EFFICIENT GENETIC ALGORITHM FOR PLACEMENT AND SIZING OF MULTIPLE

DISTRIBUTED GENERATORS IN RADIAL DISTRIBUTION SYSTEMS THROUGH

HYBRID OPTIMIZATION

Chapter 4 Computational Algorithms

Considering the literature review, we worked on amalgamating BFS and BB-BC in this
project to solve the challenging optimization problem OADG. The OADG is first solved,
considering mono-objective: real power loss minimization. Later OADG is also solved in a
multi-objective framework combining the indices of real power minimization and voltage
deviation improvement with suitable weights. Further, this thesis evaluates three different
modes of DG operation (P only, Q only or both P& Q) for the multi-objective approach.

4.1 Mono Objective Formulation

4.1.1 Real power loss

Computation of real power loss in the branch ‘k’ is the product of the current flowing
through and the corresponding line resistance ( I 2k × Rk ) on the sample distribution network;
the line loss can be formulated as:

( )
2 2
P( L (i +1) )+Q ( L (i +1))
P Loss( k )=( I ) × Rk =
2
k 2
× R k Eq. 4.1.1.1
V ((i +1) )

In the same principle, the reactive power loss in branch ‘k’ between ‘i’ and ‘i+1’ can be
expressed as:

( )
2 2
P( L ( i+1) )+Q ( L (i +1))
Q Loss( k )=( I ) × Rk =
2
k 2
× X k Eq. 4.1.1.2
V ((i +1) )

where:

P( L ( i+1 )) is the total effective active power supplied beyond the node ‘i+1’

Q( L (i+1 )) is the total effective reactive power supplied beyond the node ‘i+1’
V i is voltage at node ‘i’

Rk is the resistance of the branch ‘k’

X k is the reactance of the branch ‘k’

Real power loss minimization helps DGs supply power to the surplus loads with existing
substation power. Further, reducing Real Power loss also results in savings for DGs
regarding reduced energy loss costs. Therefore, the objective function for Real power loss
minimization is very crucial and is defined as

M 1=min ( P Loss ) Eq. 4.1.1.3

where

P Loss or P L is the real power loss of the system calculated from the two-bus equivalent of a
distribution system using Eq. 4.1.1.4 as [3]
Nb
P Loss=∑ I 2(k , k+1) × R( k ,k +1) Eq. 4.1.1.4
k=1

where

R( k ,k +1) and I (k , k+1 ) are the resistance and the current flowing through the branch connected

by the k th and ( k +1 )th nodes, respectively.

Nb represents the total number of branches of the network.

4.1.2 Voltage deviation

The loads connected towards the end of radial distribution networks experience a poor
voltage level, and an increase in load demand exacerbates the scenario. Further, the
addition of sensitive loads requires a flat voltage profile. Therefore, the following voltage
deviation objective function measures the degree to which a flat voltage profile is
maintained.

M 2=min ( VD ) Eq. 4.1.2.1

where VD is the voltage deviation of the network, which is the total of the square of the
differences in the bus voltage level of each node from the subs station and is calculated
as:
Nn
VD=∑ ( V s −V k ) Eq. 4.1.2.2
2

k =1

where V s and V k are the voltage magnitudes of the substation and the k th node,
respectively, N n represents the total number of nodes of the network.

4.2 Multi-Objective Formulation

The weighted sum multi-objective function is formulated by combining the two indices
(IRPL, IVD) is expressed below:

minM =w1 min ( P( Loss , DG) ) + w2 min ( VD ( DG) ) Eq. 4.2.1

Where w 1 and w 2 are the weights and decide the impact of an individual index on the
overall objective function, the utilities can judiciously select these weights on a priority
basis. However, this study considers the values of the weights as 1 and 1 respectively.

As a result, the objective function minimizes the associated losses and the voltage
deviation, as shown in Eq. 4.1.1.3 and Eq. 4.1.2.1. In this regard, the optimal location and
the corresponding DG unit size, which result in minimum power losses and minimum
voltage deviation from the nominal value without violating the system constraints, will be
systematically determined.

