See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/351844137

Comparison of Convexificated SQCQP and PSO for the Optimal Transmission

System Operation based on Incremental In-Phase and Quadrature Voltage
Controlled Transformers

Preprint · May 2021

DOI: 10.13140/RG.2.2.10051.32806

CITATIONS READS

0 67

4 authors, including:

Marcel Sarstedt Thomas Leveringhaus

enercity AG Leibniz Universität Hannover
29 PUBLICATIONS 136 CITATIONS 22 PUBLICATIONS 75 CITATIONS

SEE PROFILE SEE PROFILE

Lutz Hofmann
Leibniz Universität Hannover
168 PUBLICATIONS 1,194 CITATIONS

SEE PROFILE

All content following this page was uploaded by Marcel Sarstedt on 25 May 2021.

The user has requested enhancement of the downloaded file.

Comparison of Convexificated SQCQP and PSO for the
Optimal Transmission System Operation based on Incremental
In-Phase and Quadrature Voltage Controlled Transformers
Marcel Sarstedt1 , Thomas Leveringhaus1 , Leonard Kluß1 , and Lutz Hofmann1
1 Institute of Electric Power Systems, Electric Power Engineering Section, Leibniz Universität Hannover, Germany
1 [surname]@ifes.uni-hannover.de

Abstract
The optimal operation of electrical energy systems by solving a security constrained optimal power flow (SCOPF)
problem is still a challenging research aspect. Especially, for conventional optimization methods like sequential quadratic
constrained quadratic programming (SQCQP) the formulation of the incremental control variables like in-phase and
quadrature voltage controlled transformers in a solver suitable way is complex. Compared to this, the implementation of
these control variables within heuristic approaches like the particle swarm optimization (PSO) is simple but problem
specific adaptations of the classic PSO algorithm are necessary to avoid an unfortunate swarm behavior and local
convergence in bad results. The objective of this paper is to introduce a SQCQP and a modified PSO approach in detail to
solve the SCOPF problem adequately under consideration of flexible incremental in-phase and quadrature transformers
tap sets and to compare and benchmark the results of both approaches for an adapted IEEE 118-bus system. The case-
study shows that both approaches lead to suitable results of the SCOPF with individual advantages of the SQCQP
concerning the quality and the reproducibility of the results while the PSO lead to faster solutions when the complexity
of the investigation scenario increases.

Keywords— Redispatch, Grid Control Optimization, Security Constrained Optimal Power Flow, Sequential Quadratic
Constrained Quadratic Programming, Particle Swarm Optimization

1 Motivation focusing a grid loss reduction in the interest of common

welfare can be described by an security constrained
The massive integration of decentral energy resources optimal power flow problem (SCOPF) [1, 2]. For the
especially to the distribution grid level leads to a transition solution of the SCOPF suitable robust optimization
of the electric power system. The share of conventional methods are necessary that can deal with the specific
thermal power plants in the energy mix is reduced because requirements of the meshed transmission grid level and a
of the power supply priority of renewables and the variety of different flexibilities to avoid local convergence.
phase-out of nuclear and fossil fuel based generation in Operational degrees of freedom within this paper are
Germany. The former unidirectional active and reactive represented by the active and reactive power redispatch and
power flows from the transmission to the distribution grid voltage control of thermal power plants, the curtailment of
level become bidirectional and volatile and the directions renewable power supply as well as incremental in-phase
of active and reactive power flows are decoupled. The and quadrature voltage control of transformers.
existing grid is more stressed in operating points with Solving a SCOPF is challenging, due to its non-convexity,
a high share of renewables and partially critical system np-hardness [3] and high number of degrees of
states may arise due to long distances between power freedom [4]. Therefore, this paper presents the comparison
generation and loads. A variety of grid control actions of two, independently developed optimization methods
by the system operators are necessary to guarantee a (see section 3) for the solution of the SCOPF problem
secure and reliable energy supply. The distribution grid described in section 2. The first approach is a solution
level becomes increasingly active due to the flexibility by a sequentially solved convexificated Quadratically
potentials of converter coupled energy resources, the Constrained Quadratic Program (SQCQP) based on [5–8],
massive integration of measurement infrastructure and which is presented in section 3.1. The second approach
information and communication technologies. In contrast solves a mixed integer, non-linear optimization problem
to that, the flexibility potentials regarding the provision using a modified particle swarm optimization (PSO) e.g.
of ancillary services for the grid control decrease at with an enhanced velocity control of the swarm (see
the transmission grid level. Cost-intensive grid expansion section 3.2, [9]). For the evaluation of the swarm fitness in
measures and a large number of redispatch actions by the PSO as well as of the quality of approximation in the
the transmission system operators are the consequence. SQCQP a Newton Raphson power flow calculation is used.
The optimal planning of these redispatch actions and The focus of the PSO approach is a robust application to
additional grid control measures (e.g. transformer tap sets) the SCOPF problem by avoiding local optima without an
optimization of the PSO hyperparameters. The intention Ii Yi Y´j I´j  Ij j
i j´
of the comparison and the main contribution of this
paper are the benchmark and reproducibility of both
~ ~
approaches for the solution of SCOPF based problems

~
Vi Y s,i , j V´j Vj
at the transmission grid level and the identification of
individual advantages (e.g. computation time, quality
of the results). The benchmark scenario is based on an
adaption of the IEEE 118-bus transmission grid regarding Figure 1 T-equivalent circuit of a transformer
German transmission grid characteristic, introduced in
[10]. A MathWorks MATLAB dataset of the grid and the
results of the case study will be accessible at [11] for The electric behavior at the higher voltage level node i
reproducibility reasons. The scenarios for the case-study and the lower voltage level node j is described by the
and the results for the comparison of both approaches are terminal currents (I i , I j ), the terminal voltages (V i ,V j )
presented in section 4. interconnected by the terminal admittance matrix Y i j
within the two-port equation:

