You are on page 1of 10

IET Generation, Transmission & Distribution

Research Article

Mixed integer linear formulation for ISSN 1751-8687


Received on 14th July 2017
Revised 10th December 2017
undervoltage load shedding to provide Accepted on 26th January 2018
E-First on 7th March 2018
voltage stability doi: 10.1049/iet-gtd.2017.1118
www.ietdl.org

Masoud Javadi1, Turaj Amraee1


1Faculty of Electrical Engineering, K. N. Toosi University of Technology, Tehran, Iran
E-mail: amraee@kntu.ac.ir

Abstract: Undervoltage load shedding (UVLS) is the last resort against voltage instability in critical situations. Here, the UVLS
scheme is formulated as a mixed integer programming (MIP) model. The aim of the proposed UVLS model is to provide a
predetermined value of loading margin with minimum amount of load shed. The full non-linear AC power flow equations are
linearised using a piecewise linear technique. The proposed linear AC power flow is then integrated into the UVLS problem. A
ZIP load model is utilised to demonstrate the fulfilment of the MIP-based UVLS model under non-linear static loads. The efficacy
of the developed linear AC power flow is verified under different operational conditions and contingencies. The results of the
proposed MIP-based UVLS model are compared with the original non-linear programming UVLS formulation. To verify the
performance of the developed load shedding strategy, the proposed MIP-based UVLS model is implemented in IEEE 14-bus
and IEEE 118-bus test systems.

^
 Nomenclature δ voltage angle approximation
^+
Indices and subscripts δ positive voltage angle approximation
NS number of line segments for cosine approximation
i, j index of buses
k index of piecewise segment in linear AC power flow model Sk+ slope of the kth line segment located in the positive
argument of cosine function
c subscript for critical operating point
0 subscript for normal operating point dk+ value of the kth segment of positive voltage angle
approximation
M auxiliary variable used in the proposed linear model
Sets NS′ number of initial segments in proximity of zero point
ΩB set of all buses in the proposed linear model
ΩG set of all voltage-controlled nodes y approximation of cosine function
ΩL set of all load buses W+ binary decision variable indicating the sign of voltage
angle difference
ΩTL set of all transmission lines
Uk+ binary decision variable for the kth segment of
positive voltage angle approximation
Parameters x+ / x− auxiliary variables used in linearisation of positive/
negative voltage angle approximation
NB number of buses
λLIB LIB-based loading margin
NG number of voltage-controlled nodes
NL number of load buses λSNB SNB-based loading margin
ε̄ mean absolute error of output variables
NTL number of transmission lines
λmin predefined minimum loading margin
B susceptance matrix 1 Introduction
Bsh shunt susceptance matrix Voltage instability has been one of the major concerns in power
G conductance matrix system operation due to several reported power outages related to
φ power factor angle this phenomenon [1]. Power systems must be able to maintain
δmax predefined upper bound of voltage angle differences steady voltages following large disturbances including system
ω weighting factor for prioritising load shedding faults, loss of generating unit, line outage, and small disturbances
such as incremental changes in load demand. In normal operating
conditions, all voltage magnitudes must remain within the
Variables
allowable safe range. However, in critical conditions (e.g. under
Pd /Qd active/reactive load demand severe contingencies), voltage magnitudes may experience large
Pdshed /Qdshed amount of active/reactive load shedding and progressive falls. This phenomenon is referred to as voltage
instability and may result in progressive and uncontrolled voltage
Pdshed, max maximum amount of allowable load shedding
decline or voltage collapse. Serious consequences of voltage
Pg /Qg active/reactive power generation collapse are loss of load, tripping of transmission lines, and
Pgmin /Pgmax minimum/maximum active power generation of generating units due to activation of related protective relays [2, 3].
generating units Conventional protective relays may fail to stop the partial or
PL, i j /QL, i j active/reactive power flows across transmission line widespread voltage collapses. Furthermore, system conditions may
between nodes i and j deteriorate with some local protective schemes due to undesired
v∠δ voltage phasor at critical point tripping [4].
vmin /vmax minimum/maximum of voltage magnitude

