Electrical Engineering

https://doi.org/10.1007/s00202-020-00974-z

ORIGINAL PAPER

Calculating the locational marginal price and solving optimal power

flow problem based on congestion management using GA-GSF
algorithm
Masoud Dashtdar1 · Mojtaba Najafi1 · Mostafa Esmaeilbeig1

Received: 26 September 2019 / Accepted: 6 March 2020

© Springer-Verlag GmbH Germany, part of Springer Nature 2020

Abstract
An important factor in reviewing the performance of generation units and calculating their profit is the calculation of their
locational marginal price (LMP), and this depends on our knowledge on the capacity of transmission lines and optimal power
flow (OPF) based on reality by which we aim to minimize the total cost of the generators, solve the congestion of transmission
lines, and hence, reduce the price of electricity in the market. Since power flow equations are nonlinear, they should be
solved using numerical and repetition-based methods. In this paper, genetic algorithm (GA) has been employed to solve these
equations, and in order to improve the performance of GA in its structure, generating scaling factor (GSF) has also been
used for simultaneous calculations of power passing through in transmission lines so that by gaining some knowledge on the
capacity of transmission lines, in addition to optimal power flow becoming real, we could determine the electricity price by
uniform market pricing or LMP methods depending on using the full capacity of lines and generating power of the units, and
thus, we can calculate the profit of generators. Finally, the output of the proposed GA-GSF algorithm would include values
of buses voltages, lines losses, injected power to buses, power passing through lines, total generation cost, and generators’
profits. Also, the proposed algorithm in this paper has been tested on IEEE 14-BUS, IEEE 30-BUS, IEEE 57-BUS network,
and the results show improvements on the OPF problem.

Keywords Locational marginal price · Optimal power flow · Congestion management · Genetic algorithm · Generating
scaling factor

1 Introduction and reactive powers, power losses, power exchange between

different power systems, generation and consumption bal-
Power systems are usually employed in three balanced phases ance in the system, transmitted powers, and other features
and in stable sinusoidal conditions. In such a system, infor- gained from measuring current and voltage in different parts.
mation on power flow through lines and voltages in different In power flow calculations, the node voltage method which is
buses in the system is very important for power system oper- the most appropriate method in many power system analyses
ators. The goal of power flow studies is to gain knowledge on is used. Using this method will lead to a series of complex
the way generated power is distributed on network lines and algebraic equations based on nodes currents, and if we know
to determine the voltage of buses, etc. Power flow parameters the currents, we can solve these equations to obtain the nodes
under study include items such as voltages, currents, active voltages, and since these equations are nonlinear, they should
be solved with numerical and repetition-based methods. Two
B Mojtaba Najafi common methods in solving these equations are Gauss–Sei-
[email protected]
del and Newton–Raphson methods. Power flow studies are
Masoud Dashtdar the main bone of power system design and analyses, and
[email protected]
conducting these researches is essential for utilization and
Mostafa Esmaeilbeig planning economic scheduling between power companies
[email protected]
[1].
1 Electrical Engineering Department, Bushehr Branch, Islamic In power systems and due to the existence of elements such
Azad University, Bushehr, Islamic Republic of Iran as resistance, inductor, and capacitor, numerical methods are

123
 Electrical Engineering

mostly used to solve nonlinear equations in these systems. best response. This process reduces the convergence speed,
In these methods, the equations are solved in sequential rep- makes the power flow real, and also reduces the costs com-
etitions and finally, by reaching a converging point, the end pared to other methods. In the following and after introducing
of the calculation. Also, in order to start the calculations GA-GSF algorithm, a 24-h power flow was conducted on an
in power flow equations, some guesses should be made for IEEE 14-BUS network and eventually by determining the
unknown parameters. If these primary values were guessed electricity price through UMP or LMP, the profit of gener-
more accurately and were close to the actual power system ators is calculated and the results are compared with other
performance, the equations would converge sooner. methods such as economic dispatching (ED), quadratic pro-
Usually, optimal power flow (OPF) is used for studies on gramming using Lagrangian coefficients. The results indicate
overload fix in transmission lines, transmission systems con- the proper performance of the GA-GSF algorithm, compared
trol, calculating the available transmission capacity (ATC), to other algorithms.
lines congestion management, determining the price of active Organization of this article is as follows: In the second part
and reactive power and locational marginal price (LMP). In of this article, we introduce the problem of optimal power
OPF, control variables are determined in a way that the objec- flow and how to formulize of this problem; in the third part,
tive variable is minimized, and simultaneously, a series of the implementation procedures of power flow process in the
equal constraints (i.e., power flow equations) and unequal proposed algorithm are defined, and we explain how to deter-
constraints (including control and utilization limitations) are mine the market place for electricity and how to calculate the
established. In this series, the objective function equals the profits of generators. In the 4th part, the optimization algo-
total cost of all of the generators and control variables are rithm of GA-GSF is presented; in the 5th part, the results of
the active powers produced by generators and state vari- the implementation of the proposed algorithm are presented
ables are the domain and angle of buses’ voltages. Also, and conclusions are made in part 6.
unequal constraints include minimum and maximum limi-
tations of control variables, minimum and maximum voltage
of buses, and maximum transmission powers of lines [2,
3]. Many classical optimization techniques such as linear
2 Defining the problem of OPF
programming [4], quadratic programming [5, 6], gradient
The goal of OPF is to minimize the total fuel cost of thermal
and Lagrange methods, and also many artificial intelligence
units and also to maintain optimal system performance in
(AI) techniques such as classical evolutionary programming
terms of limitation in generator’s active and reactive power
(CEP) [7], improved fast evolutionary programming (IFEP)
output, bus voltage, capacitor/shunt reactor, transformer tap
[7], genetic algorithm (GA) [8], particle swarm optimization
adjustment, and power flow of transmission lines. In fact,
(PSO) [9–20], differential evolution (DE) [21], ant colony
OPF is an important issue in economic evaluation of power
optimization (ACO) [22] are used to solve optimal power
systems that can be defined as an optimization problem con-
flow problem. Usually, classical optimization methods do not
sisted of the objective function and some constraints and
perform well due to congestion in local minimum points and
its limitations are related to physical laws governing trans-
early convergence and although metaheuristic methods are
mission systems and energy generation and utilization and
successful in finding the optimal answer, they are relatively
equipment limitations.
slow in reaching convergence.
Genetic algorithm (GA) is among the evolutionary algo-
rithm that employs techniques inspired by natural evolution 2.1 Problem formulation
such as heredity, crossover, selection, and intercourse to find
solutions for optimization problems. In Refs [23–26] and OPF problem’s standard can be written as following (1):
[27–29], genetic and PSO algorithms have been used to solve
optimal power flow problems.
Min {F(x)}, h(x)  0; g(x) ≥ 0 (1)
Genetic algorithm is one of the strongest methods in solv-
ing optimization problems of power systems, but due to
operational processes in it, it has a high processing speed. in which F(x) is the objective function, h(x) indicates equal
In this paper and in order to improve GA in solving the prob- constraints, g(x) is unequal constraints, and x is the vector
lem of optimal power flow, generating scaling factor (GSF) of control variables which gets changed by the control cen-
was used in its structure not only to prevent congestion in ter operator (including active and reactive power, amount of
minimum local points but also to act in a way that calcu- generation bus voltage, transformer taps, etc.). The nature of
lates the power flow for each change in control variables, the OPF problem is to reduce the objective function and to
and in case of violation of lines capacity, the algorithm will simultaneously establish power flow equations (equal con-
exit this optimal point and continues this process to reach the straint) without violating unequal constraints.

