You are on page 1of 12

Electrical Power and Energy Systems 51 (2013) 178–189

Contents lists available at SciVerse ScienceDirect

Electrical Power and Energy Systems


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

Civilized swarm optimization for multiobjective short-term


hydrothermal scheduling
A. Immanuel Selvakumar ⇑
Department of Electrical and Electronics Engineering, Karunya University, Coimbatore 641 114, India

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

Article history: This paper presents a solution technique for multiobjective short-term hydrothermal scheduling
Received 10 October 2012 (MSTHTS) through civilized swarm optimization (CSO) which is the hybrid of society–civilization algo-
Received in revised form 17 February 2013 rithm (SCA) and particle swarm optimization (PSO). The intra and inter society communication mecha-
Accepted 2 March 2013
nisms of SCA have been embedded into the food-searching strategy of PSO to form CSO. The MSTHTS
Available online 2 April 2013
problem is formulated by considering economic and emission objectives. A new ideal guide method
has been proposed to find out the Pareto-optimal front. Multi-reservoir cascaded hydro power plants
Keywords:
having nonlinear generation characteristics and thermal power plants with non-smooth cost and emis-
Society
Civilization
sion curves are considered for analysis. Other aspects such as, water transport delay, water availability,
Particle swarm optimization storage conformity, power loss and operating limits are fully accounted in the problem formulation. The
Hydrothermal scheduling performance of the proposed CSO is demonstrated through two MSTHTS problems and the results are
Multiobjective optimization compared with those presented in the literature. CSO along with the new ideal guide method outper-
Economic dispatch forms all the previous approaches by providing quality Pareto-optimal fronts.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction tation techniques have been given much attention to address


STHTS. Stochastic search algorithms such as, simulated annealing
Short-term hydrothermal scheduling (STHTS) is one of the most (SA) [8], genetic algorithm (GA) [9,10], and evolutionary program-
significant optimization problems in economic operation of inter- ming(EP) [11] have been successfully applied to STHTS. Neural net-
connected power systems with hydroelectric plants. The primary works have also been used to solve STHTS [12,13]. However, in the
objective of STHTS is to allocate the available water for hydro gen- above-mentioned works, the thermal plants are represented by
eration among the various time intervals along the scheduling quadratic cost functions. This simplified model may lead to subop-
horizon and to dispatch the thermal units to minimize fuel cost timal solution. Hence, the trend in recent times is to use more real-
for thermal generation while satisfying diverse constraints. The istic model for both hydroelectric and thermal plants by including
short-term planning resolution is hourly and the scheduling length prohibited discharge zones and valve-point effects. While doing so,
is normally 1 day to 1 week. When several hydroelectric plants are the STHTS problem becomes highly nonlinear and nonconvex. To
located in the same river, the water outflow from one plant may be solve such a complex STHTS, techniques like EP [14], improved fast
a very significant portion of the inflow to other plants which are lo- EP [15], clonal selection algorithm [16] and quantum-inspired evo-
cated downstream. Scheduling such coupled hydroelectric plants is lutionary algorithm (QEA) [17] have been applied.
a formidable task; both electric and hydraulic couplings form a All the previously mentioned works consider only the economic
multi-dimensional, nonlinear programming problem. Conven- objective. Nevertheless, the generation of electricity from fossil
tional methods based on LaGrange multiplier and gradient search fuel releases several pollutants, such as sulfur dioxide, nitrogen
techniques were used to solve the STHTS, but they have the diffi- oxides and carbon dioxide, which are harmful to the life-forms. It
culty in dealing with the constraints [1]. is necessary, therefore, to consider the emission along with eco-
The STHTS has been the subject of intensive investigations for nomic operation of power plants [18]. In view to the pollution con-
several decades. Different methods have been proposed for the trol, the STHTS problem formulation includes the emission as
solution of this problem. Some of these solution methods include another objective. Hence, the problem becomes multiobjective
dynamic programming [2–4], network flow concept [5,6] and short-term hydrothermal scheduling (MSTHTS) and requires com-
decomposition technique [7]. In recent years, evolutionary compu- bined economic and emission dispatch (CEED) of thermal plants
along with scheduling of hydroelectric power plants. Various ef-
⇑ Tel.: +91 422 2614392. forts have been taken to solve MSTHTS with different levels of suc-
E-mail address: iselvakumar@yahoo.co.in cess. In [19], an interactive fuzzy satisfying method based on EP

