Sunteți pe pagina 1din 5

Optimal Load Shedding Scheme under Contingency

Condition Considering Voltage Stability Problem


Bundid Charoenphan and Kulyos Audomvongseree
Department of Electrical Engineering, Faculty of Engineering, Chulalongkorn University
E-mail: pungya_en@hotmail.com, kulyos.a@eng.chula.ac.th

Abstract optimization model to an approximated linearized model.


Rapidly increasing demands in power systems at present With the approximated linear optimization model, it can solve
may cause overloading of power system equipment. In some the load shedding problem under contingency condition with
situations, these overloading lead to severe blackout unless faster speed while retaining acceptable accuracy; therefore it
proper load shedding is performed in time. The load shedding is more suitable to be used in practice. In addition, rather than
scheme must be effective and optimal in order to protect the considering load as a continuous constant power demand, this
power system from voltage instability. This paper proposes a paper also considers load as a mix of discrete constant
new load shedding scheme under contingency condition impedance and constant power demand, as described by ZP
considering voltage stability problem. The proposed method model. A binary integer programming (BIP) algorithm is
is based on the linearized optimization method. Moreover, employed to handle the discrete load.
rather than considering load as a constant power demand, This paper is organized as follows. The load model is
more accurate model is used. The proposed method has been explained in section 2. The load shedding problem in
tested with the EGAT Thailand system. Satisfactory results linearized form is shown in section 3. The summary of the
have been obtained. proposed methodology is summarized in section 4. Numerical
example is illustrated in section 5. Finally, conclusion is
Keywords: Load Shedding, Voltage Stability Index, given in section 6.

1. Introduction 2. Load model


Generally, a huge power system may have high risk of The constant power load model is typically used in many
overloading of power system equipment. These overloading power system studies. However, the assumption behind this
lead to severe blackout unless proper load shedding is model is not true in practice because load demand varies with
performed in time. The load shedding scheme must be voltage and also frequency. Many researchers attempt to
effective and optimal in order to protect the power system introduce several load models [2-3] to improve accuracy of
from voltage instability. Currently, voltage stability problem power flow studies. One of interesting load models is the ZP
is one of the most important problems receiving high interest. model which decomposes load into two main components,
Methodologies for investigating and monitoring the voltage which are constant impedance (Z), and constant power (P).
stability problem have been developed by many researchers. The ZP model has following form:
One of the most popular methodologies is the continuation
power flow (CPF) [1] which uses a continuum of power flow P = P0 ⎡⎢a V + c ⎤⎥
2
(1)
solutions to find the voltage collapse point. However, the CPF ⎣ ⎦
Q = Q0 ⎡⎢ d V + f ⎤⎥
is not applicable in practice because of its computation time 2
(2)
⎣ ⎦
limitation, especially in the contingency condition. Another
method that can be used to observe the problem more rapidly
and is widely used in practice is voltage stability indices Parameters a, c, d, and f can be estimated from the
(VSI). There are many indices proposed in many research operating conditions pre- and post- contingency by following
works [5]. However, in this paper, an index call PQVSI [6], equations:
⎡ 2 ⎤
which has been proven as an accurate index among many ⎡ Ppre ⎤ ⎢ V pre 1⎥ ⎡a0 ⎤
others, is used. ⎢ ⎥=⎢ 2 ⎥⎢ ⎥ (3)
⎣⎢ Ppost ⎦⎥ ⎢ V post 1⎥ ⎣ c0 ⎦
T. Amraee, B. Mozafari, and A.M. Ranjbar [7] introduce an ⎣ ⎦
optimal load shedding scheme by a nonlinear optimization ⎡ 2 ⎤
⎡ Q pre ⎤ ⎢ V pre 1⎥ ⎡d 0 ⎤
model, considering long term voltage stability. Although, ⎢ ⎥=⎢ 2 ⎥⎢ ⎥ (4)
nonlinear optimization model can solve a problem accurately; ⎣⎢Q post ⎦⎥ ⎢ V post 1⎥ ⎣ f 0 ⎦
⎣ ⎦
however it is not appropriated to be used in practice because
of its computation time. Thus, this paper proposes a multi- P0 = a0 + c0 (5)
step linearization technique to transform the nonlinear Q0 = d 0 + f 0 (6)
⎡ 2 ⎤ The proposed load shedding scheme will be iteratively
⎡ Ppre ⎤ ⎢ P0 ⋅ V pre P0 ⎥ ⎡a ⎤
⎢ ⎥=⎢ 2 ⎥⎢ ⎥ (7) solved by a binary integer programming (BIP) based on
⎣⎢ Ppost ⎦⎥ ⎢ P0 ⋅ V post P0 ⎥ ⎣ c ⎦ branch and bound algorithm (LP-relaxation) [8] until the error
⎣ ⎦
⎡ 2 ⎤
is acceptable.
⎡ Q pre ⎤ ⎢ Q0 ⋅ V pre Q0 ⎥ ⎡ d ⎤
⎢ ⎥=⎢ 2 ⎥⎢ ⎥ (8)
⎣⎢Q post ⎦⎥ ⎢Q0 ⋅ V post Q0 ⎥ ⎣ f ⎦ 3.1 Linearization of Voltage Level Limit
⎣ ⎦
Relationships of bus voltage and complex power under ZP
load model can be expressed as follows:
where Ppre is the real power load before disturbance. Ppost
is the real power load after disturbance. Qpre is the reactive n
power load before disturbance. Qpost is the reactive power Pi = ∑ Vi V j Yij cos(θij − δ i + δ j ) (15)
load after disturbance. Vpre is the bus voltage before j =1

