Multi-Objective Optimal Scheduling for Adrar Power System

including Wind Power Generation

Redha Djamel Mohammedi1, Mustafa Mosbah2, Abdellah Kouzou1
1
Applied Automation and Industrial Diagnostics Laboratory (LAADI), Djelfa University, Algeria
2
Engineer Operating in Algerian Distribution Electricity and Gas Company

Abstract
Recently, the high penetration of wind power and other renewable generation into the power grids has made the
resource scheduling a very interesting and challenging optimization problem. In this paper, a Multi-Objective Wolf
Optimizer (MOGWO) is proposed to solve the multi-objective optimal power flow problem (MO-OPF) in a power
system including wind farms. Three objective functions are simultaneously considered which are: reduction of
generation cost, reduction of real power loss, and enhancement of voltage stability while satisfying several equality
constraints (e.g., load flow equations) and inequality constraints (e.g., operational and security limits). The
proposed algorithm is applied in isolated Adrar power system situated in the south of Algeria. The obtained results
reveal the efficiency and effectiveness of the proposed approach.
Keywords: Optimal Power Flow, Power Losses, Voltage Stability, Wind Power Uncertainty, Multi-Objective Wolf
Optimizer

Received: September 13, 2018

To cite this article:
Mohammedi R.D., Mosbah M., Kouzou A., “Multi-Objective Optimal Scheduling for Adrar Power System
including Wind Power Generation”, in Electrotehnica, Electronica, Automatica (EEA), 2018, vol. 66, no. 4,
pp. 102-109, ISSN 1582-5175.

1. Introduction The proposed approach is applied on an existing power

Over the last decades, wind energy has demonstrated a system in the region of Adrar, situated in the south of
rapid growth as a clean and affordable energy source in the Algeria.
worldwide. Unlike conventional power generation sources, The remainder of this paper is organized as follows: In
electricity generated with wind turbines is subject to strong Section 2, the mathematical formulation of the OPF
fluctuations due to the intermittent nature of wind speed. problem is described.
Further research must be conducted to modelling the Section 3 briefs the MOGWO algorithm.
generation scheduling problems taking into consideration In Section 4, the detailed application of MOGWO in
the variability and intermittency of wind power. proposed problem is given.
One fundamental problem in the optimization of power The simulation test is presented and discussed in
systems scheduling is the optimal power flow (OPF) Section 5. Section 6 concludes the paper.
problem.
The objective of OPF is to optimize selected objective 2. Problem formulation
functions, while satisfying several equality and inequality The desired goal of the OPF is to optimize a certain
constraints. power system objectives with a set of equality and
Generally, OPF problem is a highly constrained, inequality constraints [13]. The mathematical model of
nonlinear, non-smooth, and non-convex optimization such problem can be expressed as follows
problem. The volatile nature of wind speed is modelled
𝑀𝑖𝑛𝑖𝑚𝑖𝑧𝑒 𝐹(𝑋, 𝑈) (1)
using the Weibull probability distribution function (PDF)
and both the overestimation and underestimation costs are 𝑆𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜
incorporated into the OPF problem.
A survey of early works related to the OPF is provided 𝐶𝐸 (𝑋, 𝑈) = 0 (2)
in Ref. [1-3]. Several research works have been reported on
OPF with inclusion of wind generation in Refs. [4-12]. 𝐶𝐼 (𝑋, 𝑈) ≤ 0 (3)
In this paper, a Multi-Objective Grey Wolf Optimizer where
(MOGWO) is suggested to find the optimal scheduling for 𝑋 and 𝑈 are the vectors of state and decision
Adrar power system including wind power uncertainties. variables,
Three objective functions are considered: 𝐹 is the objective functions to be optimized,
− generating cost minimization, 𝐶𝐸 and 𝐶𝐼 are the equality and inequality
− active losses minimization constraints.
− voltage stability improvement.
 ELECTROTEHNICĂ, ELECTRONICĂ, AUTOMATICĂ (EEA), 66 (2018), nr. 4 103
The decision vector 𝑈 is consisting of: 𝑣 𝑘 𝑣𝑜𝑢𝑡 𝑘
𝑓(𝑝 = 𝑝𝑟 ) = exp (− ( 𝑟) ) + exp (− ( ) ) (11)
− Active power of slack bus generator 𝑃𝑠𝑙𝑎𝑐𝑘 . 𝑐 𝑐
− Voltage level at load-buses (𝑃𝑄 buses).
− Reactive power of generator buses (𝑃𝑉 buses). 2.2. Wind generating cost
− Apparent power flow on overloaded lines 𝑆𝑙 . Because of the randomness of wind speed, the power
The state vector 𝑈 can be written as: produced by wind farms may not be equal to the scheduled
power at a given time.
𝑈 = [𝑃𝑠𝑙𝑎𝑐𝑘 , 𝑉1 , … 𝑉𝑁𝑃𝑄 , 𝑄𝐺1 , … 𝑄𝑁𝑃𝑉 , 𝑆𝑙1 , … 𝑆𝑙 𝑁𝑇𝐿 ] (4)
Regarding this, two cost factors such as overestimation
where and underestimation of available wind power will need to
be included in the total generating cost as follows [16]:
N PQ , N PQ , and N TL are number of load buses,
number of generator buses and number of 𝑤𝑖𝑛𝑑 𝑔𝑒𝑛𝑒𝑟𝑎𝑡𝑖𝑛𝑔 𝑐𝑜𝑠𝑡 = 𝐶𝑑,𝑖 (𝑝𝑠ℎ𝑒𝑑,𝑖 ) +
(12)
transmission lines, respectively. 𝐶𝑝,𝑖 (𝑝𝑠ℎ𝑒𝑑,𝑖 ) + 𝐶𝑟,𝑖 (𝑝𝑠ℎ𝑒𝑑,𝑖 )
The decision variables vector X is consisting of:
− Active power of generator buses. Cd,i (pshed,i )=di pshed,i (13)
− Active power output from wind farms.
𝑝𝑟,𝑖
− Voltage level of generator buses. 𝐶𝑝,𝑖 (𝑝𝑠ℎ𝑒𝑑,𝑖 ) = 𝐾𝑝,𝑖 ∫ (𝑝 − 𝑝𝑠ℎ𝑒𝑑,𝑖 )𝑓(𝑝)𝑑𝑝 (14)
− Tap position of tap-changing transformers. 𝑝𝑠ℎ𝑒𝑑,𝑖