123
Electrical Engineering

2.2 Objective function 2.5 Types of unequal constraints

The most important goal in the formulation of OPF is the OPF’s unequal constraints reflect the limitations in physical
minimization of the total cost of active power generation. equipment of the power system and also limitations cre-
Also, the cost function of each generation unit depends on its ated for the system’s reliability. The most common types
generated active power and it is shown as a quadratic curve. of unequal constraints are high voltage limitation in buses
Then, the objective function of the entire system is obtained connected to power and generation units and low voltage lim-
by summing the cost functions of each generator (2). itation in buses connected to power. Generation limitations
include maximum and minimum generated active power in
Ng  
 generators, maximum capacity of transmission lines, and
F(x)  ai + bi Pgi + ci Pgi2 (2) limitation in tap adjustment of TCULs and phase change.
i1
Unequal constraints of problem variables include:
in which N g is the number of generation units which include High and low bands of active power generated in buses
slack bus, Pgi is the generation active power in bus number that have generators (8).
i, and a, b, and c are the coefficients of the cost function for
each generator. min
PGi ≤ PGi ≤ PGi
max
(8)

2.3 Types of equal constraints

High and low bands of reactive power generated in buses
In minimizing the cost function, we should make sure that that have generators and reactive power injection in buses
the production equals power demand and transmission losses. with VAR compensator (9) and (10):
Therefore, power flow equations are considered as equal con-
straints (3).
Gi ≤ Q Gi ≤ Q Gi
Q min max
(9)
   
Pi Pi (V , θ ) − (PGi − PDi )
 0 (3)
Q i Q i (V , θ ) − (Q Gi − Q Di ) min
Q Ci ≤ Q Ci ≤ Q Ci
max
(10)
Injected active and reactive power in the bus i is defined
as following equations (4) and (5): High and low bands of voltage amount in all buses (11)
and (12):

Ns
 
Pi (V , θ )  Vi V j G i j cos θi j + Bi j sin θi j (4)
j1 min
VGi ≤ VGi ≤ VGi
max
(11)
Ns
 
Q i (V , θ )  Vi V j G i j sin θi j − Bi j cos θi j (5)
j1
min
VLi ≤ VLi ≤ VLi
max
(12)

in which Gij is conductance, Bij is susceptance, Vi is the

magnitude of voltage in the bus i, and θ ij is the voltage angle High and low bands in phase angle of bus voltage (13):
of the bus phase.
θGi
min
≤ θGi ≤ θGi
max
(13)
2.4 Power balance limitation

One of the important factors in the OPF is the power loss Maximum passing capacity for each transmission line
in the network which reduces the power from generation to can be defined based on MW/MVAR/MVA considering the
consumption. Therefore, the power produced by the gener- conductor’s temperature or because of concerns for system
ator is equal to power consumption and network losses (6) stability (14):
and (7).


N Pi, j ≤ Pi,max
j (14)
Pi  P D + PLoss (6)
i1
Now we can see that the objective function F is a nonlinear

Nbus 
Nbus
 
PLoss  |Vi | V j Ybusi j cos θbusi j − δi + δ j (7) function and the number of equal and unequal constraints
i1 j1 increases depending on the network size.

123
 Electrical Engineering

⎡ ⎤
3 Implementation stages of OPF process S1,2 . . . S1, j
and determining the market price ⎢ .. .. ⎥
⎢. . ⎥
of electricity ⎢ ⎥ j  2, . . . , Nbus
SFI  ⎢ Sk,2 . . . Sk, j ⎥
⎢
⎥, (21)
⎢. . ⎥ i  1, . . . , Nbus
3.1 Implementing the process of optimal power flow ⎣ .. .. ⎦
Si,2 . . . Si, j
Here, the solution to the power flow problem is divided into 5
stages that are implemented by the proposed algorithm. The 5. Repetitive calculations of the following stages to reach
stages are as follows: convergence:
(A) Forming a broad form of the problem using a Jaco-
1. Per-unitization of values and forming network’s admit- bian matrix:
tance matrix (Y bus ) (15):    
δi −1 Pi
⎡ ⎤  [J ] (22)
Y11 . . . Y1 j . . . Y1n  Vi Q i
⎢ .. .. .. ⎥
⎢. . . ⎥ B. Calculating power changes:
⎢ ⎥
Ybus  ⎢ Yk1 . . . Yk j . . . Ykn ⎥
⎢
⎥ (15)
⎢. .. .. ⎥ 
Ns
 
⎣ .. . . ⎦ Pi  |Vi ||Yin | V j cos δ j − δi + θin + P Di ,
Yn1 . . . Yn j . . . Ynn j1

Pi  Piold − Pi (23)

2. Forming the table of initial voltage values (state vari-
ables) (16): 
Ns
 
Qi  − |Vi ||Yin | V j sin δ j − δi + θin + Q Di ,
⎡ ⎤ ⎡ ⎤
|V1 | δ1 j1
⎢ .. ⎥ ⎢ .. ⎥ Q i  Q iold − Q i (24)
|Vbus |  ⎣ . ⎦, δbus  ⎣ . ⎦ (16)
|Vn | δn
C. Calculating the power passing through lines: Note
that lines losses have been considered.
3. Forming Jacobian matrix (17): In this stage, MVAR of ⎡ ⎤ ⎡ ⎤
buses with generators is tested and if the voltage value P1,2 P I2
⎢ ⎥ ⎢ ⎥
was not in the range, in the position of 0.01 pu, the gen- Pi, j  ⎣ ... ⎦  SFI × ⎣ ... ⎦, N  2, . . . , Nbus
erator’s MVAR will be placed in the specific range. Pi, j P IN
⎡⎧ ⎤ (25)
⎧
⎨ ∂ Pi ⎨ ∂ Pi
P Ii  Pi − P Di − PLoss (26)
⎤ ⎢ ⎥
∂δi ∂ Vi
⎡ ⎢ ⎩ ∂ Pi ⎥
⎢ ⎩ ∂ Pi ⎥
⎦⎢ ⎥
H M ∂δn ∂ Vn
[J ]  ⎣ ⎢⎧ ⎧ ⎥ (17) Eventually, these calculations will repeat until the gen-
N L ⎢ ⎨ ∂ Qi ⎨ ∂ Qi ⎥ eration power limit of generators and limits of power flow
⎢ ∂δi ∂ Vi ⎥
⎣ ⎦ through lines are not violated.
⎩ ∂ Qi ⎩ ∂ Qi
∂δn ∂ Vn
3.2 Calculating the market price of electricity
4. Forming the GSF matrix: Using this matrix, we can relate
the changes in the power passing through the lines to the After OPF calculations and obtaining the power flow in lines,
change in the power injected to buses. we can calculate the market price of electricity using two
methods. In the first method (UMP), power flow results are

  −1 obtained in a way that did not involve any congestion, and
X i, j  Bi, j (18) thus, electricity price is obtained from the total cost of work-
 ing generators. Also, in this case, the electricity price for each
 
 0 i or j  1 0 0 bus will be the same. The second method (LMP) is for times
X i, j    (19)
X i, j i or j  2, . . . , Nbus 0 X  when one or some of the transmission lines have reached
 
their maximum capacity, and thus, the electricity price for
X i,k − X j,k each bus will not be the same and the price should be calcu-
Si− j,k  (20)
xi, j lated depending on the production of generators.

123
Electrical Engineering

(A) UMP price Pi  Pimin , Pi  Pimax , i ∈ {1, 2, 3, 6, 8} (29)

In this case, the final cost of units is calculated by deriv- Here, i is an indicator for buses that have a final generator.
ing from the functions of generators’ costs and putting the The second case: for buses that do not have a generator
minimum generation power obtained in (27); considering the or those whose generator has reached their generation limits
information of generation units of the IEEE 14-BUS network (considering IEEE 14-BUS network)
given in [1], final cost of the generators for the minimum gen-
erating power will be as follows: [30] 
π j  LMP j  K i · LMPi ,
  dFi  P min   $  i
i
MCi Pi min
 , i  1, 2, 3, 6, 8 j ∈ {{1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14} − i}
dPimin MWh
(30)
(27)

