Electrical Power and Energy Systems 83 (2016) 158–174

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

journal homepage: www.elsevier.com/locate/ijepes

Ant lion optimization for short-term wind integrated hydrothermal

power generation scheduling
Hari Mohan Dubey a, Manjaree Pandit a,⇑, B.K. Panigrahi b
a
Department of Electrical Engineering, M.I.T.S., Gwalior, India
b
Department of Electrical Engineering, IIT Delhi, India

a r t i c l e i n f o a b s t r a c t

Article history: A novel nature inspired (NI) optimization algorithm, known as ant lion optimization (ALO) is used in this
Received 11 July 2015 paper for solving practical hydrothermal power generation scheduling (HTPGS) problem with wind
Received in revised form 20 March 2016 integration. The ALO algorithm mimics the unique, 6-step hunting activity of ant lions in nature which
Accepted 31 March 2016
is modelled by (i) constructing ant lion traps using roulette wheel, (ii) creating random walk of ants,
(iii) entrapment of ants in pits, (iv) adaptive shrinking of traps for sliding ant towards ant lion, (v)
catching ants and rebuilding the pits, and (vi) applying elitism. The random walk mechanism and roulette
Keywords:
wheel operation for building traps provide the ALO with a high exploration capability. The shrinking of trap
Ant lion optimization (ALO)
Ant lion trap
boundaries and elitism operations increase exploitation efficiency of the ALO, making it a very powerful
Hydrothermal power generation scheduling search technique for complex domains.
(HTPGS) The wind integrated HTPGS is a non linear, non convex and highly complex optimization problem due
Nature inspired (NI) optimization to composite operational constraints associated with hydro, thermal and wind units. To demonstrate
Random walk mechanism the applicability of the ALO algorithm for real-world problems, it is tested on four standard test systems.
Wind power uncertainty The obtained simulation results are compared with results of other algorithms reported in most recent
literature. It is found that the proposed method is proficient in producing encouraging solutions for
real-world problems.
Ó 2016 Elsevier Ltd. All rights reserved.

Introduction
Modern power system is a large complex network consisting of
electrical power generation, its transmission and distribution. With
increase in demand of electrical power and phenomenal rise in the
prices of fossil resources, economic dispatch becomes a crucial task
in power system operation whose key objective is to minimize the
fuel cost while all operating constraints are satisfied. Though the
world is slowly shifting its focus towards renewable energy
resources, majority of the power is still being generated from fossil
fuels, which are also the major sources of atmospheric pollution.
Optimal allocation and dispatch of the available hydro generation
capacity is a very attractive alternative for meeting the growing
demand for power, without increasing the fuel cost and emissions.
Therefore the integrated optimal dispatch of hydro–thermal gener-
ating units assumes a great importance in power system operation.
The idea behind short-term hydro–thermal scheduling is to
allot available water resources to hydro generators in each time
interval and to dispatch the thermal generators in such a manner
⇑ Corresponding author. Tel.: +91 0751 2665962, +91 0751 2409380.
E-mail address: manjaree_p@hotmail.com (M. Pandit).
that the fuel cost of thermal units is minimized and operational
constraints of both, thermal and hydro units are fully satisfied.
The major constraints of the HTPGS problem are water discharge
constraints between consecutive time intervals, practical limits
on reservoir storage and turbine flow rate, the changing hourly
reservoir inflows, the cascaded hydraulic network, prohibited
operational zones of hydro units, ramping limits of thermal units,
maintaining power balance with changing system load and the
capacity limits of both thermal and hydro units.
In literature the HTPGS solutions are presented using analytical
methods like dynamic programming (DP) [1], Lagrange relaxation
(LR) [2] and Newton’s method [3]. But these approaches may not
perform satisfactorily due to the nonlinearity and composite
constraints of the problem. Therefore, over the past few years,
many populations based meta-heuristic techniques are introduced
as alternatives to classical methods. These techniques begin with a
set of randomly generated solutions which are iteratively
improved using random operators. Due to the parallel search
mechanism and randomness involved, the meta-heuristic tech-
niques have a higher probability of escaping local minima and
achieving the global best solution. These methods are unaffected
by the nature of the problem because optimization is carried out

http://dx.doi.org/10.1016/j.ijepes.2016.03.057
0142-0615/Ó 2016 Elsevier Ltd. All rights reserved.
 H.M. Dubey et al. / Electrical Power and Energy Systems 83 (2016) 158–174
independently using a black box concept, by utilizing only the
inputs and associated outputs, without going for the problem
derivative.
Amongst the meta-heuristics, evolutionary programming (EP)
[4], [5] is proposed for the solution of the HTPGS problem based
on adoption and evolution theory which follows natural selection
process. Solution of HTPGS is also done using GA [6], which is
based on selection, crossover and mutation operations.
Differential Evolution (DE) and its variants [7–12], particle
swarm optimization (PSO) [13,14], teacher learner based optimiza-
tion (TLBO) [15,16], chemical reaction optimization (CRO) [17,18],
fish swarm optimization (FSO) [19], Artificial Bee Colony (ABC)
algorithm [20] and clonal selection algorithm (CSO) [21] are suc-
cessfully implemented for the HTPGS problem. Recently, covari-
ance matrix adaptation evolution strategy (CMA-ES) improved
with directed target to best perturbation (DTBP) approach has also
been used very effectively [22].
Lately many new meta-heuristic algorithms have been pro-
posed taking inspiration from natural organism and biological phe-
nomena. The main challenge in these techniques is how to
maintain balance between global (exploration) and local search
(exploitation) so that population diversity is maintained and early
convergence to a sub-optimal solution can be avoided [23].
The first generation of the nature inspired (NI) meta-heuristics,
like genetic algorithm (GA), particle swarm optimization (PSO) and
Differential Evolution (DE), experienced premature convergence
issues which were subsequently handled by enhanced versions of
these algorithms [7,8,10–12,14,20,24]. These improved variants
proposed either parameter automation [11,12,20] or hybridization
of different algorithms [25] to provide balance between explo-
ration and exploitation capabilities. Moving one step ahead, the
new NI methods provide separate operations for carrying out
exploration and exploitation within the algorithm itself. For exam-
ple, in Artificial Bee Colony (ABC) algorithm two separate phases,
employed bee phase and onlooker bee phase are used for carrying
out exploration and exploitation respectively. The ant lion opti-
mization (ALO) developed by Mirjalili [26] is also one such algo-
rithm where features for both, exploration through global search
and exploitation through local search, are embedded in the algo-
rithm. ALO is a nature inspired optimization tool which formulates
the intelligent behaviour of ant lion while hunting ants in their
constructed trap during optimization process. In the ALO, there is
a powerful inherent mechanism for improving the search effi-
ciency through extensive global search and intensive local search.
Some other newly developed NI techniques are water wave
optimization (WWO) [27], moth-flame optimization (MFO) [28],
stochastic fractal search (SFS) [29], social spider algorithm (SSA)
[30] and lightning search algorithm (LSA) [31]. The WWO [27]
which is inspired by shallow water wave model employs propaga-
tion, refraction, and breaking operations for improving exploration
and exploitation. The MFO [28] is based on how moths take a spiral
path around source of light which finally brings them to the source.
The SFS [29] makes use of the concept of creating fractals for con-
ducting a search through the problem domain. The random fractals
are generated by using any mathematical method like Levy flight,
Gaussian walks, percolation clusters or Brownian motion. The
SSA [30] is inspired by the foraging behaviour of social spiders,
which locate the prey on their web by sensing vibrations. The
LSA [31] is inspired by the lightening phenomenon in sky and
the mathematical model for search and exploration is based on
the manner in which the projectiles propagate towards ground.
Wind power is a clean and renewable energy which is gaining
acceptance for solving energy demand at low cost without produc-
ing any harmful emissions. Due to these advantages now-a-days
various researchers are focussing on issues arising due to integra-
tion of renewable energy such as wind and solar power with
159
conventional thermal units [32–42]. Due to the highly unpre-
dictable and random nature of wind power, the wind uncertainty
costs also need to be included in the optimization problem.
Suitable probability density functions (pdf) are used for wind
power characterization and for the computation of over and under
estimation penalties [32]. Wind-thermal scheduling models for
Gencos [33], for multi-area operation [34] and for mid-term
generation scheduling [35] have been recently presented. The
unpredictability of renewable generation makes it difficult to
accommodate economic considerations in real time economic dis-
patch, therefore the ‘best-fit’ participation factors are evaluated by
considering the minute-to-minute volatility of solar and wind
power and load demand [36]. The best fit day-ahead optimal
scheduling for hybrid power systems consisting of thermal, wind
and solar PV has been conducted in [37] taking into consideration
the uncertainties in load and generation to minimize real-time
costs, including the revenue from renewable energy certificates.
Some researchers have also shown the impact of wind integra-
tion on cost and emission [38,39] and reserve dispatch [40]
through multi-objective models. In [40] the energy/spinning
reserve costs and emission reduction objectives are included in
the multi-objective model. A multi-objective market clearing
model to simultaneously minimize cost as well as system risk level
is proposed in [41]. Wind uncertainty is modelled using Weibull
distribution [38–41] where as load forecasting uncertainty is
represented by a normal distribution [40,41]. In [42] a day ahead
and real time market clearing scheme is proposed to minimize
the adjustment costs and to maximizes social welfare, using a
two stage optimization approach based on GA and two-point
estimate OPF and GA and Monte Carlo simulation.
In this paper ALO algorithm is applied to solve HTPGS problem
with wind integration with complex practical operating con-
straints. To demonstrate the efficiency and applicability of ALO
for HTPGS problems its performance is compared and validated
with many different algorithms available in recent literature. To
the best of authors’ knowledge, the ALO algorithm has not yet been
tested on a complex, constrained optimization problem such as the
hydro–thermal generation scheduling problem. Wind power
uncertainty modelling further complicates the already complex
HTPGS problem.
The results demonstrate the ability of ALO to get good solutions
while satisfying a large number of practical operating constraints of
hydro, thermal and wind generation. The simplicity, simultaneous
focus on exploration and exploitation, guaranteed convergence,
ability to avoid local optima, ability to maintain population
diversity, absence of tuning parameters and high solution quality
present ALO as an attractive alternative methodology for solving
complex, real-world optimization problems.
Problem formulation of hydrothermal power generation
scheduling with wind integration
The HTPGS problem with wind integration has complex equal-
ity and inequality constraints associated with thermal, hydro and
wind generating units. Due to zero fuel cost of hydro power gener-
ating units, the prime objective becomes minimization of fossil fuel
cost of thermal units along with cost of wind power generating
units. The objective function to be minimized is
( )
X
T XNs X
Nw
FT ¼ ½f it ðPsit Þ þ ½f wkt ðPwkt Þ ð1Þ
t¼1 i¼1 k¼1
Here F T is the total operating cost of thermal and wind generators,
the total number of thermal and wind units is Ns and Nw
respectively. The interval length is T which consists of several sub
intervals. At tth interval, the scheduled power of ith thermal and
160 H.M. Dubey et al. / Electrical Power and Energy Systems 83 (2016) 158–174

