1 6.2000 1.6000 1.0420 16.1300 33.00
15 1 8.2000 2.5000 1.0380 16.2200 33.00
16 1 3.5000 1.8000 1.0450 15.8300 33.00
17 1 9.0000 5.8000 1.0400 16.1400 33.00
18 1 3.2000 0.9000 1.0280 16.8200 33.00
19 1 9.5000 3.4000 1.0260 17.0000 33.00
20 1 2.2000 0.7000 1.0300 16.8000 33.00
21 1 17.5000 11.2000 1.0330 16.4200 33.00
22 1 0 0 1.0330 16.4100 33.00
23 1 3.2000 1.6000 1.0270 16.6100 33.00
24 1 8.7000 6.7000 1.0210 16.7800 33.00
25 1 0 0 1.0170 16.3500 33.00
26 1 3.5000 2.3000 1.0000 16.7700 33.00
27 1 0 0 1.0230 15.8200 33.00
28 1 0 0 1.0070 11.9700 132.00
29 1 2.4000 0.9000 1.0030 17.0600 33.00
30 1 10.6000 1.9000 0.9920 17.9400 33.00
Table 11
Branch data of IEEE 30 bus test system.
Line No From bus To bus Resistance (p.u.) Reactance (p.u.) Susceptance (p.u.) MVA rating
Table 12
Generator data of IEEE 30 bus test system.
Real power (MW) Reactive power (MVar) Pmin (MW) Pmax (MW) Qmin (MVar) Qmax (MVar)
greece, Sol. Energy 30 (6) (1983) 497e505. [40] S. He, Q.H. Wu, J. Saunders, Group search optimizer: an optimization algo-
[32] A. Garcia, J. Torres, E. Prieto, A. De Francisco, Fitting wind speed distributions: rithm inspired by animal searching behavior, Evol. Comput. IEEE Trans. 13 (5)
a case study, Sol. Energy 62 (2) (1998) 139e144. (2009) 973e990.
[33] A. Rabiee, A. Soroudi, B. Mohammadi-ivatloo, M. Parniani, Corrective voltage [41] M. Moradi-Dalvand, B. Mohammadi-Ivatloo, A. Najafi, A. Rabiee, Continuous
control scheme considering demand response and stochastic wind power, quick group search optimizer for solving non-convex economic dispatch
IEEE Trans. Power Syst. 29 (6) (2014) 2965e2973. problems, Electr. Power Syst. Res. 93 (2012) 93e105.
[34] A. Soroudi, B. Mohammadi-Ivatloo, A. Rabiee, Energy hub management with [42] O. Alsac, B. Stott, Optimal load flow with steady-state security, Power App.
intermittent wind power, in: Large Scale Renewable Power Generation, Syst. IEEE Trans. 3 (1974) 745e751.
Springer, 2014, pp. 413e438. [43] M. Ghasemi, S. Ghavidel, M. Gitizadeh, E. Akbari, An improved teach-
[35] D. Simon, Biogeography-based optimization, IEEE Trans. Evol. Comput. 12 (6) ingelearning-based optimization algorithm using le vy mutation strategy for
(2008) 702e713. non-smooth optimal power flow, Int. J. Electr. Power & Energy Syst. 65 (2015)
[36] M. Rosenblatt, Remarks on a multivariate transformation, Ann. Math. Stat. 375e384.
(1952) 470e472. [44] M. Rezaei Adaryani, A. Karami, Artificial bee colony algorithm for solving
[37] H. Li, Z. Lü, X. Yuan, Nataf transformation based point estimate method, Chin. multi-objective optimal power flow problem, Int. J. Electr. Power Energy Syst.
Sci. Bull. 53 (17) (2008) 2586e2592. 53 (2013) 219e230.
[38] P.L. Liu, A. Der Kiureghian, Multivariate distribution models with prescribed [45] T. Niknam, M. Jabbari, A.R. Malekpour, et al., A modified shuffle frog leaping
marginals and covariances, Probab. Eng. Mech. 1 (2) (1986) 105e112. algorithm for multi-objective optimal power flow, Energy 36 (11) (2011)
[39] A. Der Kiureghian, P.L. Liu, Structural reliability under incomplete probability 6420e6432.
information, J. Eng. Mech. 112 (1) (1986) 85e104.