IET Gener. Transm. Distrib., 2018, Vol. 12 Iss. 9, pp. 2095-2104 2095
© The Institution of Engineering and Technology 2018
In order to avoid voltage collapse, many remedial actions such In Section 3, the simulation results of the developed mixed integer
as shunt compensation, secondary voltage controls, or load tap programming (MIP)-based algorithm are discussed under different
changer blocking can be taken [5]. However, in some conditions, loading conditions and topological changes. Moreover, in this
these controls may not be effective enough to prevent voltage section, the performance of the proposed linearised UVLS is
collapse. Undervoltage load shedding (UVLS) scheme can be compared with the conventional non-linear UVLS plan. Finally, the
executed as the last resort to prevent voltage instability. paper is concluded in Section 4.
UVLS relays are designed to shed a predetermined amount of
load at each bus under undesired voltage declines. UVLS plan 2 UVLS formulation
could be designed either to restore the equilibrium point or to
provide a predetermined voltage stability margin. In [6], authors In this paper, the loading margin (LM) is utilised as the voltage
have presented a method for load shedding with considering risk of stability criterion. This criterion represents the proximity of current
voltage instability. In [7], two UVLS strategies have been loading point to the voltage collapse point. For a particular
developed considering dynamic load characteristics. In [8], operating point, the LM is defined as the amount of additional load
emergency secondary voltage control and load shedding approach in a specific load increase pattern that would cause voltage
has been combined as an optimisation model which has resulted in collapse. LM can be calculated using an optimisation problem
reduction in the overall curtailed load with lower risk of system either considering reactive power limits of generators or not. The
blackout. In [9], a distributed robust UVLS plan has been presented bifurcation with and without considering reactive power limits of
in which a set of controllers are used to shed load. A load reduction generators are referred to as limit-induced bifurcation (LIB) and
optimisation with relaxed constraints is proposed in [10]. In this saddle node bifurcation (SNB), respectively [23, 24].
scheme, the violation of voltage and power transmission limits is The overall structure of the proposed load shedding strategy is
allowable for a very short time to minimise the total amount of illustrated in Fig. 1. Based on this scheme, in the case of significant
load shedding. In [11], an adaptive multistage load shedding reduction in voltage stability margin or disappearance of steady-
scheme has been developed using model predictive control. To state equilibrium point of the network, the UVLS plan is executed
solve the UVLS problem, a particle swarm optimisation algorithm as the last resort to normalise the critical conditions. Two LIB or
has been proposed in [12], with considering sensitivity analysis of SNB formulations may be utilised for calculating voltage stability
voltage stability margin at the maximum loading point. margin. The MILP-based and NLP-based UVLS models can be
Recently, the uncertainties of load and generation have been optimised considering either LIB or SNB formulations.
considered in UVLS design. To handle the load uncertainty, a The objective function of UVLS problem is expressed as
probabilistic UVLS plan is introduced in [13] which is expressed follows:
as a non-linear programming (NLP) problem. Owing to the non-
NL
linearity of the AC power flow formulation, the UVLS model is
inherently a mixed integer non-linear programming (MINLP) Min OF = ∑ ωi × Pdished (1)
i=1
problem. In previously proposed UVLS models, no effort has been
done to simplify the MINLP-based UVLS model. To solve non-
linear steady-state power system studies, various linearisation The goal of this problem is to minimise the total load curtailment
techniques have been suggested in previous researches. The with considering load priorities. This objective function is subject
utilisation of cutting planes and Taylor series expansions has to various constraints such as AC power flow equations, load
already been discussed in [14–16] to linearise the optimal power shedding limits, operational constraints, and voltage stability limit.
flow (OPF) problem. As discussed in [15], there are three types of For the sake of simplicity, in this paper, it is assumed that all load
power flow formulations: the polar power–voltage power flow points have the same priority. The constraints of non-linear UVLS
formulation, the rectangular power–voltage power flow are described below.
formulation, and the rectangular current injection formulation.
Recently, a successive linear programming (SLP) formulation of 2.1 Non-linear UVLS model
AC OPF has been proposed in [17]. This approach uses bus voltage The first group of constraints in UVLS model is the AC power
and current injections in IV formulation of AC power flow. The balance equations as given below
error of the proposed linear approximation in [17] depends
significantly on the initial point provided by the previous iteration shed
and step size of the algorithm. The advantages and disadvantages 1 + λmin Pgi
0 0
− Pdi + Pdi
of the IV power flow formulations have been discussed in [18]. NB (2)
According to [18], polar coordinates are more advantageous than = vic × ∑ vcj (Gi jcos(δicj) + Bi jsin(δicj)), i ∈ ΩB
rectangular coordinates. Also, compared with rectangular j=1
formulation, polar formulation of power flow equations has fewer
c shed
number of equations. Therefore, this may improve the efficacy of Qgi − 1 + λmin Qdi
0
− Qdi
the power flow model, especially in large-scale power systems. NB
Moreover, the rectangular formulation is also non-convex and the (3)
number of local optima in rectangular coordinates is the same as
= vic × ∑ vcj (Gi jsin(δicj) − Bi jcos(δicj)), i ∈ ΩB
j=1
the number of local solutions in polar coordinates [19, 20]. In [21],
a linear approximation of AC power flow equations is introduced It is noted that the power balance equations are expressed at the
using Taylor's expansion of line flows at normal operating point. In critical bifurcation point.
[22], an MILP-based model of OPF has been developed. However, The second group of constraints includes the operational limits
this method introduces too many binary variables and increases the on voltage magnitudes, active power generations, reactive power
complexity of the method. generations or absorptions, and power flows of transmission lines:
Non-linear UVLS models may give local optimum. In this
paper, an efficient linear UVLS model is developed in order to find
the global optimum location and amount of load shedding. A linear vic, min ≤ vic ≤ vic, max, i ∈ ΩB (4)
AC power flow model is developed and is then integrated in UVLS
c, min c c, max
scheme. The optimal solution of the proposed MILP-based UVLS Qgi ≤ Qgi ≤ Qgi , i ∈ ΩG (5)
model is obtained using CPLEX solver. Also, the efficacy of the
load shedding strategy is investigated in the presence of non-linear c . min c c . max
Pgi ≤ Pgi ≤ Pgi , i ∈ ΩG (6)
loads.
The present paper is organised as follows. In Section 2, details
PLc , i j ≤ PLc,,max
ij , i j ∈ ΩTL (7)
of the NLP-based and the proposed MILP-based UVLS models
including the linearised AC power flow formulation are introduced.

