A solution to optimal power flow problem using Meta-heuristic

Bat algorithm

By: Yeshitela Shiferaw

Research Guide: Dr. K.Padma

Defense University, College of Engineering

Bishoftu
July 2019 1
Parts of presentation:

1. Introduction

2. Differential Evolutions

3. Particle Swarm Optimization

4. Voltage stability indices

5. Bat algorithm

6. Result and discussion

7. Conclusions

CoE,DU National conference By: Yeshitela

12/14/2019 2
Shiferaw
 1.Introduction
The optimization can be defined as the process of finding the
conditions that provide the minimum or maximum value of a chosen
objective function.
• The optimal power flow was at first developed by Carpentier in
1962 to minimize generator cost.
• The power flow analyses forecast power flows and voltages in the
network as a given category of generators and loads.
• With the ever-increasing demand for electricity and the complexity
of the power system, optimal power system operation issue becomes
important.

CoE,DU National conference By: Yeshitela

12/14/2019 3
Shiferaw
 1.Introduction

• Conventional methods are useful to solve optimal power flow as

stated in the literature, such as Interior point methods , Newton’s
methods, linear programming,

• nonlinear programming and gradient method are regularly facing

problems of convergence and complications in obtaining the overall
optimal solution.

• Due to the above problems, research focus has already shifted to

Meta-heuristic algorithm.

• Meta-heuristic model is a way to solve complex optimization

problems by using simple mathematic rules and powerful
computers.
CoE,DU National conference By: Yeshitela
12/14/2019 4
Shiferaw
 2. Differential Evolution

• DE is an evolutionary algorithm
• The Differential evolution (DE) algorithm is a stochastic
population based optimization for solving nonlinear optimization
problem. This algorithm was developed by Storn and Price in
1996.

Initialization Mutation Recombinations Selection

General Differential Evolution algorithm procedure

• Initialization: define upper and lower bounds of each parameter.

12/14/2019
CoE,DU National conference 5
By: Yeshitela Shiferaw
&
2. Differential Evolution

Mutation:
• Mutation expands the search spaces.
𝑔 𝑔
• selecting three new vector from each parameter vector 𝑋𝑟1𝑛 , 𝑋𝑟2𝑛
𝑔
And 𝑋𝑟3𝑛
The weighted difference of the two vectors is added to the third.

𝑔+1 𝑔 𝑔 𝑔
𝑉𝑛 = 𝑋𝑟1𝑛 + 𝐹(𝑋𝑟2𝑛 − 𝑋𝑟3𝑛 ൯ F is constant from [0,2]
Recombination:
• Incorporate successful solutions from the previous generations.

• The process is generating of trial vector

CoE,DU National conference By: Yeshitela

12/14/2019 6
Shiferaw
 2. Differential Evolution
𝑔 Target vector
𝑔+1
𝑈𝑛,𝑖 from 𝑋𝑛,𝑖 & 𝑉 𝑔+1
𝑛,𝑖
𝑔+1 Donor vector
𝑔+1 𝑉𝑛,𝑖 , 𝑖𝑓𝑟𝑎𝑛𝑑(0,1) ≤ 𝐶𝑅
𝑈𝑛,𝑖 = ቐ 𝑔
𝑋𝑛,𝑖 , 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒𝑠

Selection:
• comparing the target vector and the trial vector. The lowermost
function value is selected for the succeeding generations.
𝑔+1 𝑔+1 𝑔
𝑔+1 𝑈𝑛,𝑖 , 𝑖𝑓𝑓(𝑈𝑛 ) ≤ 𝑓(𝑥𝑛 ൯
𝑋𝑛 =ቐ 𝑔
𝑋𝑛 , 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒𝑠

• mutation,recombinations,and selection continuous until the setting

criterion is reached.
CoE,DU National conference By: Yeshitela
12/14/2019 7
Shiferaw
 3. Particle Swarm Optimization
• The particle swarm optimization algorithm was developed by
James Kennedy and Eberhart in 1995.
• Each particle tries to modify its position using the following
information: the current position, the current velocity, the distance
between the current position and pbest, and the distance between the
current position and gbest.

