Optimal Power Flow Using Genetic Algorithm

Fernando Buzzulini Prioste

Dept. of Electrical Engineering
Santa Catarina State University
Joinville, Brazil
fernando.prioste@udesc.br

Abstract—This paper presents a genetic algorithm (GA) to [15], [16], [21]. The Alternating Current Optimal Power Flow
solve Optimal Power Flow (OPF) problems, optimizing electricity (ACOPF) is at the heart of Independent System Operator (ISO)
generation fuel cost. The GA based OPF is a derivative free power markets, and is solved for system planning, day-ahead
optimization technique that relies on the evaluation of several
points in the parameter search space strictly on the objective markets and even at every 5 minutes. Even nowadays, 60
function. A 3 bus system and the IEEE 30 bus test system are years after the problem was formulated, there are not a fast
used to validate the developed GA based OPF by means of and robust solution technique available for its full solution.
comparisons with an interior point based optimal power flow. Depending on the amount of electric energy production of a
country, finding a good solution technique for the full ACOPF
Index Terms—Electric Power Systems, Power Flow, Optimal
Power Flow, Nonlinear Optimization, Genetic Algorithm, Interior could save tens of billions of dollars annually. The OPF
Point problem is a large, highly nonlinear, multimodal, constrained
optimization problem that may have nonconvex, nonsmooth,
I. I NTRODUCTION and nondifferentiable objective functions, in which there may
be more than one optimal local solution.
Since the emergence of electric power systems, there has
The methods or techniques that have been implemented to
always been a search for increasingly efficient methods to
optimize the OPF are divided into two categories: deterministic
solve the power flow problem. The power flow consists of
methods and evolutionary methods. Deterministic methods in-
solving a set of nonlinear algebraic equations with quadratic
clude linear, nonlinear programming, quadratic programming
and trigonometric terms. When its solution is found, state vari-
and interior point method [17], [18]. These methods may
ables of a power system are available and several information
have problems in handling many local minima because of
about the system may also be used, for example, to system
the nonconvexity of OPF problems. Due to limitations of
planning and stability studies. Among the most successful
deterministic methods, evolutionary methods such as genetic
algorithms applied on the power flow solution are those that
algorithm (GA) and particle swarm optimization have been
use the Newton-Raphson method, proposed by Van Ess and
proposed for solving the OPF as alternative [24], [26]–[29].
Griffin [4]. This method underwent improvements in [5], who
Genetic algorithms offer a powerful approach to these opti-
introduced matrix techniques using sparsity as a means of
mization problems, because their global search techniques are
implementing the Newton-Raphson method for large networks.
more likely to converge toward the global solution. Also GA,
Early developments of digital power flow calculations were
simultaneously, evaluates several points in the parameter space
reviewed by Tinney and Powell [6]. Currently there are several
strictly on the objective function and do not need to assume
books and computer programs that explain and solve the power
that the search space is differentiable or continuous.
flow problem [7]–[9], [19].
This work aims to develop an OPF algorithm whose solution
Prior to the development of algorithms for the power flow
is based on the use of Genetic Algorithm. The development
solution, there was already a concern with increasing the
of such a program takes into account the fact that writing
operational efficiency of electrical power systems [1], since
nonlinear optimization programs with restrictions applied to
the costs, mainly of fossil fuels, increased the price of tariffs
electrical power systems is a cumbersome process, mainly
of electricity. In order to optimize operating costs, in [2]
due its dimensions. Also, the use of evolutionary algorithms
the economic dispatch of generating units was presented.
in OPF solution is not such a widespread idea in Brazil if
The economic dispatch problem, as it takes only generation
compared to deterministic methods. To validate the developed
costs, was extended by Carpentier [3], in order to take into
program, Matpower [19], which uses interior point method to
account not only the generation costs, but also the losses in
solve OPF will be used. Two test systems are used for this.
the transmission systems. That’s where the term optimal power
The document is structured as follows: in Section II, the
flow [10], [12]–[14] arose and since then, the use of new
problem formulation is presented. In this Section the conven-