2096 IET Gener. Transm. Distrib., 2018, Vol. 12 Iss. 9, pp. 2095-2104
© The Institution of Engineering and Technology 2018
shed
1 + λmin Pgi
0 0
− Pdi + Pdi
NB (12)
∑ (Gi j(vic + vcj + cos(δi j) − 2) + Bi jδi j),
^ ^
= i ∈ ΩB
j=1

c shed
Qgi − 1 + λmin Qdi
0
− Qdi
NB (13)
∑ (Gi jδi j − Bi j(vic + vcj + cos(δi j) − 2)),
^ ^
= i ∈ ΩB
j=1

PLc . i j = − 2vic − 1 Gi j
^ ^ (14)
+Gi j vic + vcj + cos(δi j) − 2 + Bi jδi j, i j ∈ ΩTL

QLc . i j = − 2vic − 1 (Bishj − Bi j)


^ ^ (15)
+Gi jδi j − Bi j vic + vcj + cos(δi j) − 2 , i j ∈ ΩTL

^
Owing to the cosine function (i.e. cos(δi j)), the above formulation
^
is still non-linear. For simplicity, yi j is substituted for cos(δi j) and is
approximated by a piecewise linear function consisting of several
line segments as shown in Fig. 2a. The number of line segments
may be selected according to the required precision. It should be
noted that due to the low slope of the cosine function around zero
^
(i.e. δi j = 0), for better and more accurate approximation, the
length of line segments around this point (i.e. NS′ initial segments)
must be lower than other segments (e.g. half the length of other
segments).
For an NS segments approximation, the cosine term is
approximated via (16)–(29).
The sign of voltage angle differences is determined using (16).
Owing to the evenness of cosine term, this function is divided into
^+ ^+
two parts with positive (i.e. W i+jδi j) and negative (i.e. −(1 − W i+j)δi j)
arguments. A binary variable (i.e. W i+j) is defined to select either
the positive or the negative part, where W i+j = 1 corresponds to the
positive (i.e. right side) and W i+j = 0 corresponds to the negative
(i.e. left side) as given in the below equation
^ ^+ ^+
δi j = W i+jδi j − (1 − W i+j)δi j (16)

As illustrated in Fig. 2b, by symmetry, the piecewise linear


approximation of the positive part of cosine function is sufficient
Fig. 1  Overall structure of the proposed algorithm for modelling the whole function. From this point on, the right-side
function of the cosine function is modelled.
2
PLc . i j = − vic Gi j + vicvcj (Gi jcos(δicj) + Bi jsin(δicj)), ij ^+
According to (17), the variable δi j is divided into Ns /2 parts
(8)
∈ ΩTL
^+
2
δi j = di+j, 1 + di+j, 2 + ⋯ + di+j, (NS /2) (17)
QLc . i j = − vic (Bishj − Bi j)
(9)
+vicvcj Gi jsin(δicj) − Bi jcos(δicj) , i j ∈ ΩTL The group of constraints given by (18)–(21) represents the low
slope part of cosine function (i.e. the first NS′ segments). It is noted
The amount of load shedding at each load point is constrained as that due to the priority considered, di+j, k does not take on value as
given by (10). Furthermore, according to (11), it is assumed that long as its prior segment (i.e. di+j, k − 1) has not reached its upper
the power factors of the load demands remain constant during load bound. If the final solution is located in the kth segment, then the
shedding variables of the previous segments have reached their upper bounds
and the corresponding variables of the subsequent segments must
shed shed, max be zero. It is clear that the variable of first segment is always active
0 ≤ Pdi ≤ Pdi , i ∈ ΩL (10)
and its binary decision variable is fixed at 1 as given in the below
shed shed equation
Qdi = Pdi × tan(φi), i ∈ ΩL (11)
δmax + δmax
2.2 Linearised UVLS model U ≤ di+j, 1 ≤ (18)
2M i j, 2 2M
In this section, the AC power flow equations are linearised using a
piecewise linear technique. Power flow equations can be expanded According to (18), two conditions are possible: (i) the
(i.e. the first-order approximation of Taylor's expansion) around the approximated angle is located in the first segment. Therefore, di+j, 2
normal operating point (i.e. v = 1 p . u ., δ = 0∘) as follows: and subsequent corresponding variables must be zero, and (ii) the
approximated angle is not located in the first segment. Therefore,
the approximated angle is located in segment 2 or subsequent

IET Gener. Transm. Distrib., 2018, Vol. 12 Iss. 9, pp. 2095-2104 2097
© The Institution of Engineering and Technology 2018
represents the remaining segments of cosine function with upper
bound δmax /M

δmax + δmax + NS
U ≤ di+j, k ≤ U , k = NS′ + 1, …, − 1 (22)
M i j, k + 1 M i j, k 2