CoE,DU National conference By: Yeshitela

12/14/2019 8
Shiferaw
 3. Particle Swarm Optimization

The velocity of each particle is updated by the following equation.

𝑘+1
𝑉𝑖𝑑

𝑊 max − 𝑊 min
𝑊 𝑘 = 𝑊 max − ( max
) ∗ 𝑖𝑡𝑒𝑟
𝑖𝑡𝑒𝑟
Each particle’s position is updated as

𝑘+1 𝑘 𝑘+1
𝑋𝑖𝑑 = 𝑋𝑖𝑑 + 𝑉𝑖𝑑

CoE,DU National conference By: Yeshitela

12/14/2019 9
Shiferaw
 4.voltage stability indices

• The load flow algorithms incorporate load and generator control

characteristics.
• The voltage stability indices value is changed between zero and one.
• Its value is zero and one means the system has been at no load and
voltage collapse conditions respectively.

• the L index can be computed as :

𝑛𝑔
𝑉𝑖
𝐿𝑗 = |1 − ෎ 𝐹𝑗𝑖 ∠𝜃𝑖𝑗 + 𝛿𝑖 − 𝛿𝑗 |
𝑉𝑗
𝑖=1

CoE,DU National conference By: Yeshitela

12/14/2019 10
Shiferaw
 5.Problem Formulations

• The optimal power flow problem is mainly concerned with the

minimizations of the fuel cost of active power generations and
power losses in the power system.
• In this paper, the proposed bat algorithm goes to minimize the
objective function shown in equation.

𝑁𝐺

• Cost Minimize 𝐹 = ෍ 𝑎𝑖 𝑃2 𝑔𝑖 + 𝑏𝑖 𝑃𝑔𝑖 + 𝑐𝑖

𝑖=1

CoE,DU National conference By: Yeshitela

12/14/2019 11
Shiferaw
 5. Problem Formulations

Real power loss minimization using equation

𝑁𝐿

𝐹𝑙𝑜𝑠𝑠 = min(𝑃𝑙𝑜𝑠𝑠 ) = min ෍ 𝐺𝑖,𝑗 (𝑉𝑖2 + 𝑉𝑗2 − 2𝑉𝑖 𝑉𝑗 cos(𝛿𝑖 − 𝛿𝑗 )ቍ

𝑖=1

fitness functions are defined as follows

1
𝑓𝑖𝑡𝑛𝑒𝑠𝑠 = 𝑓max =
𝐹𝑋

CoE,DU National conference By: Yeshitela

12/14/2019 12
Shiferaw
 5.Problem Formulations
Equality constraints
𝑛

𝑃𝐺𝑖 − 𝑃𝐷𝑖 − ෍ |𝑉𝑖 ||𝑉𝑗 | |𝑌𝑖𝑗 |cos 𝜃𝑖𝑗 − 𝛿𝑖 + 𝛿𝑗 = 0

𝑗=1

𝑄𝐺𝑖 − 𝑄𝐷𝑖 + ෍ |𝑉𝑖 ||𝑉𝑗 | |𝑌𝑖𝑗 |sin 𝜃𝑖𝑗 − 𝛿𝑖 + 𝛿𝑗 = 0

𝑗=1

Inequality constraint
min max
Real power generation limits 𝑃𝐺𝑖 ≤ 𝑃𝐺𝑖 ≤ 𝑃𝐺𝑖

min max
Voltage limits for generator bus 𝑉𝐺𝑖 ≤ 𝑉𝐺𝑖 ≤ 𝑉𝐺𝑖

min max
Reactive power generation limits 𝑄𝐺𝑖 ≤ 𝑄𝐺𝑖 ≤ 𝑄𝐺𝑖
CoE,DU National conference By: Yeshitela
12/14/2019 13
Shiferaw
 5.Problem Formulations

Transmission line loading limits 𝑆𝑖 ≤ 𝑆𝑖max

Voltage stability indices 𝐿𝑗𝑖 ≤ 𝐿𝑗𝑖max

• The equality constraints are determined after

solving the power flow program.