kth wind power generating unit is Psit and Pwkt respectively, and fit(.), where V Hjt and Q Hjt are the storage volume and water discharge rate
fwkt(.) are the respective costs. of reservoir respectively; C1j, C2j, C3j, C4j, C5j and C6j are the coeffi-
The fuel cost of ith thermal unit at tth interval with valve point cients of the hydro power plant which express the hydro generation
loading (VPL) is represented as, in terms of water discharge and reservoir storage volume.
n  h  io
 
f it ðPsit Þ ¼ asi þ bsi Psit þ csi P2sit þ dsi  Sin esi  Pmin
si  Psit  Power capacity limits
ð2Þ
where asi , bsi , csi , dsi and esi are thermal cost coefficients. The cost of Pmin
si 6 Psit 6 Pmax
si ð9Þ
wind power (WP) is computed by taking (i) a direct cost component Pmin 6 PHjt 6 Pmax ð10Þ
Hj Hj
and (ii) a component to account for the wind uncertainty cost as the
WP is an uncertain resource. If the scheduled WP at tth interval P wkt Pmin
wk 6 Pwkt 6 Pmax
wk ð11Þ
is less than the available wind power W kt;av then penalty cost will
The optimal dispatch of generating units, at each time instant ‘t’,
be imposed for not utilizing all available WP. A penalty function
must satisfy the specified minimum and maximum operating limits
Cpk,t will be applicable to account for this situation arising due to
given by Pmin
si & P max min
si , P Hj & P max
Hj and P min min
wk & P wk for the thermal,
the underestimation of WP.
Similarly, if the available WP is less than the scheduled WP, the hydro and wind power generating units respectively.
deficit is fulfilled from thermal reserve. This situation is modelled
using a reserve cost function Crk,t to compute the cost of reserve. Ramp rate limits (RRL) of thermal power generating units
Accounting for these three factors affecting the WP cost, the cost
of kth wind turbine at tth time can be computed as [32]. Psit  Psiðt1Þ 6 URi
     ð12Þ
f wkt ðPwkt Þ ¼ ðK k  Pwkt Þ þ C pkt W kt;av  Pwkt þ C rkt Pwkt  W kt;av Psiðt1Þ  Psit 6 DRi
ð3Þ Between two consecutive time periods, the change in thermal
The under scheduling penalty cost will be proportional to the differ- power generation is limited by the up and down ramp rate limits
ence between available and scheduled wind power. It can be of generating units. For the ith thermal unit up/down ramp rate lim-
expressed as [32]: its are given by URi/DRi MW per hour respectively.

C pkt ðW kt;av  Pwkt Þ ¼ kpk  ðW kt;av  Pwkt Þ Hydraulic dynamic balance constraints
Z wr
¼ kpk  ðw  Pwkt Þf w ðwÞdw ð4Þ
Pwkt
V Hjt ¼ V Hj;t1 þ IHjt  Q Hjt  SHjt
Here kpk is the penalty cost coefficient or under estimation coeffi- Ruj 
X 
cient. The wind power characterization is done using Weibull pdf, þ Q Hm;ðtsmj Þ þ SHm;ðtsmj Þ ð13Þ
fw(w). Similarly the cost of reserve will be proportional to the short m¼1
fall (occurring due to scheduled wind power being more than avail-
IHjt represents natural inflow of jth hydro reservoir at time t, SHjt is
able wind power).
the spillage discharge rate of jth hydro generating unit at time t,
C rkt ðP wkt  W kt;av Þ ¼ krk  ðP wkt  W kt;av Þ smj is the water transport delay from reservoir m to j and Ruj is
Z Pwkt the number of upstream hydro generating plants immediately
¼ krk  ðPwkt  wÞf w ðwÞdw ð5Þ above the jth reservoir.
0

The reserve cost coefficient, also called as over estimation coeffi- Reservoir storage volume limits
cient, is denoted by krk. The total cost to be optimized can be com-
puted by substituting (2)–(5) in (1).
V min max
Hj 6 V Hjt 6 V Hj ð14Þ
Power balance constraint for demand-generation balance
where V min max
Hj and V Hj are the minimum and maximum reservoir vol-
ume of jth hydro plant respectively.
X
Ns X
NH X
Nw
Psit þ PHit þ P wkt ¼ PDt þ PLt ð6Þ
Water discharge rate limits
i¼1 j¼1 k¼1

Here NH represents the number of hydro plants, PHjt is the active

power generation from hydro plants; PDt and PLt are the power Q min max
Hj 6 Q Hjt 6 Q Hj ð15Þ
demand and transmission loss respectively at time interval t such
P where Q min max
Hj and Q Hj represents the minimum and maximum water
that t = T. Total transmission loss PLt can be calculated using
th
B-loss coefficient as below. discharge of j hydro plant.

X
NT X
NT X
NT
Prohibited operating zones (POZ) of water discharge rate
PL;t ¼ Pit Bij Pjt þ B0i Pit þ B00 ð7Þ
i¼1 j¼1 i¼1
8 min
Here total number of plants NT = NS + NH + NW and Pit is the respec- >
> Q 6 Q Hj 6 Q LB
< Hj Hj;1
tive thermal, hydro and wind power generation. The hydro power Q Hj 2 Q UB 6 Q Hj 6 Q LB ; m ¼ 2; 3; . . . ; nj ð16Þ
>
>
Hj;m1 Hj;m
generation from plant j at interval t can be expressed as: : max
Q Hjm 6 Q Hj 6 Q Hj
PHjt ¼ C 1j  V 2Hjt þ C 2j  Q 2Hjt þ C 3j  V Hjt  Q Hjt þ C 4j  V Hjt
where Q LB UB
Hjm and Q Hjm are the lower and upper bonds of the m
th
pro-
þ C 5j  Q Hjt þ C 6j ð8Þ th
hibited zone of the j hydro power generating unit respectively.
 H.M. Dubey et al. / Electrical Power and Energy Systems 83 (2016) 158–174 161

Total number of prohibited operating zones for the jth hydro gener- The hunting of ant by antlion can be understood by observing
ator are nj. their interactions. The different steps taken by the ant and ant lion
during hunting are modelled by defining six operations for the ALO
Wind power uncertainty constraints algorithm, which are explained later in this section. On the basis of
these operations the optimization model is built. To establish an
The Weibull pdf of wind speed v can be expressed as: analogy between the natural hunting phenomenon and the ALO
 k
algorithm, some guidelines are applied, which are: (i) Ants
k v k1 v
pdf ðv ; k; cÞ ¼ exp  ð17Þ take random walks around the search space. (ii) Random walks
c c c are taken by all ant dimensions. (iii) The ant lion traps influence
Here k and c are the shape and scale factor respectively. The cumu- the random walks of ants. (iv) The size of traps built by the ant
lative distribution function (cdf) of wind speed v can be expressed lions is proportional to their hunting ability/fitness. (v) Probability
as: of catching ant is higher for ant lions having larger pits. (vi) In each
iteration, there are chances of an ant being caught by either the
v
cdf ðv ; k; cÞ ¼ 1  expð c Þ
k
ð18Þ elite or some other fit ant lion. (vii) To simulate sliding of ant
towards ant lion, the range of random walk is decreased adap-
Wind power at any time can be described as a stochastic variable w tively. (viii) When an ant is pulled by an ant lion under the sand
and calculated from wind speed v as: and consumed, the ant lion becomes fitter than the ant and (ix)
8
> 0; ðv < v in or v P voÞ the ant lion takes the position of the consumed ant.
<
w ¼ PwR ðv r 6 v < v o Þ ð19Þ
Similar to ants, the ant lions are also supposed to be hidden in
>
: ðv v in ÞPwR the same search space. The matrices MAnt and MAntlion as indicated
v r v in ðv in 6 v < v r Þ below are used for the purpose of storing position of ants and ant
lions during the optimization process.
In above expression vin, vo and vr are the cut-in, cut-out and
rated wind speeds respectively and PwR is the rated power of the
2 3
wind generating unit. For the discrete region given by the first A1;1 A1;2 . . . . . . A1;d
and second condition in (19), the probability of wind power being 6 A2;1 A2;2 :Ai;j :: . . . A2;d 7
6 7
0 or PwR can be calculated as (20) and (21) respectively. M Ant ¼6
6 .. .. .. ..
7
.. 7 and
4 . . . . . 5
qfw ¼ 0g ¼ cdf ðv in Þ þ ð1  cdf ðv o ÞÞ
An;1 An;2       An;d
¼ 1  expððv in =cÞk Þ þ expððv o =cÞk Þ ð20Þ 2 3 ð23Þ
AL1;1 AL1;2 . . . . . . AL1;d
qfw ¼ PwR g ¼ cdf ðv o Þ þ ð1  cdf ðv r ÞÞ 6 AL2;1 AL2;2 . . . . . . AL2;d 7
6 7
¼ expððv r =cÞk Þ  expððv o =cÞk Þ ð21Þ M Antlion ¼6
6 .. .. .. ..
7
.. 7
4 . . . . . 5
The aggregate sum of probability of all variables will be one at any ALn;1 ALn;2       ALn;d
instant of time. For the continuous region denoted by the third con-
dition in (19), for wind speed between vin and vr, the wind power w
is a continuous variable and its pdf can be written as: where Aij denotes the value of jth dimension of ith ant, n is the num-
ber of ants and d denotes the ant dimension, i.e. number of problem