disturbance. Vpost is the bus voltage after disturbance. P0 is n


Qi = − ∑ Vi V j Yij sin(θij − δ i + δ j ) (16)
the real power load at 1 per unit voltage. And Q0 is the j =1
reactive power load at 1 per unit voltage.
PiL = P0Li ⎡ai Vi + ci ⎤
2
⎢⎣ ⎥⎦
(17)
3. Mathematical Formulation of the Proposed Load
QiL = Q0Li ⎡ di Vi + fi ⎤
2
Shedding Scheme (18)
⎢⎣ ⎥⎦
The linearized model of the proposed load shedding
scheme can be formulated as follows. In this paper, multi-step The linearized forms of (15) - (18) with respect to the state
load shedding scheme is developed. At each bus, load variables, Δδ , ΔV , ΔP0 , ΔQ0 , can be shown in (19)-(20).
demand is divided into many steps depending on the number
of transformers installed.
⎡ dP dP ⎤
⎡ ΔP ⎤ ⎢⎢ dδ dV ⎥ ⎡ Δδ ⎤
⎥⎢
⎛N ⎞ ⎢ ⎥ = ⎢ dQ ⎥ (19)
Min ⎜⎜ ∑ Ci PLi X i ⎟⎟ ⎣ ΔQ ⎦ dQ ⎥ ⎣Δ V ⎦
(9) ⎢ dδ ⎥
⎝ i =1 ⎠ ⎣ dV ⎦
Subject to ⎡ dPL dPL ⎤ ⎡ dPL dPL ⎤
N ⎢ ⎥ ⎢ ⎥
Vkmin ≤ Vk0 + ∑ [[Jcki ] + αi [Jdki ]][PLi X i ] ≤ Vkmax (10) ⎡ ΔP L ⎤ ⎢ dδ d V ⎥⎡ Δδ ⎤ ⎢ dPo
+
dQo ⎥ ⎡ ΔP0 ⎤ (20)
⎢ L⎥ =⎢ L ⎥⎢ ⎥ ⎢ ⎥
i =1
⎢⎣ΔQ ⎥⎦ ⎢ dQ dQL ⎥ ⎣Δ V ⎦ ⎢⎢ dQL dQL ⎥⎥ ⎣ΔQ0 ⎦
[[ ] [ ]]
N ⎢ dδ
Qg ≤ Q g0 + ∑ Gc gi + α i Gd gi [PLi X i ] ≤ Q gmax ⎣ d V ⎥⎦ ⎢⎣ dPo dQo ⎥⎦
min
(11)
i =1
N Because change of net injected power described in (19) can
Il0 + ∑ [[JIali ] + αi [JIbli ]][PLi X i ] ≤ Ilmax (12)
i =1 be decomposed as change from generation and load in (21).
N Assumed there is no change from generation; thus, we can
λl0 + ∑ [Jλli ][PLi X i ] ≤ λlmax (13) write the impact of load change to system voltage as in (22).
i =1