2 Definition of the SCOPF Problem

   
Ii V
=Y ij i (7)
Ij Vj
In general the formulation of the SCOPF problem within 1 h 0
Y i Y j +Y s,i, j


−τ j Y i Y 0j
i
this paper is based on [2, 4]. The first summand of the Y ij = 0 −τ ∗
Y Y 0 2 0
Y Y +Y

(8)
Y i +Y j +Y s,i, j j i j τ j j i s,i, j
SCOPF objective function (see Eq. 1) to be minimized
considers the grid losses Ploss monetized with the cost The rated terminal admittance matrix Y i j,r without in-phase
factor closs . The second summand is the monetarization and quadrature voltage control is based on Eq. 9. The
of nodal active power adaptations ∆ppN using the energy rated phase shifting is specified by the vector group code
resource specific redispatch cost vector c N . The grid losses number k (e.g. k = 5 for vector group Yd5):
are a function of the complex nodal voltages v N :
Vi,r jk30◦
τ j,r = e with Vi,r > V j,r (9)
min closs · Ploss (vvN ) +ccTN · ∆ppN


(1) V j,r

s.t. The adaptation of Y i j,r in case of in-phase and quadrature

voltage control is described with the help of a diagonal tap
v N,min ≤ v N (vvN ) ≤ v N,max (2) changing matrix T i j,rel and the relative tap changings τ i,rel
i T (vvN ) ≤ i T,max (3) and τ j,rel in Eq. 10, wheras τ i,rel typically equals 1 because
(ppN0 + ∆ppN ) + j (qqN0 + ∆qqN ) = 3 ·V V N ·YY ∗NN ·vv∗N (4) only one side of the transformer has tap changers.
 
The constraints (see Eq. 2, 3 and 4) of the SCOPF are τ =1 0
T i j,rel = i,rel (10)
represented by the minimum and maximum of the absolute 0 τ j,rel
value of the nodal voltages (vvN,min , v N,max ), the maximum
Thereby, the in-phase voltage control is specified by the
admissible absolute value of the terminal current of the
voltage change integer n and the increment of the relative
lines and transformers (iiT,max ) and the balances of nodal
voltage change ∆V j,inc at the lower voltage side. The
active and reactive power (ppN , q N ) within the power
quadrature voltage control is given by the phase shift
equation of Eq. 4. Thereby, V N is the diagonal matrix of the
integer m and the increment of the phase shift ∆φ j,inc . Both
nodal voltages v N and Y NN is the nodal admittance matrix
are combined in Eq. 11 and reformulated in Eq. 12, based
of the system.
on the transformer model in [12]:
The first two categories of operational degrees of freedom
are redispatch measures represented by the change of ejm j ∆φ j,inc
the active and reactive power supply (∆ppN , ∆qqN ) of the τ j,rel = (11)
1 + n j ∆V j,inc
different energy resources within the limits:
τ j,rel · (1 + n j ∆V j,inc ) = ejm j ∆φ j,inc (12)
∆ppN,min ≤ ∆ppN ≤ ∆ppN,max (5)
The increment of the relative voltage change ∆V j,inc and
∆qqN,min ≤ ∆qqN ≤ ∆qqN,max (6)
the increment of the phase shift ∆φ j,inc are constant
At the slack and generation buses the voltage magnitudes parameters of a specific transformer type. The additional
are control variables. The flexibility range for the voltage non-convexity resulting from the integer property of the
control are specified by Eq. 2 and constraints of the transformer steps complicates the solution process. The
SCOPF are given by Eq. 6. Further operational degrees limits of the in-phase and quadrature voltage control of
of freedom originate from the flexible adaptation of tap each transformer are given by the minimum and maximum
changer positions regarding an in-phase and quadrature phase shift and voltage change integers, respectively:
voltage control of the transformers at their lower voltage
m j,min ≤ m j ≤ mj,max , m j ∈ Z (13)
side j. The implementation of both is based on the
general T-equivalent circuit of a transformer (see Fig. 1). n j,j,min ≤ n j ≤ nj,max n j ∈ Z (14)
The resulting terminal admittance matrix Y i j of a of two. Therefor, auxiliary equations and variables are
transformer can be calculated with Eq. 15. implemented. In the following these auxiliary variables
will have an additional index a. All equations are split
Y i j = T ∗i j,rel ·Y
Y i j,r ·T
T i j,rel (15) into their real and imaginary part in Cartesian coordinates.
Eq. 17 takes a special role, because if proceeding in the
To combine the two-port equations of all transmission aforementioned way, the variables m j would be part of
assets, the two-port matrices Y i j,r are arranged in a block the argument of sine and cosine functions, and therefore
diagonal matrix Y TT,r and the tap changing matrices T i j,rel there would be non-quadratic functions. Due to the
are arranged in a block diagonal matrix T T,rel . So in Eq. 16 integer property of m j , exact reformulations as piecewise
the matrix form of Eq. 15 and the terminal admittance linear functions with vertices at all integer values of m j
matrix Y TT result. In Eq. 17 the additionally necessary are a theoretical option, but the Special-Ordered-Sets
matrix form of Eq. 12 is given, in which e is a one vector, required for these approximations would massively
M T and N T are the diagonal matrices of all m j and n j and slow down the solution process of each sequential
∆v T,inc and ∆ϕ T,inc are vectors of the increment parameters step and massively extend the memory requirements.
∆V j,inc and ∆φ j,inc : Approximating these functions with first or second-order
Y TT = T ∗T,rel ·Y
Y TT,r ·T
T T,rel (16) Taylor series approximations in each sequential step of
the SQCQP has been tested but led to high approximation
M T ·∆ϕ
jM ∆ϕ T,inc
T T,rel · (11T +N
N T ·∆v
∆vT,inc ) = e (17) deviations and bad convergence. Deeper research has led
to a somewhat counterintuitive way, but that has proven
For transmission lines and transformers terminals without to be advantageous: Eq. 17 is not to be split into its real
voltage control, the matrix T T,rel only contains ones on its and imaginary parts in Cartesian coordinates, but into its
corresponding diagonal elements and: squared absolute values and into the tangents of its angles,
whereas the transformer tap changes in T T,rel are still to be
m j,min = m j,max = n j,min = n j,max = 0 (18)
split into their real parts T T,rel,r and imaginary parts T T,rel,i
∆V j,inc = 0 (19) (likewise for the vector τ T,rel ). This procedure does not
∆φ j,inc = 0 (20) yet require approximations at this stage and so Eq. 22
results for the squared absolute values and Eq. 23 results
To consider the topology of the network the node-terminal- for the tangents (whereas t T (m mT ) is an abbreviation for the
incidence matrix I NT is used to calculate the nodal tangents):
admittance matrix Y NN in Eq. 21. The incidence matrix I NT
T 2T,rel,r +T T 2T,rel,i · (11T +N ∆vT,inc )2 = 1 2T