k1
khv in ð1 þ hw=PwR Þv in variables. Similarly, ALij denotes the jth dimension (variable) of ith
f w ðwÞ ¼
PwR c c ant lion. Each row (ant) represents a d-dimensional solution (deci-
( " #) sion variable) of the optimization problem.
ð1 þ hw=P wR Þv in
k
 exp  ð22Þ The ants and ant lions are evaluated and the evaluation func-
c tions are stored in matrices MOA and MOAL respectively.
where h = (vr/vin)  1. 2 3 2 3
f ð½A1;1 ; A1;2 ;   ; A1;d Þ f ð½AL1;1 ; AL1;2 ;   ; AL1;d Þ
6 7 6 7
Ant lion optimization 6 f ð½A2;1 ; A2;2 ;   ; A2;d Þ 7 6 f ð½AL2;1 ; AL2;2 ;   ; AL2;d Þ 7
6 7 6 7
6 .. 7 6 .. 7
M OA ¼ 6
6 . 7 and M OAL ¼ 6
7 6 . 7
7
The ALO algorithm [26] is inspired by the hunting of ants and 6 . 7 6 .. 7
6 .
. 7 6 . 7
other prey by the ant lions found in nature. It is a population based 4 5 4 5
random search algorithm. f ð½An;1 ;An;2 ;   ; An;d Þ f ð½ALn;1 ; ALn;2 ;   ; ALn;d Þ
Ant lions, also known as doodlebugs, look like small dragon ð24Þ
flies. Their lifespan is around three years, out of which most of
the lifecycle occurs in larvae phase and only 3–5 weeks are spent
in adult phase. During larvae stage they catch and eat prey, mainly The ALO can be explained with six operators, discussed below,
ants, by their unique hunting behaviour. The ant lions dig a conical which are designed to mimic the different steps of the actual hunt-
hole in the sand, hide at the bottom of the cone and wait for their ing behaviour of ant lion as observed in nature.
prey to be trapped in the hole. The elite ant lion has the best trap
and hence it has the highest probability of catching prey. The ant
falls in the sharp edged hole and ant lion throws sand towards it Constructing ant lion traps using roulette wheel
so that the ant/pray is not able to escape. Then the ant is consumed
and the pit is rebuilt for catching other ants. In the process, the For each ant an ant lion is selected because ants are assumed to
quality of trap and the fitness of the ant lion keep improving. be trapped around only one selected ant lion. The selection is based
In the ALO algorithm, ants are search agents which wander over on its evaluation/hunting ability. A roulette wheel operator is used
the search space and ant lions dig pits in the ground to trap and for selecting an ant lion based on fitness. This selection offers
consume the ants. The objective function to be optimized is higher chances to the fitter ant lions for catching ants because a fit-
modelled to reflect the hunting ability of the ant lion. ter ant lion is assumed to build a better trap.
162 H.M. Dubey et al. / Electrical Power and Energy Systems 83 (2016) 158–174

Creating random walk of ants where U is a constant which varies between 1 and 6 as shown
below. With each time step, R increases, resulting in reduction of
The ALO algorithm imitates the actual interaction between the the boundary of random walk. Thus the exploitation level slowly
ant lions and their prey in the pits. The interactions are modelled increases with increase in iteration count.
through movement of ants over the search space and their capture 8
>
> 2 if iter > 0:1  itermax
by ant lions using traps. In nature ants move randomly in search of >
>
>
> 3 if iter > 0:5  itermax
food, therefore the ants’ movement is modelled based on random >
>
>
>
walk. The ant position in the ith dimension of a d-dimensional < if iter > 0:75  itermax
> 4
space, Xi is modelled using random walk as follows [26]: U¼ 5 if iter > 0:9  itermax ð33Þ
>
>
X i ¼ ½0; cumsumð2  f  1Þ ð25Þ > 6 if iter > 0:95  itermax
>
>
>
>
>
>
> else
where cumsum denotes cumulative sum, maxiter denotes the max- >
:
imum iteration count and f is a function which randomly returns 1
either 0 or 1, as defined below
Here f(.) is a random function defined as Catching ants and rebuilding the pits

1 if randðmaxiter; 1Þ > 0:5
f ¼ ð26Þ In the final stage of hunt the ant reaches the bottom of the pit
0 if randðmaxiter; 1Þ 6 0:5
and it is caught by the ant lion. Then the ant lion drags the ant
Here rand represents a uniformly distributed random number inside the sand and consumes it. To model this process, it is
between 0 and 1. During optimization, at each step positions of ants assumed that consumption of ant by ant lion is synonymous to
are updated using random walk. To limit the random walk within ant becoming fitter (as it goes inside sand) than its corresponding
the problem domain, the random values generated in (25) are nor- ant lion. Then the ant lion updates its position by taking the posi-
malized to lie between the minimum and maximum limits of the tion of the consumed ant, which is synonymous to ant lion rebuild-
each dimension. The upper/lower limits for ith dimension at current ing its pit to improve the chances of catching new prey. If the ith ant
iteration are represented by citer
i and diter
i respectively. The normal- is consumed by the jth ant lion, at a iteration represented by ‘iter’
ized ant position in ith dimension is represented as this phenomenon can be understood as Antlioniter j = Antiter
i . It can
be modelled as
ðX iter  ai Þðditer  citer
i Þ
X iter ¼ i i
þ citer ð27Þ
i
ðbi  ai Þ i ½ALj1; ALj2 ; . . . ; ALjd  ¼ ½Ai1 ; Ai2 ; . . . ; Aid  if f ðAntiter
i Þ
iter
where ai and bi are the minimum and maximum random walk of ith > f ðAntlionj Þ ð34Þ
variable. Similarly random walks are generated for each dimension
of ant movement.
Applying elitism
Entrapment of ants in pits
To preserve the best solution obtained at each stage, the posi-
tion of the best (fittest) ant lion is saved as elite. Being the best,
As previously discussed, ant lion traps, affect the random walks
the elite ant lion is considered to influence the movement of each
of ants. To model this behaviour, the boundary of ant movement is
ant. Therefore, each ant is assumed to take a random walk
adjusted in each step (iteration) such that the ant moves in a hyper
region around the selected ant lion trap. The lower and upper (RW iter
ant lion ) around the ant lion selected using Roulette wheel and
bounds of the ant dimension are computed at each step as [26] the elite ant lion (RW iter
elite ). The position of i
th
ant at any time
specified by iter, is calculated using these two random walks, as
citer ¼ Antlioniter
j þ citer1 ð28Þ given by
iter iter iter1
d ¼ Antlionj þd ð29Þ
Antti ¼ ðRW iter iter
ant lion þ Relite Þ=2 ð35Þ
iter
Vectors citer , d represent the lower and upper boundary values of
iter
all ant dimensions at current iteration ‘iter’ and vector Antlionj Implementation of wind integrated HTPGS problem using ALO
denotes the position of the jth selected ant lion at current iteration.
Thus to trap the ant in ant lion pit, the ALO restricts ant movement Main steps for implementing ALO algorithm for the solution of
 
within a d-dimensional hyperspace, the boundaries citer ; diter of wind integrated HTPGS problem includes initialization, constraints
handling, evaluation of fitness and updating of ant position, which
which are decided by the position of ant lion pit as given by (28)
is presented in Fig. 1 and described below.
and (29).

Initialisation
Adaptive shrinking of traps for sliding ant towards the ant lion

Cost and other coefficients of hydro, thermal and wind power

When an ant takes a random walk and falls in an ant lion trap,
units, water discharge, minimum/maximum limits of reservoir
the ant lion throws sand towards the ant. Due to this the trapped
storage volume, minimum/maximum power limits of hydro,
ant is not able to escape from the trap and it slides down towards
thermal and wind generating units, hourly power demand,
the bottom of the pit. To model this behaviour the lower and upper
initial and final reservoir volumes and maximum iteration count
limits of ant’s random walk are reduced iteratively as shown below
is taken from a standard system data set. Number of search agents
[26]:
and maximum iteration count is selected to suit problem
citer ¼ citer =R ð30Þ requirement.
iter iter
d ¼d =R ð31Þ The water discharge, thermal power and wind power are
U initialised randomly between the specified minimum/maximum
where R ¼ 10  iter=maxiter ð32Þ
limits as:
 H.M. Dubey et al. / Electrical Power and Energy Systems 83 (2016) 158–174 163

Ant Lion Opmizaon algorithm (ALO)

Initial ant Population

While iter < itermax

Population Evaluation for minimizing (1) subject
to constraints (6), (9)-(22)
Thermal
Schedule

Construcon of Traps

Hydro
Random walk of ant
Schedule

Entrapment of ant in ant lion pits

Wind
Adapve shrinking of trap Schedule

Catching ant and rebuilding trap

Updated Ant lion populaon

Fig. 1. Schematic diagram for wind integrated hydrothermal power scheduling using ant lion optimization.

 
Q Hjt ¼ Q min max min Evaluate the ant lion population
Hj þ r H Q Hj  Q Hj ð36Þ
 
PSit ¼ P min
si þ r S Pmax
si  P min
si ð37Þ The merit of a solution string is computed by calculating its
  evaluation function which is defined for wind integrated HTPGS,
Pwkt ¼ Pmin max
wk þ r w P swk  P wk
min
ð38Þ as:
2 !2
Here rH , rS and rw are uniformly distributed random numbers X
T X
Ns X
Nw
between [0, 1]. The reservoir volume is evaluated using (13) which EðF T Þ ¼ 4F T þ n  Psit þ Pwt  PD ðtÞ  PL ðtÞ
must satisfy (14). The active power generation PHjt of each hydro t¼1 i¼1 k¼1
#
power unit for every interval of time is computed using (8) which X
NS X
NH X
Nw

must satisfy (10). Similarly wind power is also initialized randomly. þw ðCV i;Thermal Þ2 þ  ðCV j;Hydro Þ2 þ f  ðCV K;wind Þ2
i¼1 j¼1 K¼1
After subtracting wind and hydro power from the power demand,
 
the remaining active power demand P sD ðtÞ is evaluated which ð41Þ
would be fulfilled by thermal power generating units. where the terms CV, CVj and CVk represent constraint violations of ith
X
NH X
Nw thermal, jth hydro and kth wind power generating units respectively.
PsDt ¼ P Dt  PHjt þ Pwkt ð39Þ The evaluation function converts the constrained optimization
j1 K¼1 problem into an unconstrained one. If there is no constraint viola-
tion the CV term will be equal to zero. Otherwise a penalty propor-
The decision variable (search agents/ant lions) over entire time tional to the constraint violation is imposed to increase the
frame can be expressed as evaluation function (cost). The penalty coefficients n, w and f are
2 3 selected such that an infeasible solution gets higher cost as com-
Q H11 Q H21   Q HNH 1 P S11 P S21   P SNs1    P w11 P w21    P wNw1
6 7 pared to a feasible solution.
6 Q H12 Q H22   Q HNH 2 P S21 P S22   P SNs2    P w12 P w22    P wNw2 7
6 7
6 . .. .. .. .. .. .. .. .. 7
6 .. .   . . .   .    . .    . 7 Sort ant lions in merit order and initialize best ant lion
4 5
Q H1T Q H2T   Q HNH T P S1T P S2T   P SNsT    P w1T P w2T    P wNwT
After fitness evaluation all ant lions are sorted in descending
ð40Þ
order of their evaluation function. For a minimization problem, this
164 H.M. Dubey et al. / Electrical Power and Energy Systems 83 (2016) 158–174

means that the first ant lion is the fittest and it is termed as the 30
QH1
elite ant lion (local best solution) amongst the whole ant lions
QH2
(solutions) in the population.
QH3
25
Iterative steps of ALO algorithm QH4

Selection of ant lion 20

Dimension
For each ant an ant lion is selected using Roulette wheel selec-
tion mechanism such that the probability of selection of an ant lion
is proportional to the fitness of that ant lion. 15

Adjusting the movement of ant

Adjustment of ant movement during iteration can be controlled 10
by the two sub step as follows.