X i ∈ {0,1} (14) ⎡ ΔP ⎤ ⎡ ΔPG − ΔP L ⎤


⎢ ⎥=⎢ G (21)
L⎥
⎣ΔQ ⎦ ⎢⎣ΔQ − ΔQ ⎥⎦
where X i is a binary number representing a status
⎡ Δδ ⎤ ⎡ Ja Jb ⎤ ⎡ ΔP0 ⎤
ON/OFF of the transformer i . PLi is amount of base load at ⎢ ⎥=⎢ ⎥⎢ ⎥ (22)
the loading transformer i . Ci is value of load loss of ⎣ Δ V ⎦ ⎣ Jc Jd ⎦ ⎣ ΔQ0 ⎦
customers connected at transformer i . N is the number of −1
⎛ ⎡ dP ⎤ ⎡ dP L dP L ⎤ ⎞⎟ ⎡ dP L
⎜ dP dP L ⎤
transformers. g is number of generators. k is number of
⎡ Ja Jb ⎤ ⎜ ⎢ dδ ⎥ ⎢ dδ ⎥
d V ⎥⎟
⎢ ⎥
buses or substations; and l is number of transmission lines. where ⎢ ⎥ = − ⎜ ⎢⎢ dQ
dV ⎥ +⎢ ⎢ dPo dQo ⎥
⎥ ⎢ dQ L ⎥⎟ ⎢ dQ L
⎣ Jc Jd ⎦ ⎜ dQ dQ L ⎥ ⎟ dQ L ⎥⎥
There are four constraints focused in this paper. The first ⎜⎜ ⎢ dδ dV ⎥ ⎢ ⎟ ⎢
one is the voltage level limit presented in (10). The second ⎝⎣ ⎦ ⎢⎣ dδ d V ⎥⎦ ⎟

⎢⎣ dPo dQo ⎥⎦
one is the reactive power generation limit presented in (11).
Another one is the transmission line thermal limit or current It is assumed here that the load shedding will be
limit presented in (12). The other is the voltage stability limit implemented such that it will shed the real and reactive power
indicated by the PQVSI index [6] presented in (13). Detailed loads at fixed proportion as shown in (23). Hence (22) will be
equation of each linearized constraint will be shown next sub- reduced to (24).
section.
ΔQ0 system index. It lies between 0 and 1, and equal to 1 at the
=α (23)
ΔP0 voltage collapse point.
[Δ V ] = [[Jc ] + α [Jd ]][ΔP0 ] (24)
P ji0
λ = PQVSI = (32)
P jiNP
3.2 Linearization of Reactive Power Generation Limit
2
Linearization of reactive power generation limit will be Vi
performed by substitute (19)-(20) to (21) as shown in (25). P jiNP = (33)
2[Im{ AB * } tan θ ji − Re{ AB * } + A B sec θ ji ]

⎡ ΔPG ⎤ ⎡Ga Gb ⎤ ⎡ ΔP0 ⎤


⎢ ⎥=⎢ ⎥⎢ ⎥ (25) where A and B is the transmission line parameter in the
⎣ΔQG ⎦ ⎣Gc Gd ⎦ ⎣ΔQ0 ⎦ o
form of 2-port model, where A=1+ZYc, and B=Z. Pji is real
where power flow out of the receiving end at current operating
⎛ ⎡ dP ⎤ ⎡ dP L dP L ⎤ ⎞⎟
NP
⎜ dP ⎡ dP L dP L ⎤ point. Pji is real power flow out of the receiving end at
⎡Ga Gb ⎤ ⎜ ⎢⎢ dδ ⎥ ⎢ dδ ⎥ ⎢
d V ⎥ ⎟ ⎡ Ja Jb ⎤ ⎢ dPo