4. 3 Constraints
The objective functions defined in Eq. 4.1.1.3, Eq. 4.1.2.1, Eq. 4.2.1 are optimized for the
following pragmatic constraints.

 Bus voltage constraint

The bus voltage is allowed to vary within -5% to +5% of the nominal voltage

V min ≤ V k ≤V max Eq. 4.3.1

 Branch flow constraint

 DGs can affect the branch flow of the network. Therefore, the following constraint is
defined to restrict the branch flow in the presence of DGs within safe limits.
max
I (k , k+1 ) ≤ I ( k ,k +1) Eq. 4.3.2

max
Where I (k , k+1 ) is the branch current, and I (k , k+1 ) is the maximum permissible branch
current

 DG position constraint

The nodes for DG integration are generated using Eq. 4.3.3 without any repetition:

2 ≤ DG loc ≤ N n Eq. 4.3.3

 DG capacity constraints

As defined in eq. 4.3.4, the DG sizes must be within the capacity limit,

0 ≤ PDG ≤ 3000 kW ; 0 ≤ QDG ≤ 3000 kVAR Eq. 4.3.4

4.4 Particle swarm optimization algorithm

The inertia weights are fixed throughout the entire iteration in the conventional PSO
algorithm. Fixed inertia weights slow down the searching speed of the particles and hence
take more time to reach the optimal solution [1]. Therefore, updating the inertia weights in
all iterations is necessary to minimize the searching time and simplify the manipulation.
This algorithm is called IPSO and is applied in this project. In this condition, the inertia
weight factor w is varied at each step by multiplying its updated value with the velocity of
the particles obtained in the previous iteration, i.e., V tid . The particle velocity is a sum of the
inertia factor, the personal and social influence in the swarm and can be altered using the
following equation.

V (idt+1 )=w × V tid +c 1 × rand 1 × ( Ptid − X tid ) + c 2 × rand 2 × ( Ptgbd − X tid ) Eq. 4.4.1

Hence, the evaluation of PSO is modified from iteration to iteration. In other words, unlike
the conventional PSO, the linear recursive function is modified, and the i th particle position
is also updated using the following equation.

( t +1) t ( t +1)
X id = X id + V id , fori=1 , 2 ,... , n and d=1, 2 , ..., m Eq. 4.4.2
However, velocity values are restricted to some minimum and maximum values [
velmin , velmax ] using the following equation.

( t +1) ( t +1)
v id = velmax if v id > velmax; Eq. 4.4.3

( t +1) ( t +1 ) ( t +1)
v id =v id if vid ≤ velmax

( t +1) ( t +1)
v id =velmin if v id ≤ velmin

Generally, the procedures for the IPSO algorithm are presented in the flow chart depicted
in Fig.

<Flow Chart>

4.5 Artificial Bee Colony Algorithm

ABC algorithm is centered around the behaviours of three types of bees, including
employee, onlooker and scout bees. The number of employee bees equals the number of
onlooker bees in a hive. The structure of the basic algorithm is in three phases: the
employee bees' phase, the onlooker bees' phase, and the scout bees' phase. In this
project, a food source represents a solution that includes the size and location of a DG in a
distribution network. The food source's nectar amount (quality) represents the objectives,
including total active power loss, average voltage drop, and total energy cost. The pseudo-
code of the ABC algorithm for the DG sizing and placement problem is as follows:

Step I: Read the line and load data, perform load flow analysis, and find the values of
objectives without adding the DGs.

Step II: Define parameters of ABC algorithm:

number of employee bees,

number of onlooker bees,

number of limit cycles,

initial food sources (random solutions) and

number of iterations (MI)

Step III: Read the line and load data and analyze load flow by incorporating algorithm-
controlled DG sizes and locations.

Again, find and store the values of objectives (employee bees’ phase)

Evaluate the values of objectives and select the best ones (onlooker bees’ phase)

Set iteration to 1

Step IV: Initiate more food sources (random solutions). Read the line and load data, and
perform load flow analysis incorporating new DG sizes and locations.