techniques for solution have been proposed over the years
tional power flow and the optimal power flow are described.
The author thanks FAPESC (Fundação de Apoio à Pesquisa e Inovação do In Section II-C, a general ideia of the developed Genetic
Estado de Santa Catarina). Algorithm Optimal Power Flow (GAOPF) is presented. In
Section III, the 3-bus and 30-bus test systems are presented Each bus provides two equations and four unknowns, which
together with the obtained results, which validates the devel- means that two variables per bus must be specified to solve the
oped GAOPF program. Finally, comments and conclusions are resulting load flow equations. Depending on which variables
presented. are specified, three main types of buses can be distinguished:
sp
Load or PQ buses: Both active PD i
and reactive Qsp
Di power
II. P ROBLEM FORMULATION absorbed by the sum of loads connected at the bus are
In this section, the basic idea and formulation of the power specified. It must be verified if the difference between the
flow and optimal power flow are presented. Also the basic idea specified power and the evaluated power through equations is
of how to implement a Genetic Algorithm based Power Flow null. This leads to the following bus constraints:
is explained.
Pi = Pisp = −PD
sp
i
; Qi = Qsp sp
i = −QDi . (3)
A. Basic Power Flow Problem
leaving the phasor voltage components, V̇i = Vi ∠θi , as the
The load flow or power flow problem consists of finding remaining unknowns.
the steady-state operating point of an electric power system. Generation or PV buses or also voltage controlled bus:
The the aim is to obtain all bus voltages and complex power These are buses where the voltage regulator of a local gener-
flowing through all network components such as generating ator keeps the voltage magnitude to a specified value (Vspi ).
units, transmission lines and loads. The power flow is the Furthermore, the active power PGspi injected by the generator
most widely used application in electric power systems, both is specified. This specification may be according to certain
in operating and in planning environments, either as a stand- economic criteria. Taking into account the possible local load
alone tool or as a subroutine within more complex processes demand, the resulting net constraints are:
such as stability analysis, optimization problems, etc. It can be
formulated as a set of nonlinear algebraic equality or inequality Pi = Pisp = PGi
sp sp
− PDi
; Vi = Visp . (4)
constraints. These constraints represent both Kirchhoff’s laws
and network operation limits. leaving Qi and θi as unknowns. A particular case of PV bus
arises when a reactive power compensator (rotating or static),
Pi (V, θ) = PGi − PDi ∀i = 1, m (1a) equipped with voltage regulator, is connected to a bus. In such
sp
Qi (V, θ) = QGi − QDi ∀i = 1, m (1b) case, PGi = 0.
PGimin ≤ PGi (V, θ) ≤ PGimax ∀i = 1, NG (1c) If only those two types of buses were considered, all injected
complex powers should be specified a priori, which requires
QGimin ≤ QGi (V, θ) ≤ QGimax ∀i = 1, NG (1d)
that active and reactive losses be known in advance. However,
Vimin ≤ Vi ≤ Vimax ∀i = 1, m (1e) power losses depend on the resulting power flows and cannot
be accurately determined until the load flow itself is solved.
where
Therefore, the complex power of at least a generator should
PGi : is the generator real power output at to bus i;
be left as an unknown. For convenience, the voltage phasor of
QGi : is the generator reactive power output at to bus i;
the generating bus whose complex power remains unspecified
PDi : is the real load power at bus i;
is taken as reference for phase angles. This particular PV bus,
QDi : is the reactive load power at bus i;
is the third type of bus and it is known as the slack or swing
Pi : is the real net power injection at bus i;
bus. Such bus is usually chosen among those generating buses
Qi : is the reactive net power injection at bus i;
with largest capacity, frequently being in charge of frequency
Vi : is the voltage magnitude at bus i;
regulation duties. For the slack bus, the complex voltage is
θi : is the voltage angle at bus i;
fully specified whereas both power components, active and
NG is the number of generators on the system;
reactive, belong to the set of unknowns. Thus, the solution
m is the number of power system buses.
of the set of nonlinear algebraic equations (1a) − (1e), is
Also (1a) and (1b) are the power flow equations represented,
only possible if there exists a slack bus. On the slack bus
respectively, as follows:
the generation level is calculated so as to meet the active and
m
X reactive power balance of entire power system using (5), with
Pi (V, θ) = Vi Vj (Gij cosθij + Bij sinθij ) (2a) Si = Pi ± jQi :
j=1
m m
X m
X NX
G−1
X
Qi (V, θ) = Vi Vj (Gij sinθij − Bij cosθij ) (2b) SGenSlack = SDj + Slossj − SGenj (5)
j=1 j=1 j=1 j=1