(i) Trapping ant in ant lion pits: The lower and upper limit of 5
0 20 40 60 80 100 120
each ant dimension is modified dynamically around the
location of the selected ant lion pit as explained in (28) Iteration
and (29).
Fig. 3. Random walks of four hydro power variables (Test Case I).
(ii) Sliding ant towards ant lion: Boundaries of ant movement are
decreased adaptively as shown in (30) and (31) to converge
ant location towards ant lion pit. assumed to be consumed by the ant lion, and the ant lion builds
a new trap by taking the position of the consumed ant.
Ant position update through random walk and elitism

(i) For each ant dimension random walks are generated around
the (a) selected ant lion and (b) the elite ant lion by applying
(25).
(ii) The ant position is update using the above two random
walks as shown in (35).
The random walks for the different ant dimensions of the
HTPGS problem in 100 iterations are shown in Figs. 2 and 3 respec-
tively. The random walks can be seen to follow the upper and
lower operating boundaries.
Stopping criterion
The steps from Section ‘Selection of ant lion’, ‘Adjusting the
movement of ant’, ‘Ant position update through random walk
and elitism’, ‘Implementation of binding constraints’, and ‘Re
building the ant lion pits’ are repeated iteratively till the stopping
criterion, specified by maximum iteration count, is reached.
Simulation results and discussion

The ALO algorithm is tested on four different HTPGS test cases

Implementation of binding constraints
to inspect its efficiency to solve constrained real word problems.
If the updated ant position is found to wander beyond specified
Each test case has scheduled load pattern of one day with 24 inter-
upper and lower limit/boundaries, they are forced back inside the
vals of 1 h each. The ALO algorithm is a gradient free optimization
feasible region.
technique with no tuning coefficients and only two variable
parameters, which are number of search agents (population) and
Re building the ant lion pits
maximum iteration count.
The updated ant positions are evaluated using (41); if the new
ant position is found to be superior to the ant lion position then
the position of the ant lion is updated using (34). The ant is Effect of search agents size

Effect of number of search agents on performance was analyzed

500
Ps1
for all four test cases. The algorithm was run for 25 trials for search
450 agent sizes of 10, 30, 50 and 100. It was observed that 30 numbers
Ps2
of agents produced the best results. In Table 1 the results for Test
400 Ps3
Case II are reported. With search agent size of 10 the ALO produces
350 inferior results, while for search agent size more than 30 i.e. 50 and
100, a higher CPU time was required while there was no significant
300
Dimension

improvement in operating cost. The maximum iteration count was

250 set at 1500 for all systems.

200
Table 1
150
Effect of number of search agents.
100 No. of search Min cost Max cost Mean cost CPU time
agents (NP) ($) ($) ($) (sec)
50
10 42867.8383 42954.8802 42886.7107 15.38
0 30 42833.9085 42900.1924 42867.3120 55.63
0 20 40 60 80 100 120
50 42833.9101 42900.1995 42867.97196 105.23
Iteration 100 42833.9086 42919.3024 42871.4310 146.75

Fig. 2. Random walks of the three thermal power variables (Test Case 1). The best results are shown in bold.
 H.M. Dubey et al. / Electrical Power and Energy Systems 83 (2016) 158–174 165

Test Case I variables fulfil their inequality operating constraints like RRL
related to thermal unit (12) and POZ of hydro unit (16) including
This system consists of four hydro and three thermal power power balance constraints (6).
generating units. The hydro plant generation coefficients, reservoir The status of reservoir volume and water discharge over the
volume limits, reservoirs inflows and minimum/maximum limits entire scheduling horizon is presented in Figs. 8 and 9 respectively,
along with fuel cost coefficients of thermal units are adopted from which satisfies the reservoir volume constraints given by (13) and
[20]. The thermal cost characteristic is non-convex due to valve (14) and water discharge constraints given by (15) and (16) respec-
point loading effects as shown in (2). However the transmission tively. Fig. 10 shows the computed values of optimal hydro and
losses are neglected here. thermal power generation schedule against the hourly power
In Table 2, the optimal results obtained by ALO are compared demand.
with recently reported meta-heuristic methods like modified chao-
tic differential evolution (MCDE) [12], disruption based gravita- Test Case III
tional search algorithm (DGSA) [43], oppositional real coded
chemical reaction based optimization (ORCCRO) [18], hybrid real It is a comparatively larger HTPGS problem with multi chain
coded genetic algorithm and artificial fish swarm algorithm cascade of four reservoir plants and ten thermal power generating
(RCGA_AFSA) [19] and many others. The ALO algorithm can be seen units. The cost curve with VPL constraints is considered for thermal
to produce the best results in terms of minimum, maximum as units. The complete data for this system is adopted from [9]. For
well as mean cost, quite fast. Minimum cost obtained by ALO is the sake of comparison with [9], constraints like RRL of thermal
$40780.0523 for supplying the given load pattern of one day. power generating units, POZ of hydro power generating units
Table 3 presents the complete hourly results of the optimal and transmission loss are not considered in this case.
variables for the given hourly power demand. The optimal water The optimal hydro and thermal power discharge for the hourly
discharge, hydro and thermal power generation are listed. Here it forecast load demand computed for this case using ALO is listed in
is found that all the variables fulfil their operating constraints. Tables 7 and 8 respectively. Fig. 11 depicts the pattern of the
The hourly reservoir volume and water discharge corresponding cascade reservoir storage volume, while Fig. 12 shows the water
to the best cost solution are plotted in Figs. 4 and 5 respectively discharge over the entire scheduled time horizon for 24 h. The
for the four hydro power plants. The optimal hourly hydro and minimum cost, maximum cost and mean cost is computed out of
thermal power generation solution of HTPGS corresponding to 25 trials and compared with MDE [13], SPPSO [13], ORCCRO [18],
hourly power demands are plotted in Fig. 6. The ALO has stable RCCRO [18], IDE [5] and most recently reported MCDE [12] in
convergence characteristic as shown in Fig. 7. Table 6. The optimal cost attained by ALO for this larger test case
is $161353.9606, which is found to be superior to all other
Test Case II reported methods. Optimal hourly generation scheduling to fulfil
the corresponding load demand is presented with the help of
This test case also consists of four hydro and three thermal Fig. 13.
power generating units but this system is more complex than the
first test case due to the practical constraints of thermal and hydro Test Case IV (wind integrated HTPG system)
generating units and inclusion of transmission loss. The thermal
units have VPL effects as given by (2), ramp rate limits (RRL) given The data for this system is similar to Test Case III except the last
by (12) and Prohibited operating zone (POZ) constraints for hydro two thermal power generating units are replaced by wind power
power generating units given by (16). The transmission losses are generators. The wind power availability is modelled as probabilis-
also included here with the help of loss coefficient matrix [7]. The tic constraint in power balance equation. The Weibull shape factor
data for this system is taken from [10]. and scale parameter constants for the two wind power generators
For this complex test case the optimal cost solution obtained by considered for analysis are k1 = 1.5; k2 = 1.5 and c1 = 15; c2 = 15
ALO is $42833.9085 and total transmission loss is found to be respectively. The reserve and penalty cost coefficients for the above
350.2342 MW which is found to be better than improved differen- two wind power generating units are taken as kr1 = 5, kr2 = 5,
tial evolution (IDE) [10] as shown in Table 4. kp1 = 5, kp2 = 5 respectively. The rating of wind generator is
The optimal hourly water discharge and power generation PwR1 = PwR2 = 175 MW. The cut in, cut out and rated wind speeds
sharing between hydro/thermal power units for the hourly power are vin = 5, vo = 45 and vr = 15 respectively.
demand is presented in Table 5. Here it is found that all the For the Test Case IV, no previous results were available,
therefore the performance of ALO is validated using Differential
Table 2 Evolution (DE) [44], Artificial Bee Colony (ABC) [45] and Flower
Comparison of results (Test Case I).
Pollination Algorithm (FPA) [46] and ALO is found to be better in
Method Min cost Max cost Mean cost CPU time terms of solution quality as well as speed.
($) ($) ($) (sec) The optimal water discharge and hydro power generation com-
ALO 40780.0523 41094.3414 40905.8259 15.01 puted by ALO is listed in Table 9. The thermal power generation
GSA [30] 42032.35 42561.53 42292.12 32.29 and wind power generation sharing for this test case is listed in
DGSA [30] 41751.15 41989.02 41821.49 31.99
Table 10. The optimal dispatch results in Tables 9 and 10 can be
CSA [21] 422440.57 NA NA 109
ACABC [20] 41074.42 NA NA 16 seen to satisfy all the complex operating constraints associated
RCGA_AFSA [19] 40913.82 41362.57 41235.72 21 with hydro, thermal and wind power generating units.
ORCCRO [18] 40936.6526 41127.6819 40944.2938 10.48 Fig. 14 shows the pattern of the cascade reservoir storage vol-
RCCRO [17] 41497.8517 41502.3669 41498.2129 15.51
ume, Fig. 15 depicts the water discharge over the entire scheduled
QTLBO [16] 42187.49 42202.75 42193.46 6.83
TLBO [15] 42385.88 42441.36 424407.23 NA
time horizon for 24 h. The optimal cost attained by ALO for this
MCDE [12] 40945.75 41977.04 41380.54 50.8 wind integrated HTPGS problem is 151329.0340($) whereas wind
PSO [9] 44740 NA NA 232.73 over estimation and under estimation costs are found to be
MHDE [8] 41856.50 NA NA 31 695.9574($) and 23.0162($) respectively. Power generation by
MDE [7] 42611.14 NA NA 125
wind integrated HTPGS is plotted in Fig. 16 along with the given
The best results are shown in bold. hourly load demand. Comparison of Test Case III and Test Case IV
166 H.M. Dubey et al. / Electrical Power and Energy Systems 83 (2016) 158–174

Table 3
Optimal water discharge, hydro and thermal power generation and thermal power generation for Test Case I.