δmax +
0 ≤ di+j, (NS /2) ≤ U (23)
M i j, (NS /2)
^
Owing to Uδ term in (16), the problem is still non-linear. Let xi+j
and xi−j be the products of binary decision variables W i+j and
^+
(1 − W i+j) and bounded continuous variable δi j, respectively. To this
^+ ^+
end, W i+jδi j and (1 − W i+j)δi j are represented by new variables xi+j and
xi−j, respectively. The following constraints are the linearised
equivalents of (16) using the technique introduced in [25]
^
δi . j = xi+j − xi−j (24)

0 ≤ xi+j ≤ W i+j (25)

^+ ^+
δi j − (1 − W i+j) ≤ xi+j ≤ δi j (26)

0 ≤ xi−j ≤ 1 − W i+j (27)

^+ ^+
δi j − W i+j ≤ xi−j ≤ δi j (28)

It can be seen that if W i+j = 0, then according to (25), xi+j must be


^+
zero and according to (28), xi−j must be equal to δi j. Similarly, if
W i+j = 1, then according to (27), xi−j must be zero and according to
^+
(26), xi+j must be equal to δi j.
Finally, the approximated cosine function in power flow
equations equals to the summation of its linear segments as given
in (29). According to (29), the value of cosine function at zero
point is equal to 1
(NS /2)
yi j = 1 + ∑ Sk+di+j, k (29)
k=1

Fig. 2  Piecewise linear approximation of cosine function By increasing the number of segments (e.g. 8 segments or more as
(a) Approximation of cosine function with 16 lines, (b) Piecewise approximation of will be discussed in later sections), the accuracy of the AC power
the right side of cosine function flow approximation improves significantly. The lowest possible
amount of load shedding may be achieved using the MILP
segments. As a result, the variable of the first segment has reached formulation. These features make this method suitable for medium-
to its upper limit and the variable of the second segment or or large-scale power system problems.
subsequent segments (if needed) are activated.
This idea is generalised to the subsequent segments as given in 2.3 Voltage-dependent load modelling
the below equation
Load modelling is of great importance in realistic power system
δmax + δmax + analysis. Different load sectors such as industrial, commercial,
Ui j, k + 1 ≤ di+j, k ≤ U , k = 2, 3, …, NS′ (19)
2M 2M i j, k residential, and agricultural have distinct dependencies on voltage
and frequency. Moreover, each load sector comprises different load
where M is an auxiliary variable and is defined as follows: components such as lighting, heating, industrial motors,
agricultural pumps etc. Actual loads are mixtures of different load
NS NS components which must be considered in load modelling [26].
M= − (20) In this paper, a ZIP model is used to model non-linear
2 8
dependencies of loads on voltage deviations. As given by (30) and
It is noted that NS′ refers to the number of initial segments with (31), in the ZIP model, active and reactive load demands consist of
lower length and is defined as follows: three components including constant impedance (i.e. Pd = f (v2)),
constant current (i.e. Pd = f (v)), and constant power (i.e. Pd = cte),
NS where cte is a constant value
NS′ = 2 × (21)
8
Pd v v 2
= a0 + a1 + a2 (30)
In (20)–(21), ∙ represents the floor function. It should be noted P0 v0 v0
that NS must be an even number to satisfy the proposed equations
and constraints. The second group of constraints (i.e. (22)–(23)) Qd v v 2
= b0 + b1 + b2 (31)
Q0 v0 v0

2098 IET Gener. Transm. Distrib., 2018, Vol. 12 Iss. 9, pp. 2095-2104
© The Institution of Engineering and Technology 2018
Table 1 Errors of the proposed MIP AC power flow model in the base case point
Case study ε̄v, p . u . ε̄Qg, p . u . ε̄PL, i j, p . u . ε̄QL, i j, p . u .
IEEE 14-bus 0.0008 0.0114 0.0020 0.0031
IEEE 30-bus 0.0004 0.0024 0.0006 0.0005
IEEE 39-bus 0.0007 0.0542 0.0085 0.0127
IEEE 57-bus 0.0095 0.0791 0.0044 0.0163
IEEE 118-bus 0.0002 0.0175 0.0080 0.0054
IEEE 300-bus 0.0017 0.0687 0.0158 0.0197

Table 2 Errors of the proposed MIP AC power flow model in proximity of saddle node point
Case study ε̄v, p . u . ε̄Qg, p . u . ε̄PL, i j, p . u . ε̄QL, i j, p . u .
IEEE 14-bus 0.0009 0.0132 0.0055 0.0038
IEEE 30-bus 0.0019 0.0151 0.0045 0.0028
IEEE 57-bus 0.0095 0.0762 0.0047 0.0156
IEEE 118-bus 0.0004 0.0199 0.0100 0.0063
IEEE 300-bus 0.0022 0.0819 0.0179 0.0226