describes the topology of the grid by the specification of N T ·∆v (22)
the connection nodes i and j of each two-port: τ T,rel,i
= tan (M M T ·∆ϕ mT )
∆ϕ T,inc ) = t T (m (23)
Y TT ·II TNT τ T,rel,r
Y NN = −II NT ·Y (21)
Eq 22 is of fourth degree but can be reformulated into
Eq. 16, 17 and 21 need to be added to the SCOPF in
two equations of second degree polynomial. The result is
Eq. 1-4 to consider the additional degrees of freedom of
shown in Eq. 24 and 25.
transformer tap changings.
τ 2T,rel,r +ττ 2T,rel,i = τ 2T,rel (24)
3 Optimization Methods for the T T,rel · (11T +N
N T ·∆v
∆vT,inc ) = 1 T (25)
Solution of the SCOPF The geometry of the function in Eq 23 has been
investigated more in detail and the tangent function can be
The resulting formulation of the SCOPF has been derived well approximated for small angles - and only those are
in the previous section. As described in section 1, this paper relevant at quadrature voltage control - by a linearization.
presents the comparison of two, independently developed It should be emphasized that thus approximations are used
optimization methods for the solution of the SCOPF in each step of the SQCQP. To enable convergence, the
problem. Adaptions of the problem formulation according point of linearization must be updated in each sequential
to specific needs of the approach and adaptions of solver step ν. The result is shown in Eq. 26. The overall system
parameters are described in the following sections. of equations is shown in Eq. 27 and 28.
!
3.1 SQCQP Approach ∂tt (m
mT )
τ T,rel,i −ττ T,rel,r t T (m
mT )|ν + · ∆m
mT ≈ 0 T
The first approach applies a sequentially solved novel ∂mmTT ν
convexificated Quadratically Constrained Quadratic (26)
Program (SQCQP) based on [5–8] that is further developed
in this paper to consider in-phase and quadrature voltage
controlled transformers. For the convexification, the
equations of the SCOPF need to be reformulated as
real-valued system of equations with equal number of
equations and variables and a maximal polynomial degree
    
0T τ T,rel,r 3.2 Particle Swarm Optimization Approach
 1T  τ T,rel,i 
0 
 T
 τ
 T,rel 
 In the field of electric power system, optimization
m 
 T
 m 
 T  metaheuristics are used for solving a variety of
 nT 
 
 n T 
  optimization problems concerning system planning
 0T 
 
 u T,a,r 
  and operation [13–19]. For solving SCOPF problems the
 0T   u T,a,i  PSO (cf. [20, 21]) shows an appropriate convergence
   
 0T   i behavior and good performance characteristics by the
 0 = f  T,a,r  (27)

 T  i T,a,i  adaptation of hyperparameters [13, 16, 17]. The PSO
0   i 
 T  T,r  algorithm used for the investigations within this paper is
 0T   i T,i 

 0T 
 
 i T 
 described in detail at [9]. For a better understanding of

∆ppN 
 
 u N,r 
 the specific adaptations for solving the SCOPF presented
 ∆qqN 
 
 u N,i 
  in section 2 the main equations of the PSO and the
0 
N
 u 
N
general procedure are introduced. At the beginning of the
0 Ploss iterative solution process the particle swarm consisting of
i = 1, ..., n individuals and j = 1, ..., m control variables xi, j
  is initiated randomly within the corresponding flexibility
τ 2T,rel,r +ττ 2T,rel,i −ττ 2T,rel limits of Eq. 2, 5, 6, 13 and 14. Analogously to the control

 T T,rel ·(11T +N N T ·∆v
∆vT,inc ) 

variables, the velocity vector of the swarm particles v i
∂tt (m
mT )
τ −τ
τ ·
 T,rel,i T,rel,r T T νt m
(m )| + ∂m mT
· ∆m
mT 
 is generated randomly for the initial iteration step t = 0.
T ν
For each swarm particle a power flow calculation based
 
 mT 
nT on the Newton Raphson algorithm is performed and the
 
 
T T,rel,r ·II TNT ·uuN,r −T T T,rel,i ·II TNT ·uuN,i −uuT,a,r fitness value is evaluated by Eq. 1 [13, 16]. By this,
 
 