Consequently, the decision variables vector 𝑋 can be

𝑝𝑠ℎ𝑒𝑑,𝑖
written as follows: 𝐶𝑟,𝑖 (𝑝𝑠ℎ𝑒𝑑,𝑖 ) = 𝐾𝑟,𝑖 ∫ (𝑝𝑠ℎ𝑒𝑑,𝑖 − 𝑝)𝑓(𝑝)𝑑𝑝 (15)
0
𝑋 = [ 𝑃1 , … 𝑃𝑁𝑃𝑉 , 𝑃1 , … 𝑃𝑁𝑊𝐺 , 𝑉1 , … 𝑉𝑁𝑃𝑉 +𝑁𝑊𝐺 , 𝑇1 , … 𝑇𝑁𝑡 ] (5) where
where 𝑑𝑖 is the direct cost coefficient (in $/h),
𝑁𝑡 and 𝑁𝑊𝐺 are the total number of tap 𝐶𝑝,𝑖 and 𝐶𝑟,𝑖 are overestimation and
transformers and the total number of the wind underestimation costs (in $/h),
farms, respectively. 𝑝𝑠ℎ𝑒𝑑,𝑖 is the schedule wind power output (𝑝𝑠ℎ𝑒𝑑,𝑖
ranges from 0 to 𝑝𝑟,𝑖 ),
2.1. Wind energy system 𝐾𝑝,𝑖 and 𝐾𝑟,𝑖 are penalty cost coefficients for
The power developed by a wind turbine is given by [14]: overestimation and underestimation of
0 𝑓𝑜𝑟 𝑣 < 𝑣𝑖𝑛 𝑎𝑛𝑑 𝑣 > 𝑣𝑜𝑢𝑡 available power for the ith wind farm.
𝑣−𝑣𝑖𝑛
𝑝 = {𝑝𝑟 (𝑣 ) 𝑓𝑜𝑟 𝑣𝑖𝑛 ≤ 𝑣 ≤ 𝑣𝑟 (6) 2.3. OPF objective functions
𝑟 −𝑣𝑖𝑛
𝑝 𝑓𝑜𝑟 𝑣𝑟 ≤ 𝑣 ≤ 𝑣𝑜𝑢𝑡 The following three objective functions are considered
in this paper:
where − minimization of total cost,
𝑝 is the electric power (in MWs), − minimization of active power loss,
𝑝𝑟 is the rated power, − enhancement of voltage stability.
𝑣 is the wind speed (in m/s),
𝑣𝑖𝑛 , 𝑣𝑜𝑢𝑡 , 𝑣𝑟 are cut-in, cut-out and rated wind 2.3.1. Objective 1: Operating cost minimization
speeds of the wind turbines (in m/s). The total cost (including fuel cost and wind farm cost)
The wind speed is statistically modelled by using is considered as the main objective function of practical
Weibull Probability Density Function (PDF), and is OPF problem. It can be expressed as:
expressed as [15], Minimize,
𝑘 𝑘 𝑘−1 𝑘 𝑘 𝑁𝑇𝐺
𝑓(𝑣) = ( ) ( ) exp [− ( ) ] 0<𝑣<∞ (7) 𝐹1 = ∑ 𝑊𝐺
𝐶𝑖 (𝑃𝐺𝑖 ) + ∑𝑖=1
𝑁
𝐶𝑑,𝑖 (𝑝𝑠ℎ𝑒𝑑,𝑖 ) +
𝑐 𝑐 𝑐
𝑖=1 (16)
Cumulative distribution function (CDF) of Weibull 𝑁
𝑊𝐺
∑𝑖=1 𝑊𝐺
𝐶𝑝,𝑖 (𝑝𝑠ℎ𝑒𝑑,𝑖 ) + ∑𝑖=1
𝑁
𝐶𝑟,𝑖 (𝑝𝑠ℎ𝑒𝑑,𝑖 )
distribution is given by
The first term of 𝐹1 consist of the sum of fuel costs of
𝑘 𝑘
𝐹(𝑣) = 1 − exp [− ( ) ] (8) thermal generators.
𝑐
The second term is the direct cost of the wind power.
From Eq. (7) and (8), the probability density function The third and fourth are overestimation and
𝑓(𝑝) for the wind power in the interval 0 < 𝑝 < 𝑝𝑟 , is underestimation costs of the available wind power,
described as: respectively.
The fuel cost of thermal unit is approximated by a
𝑘(𝑣𝑟 −𝑣𝑖𝑛 ) 𝑣𝑖𝑛 𝑝𝑟 +𝑝(𝑣𝑟 −𝑣𝑖𝑛 ) (𝑘−1)
𝑓(𝑝) = ( )( ) × quadratic function as follows [13]:
𝑐𝑝𝑟 𝑐𝑝𝑟
(9)
𝑣𝑖𝑛 𝑝𝑟 +𝑝(𝑣𝑟 −𝑣𝑖𝑛 ) 𝑘 𝐶𝑖 (𝑃𝐺𝑖 ) = 𝑎𝑖 + 𝑏𝑖 𝑃𝐺𝑖 + 𝑐𝑖 𝑃𝐺2𝑖 (17)
exp (− ( ) )
𝑐𝑝𝑟
𝑎𝑖 , 𝑏𝑖 and 𝑐𝑖 are fuel cost function coefficients,
The probability of wind power being 0 or 𝑝𝑟 can be which are found from the input–output curve of
calculated as (10) and (11): the ith thermal unit.
𝑣𝑖𝑛 𝑘 𝑣𝑜𝑢𝑡 𝑘 2.3.2. Objective 2: Transmission losses minimization
𝑓(𝑝 = 0) = 1 − exp (− ( ) ) + exp (− ( ) ) (10)
𝑐 𝑐
The second objective function is minimizing the real
transmission losses. This function is the sum of power losses
in the transmission system, it can be formulated as follow:
104 ELECTROTEHNICĂ, ELECTRONICĂ, AUTOMATICĂ (EEA), 66 (2018), nr. 4