where [8], [9]: Such condition, if attended, assures that the power system
Gij : is the series conductance between buses ij; operates at its nominal frequency 60 Hz or 50 Hz. The slack
Bij is the series susceptance between buses ij; bus concept is a mathematical artifact, without any direct link
The inequalities (1c) − (1d) reflect the limits on physical to the real world, as no bus in the system is explicitly in charge
devices and (1e) the limits created to ensure system security. of balancing (5).
 In order to solve the power flow problem, several itera- C. Genetic Algorithm Optimal Power Flow Problem
tive methods where employed through the years, but one of
The Genetic Algorithms (GAs) are part of the evolutionary
the most effective is the Newton-Raphson method [5], [7],
algorithms family which in turn are inspired in the Nature.
[8]. This method successively improves the unknown values
Such algorithms are powerful stochastic search tools based on
through first-order approximations of the set of nonlinear alge-
natural selection and genetics [22], [23]. They offer a powerful
braic equations. For initial guess values x0 , close to the solu-
approach to nonlinear constrained optimization problems solu-
tion, also known as flat start, i.e., V̇i = Vi ∠θi = 1, 0∠0 pu, the
tions such as ACOPF and have found extensive applications in
Newton–Raphson method converges quadratically. Irrespective
solving global optimization searching problems. Their advan-
of the network dimension, starting from the flat voltage profile,
tage relies on the evaluation of many points in the parameter
takes from three to five iterations to attain convergence [11].
space and such search space does not need to be differentiable
B. Optimal Power Flow Problem or continuous as methods in [17], [18].
An OPF is an optimization problem applied to power flow, To solve ACOPF with a simple GA, some details must
which determines the optimal settings of the power system be addressed: First it is necessary to generate an initial
variables in a way that the equality and inequality constraints binary string (chromosome) which associates a binary code
are satisfied [3], [10], [12], [14]. OPF objective function and to each active power (PGi ) for all the system generators. An
constraints can be formulated as (6a) − (6f) initial population is created with several (PGi ∈ PGimin <
NG PGi < PGimax ) and these chromosomes originate the initial
X
min F = fi (PGi ) (6a) search space or population. Then the initial population passes
i=1 through the processes of reproduction, crossover, mutation and
s.t.: Pi (V, θ) = PGi − PDi ∀i = 1, m (6b) immigration so that new intermediate populations are created.
In the reproduction process, the chromosomes (or possible
Qi (V, θ) = QGi − QDi ∀i = 1, m (6c)
combinations of solutions PGi ) are evaluated in the objective
PGimin ≤ PGi (V, θ) ≤ PGimax ∀i = 1, NG (6d) function and the best chromosomes are selected. These se-
QGimin ≤ QGi (V, θ) ≤ QGimax ∀i = 1, NG (6e) lected chromosomes (sets of PGi ) then pass to the crossover
Vimin ≤ Vi ≤ Vimax ∀i = 1, m (6f) process where segments of their bits are exchanged with each
other to generate their offspring. The chromosomes with larger
where equality and inequality constraints, functions and vari- scores have more chances to be mixed with other ones.
ables are the same defined in the Basic Power Flow Problem, Mutation is the secondary operator and prevents the pre-
Section II-A. There are several kinds of objective functions, mature stopping of the algorithm in a local solution. The
among the most common are: Generation cost objective func- mutation operator is defined by a random bit value change
tion, Active power loss objective function, Pollutant Gases in a chosen string (chromosome) with a low probability of
Emission objective function and combinations of the men- such change. The mutation adds a random search character to
tioned ones in order to obtain multi-objective functions [29]. the genetic algorithm, and it is necessary to avoid that, after
In this work, the generator cost function was chosen to
some generations, all possible solutions become very similar
minimize the total energy production cost of the power system.
ones [23], [25].
This kind of objective function for the OPF reflects the
Another procedure which may be used in GA is the process
costs associated with generating power in the system. The
of immigration [30]. This process raises the fitness of the
quadratic cost model for thermal power generation will be
population as a whole and enriches its diversity, i.e., random
used and is defined in (7), fi (PGi ) represents the active power
chromosomes are created and added to the population in order
generation cost by each generator and ai , bi and ci , the ith
to improve its diversity and the possibility of finding a better
cost function coefficients of the quadratic cost model at each
minima.
thermal generator.
After these procedures, the resulting chromosomes (PGi )
2
fi (PGi ) = ai + bi PGi + ci PGi (7) that belong to the next Generation/population are tested on
Such objective function (6a), where NG is the total number the objective function. If this is the optimum Generation, the
∗
of generation units, is used to minimize the total system costs best/optimal chromosome (PGi ) is chosen and the process is
associated with active power generation PGi , and does not stopped as the optimum solution is found, otherwise it is nec-
necessarily minimize the costs for a particular area within the essary to repeat the steps above. In Fig. 1 there is GA-Based
power system. Optimization Approach Flowchart. Finally, if is necessary to
As a nonlinear optimization problem, several solution meth- know all the active and reactive power flowing through the
ods have been proposed through the years [5], [12], [14]–[18]. system, a conventional power flow must be performed. All
In order to solve the classical OPF problem, in this work the simulations by GAOPF on this work have used the Matlab
the Matlab Toolbox Matpower [19] was used for comparison ga(x) function as well all as the following control parameters:
purposes. Matpower has interior point method for nonlinear • Population size: 50
optimization [18], [20], [21] as default algorithm for the OPF • Number of generations: 300
solution. • Crossover probability: 0.8 - Matlab default
 Generators data and Cost coefficients are given on Table I.
According to the presented nomenclature, Section II-B, the
nonlinear optimization problem (6a)-(6f), has NG = 2, m = 3
and the objective function is a sum of two distinct second order
polynomials given by (7).
The prime objectives of using the 3 Bus system is to
compare results of an optimal load-flow solution, which gives
the optimal active and reactive power dispatch for a static
power-system loading condition, with the ones obtained by
the developed Genetic Algorithm Optimal Power Flow. This
comparison is made in order to validate the developed GAOPF,
besides understanding some important aspects of the OPF
problem. The validation of the GAOPF was done comparing
some crucial results with the ones obtained by the Matpower
[19].