Again, find and store the values of objectives (employee bees' phase)
Evaluate the values of objectives and select the best ones (onlooker bees' phase)
Find scout bees (some random solutions as of the first iteration)

Rnew =R oldn +u ( Roldm − R oldn ); m ≠n Eq. 4.5.1

where Rnew is the updated solution,

Roldm is the old solution at random food source position m

Roldn is the old solution at food source position n

u is the random number between [-2, 2]

The probability R R of the fitness of the food sources is calculated by using

f itness R
R R= Ne
Eq. 4.5.2
∑ fitness R
R =1

Set iteration = iteration +1.

Step V: Repeat Step IV until iterations = MI

Step VI: Find the probability RR of fitness of solutions using Eq. 4.5.2

4.6 Big Bang Big- Crunch Algorithm

Erol and Eksin developed a simple nature-based or physics-based meta-heuristic. Energy
dissipation in the form of transformation from an ordered state to a disordered or chaotic
state is the inspiration for this algorithm. In the big-bang phase, particles are drawn toward
irregularity by losing energy, while in the Big Crunch phase, they converge toward a
specific direction.
Like other population-based meta-heuristics, the BB-BC algorithm starts with a set of
randomly generated initial candidate solutions in the search phase, known as the initial Big
Bang. Each cycle of the algorithm is composed of two phases: the Big Crunch phase, in
which a converging operator is formed, and the Big Bang phase, in which particles are
updated in the search space with the step sizes in the vicinity of the converging operator
generated in the first phase.

Consider a certain number of particles ( nP ) as the population or candidate solutions matrix

( P), its corresponding penalized objective function ( PFit ) vector, and the best-observed
particle in each iteration (bestP ) with the smallest value of penalized objective function.

After each Big Bang phase, a Big Crunch phase should take place to determine the
convergence operator by which particles will be drawn into an orderly fashion in the
subsequent Big Bang phase.

The convergence operator based on the Big Crunch phase can be defined as the weighted
average of the candidate solution positions known as the centre of mass ( CM ) or the
position of the best candidate solution (bestP ). For minimization problems, CM is
formulated as follows:

( )
nP
P ( j ,i )
∑ PFit ( j )
j=1
CM ( i )= nP , i=1 , ..., nv Eq. 4.6.1
∑( 1
PFit j( ))
j=1

The Big bang phase occurs now. In the original BB-BC, the particles are updated simply
for the previously determined centre of mass (CM ) or the position of the step size defined
by the upper (Ub) and Lower ( Lb ) bound of design variables:

rand × ( Ub− Lb )
newP=( CM ∨best P ) + Eq. 4.6.2
nIT

where a rand is a random number uniformly distributed in (0,1), the step size is also
divided by the number of algorithm iterations or the number of the Big Bang phases ( nIT s)
to generate the effective search range about the global optimum or centre of mass with the
aim of narrowing the search with the progress of the algorithm.
These two contraction (Big Crunch) and dispersing (Big Bang) phases are repeated in the
cyclic body of the algorithm in succession to satisfy a stopping criteria to steer the particles
toward the global optimum of the problem. [3] The two parameters that are required for all
meta-heuristics as stopping criteria are:

1. number of the algorithm population

2. maximum number of algorithm iterations

However, Camp [4] presented a new formulation with two additional parameters for the Big
Bang phase and showed the efficiency of the newly proposed formulation. They are:

3. To limit the size of the search space (α )

4. To consider both CM and bestP as the converging operators, β

The new formulation was proposed as:

rand × α × ( Ub− Lb )
newP ( i )=( β ×CM + (1 − β ) ×bestP ) + , i=1 , ..., nP Eq. 4.6.3
nIT

This modified formulation is used and coded in this project. It should be noted that the BB-
BC algorithm does not need a replacement strategy. In other words, particles leave their
position no matter if their current position is better.

The pseudo code of algorithm is provided as follows, and the flow chart of BB-BC is
illustrated in Fig.
According to this flowchart, it is evident that one cannot explicitly distinguish between
intensification and diversification because the BB-BC involves these two features together.