Hour Water discharge (104 m3) Hydro power generation (MW) Thermal power generation (MW)
Q1 Q2 Q3 Q4 PH1 PH2 PH3 PH4 PT1 PT2 PT3
1 10.6625 6.4716 20.0912 13.0013 88.6096 53.1849 40.8069 200.1038 102.6382 124.9024 139.7542
2 8.1430 8.1909 29.4714 13.0012 76.0685 63.9413 0 187.7626 102.6572 209.8141 139.7563
3 11.6110 8.9167 29.8410 13.0013 90.8765 68.0488 0 173.7393 102.6732 124.9078 139.7545
4 7.6824 6.3939 29.8948 13.0003 71.8219 54.0849 0 156.7886 102.6523 124.9017 139.7506
5 6.3034 7.1938 22.4277 13.0013 62.0784 59.8931 7.4379 166.1622 20.0017 124.9079 229.5188
6 6.4913 9.3541 19.0653 13.0012 63.6338 71.0019 26.4008 186.7542 102.6352 209.8158 139.7584
7 5.1864 6.1235 21.3664 13.0013 53.8880 51.2442 11.8290 206.1271 102.6727 294.7197 229.5193
8 7.4998 6.6559 17.9320 13.0017 71.9927 55.1180 27.3479 223.8548 102.5925 209.8148 319.2793
9 5.1936 6.0010 15.9629 13.0000 55.2011 51.6910 33.5325 232.9853 102.6108 294.7234 319.2559
10 11.7136 8.5782 11.7877 14.3026 94.4204 68.4253 40.4653 249.7593 102.6745 294.7355 229.5196
11 5.0009 6.4499 17.5616 13.0007 54.3692 56.4756 27.6387 244.8410 102.6734 294.7229 319.2793
12 8.7534 7.3628 14.6209 13.0458 83.0452 62.9140 37.9139 249.4484 102.6735 294.7256 319.2794
13 9.8538 9.6810 13.8555 15.4878 89.4512 75.1702 40.6133 272.9818 102.6779 209.8259 319.2797
14 9.9877 10.7362 15.2407 16.7633 90.6036 79.0347 39.4665 278.8825 102.6748 209.8162 229.5216
15 9.7986 9.5836 15.4675 14.6865 89.8850 73.2893 40.8854 263.9429 102.6622 209.8158 229.5195
16 6.3211 9.5302 16.3792 13.4844 66.7677 72.0802 40.7818 253.4560 102.6733 294.7217 229.5193
17 7.6027 7.7623 18.7221 13.0564 76.8836 61.8061 34.5780 249.8155 102.6735 294.7236 229.5196
18 9.3442 10.3761 16.4737 17.1830 88.0728 72.7545 42.5371 284.8669 102.6735 209.8159 319.2794
19 7.7370 7.5275 14.3435 14.1542 77.6156 57.3411 48.0082 260.1181 102.6736 294.7239 229.5195
20 9.4522 11.2414 13.1029 19.8131 87.9124 73.7229 50.6628 300.5445 102.6736 294.7238 139.7599
21 6.3427 7.2041 10.0031 17.3012 66.6334 54.4032 51.6251 285.0890 102.6735 209.8159 139.7598
22 9.4513 8.2412 10.0591 17.8789 87.7431 61.1245 53.5799 287.9462 20.0004 209.8250 139.7809
23 6.0550 9.3672 10.0048 20.3020 64.3761 66.2419 55.0414 296.9991 102.6736 124.9080 139.7598
24 8.8124 13.0569 10.0923 22.9385 84.7521 76.8055 56.2310 297.5323 20.0101 124.9091 139.7598

170
Plant 1
Reservior storage volume ( X105 m 3 )

160 Plant 2
Plant 3
150 Plant 4

140

130

120

110

100

90

80

70
0 5 10 15 20 25
Time (Hour )

Fig. 4. Optimal hourly hydro reservoir storage volume for Test Case I computed using ALO.

30
Plant 1
Water discharge (x 10 m )
3

Plant 2
25 Plant 3
4

Plant 4

20

15

10

5
2 4 6 8 10 12 14 16 18 20 22 24
Time (Hour)

Fig. 5. Optimal hourly water discharge for hydro plants of Test Case I computed using ALO.

reveals that the total operating cost is reduced by $10024.92659 HTPGS problem are presented in Tables 2, 4, 6 and 11 respec-
per day for the given daily load profile, due to wind integration. tively. For the Test Case 1 minimum fuel cost obtained by ALO
($40780.0523) is the best out of the results reported so far.
Comparison of ALO with established methods Though the minimum cost reported by MDNLPSO [14] is lesser
at $40,179, it is infeasible because the given water discharge
The optimal fuel costs obtained by ALO algorithm for three constraint given by (15) is not satisfied for some time intervals
different cases of HTPGS problem and for a wind integrated in [14].
 H.M. Dubey et al. / Electrical Power and Energy Systems 83 (2016) 158–174 167

Hydro Power Generation Thermal Power Generation Power Demand

1400

Power Generation (MW)

1200
1000
800
600
400
200
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
Time (Hour)

Fig. 6. Optimal hourly hydro and thermal power generation with demand for Test Case I computed using ALO.

x 10 4
5.6
4 HT & 3 Th with VPL
5.4
Power generation cost ($)

5.2

4.8

4.6

4.4

4.2

4
0 500 1000 1500
Iteration

Fig. 7. The ALO convergence characteristic for Test Case I.

Table 4
Comparison of results (Test Case II).

Method Min cost ($) Max cost ($) Mean cost ($) CPU time (sec)
ALO 42833.9085 42900.1924 42867.3120 55.63
IDE [10] 43790.33 43812.01 43800.51 782.23

The best results are shown in bold.

Table 5
Optimal water discharge, hydro and thermal power generation for Test Case II.

Hour Water discharge ( 104 m3) Hydro power generation (MW) Thermal power generation (MW)
Q1 Q2 Q3 Q4 PH1 PH2 PH3 PH4 PT1 PT2 PT3
1 11.1132 6.9129 20.4236 13.0983 90.1571 55.8831 39.2203 200.8544 104.4566 126.1527 139.8107
2 6.8967 6.0002 21.8576 13.0304 67.9402 50.7846 26.9851 187.8629 102.4586 124.8676 229.5195
3 7.4165 6.4231 20.6841 13.0043 71.7468 55.1996 30.9822 173.6008 22.4644 124.9082 229.5196
4 11.2055 10.8697 17.7985 13.0540 89.8632 78.9188 42.1172 156.9692 102.4643 45.5688 139.7607
5 10.3902 6.6192 29.9979 13.1975 84.9137 56.2754 0 167.6884 102.6769 124.9435 139.8489
6 12.7670 10.1268 29.9950 14.5756 89.1599 74.5095 0 187.7790 103.7004 126.0326 229.5196
7 10.7645 8.9983 29.9844 13.1670 82.2992 66.9840 0 186.0913 102.6760 209.8172 320.3686
8 6.0001 8.0005 19.4942 13.0009 57.2060 60.7783 24.6910 190.4217 169.5775 209.8158 319.2794
9 7.1062 8.3662 14.1452 14.3884 66.1405 62.6497 43.9192 220.1765 102.6735 294.7238 319.2794
10 6.5714 6.3502 15.7954 13.0755 63.8311 51.6473 40.9884 226.4210 102.6735 294.7239 319.2794
11 6.3188 6.0000 12.0616 13.0363 63.4864 50.9220 46.8013 241.9340 102.4194 294.7236 319.2793
12 12.9142 11.9865 13.2679 21.2287 94.7103 80.3067 47.5791 301.0575 115.6695 210.4789 319.2794
13 5.7138 6.3247 14.9529 14.7207 59.1870 51.8773 46.1660 255.7164 102.6734 294.7231 319.2766
14 11.5242 6.9697 29.4933 15.9335 93.3279 57.4106 0 265.9015 102.6735 209.8159 319.2794
15 5.1062 8.7404 19.1466 13.3565 55.0345 68.2500 31.3309 241.7091 102.6735 209.7763 319.2794
16 6.9444 10.0986 15.0858 15.1600 70.6792 73.8449 45.4914 256.3423 102.6736 209.8158 319.2794
17 6.4080 9.6737 16.1389 15.2571 66.8759 70.0836 42.5175 256.8685 102.6735 209.8159 319.2794
18 6.3046 8.7572 12.1465 14.4667 66.2368 63.5613 49.1937 263.7516 170.0000 209.8158 319.2793
19 11.7091 12.9582 10.7140 23.0370 96.9112 76.3980 50.3760 316.7874 102.6736 209.8161 229.5196
20 7.4231 6.7784 13.8433 14.8467 73.8435 48.6364 51.4789 264.0038 100.9615 209.8158 319.2793
21 7.9858 6.9963 11.2611 15.9970 77.4793 51.3778 54.3573 274.0977 23.1578 209.8159 229.5196
22 6.2386 10.1210 11.3079 23.4581 64.9933 67.6424 56.6792 305.6048 20.0000 124.9081 229.5196
23 5.1579 6.3560 10.3658 14.4335 56.1569 47.4976 56.3794 246.2849 99.9996 124.8039 229.3683
24 5.0200 11.5722 11.1114 15.3030 55.2078 72.1120 57.7789 252.2910 102.6837 40.4686 229.5196
168 H.M. Dubey et al. / Electrical Power and Energy Systems 83 (2016) 158–174

180
Plant 1

Reservoir storage volume (X 10 5 m3 )

Plant 2
160 Plant 3
Plant 4

140

120

100

80

60
0 5 10 15 20 25
Time (Hour)

Fig. 8. Optimal hourly hydro reservoir storage volume for Test Case II computed using ALO.