 T T,rel,i ·II TNT ·uuN,r +T T T,rel,r ·II TNT ·uuN,i −uuT,a,i 
 the implementation of different control variables to the
 G TT ·uuT,a,r −B BTT ·uuT,a,i −iiT,a,r 
SCOPF problem is simple and is done by the integration
=  (28)

 G TT ·uuT,a,i +B BTT ·uuT,a,r −iiT,a,i 
 of the variable to the power flow calculation algorithm.
T T,rel,r ·iiT,a,r +T T T,rel,i ·iiT,a,i −iiT,r
An additional punishment summand g is added to Eq. 1
 
T T,rel,r ·iiT,a,i −T T T,rel,i ·iiT,a,r −iiT,i
 
for the non compliance of technical constraints. See [9]
 
i 2T,r +ii2T,i −ii2T
 
for details regarding the determination of the punishment


3 · U N,r ·II NT ·iiT,r +U U N,i ·II NT ·iiT,i
− p N0
 
summand g [22, 23]. Based on the fitness values the global
 
 3 · U N,i ·II NT ·iiT,r −U U N,r ·II NT ·iiT,i −qqN0 
best swarm particle position pgb of all particles as well
 
 u 2N,r +uu2N,i −uu2N 
    as the individual best position of each particle pb,i during
3 · u TN,r ·II NT ·iiT,r +uuTN,i ·II NT ·iiT,i − Ploss
all iterations steps are updated. For the next iteration step
the movement of the swarm and by this the change of the
With this system of equations, that only has real-
control variables are determined:
valued polynomial functions of degree two and whose
number of equations equals the number of variables, the v i,t+1 = wvvi,t + c1r 1 (ppb,i −xxi ) + c2r 2 p gb,i −xxi


(32)
SQCQP approach from [8] can be applied. As reasoned
and undertaken in [8], a distributed slack needs to be with r 1 ,rr 2 = (nx1),vvi ,xxi = (1xm) (33)
inserted additionally before inverting and convexifying. x i,t+1 = x i,t + kvvi,t+1 (34)
The resulting convexificated quadratically constrained
quadratic program is shown in Eq. 29 to 31. The acceleration coefficients c1 and c2 describe the social
and the cognitive interactions within the swarm. The
min closs · Ploss m TT ,nnTT , ∆ppTN , ∆qqTN +ccTN · ∆ppN
 


(29) vectors r 1 and r 2 consist of random numbers in the interval
of [0, 1] representing the stochastic nature of the PSO. The
s.t. inertia indicates how the velocity of the swarm v i,t+1 is
affected by the current velocity v i,t . The initial inertia w
v N,min ≤ v N m TT ,nnTT , ∆ppTN , ∆qqTN ≤ v N,max
 

(30)
decreases within the iteration process from wstart to wend to
iT mTT ,nnTT , ∆ppTN , ∆qqTN ≤ iT,max
 

(31) guarantee at the beginning of the solution process a global
and at the end (t = tmax ) a local search behavior of the
To enhance the speed of the sequential approach, the swarm [13, 24]:
necessary relative MIPgap to be is dynamically adjusted
in each sequential step: In the first step, for which a
 
wstart − wend
heightened forecast error is to be expected due to the w = wstart − t (35)
tmax
high state changings by the optimizer and the used
approximations, it is set to a value of 10 %. In the The constriction factor k ensures a convergence of the
following steps it is set to the tenth of the MIPgap of swarm in a reliable solution of the optimization problem
the respective preceding step. Furthermore an dynamically [13, 24]:
adjusted time-limit is set for the optimizer, that starts
2
with 128 s in the first step and doubles with each step. The k= p , with ψ = c1 + c2 ≥ 4 (36)
total time for the duration one step can be longer, due to 2−ψ − ψ 2 − 4ψ
the convexification and further scripts.
The iterative solution process stops after the maximum Table 1 Hyperparameters of the PSO
iteration step tmax is reached. In the following the
modifications of the classic PSO [20] approach regarding zTQ
c1 zAR
the solution of the SCOPF are introduced. Within the n tmax wstart wend zTIP λ
c2 zRR
classic PSO only continuous variables are implemented to zVC
avoid a negative influence on the swarm behavior and to 200 500 2 0.9 0.4 1/10 1/5 100
guarantee appropriate convergence. To consider integers
for the in-phase and quadrature transformer tap sets the Table 2 Limits of the control variables and incremental
corresponding variables z are considered as rounded values voltage and in-phase change per transformer tap change
only during the evaluation of the swarm fitness [19]:
∆ppN,min
d2xi,z e ∆ppN,max v N,min m min n min
xi,z = b c (37) V inc
∆V ∆φφ inc
2 ∆qqN,min v N,max m max n max
To avoid unfeasible solutions the movements of the ∆qqN,max
particles is limited in front of the next iteration step within see [11] 0.9 p.u. -10 -10
0.25 % 1°
a set-to-limit operator [14]: see Fig. 7 1.1 p.u. 10 10

xi, j (xi, j < xi, j,min ) = xi, j,min (38)