Fig. 3  Voltage profile of IEEE 118-bus system in normal operating condition

In this model, a0, a1, a2, b0, b1, and b2 are load model parameters. P0 synchronous generators, and active and reactive flow of
and Q0 are nominal values of active and reactive load demands at transmission lines) are calculated. In the MIP model, 32 segments
nominal voltage magnitude (i.e. v = v0) [27, 28]. are used for piecewise linear approximation.
The non-linear load models given in (30) and (31) can be The mean absolute errors (MAE) of voltage magnitudes,
linearised around the nominal operating point by the Taylor reactive powers, and active/reactive flows are calculated as
expansion as follows: follows:
NL
Pd a 1
NL i∑
= a0 + a2(1 − 2v0) + 1 + 2a2 v (32) ε̄v p . u . = viMIP − viNLP , i ∈ ΩL (34)
P0 v0
=1

Qd b NG
= b0 + b2(1 − 2v0) + 1 + 2b2 v (33) 1
NG i∑
Q0 v0 ε̄Qg p . u . = MIP
Qgi NLP
− Qgi , i ∈ ΩG (35)
=1

3 Simulation results 2NTL


1
2NTL i∑
The simulation results are presented in two different parts. In the ε̄PL . i j p . u . = PLMIP NLP
, i j − PL, i j , i j ∈ ΩTL (36)
first part, the accuracy of the proposed linear AC power flow is =1
investigated. Different IEEE test systems are utilised and the
accuracy of the developed linear AC power flow is compared to the 2NTL
1
2NTL i∑
full non-linear AC power flow under different loading conditions ε̄QL . i j p . u . = QLMIP NLP
, i j − QL, i j , i j ∈ ΩTL (37)
and contingencies. In the second part, the results of the proposed =1
linear UVLS scheme are presented. Also, in this section, the
efficacy of the linearised UVLS model is compared with non-linear The resulted accuracies using the developed linear AC power flow
UVLS model. The data of these test cases could be found in [29]. at base case operating point have been reported in Table 1. The
worst errors among all simulated test cases, for voltage
magnitudes, reactive power generations, and active and reactive
3.1 Accuracy of the proposed MIP AC power flow
power flows are 0.0095, 0.0791, 0.0158, and 0.0197, respectively.
In this section, the accuracy of the proposed MIP model is The AC power flow equations near the collapse point (i.e. near the
compared with the results of the conventional full non-linear AC maximum LM) are highly non-linear. The accuracies of the
power flow model. To this end, the errors of the outputs (i.e. proposed linear AC power flow at such stressed operating point
voltage magnitude of load buses, reactive power generation of have been reported in Table 2. It can be seen that the resulted errors
in proximity of voltage collapse point still remain in a relatively
IET Gener. Transm. Distrib., 2018, Vol. 12 Iss. 9, pp. 2095-2104 2099
© The Institution of Engineering and Technology 2018
Fig. 4  Reactive power generation of IEEE 118-bus system in the normal operating condition

Table 3 Errors of the linear AC power flow versus number of piecewise segments for IEEE 14-bus and IEEE 118-bus
Case study NS ε̄v, p . u . ε̄Qg, p . u . ε̄PL, i j, p . u . ε̄QL, i j, p . u .
IEEE 14-bus 1 0.0028 0.1120 0.0083 0.0069
8 0.0031 0.0262 0.0079 0.0055
12 0.0018 0.0153 0.0076 0.0051
16 0.0014 0.0114 0.0076 0.0050
32 0.0008 0.0114 0.0020 0.0031
IEEE 118-bus 1 0.0019 0.1458 0.0357 0.0345
8 0.0010 0.0470 0.0170 0.0121
12 0.0007 0.0234 0.0086 0.0066
16 0.0002 0.0174 0.0075 0.0055
32 0.0002 0.0175 0.0080 0.0054