Minimize, The equality constraints in (21)-(24) are a set of

𝑁𝑇𝐿
nonlinear equations, which can be solved using Newton-
𝐹2 = ∑ 𝐺𝑞(𝑖,𝑗) [𝑉𝑖2 + 𝑉𝑗2 − 2 𝑉𝑖 𝑉𝑗 cos(𝛿𝑖𝑗 )] (18) Raphson method [18].
𝑞=1
2.4.2. Inequality Constraints
where (i) Slack bus constraint
𝛿𝑖𝑗 = 𝛿𝑖 - 𝛿𝑗 indicates the voltage phase angle Real power generation output at slack bus must lie
difference between buses 𝑖 and 𝑗, between its upper and lower bounds:
𝐺𝑞(𝑖,𝑗) is the conductance of qth line connecting
min max
between buses 𝑖 and 𝑗. 𝑃𝑠𝑙𝑎𝑐𝑘 ≤ 𝑃𝑠𝑙𝑎𝑐𝑘 ≤ 𝑃𝑠𝑙𝑎𝑐𝑘 (25)

2.3.3. Objective 3: Voltage stability improvement (ii) Security constraints

Voltage of each load bus (PQ bus) must be within the
A fast indicator of voltage stability (LVSI Index)
allowable minimum and maximum limits. Also, the
proposed by Moghavvemi and al [17] is selected voltage
apparent power flow through each transmission line must
stability assessment. The LVSI index for the kth line
not exceed the maximum safe capacity. These are
connecting bus i and j is obtained as follows:
described as follows:
4𝑋𝑖𝑗 𝑄𝑗
𝐿𝑉𝑆𝐼𝑘 = 2 ≤ 1.0 (19) 𝑉𝑖min ≤ 𝑉𝑖 ≤ 𝑉𝑖max , 𝑖 = 1, … , 𝑁𝑃𝑄 (26)
[𝑉𝑖 sin(𝜃𝑖𝑗 −𝛿𝑖𝑗 )]

The line that gives index value closest to 1 will be the

|𝑆𝑙𝑖 | ≤ 𝑆𝑙𝑖max , 𝑖 = 1, … , 𝑁𝑇𝐿 (27)
most critical line and may lead to the whole system
instability. The mathematical formulation of third where
objective function can be written as: 𝑉𝑖min and 𝑉𝑖max the upper and lower voltage limits
Minimize, of the ith load bus,
𝑆𝑙𝑖max is the maximum capacity limit of the ith
𝐹3 = max(𝐿𝑉𝑆𝐼𝑘 ), 𝑘 = 1, … , 𝑁𝑇𝐿 (20)
transmission line.
In order to handle the inequality constraints in Eqs.
2.4. Constraints (25)-(27), a penalty term is added to the objective function
which degrades the fitness in cases with violated constraint
2.4.1. Equality Constraints
[19]:
The optimal solution should satisfy the real and reactive
lim 2
power balance equations at each bus. Also, the generated 𝑃𝑒𝑛𝑎𝑙𝑖𝑡𝑦 = 𝜆𝑆𝐿 (𝑃𝑠𝑙𝑎𝑐𝑘 − 𝑃𝑠𝑙𝑎𝑐𝑘 ) +
power should be equal to the summation of the total load 𝑁𝑃𝑄 2 𝑁𝑇𝐿 2 (28)
demand and total transmission losses. 𝜆𝑉 ∑ (𝑉𝑖 − 𝑉𝑖lim ) + 𝜆𝑆 ∑ (𝑆𝑙𝑖 − 𝑆𝑙𝑖lim )
𝑖=1 𝑖=1