xi, j (xi, j > xi, j,max ) = xi, j,max (39) Matlab. The simulations are performed on computers with
a 2.7 GHz QuadCore and 16 GB RAM. For the comparison
The limitation of the swarm velocities is a common of the SQCQP and the PSO approach an adaptation of
procedure to avoid alternating jumps between the control the IEEE 118-bus transmission grid regarding German
variable limits (xi, j,min , xi, j,max ) and to guarantee a transmission grid characteristics (see Fig. 2, cf. [10]) is
more detailed global solution search especially at the used. A MathWorks Matlab dataset of the grid model
beginning of the PSO [25]. The speed coefficient z is and the results of the case study for the SQCQP and
specified individually for each control variable (see Tab. 1, the best PSO run are available at [11]. The limits of the
active power redispatch zAR , reactive power redispatch control variables as well as the incremental voltage and
zRR , quadrature voltage control transformer zTQ , in-phase phase change per transformer tap change n and m are
voltage control transformer zTIP , voltage control power given in Tab. 2. The costs for grid losses closs are set
plants zVC ): to 1 while the costs for active and reactive redispatch cN
vi, j (|vi, j | > vi, j,max ) = sgn(vi, j )r3 vi, j,max with (40) are set to 0. Bus number 63 is selected as slack and the
voltage phase is set to 0°. The initial grid losses with the
vi, j,max = z(xi, j,max − xi, j,min ), r3 = rand([0, ..., 1]) (41)
voltage control from [10] are Ploss,0 = 189.90MW. The
The movement of the swarm particles to unfeasible investigations within the case-study are divided into three
solutions and an accumulation of the swarm at the control scenarios. In scenario 1 operational degrees of freedom are
variable limits are restricted by [20]: only provided by power plants that are able to contribute

to active and reactive power redispatch and to perform
vi, j xi, j = xi, j,min ∧ vi, j < 0 = −vi, j (42) the voltage control at the generation buses, respectively.
vi, j (xi, j = xi, j,max ∧ vi, j > 0) = −vi, j (43) In scenario 2 additional flexibilities are provided by the
incremental in-phase voltage control of the transformers.
At the beginning of the PSO the high inertia of the
In scenario 3 additionally the incremental quadrature
swarm enables a global solution search. As a result of the
voltage control of the transformers is considered.
decreasing inertia over PSO iterations the search behavior
becomes more local. For a better local solution search a
mutation operator is introduced that manipulates a random 4.1 Scenario 1
control variable of each swarm particle [13]: In Fig. 4 the convergences of the SQCQP and the PSO
for scenario 1 are presented. The convexificated quadratic
xi,r4 = r5 vi,r4 ,max ∀ i ∈ [1, ..., n] with (44)
approximations of the SQCQP approximate the non-linear
r4 = rand([0, ..., m]), r5 = rand([−1, ..., 1]) (45) system behavior well except for the second sequential
The hyperparameters of the PSO (see Table 1) are step (see Fig. 4 left). The SQCQP finds a slightly
selected manually based on references in the literature better solution (∆Ploss = 0.06 MW) and the computation
and experiences during the case study [9, 17, 18, 20, 26]. is faster (∆t = 516 s). Within the λ PSO runs the average
Multiple PSO runs λ are performed in parallel due to the result is Ploss,av = 46.34 MW and the worst result is
stochastic nature of the PSO and the possibility of local Ploss,w = 46.83 MW. In contrast to that, another advantage
convergence [25]. of the SQCQP in scenario 1 is the reproducibility of the
results.
The exclusive monetarization of grid losses in the objective
4 Case-Study function leads to significant active and reactive power
redispatches and an increase of the control voltages at
The SQCQP and the PSO approach presented in the generation buses (see [11]). For the resulting absolute
sections 3.1 and 3.2 are both implemented in MathWorks values of the nodal voltages v N in Fig. 3 just small
deviations between the SQCQP and the PSO are identified. 200 Load Flow Results 200 Load Flow Results
Quadratic Approximation
This observation also applies to the following case studies, t 1 2 3 t 500

Ploss in MW
Ploss in MW
160 189.90 69.69 46.46 160 45.66
so the evaluation of the voltages is not presented again. 200 189.90 11.19Results
Load Flow 47.85 200 TimeFlow
Load in s Results
573
TimeQuadratic
in s - Approximation
12 11
120 tt 120
14 5
2 6
3 t 500

in MW
in MW
160 45.61 69.69
189.90 160
45.61 46.46
45.60 45.66

Plosslosses
Plosslosses
45.62 45.61 45.60
200
4.2 Scenario 2 200
80 Time in
Time in ss
189.90
Load
Quadratic
9-
11.19Results
Flow
11
12
47.85
14
11
80
Approximation
TimeFlow
Load in s Results
573
120 t 4 5 120
6
t 1 2 3 t 500

MW
MW

in Grid
in Grid
In scenario 2 again the SQCQP gives a slightly better 160
40 45.61 45.61 45.60
189.90 69.69 46.46160
40 45.66

Plosslosses
Plosslosses
45.62 45.61 45.60
result then the PSO (∆Ploss = 0.02 MW) in a lower 80 Time in s 189.90
9 11.19
11 1480
47.85 Time in s 573
Time in s - 12 11
120
0 t 4 5 120
6 0
computation time (∆t = 418 s). The scattering of the

Grid
Grid
40 1 2 3 45.61
Iterationstep
4 45.61 5 45.60
6
PSO40
0 250
Iterationstep t
500
SQCQP 45.62 t45.61 45.60

Ploss in MW Grid losses

Ploss in MW Grid losses
PSO results is reduced compared to scenario 1 with 80 Time in s 9 11 1480
0 0
an average result of Ploss,av = 45.54 MW and the 200
Figure 1
40 4
2
Results3 Load4 Flow
for 5 Results
6
scenario 12000 250Load Flow Results
500
SQCQP Iterationstep t
Quadratic Approximation PSO40 Iterationstep t
t 1 2 3 t 500
160 189.90 63.69 46.51 160 45.43
0
200 0
200
1 2 3 189.90
4 Flow
Load 5 Results
09.76 6
47.78 0 250Load
TimeFlow 559500
in s Results
, 380 kV , 220 kV SQCQP Iterationstep
TimeQuadratic
in s t 11 24 PSO
- Approximation Iterationstep t
120 tt 120
14 5
2 6
3 t 500

in MW
in MW
160 45.43 63.69
189.90 45.41 46.51
45.41 160 45.43

Plosslosses
Plosslosses
200 45.44 45.41
189.90 45.41
09.76Results
47.78 200 TimeFlow
in s Results
559
80 Time in
Time in ss Load
31-
Flow
43
11 32
24
80 Load
120 Quadratic Approximation 120
t 4 5 6
t 1 2 3 t 500