Table 4 Errors of the proposed linear AC power flow under [31]. The NLP-based UVLS model is solved using CONOPT
important N − 1 contingencies for IEEE-118 bus system solver [32]. In the following, two different severe N − 2
Contingency ε̄v, p . u . ε̄Qg, p . u . ε̄PL, i j, p . u . ε̄QL, i j, p . u . contingencies are applied to both test systems. The assumed
disturbance for IEEE 14-bus system is the simultaneous outage of
generator (bus: 10) 0.0008 0.0339 0.0137 0.0104
lines 1–2 and 2–4. The input contingency for IEEE 118-bus test
generator (bus: 66) 0.0002 0.0178 0.0088 0.0056 system is the simultaneous outage of generator 5 (i.e. generator
generator (bus: 89) 0.0005 0.0229 0.0118 0.0080 located at bus 10) and lines 8–30 (i.e. line outage due to resulted
line 26–30 0.0009 0.0282 0.0130 0.0115 overload). Owing to these contingencies, the steady-state operating
line 4–5 0.0006 0.0233 0.0107 0.0089 points of both networks disappear (i.e. the LM will be negative).
line 100–103 0.0009 0.0305 0.0139 0.0121 According to NERC (North American Electric Reliability
line 25–27 0.0004 0.0189 0.0087 0.0071
Corporation) reliability standard [33], in the case of events
resulting in the loss of two or more elements, load shedding is
line 8–30 0.0009 0.0287 0.0127 0.0112
allowed to restore the secure operating conditions.
In this section, the MIP-based UVLS scheme is applied to the
test systems using 32 line segments for linear AC power flow.
acceptable range. As an example, the voltage profile and reactive In IEEE 14-bus system, it is assumed that the system loading is
power generations of IEEE 118-bus system near the SNB have at a heavy loading condition (i.e. 125% of base case loading).
been illustrated in Figs. 3 and 4, respectively. It can be seen that the Following the outages of lines 1–2 and 2–4 in this system, the LM
obtained results using the linear AC power flow are very close to
changes from λLIB = 0067 p . u . and λSNB = 0121 p . u . in pre-
the results of full non-linear AC power flow. According to Table 3,
although by increasing the number of segments for cosine contingency normal operating condition to λLIB = − 0323 p . u .
approximation, the accuracy is promoted; however, it can no longer and λSNB = − 0028 p . u . in post-contingency critical condition,
improve the accuracy for higher number of segments. respectively. A negative LM indicates that there is no steady-state
The accuracy of the developed linear AC power flow model is equilibrium point after the contingencies. The obtained results are
now verified under important N − 1 contingencies (i.e. single given in Table 5 and Fig. 5. According to Table 5, it can be seen
outages). The accuracies of the proposed model under important that the proposed MIP-based UVLS model gives a lower amount of
contingencies have been reported in Table 4, while the loading is total load shedding. This verifies the optimality of the proposed
assumed to be at the base case operating point. According to MIP-based load shedding compared with the NLP-based UVLS
Table 4, it can be seen that even in the presence of contingencies, model.
the proposed model keeps its accuracy. As depicted in Fig. 5, for IEEE-14 bus system, there is a
significant difference between the non-linear and linear UVLS
3.2 Results of the proposed UVLS scheme methods in terms of locations and amount of load to be shed.
For IEEE 118-bus test system, due to the outages of generator 5
The proposed linearised UVLS model is implemented in IEEE 14- and lines 8–30, the LM decreases from λLIB = 0277 p . u . and
bus and IEEE 118-bus test systems. The obtained results are then
compared with the non-linear UVLS model in order to verify the λSNB = 0278 p . u . to λLIB = − 0196 p . u . and
SNB
performance of the MIP-based UVLS model. The MIP-based λ = − 0193 p . u ., respectively.
UVLS scheme is solved using the CPLEX solver [30] in GAMS

2100 IET Gener. Transm. Distrib., 2018, Vol. 12 Iss. 9, pp. 2095-2104
© The Institution of Engineering and Technology 2018
Table 5 Total amount of post-contingency LIB-based load shedding versus LM in IEEE 14-bus system
LIB-based SNB-based
LM NLP Proposed MIP NLP Proposed MIP
0 84.4158 78.8090 7.4869 2.7208
0.01 86.2358 80.6390 10.0294 5.2624
0.02 88.0212 82.4378 12.5250 7.7541
0.03 89.7727 84.2016 14.9753 10.2092
0.04 91.4915 85.9316 17.3819 12.6334
0.05 93.1784 87.6285 19.7458 15.0114
0.06 94.8343 89.2935 22.0684 17.3445
0.07 96.4602 90.9273 24.3508 19.6340
0.08 98.0568 92.5309 26.5942 21.8811
0.09 99.6250 94.1051 28.7996 24.0870
0.1 101.1650 95.6506 30.9680 26.2527

Fig. 5  Total amount of post-contingency load shedding at each bus in IEEE 14-bus system

This negative margin requires the load shedding to restore the In this situation, due to voltage dependency of loads, the voltage
secure operating point. stability margin is improved. In addition, the total amount of load
As shown in Fig. 6, assuming no automatic voltage control, the shedding is decreased significantly. Again, it can be seen that the
obtained load shedding scenario will restore the unstable post- proposed MIP-based UVLS model gives a lower amount of load
contingency operating point to a normal operating point with a shedding.
predetermined LM of 0.1. Considering voltage control of
generators for IEEE 118-bus system, the results of the LIB-based 4 Conclusion
UVLS method are given in Table 6 and Fig. 7. It can be seen that
by considering the voltage control, the developed MIP-based In this paper, the UVLS model was formulated as a mixed integer
UVLS restores the secure operating point with a lower amount of linear programming problem to restore the voltage stability of the
load shedding. system under severe contingencies. Owing to the non-linearity of
The voltage profile of the system has been visualised in Fig. 8 the full AC power flow equations, it was shown that the MIP-based
at LIB-based LM of 0.1. It can be seen that the voltage profile UVLS approach gives lower amount of load shedding compared
using the MIP-based UVLS differs considerably from the NLP- with the NLP-based UVLS model. The efficacy of the proposed
based voltage profile, as shown in Fig. 9. Moreover, the reactive linear AC power flow was investigated for different IEEE test
power outputs of generators for both UVLS plans have been systems under normal and heavy loading conditions. Also, the
illustrated in Fig. 10. It is observed that there is also a significant performance of the proposed linear power flow model was verified
difference between linear and non-linear UVLS strategies in under the important N − 1 contingencies. According to the obtained
distribution of reactive power generations. results, the piecewise linear approximation of cosine function
keeps the accuracy of power flow using a reasonable number of
3.3 UVLS under non-linear loads line segments. Furthermore, the developed MIP-based UVLS
model presents an acceptable accuracy under non-linear load types.
In this part, the efficacy of the proposed model is investigated
under non-linear loads. IEEE 118-bus test system is used for UVLS
simulation with the same contingencies mentioned in the previous
section. It is assumed that the system is operating in heavy loading
condition (i.e. 110% of base case loading) and all loads are
replaced with non-linear ZIP load models. Parameters of the load
model are assumed to be a0 = b0 = 0.2, a1 = b1 = 0.3, and
a2 = b2 = 0.5. The results of LIB-based UVLS are given in Table 7.