30
Plant 1
Plant 2
Water discharge ( X 10 4 m 3 )

Plant 3
25 Plant 4

20

15

10

5
2 4 6 8 10 12 14 16 18 20 22 24
Time ( Hour)

Fig. 9. Optimal hourly water discharge pattern for Test Case II computed using ALO.

Hydro power generation Thermal power generation Power demand

1400
Power generation (MW)

1200

1000

800

600

400

200

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
Time (Hour)

Fig. 10. Optimal hourly hydro and thermal power generation with demand for Test Case II computed using ALO.

Table 6
Comparison of results for Test Case III.

Method Min cost ($) Max cost ($) Mean cost ($) CPU time (sec)
ALO 161353.9606 161363.0615 161356.1740 20.67
ORCCRO [18] 163066.0337 163134.5391 163068.7739 15.74
RCCRO [18] 164138.6517 164182.3520 164140.3997 22.02
SPPSO [13] 167710.56 170879.30 168688.92 24.8
SPSO [13] 189350.63 191844.28 190560.31 108.1
MDE [13] 177338.60 182172.01 179676.35 86.5
DE [13] 170964.15 NA NA 96.4
MCDE [12] 165331.7 167060.6 166116.4 178.5
IDE [10] 170576.5 170608.3 170589.6 727.06

The best results are shown in bold.

 H.M. Dubey et al. / Electrical Power and Energy Systems 83 (2016) 158–174 169

Table 7
Optimal water discharge and hydro power generation for Test Case III.

Hour Hydro discharge (104 m3) Hydro power generation (MW)

Q1 Q2 Q3 Q4 PH1 PH2 PH3 PH4
1 9.6025 8.6024 30.0000 13.0094 84.2428 65.0649 0 200.1667
2 5.2413 7.2247 29.9997 13.0000 55.4293 57.3505 0 187.7440
3 6.0827 6.7290 29.9978 13.0127 62.8532 55.4612 0 173.8108
4 9.9563 6.0980 17.5730 13.0054 86.8755 52.7405 31.3955 156.7977
5 7.8210 9.4444 15.1840 13.0289 74.3573 72.3386 38.5806 178.9151
6 10.0104 7.6688 29.9845 13.1235 85.4178 61.9594 0 199.8443
7 9.2659 6.5544 16.6040 13.1625 81.3523 54.4178 31.0052 218.5906
8 10.7780 10.0532 15.5814 13.2718 87.4055 72.6046 36.3896 224.0373
9 7.3411 7.8728 17.7416 13.7460 70.1930 61.0913 29.9753 229.8444
10 9.6088 10.1153 15.4131 13.5163 83.8875 72.2422 38.0890 243.7035
11 9.1727 7.8978 16.9023 13.0000 82.5984 61.2323 34.7694 241.8836
12 7.6453 8.0923 16.2850 15.9540 74.0502 62.3221 37.9000 268.4307
13 8.5916 10.5158 15.4728 15.0654 80.6762 73.0245 42.9177 263.4127
14 7.8414 7.7486 17.0441 15.5754 76.8612 59.4906 39.5054 267.6706
15 6.0777 10.5912 18.7319 13.6024 64.2941 73.1185 33.9832 252.6162
16 8.5601 8.8268 13.4566 14.8029 82.7470 64.0558 49.0089 265.2579
17 8.4077 8.5644 16.6713 13.7894 81.8692 61.5896 43.2284 257.1679
18 8.3588 8.1841 12.7977 18.6721 81.4936 57.9861 51.5982 295.8560
19 8.9953 8.5012 15.9232 17.8138 85.0863 58.7102 48.6782 290.7699
20 7.1764 7.9238 16.7665 18.0213 72.7662 55.5104 47.0943 287.7640
21 7.8832 7.6506 10.0802 17.8930 77.6330 54.8231 52.8840 285.6400
22 7.3289 6.7497 10.0770 18.5046 73.8520 50.6558 54.3609 283.9192
23 5.5675 8.6126 10.1853 19.2412 60.0601 61.4322 55.6895 284.9545
24 7.6854 11.7781 10.4337 21.1185 77.2340 72.8419 56.8189 290.0297

Table 8
Thermal power generation for Test Case III.

Hour Thermal power generation (MW)

PT1 PT2 PT3 PT4 PT5 PT6 PT7 PT8 PT9 PT10
1 319.0766 199.1324 93.9451 119.6141 174.2308 139.7329 45.0000 35.0000 97.8320 176.9617
2 319.1684 272.5334 94.3937 119.6391 174.3365 88.5497 45.0000 35.0000 159.7959 171.0595
3 318.5697 273.0580 94.7550 119.7154 122.6582 139.7324 45.0000 35.0000 98.0522 161.3339
4 229.5196 199.4526 94.7998 119.7331 124.4747 139.6721 104.2754 35.0000 98.0623 177.2011
5 319.2793 199.5996 94.5643 119.7331 74.8673 139.7330 45.0000 35.0000 98.0606 179.9711
6 229.5196 274.4033 94.8103 119.7525 174.6144 139.7335 45.0001 35.0000 160.0000 179.9447
7 319.2794 349.2039 94.8008 119.7331 174.6000 89.8666 104.2754 35.0000 98.0612 179.8136
8 319.2794 274.3995 94.8002 119.7332 224.4662 139.7331 104.2754 35.0002 98.0608 179.8149
9 319.2794 274.3997 94.8005 120.0617 224.4662 189.5997 104.2754 35.0000 160.0000 177.0133
10 319.2794 274.3996 94.7998 119.7331 224.2301 189.6768 45.0000 35.0000 159.9685 179.9905
11 319.1822 274.3978 94.7948 119.7330 224.2909 189.5980 104.2753 35.0000 159.9996 158.2446
12 319.2794 274.4004 94.8000 119.8082 224.5058 189.5997 163.5508 35.0000 159.9998 126.3529
13 319.2794 274.3999 94.8063 119.7331 224.4662 139.7333 104.2754 35.0000 160.0000 178.2752
14 319.2793 274.1190 94.7991 119.7330 224.4660 139.7331 104.2753 35.0000 98.0598 177.0076
15 319.2794 274.3999 94.8318 119.7331 174.5997 239.4662 104.2755 35.0000 98.0604 126.3420
16 319.2794 274.3996 94.8001 119.7218 174.5998 139.7331 104.2754 35.0000 160.0000 177.1212
17 229.5196 274.3999 94.7999 119.7331 224.4662 189.5997 163.5507 35.0000 98.0622 177.0135
18 319.2794 199.5996 94.7964 119.7331 174.5996 189.4266 163.5507 35.0000 159.9998 177.0809
19 319.2793 274.3976 94.7999 119.7319 224.4662 139.7329 104.2753 35.0000 98.0602 177.0120
20 319.2795 274.4027 94.8015 119.7892 174.6013 189.5997 104.2754 35.0000 98.0702 177.0456
21 319.2794 274.3995 94.8000 119.7331 124.7331 139.7331 45.0000 35.0000 160.0000 126.3417
22 229.5195 274.3977 94.7973 119.7331 124.7331 139.7330 104.2273 35.0000 98.0602 177.0108
23 229.5196 274.3993 94.7986 119.6872 174.5984 139.7177 45.0000 35.0000 98.0602 177.0828
24 319.2794 199.5996 94.7996 119.7326 124.7330 89.8665 45.0000 35.0000 98.0550 177.0098

For case II and case III also the cost obtained by ALO is better
than the cost reported recently by algorithms (IDE) [10] and
ORCCRO [18] respectively. The ALO happens to be faster than IDE
[10].
Table 6 shows that the maximum cost $161363.0615 attained
by ALO is found to be better than the best cost $163066.0337
reported by ORCCRO [18], SPPSO [13] $167710.56 and other NI
techniques from literature. From these observations, the better
search capability and robustness of ALO algorithm is visible.
The performance of the ALO algorithm for wind integrated
HTPGS problem is compared and validated with standard estab-
lished methods such as Differential Evolution (DE) [44], Artificial
Bee Colony (ABC) [45] and Flower Pollination Algorithm (FPA)
[46]. The best results obtained by the different methods are pre-
sented in Table 11. The minimum cost, maximum cost and mean
cost is compiled out of 25 trials. Control parameters to get the best
performance out of DE, ABC and FPA are listed in Table 12. These
values are selected by conducting some initial trials.
The convergence characteristic of ALO is compared with DE,
ABC and FPA in Fig. 17. In every trial the ALO records a superior
convergence as compared to DE, ABC and FPA for the tested
system.
The reasons for getting high quality solution from ALO are iden-
tified as:
170 H.M. Dubey et al. / Electrical Power and Energy Systems 83 (2016) 158–174

170
Plant 1
Plant 2

Reservoir stirage volume (X 10 5 m 3)

160
Plant 3
150 Plant 4

140

130

120

110

100

90

80

70
0 5 10 15 20 25
Time (Hour)

Fig. 11. Optimal hourly hydro reservoir storage volume for Test Case III computed using ALO.

30
Plant 1
Plant 2
Water discharge (X 10 4 m 3 )

25 Plant 3
Plant 4

20

15

10

5
2 4 6 8 10 12 14 16 18 20 22 24
Time (Hour)

Fig. 12. Hourly water discharge pattern Test Case III.

Hydro power generation Thermal power generation Power demand

2500
Power generation (MW)

2000

1500

1000

500

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
Time (Hour)

Fig. 13. Variation of optimal hydro and thermal power generation with power demand for Test Case III.

(i) Unique operators: The ALO has some unique features such as (v) Exploitation: The boundary of ant movement in an ant lion
random walk and adaptive shrinking of boundaries which trap is reduced iteratively; the adaptive boundary shrinking
make it an efficient optimization tool. process along with elitism promotes effective exploitation
(ii) Exploration: The random selection of ant lion for each ant and convergence.
using roulette wheel and random walk of ant around (vi) Flexibility: The ALO algorithm has no tuning coefficients, so
selected ant lion pit ensures extensive exploration of the there is no need of finding the problem dependent control
search space. parameters.
(iii) Maintaining population diversity: Random walks are calcu-
lated or each dimension of ant movement, therefore popula- Effect of wind power cost coefficients
tion diversity is maintained.
(iv) Preserving good solutions: Ant lion rebuilds its trap by acquir- If the available WP is less than the scheduled value (over-
ing the position of the best ant, therefore good solutions are estimation), the operator needs to purchase additional reserve
carried over to the next iteration. Also, the best ant lion posi- power. Conversely, if available WP is more than the scheduled
tion is saved due to elitism. WP (under estimation), then the operator has to pay a penalty for
 H.M. Dubey et al. / Electrical Power and Energy Systems 83 (2016) 158–174 171

Table 9
Optimal water discharge and hydro power generation for Test Case IV.