𝑃𝐺𝑖 + 𝑃𝑊𝑖 − 𝑃𝐷𝑖 = 𝑉𝑖 ∑ 𝑉𝑗 (𝐺𝑖𝑗 cos(𝛿𝑖𝑗 ) + 𝐵𝑖𝑗 sin(𝛿𝑖𝑗 )) where, (𝜆𝑆𝐿 ), (𝜆𝑉 ), and (𝜆𝑆 ). are penalty factors.
𝑗∈𝛺𝑖 The limit values of above variables can be obtained
𝑓𝑜𝑟 𝑖 = 1, … , 𝑁 using:
(21)
𝑧 max 𝑖𝑓 𝑧 > 𝑧 max
𝑧 lim = {𝑧 min 𝑖𝑓 𝑧 < 𝑧 min (29)
𝑄𝐺𝑖 + 𝑄𝑊𝑖 − 𝑄𝐷𝑖 = 𝑉𝑖 ∑ 𝑉𝑗 (𝐺𝑖𝑗 sin(𝛿𝑖𝑗 ) − 𝐵𝑖𝑗 cos(𝛿𝑖𝑗 )) 𝑧 𝑖𝑓 𝑧 min ≤ 𝑧 ≤ 𝑧 max
𝑗∈𝛺𝑖
𝑓𝑜𝑟 𝑖 = 1, … , 𝑁 where 𝑧 can be 𝑃𝑠𝑙𝑎𝑐𝑘 , 𝑉𝑖 or 𝑆𝑙𝑖 .
(22) In this study, the penalty factors were 1000 for the
slack bus real power (𝜆𝑆𝐿 ), 10000 for voltage limits (𝜆𝑉 ),
𝑁𝑃𝑉 𝑁𝑊 𝑁 and 10000 for apparent transmission power limits (𝜆𝑆 ).
𝑃𝑠𝑙𝑎𝑐𝑘 + ∑ 𝑃𝐺𝑖 + ∑ 𝑃𝑊𝑖 = ∑ 𝑃𝐷𝑖 +
𝑖=1 𝑖=1 𝑖=1
𝑁 3. Multi-Objective Wolf Optimizer (MOGWO)
∑𝑖,𝑗=1𝑃𝑙𝑜𝑠𝑠 (𝑖𝑗) (23)
3.1. Overview
𝑁𝑃𝑉 𝑁𝑊 𝑁
In Ref. [20], S. Mirjalili et al. proposed a multi-objective
𝑄𝑠𝑙𝑎𝑐𝑘 + ∑ 𝑄𝐺𝑖 + ∑ 𝑄𝑊𝑖 = ∑ 𝑄𝐷𝑖 +
𝑖=1 𝑖=1 𝑖=1 version of new metaheuristic, population-based
𝑁 optimization technique, called “Grey Wolf Optimizer”
∑𝑖,𝑗=1𝑄𝑙𝑜𝑠𝑠 (𝑖𝑗) (24)
(GWO).
where This technique emulates the hierarchy leadership and
𝑃𝐺𝑖 and 𝑄𝐺𝑖 are real and reactive powers production hunting strategy of grey wolves. Similarly to other meta-
of generator at bus 𝑖, heuristics, it starts the optimization procedure by
𝑃𝑊𝑖 and 𝑄𝑊𝑖 are real and reactive power generating a set of random candidate solutions (wolves).
The fittest solution is called the alpha (α) and referred
generations of wind farm at bus 𝑖,
as the global best solution.
𝑃𝐷𝑖 and 𝑄𝐷𝑖 real and reactive power consumed at
The second and third best solutions are named beta (β)
the ith bus, and delta (δ) respectively.
𝐵𝑖𝑗 and 𝐺𝑖𝑗 are the susceptance and conductance The rest of the candidate solutions are assumed to be
between buses 𝑖 and 𝑗, omega (ω).
𝛺𝑖 is the set of buses incident to the ith bus. The hunting, which represents the optimization
process, is directed by the three best solutions α, β and δ
toward promising areas of the search space (Figure 1).
 ELECTROTEHNICĂ, ELECTRONICĂ, AUTOMATICĂ (EEA), 66 (2018), nr. 4 105
It should be noted that the parameters 𝐴 and 𝐶 forces
the GWO algorithm to explore and exploit the entire search
space.
Half of the iterations is devoted to exploration
(diversification) when |𝐴| ≥ 1, and the rest of iterations is
devoted to exploitation (intensification) when |𝐴| < 1. The
parameter 𝐶 switches randomly to escape from local
optima traps during the iterative process of optimization.
3.2. Multi-objective solution methodology

3.2.1. Pareto optimal solutions

Multiobjective optimization (MOO) is the procedure of
simultaneously optimizing two or more conflicting
objectives subject to a set of constraints.
Instead of single objective optimization which yields a
single best solution, the MOO search for finding a set of
optimal solutions called Pareto optimal solutions. The set
of Pareto optimal solutions is consisted of all non-
dominated solutions which are not dominated by other
Figure 1. Hunting strategy solutions.
When solution 𝑆1 dominates solution 𝑆2 , it means:
Grey wolves tend to encircle the prey before hunting. − 𝑆1 is no worse than 𝑆2 in all objectives.
The encircling procedure performed by the grey wolves is − 𝑆1 is strictly better than 𝑆2 in at least one
mathematically defined by: objective.

𝑋𝑝 (𝑖𝑡𝑟) − 𝐴⃗ ⋅ 𝐷
𝑋⃗(𝑖𝑡𝑟 + 1) = ⃗⃗⃗⃗⃗ ⃗⃗ (30) 3.2.2. Best compromise solution
Once, a set of Pareto optimal solutions is obtained
where through MOGWO, the network manager can select one
⃗⃗ = |𝐶⃗⃗⃗⃗⃗⃗ optimal solution according to his preference.
𝐷 𝑋𝑝 (𝑖𝑡𝑟) − 𝑋⃗| (31)
Otherwise, a decision-making based on fuzzy theory
where [21] can be utilized to recognize the most appropriate
solution within the set of Pareto-optimal solutions.
𝐴⃗ = 2𝑎⃗ ⋅ 𝑟𝑎𝑛𝑑1 − 𝑎⃗ (32) The membership function of each objective function is
defined as [21]:
𝐶⃗ = 2 ⋅ 𝑟𝑎𝑛𝑑2 (33) 1 ; 𝐹𝑛 ≤ 𝐹𝑛min
where 𝐹𝑛max −𝐹𝑛
𝜇𝑛 = { ; 𝐹𝑛min ≤ 𝐹𝑛 ≤ 𝐹𝑛max 𝑛 = 1, 2, 3 (38)
𝑖𝑡𝑟 is the iteration number, 𝐹𝑛max −𝐹𝑛min