dV ⎥ +⎢ ⎟
dQo ⎥ voltage collapse point. Vi is sending end bus voltage, θji is
⎢ ⎥=⎜ ⎥ ⎢ dQ L
+
L ⎥ ⎢ Jc Jd ⎥ ⎢ L⎥
⎣Gc Gd ⎦ ⎜ ⎢ dQ dQ L
dQ ⎥ ⎟ ⎣ ⎦ ⎢ dQ dQ angle of complex power flow out of the receiving end at
⎜⎜ ⎢ dδ dV ⎥ ⎢ ⎟ ⎥
⎦ ⎢⎣ dδ d V ⎥⎦ ⎟ ⎢⎣ dPo dQo ⎥⎦ current operating point, Z is transmission line impedance,
⎝⎣ ⎠
and Yc is transmission line admittance.
In addition, it is assumed here that the load shedding will Because of the PQVSI index depends on the complex
be implemented such that it will shed the real and reactive power flow in transmission line, linearized form of this index
power loads at fixed proportion. Hence (25) will be reduced can be derived as follows:
to (26).
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
[ΔQG ] = [[Gc ] + α [Gd ]][ΔP0 ] [Δλ] = ⎢ ∂λ ⎥[ΔPji ] + ⎢ ∂λ ⎥[Δθ ji ] + ⎢ ∂λ ⎥[Δ Vi ] (34)
(26) ⎢⎣ ∂Pji ⎥⎦ ⎢⎣ ∂θ ji ⎥⎦ ⎣⎢ ∂ Vi ⎦⎥
2
Pij = Vi V j Yij cos(θij + δ j − δi ) − Vi Yij cos(θij ) (35)
3.3 Linearization of Current Limit
Relationships of the current flowing in a transmission line 2
Qij = −Vi Vj Yij sin(θij + δ j −δi ) + Vi Yij sin(θij ) − Vi Ysh
2
(36)
can be expressed as follows:
tan(θ ij ) = Qij / Pij (37)
Ireal = V j Yij cos(θij + δ j ) − Vi Yij cos(θij + δi ) + Vi Ysh cos(δi + 90) (27)
[ΔPij ] = ⎡⎢ ∂∂Pδij ∂Pij ⎤ ⎡ Δδ ⎤
⎥⎢ ⎥
∂ V ⎥⎦ ⎣Δ V ⎦
(38)
Iimag = V j Yij sin(θij + δ j ) − Vi Yij sin(θij + δi ) + Vi Ysh sin(δi + 90) (28) ⎣⎢
⎡ ∂(Qij / Pij ) ∂(Qij / Pij ) ⎤ ⎡ Δδ ⎤
I ij = I 2 real + I imag
2
(29) [ ]
sec2(θij ) Δθij = ⎢ ⎥⎢ ⎥
∂ V ⎦⎥ ⎣ΔV ⎦
(39)
⎣⎢ ∂δ

The linearized forms of (27) - (29) with respect to If it is assumed here that the load shedding will be
ΔP0 , ΔQ0 , can be shown in (30). implemented such that it will shed the real and reactive power
loads at fixed proportion. Hence (38)-(39) will be reduced to
⎡ ∂ I ij ⎤ ⎡ Ja Jb ⎤ ⎡ ΔP0 ⎤
[Δ Iij ]= ⎢⎢ ∂∂Iδij ⎥
⎢ ⎥⎢
∂ V ⎥ ⎣ Jc Jd ⎦ ⎣ΔQ0 ⎦
⎡ ΔP0 ⎤
⎥ = [JIa JIb]⎢ ⎥ (30) (40)-(41).
⎣ ΔQ0 ⎦
⎣ ⎦
[ΔPji ] = [[JPa] + α [JPb]][ΔP0 ] (40)
The load shedding will be implemented such that it will [Δθji ] = [[Jθa ] + α [Jθb ]][ΔP0 ] (41)
shed the real and reactive power loads at fixed proportion.
Hence (30) will be reduced to (31). ⎡ ∂Pij ∂Pij ⎤ ⎡Ja Jb⎤
where [JP] =[JPa JPb] = ⎢ ⎥⎢ ⎥
⎣⎢ ∂δ
[Δ I ij ] = [[JIa ] + α [JIb ]][ΔP0 ] (31)
∂ V ⎥⎦ ⎣ Jc Jd⎦