Here, j is an indicator for all the buses except those with

PGenerators P1 P2 P3 P6 P8
final generators. Also, K i is the cost of 1-MW surplus con-
sumption in the bus i which is calculated by solving the
Final cost (MCi) 18.52 19.6 23 27.334 28 following equations.
Generators that have reached their generation limit in bus
j:
In the end, the electricity price (π ) in this case will be
determined based on the final cost of the more expensive
P I j  −1 (31)
generator; because if we determine the price based on the
cost of the less expensive one, more expensive generators
would cause losses and keeping them on will not be reason- Off generators or those without generators in bus j:
able. Also to keep the market price low, electricity price has
been determined based on the least generation power of the P I j  0 (32)
generators.
    In case of full capacity in each line:
π  max MCi Pimin $/MWh (28)
⎡ ⎤
P I2
⎢ ⎥
(B) LMP price PLine  0 → SFI × ⎣ ... ⎦  0 (33)
P I14
In case the capacity of transmission lines reaches its max-
imum, it is not possible to use the full generation capacity
Maintaining the power balance by the slack bus:
of the network’s generators; therefore, the electricity price
in network different locations will not be the same, which
is called locational marginal pricing (LMP). In fact, LMP P I1 + P I2 + P I3 + P I6 + P I8  0 (34)
means providing a surplus 1-MW load in place through the
cheapest generators that are capable of generation without Finally by calculating power changes in buses that have a
violating the transmission line’s limitations. Therefore, a way final generator (i.e., PI i )
of calculating LMP is paying attention to generators that have
not reached their upper or lower limits. These generators, K i  P Ii (35)
those that have a part of their capacity remaining, are called
final generators. Therefore, LMP in buses that have a final
There the LMP price is π  {π i , π j }.
generator will be equal to the final cost of this type of genera-
In the following, the profit of generators can be calculated
tors. Also, LMP of other buses that do not have generators or
through the following equation:
those whose generators have reached their generation limit
will depend on LMP of buses that have a final generator.
Therefore, LMP of buses should be calculated in two ways. Ωi  Pi × π − Fi (Pi ) (36)
The first case: for buses that have a final generator (con-
sidering IEEE 14-BUS network) Finally, Fig. 1 shows the flowchart of all the defined stages
in part 3 in which green blocks are the outputs of the proposed
πi  LMPi  MCi (Pi ), ∀Pimin < Pi < Pimax , algorithm.

123
 Electrical Engineering

Fig. 1 Block diagram of the stages

4 Proposed algorithm GA-GSF ators should be used and this process is repeated until the
capacity of lines is in its range. Also, network losses have
In this part and before introducing GA-GSF algorithm, the not been considered in this method. Finally, the output of
structure of ED-GSF methods, Lagrangian-GSF is defined the ED-GSF method will be the generation capability of
and in the following and in order to evaluate their per- generators, power passing from lines, and total generation
formance, considering the power demand curve in 24 h, costs.
power flow problem is distinctly implemented by these
algorithms, and finally, utilization cost and profit in 24 h
are calculated. Also, it should be noted that alongside
these methods, generating scaling factors (GSF) is used 4.2 Lagrangian-GSF
to establish relations between change in power injected
to buses with the change in power passing in network’s In this method, a Lagrange function is created as a mixture of
lines to properly manage power transmitting from branches objective function and functions of equal and unequal con-
and to obtain real power flow. Also, this matrix can straints along with proper Lagrange coefficients. The optimal
be calculated using the information of lines (line’s reac- point of Lagrange functions is the solution to the OPF prob-
tance). lem. Also to find the optimal point of Lagrange function, this
fact is used that in the optimal point, the gradient of Lagrange
4.1 ED-GSF function is zero in relation to all variables and Lagrange
coefficients. Therefore, solving the OPF problem depends on
In this method, the power supply is done based on full finding the answer to an equation system which is obtained
capacity from cheaper to more expensive generators, mean- by placing a zero for the gradient. So, in order to solve this
ing that by calculating the final cost of the generators, we system, first default values are considered for variables and
can obtain knowledge about the performance of genera- Lagrange coefficients as the default optimal point. Then, the
tors and supply power using cheaper generators and this amount of variables changes to make the gradient zero is cal-
reduces the operation costs. But sometimes, because of the culated and default values of variables get corrected and this
capacity of transmission lines, it is not possible to use the process continues until the answers converge. Now, consider
full capacity of a cheap generator. Therefore, a mecha- Eqs. (37).
nism should be designed so that the passing power from In the equation, x, beq, Ib, and ub are the vectors, A and
each line could be calculated for the amount of injected Aeq are the matrices, and c(x), ceq(x), and F(x) are the func-
power and this is possible by using the GSF matrix. So tions that change the vectors. F(x), c(x), and ceq(x) can be
first, the GSF matrix is formed according to stage 4, and nonlinear functions. In this method, constraints 1 and 2 have
in the following, we supply power using cheaper generators not been considered, constraint 3 is the indicator of power
and then we calculated the capacity of lines by stage 5c. passing capacity from each line, constraint 4 is the indicator
Now, if the capacity of lines is violated, again other gener- of total generation equal to power demand, and constraint 5

123
Electrical Engineering

indicates the lower and upper limit of generation capacity of IEEE 14 bus System
bus13
generators.
⎧ bus12 bus14
⎪
⎪ c(x) ≤ 0 (1)
⎪
⎪
bus11
⎨ ceq(x)  0 (2)
bus8
min(x)max(i)Fi (x) A · x ≤ b (3) (37) bus9
⎪
⎪
⎪
⎪ Aeq · x  beq (4) bus10
⎩ bus7
lb ≤ x ≤ ub (5) bus6
bus1 bus4
This method has been implemented on MATLAB using
fminimax command like the following:
bus5
(Algorithm type: ‘minimax SQP, quasi-Newton,
line_search’)
  bus2 bus3
x, fval, maxfval, exitflag, output, lamda
 fminimax(fun, x0, A, B, Aeq, Beq, lb, ub)
Fig. 2 Sample network

in which fun is the cost function (F T ), ×0 is the default value,

A and B, respectively, indicate GSF matrix and maximum
• Calculating the objective function corresponding to each
passing power capacity of each line, Aeq is a 1 × 5 matrix
chromosome which is the chromosome’s fitness.
with matrix element of 1, Beq equals the total power demand,
• Selecting a specific number of initial population chromo-
and Ib and ub are, respectively, the lower and upper limits of
somes based on the principle of survival of the fittest.
generation capacity of generators.
• Applying genetic operators (mutation and crossover) with
Also, lamda output includes four variables (lower ‘upper’
specific possibilities for the new population.
‘eqlin,’ ‘eqnonlin,’ ‘ineqlin,’ ‘ineqnonlin’), and the out-
• Calculating the fitness corresponding to each chromosome
put includes the number of iterations, algorithm type, etc.
of the new population, and reviewing convergence and
maxfval output indicates the generation cost, and x indi-
making a decision whether to continue or stop the algo-
cates the generation capacity of each generator. Also in
rithm.
the Lagrangian-GSF method, network losses have not been
considered and the output of this method is the generation
capacity of generators, power passing through lines, and total Genetic algorithm is an algorithm with a parallel search
generation cost. that works on a population of answers. This characteristic
NR-Lagrangian-GSF method In order to consider losses in helps the genetic algorithm search different areas of answer
the Lagrangian-GSF method, we can obtain the buses voltage space simultaneously, and this reduced the possibility of con-
using Newton–Raphson (NR) method, and finally by cal- gestion in local optimal points. Also, genetic algorithms do
culating the losses and applying it to the Lagrangian-GSF not need prior information about the answer space such as
method, we should repeat the calculations. convexity and derivability. The only thing that is needed is
to calculate the fitness function for each answer.
4.3 GA-GSF Now in order to do optimal power flow using genetic algo-
rithm, we should be able to introduce the generators’ fuel cost
A genetic algorithm is a statistical and random search tech- function through equal and unequal constraints along with
nique to obtain the minimum or maximum point of the network calculations as an objective function to GA, simi-
objective functions. This algorithm can model any objec- lar to Lagrangian-GSF method in which the constraints get
tive function, and that is why it has a good efficiency in included in the objective function through Lagrange coeffi-
solving complicated and multipurpose problems. Also, the cients. But contrary to the Lagrangian-GSF method in which
principles of this algorithm are inspired by genetic systems only calculations for the power of generators and passing
of living beings that act based on the following computational power took place, this method also calculates the bus voltage
processes: and losses of each network line. These stages are the same 5
stages that were defined in the article’s part 3.
• Coding each answer as a series of variables that are known Therefore, in order to take into account all equal and
as a chromosome. unequal constraints and to prevent the violation of limits
• Selecting a specific number of chromosomes in the prob- of variables, the objective function f (x) is designed as the
lem’s workspace as the initial population. following. Equal and unequal constraints with respective λj

123
 Electrical Engineering

Best: 6235.97 Mean: 6236.19 Current Best Individual x 10

4
Best, Worst, and Mean Scores

Current best individual

12000 6 2
Best fitness
Fitness value

10000 Mean fitness 4 1.5

8000 2 1

6000 0 0.5
0 20 40 60 80 100 1 2 3 4 5 20 40 60 80 100
Generation Number of variables (5) Generation

Selection Function Fitness of Each Individual Stopping Criteria

20 8000
Number of children

15 6000
Stall (T)
10 4000 Stall (G)
Time
5 2000
Generation
0 0
0 10 20 30 40 50 0 10 20 30 40 50 0 20 40 60 80 100
Individual % of criteria met
Score Histogram transmission Power Power Loss
Number of individuals