𝑋⃗ is the position vector of a grey wolf, 0 ; 𝐹𝑛 ≥ 𝐹𝑛max

⃗⃗⃗⃗⃗
𝑋𝑝 is the position vectors of prey, where
𝐴⃗ and 𝐶⃗ are coefficient vectors obtained using Eq. 𝐹𝑛max and 𝐹𝑛min are the upper and lower values of
(32) and (33), each objective function.
𝑟𝑎𝑛𝑑1 and 𝑟𝑎𝑛𝑑2 are random vectors in [0, 1]. 𝑚
∑ 𝜇𝑖𝑘
The variable 𝑎⃗ decreases linearly from 2 to 0 over the 𝜇𝑘 = 𝑖=1
(39)
𝐷
course of iterations: 𝑚
∑ ∑ 𝜇𝑖𝑘
𝑖=1
𝑎⃗ = 2 − 𝑖𝑡𝑟⁄𝑖𝑡𝑟max (34) 𝑘=1

where
So, the three best solutions α, β and δ are stored and
𝑚 is the total number of objectives to be
the other agents ω are forced to update their positions
optimized,
according to the best agents positions as follows.
𝐷 is the total number of non-dominated solutions.
⃗⃗⃗⃗⃗⃗
𝐷𝛼 = |𝐶⃗⃗⃗⃗⃗1 ⃗⃗⃗⃗⃗
𝑋𝛼 − 𝑋⃗| The best compromise solution is that having highest
⃗⃗⃗⃗⃗ value of 𝜇𝑘 .
{𝐷𝛽 = |𝐶2⃗⃗⃗⃗⃗
⃗⃗⃗⃗⃗ 𝑋𝛽 − 𝑋⃗| (35)
⃗⃗⃗⃗⃗
𝐷𝛿 = |𝐶⃗⃗⃗⃗⃗3 ⃗⃗⃗⃗⃗
𝑋𝛿 − 𝑋⃗| 4. Application of MOGWO to the proposed problem
The steps of the MOGWO algorithm are given below.
⃗⃗⃗⃗⃗ 𝑋𝛼 − ⃗⃗⃗⃗⃗
𝑋1 = ⃗⃗⃗⃗⃗ 𝐴1 ⋅ ⃗⃗⃗⃗⃗⃗
𝐷𝛼 Step 1: Read the data of the power system including
⃗⃗⃗⃗⃗2 = ⃗⃗⃗⃗⃗
{𝑋 𝑋𝛽 − 𝐴⃗⃗⃗⃗⃗2 ⋅ 𝐷
⃗⃗⃗⃗⃗ (36) line data, bus data, thermal generators data,
𝛽
wind farms data and tap setting of the
⃗⃗⃗⃗⃗ 𝑋𝛿 − ⃗⃗⃗⃗⃗
𝑋3 = ⃗⃗⃗⃗⃗ 𝐴3 ⋅ ⃗⃗⃗⃗⃗
𝐷𝛿 transformers.
Step 2: Initialize the MOGWO parameters such as
(
X (itr + 1) = X 1 + X 2 + X 3 )3 (37)
population size 𝑃𝑆 , maximum number of
iterations 𝑖𝑡𝑟𝑚𝑎𝑥 and the coefficient vectors
(𝑎⃗, 𝐴⃗ and 𝐶⃗).
Step 3: Specify the lower and upper bounds of control
variables (active power of generator buses,
106 ELECTROTEHNICĂ, ELECTRONICĂ, AUTOMATICĂ (EEA), 66 (2018), nr. 4

active power output from wind farms, voltage (28) to the objective functions 𝐹1 (𝑋), 𝐹2 (𝑋)
level of generator bus, tap position of tap- and 𝐹3 (𝑋).
changing transformers). Step 10: Select the new leader wolves 𝑋𝛼 , 𝑋𝛽 and 𝑋𝛿
Step 4: Set the iteration count t =0, initialize the from the repository.
repository (the repository is an archive of non- Step 11: Use eq. (32), eq. (33) and eq. (34) to
dominated solutions).
calculate the coefficient vectors ( A , C and
Step 5: Initialize the grey wolf population within
operating limits. Therefore, if the ith a ).
parameter of the control variables has its Step 12: Update the wolves’ position using eq. (37).
lower and upper bound as 𝑥𝑖𝐿 and 𝑥𝑖𝑈 , then the Step 13: Find the non-dominated solutions and saving
ith component of the jth population members those in the repository.
may be initialized as, Step 14: Set t=t+1;
Step 15: Repeat the procedure from step 6 to 14 until
𝑥𝑖,𝑗 (𝑡) = 𝑥𝑖𝐿 + 𝑟𝑎𝑛𝑑 × (𝑥𝑖𝑈 − 𝑥𝑖𝐿 ) (40) the maximum number of iteration 𝐼𝑡𝑟max is
reached.
where
Step 16: Choose the best compromise solution from
𝑟𝑎𝑛𝑑 is a random number between 0 and 1. the Pareto set obtained using (38) and (39).
Step 6: Perform the power flow analysis using
Newton-Raphson method for each search 5. Results and Discussion
agent (wolf) to obtain the state variables of In order solve the optimal power flow of the isolated
the power system eq. (4). Adrar power system located in the south of Algeria the
Step 7: Calculate the wind power generating cost MOGWO has proposed in this paper.
using (13), (14) and (15). The network consists of 7 thermal power plants with a
Step 8: According to the results of the power flow and size of 425.6 MW and 3 wind farms with a size of 30 MW
the previous step, the objective functions (see the Appendix).
values and are evaluated using (16), (18) and The proposed methodology is programmed in MATLAB
(20). R2017a. The experiments are performed on a desktop
Step 9: Check the inequality constraints (25), (26) and computer with Intel Core i7-8700K CPU, 3.70 GHz, 16GB
(27) considering the results of load flow RAM, running on Windows 10.
analysis. If all constraints are satisfied, skip to Figure 2 shows illustrates the single-line diagram of the
the next step, otherwise add a penalty term simulated power system.