0142-0615/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijepes.2013.03.002
A. Immanuel Selvakumar / Electrical Power and Energy Systems 51 (2013) 178–189 179

Nomenclature

k iteration count objective(k) minimum objective function value until kth iteration
kmax maximum number of iterations L(k) swarm leader at kth iteration
M swarm size Lit(k) i, t-element of the swarm leader at kth iteration
Ns number of society leaders Y Lit (k) i, tth element of the velocity of swarm leader at kth iter-
Nr number of society members ation
L
w inertia weight; decreased from wmax to wmin during Pbestit (k) i, tth element of the own best position of swarm leader
optimization until kth iteration
Cl coefficient, which accelerates the swarm leader towards SL(k)/SM(k) set of society leaders/society members at kth itera-
its own best position tion
Csl1 coefficient, which accelerates the society leader towards SLs(k)/SMr(k) society leader ‘s’/society member ‘r’ at kth iteration
its own best position SLsit =SMrit (k) i, tth element of SLs(k)/SMr(k)
Csl2 coefficient, which accelerates the society leader towards Y sit (k) i, tth element of the velocity of society leader ‘s’ at kth
the swarm leader iteration
Csm1 coefficient, which accelerates society member towards Y rit (k) i, tth element of the velocity of society member ‘r’ at kth
its own best position iteration
s
Csm2 coefficient, which accelerates society member towards Pbestit (k) i, tth element of the own best position of society leader
its society leader ‘s’ until kth iteration
r
r1, r2, r3, r4, r5 random numbers chosen from a uniform distribu- Pbestit (k) i, tth element of the own best position of society mem-
tion on the interval [0, 1] ber ‘r’ until kth iteration
Swarm(k) swarm at kth iteration XL new location of the swarm leader
Xm(k) position of mth particle at kth iteration XSLs =Xr
SM
new location of SLs/SMr
Ym(k) velocity of mth particle at kth iteration L
Y (k) velocity of the swarm leader at kth iteration
Pbestm(k) personal best position of mth particle until kth iteration Ys(k)/Yr(k) velocity of the society leader ‘s’/ society member ‘r’ at
w(Xm(k)) penalized objective function value for Xm(k) kth iteration

(fuzzy EP) has been proposed for MSTHTS. A simulated annealing and emission objectives. CSO is used as the optimizer to obtain a
based goal-attainment (SAGA) method, in which the multiobjec- solution closer to ideal guides by minimizing the Euclidean dis-
tive problem is transformed into single objective optimization by tance. CSO has been developed by the author for nonconvex static
goal attainment and then solved by SA, has been discussed in economic dispatch (NSED) problems [31]. In this paper, CSO is
[20]. Improved GA with multiplier updating (IGAMU) [21] is an- modified to be suitable for STHTS, which is dynamic and more
other evolutionary computation technique that has been applied complex than NSED. Metaheuristic techniques for optimization
for MSTHTS with e-constraint technique to manage the multiple can be broadly classified into purebred and hybrid. In the purebred
objectives. In the e-constraint technique, one of the objectives category, a particular evolutionary computation or swarm intelli-
(emission in this case) is redefined as a constraint. Many research- gence technique is used as a stand-alone tool to carry out the opti-
ers have used price penalty factor to combine the economic and mization. Purebred metaheuristic algorithms like GA, EP, SA and
emission objectives; any one of the algorithms can solve the resul- PSO are well performing algorithms for small dimensional and less
tant single objective optimization problem. In this way, modified complicated problems. They are able to find out the global
differential evolution (MDE) [22], quantum particle swarm optimi- minimum for less complicated functions with few local minima.
zation (QPSO) [23] and PSO [24] based methods have been applied Nevertheless, they fail to locate global minimum for complex mul-
to schedule hydrothermal power systems. Quadratic approxima- ti-minima functions. Although they locate promising area, they fail
tion based differential evolution with valuable trade-off (QADEVT) to exploit the promising area to get quality solutions. Several at-
[25] is another approach in which economic and emission objec- tempts have been made to control and to strike a balance between
tives are integrated into a single objective using valuable trade- exploration of whole search space to locate promising area and
off ratio and price penalty factor. Two objectives are combined exploitation of the promising area to get global minimum. With
using weight and normalization factors and the resultant single a purebred technique, the goal is difficult to achieve. Hybrid tech-
objective problem has been solved by GA [26]. However, it is diffi- niques are attractive alternatives in which global and local search
cult to find out the uniformly distributed set of Pareto-optimal can be carried out by two constituent techniques. CSO is such a
solutions with these approaches. Another trend is finding all the hybrid technique, which has been developed by integrating soci-
Pareto-optimal solutions in one single optimization run using evo- ety–civilization algorithm (SCA) [32] and PSO [33]. SCA, introduced
lutionary algorithms. Nondominated sorting GA-II (NSGA-II) [27], by Ray and Liew, finds the optimal solution based on intra and
multiobjective cultural algorithm based on PSO (MOCAPSO) [28] intersociety interactions. Kennedy and Eberhart introduced PSO
and multi-objective differential evolution (MODE) [29] are few in 1995. It is a population-based, self-adaptive search optimization
examples. technique and has been successfully applied for many power sys-
In this context, this paper proposes civilized swarm optimiza- tem problems [34]. PSO simulates the food searching activity of a
tion (CSO) based MSTHTS with cost and emission as competing swarm of birds to perform optimization. The optimum searching
objectives. A new ideal guide based method has been proposed mechanisms of SCA and PSO have been combined to form an
to find the Pareto- optimal front. In this approach a set of ideal efficient swarm intelligence technique, CSO, which has a perfect
guides are generated and a solution which is closer to a particular balance between local and global searching abilities. To validate
ideal guide is found out. By this way, the required number of solu- the CSO based technique for MSTHTS, it has been illustrated with
tions in the Pareto-optimal front can be determined. A global crite- two hydrothermal test systems. Performance comparison with
rion function concept [30], which is based on Euclidean distance, is existing methods on solution optimality, consistency and execu-
utilized in formulating a single objective function from economic tion time is presented for economic, emission and combined
180 A. Immanuel Selvakumar / Electrical Power and Energy Systems 51 (2013) 178–189