50 100 2.5

40 2
50
30 1.5

MW
MW

20 1
0
10 0.5

0 -50 0
1-2 1-5 2-3 2-4 2-5 3-4 4-5 4-7 4-9 5-66-116-126-137-8 7-99-109-1410-11
12-13
13-14 1-2 1-5 2-3 2-4 2-5 3-4 4-5 4-7 4-9 5-66-116-126-137-8 7-99-109-1410-11
12-13
13-14
6235 6236 6237 6238 6239 6240 6241
Score (range) Line Line

Generators Power bus Voltage

150 1.5

100 1
PU
MW

50 0.5

0 0
P1 P2 P3 P6 P8 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Power Bus

Fig. 3 Results from the proposed algorithm

and μi coefficients and limit violations through the penalty be more optimal when this value is between 0.8 and 0.85 and
function are included in the objective function. the smaller this value, the farther we are from the optimal
answer. In the GA-GSF method, a crossover value of 0.85
Ng  
 has been considered and mutation is of the Gaussian type.
FT  F(x)  ai + bi Pg i + ci Pg i 2 (38)
(i1)
Neq Nineq
 
f (x)  FT + μi |gi (x)| + λ j h j (x) + penalty (39) • Work stages All the stages defined in Fig. 1 will be imple-
i1

i1
 mented as the following by GA:

n
Penalty  100 ∗ abs Pi − P D − PLoss (40) (A) Creating an initial population
i1 (B) Calculating the fitness function f (x) using OPF (i.e.,
the objective function designed with 5 stages)
An important factor in GA is crossover that causes the (C) Selecting a specific number of initial population
creation of new generations with better characteristics and chromosomes and applying genetic operators (muta-
has a possibility between 0 and 1, and usually, the answer will tion and crossover)

123
Electrical Engineering

Table 1 Comparing the results of OPF in an IEEE 14-BUS network

Power Min Max PSO [1] IPSO [1] SQP [31] IHS [31] GA [31] ED-GSF LA-GSF NR-LA GA-GSF

P1 (MW) 50 200 161.4592 154.7555 133.6103 138.7025 162.7528 135.75 131.7265 139.0018 132.6016
P2 (MW) 20 80 20 20 56.8206 59.5883 60.1288 20 23.3831 24.8658 26.6337
P3 (MW) 15 50 50 50 36.4725 33.2682 40.5898 50 50 50 50
P6 (MW) 10 30 10 30 24.9509 23.9271 21.6584 30 30 30 30
P8 (MW) 10 35 23.5941 10 28.8968 23.901 27.0935 23.25 23.8904 24.1745 24.5135
P1,2 (MW) 120 105.5066 101.8434 71.8454 75.8468 69.2586 93.1343 89.8782 88.4489 86.7446
P1,5 (MW) 65 49.8993 47.1566 40.0138 42.4686 40.2709 42.6157 41.8483 41.5107 41.1082
P2,3 (MW) 36 38.3397 37.8898 44.0003 46.6160 42.5510 35.9989 36 36 36
P2,4 (MW) 65 37.3628 36.4213 36.0876 38.4536 37.0483 32.4641 32.4664 32.4664 32.4663
P2,5 (MW) 50 28.1040 25.8323 26.8782 28.6654 28.0882 22.9713 23.0949 23.1483 23.212
P3,4 (MW) 63 − 5.8603 − 6.3102 − 13.7272 − 14.3158 − 11.0592 − 8.2011 − 8.2 − 8.2 − 8.2
P4,5 (MW) 45 − 40.3963 − 45.8344 − 40.1182 − 42.6453 − 39.1445 − 41.0784 − 40.5778 − 40.3572 − 40.0941
P4,7 (MW) 55 12.3473 16.6271 5.6951 9.0582 7.6087 8.2239 7.8275 7.6516 7.4417
P4,9 (MW) 32 11.7515 11.5184 8.9834 9.9250 9.7248 9.3175 9.2167 9.172 9.1186
P5,6 (MW) 45 30.0071 19.5545 19.1738 20.8887 21.6146 16.9086 16.7654 16.7019 16.6262
P6,11 (MW) 18 5.0543 10.8036 7.5338 7.9500 7.0210 9.2103 9.124 9.0858 9.0402
P6,12 (MW) 32 7.3615 8.2059 7.7257 7.7868 7.6503 7.9719 7.9592 7.9536 7.9469
P6,13 (MW) 32 16.3913 19.3450 17.6652 17.8790 17.4017 18.5265 18.4821 18.4625 18.4391
P7,8 (MW) 32 − 23.5941 − 10.0000 − 28.8968 − 23.9010 − 27.0935 − 23.25 − 23.8904 − 24.1745 − 24.5135
P7,9 (MW) 32 35.9414 26.6271 34.5919 32.9592 34.7022 31.4739 31.7179 31.8261 31.9552
P9,10 (MW) 32 7.4457 1.6964 4.9662 4.5500 5.4790 3.2897 3.376 3.4142 3.4598
P9,14 (MW) 18 10.7473 6.9491 9.1092 8.8342 9.4480 8.0017 8.0587 8.0839 8.114
P10,11 (MW) 12 − 1.5543 − 7.3036 − 4.0338 − 4.4500 − 3.5210 − 5.7103 − 5.624 − 5.5858 − 5.5402
P12,13 (MW) 12 1.2615 2.1059 1.6257 1.6868 1.5503 1.8719 1.8592 1.8536 1.8469
P13,14 (MW) 12 4.1527 7.9509 5.7908 6.0658 5.4520 6.8983 6.8413 6.8161 6.786
V1 (pu) 1.1000 1.1000 0.9558 0.9819 0.9965 – – 1.0600 1.0600
V2 (pu) 1.0804 1.0979 0.9565 0.9788 0.9860 – – 1.0350 1.0450
V3 (pu) 1.0544 1.0734 0.9592 0.9894 0.9810 – – 1.0000 1.0100
V4 (pu) – – – – – – – 1.0061 1.0232
V5 (pu) – – – – – – – 1.0098 1.0264
V6 (pu) 1.1000 1.0686 0.9512 0.9833 0.9945 – – 1.0600 1.0700
V7 (pu) – – – – – – – 1.0409 1.0540
V8 (pu) 1.0756 1.0354 0.9922 0.9774 0.9904 – – 1.0800 1.0900
V9 (pu) – – – – – – – 1.0241 1.0365
V10 (pu) – – – – – – – 1.0228 1.0348
V11 (pu) – – – – – – – 1.0376 1.0485
V12 (pu) – – – – – – – 1.0434 1.0538
V13 (pu) – – – – – – – 1.0370 1.0474
V14 (pu) – – – – – – – 1.0113 1.0230
Power losses 6.0533 5.7555 21.7511 20.3871 53.2233 – – 9.0422 4.7488
(MW)
Generation 6170.6 6165.6 6342.8 6203.1 7159.5 6128.5 6126.4 6332.7 6235.4
cost ($)
Processing 8.814 s 8.9 s 9.2 s 28.5 s 100.1 s 5.6 s 3.42 s 10.9 s 20.97 s
speed

123
 Electrical Engineering

Table 2 Comparing OPF results

in the IEEE 30-BUS network Bus power Min Max PSO [1] IPSO [1] GA-GSF

P1 50 200 175.1820 177.0906 191.1802

P2 20 80 49.2678 48.6662 48.1101
P5 15 50 21.7230 21.3156 19.4706
P8 10 35 22.0514 21.1555 11.0221
P11 10 30 11.8301 11.8729 10.0000
P13 12 40 12.0000 12.0000 12.0000
PLoss 8.6543 8.7008 8.3831
F 799.6711 799.3417 796.2202
Processing speed (CPU) 10.856 s 10.894 s 23.21 s

Table 3 Comparing OPF results in the IEEE 57-BUS network

Bus power Min Max PSO [1] IPSO [1] SQP [31] IHS [31] GA [31] GA-GSF

P1 0 575.88 139.1571 145.3988 229.8185 272.4235 274.93 142.2828

P2 0 100 100 79.0190 143.1449 125.4842 138.26 100
P3 0 140 75.8451 42.6999 96.1701 77.8435 93.2667 43.9821
P6 0 100 38.4932 75.4517 243.6877 266.3847 284.3433 73.0015
P8 0 550 455.56 459.2791 182.2301 170.8372 181.3667 472.5673
P9 0 100 100 99.5688 183.6253 167.9762 187.8567 88.6241
P12 0 410 360.254 364.5374 223.4802 238.4512 237.7633 343.5958
P8,9 200 176.2036 187.9362 81.8501 83.6866 91.9319 200
P9,11 50 46.3299 50.4899 36.5837 33.2160 41.7931 50
P6,7 40 − 21.4157 − 15.3374 67.0019 73.9498 75.0824 − 15.4533
P3,15 100 59.8149 49.5377 111.4366 105.3258 113.3175 55.0204
PLoss 73.5094 70.1547 106.3568 123.6005 201.9867 68.2536
F 42,109.7231 41,688.5004 47,346.9008 48,489.299 52,110.706 41,570.562
Processing speed 18.814 s 18.90 s 460.8 s 295.4 s 1436.2 s 29.74 s

(D) Repeating part B and reviewing the establishment erators that have caused increased generation costs. Table 1
of the conditions and in case of violation, repetition compares the results from the proposed method with other
until reaching a more optimal answer. methods.