Figure 2. One-line diagram of Adrar power system

The parameters of Weibull distribution of wind speed considered: Seven thermal generator real power outputs,
are: scale factor 𝑐 = 15, while the shape factor is 𝑘 = 1. three wind generators active power output and Seven
The wind turbines model is Gamesa G52-850, their thermal generator-bus voltage magnitudes.
characteristics are: 𝑝𝑟 = 850𝑘𝑊 , 𝑣𝑟 = 13 𝑚⁄𝑠 , 𝑣𝑖𝑛 = The objective is to determine the set of control
4 𝑚⁄𝑠, 𝑣𝑜𝑢𝑡 = 24 𝑚⁄𝑠. variables, which minimize all the objective functions
The direct cost coefficients if the wind power 𝑑 = considered.
1 $⁄𝑀 𝑊, the reserve cost coefficients 𝐾𝑟 is selected as In this paper, the MOGWO parameters considered are:
1 $⁄𝑀 𝑊 and the penalty cost coefficients 𝐾𝑝 is selected Swarm size is 200, size of search agents is 15, and maximum
as 5 $⁄𝑀 𝑊. In this paper, 15 control variables are number of generations is 500.
 ELECTROTEHNICĂ, ELECTRONICĂ, AUTOMATICĂ (EEA), 66 (2018), nr. 4 107
Four case studies are performed, and they are:
Case 1: OPF with total generation cost reduction as Figure 3 shows the resulting Pareto optimal front. In this
objective. figure, the generating solutions by employing the proposed
Case 2: OPF with active power losses reduction as multi-objective approach are diverse and well distributed.
objective.
Case 3: OPF with voltage stability improvement as
objective.
Case 4: MO-OPF with total generation cost, active
power losses reduction and voltage stability
improvement as objectives.
Table 1 includes the control settings and objective
functions evaluations for base case (under any optimization
objective) and with single/multiple objective optimization
using the proposed approach MOGWO.

Table 1. Optimal results of control variables

Control Base case Case 1 Case 2 Case 3 Case 4
Variables
𝑃𝐺1 (in MW) 5.00 22.00 18.32 13.89 14.29
𝑃𝐺2 (in MW) 4.00 13.60 12.97 6.30 12.17
𝑃𝐺3 (in MW) 10.00 34.64 32.56 29.63 26.64
𝑃𝐺4 (in MW) 10.00 49.95 50.00 50.00 49.11
𝑃𝐺5 (in MW) 25.00 66.78 82.11 73.23 82.73
Figure 3. Three-dimensional Pareto-front
𝑃𝐺6 (in MW) 10.00 35.37 19.77 33.97 17.49
𝑃𝑊1 (in MW) 10.00 10.00 9.61 2.86 9.85
𝑃𝑊2 (in MW) 10.00 10.00 1.29 0.77 2.93 Also, Figure 4 demonstrates that the system loadability
𝑃𝑊3 (in MW) 10.00 10.00 6.47 10.00 8.51 [22] has been improved from 1.55 in the base case to 1.76
𝑉𝐺1 (in p.u.) 1.0600 1.0979 1.045 1.1000 1.0592 after optimization, so the system will be able to tolerate
𝑉𝐺2 (in p.u.) 1.0600 1.0990 1.0880 1.1000 1.0329 more load demand without voltage collapse.
𝑉𝐺3 (in p.u.) 1.0600 1.0638 1.0268 1.1000 0.9811
𝑉𝐺4 (in p.u.) 1.0500 1.1000 1.1000 1.0833 1.0682
𝑉𝐺5 (in p.u.) 1.0500 1.1000 1.1000 1.1000 1.0990
𝑉𝐺6 (in p.u.) 1.0500 1.0700 1.0230 1.1000 0.9762
Wind Gen. Cost 13.79 13.80 28.94 13.81 14.29
(in $/hr)
(in $/hr)
Total Gen. Cost 8664.15 8270.84 8902.75 8825.50 8465.74
(in $/hr)
(in $/hr)
Total Loss 356.45 442.9 264.3 340.9 290.8
(in kW)
Voltage Stability 0.3449 0.3541 0.2429 0.1635 0.1827
Index

Case 1: Reduction of active power loss as objective in

the bold values in the table highlight the optimum value of
the chosen objective in the case of single objective
optimization.
Case 2: Optimizing the total generating cost causes the
system losses to increase to its maximum.
The control settings obtained in this case result in a
minimization of 4.54% in total costs but causes the system
losses to increase by 24.25% comparing with base case Figure 4. Loadability curve before and after optimization
values.
The system stability index is also degraded by 2.6674%.
OPF results in minimization of 25.85% in total active losses. 6. Conclusion
Also, the voltage stability index is reduced by 29.57%. This paper solves the multi-objective optimal power
Otherwise, the total cost increased by 2.75% over the base flow problem of Adrar power system in the presence of
case. wind power generations using MOGWO.
Case 3: Control settings based on stability improvement It gives the optimal generation schedules by optimizing
as objective in OPF yields an improvement of 52.59% in the three different objective functions, i.e., total
stability index, but results in loss reduction of only 4.36%. generation cost minimization, transmission losses
On the other hand, the total generating cost is minimization and voltage stability enhancement while
increased by 52.60%. Therefore, such optimization satisfying all the equality and inequality constraints.
objectives cannot be treated independently. Here, the stochastic nature of wind power outputs is
Case 4: With simultaneous optimization of all the modeled using the Weibull probability distribution
objectives, total generating cost is reduced by 52.50%, the function.
total active power loss is reduced by 34.34% and voltage The simulation results have demonstrated the
stability index is improved by 47.03% comparing with base efficiency and effectiveness of the proposed approach.
case values.
108 ELECTROTEHNICĂ, ELECTRONICĂ, AUTOMATICĂ (EEA), 66 (2018), nr. 4