Hour Water discharge (104 m3) Hydro power generation (MW)

Q1 Q2 Q3 Q4 PH1 PH2 PH3 PH4
1 9.0818 6.6749 30.0000 13.0000 81.7239 54.4433 0 200.0937
2 5.8202 8.4909 29.2918 13.0001 60.3261 65.3446 0 187.7560
3 6.5396 6.7501 28.2109 13.0115 66.3472 55.9732 0 173.8141
4 10.4915 6.1927 20.6268 13.0000 88.9161 53.7112 16.6142 156.7761
5 9.5740 10.5532 15.1233 13.2261 83.4313 77.2612 38.4396 180.3333
6 7.9055 9.0875 29.5685 13.1117 73.7814 68.8510 0 198.7359
7 9.9627 6.0000 11.8302 13.0001 84.1772 49.7317 41.4317 214.4676
8 8.7774 7.3921 16.1982 13.0126 78.3332 58.8241 37.1048 222.4319
9 7.2651 7.5407 20.7543 13.8111 69.8755 60.0279 18.4278 231.1318
10 10.3005 11.1127 16.7950 13.9687 87.0771 76.9258 34.6698 247.8156
11 9.1781 8.5573 15.2322 13.1069 82.6224 64.9732 38.8624 238.3868
12 7.3327 7.2356 17.1837 13.9029 71.9825 57.5650 34.7816 248.1456
13 8.8989 11.2305 13.1827 14.6093 82.4743 76.0135 46.8512 260.3080
14 10.1175 9.3518 17.1229 14.5549 89.3953 67.3575 40.2284 261.8236
15 5.8544 7.2763 17.2679 13.3156 62.1759 56.6837 40.5064 251.5652
16 6.9258 8.6322 17.0138 15.9834 71.2245 64.2844 43.6095 277.2205
17 9.3505 10.3025 18.1437 16.8646 87.3351 70.3564 39.6724 280.9064
18 8.5243 8.4002 13.3196 14.8231 82.2916 59.1999 50.5730 265.8224
19 6.9294 9.5853 15.3345 15.9015 71.1487 63.4957 49.3500 276.5153
20 7.1111 7.6282 15.2183 19.6413 72.3695 53.2244 50.9734 301.1452
21 5.4606 7.6834 10.0034 18.2121 59.0221 54.4615 52.8147 291.7677
22 8.0930 7.3410 10.0036 19.0672 79.5539 53.5021 54.4786 291.5781
23 6.0726 9.0167 10.0363 20.6460 64.5758 62.3254 55.2300 294.8398
24 9.4328 9.9642 10.0111 21.4476 88.4354 65.5367 56.0808 291.5385

Table 10
Thermal power generation and wind power generation for Test Case IV.

Hour Thermal power generation (MW) Wind power generation

(MW)
PT1 PT2 PT3 PT4 PT5 PT6 PT7 PT8 PW1 PW2
1 319.2794 125.2182 117.4369 119.7333 124.7328 189.5997 45.0000 35.0000 162.7388 174.9999
2 319.2794 232.8778 95.5863 119.7332 174.5991 94.4975 45.0000 35.0000 175.0000 175.0000
3 319.2794 274.4014 94.8405 119.7370 25.8710 139.7353 45.0009 35.0000 175.0000 175.0000
4 229.5201 199.5989 94.7998 119.7331 74.8665 139.7331 104.2753 35.0376 167.4867 168.9314
5 232.8328 203.4266 130.0000 119.8792 84.4661 89.9059 45.0239 35.0000 175.0000 175.0000
6 139.8898 274.7620 129.9991 119.7973 224.4662 139.7331 45.0000 35.0000 174.9841 175.0000
7 319.2794 348.2126 94.7988 119.7331 174.5998 40.0195 104.2753 35.0000 168.2116 156.0616
8 229.5196 274.3807 94.7790 119.7331 274.3318 139.7328 104.2753 35.0000 166.8397 174.7141
9 319.2745 274.2357 94.7988 119.7331 223.7335 239.0940 104.2742 35.0000 139.9201 160.4731
10 319.2799 349.6887 100.3397 119.7411 124.8450 189.5997 45.0175 35.0000 175.0000 175.0000
11 319.2794 199.6002 94.7999 119.7331 224.4662 189.5997 104.2754 73.4016 174.9998 175.0000
12 319.2784 271.5257 89.3677 119.7331 274.3315 137.5429 163.5504 35.0000 164.5262 162.6695
13 319.2791 349.1991 94.7994 119.7329 174.5998 170.1589 45.0000 35.0000 165.8961 170.6877
14 409.0392 275.6862 121.9236 119.7333 124.7345 90.0788 45.0000 35.0000 174.9994 175.0000
15 319.2794 274.3995 94.7998 119.7334 174.5995 139.7332 104.2753 35.0000 165.1178 172.1309
16 319.2806 274.4011 94.8025 119.7358 175.0088 139.7344 104.2772 35.0000 166.4206 175.0000
17 139.7598 205.2866 94.7998 119.7329 224.4662 189.5996 163.5506 84.8655 174.9999 174.6687
18 319.2794 206.1493 94.7998 119.7331 174.5993 89.8665 222.8262 84.8661 174.9957 174.9977
19 319.2793 240.1604 94.7999 119.7331 224.4662 139.7021 104.2752 35.0345 157.0395 175.0000
20 409.0391 124.7996 94.7998 119.7331 124.7331 189.5996 104.2753 66.2768 168.4441 170.5870
21 319.2794 326.5389 94.7998 119.7331 124.7331 89.8665 45.0000 35.0000 135.4160 161.5672
22 229.5196 199.8259 95.2165 119.7332 174.5998 139.7331 45.0000 35.0001 167.2606 174.9985
23 139.7598 274.4012 94.8002 119.7338 224.4666 89.8674 45.0000 35.0000 175.0000 175.0000
24 319.2667 124.8006 94.7997 119.7318 124.7330 89.8665 45.0000 35.0000 172.0039 173.2063
Total cost ($) 151329.03406
Total thermal generation cost ($) 150610.06
Wind over estimation cost ($) 695.9574
Wind under estimation cost ($) 23.0162

wasting precious renewable resource. These economic aspects aris-
ing due to wind power uncertainty can be included in the economic
dispatch problem of wind-thermal power system using penalty and
reserve cost coefficients kp and kr. The selection of these coefficients
plays an important role in optimal scheduling of WP. If a higher
value of kr is selected, the optimization model will tend to schedule
lesser WP to avoid reserve cost in case available WP falls short. On
the other hand, if a higher kp is selected, there will be a higher pen-
alty for wasting WP so the operator will be inclined to schedule a
higher WP to avoid penalty. The effect of reserve and penalty coef-
ficients on optimal WP scheduling has been shown in Figs. 18 and
19 respectively for different values of Weibull scale factors.
172 H.M. Dubey et al. / Electrical Power and Energy Systems 83 (2016) 158–174

30
Plant 1
Plant 2
Plant 3

Water discharge (x 10 4 m 3)
25
Plant 4

20

15

10

5
2 4 6 8 10 12 14 16 18 20 22 24
Time (Hour)

Fig. 14. Hourly water discharge pattern Test Case IV computed using ALO.

180
Plant 1
Reservoir Storage volume (x 10 5 m3 )

Plant 2
160
Plant 3
Plant 4
140

120

100

80

60
2 4 6 8 10 12 14 16 18 20 22 24
Time (Hour)

Fig. 15. Variation in hydro reservoir storage volume for Test Case IV computed using ALO.

Hydro power generation Thermal power generation Wind power generation Power demand
2500
Power Generation (MW)

2000

1500

1000

500

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
Time (Hour)

Fig. 16. Optimal hydro and thermal power generation with power demand for Test Case IV computed using ALO.

Table 12
Parameter used for Test Case IV.
Table 11
Comparison of results for Test Case IV. ALO FPA DE ABC

Method Min cost ($) Max cost ($) Mean cost ($) CPU time (sec) Number of Number of flowers Population size Colony size
search (NP) = 20 (NP) = 50 (NP) = 50
ALO 151329.0340 151330.0873 151329.3469 68.32
agents Switching Scaling factor: Limit = 500
FPA 151787.7364 152676.8260 152498.9778 61.89
(NP) = 30 probability F1 = 0.5, F2 = 0.3
ABC 151536.6465 151787.8161 151606.9009 70.25
(q) = 0.75 Cross over
DE 151456.7793 151460.0161 151457.9439 74.43
probability: CR = 0.8
The best results are shown in bold.
 H.M. Dubey et al. / Electrical Power and Energy Systems 83 (2016) 158–174 173

x 10 5
2.1
FPA
ALO
2
ABC

Total Operating Cost ($)

DE
1.9

1.8

1.7

1.6

1.5
0 500 1000 1500
Iteration

Fig. 17. Comparison of convergence characteristic for Test Case IV.

360
C=1

350 C=5
Total Wind power (MW)

C=10
340
C=15

330

320

310

300
2 4 6 8 10 12 14 16 18 20
Kr

Fig. 18. Effect of reserve cost coefficient on WP scheduling for different Weibull scale factors.

360

350

340
Total Wind power (MW)

330

320

310
C=1
300

290 C=5

280 C=10
270
C=15
260
2 4 6 8 10 12 14 16 18 20
Kp

Fig. 19. Effect of wind penalty coefficient on WP scheduling for different Weibull scale factors.