TABLE I
3 B US S YSTEM G ENERATOR DATA

Bus PGmin PGmax QGmin−max Cost coefficients ∗

No. MW MW MVAr a b c
1 50 300 −200 to 200 0.0 2 0.00375
Fig. 1. GA-Based Optimization Approach Flowchart. 3 50 250 −200 to 200 0.0 1.75 0.0175
∗ f (P ) = a + b P 2
i Gi i i Gi + ci PGi in $/hr.

• Mutation probability: 0.01 - Matlab default The results of OPF and GAOPF are summarized in Table II
and Table III. Both optimization algorithms led to almost the
III. T EST S YSTEMS same results in active power generation as can be observed in
Table II. It is possible to note that PG1 achieved its maximum
In this section two test systems are presented in order
allowed capacity, limited by the PG1max bound, highlighted
to validate the developed GAOPF. The developed algorithm
in bold, when the using NLOPF. On the other way, PG1 ≈
objective is to optimize the active power dispatch in order to
PG1max when GAOPF was performed. Despite that, in both
reduce the cost of electricity production. The first test system
situations, the necessary condition imposed by (5) was satisfied
has 3 bus and is a fictitious electric power system, the second
[8], [9].
one is the IEEE 30 bus system [12]. Results of the developed
GAOPF and Matpower [19] are compared. Matpower is a
software whose main purpose is to solve optimal power flows, TABLE II
P OWER G ENERATION - 3 B US S YSTEM
using by default, the interior point method for that.
Optimization G1 G3 G1 G3
A. 3 Bus System Method MW MW MVAr MVAr
NLP-OPF 300.00 119.47 90.06 199.79
The 3 Bus test system is the same one presented in [9]. It GA-OPF 299.14 120.35 112.95 176.94
is a fictitious system, composed by 2 synchronous generators,
3 branches and a single load located at bus 2. Bus 1 is the
slack bus, the PV (voltage controlled bus) is bus 3 and the PQ Also the voltage magnitudes of the 3 Bus system have
bus is bus 2. The one-line diagram is given in Fig. 2. almost the same profile, as seen on Table III. It is important
to note that a slight difference between |V̇3 | voltages in
OPF and GAOPF implies in a quite significant change on
the reactive power generated by G1 and G3 , Table II. In
both cases the imposed bus voltage inequality constraints are
0.95 pu < Vm < 1.05 pu and there were not violations at the
end of the optimization processes.
Recalling (5) is possible to verify the active and reactive
power system losses (Sloss ). Such power losses agree with
the ones presented on Table IV. Also it is easy to see why the
production cost ($/hr) calculated by the GAOPF is bigger than
Fig. 2. 3 Bus Test System.
the one calculated by NLOPF. The c cost coefficient, Table I,
2
of cG3 ≈ 4, 66cG1 and it is proportional to PG3 , Table II.
 TABLE III respective Cost Coefficients to be used. These data are used
3 B US S YSTEM B US VOLTAGES on inequality constraints (6d), (6e) and also on the objective
Optimization |V̇1 | |V̇2 | |V̇3 | function (6a), respectively.
Method pu pu pur
NLP-OPF 1.05 0.975 1.044
TABLE V
GA-OPF 1.05 0.972 1.040
IEEE 30 B US S YSTEM G ENERATOR DATA