APPENDIX References
The line data, the bus data and the generator data of [1] M. Ebeed, S. Kamel, and F. Jurado, "Chapter 7 - Optimal
Adrar power system are given in Tables 2, Table 3 and Power Flow Using Recent Optimization Techniques," in
Table 4. Classical and Recent Aspects of Power System
Optimization, A. F. Zobaa, S. H. E. Abdel Aleem, and A. Y.
Base values of 100 MVA and 220 kV were used to
Abdelaziz, Eds.: Academic Press, 2018, pp. 157-183.
calculate the parameters in per unit in Table 3. [2] S. Frank, I. Steponavice, and S. Rebennack, "Optimal
power flow: a bibliographic survey I," vol. 3, no. 3, pp.
Table 2 Test System Bus Data 221-258, 2012.
[3] S. Frank, I. Steponavice, and S. Rebennack, "Optimal
PD QD Qinj PD QD Qinj
Bus Bus name (MW) (MW (MVAr Bus Bus name (MW) (MW (MVAr
power flow: a bibliographic survey II," vol. 3, no. 3, pp.
) ) ) ) 259-289, 2012.
1(SB) ADRAR T1 0 0 0 15 ADRAR 33.6 21.0 0 [4] S. S. J. E. E. Reddy, "Optimal power flow with renewable
2 TIMIMOUNE C energy resources including storage," vol. 99, no. 2, pp.
0 0 0 16 AIN SALAH 50.4 30.8 0
PIC 685-695, 2017.
3 REGGANE PIC 0 0 0 17 KABERTANE 16.8 9.8 0 [5] C. Mishra, S. P. Singh, J. Rokadia, Transmission, and
4 AOULEF SELF Distribution, "Optimal power flow in the presence of wind
0 0 20 18 ADRAR T2 0 0 0
1
power using modified cuckoo search," vol. 9, no. 7, pp.
5 AIN SALAH
ZAOUIET K PIC 0 0 10 19 0 0 20 615-626, 2015.
SLF
6 KABERTANE [6] A. Panda, M. Tripathy, and E. Systems, "Optimal power
0 0 0 20 ADRAR T3 0 0 0 flow solution of wind integrated power system using
PIC2
7
ADRAR 220 0 0 0 21
TIMIMOUNE
0 0 0
modified bacteria foraging algorithm," vol. 54, pp. 306-
TG1 314, 2014.
8 TIMIMOUNE [7] S. Makhloufi, A. Mekhaldi, and M. J. E. Teguar, "Three
0 0 10 22 ZAOUIET KTG 0 0 0
SLF
powerful nature-inspired algorithms to optimize power
9 AIN SALAH
ADRAR POSTE 0 0 0 23 0 0 0 flow in Algerian Adrar power system," vol. 116, pp. 1117-
TG1
10 KABERTANE 1130, 2016.
ADRAR CM 63.0 39.2 0 24 0 0 0 [8] K. Teeparthi, D. Kumar, and Applications, "Security-
TG1
11 TIMIMOUNE 35.0 21.0 0 25 ADRAR EOL 0 0 0 constrained optimal power flow with wind and thermal
12 ZAOUIET KABERTANE power generators using fuzzy adaptive artificial physics
36.4 22.4 0 26 0 0 0
KOUNT EOL optimization algorithm," vol. 29, no. 3, pp. 855-871, 2018.
13 TIMIMOUNE
AOULEF 22.4 14.0 0 27 0 0 0 [9] P. P. Biswas, P. Suganthan, G. Amaratunga, and
EOL
14 REGGANE 33.6 21.0 0 Management, "Optimal power flow solutions incorporating
stochastic wind and solar power," vol. 148, pp. 1194-1207,
2017.
Table 3 Test System Line Data [10] K. Ravi, "Optimal power flow considering intermittent
From Resistance Reactance Susceptance Rating wind power using particle swarm optimization," vol. 6, no.
Line To bus 2, pp. 504-509, 2016.
bus (p.u.) (p.u.) (p.u.) (MVA)
1 4 19 0.0084 0.0865 0.0013 600 [11] M. D. Raj, N. Muthuselvan, P. Somasundaram, and
2 3 4 0.0186 0.0868 0.0013 600 Engineering, "Swarm-inspired artificial bee colony
3 5 3 0.0149 0.0694 0.0013 600 algorithm for solving optimal power flow with wind farm,"
4 2 6 0.0050 0.1041 0.0013 600 vol. 39, no. 6, pp. 4775-4787, 2014.
5 8 2 0.0003 0.0052 0.0013 600
6 6 7 0.0029 0.0607 0.0013 600
[12] M. Güçyetmez and E. Çam, "A new hybrid algorithm with
7 7 5 0.0152 0.0709 0.0013 600 genetic-teaching learning optimization (G-TLBO)
8 7 9 0 0.1000 0 80 technique for optimizing of power flow in wind-thermal
9 3 14 0 0.1000 0 40 power systems," vol. 98, no. 2, pp. 145-157, 2016.
10 19 16 0 0.0500 0 40 [13] A. J. Wood and B. F. Wollenberg, Power generation,
11 5 12 0 0.1000 0 40 operation, and control. John Wiley & Sons, 2012.
12 6 17 0 0.0500 0 40 [14] S. Reddy, "Optimal scheduling of wind-thermal power
13 7 15 0 0.0500 0 80
14 4 13 0 0.1000 0 40
system using clustered adaptive teaching learning based
15 7 10 0 0.1000 0 80 optimization," vol. 99, no. 2, pp. 535-550, 2017.
16 8 11 0 0.1000 0 40 [15] J. Hetzer, C. Y. David, and K.. Bhattarai, "An economic
17 9 18 0 0.2000 0 30 dispatch model incorporating wind power," vol. 23, no. 2,