Conclusion
The ALO algorithm takes its inspiration from the hunting beha-
viour of ant lions in nature. The six step hunting behaviour is mod-
elled using six simple operations for solving the wind integrated
HTPGS problem with a large number of complex constraints.
The performance of the ALO algorithm is compared with other
established NI algorithms for the wind integrated hydro–thermal
scheduling problem and found to be superior. The results obtained
by the ALO algorithm fully satisfy all the complex constraints
associated with hydro, thermal and wind generating units. In
ALO, population diversity is obtained due to the thorough explo-
ration of search space through random walks, which are generated
for each ant dimension. Selection of ant lion using roulette wheel
mechanism also increases the population diversity. The unique
boundary shrinking mechanism promotes exploitation and good
solutions are preserved through the concept of elitism. The ALO
demonstrates its applicability for solving real-world optimization
174 H.M. Dubey et al. / Electrical Power and Energy Systems 83 (2016) 158–174

problems due to (i) the continual focus and control on exploration [20] Liao X, Zhou J, Ouyang S, Zhang R, Zhang Y. An adaptive chaotic artificial bee
colony algorithm for short-term hydrothermal generation scheduling. Int J
and exploitation, (ii) stable convergence, (iii) ability to maintain
Electr Power Energy Syst 2013;53:34–42.
population diversity, (iv) the absence of tuning parameters, and [21] Swain RK, Barisal AK, Hota PK, Chakrabarti R. Short-term hydrothermal
(v) high solution quality. It can be employed for the solution of scheduling using clonal selection algorithm. Int J Electr Power Energy Syst
other large scale problems of modern power system. 2011;33:647–56.
[22] Reddy S Surender, Panigrahi BK, Debchoudhury Shantanab, Kundu Rupam,
Mukherjee Rohan. Short-term hydro-thermal scheduling using CMA-ES with
directed target to best perturbation scheme. Int J Bio-Inspired Comput 2015;7
Acknowledgements (3):195–208.
[23] Dubey Hari Mohan, Panigrahi BK, Pandit Manjaree. Bio-inspired optimization
for economic load dispatch: a review. Int J Bio-Inspired Comput 2014;6
The authors sincerely acknowledge the financial support (1):7–21.
provided by UGC-MRP vide F No. 34-399/2008 (SR) dated, 24th [24] Pandit Manjaree, Srivastava Laxmi, Sharma Manisha, Dubey Hari Mohan,
December 2008. The second author acknowledges UGC research Panigrahi BK. Large scale multi-zone optimal power dispatch using hybrid
hierarchical evolution technique. IET J Eng 2013. http://dx.doi.org/10.1049/
award for post doctoral work sanctioned by UGC, New Delhi, India, joe.2013.0262. IET digital library.
vide letter no. F-30-120(SC)/2009 (SA-II). The authors also thank [25] Pandit Manjaree, Chaudhary Vishal, Dubey Hari Mohan, Panigrahi BK. Multi-
the Director and management of M.I.T.S. Gwalior, India and IIT period wind integrated optimal dispatch using series PSO-DE with time-
varying Gaussian membership function based fuzzy selection. Electr Power
Delhi, India, for providing facilities for carrying out this work. Energy Syst 2015;73:259–72.
[26] Mirjalili S. The ant lion optimizer: advances in engineering software; 2015, vol.
83, p. 80–98.
References [27] Zheng YJ. Water wave optimization: a new nature-inspired meta-heuristic.
Comput Oper Res 2015;55:1–11.
[28] Mirjalili S. Moth-flame optimization algorithm: a novel nature-inspired
[1] Wood AJ, Wollenberg BF. Power generation, operation and control. 2nd
heuristic paradigm. Knowl-Based Syst 2015;89:228–49.
ed. New York: John Wiley & Sons; 1996.
[29] Salimi H. Stochastic Fractal Search: a powerful metaheuristic algorithm.
[2] Salam Nor KM, Hamdan AR. Hydrothermal scheduling based Lagrangian
Knowl-Based Syst 2015;75:1–18.
relaxation approach to hydrothermal coordination. IEEE Trans Power Syst
[30] Yu JJQ, Li VOK. A social spider algorithm for global optimization. Appl Soft
1998;13:226–35.
Comput 2015;30:614–27.
[3] Zaghlool MF, Trutt FC. Efficient methods for optimal scheduling of fixed head
[31] Shareef H, Ibrahim AA, Mutlag AH. Lightning search algorithm. Appl Soft
hydrothermal power systems. IEEE Trans Power Syst 1988;3:24–30.
Comput 2015;36:315–33.
[4] Sinha N, Chakrabarti R, Chattopadhyay PK. Fast evolutionary technique for
[32] Hetzer J, Yu DC, Bhattarai K. An economic dispatch model incorporating wind
short-term hydrothermal scheduling. IEEE Trans Power Syst 2003;18:214–20.
power. IEEE Trans Energy Convers 2008;23(2):603–11.
[5] Basu M. An interactive fuzzy satisfying method based on evolutionary
[33] Lakshmi K, Vasantharathna S. Gencos wind–thermal scheduling problem using
programming technique for multi-objective short-term hydrothermal
artificial immune system algorithm. Electr Power Energy Syst
scheduling. Electr Power Syst Res 2004;69:277–85.
2014;54:112–22.
[6] Orero SO, Irving MR. A genetic algorithm modeling framework and solution
[34] Chen CL, Chen ZY, Lee TY. Multi-area economic generation and reserve
technique for short-term optimal hydrothermal scheduling. IEEE Trans Power
dispatch considering large-scale integration of wind. Electr Power Energy Syst
Syst 1998;13:501–18.
2014;55:171–8.
[7] Lakshminarasimman L, Subramanian S. Short-term scheduling of
[35] Niknam T, Massrur HR. Stochastic mid-term generation scheduling
hydrothermal power system with cascaded reservoirs by using modified
incorporated with wind power. Int J Electr Power Energy Syst 2015;64:
differential evolution. IEE Proc Gene Trans Distrib 2006;153:693–700.
237–46.
[8] Lakshminarasimman L, Subramanian S. A modified hybrid differential
[36] Surender Reddy S, Bijwe PR. Real time economic dispatch considering
evolution for short-term scheduling of hydrothermal power systems with
renewable energy resources. Renewable Energy 2015;83. 12151-1226.
cascaded reservoirs. Energy Convers Manage 2008;49:2513–21.
[37] Surender Reddy S, Momoh James A. Realistic and transparent optimum
[9] Mandal KK, Chakraborty N. Differential evolution technique based short-term
scheduling strategy for hybrid power system. IEEE Trans Smart Grid 2015;6
economic generation scheduling of hydrothermal systems. Electr Power Syst
(6):3114–25.
Res 2008;78:1972–9.
[38] Dubey Hari Mohan, Pandit Manjaree, Panigrahi BK. Hybrid flower pollination
[10] Basu M. Improved differential evolution for short-term hydrothermal
algorithm with time-varying fuzzy selection mechanism for wind integrated
scheduling. Int J Electr Power Energy Syst 2014;58:91–100.
multi-objective dynamic economic dispatch. Renewable Energy 2015;83:
[11] Yaoyao H, Qifa X, Shanlin Y, Aoyang H, Yang L. A novel chaotic differential
188–202.
evolution algorithm for short-term cascaded hydroelectric system scheduling.
[39] Yuan X, Tian H, Yuan Y, Huang Y, Ikram RM. An extended NSGA-III for solution
Int J Electr Power Energy Syst 2014;61:455–62.
of multi-objective hydro-thermal-wind scheduling considering wind power
[12] Zhang J, Lin S, Qiu W. A modified chaotic differential evolution algorithm for
cost. Energy Convers Manage 2015;96:568–78.
short-term optimal hydrothermal scheduling. Int J Electr Power Energy Syst
[40] Surender Reddy S, Bijwe PR, Abhyankar AR. Multi-objective market clearing of
2015;65:159–68.
electrical energy, spinning reserves and emission for wind-thermal power
[13] Zhang J, Wang J, Yue C. Small population-based particle swarm optimization
system. Electr Power Energy Syst 2013;53:782–94.
for short-term hydrothermal scheduling. IEEE Trans Power Syst 2012;27:
[41] Surender Reddy S, Abhyankar AR, Bijwe PR. Market clearing for a wind-
142–52.
thermal power system incorporating wind generation and load forecast
[14] Rasoulzadeh-akhijahani A, Mohammadi-ivatloo B. Short-term hydrothermal
uncertainties. In: IEEE Power Energy Soc General Meet. p. 1–8. http://dx.doi.
generation scheduling by a modified dynamic neighborhood learning based
org/10.1109/PESGM.2012.6345335.
particle swarm optimization. Int J Electr Power Energy Syst 2015;67:350–67.
[42] Surender Reddy S, Bijwe PR, Abhyankar AR. Optimal posturing in day-ahead
[15] Roy PK. Teaching learning based optimization for short-term hydrothermal
market clearing for uncertainties considering anticipated real-time
scheduling problem considering valve point effect and prohibited discharge
adjustment costs. IEEE Syst J 2015;9(1):177–90.
constraint. Int J Electr Power Energy Syst 2013;53:10–9.
[43] Gouthamkumar N, Sharma V, Naresh R. Disruption based gravitational search
[16] Roy PK, Paul C, Sultana S. Oppositional teaching learning based optimization
algorithm for short term hydrothermal scheduling. Expert Syst Appl
approach for combined heat and power dispatch. Int J Electr Power Energy Syst
2015;42:7000–11.
2014;57:392–403.
[44] Storn R, Price K. Differential evolution: a simple and efficient adaptive scheme
[17] Bhattacharjee K, Bhattacharya A, nee Dey SH. Real coded chemical reaction
for global optimization over continuous spaces. Berkeley, CA: International
based optimization for short-term hydrothermal scheduling. Appl Soft Comput
Computer Science Institute; 1995.
2014;24:962–76.
[45] Karaboga D, Basturk B. A powerful and efficient algorithm for numerical
[18] Bhattacharjee K, Bhattacharya A, nee Dey SH. Oppositional real coded chemical
function optimization: artificial bee colony (ABC) algorithm. J Global Optim
reaction based optimization to solve short-term hydrothermal scheduling
2007;39:459–71.
problems. Int J Electr Power Energy Syst 2014;63:145–57.
[46] Yang XS. Flower pollination algorithm for global optimization, unconventional
[19] Fang N, Zhou J, Zhang R, Liu Y, Zhang Y. A hybrid of real coded genetic
computation and natural computation. Lect Notes Comput Sci, LNCS
algorithm and artificial fish swarm algorithm for short-term optimal
2012;7445:240–9.
hydrothermal scheduling. Int J Electr Power Energy Syst 2014;62:617–29.