MW
MW
in Grid

in Grid
160
40 45.43 45.41 45.41
189.90 63.69 46.51 160
40 45.43

Plosslosses
Plosslosses
45.44 45.41 45.41
80 Time in s 189.90
31 09.7643 47.78
32 80 Time in s 559
Time in s - 11 24
120
0 t 4 5 6 120
0

Grid

Grid
40 1 2 3 45.43
Iterationstep
4 45.41 5 45.416 400 250
Iterationstep t
500
Transformer SQCQP 45.44 t45.41 45.41 PSO

Ploss in MW Grid losses

Ploss in MW Grid losses
80 Time in s 31 43 32 80
Slack bus, 380 kV, voltage control
0 0
Active and reactive power dispatch 200 1 2 3 Load 4 Flow 5 Results
6 2000 250Load Flow Results
500
, Generation bus, voltage control, active power redispatch
40
SQCQP Iterationstep t
Quadratic Approximation PSO40 Iterationstep t
t 1 2 3 t 500
160 160
0 189.90 62.90 45.63 0 45.02
Figure 1 Results for scenario 22000
200 5 2 3 189.904 Flow
Load 5 Results
08.84 6
46.75 250Load
TimeFlow 579500
in s Results
SQCQP PSO Iterationstep
TimeQuadratic
in s t 84 50
- Approximation Iterationstep t
120 120
Figure 2 IEEE 118-bus system with control variables tt 14 5
2 6
3 t 500

in MW
in MW

160 189.90 44.81

44.82 62.90 44.80
45.63 160 45.02

Plosslosses
Plosslosses

200 44.82 08.84

189.90 44.80 46.75
44.79 200 Time in s 579
80 Time in Load Flow
in ss 412
TimeQuadratic
Results
1,044
84 2,064
- Approximation
50
80 Load Flow Results
120 t 4 5 6 120
t 1 2 3 t 500

MW
MW

in Grid
in Grid

160
40 44.82 44.81 44.80
189.90 62.90 45.63 160
40 45.02

Plosslosses
Plosslosses

44.82 44.80 44.79

80 Time in s 189.90
412 08.84
1,044 46.75
2,064 80 Time in s 579
Time in s - 84 50
120
0 t 4 5 6 120
0

Grid
Grid

40 1 2 3 44.82
Iterationstep
4 44.81 5 44.80
6
PSO40
0 250
Iterationstep t
500
SQCQP 44.82 t44.80 44.79

Grid losses
Grid losses

80 Time in s 412 1,044 2,064 80

0 0
1 2 3 4 5 6 0 250 500
40
SQCQP Iterationstep t PSO40 Iterationstep t
Initial system state:
0 0
- no redispatch 1 2 3 4 5 6 0 250 500
SQCQP Iterationstep t PSO Iterationstep t
- no voltage control
Figure 6 Results for scenario 3

worst result Ploss,w = 45.61 MW. The differences between

the results of scenario 1 and 2 are small due to an already
high utilization of the voltage limits based on reactive
power optimization and voltage control in scenario 1 (see
Fig. 3) as well as the dominating influence of the in-phase
SQCQP voltage control on the same voltage magnitudes.

4.3 Scenario 3
In scenario 3 the additional consideration of the quadrature
voltage control significantly lowers the objective value.
The impact is higher than that of the in-phase voltage
control in scenario 2. This can be reasoned with the
dominating influence of the quadrature voltage control
on the distribution of the active power flows in the grid
PSO and consequently on the losses. The SQCQP reduces the
grid losses more then the PSO (∆Ploss = 0.23 MW), but
needs significantly more time, despite the time limits
Bus voltages vN in p.u. defined in section 3.1. Compared to scenario 2 the
scattering of the PSO results increases slightly with an
0.9 1 1.1 average result of Ploss,av = 45.14 MW and the worst result
Ploss,w = 45.30 MW. In general, both approaches utilize the
Figure 3 Initial and resulting nodal voltages in scenario 1 flexibility potentials of the control variables in the same
way (see Fig. 7). Significant differences only arise for the
 1000
1000
800
1000
800

redispatch
600

redispatch
redispatch
800
600
1000
400 with each other regarding the quality of the results and
redispatch
600
400
800

MW
200
the computation time within a case-study of an adapted
MW
400
redispatch
inN MW 200
600
power
0
IEEE 118-bus system. Within the three investigation
NNMW in
power

200
power

4000
-200
pin

SQCQP scenarios an increasing number of control variables are

power

-2000
inpMWpp

200
Active
in

-400 SQCQP
PSO
Active

considered. At scenario 1 only active and reactive power

Active

-200
N

-4000
power

-600 PSO
Flexibility
SQCQP limits
pN 
Active

-400
-600
-200 Flexibility
PSO limits redispatch and voltage control measures are implemented.
-800 SQCQP
-600 Flexibility limits
Active

-800
-400
-1000
4
PSO
11 15 24 28 35 48 67 73 75 87 91 96 98 103104105 106107110111 112116
In scenario 2 and 3 additional the incremental in-phase and
-800
-1000
-600
4 11 15 24 28 35 48 67 73 75 Bus
87 91number
Flexibility limits
96 98 103104105 106107110111 112116 quadrature voltage control of the transformers are taken
-1000
-800 4 11 15 24 28 35 48 67 73 75 Bus
87 91number
96 98 103104105 106107110111 112116 into account.
-1000 Bus number
500 4 11 15 24 28 35 48 67 73 75 87 91 96 98 103104105 106107110111 112116
500
The results of the case study can be concluded as
400 Bus number
follows: In comparison of the influence on the objective
redispatch