• Comparing the results

Finally, the output of the GA-GSF method will be the gen-
eration capacity of generators, power passing through lines,
total generation cost, a voltage of each bus, and network’s In this part, the results from the proposed algorithm are
losses. In the following, the GA-GSF algorithm has been compared to methods proposed in [1, 31] whose information
implemented on the IEEE 14-BUS network (Fig. 2). Also, is shown in Tables 1, 2, and 3. In the IEEE 14-BUS network,
the network’s information was taken from [1] and the results although P2 generator is cheaper than bus generators 3, 6,
from the proposed method are shown in Fig. 3, in which the and 8, due to 36-MW capacity of line 2–3, it is not possible
number of changes in optimal answer, the best location of to increase the generation capacity of generator P2 because
variables, productions, number of generations, fitness func- it has caused congestion in line 2–3, so we had to use more
tion of each generation, stoppage criterion, the number of expensive generators. Also in Table 1, the lines that are very
changes in power passing from each line, and power losses effective in optimal power flow are highlighted in italic and
in each line are shown. As the results show, although P2 lines that exceed their capacity are highlighted in bold.
generator is cheaper than P3, P6, and P8 generators, due to As you can see the results, in methods of [1, 31], the capac-
the limitation of transmission lines, it cannot generate more. ity limit of transmission lines has been violated and power
Therefore, we had no choice but to use more expensive gen- flow has practically become unreal. Therefore, the results

123
Electrical Engineering

Table 4 Network’s power

demand information Bus load Hour

1 2 3 4 5 6 7 8 9 10 11 12

1 0 0 0 0 0 0 0 0 0 0 0 0
2 10.4 13.29 14.3 14.98 15.98 16.65 17.66 18.34 19 20.01 21.02 21.7
3 82.9 85.79 86.8 87.48 88.48 89.15 90.16 90.84 91.5 92.51 93.52 94.2
4 36.5 39.39 40.4 41.08 42.08 42.75 43.76 44.44 45.1 46.11 47.12 47.8
5 0 0 0 0 1.88 2.55 3.56 4.24 4.9 5.91 6.92 7.6
6 0 2.79 3.8 4.48 5.48 6.15 7.16 7.84 8.5 9.51 10.52 11.2
7 0 0 0 0 0 0 0 0 0 0 0 0
8 0 0 0 0 0 0 0 0 0 0 0 0
9 18.2 21.09 22.1 22.78 23.78 24.45 25.46 26.14 26.8 27.81 28.82 29.5
10 0 0 1.6 2.28 3.28 3.95 4.96 5.64 6.3 7.31 8.32 9
11 0 0 0 0 0 0 0 0.14 0.8 1.81 2.82 3.5
12 0 0 0 0 0 1.05 2.06 2.74 3.4 4.41 5.42 6.1
13 0 2.05 4.06 4.96 6.56 7.5 9.02 10.04 11.1 12.01 13.12 13.5
14 0 2.1 4.54 6.96 8.58 9.3 10.8 11.6 12 13.1 14 14.9
PD (MW) 148 166.5 177.6 185 196.1 203.5 214.6 222 229.4 240.5 251.6 259
Bus load 13 14 15 16 17 18 19 20 21 22 23 24

1 0 0 0 0 0 0 0 0 0 0 0 0
2 20.01 19 18.34 16.65 14.23 16.65 18.34 20.01 18.34 16.65 14.23 13.29
3 92.51 91.5 90.84 89.15 87.38 89.15 90.84 92.51 90.84 89.15 87.38 85.79
4 46.11 45.1 44.44 42.25 41.08 42.25 44.44 46.11 44.44 42.25 41.08 39.39
5 5.91 4.9 4.24 2.55 0.88 2.55 4.24 5.91 4.24 2.55 0.88 0
6 9.51 8.5 7.84 6.15 3.08 6.15 7.84 9.51 7.84 6.15 3.08 2.79
7 0 0 0 0 0 0 0 0 0 0 0 0
8 0 0 0 0 0 0 0 0 0 0 0 0
9 27.81 26.8 26.14 24.45 22.01 24.45 26.14 27.81 26.14 24.45 22.01 21.09
10 7.31 6.3 5.44 3.45 2.18 3.45 5.64 7.31 5.64 3.45 2.18 0
11 1.9 0.9 0.3 0 0 0 0.14 1.81 0.14 0 0 0
12 4.41 3.4 2.74 1.05 0 1.05 2.7 4.5 2.7 1.05 0 0
13 11.81 10.8 10.14 8.05 6.08 8.05 10.14 11.81 10.14 8.05 6.08 2.05
14 13.21 12.2 11.54 9.75 8.08 9.75 11.54 13.21 11.54 9.75 8.08 2.1
PD (MW) 240.5 229.4 222 203.5 185 203.5 222 240.5 222 203.5 185 166.5

of [1, 31] are completely wrong. In the proposed method, any congestion in the IEEE 30-BUS network and also the
because of considering generating scaling factors, real power proposed algorithm has yielded a lower generation cost.
flow has been obtained. Also, the reason for reduced costs In Table 3, the proposed method has been compared to
in ED-GSF and LA-GSF methods is that they did not take methods of [1, 31]. As you can see, sometimes the capacity
into account the network losses. In the end, we can see that of some lines of the IEEE 57-BUS network has been vio-
in GA-GSF method, the generation costs, network losses, lated in methods of [1, 31], whereas in the proposed method,
and also the processing speed have significantly decreased in addition to not violating these limits, we have achieved
compared to GA [31] method. reduced generation costs. The lines that are very effective
Also, in order to better investigate the proposed method, in OPF are highlighted in italic and lines that exceed their
the GA-GSF method has been implemented on IEEE 30-BUS capacity limits are highlighted in bold.
and IEEE 57-BUS networks. Table 2 shows the power flow From the results of Tables 1, 2, and 3, we can conclude
results from the proposed method and [1], and there was not that the proposed algorithm not only makes the power flow

123
 Electrical Engineering

500 Load2
Consumed power of Network 5 Simulation results
Load3
450 Load4
Load5
Imagine that Table 4 and Fig. 4 are the predictor curve of
400
Load6 power demand for an IEEE 14-BUS network in 24 h where
Load9
350
Load10
the load peak occurs at 12 midday and 8 PM. Now, in order to
Demand (MW)

300 Load11 review the performance of GA-GSF algorithm, OPF problem

Load12
Load13
for each hour has been implemented according to the sug-
250
Load14 gested stages, and finally after reviewing the power passing
200
through lines, the electricity market price will be determined
150 depending on the capacity of lines as UMP or LMP and prof-
100 its of generators will be calculated. In the following, we will
50
compare the profit gained from the GA-GSF method to that of
ED-GSF and NR-LA-GSF methods’. Also, the simulations
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 in this article are all done by MATLAB software.
Hour
Here, the sample network has 6 generators, 11 loads, 3
Fig. 4 Network’s power demand curve transformers, 20 transmission lines. As you can see from the
information on the network’s load, buses 1, 7, and 8 lack
load. Also, the network’s load information is not the same
for each hour; although total power consumption is the same
in some hours and this itself will lead to a difference in the
real, it has also been able to achieve lower generation costs answer for optimal power flow for each hour.
by network expansion.