IET Gener. Transm. Distrib., 2018, Vol. 12 Iss. 9, pp. 2095-2104 2101
© The Institution of Engineering and Technology 2018
Fig. 6  SNB-based P–V curve for IEEE 118-bus system without voltage control

Table 6 Total amount of post-contingency load shedding


versus LM in IEEE 118-bus system
LIB-based SNB-based
LM NLP Proposed MIP NLP Proposed MIP
0 168.0413 159.7529 165.3303 153.3212
0.02 182.4130 173.7895 179.6723 167.2963
0.04 196.3778 187.4212 193.6048 180.8795
0.06 209.9570 200.6781 207.1468 194.0958
0.08 223.1714 213.6196 220.3211 207.0289
0.10 236.0423 226.2804 233.1467 219.6149
0.12 248.5903 238.5768 245.6421 231.8883
0.14 260.8348 250.5724 257.8262 243.8402
0.16 272.7928 262.4104 269.7167 255.4566
0.18 290.4487 279.8090 287.3720 272.6619
0.20 309.6737 298.7076 306.6063 291.3700

Fig. 8  Voltage visualisation of IEEE 118-bus system

Fig. 7  Total amount of post-contingency LIB-based load shedding at each


bus in IEEE 118-bus system

2102 IET Gener. Transm. Distrib., 2018, Vol. 12 Iss. 9, pp. 2095-2104
© The Institution of Engineering and Technology 2018
Fig. 9  Voltage profile of IEEE 118-bus system after execution of UVLS scheme for λLIB = 01

Fig. 10  Post-contingency reactive power generation of generators in different LMs


(a) MIP method, (b) NLP method

Table 7 Location and amount of post-contingency LIB-based load shedding versus LM in IEEE 118-bus system considering
load model
Constant load Non-linear load
Bus number Pdshed, MW (MIP) Pdshed, MW (MIP) Pdshed, MW (NLP)
1 51.0000 34.4556 41.58816
2 19.9581 0 3.030541
3 39.0000 23.8530 25.25249
4 39.0000 9.8948 13.73372
6 52.0000 4.5456 2.251695
7 3.7933 0 0
8 28.0000 0.8729 16.97698
11 33.2989 0 11.71333
13 7.3589 29.0187 25.6683
76 0 7.2334 0.825149
117 5.6307 8.16966 10.10274
total 279.0400 118.0437 151.1431

5 References [5] Vargas, L.S., Canizares, C.A.: ‘Time dependence of controls to avoid voltage
collapse’, IEEE Trans. Power Syst., 2000, 15, (4), pp. 1367–1375
[1] Rabiee, A., Soroudi, A., Mohammadi-Ivatloo, B., et al.: ‘Corrective voltage [6] Tuan, T., Fandino, J., Hadjsaid, N., et al.: ‘Emergency load shedding to avoid
control scheme considering demand response and stochastic wind power’, risks of voltage instability using indicators’, IEEE Trans. Power Syst., 1994,
IEEE Trans. Power Syst., 2014, 29, (6), pp. 2965–2973 9, (1), pp. 341–351
[2] Van Cutsem, T.: ‘Voltage instability: phenomena, countermeasures, and [7] Arnborg, S., Andersson, G., Hill, D.J., et al.: ‘On undervoltage load shedding
analysis methods’, Proc. IEEE, 2000, 88, (2), pp. 208–227 in power systems’, Int. J. Electr. Power Energy Syst., 1997, 19, (2), pp. 141–
[3] Kundur, P., Paserba, J., Ajjarapu, V., et al.: ‘Definition and classification of 149
power system stability IEEE/CIGRE joint task force on stability terms and [8] Popović, D., Levi, V., Gorečan, Z.: ‘Co-ordination of emergency secondary-
definitions’, IEEE Trans. Power Syst., 2004, 19, (3), pp. 1387–1401 voltage control and load shedding to prevent voltage instability’, IEE Proc.,
[4] Pourbeik, P., Kundur, P.S., Taylor, C.W.: ‘The anatomy of a power grid Gener. Transm. Distrib., 1997, 144, (3), pp. 293–300
blackout-root causes and dynamics of recent major blackouts’, IEEE Power [9] Otomega, B., Glavic, M., Van Cutsem, T.: ‘Distributed undervoltage load
Energy Mag., 2006, 4, (5), pp. 22–29 shedding’, IEEE Trans. Power Syst., 2007, 22, (4), pp. 2283–2284