500
400
300
redispatch
redispatch

400
300
500
200
value, it can be seen, that the in-phase voltage control
redispatch

300 as well as the quadrature voltage control only have a

Mvar

200
400
100
Mvar
redispatch
N Mvar

200
power

100
300 smaller impact on the objective value compared to the
0
NNMvar
power
in
power

100
0
200 reactive and specially the active power redispatch. The
in

-100
qin
power
inqMvar
Reactive

0
-100
q

100
advantages of the voltage controllers are likely to become
qin

-200
Reactive
Reactive
power

-100
-200
N

0
-300 visible, when current and voltage congestions play an
Reactive
q N 

-200
-300
-100
-400
important role in the dataset. The results of the scenario 1
Reactive

-300
-400
-200
-500
-400
-500
-300
24 28 73 75
Bus
87
number
90 96 98 116 already show differences and specific advantages of the
24 28 73 75 87 90 96 98 116
-500
-400 24 28 73
Bus number
75 87 90 96 98 116
SQCQP and the PSO in finding an optimal solution
Bus number of the SCOPF: Both algorithms nearly reach the same
p.u.

1.1
-500
voltage
Nodal

24 28 73 75 87 90 96 98 116
p.u.

1.1
vinN p.u.

Bus number
voltage

1.08
voltage

objective value, but the PSO takes a longer computational

Nodal
Nodal

vvNNp.u. in

1.1
1.08
voltage

1.06
in
Nodal

1.08
1.06 0 4 11 15 35 48 53 63 67 91 103 104 105 106 107 110 111 112 time. So the advantages of analytical approaches for
vN invp.u.
N in

1.10 4 11 15 35 48 53 63Bus
67 number
91 103 104 105 106 107 110 111 112
voltage
Nodal

1.06 Bus number continuous functions can be seen. The results of scenario 2,
1.08 60 4 11 15 35 48 53 63Bus
67 number
91 103 104 105 106 107 110 111 112
Bus number
In-phase
nn n

1.06 6
4 that additionally considers incremental in-phase voltage
n set
In-phase
In-phase

6 0 4 11 15 35 48 53 63 67 91 103 104 105 106 107 110 111 112

4 control of transformers, again reveal nearly equality of
Bus number
set
set

2
settap
In-phase

4
2
tap

6
tapntap

0 the gained objective value and an advantage of the

In-phase

20
04
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Transformer
7 8 9 10 11number
tap set

0 1 2 3 4 5 6 12 13 14 15 16 17 18 19 20 SQCQP with respect to the summed computation time

02
0 1 2 3 4 5 6
Transformer number
7 8 9 10 11 12 13 14 15 16 17 18 19 20
10 Transformer number of all sequential steps. The results of the scenario 3,
100 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
8
Transformer number
that additionally considers in-phase and quadrature voltage
108
mm

6 control of transformers, again reveal nearly equality of

68
m set
m

10
4
the gained objective value that can be also identified by
set
set
settap

68
4
2
tap
tap

the comparison of the control variables in Fig. 7. The

Quadrature

4
2
tapm

06
Quadrature
Quadrature
tap set

02
-24
convergence of both approaches to nearly the same control
Quadrature

0
-2
-42
variable utilization lead to the assumption that the solution
Quadrature