Table 5 Results from OPF in 24 h

Hour GA-GSF

P1 P2 P3 P6 P8 PD Loss Cost of production Market price (UMP or LMP)

1 80.8078 22.5119 25.7168 11.1711 10.2851 148 2.4927 2946.8 UMP  28

2 88.2674 24.8474 31.0129 15.0056 10.0820 166.5 2.7153 3439.4 UMP  28
3 96.8872 26.2725 34.6890 11.1272 11.5908 177.6 2.9667 3733 UMP  28
4 97.4981 22.6387 34.0128 19.0955 14.7843 185 3.0294 3958.1 UMP  28
5 109.8713 28.7847 40.8183 10.1425 10.0077 196.1 3.5245 4271.5 UMP  28
6 106.0794 27.4307 39.3061 17.1343 16.9268 203.5 3.3773 4485.8 UMP  28
7 116.3731 27.5184 43.2278 18.5287 12.8155 214.6 3.8635 4819.7 UMP  28
8 122.5521 31.5123 47.0333 14.8327 10.3011 222 4.2315 5070.6 UMP  28
9 116.3820 28.5180 44.3829 26.3340 17.7348 229.4 3.9517 5276 UMP  28
10 118.1341 30.7747 46.3060 27.6153 21.7568 240.5 4.0869 5633.5 UMP  28
11 130.8488 24.2545 48.2389 30 22.8655 251.6 4.6077 5989.7 UMP  28
12 132.6016 26.6337 50 30 24.5135 259 4.7488 6235.4 LMP
13 119.7318 30.7498 46.9902 30 17.2444 240.5 4.2162 5633.2 UMP  28
14 117.1643 29.2492 44.6459 20.2290 22.0340 229.4 3.9224 5282.8 UMP  28
15 113.6272 28.1377 42.9637 22.9802 18.0693 222 3.7781 5046.5 UMP  28
16 105.0814 26.5851 38.6940 19.8642 16.6349 203.5 3.3597 4486.2 UMP  28
17 102.8455 27.0021 37.4759 10.0439 10.8682 185 3.2355 394.65 UMP  28
18 114.6652 26.9695 41.7621 11.7254 12.0861 203.5 3.7083 4484.8 UMP  28
19 119.9940 26.5042 44.5293 24.2701 10.7952 222 4.0929 5052.5 UMP  28
20 121.2517 30.6871 47.3654 28.9476 16.5288 240.5 4.2806 5632.5 UMP  28
21 115.9389 29.7648 44.4101 22.1304 13.6856 222 3.9298 5049.9 UMP  28
22 107.6330 23.4966 38.4901 24.2663 13.0953 203.5 3.4814 4487.8 UMP  28
23 98.5596 26.4706 35.8064 11.0770 16.1429 185 3.0566 3953.7 UMP  28
24 95.5259 20.8454 31.5765 10.7879 10.6227 166.5 2.8584 3428 UMP  28

123
Electrical Engineering

Table 6 LMP values for network buses

Hour GA-GSF

LMP

Bus 1 Bus 2 Bus 3 Bus 4 Bus 5 Bus 6 Bus 7 Bus 8 Bus 9 Bus 10 Bus 11 Bus 12 Bus 13 Bus 14

12 22.68 21.46 63.03 34.27 30.62 31.86 33.80 33.80 33.27 33.02 32.45 31.97 32.06 32.74

Consumed power of Network

Table 5 shows the results from implementing the proposed 300
algorithm, and they include the amount of injected power,

Demand (MW)
network losses, total generation cost, and electricity price 200
for 24 h of a day. From the results, we can see that one of the
limitations of this network is the 36-MW capacity of line 2–3 100
that has reached its maximum capacity at the time 12 midday. P1
P2
Therefore, in this hour, the price will be calculated based on 0
LMP and Table 6 shows LMP values for different buses. 1 2 3 4 5 6 P3
7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
P6 Hour
Considering Table 5 information, the generation capacity P8
0
of generators and power passing through lines are, respec-
tively, shown in Figs. 5 and 6. As you can see, we can properly
-5
relate the changes in power injected to buses to the passing
MW
power in lines using GA-GSF and we could do power flow Power Loss
-10
in a way that, contrary to methods in [1, 31], does not vio-
late the capacity limits and we can obtain a real power flow.
-15
Now, the first question that comes to mind is that why bus 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
generator 2 generates less power compared to bus generators Hour
3, 6, and 8 that are more expensive? To answer this question,
Fig. 5 Generation power and network losses in GA-GSF method
Fig. 6 results can help us; in these results, the maximum gen-
eration capacity of each line and also the number of changes
in 24 h are specified. We can clearly see that the reason for ED-GSF method, generators 6 and 8 entered the circuit only
this problem is the 36-MW capacity of line 2–3 that don’t in load peak hours and were off most of the time. And in the
allow bus generator 2 to generate more power and this line NR-LA-GSF method, considering that quadratic equations
reaches its maximum capacity at 12 midday. Also, negative were solved on MATLAB to solve the problem, the answers
power columns indicate the reversed direction of power in are obtained for a maximum capacity of lines.
that line. So, considering the results of lines 3–4, power is
transferred from bus 4 to bus 3 and this is another reason for • Determining electricity price
the reduced power generation of bus generator 2. Therefore,
bus 3 power is supplied as the following:
In the GA-GSF method, considering the review of network
lines’ capacity and full capacity of lines 2–3 at 12 midday,
Load3  P2,3 + P3 + P3,4 LMP price should be calculated for buses at this hour and
for other hours (when lines are not full), UMP is considered
Figure 7 compares the results of power flow by NR-LA- and its values are shown in Fig. 8. As you can see from the
GSF, ED-GSF, and GA-GSF methods. As it was previously results of Fig. 8, the capacity of lines 2–3 was full and it
mentioned, in the IEEE 14-BUS network under study, the was not possible to generate more power by bus generators
capacity of lines 2–3 is one of the main barriers in increasing 1 and 2, and in order to compensate, we had to use the most
the production of generators 1 and 2. The condition of power economic generator, meaning generator 3. This itself has led
passing from lines 2–3 for NR-LA-GSF, ED-GSF, and GA- to the reduced price in buses 1 and 2 and increased price in
GSF methods is shown in Fig. 7. In the NR-LA-GSF method, bus 3, and its results are well shown in Fig. 8. Also in the
lines 2–3 are at its full capacity in all the hours. Therefore, NR-LA-GSF method, due to the full capacity of lines 2–3 at
for times when the capacity of lines 2–3 has reached its max, all times, there is no UMP and LMP should be calculated for
electricity price should be calculated as LMP. Figures 8 and 9 all 24 h. The results are shown in Fig. 9. As you can see in
show the electricity market price by NR-LA-GSF, ED-GSF, Fig. 8, electricity prices in buses 1 and 2 are less than other
and GA-GSF methods. Also, Fig. 7 results show that in the buses.

123
 Electrical Engineering

Capacity (MW)
Capacity (MW)
Capacity (MW)

Capacity (MW)
120 65 65
Line1-2,full 120 Line1-5,full 65 Line2-4,full 65
Line2-3,full 36
60 32.5 36 32.5

0 0 0
0 12 24 0 12 24 0 12 24 0 12 24
Hour Hour Hour Hour

Capacity (MW)
Capacity (MW)
Capacity (MW)

Capacity (MW)
50 0 0
Line2-5,full 50 10
25 -20 -20
0
Line3-4,full 63 -40 Line4-7,full 55
Line4-5,full 45
0 -40 -10
0 12 24 0 12 24 0 12 24 0 12 24
Hour Hour Hour Hour

Capacity (MW)
Capacity (MW)
Capacity (MW)

Capacity (MW)
20
Line4-9,full 32 Line6-11,full 18 Line6-12,full 32
10 10 10 10
5 0 5 5
Line5-6,full 45
0 -10 0 0
0 12 24 0 12 24 0 12 24 0 12 24
Hour Hour Hour Hour

Capacity (MW)
Capacity (MW)
Capacity (MW)

Capacity (MW)
0 10
Line6-13,full 32 40 Line7-9,full 32
20
-20 20 0
10
Line7-8,full 32 Line9-10,full 32
0 -40 0 -10
0 12 24 0 12 24 0 12 24 0 12 24
Hour Hour Hour Hour

Capacity (MW)
Capacity (MW)
Capacity (MW)