IET Gener. Transm. Distrib., 2018, Vol. 12 Iss. 9, pp. 2095-2104 2103
© The Institution of Engineering and Technology 2018
[10] Fernandes, T.S., Lenzi, J., Mikilita, M.A.: ‘Load shedding strategies using [21] Trodden, P.A., Bukhsh, W.A., Grothey, A., et al.: ‘Optimization-based
optimal load flow with relaxation of restrictions’, IEEE Trans. Power Syst., islanding of power networks using piecewise linear AC power flow’, IEEE
2008, 23, (2), pp. 712–718 Trans. Power Syst., 2014, 29, (3), pp. 1212–1220
[11] Amraee, T., Ranjbar, A., Feuillet, R.: ‘Adaptive under-voltage load shedding [22] Akbari, T., Bina, M.T.: ‘Linear approximated formulation of AC optimal
scheme using model predictive control’, Electr. Power Syst. Res., 2011, 81, power flow using binary discretisation’, IET Gener. Transm. Distrib., 2016,
(7), pp. 1507–1513 10, (5), pp. 1117–1123
[12] Amraee, T., Ranjbar, A., Mozafari, B., et al.: ‘An enhanced under-voltage [23] Avalos, R.J., Cañizares, C.A., Milano, F., et al.: ‘Equivalency of continuation
load-shedding scheme to provide voltage stability’, Electr. Power Syst. Res., and optimization methods to determine saddle-node and limit-induced
2007, 77, (8), pp. 1038–1046 bifurcations in power systems’, IEEE Trans. Circuits Syst. I, Regul. Pap.,
[13] Kaffashan, I., Amraee, T.: ‘Probabilistic undervoltage load shedding using 2009, 56, (1), pp. 210–223
point estimate method’, IET Gener. Transm. Distrib., 2015, 9, (15), pp. 2234– [24] Sadati, N., Amraee, T., Ranjbar, A.: ‘A global particle swarm-based-simulated
2244 annealing optimization technique for under-voltage load shedding problem’,
[14] O'Neill, R.P., Castillo, A., Cain, M.B.: ‘The iv formulation and linear Appl. Soft Comput., 2009, 9, (2), pp. 652–657
approximations of the ac optimal power flow problem’. Available at http:// [25] FICO: ‘MIP formulations and linearizations-quick reference’. Available at:
www.ferc.gov/industries/electric/indus-act/market-planning/opf-papers/ http://www.fico.com/en/node/8140?file=5125
acopf-2-iv-linearization.pdf, 2012 [26] Kundur, P., Balu, N.J., Lauby, M.G.: ‘Power system stability and control’
[15] Castillo, A., O'Neill, R.P.: ‘Computational performance of solution techniques (McGraw-Hill, New York, 1994)
applied to the ACOPF’, 2013 [27] Mota, L.T.M., Mota, A.A.: ‘Load modeling at electric power distribution
[16] Pirnia, M., O'Neill, R.P., Lipka, P.A., et al.: ‘A computational study of linear substations using dynamic load parameters estimation’, Int. J. Electr. Power
approximations to the convex constraints in the iterative linear IV-ACOPF Energy Syst., 2004, 26, (10), pp. 805–811
formulation’, Federal Energy Regulatory Commission, 2013, https:// [28] Martí, J.R., Ahmadi, H., Bashualdo, L.: ‘Linear power-flow formulation
www.ferc.gov/industries/electric/indus-act/market-planning/opf-papers/ based on a voltage-dependent load model’, IEEE Trans. Power Deliv., 2013,
acopf-8-preprocessed-constraints-iliv-acopf.pdf 28, (3), pp. 1682–1690
[17] Castillo, A., Lipka, P., Watson, J.-P., et al.: ‘A successive linear programming [29] Christie, R.: ‘Power systems test case archive’, University of Washington,
approach to solving the IV-ACOPF’, IEEE Trans. Power Syst., 2016, 31, (4), Electrical Engineering. Available at https://www2.ee.washington.edu/
pp. 2752–2763 research/pstca, 2000
[18] Yang, Z., Zhong, H., Xia, Q., et al.: ‘Optimal power flow based on successive [30] I. CPLEX: ‘ILOG CPLEX homepage 2009’. Available at http://
linear approximation of power flow equations’, IET Gener. Transm. Distrib., www.ilog.com, 2009
2016, 10, (14), pp. 3654–3662 [31] Rosenthal, E.: ‘GAMS-A user's guide’, GAMS Development Corporation,
[19] Tate, J.E., Overbye, T.J.: ‘A comparison of the optimal multiplier in polar and 2008: Citeseer
rectangular coordinates’, IEEE Trans. Power Syst., 2005, 20, (4), pp. 1667– [32] Drud, A.: ‘GAMS/CONOPT user's notes’, Washington, DC, 2001
1674 [33] U.F.E.R. Commission: ‘Mandatory reliability standards for the bulk-power
[20] Bukhsh, W.A., Grothey, A., McKinnon, K.I., et al.: ‘Local solutions of the system’, Order no. 693, p. 72
optimal power flow problem’, IEEE Trans. Power Syst., 2013, 28, (4), pp.
4780–4788

2104 IET Gener. Transm. Distrib., 2018, Vol. 12 Iss. 9, pp. 2095-2104
© The Institution of Engineering and Technology 2018

You might also like