economic and emission dispatch cases. Test results confirm that


the proposed CSO is faster and more consistent than the previous
Ideal Guide
approaches in giving minimum fuel cost and emission. Moreover,
CSO provides best Pareto-front in all the test cases, which has been
validated through a set of multiobjective optimization perfor-
mance metrics [35].

2. Multiobjective short-term hydrothermal scheduling


(MSTHTS)

STHTS of power systems has been extensively studied because


of its significant economical impact. In a hydrothermal power sys-
tem, the marginal cost of hydroelectric generation is negligible.
Moreover, the hydro plant by itself cannot fully meet the load de-
mand. This implies that the available water in system reservoirs
should be used as much as possible for hydropower generation
in such a way that the overall cost of thermal generation is mini- Fig. 1. The concept of ideal guides.
mized over the planning period while satisfying various
constraints.
Multiobjective STHTS, the extension of STHTS, aims at the allo- 2.3. Combined economy and emission objective
cation of available water in system reservoirs to be used for hydro-
power generation in such a way that both the fuel cost and the There are several methods available to combine and to handle
emission of thermal generation are minimized. While doing so, the economic and emission objectives. Some examples are fuzzy
many system-operating constraints, including hydraulic, thermal approach [19], goal-attainment [20], e-constraint technique [21],
and electrical aspects must be satisfied. price-penalty factor method [22–24], valuable trade-off approach
[25] and weighted-sum approach [26]. This paper proposes a
2.1. Economy objective new procedure based on ideal guides to find out the Pareto-opti-
mal front and it is outlined below.
The economy objective is to minimize the fuel cost of all the on-
line thermal units over the short-term scheduling horizon. Valve-  Step1. The minimum fuel cost (FCmin) is determined by minimiz-
point effects are included by expressing fuel cost function as the ing the cost objective and the corresponding emission release is
sum of quadratic and sinusoidal functions. This inclusion provides ERmax.
realistic model of cost function. Mathematically, the economic  Step2. The minimum emission release (ERmin) is determined by
objective is expressed as the following function. minimizing the emission objective and the corresponding fuel
cost is FCmax.
X X 2    
   Step3. The set of ideal guides, i.e., points are generated along the
min FC ¼ Nt aj PSjt þ bj PSjt þ cj þ ej sin fj Pmin
Sj  P Sjt  ð1Þ
t2T j2N th
FCmin and ERmin axes as shown in Fig. 1.
 Step4. To find out a Pareto-optimal point closer to an ideal
where FC is total fuel cost ($); T is the number of scheduling inter- guide, a global criterion function (GCF) [30] which describes
vals; Nt is the length of tth subinterval (h); Nth is the number of the measure of ‘closeness to the ideal solution’ is used. The
thermal plants; aj, bj, cj, ej, fj are fuel-cost curve coefficients of jth GCF transforms the multiobjective problem into a single objec-
thermal plant; PSjt is the real power generation of jth thermal plant tive problem and it is expressed as
during tth subinterval (MW); and Pmin Sj and P max
Sj are the minimum
and maximum power generation limits of jth thermal plant, 2 !2 !2 312
respectively. FC 1  FC ER1  ERÞ
GCF ¼ 4 þ 5 ð3Þ
FC 1 ER1
2.2. Emission objective
where FC1 and ER1 are the coordinates of ideal guide 1. The
Since thermal power stations emit high concentration of Pareto-optimal point closer to an ideal guide can be determined
pollutants, they are the major causes of atmospheric pollution. by minimizing GCF through CSO.
Among the various pollutants from the fossil-fuel based
power plants, the nitrogen oxides (NOx) emission is accounted Step4 can be repeated to determine Pareto-optimal points
in this work. Fossil-fuel based generating stations, primary closer to other ideal guides. An ideal guide directs the optimizer
sources of NOx, are strongly requested by the environmental to get a Pareto-optimal point closer to it. These points are named
protection agency to reduce their emissions. The amount of as ideal guides because of two reasons. Firstly, the ideal guides
NOx emission from each thermal unit is expressed as the are fictitious; the optimizer cannot find a solution on this location
function of its output. The emission function is accurately but it can find out a point closer to it. Second reason is that the
represented as the superposition of quadratic and exponential point ‘guides’ the optimizer to find out a solution point nearer to
functions, and the emission objective is mathematically ex- it. By minimizing the GCF with different ideal guides, the entire
pressed as in Eq. (2). Pareto optimal points could be found out one by one.

X X 2
min ER ¼ Nt aj PSjt þ bj PSjt þ cj þ gj edj PSjt ð2Þ 2.4. Constraints
t2T j2Nth

The leading constraints of the MSTHTS are demand–supply bal-


where ER is the total emission release (lb) and aj, bj, cj, gj, dj are the ance, operational limits on hydro and thermal generations, flow
emission-curve coefficients of jth thermal plant. balance or continuity equation, bounds of water discharge through
A. Immanuel Selvakumar / Electrical Power and Energy Systems 51 (2013) 178–189 181

turbines and bounds on reservoir storage. Additional constraints and emission while attempting to satisfy the various unit and
such as flood control, irrigation, recreation, fishing and wild life reservoir constraints. The main consideration is the time cou-
maintenance could be imposed depending on contractual or legal pling effect of the hydro reservoirs, where the water flow in an
obligations with respect to the hydro system network. The hourly earlier time interval affects the discharge capability at a later
demands and inflow of the day under consideration can be forecast period. The time coupling effect is accounted by introducing
with good accuracy; hence, the usual assumption is that load de- the continuity equation, which is given by
mands and river inflows are known. The constraints involved in X
MSTHTS problem are described below:
V it ¼ V iðt1Þ  qit  Sit þ Iit þ ðquðtsui Þ þ Suðtsui Þ Þ ð8Þ
u2Rui