-2
-4 of the optimization problem has no pareto optimum. In
-60
-4
-6
-2
-8 contrast to scenario 1 and 2 a significant advantage of the
-6
-8 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
-4
0 1 2 3 4Transformer
7 8 9 10 11number
5 6 12 13 14 15 16 17 18 19 20 PSO with respect to computation time arise in scenario 3.
-8
-6 0 1 2 3 4 5 6 Transformer
7 8 9 10 11number
12 13 14 15 16 17 18 19 20 Comparing all scenarios the computation times of the
Figure 7-8 0 Transformer
Utilization of control number in scenario 3
variables
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 PSO are constant because it scales only with the number
Transformer number
of power flow calculations and the performance of the
computer in case of background processes. In contrast
in-phase voltage control and especially for the quadrature to that, the integration of the incremental flexibilities to
voltage control of the transformers. the SQCQP leads to a more complex problem and by
this to an increased computation time (∆t = 3075 s).
It was observed that the SQCQP has stuck in lowering
5 Conclusion and Outlook the MIPgap without finding new solutions, but trying to
increase the lower bound, when considering quadrature
The flexibilities in classical Security Constrained Optimal voltage control. Equivalent or slightly different solutions
Power Flows (SCOPF) are often only represented by of the SCOPF are suspected in this context and this aspect
active and reactive power redispatch and voltage control as well as performance and accuracy improvements of
measures of thermal power plants or aggregated renewable the SQCQP will be part of future research. Similar future
energy resources. Within this paper, incremental in- goals hold for the hyperparameter tuning of the PSO
phase and quadrature voltage controlled transformers are (e.g. swarm size, see Tab. 1) or the introduction of a
considered as additional, mixed-integer control variables. sequential solution process. The optimization approaches
For the solution of this more complex SCOPF problem and investigations are aimed to be extended by network
two different powerful optimal power flow solvers, configurations, flexibility potentials of underlying voltage
namely a Sequential Quadratic Constrained Quadraic levels or costs for redispatch by linking the simulation to a
Programming (SQCQP) approach and a modified particle market simulation.
swarm optimization (PSO), are introduced and compared
 6 References [14] B. Zhao, C. Guo, and Y. Cao, “A multiagent-based
particle swarm optimization approach for optimal
[1] J. Carpentier, “Contribution a l’etude du dispatching reactive power dispatch,” IEEE transactions on
economique,” Bulletin de la Societe Francaise des power systems, vol. 20, no. 2, pp. 1070–1078, 2005.
Electriciens, vol. 3, no. 1, 1962. [15] A. Kaviani, H. Baghaee, and G. Riahy, “Optimal
[2] H. W. Dommel and W. F. Tinney, “Optimal Sizing of a Stand-alone Wind/Photovoltaic
power flow solutions,” IEEE Transactions on Power Generation Unit using Particle Swarm Optimization,”
Apparatus and Systems, vol. PAS-87, no. 10, pp. Simulation, vol. 85, pp. 89–99, Feb. 2009.
1866–1876, 1968. [16] T. Sharma, A. Yadav, S. Jamhoria, and R. Chaturvedi,
[3] D. Bienstock and A. Verma, “Strong np-hardness of “Comparative study of methods for optimal
ac power flows feasibility,” 2019. reactive power dispatch,” Electrical and Electronics
[4] S. Frank, I. Steponavice, and S. Rebennack, “Optimal Engineering, vol. 3, no. 3, pp. 53–61, 2014.
power flow: a bibliographic survey,” Energy systems, [17] M. Sarstedt, S. Garske, and L. Hofmann,
2012. “Application of PSO-Methods for the Solution
[5] T. Leveringhaus and L. Hofmann, “Combined of the economic Optimal Reactive Power Dispatch
and optimized redispatch management of multiple Problem,” in 2018 IEEE Electronic Power Grid
congestions and voltage deviations with active and (eGrid), Nov. 2018, pp. 1–6.
reactive power based on ac-ptdfs with distributed [18] M. A. Abido, “Optimal power flow using particle
slack,” in 2015 IEEE Power Energy Society General swarm optimization,” International Journal of
Meeting, 2015, pp. 1–5. Electrical Power & Energy Systems, vol. 24, no. 7,
[6] ——, “Optimal power flow comprising congestions pp. 563–571, 2002.
and voltage management by global quadratic [19] B. Zhao, Q. Jiang, C. Guo, and Y. Cao, A Novel
optimization of active and reactive power,” in 2016 Particle Swarm Optimization Approach for Optimal
IEEE International Conference on Power System Reactive Power Dispatch.
Technology (POWERCON), Sep. 2016, pp. 1–6. [20] J. Kennedy and R. Eberhart, “Particle swarm
[7] T. Leveringhaus, T. Breithaupt, S. Garske, and optimization,” in Proceedings of ICNN’95 -
L. Hofmann, “Modelling of sequential optimal power International Conference on Neural Networks,
flow by piecewise linear convexificated quadratic vol. 4, Nov. 1995, pp. 1942–1948 vol.4.
approximations,” in 2018 International Conference [21] R. Eberhart and J. Kennedy, “A new optimizer using
on Power System Technology (POWERCON), 2018, particle swarm theory,” in MHS’95. Proceedings of
pp. 87–92. the Sixth International Symposium on Micro Machine
[8] T. Leveringhaus and L. Hofmann, “Detailed and Human Science, Oct. 1995, pp. 39–43.
reasoning and derivation of a convexificated [22] G. Coath and S. K. Halgamuge, “A comparison of
quadratic approximation approach for the constraint-handling methods for the application of
security constrained optimal power flow,” in particle swarm optimization to constrained nonlinear
2020 International Conference on Smart Grids and optimization problems,” in The 2003 Congress on
Energy Systems (SGES), 2020. Evolutionary Computation, 2003. CEC’03., vol. 4.
[9] M. Sarstedt, L. Kluß, J. Gerster, T. Meldau, IEEE, 2003, pp. 2419–2425.
and L. Hofmann, “Survey and comparison of [23] A. W. Mohemmed and M. Y. Alias, “Particle Swarm
optimization-based aggregation methods for the Optimization for Constrained and Multiobjective
determination of the flexibility potentials at vertical Problems: A Brief Review,” p. 5.
system interconnections,” Energies, 2021.
[24] K. E. Parsopoulos and M. N. Vrahatis, “Particle
[10] H. Barrios, A. Roehder, H. Natemeyer, and Swarm Optimization Method for Constrained
A. Schnettler, “A benchmark case for network Optimization Problems,” p. 7.
expansion methods,” in IEEE PowerTech, 2015.
[25] J. F. Schutte, J. A. Reinbolt, B. J. Fregly, R. T. Haftka,
[11] M. Sarstedt, N. Majumdar, C. Blaufuß, and and A. D. George, “Parallel global optimization with
L. Hofmann, “Dataset: Multi-voltage-level electric the particle swarm algorithm,” International Journal
power system data sets,” 2020. for Numerical Methods in Engineering, vol. 61,
[12] J. Holzer, C. Coffrin, C. DeMarco, R. Duthu, no. 13, pp. 2296–2315, Dec. 2004.
S. Elbert, S. Greene, O. Kuchar, B. Lesieutre, H. Li, [26] J. Polprasert, W. Ongsakul, and V. N. Dieu, “Optimal
W. K. Mak, H. Mittelmann, R. O’Neill, T. Overbye, Reactive Power Dispatch Using Improved Pseudo-
A. Tbaileh, P. V. Hentenryck, A. Veeramany, and gradient Search Particle Swarm Optimization,”
J. Wert, “Grid Optimization Competition Challenge Electric Power Components and Systems, vol. 44,
2 Problem Formulation,” p. 79. no. 5, pp. 518–532, Mar. 2016.
[13] J. Zhu, Optimal Power Flow. John Wiley Sons, Ltd,
2015, ch. 8, pp. 297–364.

View publication stats