1 ⎡ ∂(Qij / Pij ) ∂(Qij / Pij ) ⎤ ⎡ Ja Jb⎤


3.4 Linearization of PQVSI index [Jθ ] =[Jθa Jθb] = ⎢ ⎥⎢ ⎥
In this paper, an index so called PQVSI [6] is used to sec (θij ) ⎣⎢ ∂δ
2 ∂V ⎦⎥ ⎣ Jc Jd⎦
measure how close the current operating condition to the
voltage instability. The mathematical formula of this index Substituting (24), (40) and (41) into (34), we will get the
can be shown in (32)-(33). This formula will be applied to all linearized form of PQVSI as follow:
transmission line. The maximum one is selected to be the
[Δλ ] = [Jλ ][ΔP0 ] (42) Table 1 Results obtained from the SQP method
where Bus Voltage before load Voltage after load
⎡⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎤ shed (pu) shed (pu)
[Jλ] = ⎢⎢ ∂λ ⎥[[JPa] + α[JPb]] + ⎢ ∂λ ⎥[[Jθa] + α[Jθb]] + ⎢ ∂λ ⎥[[Jc] + α[Jd]]⎥ PKC 0.9448 0.9518
⎢⎣⎢ ∂Pji ⎦⎥
⎣ ⎣⎢ ∂θ ji ⎦⎥ ⎣⎢ ∂ Vi ⎦⎥ ⎥
⎦ CBD 0.9481 0.9499
DBN 0.9499 0.9517
4. Proposed Methodology Generator QGmax-QG (Mvar) QGmax-QG (Mvar)
The proposed methodology of this paper can be TL3 -10.79 1.73
summarized in the following flowchart. Line VSI before load VSI after load shed
shed
Occurrence of TTK-CYP 0.7698 0.7033
contingency Bus Amounts of load at Load shed at bus
bus (MW) (MW)
CYP-115 kV 62.49 20.26
Define k = 1 CYP-230 kV 100.65 100.01
m = linearization
occurrence number Amounts of load shed 120.27

Table 2 Results of the case without considering discrete load


Calculating of ZP obtained from the Simplex method
load models (3)-(8)
Bus Voltage before load Voltage after load
shed (pu) shed (pu)
PKC 0.9448 0.9518
YES
CBD 0.9481 0.9499
Xn+1=Xn+X
Constraints are met ? DBN 0.9499 0.9517
k=k+1
or k > m ?
Generator QGmax-QG (Mvar) QGmax-QG (Mvar)
NO TL3 -10.79 1.73
Line VSI before load VSI after load shed
Determine amounts
and locations of load shed
Stop
Calculate power flow shedding (9) – (13) TTK-CYP 0.7698 0.7033
with ZP load model
Bus Amounts of load at Load shed at bus
bus (MW) (MW)
CYP-115 kV 62.49 20.96
Calculating of linearized
Extract bus which has low CYP-230 kV 100.65 99.32
sensitivity to contingency
parameters
parameters
Amounts of load shed 120.28
α , Jc, Jd , JIa, JIb, Jλ , Gc, Gd