Bus PGmin PGmax QGmin−max Cost coefficients ∗

TABLE IV No. MW MW MVAr a b c
3 B US S YSTEM P RODUCTION C OST AND L OSSES 1 50 200 −20 to 250 0.0 2 0.02
2 20 80 −20 to 100 0.0 1.75 0.0175
Optimization Total Cost PLosses QLosses 5 15 50 −15 to 80 0.0 1.0 0.0625
Method $ /hr MW MVAr 8 10 35 −15 to 60 0.0 3.25 0.00834
NLP-OPF 1396.00 19.47 39.84 11 10 30 −10 to 50 0.0 3.0 0.025
GA-OPF 1397.90 19.46 39.88 13 12 40 −15 to 60 0.0 3.0 0.025
∗ f (P ) = a + b P 2
i Gi i i Gi + ci PGi in $/hr.

B. IEEE 30 Bus System The optimal power flow was performed with Matpower
An adaptation of the IEEE 30-bus standard test-system, (NLP-OPF) and also with the developed algorithm (GA-OPF).
Fig. 3, is used to compare the Matpower [19] OPF solution On Table VI it is possible to compare the optimal resulting
with the developed GAOPF algorithm in Matlab. The data active power generation by generator and method. The bold
used to represent the 30 Bus System may be found in [12]. numbers in Table VI indicate that the maximum active power
The objective function to be minimised is again the total (PGmax ) was reached and limited by its respective constraints
system active-power generation cost, allowing all the generator presented on Table V. Optimal MW-outputs achieved by both
(PGi ) MW-outputs to be controllable and considered as one algorithms were quite similar. It can be observed that only G1
of the inequality constraints. The IEEE 30 Bus System has and G5 are far from their respective maximum active power
6 generators, 41 branches and its total amount of load is limits, mainly because of their contribution to the total amount
SD = 283.4 + j126.2 MVA. Observing the nonlinear opti- cost of energy production.
mization problem, set of equations (6a)- (6f), with NG = 6
and m = 30 it is possible to determine the problem dimension TABLE VI
and envision the complexity that may arise when applying P OWER G ENERATION - IEEE 30 B US
nonlinear optimization methods in a real electric power system Optimization G1 G2 G5 G8 G11 G13
such as the Brazilian Interconnected Power System (BIPS). Method MW MW MW MW MW MW
NLP-OPF 70.95 80.00 33.17 35.00 30.00 38.88
GA-OPF 70.12 80.00 32.85 35.00 30.00 40.00

The IEEE 30 Bus System voltage profiles, by bus number

and algorithm, are presented in Fig. 4. The voltage profiles
are very similar, which implies that the reactive power (QGi )
injected by each generator, established by the optimization
method, in the system also have close values. It is also
important to mention that the voltage inequality constraints
(0.95 pu < Vm < 1.085 pu) were not violated at the end of
the optimization processes.