18 15 1 0 0.1000 0 240 pp. 603-611, 2008.
19 15 20 0 0.1000 0 30 [16] H. M. Dubey, M. Pandit, and B. J. R. E. Panigrahi, "Hybrid
20 2 21 0 0.2000 0 106 flower pollination algorithm with time-varying fuzzy
21 12 22 0 0.2000 0 38
22 19 23 0 0.4000 0 212
selection mechanism for wind integrated multi-objective
23 6 24 0 0.2000 0 50 dynamic economic dispatch," vol. 83, pp. 188-202, 2015.
24 15 25 0 0.2000 0 12 [17] M. Moghavvemi, F. Omar, Transmission, and Distribution,
25 17 26 0 0.2000 0 12 "Technique for contingency monitoring and voltage
26 2 27 0 0.2000 0 12 collapse prediction," vol. 145, no. 6, pp. 634-640, 1998.
[18] J. D. Glover, M. S. Sarma, and T. Overbye, Power System
Table 4 Test System Generator Data Analysis & Design, SI Version. Cengage Learning, 2012.
[19] R. Mallipeddi, S. Jeyadevi, P. N. Suganthan, S. J. S. Baskar,
Pmin Pmax Qmin Qmax Cost coefficients and e. computation, "Efficient constraint handling for
Bus Bus name (MW) (MW) (MVAr) (MVAr) a b c optimal reactive power dispatch problems," vol. 5, pp. 28-
($/MW2) ($/MW) ($) 36, 2012.
1(SB) ADRAR T1 25 125 -45 105 0.01 31 0
[20] S. Mirjalili, S. Saremi, S. M. Mirjalili, and L. d. S. Coelho,
18 ADRAR T2 5 22 -9 22 0.01 31 0
20 ADRAR T3 4 13.6 -16 8 0.01 31 0 "Multi-objective grey wolf optimizer: A novel algorithm for
21 TIMIMOUNE 10 44 -18 42 0.01 31 0 multi-criterion optimization," Expert Systems with
TG1 Applications, vol. 47, pp. 106-119, 2016/04/01/ 2016.
22 ZAOUIET KTG 10 50 -18 42 0.01 31 0 [21] M. Mosbah, S. Arif, and R. Mohammedi, "Multi-objective
23 AIN SALAH TG1 25 125 -45 105 0.01 31 0 optimization for optimal multi DG placement and sizes in
24 KABERTANE 10 44 -16 42 0.01 31 0 distribution network based on NSGA-II and fuzzy logic
TG1
combination," in Electrical Engineering-Boumerdes (ICEE-
 ELECTROTEHNICĂ, ELECTRONICĂ, AUTOMATICĂ (EEA), 66 (2018), nr. 4 109
B), 2017 5th International Conference on, 2017, pp. 1-6: Mustafa Mosbah was born in Algeria, on
IEEE. January 15, 1982. He received the state
[22] R. Mohammedi, A. Hellal, S. Arif, and M. Mosbah, "Optimal engineering degree in electrical engineering in
DG placement and sizing in radial distribution systems 2006, the Magister degree in 2011 and the PhD
using NSGA-II for power loss minimization and voltage in 2018 on Power system from Amar Telidji
stability enhancement," vol. 8, pp. 1806-14, 2013. Laghouat University, Algeria. Since 2008,
Engineer operating in Electricity and Gas
Acknowledgments Distribution Algeria Company.
We would like to thank Sonelgaz (Algeria’s national His research interests are focused on : Distribution Power
electricity and gas company) for providing data for this Systems Analysis and Control, Renewable Energy, Smart Grids,
Artificial Intelligence applied to Distribution Power Systems
work.
Corresponding-address: [email protected]
Biography
Abdellah Kouzou (IEEE Senior member,
Redha Djamel Mohammedi was born in IACSIT Senior member, IFAC, IAENG, IISRO
Algeria, on September 21, 1982. He received member, IEEE-HKN Alumni Member) was born
the state engineering degree in electrical in Djelfa, Algeria in 1964.
engineering from Djelfa University, Algeria in He is a collaborator researcher at Texas A A&M
2006, the Magister degree in 2008 and the University at Qatar.
Magister and Ph.D degrees from Laghouat in
His main research interests include Active Power Filtering
2008 and 2016, respectively. He is an assistant
techniques, Power Quality issues, Power Electronics Devices,
professor at Djelfa University.
Application of Power electronics in Renewable Energies, Smart
His areas of interest are the Application of Meta-Heuristic Grid, reliability and diagnostics in power electronics
Optimization Algorithms in Electric Power Systems, Artificial converters.
Intelligence in Renewable Energetic Systems, FACTS modelling Corresponding-adresse: [email protected]
and integration in practical power systems, and Voltage
Stability and Security Analysis.
Corresponding-address: [email protected]