2.4.1. Demand–supply balance where Sit is spillage from ith reservoir during tth subinterval;
The total active power generation from hydro and thermal Iit is the natural water inflow to ith reservoir during tth sub-
plants must balance the predicted power demand and losses at interval; Rui is the set of upstream hydro plants directly above
each time interval over the scheduling horizon. ith hydro plant and sui is the water transport delay between
X X ith reservoir and its upstream reservoir u. The suffix ‘u’ de-
PHit þ PSjt ¼ PLoadt þ PLosst ; 8t 2 T ð4Þ
notes upstream reservoir.
i2Nhy j2N th
(b) Initial and final reservoir storage (storage conformity): The
where Nhy is the number of hydro plants; PHit is the real power gen- reservoir storage levels at the initial and final scheduling
eration of ith hydro plant during tth subinterval (MW); PLoadt is the intervals must match with the initial and target dam levels.
total load in the system during tth subinterval (MW) and PLosst is the
V i0 ¼ V initial
i ; 8i 2 Nhy ; V iT ¼ V end
i ; 8i 2 N hy ð9Þ
network loss during tth subinterval (MW), which is determined
using B-matrix loss formula. where V initial
iand V end
are the initial and target levels of ith
i
The power generated from a hydro plant is related to the reser- reservoir, respectively.
voir characteristics as well as the water discharge rate. A number (c) Bounds on water discharge:
of models have been used to represent this relationship. In general,
qmin
i 6 qit 6 qmax
i ; 8i 2 Nhy ; 8t 2 T ð10Þ
the hydro generator power output is a function of the net hydraulic
head (H), reservoir volume V, and the rate of water discharge (q). where qmin
and qmax
are the lower and upper bounds of water
i i
Constant water head is generally assumed in short-term schedul- discharge of i th reservoir, respectively.
ing. However, this assumption is true only in the case of large (d) Bounds on reservoir storage:
capacity reservoirs. Head variation cannot be ignored if there is a
strong relationship between inflow and capacity. Since the net V min
i 6 V it 6 V max
i ; 8i 2 Nhy ; 8t 2 T ð11Þ
head is a function of volume of stored water, hydropower genera- where V min
and V max
are the lower and upper bounds of stor-
i i
tion can be written in terms of turbine discharge rate and reservoir age volume of i th reservoir, respectively.
volume as given by (e) Water availability: In an alternative problem formulation,
PHit ¼ c1i V 2it þ c2i q2it þ c3i V it qit þ c4i V it þ c5i qit þ c6i ð5Þ hydraulic continuity and storage conformity constraints are
combined to form water availability constraint [20].
where c1i ; c2i ; c3i ; c4i ; c5i and c6i are the power generation coefficients
X  
of ith hydro plant; Vit and qit are storage and water discharge of ith
Nt aoi P2Hit þ a1i P Hit þ a2i ¼ W i ; 8i 2 Nhy ð12Þ
reservoir during tth subinterval, respectively. t2T

2.4.2. Bounds on hydro and thermal generations where a0i, a1i and a2i are the coefficients for water discharge
The operational ranges of hydro and thermal plants are rate function of ith hydro plant and Wi is the pre-specified
bounded by their capability (loading) limits. volume of water available for ith hydro plant during the
scheduling horizon.
Pmin max
Hi 6 P Hit 6 P Hi ; 8i 2 Nhy ; 8t 2 T ð6Þ
3. Civilized swarm optimization (CSO) and its implementation
Pmin
Sj 6 PSjt 6 Pmax
Sj ; 8j 2 Nth ; 8t 2 T ð7Þ to MSTHTS

where Pmin max


Hi and P Hi are the minimum and maximum power gener- 3.1. Concept of CSO
ation limit of ith hydro plant, respectively.
Civilized swarm optimization (CSO), the new optimization
2.4.3. Hydraulic network constraints strategy, is a hybrid swarm intelligence technique, which has been
The hydraulic constraints are determined by the reservoir and developed by integrating society–civilization algorithm (SCA) [32]
plant limitations as well as the multipurpose requirements of the with PSO [33]. Ranges of nature-inspired algorithms are available
hydro system. Following are the important hydraulic constraints: to solve complex optimization problems. Examples of such algo-
rithms include GA, EP, evolutionary strategy (ES), and ant colony
 Continuity or water balance equation for the hydro reservoir model. In this series, SCA and PSO are emerged from socio-behav-
network. ioral dynamics of the nature. The basis for SCA is derived out of two
 Desired storage volume. observations from nature [32].
 Physical limitations on reservoir discharge rates and storage
volumes.  A civilization progresses due to cooperative relationships
(a) Hydraulic continuity: In short-term scheduling, the target among its societies.
dam levels at the end of the scheduling period are set by a  Individuals of societies improve due to interaction among soci-
medium term scheduling process that takes into account ety members.
long-term river inflow modeling and load predictions. The
short-term scheduler then allocates this water to the various In SCA, a set of individuals are grouped into small societies
time intervals in an effort to minimize thermal generation cost (clusters); each cluster has better performing individuals as the
182 A. Immanuel Selvakumar / Electrical Power and Energy Systems 51 (2013) 178–189