• The generator bus terminal voltages are the

control variables

• Stability indexes are state variables which are limited

through the penalty determination method.

12/14/2019 CoE,DU National conference By: Yeshitela 14

Shiferaw
 6. Proposed Bat Algorithm
• Bat algorithm is one of a Meta-heuristic optimization algorithm,
proposed by Xin-She Yang in 2010 based on the echolocation
behavior of microbats.
• The bat use sonar echoes to detect and avoid obstacles.

• Sound pulses are transformed in to frequency which reflects

from obstacles.

• The bat navigate by using the time delay from emission to the
reflection.

• After hitting and reflecting ,the bats transform their own pulse
in to useful information to explore how far away from the prey
is.
CoE,DU National conference By: Yeshitela
12/14/2019 15
Shiferaw
 6.Proposed Bat Algorithm

bat Algorithm is developed based on the following approximated rules.

• Sensing distance.
All bats use echolocation to sense distance and they also know the
difference between food and barriers in some magical way
• Bats fly randomly with velocity Vi at position Xi

They can automatically adjust the wavelength (or frequency) of

their emitted pulses.

• loudness varies from a large positive value to a small

constant value

CoE,DU National conference By: Yeshitela

12/14/2019 16
Shiferaw
 6.Proposed Bat Algorithm

For the movement of virtual bats, the rules of updating velocities

position and frequency are;

𝐹𝑖 = 𝐹min + (𝐹max − 𝐹min ). 𝛽

𝑉𝑖𝑘+1 = 𝑉𝑖𝑘 + (𝑋𝑖𝑘 − 𝑋 ∗ ). 𝐹𝑖

𝑋𝑖𝑘+1 = 𝑋𝑖𝑘 + 𝑉𝑖𝑘+1

𝛽 is a random number within [0, 1]

• A new solution for each bat is

generated locally using random 𝑋𝑛𝑒𝑤 = 𝑋𝑜𝑙𝑑 + 𝜀𝐴𝑡
walk given by equation
CoE,DU National conference By: Yeshitela
12/14/2019 17
Shiferaw
 6. Proposed Bat Algorithm

Step of BAT algorithm for proposed system

Step 1: Reading of input parameter (transmission line, generator and
load data)

Step 2: Initializing the populations of micro-bats randomly (maximum

iteration, velocity, loudness, pulse rate, number of dimension,
minimum and maximum frequency).

Step 3: initializing the best solutions of fits, costs, and real power
generation
Step 4: Evaluating the cost function and fitness value for the random
number of micro-bats.

CoE,DU National conference By: Yeshitela

12/14/2019 18
Shiferaw
 6.Proposed Bat Algorithm

Step of BAT algorithm for proposed system

Step 5: Finding the best solution according to the objective

function and store the current population.

Step 6: Randomly updating the current micro-bats population to

update position vector and velocity vector of the micro-bats.

Step 7: Evaluating the objective of the new micro-bats population

and select the best solution among the solution.

CoE,DU National conference By: Yeshitela

12/14/2019 19
Shiferaw
 6.Proposed Bat Algorithm

Step of BAT algorithm for proposed system

Step 8: Finding the power loss, voltage stability indices, and load
flow solutions of the best solution. The penalty factors are added to
the power loss if the constraints are violated.

Step 9: Checking the termination criterion.

Step 10: Generating new agents to generate new solutions. If not

satisfied. Go to Step 4.

Step 11: If the maximum iteration touched, then display the results
otherwise repeat step 4-10.