Fig. 1 Flowchart of load shedding procedure


Table 3 Results of the case considering discrete load obtained
from the branch and bound method
5. Numerical Example Bus Voltage before load Voltage after load
The proposed method is applied to the EGAT Thailand
shed (pu) shed (pu)
system in peak load situation with one line contingency. For
PKC 0.9448 0.9523
the load model, we assume all system load comprise constant
CBD 0.9481 0.9501
impedance, constant current and constant power components;
DBN 0.9499 0.9518
and estimate them with the ZP models by using the fitting
equations (3)-(8). There are 3 cases to be investigated. First, Generator QGmax-QG (Mvar) QGmax-QG (Mvar)
result from the conventional nonlinear optimization method. TL3 -10.79 2.72
Another is the case using the linearized model but considering Line VSI before load VSI after load shed
load as continuous constant power demand. Other is the result shed
from the proposed method considering discrete ZP load TTK-CYP 0.7698 0.6975
model. The binary linear optimization model is used to solve Bus Amounts of load at Load shed at bus
the case of discrete load. For the nonlinear optimization bus (MW) (MW)
method, the Sequential Quadratic Programming (SQP) is CYP-115 kV 62.49 30.64
used. Three system limits comprising 0.95 ≤ Vk ≤ 1.1 , CYP-230 kV 100.65 100.65
Q g min ≤ Q g ≤ Q g max , I l ≤ I max , and λl ≤ 0.7 are of interest. Amounts of load shed 131.29
Table 4 Computation time of each case References
Computation time EGAT Thailand system [1] V. Ajjarapu, C. Christy, “The Continuation Power Flow:
SQP 3 hours A Tool for Steady State Voltage Stability Analysis”, IEEE
Simplex 6 seconds Transactions on Power Systems, Vol. 7, No. 1, February
Binary programming 16 seconds 1992.
[2] IEEE Task Force on Load Representation for Dynamic
It can be seen that load shedding results obtained from the Performance, “Standard Load Models for Power Flow and
nonlinear optimization and the approximated linearization Dynamic Performance Simulation”, IEEE Trans. Power
shown in Table 1 and 2 are close to each other. It means that Syst., vol. 10, No. 3, pp. 1302-1313, Aug 1995.
the error from linearization is acceptable when we apply the [3] Y. Li, H.D. Chiang, B.K. Choi, Y.T. Chen, D.H. Huang,
iterative linearization method. However, computation times of M.G. Lauby, “Representative static load models for
both methods are far different. It should be noted again that transient stability analysis: development and
both load shedding schemes consider load demand as pure examination”, IET Gener. Transm. Distrib. 2007, 1, (3),
constant power. This load model error causes some pp. 422-431.
constraints violation after load shedding, especially the [4] G. Dantzig, A. Orden, and P. Wolfe, "Generalized
voltage stability limit as shown in the tables. Thus, more Simplex Method for Minimizing a Linear from Under
accurate load model is needed. Moreover, although, the linear Linear Inequality Constraints," Pacific Journal Math., Vol.
optimization can solve the problem with satisfactory 5, pp. 183–195.
computation time, it is still not applicable in practice. It is [5] M.V. Suganvadevi, C.K. Babulal, “Estimating of
because, indeed, load at each bus is discrete rather than Loadability Margin of a Power System by comparing
continuous. The proposed method considering ZP load model Voltage Stability Indices”, International Conference on
and discrete load is applied to the same problem. The Control Automation Communication and Energy
obtained results are shown in Table 3. It can be seen that with Conservation 2009.
this proposed method all system constrained after load [6] P. Tiabrat, V. Yokyong, K. Audomvongseree, “PQ
shedding are kept within their limits. However, the amount of Voltage Stability Index (PQVSI) for Voltage Stability
load shed in this case is higher. This is because, in practice, Analysis”, EE Conference 2008.
we have to shed all loads connected at a transformer, not only [7] T. Amraee, B. Mozafari, and A.M. Ranjbar, “An
some parts as done in the previous 2 cases. Improved Model for Optimal Under Voltage Load
Shedding: Particle Swarm Approach”, Power India
6. Conclusion Conference, 2006 IEEE.
This paper proposes the multi-step linearization method to [8] Frederick S. Hillier, Gerald J. Lieberman, “Introduction to
solve the load shedding problem. There are four system Operations Research”, McGraw-Hill, Inc., 2001.
constraints focused in this method, the voltage level limit, the [9] Stephen G. Nash, Ariela Sofer, “Linear and Nonlinear
reactive power generation limit, the transmission line thermal Programming”, McGraw-Hill, Inc., 1996.
limit, and the voltage stability limit. The voltage stability
limit is measured through the PQVSI index which has been Biographies
proven as one of the most accurate index. In addition, a
discrete ZP load model which is more suitable in practice is Bundid Charoenphan received the B.Eng
used. The optimization technique used for solving this from Khonkean University, Thailand, 2008.
At present, he is pursuing the M.Eng at
proposed model is the binary programming based on branch
Chulalongkorn University. He is interested in
and bound algorithm. With this proposed method, the load
power system planning and operation.
shedding solution can be obtained with faster computation
time compared to the conventional nonlinear optimization
Kulyos Audomvongseree received the B.Eng
method. The proposed method is applied to the EGAT and M.Eng from Chulalongkorn University,
Thailand system. Satisfactory results were obtained. Thailand in 1998 and 2000 respectively. He
obtained a Ph.D from University of Tokyo,
Japan in 2004. He is now an assistant
professor at Chulalongkorn University.

S-ar putea să vă placă și