society leaders (SLs). The societies form the civilization and the 3.2. Implementation of CSO for MSTHTS
best performer among all the society leaders is the civilization
leader (CL). During the optimization, the society leaders help In general, the implementation of a population based technique
their society members (SMs) to improve. On the other hand, is very difficult in the case of MSTHTS. Because, a single individual
the CL guides all society leaders. In PSO, each particle (bird) of of the population has to represent the water discharge schedules of
the swarm represents a possible solution to the optimization all the hydro reservoirs and the power generations of all the ther-
problem. Particles move through random trajectories towards mal power plants during all the scheduling intervals. Following are
the optimal point with the help of velocities. Three components, the important factors to be considered while implementing the
previous velocity, bird’s memory about its previous best position, proposed CSO for the solution of MSTHTS.
and memory about swarm’s best position influence the velocity
of the particle. The SCA differs with PSO in the fact that the indi- 3.2.1. Representation of the swarm
viduals of SCA follow only their leaders neglecting their own In general, water discharge rates of reservoirs (q) and thermal
experiences. Based on the optimum seeking mechanisms of generations (PS) are the decision variables for short-term hydro-
SCA and PSO, their relative merits and demerits are discussed thermal scheduling. The reservoir storage and hydro generation
below. are calculated using Eqs. (8) and (5), respectively. Therefore, q
and PS are used to form the swarm. The swarm with M particles
 The intrasociety interaction between SMs and the correspond- is represented as a matrix below.
ing SL enhances the local searching ability of the SCA. SMs
Swarm ¼ ½ ½X1  ½X2     ½Xm     ½XM   ð13Þ
can exploit a promising region efficiently. The communication
between the SLs and the CL (intersociety interaction) provides where the matrix Xm represents the mth particle of the swarm (mth
exploration capability to SCA. SLs explore new areas, which possible solution, i.e., hydrothermal schedule) and it is defined as:
 m 
are exploited by SMs. Thus, SCA has a perfect balance ½Q 
Xm ¼  m  ; where Qm is a Nhy  T sub-matrix and the element
between local and global searching abilities. However, PSO PS
m
suffers from the imbalance between local and global search. qm
it of Q is the i th reservoir discharge rate at tth time interval
During the earlier optimization stages, it explores the search for the mth solution; Pm m
S is a Nth  T matrix and the element P Sjt of
space well; but it fails to exploit a particular promising PmS is the jth thermal generation at tth time interval for the mth pos-
region. During later stages of optimization, exploitation dom- sible solution. There are (Nhy + Nth) rows and (T  M) columns in the
inates exploration. Nevertheless, SCA has both exploration swarm matrix.
and exploitation throughout the optimization.
 In the strategy of SCA, individuals are guided only by others, i.e., 3.2.2. Constraint handling
SMs are guided by SLs and SLs are guided by CL. In PSO individ- The elements qit and PSjt of the swarm are clamped between
uals (particles) seek the optimum through the guidance of the minimum and maximum values to satisfy the constraints given