Capacity (MW)
10 0 4 10
Line12-13,full 12 Line13-14,full 12
0 -5 2 5
Line9-14,full 18
Line10-11,full 12
-10 -10 0 0
0 12 24 0 12 24 0 12 24 0 12 24
Hour Hour Hour Hour

Fig. 6 Power passing from network lines in GA-GSF method

GA-GSF method ED-GSF method P1 NR-LA-GSF method

300 300 300
P2
Demand (MW)

P3
P6
200 200 200
P8

100 100 100

0 0 0
0 12 24 0 12 24 0 12 24
Hour Hour Hour
Capacity (MW)

Line2-3,full 36 Line2-3,full 36 Line2-3,full 36

36 36 36

0 12 24 0 12 24 0 12 24
Hour Hour Hour

Fig. 7 Power flow results for NR-LA-GSF, ED-GSF, and GA-GSF methods

Now that the market price has been determined, we could each generator along with total GENCO profit in the GA-
calculate the profit for each generator. Table 7 shows the GSF method. As you can see, at 12 middays when there
calculations and the total profit of generators in 24 h in the was congestion in lines 2–3, the profits of bus genera-
GA-GSF method. tors have decreased because of reduced LMP, and at the
In the GA-GSF method, the total profit of generators same hour, because of increased LMP in buses 3, 6, and
equals 30,788.77$. Also, the highest profit belongs to bus 8, their profits have increased. These changes can be seen in
generator 1 and the least profit belongs to bus genera- Fig. 10.
tors 6 and 8 and this is because we tried to keep the Figure 11 shows the total profit made by NR-LA-GSF,
electricity price low and use expensive generators to the ED-GSF, and GA-GSF methods. By comparing the results,
least extent. Figure 10 shows the profit by the hour for we can see that the factors of loss, power flow method, and

123
Electrical Engineering

GA-GSF method ED-GSF method

60 29 UMP,bus 1,2,3,4,5,6,7,8,9,10,11,12,13,14
UMP,bus 1,2,3,4,5,6,7,8,9,10,11,12,13,14
40 27
$/MWh

UMP UMP
25
20

LMP
23 LMP LMP
0
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 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Hour Hour

63.03
57.12
55.04
52
LMP,Hour = 12 LMP,Hour = 10 LMP,Hour = 11 LMP,Hour = 15
$/MWh

31.515
27.52 28.56
26

1 2 3 4 5 6 7 8 9 1011 12 13 14 1 2 3 4 5 6 7 8 9 1011 12 13 14 1 2 3 4 5 6 7 8 9 1011 12 13 14 1 2 3 4 5 6 7 8 9 10 11 12 13 14

Bus Bus Bus Bus

Fig. 8 Electricity price in ED-GSF and GA-GSF methods

64 64 64 64
LMP,bus 3
$/MWh

$/MWh

$/MWh

$/MWh
LMP,bus 1 LMP,bus 2 LMP,bus 4
32 32 32 32

0 0 0 0
0 12 24 0 12 24 0 12 24 0 12 24
Hour Hour Hour Hour
64 64 64 64
$/MWh

$/MWh

$/MWh

$/MWh
LMP,bus 5 LMP,bus 6 LMP,bus 7 LMP,bus 8
32 32 32 32

0 0 0 0
0 12 24 0 12 24 0 12 24 0 12 24
Hour Hour Hour Hour
64 64 64 64
$/MWh

$/MWh

$/MWh

$/MWh

LMP,bus 9 LMP,bus 10 LMP,bus 11 LMP,bus 12

32 32 32 32

0 0 0 0
0 12 24 0 12 24 0 12 24 0 12 24
Hour Hour Hour Hour
64 64
$/MWh

$/MWh

LMP,bus 13 LMP,bus 14
32 32

0 0
0 12 24 0 12 24
Hour Hour

Fig. 9 Electricity price in NA-LA-GSF method

determining electricity price have significantly influenced the the profit of generators by reducing losses and increasing
profit of generators, in a way that the profit from the GA-GSF electricity prices.
method is 38$ more than the NR-LA-GSF method that had Finally, according to the results, the advantage of the pro-
an LMP price during all hours and 12,081 $ more than ED- posed method compared to other methods can be described
GSF which has not considered the losses. Also, this profit as follows:
is obtained in a way that network losses have increased and
congestion occurs only at 12 midday. Therefore, we can say • Providing optimal power flow reality-based: using gener-
that apart from reduced generation costs, we can increase ating scaling factors in the algorithm structure to manage

123
 Electrical Engineering

Table 7 Calculations for profits of generators in GA-GSF method

Hour GA-GSF

P1 × π − P2 × π − P3 × π − P6 × π − P8 × π − Loss cost Total profit of π

C1(P1) C2(P2) C3(P3) C6(P6) C8(P8) generators

1 805.1 244.2166 183.7 14.0185 19.9837 − 25.6925 1241.3 28

2 862.9 261.4285 139.4 14.9925 19.9987 − 27.9643 1270.7 28
3 926.1 271.1808 092.1 13.996 19.4939 − 30.5256 1292.4 28
4 930.4 245.1903 101.8 13.8696 15.4221 − 31.1627 1275.6 28
5 1014.2 286.9875 − 16.7 13.4236 20 − 36.1918 1281.8 28
6 989.4 278.6877 13.6 14.6865 10.4039 − 34.6998 1272.1 28
7 1055.2 279.2409 − 69.8 14.1584 18.4146 − 39.6240 1257.6 28
8 1092.1 302.1487 − 165.4 14.9908 19.9819 − 43.3412 1220.5 28
9 1055.3 285.3933 − 097.3 6.4129 8.0346 − 40.5146 1217.3 28
10 1065.9 298.2543 − 146 4.3649 − 7.6445 − 41.8811 1173 28
11 1138.7 257.2037 − 198.7 − 0.03 − 13.1042 − 47.1287 1136.9 28
12 462.5727 80.7343 1501.5 115.77 115.3924 − 79.0247 2196.9 LMP
13 1075.5 298.1202 − 164.2 − 0.03 9.5037 − 43.1858 1175.7 28
14 1060 289.7166 − 103.7 13.1635 − 8.9634 − 40.2195 1210 28
15 1038.2 283.0856 − 063.7 10.7368 6.9773 − 38.7602 1236.5 28
16 982.7 273.2439 25.3 13.4095 11.1956 − 34.5201 1271.3 28
17 967.6 275.9535 47.3 13.3591 19.8492 − 33.2590 1290.8 28
18 1044.7 275.7434 − 36.8 14.2806 19.1296 − 38.0537 1279 28
19 1077.1 272.7126 − 100.8 9.2514 19.8735 − 41.9417 1236.1 28
20 1084.5 297.7817 − 174.4 2.003 11.475 − 41.1935 1180.2 28
21 1052.5 292.6751 − 97.9 11.5942 17.2833 − 40.2943 1235.9 28
22 999.7 251.6598 29.1 9.2561 18.0838 − 35.7548 1272 28
23 937.9 272.4914 75.1 13.9699 12.453 − 31.4404 1280.5 28
24 916.4 231.0013 133 13.8133 19.9224 − 29.4228 1284.7 28
Total 23,634.7671 6404.8519 906.3272 365.4611 403.1601 − 925.7965 30,788.77

the congestion of lines. As opposed to the quasi-heuristic calculation to be brought closer to reality and with network
and quadratic programming methods introduced, no con- monitoring to identify network weaknesses.
gestion of the capacities of the lines occurred and actual • In addition to reducing production costs, generators profit
optimal power flow was presented. can be increased by reducing losses and optimizing elec-
• Presenting the weaknesses of the proposed designs using tricity prices over other methods.
metaheuristic algorithms. • Increasing the accuracy of the algorithm for the larger net-
• Calculating electricity market prices: In this project the works: while the accuracy of the algorithms in the IEEE
goal of reducing electricity prices based on the elimination 14-BUS network that has some restrictions compared to
of congestion lines. other methods is better.
• Increasing the computational speed of the algorithm with
considering the generating scaling factor: Such that for the
generation power, one can quickly check the condition of
the power passing on the lines so that the algorithm does 6 Conclusion
not get stuck at this point in case of violation the capacity
of the lines and can calculate the appropriate optimal point. In this article, we used the metaheuristic algorithm GA-GSF
• Including the terms of the losses in the power flow calcula- to solve the OPF problem and to calculate the locational
tions and monitoring of all network lines: This allows the marginal price (LMP). First, we introduced ED, NR, and
Lagrangian methods and then we used these methods in
genetic algorithm. In order to improve the structure of the