Voltage Profile Comparison

1.1
NLP-OPF
1.08 GA-OPF

1.06
Bus Voltage [pu]

1.04

1.02

0.98

Fig. 3. IEEE 30 Bus Power System. 0.96

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
Bus Number

The 30 Bus System generator data are presented on Table V. Fig. 4. Voltage Profile by Bus - IEEE 30 Bus System.
All the complex power (SGi ) limits are presented as the
 As the active and reactive power generated and also the [8] J. Grainger and W. Stevenson, ”Power System Analysis”, McGraw-Hill
voltage profiles are very similar, Education; 1st edition, January 1, 1994.
P6 it is expected that the total [9] H. Saadat, ”Power Systems Analysis”, Mc Graw Hill; 2nd edition, 2002.
cost of energy production i=1 fi (PGi ) on the objective [10] H. W. Dommel and W. F. Tinney, ”Optimal Power Flow Solutions,” in
function (6a) will also be. The minimum optimal energy IEEE Transactions on Power Apparatus and Systems, vol. PAS-87, no.
production cost are shown on Table VII as well as active and 10, pp. 1866-1876, Oct. 1968, doi: 10.1109/TPAS.1968.292150.
[11] B. Stott, Effective starting process for Newton-Raphson load flows, IEE
reactive power losses through the 30 Bus System. Proceedings, 118, 1971, 983–987.
[12] O. Alsac and B. Stott, ”Optimal Load Flow with Steady-State Security,”
TABLE VII in IEEE Transactions on Power Apparatus and Systems, vol. PAS-93,
IEEE 30 B US S YSTEM P RODUCTION C OST AND L OSSES no. 3, pp. 745-751, May 1974, doi: 10.1109/TPAS.1974.293972.
[13] H. H. Happ, ”Optimal power dispatch: A comprehensive survey,” in
Optimization Total Cost PLosses QLosses IEEE Transactions on Power Apparatus and Systems, vol. 96, no. 3, pp.
Method $ /hr MW MVAr 841-854, May 1977, doi: 10.1109/T-PAS.1977.32397.
NLP-OPF 987.37 4.59 22.42 [14] D. I. Sun, B. Ashley, B. Brewer, A. Hughes and W. F. Tinney, ”Optimal
GA-OPF 986.63 4.51 22.46 Power Flow By Newton Approach,” in IEEE Transactions on Power
Apparatus and Systems, vol. PAS-103, no. 10, pp. 2864-2880, Oct. 1984,
doi: 10.1109/TPAS.1984.318284.
[15] A., Jr. Santos 1 and G.R.M. da Costa , ”Optimal-power-flow solution
IV. C ONCLUSIONS by Newton’s method applied to an augmented Lagrangian function”
Source: Volume 142, Issue 1, January 1995, p. 33 – 36, doi 10.1049/ip-
Results of the developed GA based OPF has been shown gtd:19951586
in this work. Solutions to the OPF problem has been applied [16] Xihui Yan and V. H. Quintana, ”Improving an interior-point-based
OPF by dynamic adjustments of step sizes and tolerances,” in IEEE
and presented to a 3 bus and to the IEEE 30 bus electric Transactions on Power Systems, vol. 14, no. 2, pp. 709-717, May 1999,
power systems. The developed algorithm main objective is to doi: 10.1109/59.761902.
optimize the active power dispatch. Results such as optimal [17] J. A. Momoh, R. Adapa and M. E. El-Hawary, ”A review of selected
optimal power flow literature to 1993. I. Nonlinear and quadratic
fuel costs, complex generated power, voltage magnitudes as programming approaches,” in IEEE Transactions on Power Systems, vol.
well systems losses were compared with the ones obtained 14, no. 1, pp. 96-104, Feb. 1999, doi: 10.1109/59.744492.
by Matpower, which solves optimal power flows using an [18] J. A. Momoh, M. E. El-Hawary and R. Adapa, ”A review of selected
optimal power flow literature to 1993. II. Newton, linear programming
interior point method. Comparison between both algorithms and interior point methods,” in IEEE Transactions on Power Systems,
results shown that all were quite similar. An advantage of vol. 14, no. 1, pp. 105-111, Feb. 1999, doi: 10.1109/59.744495.
GA over interior point optimization method is the fact that [19] R. D. Zimmerman, C. E. Murillo-Sanchez (2020). MATPOWER (Ver-