particle with the maximum fitness, i.e., Gbest particle. Also, by Eqs. (10) and (7), respectively. In order to exactly meet the ini-
they utilize their own experiences in seeking the optimum. tial and final reservoir storage (Eq. (9)), hydro plant discharge at a
Valid information about the search space gathered by individu- dependent interval d is randomly selected among the intervals [1,
als (particles’ personal best experiences) is effectively utilized in 2, . . . , T]. Then the discharge at the dependent interval is calculated
PSO to decide the search path. Whereas in SCA, individuals’ own from Eq. (8) with the assumption of zero spillage.
experiences are neglected. X X XX
 SCA simulates only two movements: SMs towards SL and qid ¼ V i0  V iT  qit þ Iit þ quðtsui Þ ; 8i 2 Nhy ð14Þ
SLs towards CL. Whereas the CL remains idle without doing t2T
t–d
t2T t2T u2Rui
any search. In PSO, the Gbest particle participates in the
search. All particles of PSO are involved in the optimization In addition, to meet the demand–supply balance constraint ex-
process. actly, a dependent thermal plant is selected among the Nth thermal
units and the generation level of the dependent thermal plant is
The strength of SCA is the perfect local–global search balance calculated by
and the PSO has the merit of utilizing personal experiences. CSO X X
combines the strength of each algorithm. In CSO, the swarm is con- PSdt ¼ PLoadt  PHit  PSjt ; 8t 2 T ð15Þ
sidered to be a civilization and clustered into small societies. The i2N hy j2N th
j–d
best-performing particle of the society is the SL. Among the various
SLs, the best performer is the leader (L) of the swarm. The particles If losses are included, the RHS of Eq. (15) will consists of PLosst
belonging to a society are made to follow their SL and own experi- term which is a function of PSdt. In that case, PLosst is expressed as
ences; hence, they will move within small clusters to exploit a a function of all the power generations including the dependent
‘‘promising area’’. All the society leaders are made to explore power generation. Then Eq. (15) will be a quadratic equation with
new promising areas with the help of their own experiences, and PSdt as the unknown and can be solved for PSdt [20].
that of swarm leader. Therefore, the proposed CSO is well balanced The above two adjustments are made for all the particles of the
between local and global searching abilities which are the vital swarm. Moreover, all the dependent variables must satisfy the cor-
requirement of a heuristic algorithm. Moreover, the swarm of responding inequality constraints; otherwise, penalty terms are
CSO possesses the detail knowledge of the search space. In the pro- added to the objective function.
posed CSO, a particle in the swarm is made to have the following X X XX
attributes. w ¼ GCF þ k1 MVQ id þ k2 MVPSdt þ k3 MVPHit
i2Nhy t2T t2T i2Nhy
XX
(1) A position in search space – X. þ k4 MVVit ð16Þ
(2) Its status in the swarm – an individual may have any one of t2T i2N hy
the following status: leader (L); SL and SM.
(3) A velocity to fly – Y. where w is the penalized objective function; k1 ; k2 ; k3 ; k4 are the
(4) Personal best position encountered by it so far- Pbest. penalty factors; MVQid is the magnitude of violation in water dis-
A. Immanuel Selvakumar / Electrical Power and Energy Systems 51 (2013) 178–189 183