CoE,DU National conference By: Yeshitela

12/14/2019 20
Shiferaw
 7. Result and disussion
The buses systems have 6 generators, 21 PQ loads and a connected bus.
The system base is 100MVA, 60Hz and the total load demand is
283.4MW

CoE,DU National conference By: Yeshitela

12/14/2019 21
Shiferaw
 7. Result and discussion
Parameter of DE algorithm Optimal values for
Parameter of PSO algorithm Optimal values for IEEE-
IEEE-30
30

Number of particles(D) 6
Number of particles 6
Number of population (N) 10
Number of population (N) 40
Number of iterations 150
Number of iterations 150
Differentiation constant (F) 0.9
Cognitive parameter (C1) 2
Crossover constant (CR) 0.5
Social parameter (C2) 2
Number of generation (G) 10

Parameter of bat algorithm Optimal values for IEEE-30

Number of particles 6

Number of population 40

Number of iterations 150

Loudness(Ai) 0.5

Pulse rate(r) 0.5

Frequency minimum 0

Frequency maximum 2

CoE,DU National conference By: Yeshitela

12/14/2019 22
Shiferaw
 7. Result and discussion
Generator cost comparisons

CoE,DU National conference By: Yeshitela

12/14/2019 23
Shiferaw
 7. Result and discussion
Fitness comparisons

CoE,DU National conference By: Yeshitela

12/14/2019 24
Shiferaw
 7. Result and discussion

• Optimal setting of control variables • Voltage, Real and Reactive power

for IEEE-30 bus system network generator limits for IEEE-30 bus
system network
Control Limits (pu) PSO DE BAT Bus No Pmin Pmax Qmin Qmax Vmin Vmax
variables
Min Max

PG1 0.50 2.00 1.7642 1.7561 1.8227 1 50 200 -40 200 0.95 1.05
PG2 0.20 0.80 0.4898 0.2653 0.4361

PG5 0.15 0.50 0.2246 0.2140 0.1500 2 20 80 -20 100 0.95 1.05

PG8 0.10 0.35 0.1228 0.1000 0.1000

PG11 0.10 0.30 0.2154 0.2149 0.2510 5 15 50 -15 80 0.95 1.05

PG13 0.12 0.40 0.1200 0.1611 0.1200

Cost ($/hrs.) 803.84 805.55 768.51 8 10 35 -15 60 0.95 1.05

Ploss(pu) 0.0993 0.1071 0.0807 11 10 30 -10 50 0.95 1.05

Lj(max) 0.1488 0.1487 0.1478 13 12 40 -15 50 0.95 1.05

The global cost Using MATPOWER= 885.98 $/hrs.

CoE,DU National conference By: Yeshitela

12/14/2019 25
Shiferaw
 8. Conclusions
This paper dealt with comparisons of BAT algorithm with PSO
and DE algorithm optimization techniques. The following
conclusions are drawn from the work carried out:-.

 The optimal power flow is mathematically formulated in the

power system network with equality and inequality constraints.

 Voltage stability index is formulated.

 Initialization of Bat algorithm was done and incorporated with

Newton Raphson load flow equations.

CoE,DU National conference By: Yeshitela

12/14/2019 26
Shiferaw
 8. Conclusions
 The proposed algorithm result was compared with
MATPOWER toolbox, PSO, and DE algorithm results in terms
of the generator cost, real power loss and voltage stability
indexes.
 The simulation results verified that, the effectiveness of bat
algorithm approach to solve the OPF problem. Hence, Bat
algorithm is a powerful method to solve optimal power flow on
standard IEEE-30 bus test system than MATPOWER, PSO, and
DE algorithm because it frequently runs like a fast convergence
rate,and frequency tuning to get the optimal value.

CoE,DU National conference By: Yeshitela

12/14/2019 27
Shiferaw
 Thank you !

CoE,DU National conference By: Yeshitela

12/14/2019 28
Shiferaw