sion 7.1) [Software]. Available: https://matpower.org
GA is derivative free and may be applied to nonconvex cost [20] Anders Forsgren, P. E. Gill, M. H. Wright, ”Interior Methods for
functions, besides the possibility of being coded to work on Nonlinear Optimization”, SIAM REVIEW, Vol. 44, No. 4, pp. 525–597,
parallel computers. On the other way, the main disadvantage 2006.
[21] Capitanescu, F.; Glavic, M.; Ernst, D.;Wehenkel, L., ”Interior-point
of GA is its stochastic nature, which means that not always based algorithms for the solution of optimal power flow problems.”
the solution they provide to the OPF problem is guaranteed Electr. Power Syst. Res. 2007, 77, 508–517.
to be the optimum. As compared with the results obtained by [22] J. H. Holland, ”Adaptation in natural and artificial systems”. Ann Arbor,
MI: The University of Michigan Press; 1975.
the interior point method, which always finds the same deter- [23] D. E. Goldberg, ”Genetic Algorithms in Search, Optimization and
ministic solution, the GAOPF optimal solutions are always in Machine Learning”, Addison Wesley Publishing Company, Ind. USA,
a region that presents small deviations at the vicinity of the 1989.
[24] A. G. Bakirtzis, P. N. Biskas, C. E. Zoumas and V. Petridis, ”Optimal
optimum solution points. power flow by enhanced genetic algorithm,” in IEEE Transactions on
Power Systems, vol. 17, no. 2, pp. 229-236, May 2002, doi: 10.1109/TP-
ACKNOWLEDGMENT WRS.2002.1007886.
The author thanks FAPESC (Fundação de Apoio à Pesquisa [25] T. Bouktir, L. Slimani, M. Belkacemi, ”A Genetic Algorithm for Solving
the Optimal Power Flow Problem”, Leonardo Journal of Sciences, Issue
e Inovação do Estado de Santa Catarina) for all support. 4, January-June 2004
[26] D. Devaraj and B. Yegnanarayana, ”Genetic-algorithm-based optimal
R EFERENCES power flow forsecurity enhancement”, IEE Proc.-Gener. Transm. Dis-
[1] STAHL, E. C. M. Load division in interconnections. Electrical World, trib., Vol. 152, No. 6, November 2005
Brooklyn, p. 434–438, 1930. [27] J. Yuryevich and Kit Po Wong, ”Evolutionary programming based
[2] KIRCHMAYER, L. K. Economic operation of power systems. New optimal power flow algorithm,” in IEEE Transactions on Power Systems,
York: Wiley, 1958. vol. 14, no. 4, pp. 1245-1250, Nov. 1999, doi: 10.1109/59.801880.
[3] Carpentier, J. Contribution á l’étude du dispatching économique. Bulletin [28] M.A. Abido, ”Optimal power flow using particle swarm optimization”,
de la Societe Francaise des Electriciens, London, v. 3, n. 1, p. 431–447, Electrical Power and Energy Systems 24, 2002
1962. [29] S. Kahourzade, A. Mahmoudi, H. B. Mokhlis, ”A comparative study of
[4] Van Ess, J. E. and Griffin, J. H., ”Elimination methods for load flow multi-objective optimal power flow based on particle swarm, evolution-
studies”, AlEE Transactions, Vol. 80, pp. 299-304, 1961. ary programming, and genetic algorithm, Electrical Engineering (Archiv
[5] W. F. Tinney and C. E. Hart, ”Power Flow Solution by New- fur Elektrotechnik) Volume 97 issue 1 2015.
ton’s Method,” in IEEE Transactions on Power Apparatus and [30] F. Ornelas, M. Meza, A. Padilla, F. Padilla, J. Ponce and A. Ochoa, ”Ge-
Systems, vol. PAS-86, no. 11, pp. 1449-1460, Nov. 1967, doi: netic Algorithm with Immigration Like Strategies of Diversification,”
10.1109/TPAS.1967.291823. 2010 Ninth Mexican International Conference on Artificial Intelligence,
[6] Tinney, W. F. and Powell, W. L., ”Notes on Newton-Raphson method 2010, pp. 11-15, doi: 10.1109/MICAI.2010.33.
for solution of AC power flow problem”, IEEE Short Course, Power
Systems Planning, 1971.
[7] G.W. Stagg (Autor), A.H. El-Abiad, ”Computer Methods in Power
System Analysis”, McGraw-Hill Inc.,US - 1968.