123
Electrical Engineering

G1 benfit G2 benfit
1500 300
genetic algorithm, we used GSF in the structure of GA, so
250 that we can relate the changes in the power passing through
1000 200 network lines to changes in the power of generators. And we
simultaneously check the conditions for establishing flow
$

150

500 100
power in the network for each variable and in case the
50
obtained minimum point violated the conditions, the process
0 0
0 2 4 6 8 10 12 14 16 18 20 22 24 0 2 4 6 8 10 12 14 16 18 20 22 24
will be repeated. In the end, the output of the proposed algo-
Hour Hour
G3 benfit G6 benfit rithm will be the power of generation units, network losses,
2000
100
bus voltage, generation cost, and power passing through
1500
lines, and by checking the capacity of lines, we could also
1000 50
calculate the market price of electricity and profit of genera-
$

500
0
tors.
0 An important issue in OPF is the lack of congestion in
-500
0 2 4 6 8 10 12 14 16 18 20 22 24
-50
0 2 4 6 8 10 12 14 16 18 20 22 24
network lines. Simulation results show the proper perfor-
Hour Hour mance of the GA-GSF algorithm compared to methods used
G8 benfit Genco=G1+G2+G3+G6+G8
in [1, 31], in a way that it has resulted in reduced losses, low
100 2000
processing time, reduced generation cost, and reality-based
50
1500 OPF.
$

1000

0 Acknowledgements This work was supported by the KIEE.

500

-50 0
0 2 4 6 8 10 12 14 16 18 20 22 24 0 2 4 6 8 10 12 14 16 18 20 22 24
Hour Hour

Fig. 10 Generators’ profit per hour in GA-GSF method

Fig. 11 Comparing the total Lagrangian-GSF (without Loss) NR-Lagrangian-GSF

profits of generators
benfit Total = 32081 $ benfit Total = 30751 $
2000 2000
$

1000 1000

0 0
0 2 4 6 8 10 12 14 16 18 20 22 24 0 2 4 6 8 10 12 14 16 18 20 22 24
Hour Hour
ED-GSF (without Loss) GA-GSF

benfit Total = 18708 $ benfit Total = 30789 $

2000 2000
$

1000 1000

0 0
0 2 4 6 8 10 12 14 16 18 20 22 24 0 2 4 6 8 10 12 14 16 18 20 22 24
Hour Hour
x 10
4 Comparison of methods

3
benfit Total ($)

2.5

ED-GSF Lagrangian-GSF NR-LA-GSF GA-GSF

methods

123
 Electrical Engineering

References 18. Park J-B, Jeong Y-W, Lee W-N, Shin J-R (2006) An improved
particle swarm optimization for economic dispatch problems with
1. Vasant PM (2012) Meta-heuristics optimization algorithms in engi- non-smooth cost functions. In: IEEE power engineering society
neering, business, economics, and finance. p 734. https://doi.org/1 general meeting
0.4018/978-1-4666-2086-5 19. Thanushkodi K, Muthu Vijaya Pandian S, Dhivy Apragash RS,
2. Shahidehpour M, Alomoush M (2001) Restructured electrical Jothikumar M, Sriramnivas S, Vindoh K (2008) An efficient particle
power systems: operation, trading, and volatility. Marcel Dekker, swarm optimization for economic dispatch problems with nons-
New York mooth cost functions. WSEAS Trans Power Syst 3(4):257–266
3. Alomoush M (2003) Derivation of UPFC DC load flow model with 20. Meng Ke et al (2010) Quantum-inspired particle swarm optimiza-
examples of its use restructured power systems. IEEE Trans Power tion for valve point economic load dispatch. IEEE Trans Power
Syst 18(3):1173–1180 Syst 25(1):215–222
4. Stott B, Marinho JL (1979) Linear programming for power system 21. Vaisakh K, Srinivas LR (2005) Differential evolution approach for
network security applications. IEEE Trans Power Appar Syst PAS- optimal power flow solutions. J Theor Appl Inf Technol 2:261–268
98:837–848 22. Allaoua B, Laoufi A (2009) Optimal power flow solution using ant
5. Reid GF, Hasdorf L (1973) Economic dispatch using quadratic manners for electrical network. Adv Electr Comput Eng 9(1):34–40
programming. IEEE Trans Power Appar Syst PAS-92:2015–2023 23. Haibo Z, Lizi Z, Fanling M (1998) Reactive power optimization
6. Dommel HW, Tinney WF (1968) Optimal power flow solutions. based on genetic algorithm. In: International power conference on
IEEE Trans Power Appar Syst PAS-87:1866–1876 power system technology. pp 1448–1453
7. Sinha N, Chakrabarti R, Chattopadhyay PK (2003) Evolutionary 24. Sulaiman MH, Mustafa MW, Aliman O (2009) Transmission loss
programming techniques for economic load dispatch. IEEE Trans and load flow allocations via genetic algorithm technique. IEEE
Evol Comput 7(1):83–94 Xplore, TENCON
8. Bouktir T, Slimani L, Belkacemi M (2004) A genetic algorithm for 25. Hazra J, Sinha AK (2007) A study on real and reactive power
solving the optimal power flow problem. Leonardo J Sci 4:44–58 optimization using particle swarm optimization. In: International
9. Kennedy J, Eberhart RC (1995) Particle swarm optimization. In: conference on industrial and information systems. pp 323–328
Proceedings of IEEE international conference on neural networks 26. Rahul J, Sharma Y, Birla D (2012) A new attempt to opti-
(ICNN’95). Perth, Australia, vol IV, pp 1942–1948 mize optimal power flow based transmission losses using genetic
10. Chen CH, Yeh SN (2006) Particle swarm optimization for eco- algorithm. In: Computational intelligence and communication net-
nomic power dispatch with valve-point effects. In: 2006 IEEE works (CICN), IEEE Xplore
PES transmission and distribution conference and exposition Latin 27. Vu PT, Le DL, Vo ND (2010) A novel weight-improved particle
America, Venezuela swarm optimization algorithm for optimal power flow and eco-
11. Chaturvedi KT, Pandit M, Srivastava L (2008) Self organizing nomic load dispatch problems. In: Transmission and distribution
hierarchical particle swarm optimization for nonconvex economic conference and exposition
dispatch. IEEE Trans Power Syst 23(3):1079–1087 28. Kumar S, Chaturvedi DK, Grant L (2013) Optimal power flow
12. Yoshida H, Kawata K, Fukuyama Y, Nakanishi Y (2001) A par- solution using fuzzy evolutionary and swarm optimization. Int J
ticle swarm optimization for reactive power and voltage control Electr Power Energy Syst 47:416–423
considering voltage security assessment. IEEE Trans Power Syst 29. Goswamia SK, Acharjee P (2010) Multiple low voltage power flow
15(4):1232–1239 solutions using hybrid PSO and optimal multiplier method. Expert
13. Krost G, Venayagamoorthy GK, Grant L (2008) Swarm intelli- Syst Appl 37(3):2473–2476
gence and evolutionary approaches for reactive power and voltage 30. Najafi M, Ahmadi S, Dashtdar M (2015) Simultaneous energy and
control. In: 2008 IEEE swarm intelligence symposium reserve market clearing with consideration of interruptible loads as
14. Mo N, Zou ZY, Chan KW, Pong TYG (2007) Transient stability one of demand response resources and different reliability require-
constrained optimal power flow using particle swarm optimization. ments of consumers. Int J Emerg Electr Power Syst. https://doi.org/
IEEE Gener Transm Distrib 1(3):476–483 10.1515/ijeeps-2019-0018
15. Swarup KS (2006) Swarm intelligence approach to the solution of 31. Sinsuphan N, Leeton U, Kulworawanichpong T (2013) Optimal
optimal power flow. Indian Inst Sci 86(5):439–455 power flow solution using improved harmony search method.
16. Abou El-Ela AA, El-Sehiemy RAA (2007) Optimized generation Power Syst Res Unit 13(5):2364–2374
costs using modified particle swarm optimization version. WSEAS
Trans Power Syst 10(2):225–232
17. Abido MA (2002) Optimal power flow using particle swarm opti- Publisher’s Note Springer Nature remains neutral with regard to juris-
mization. Electr Power Energy Syst 24(2002):563–571 dictional claims in published maps and institutional affiliations.

123