charge limits for ith reservoir during dependent subinterval; MVPSdt Y rit ðk þ 1Þ ¼ wY rit ðkÞ þ C sm1 r 4 Pbestit ðkÞ  SMrit ðkÞ
is the magnitude of violation in thermal generation limits for

þ C sm2 r5 SLsr r
it ðkÞ  SM it ðtÞ for 8i
dependent thermal plant during tth subinterval; MVPHit is the mag-
nitude of violation in hydro generation limits for ith hydro plant 2 ðNhy þ Nth Þ; 8t 2 T; 8r 2 N r ð21Þ
during tth subinterval; and MVVit is the magnitude of violation in r
where sr is the society in which SM is attached.
storage limits for ith reservoir during tth subinterval.
The step-wise procedure of CSO based hydrothermal scheduling
 Step9. Updating the personal best: The Pbest of leader, society
is described below.
leaders and society members are updated as below:

 Step1. Representation of the swarm: The swarm is represented m


)
Pbest ðk þ 1Þ ¼ Xm ðkÞ þ Ym ðk þ 1Þ; if wm ðk þ 1Þ < wm ðkÞ
as explained before. m m
 Step2. Initialization of the swarm: Each element of the swarm Pbest ðk þ 1Þ ¼ Pbest ðkÞ; ifwm ðk þ 1Þ P wm ðkÞ
matrix is created randomly from the feasible range of each var- ð22Þ
iable. The velocity matrix Y with the same size of swarm matrix m m
is initialized with zeros and the personal best positions are set where w (k) = w(X (k)).
to the present positions, i.e. Pbestm(k) = Xm (k) for m =
1, 2, . . . , M with k = 1.  Step10. Formation of new swarm: The swarm for next genera-
 Step3. Constraint handling: Adjustments are made in the swarm tion is initialized as an empty set, i.e. Swarm(k + 1) £. Then
to enforce hydraulic continuity and power balance as explained the leader of the swarm at kth iteration is moved to new posi-
earlier. For each individual Xm, w(Xm) is calculated using tion and included to the new swarm.
Eq. (16). XL ¼ LðkÞ þ YL ðk þ 1Þ; Swarmðk þ 1Þ
 Step4. Identification of society leaders: Pbest(k), Y(k) and X(k) of L
Swarmðk þ 1Þ [ X ð23Þ
Swarm(k) are arranged in the ascending order of w values. The
first Ns particles are selected as society leaders, i.e. Then the society leaders and the society members of the swarm at
SLðkÞ ¼ fSL1 ; SL2 ; . . . ; SLs ; . . . ; SLNs g. kth iteration are moved to new positions and included to the new
 Step5. Identification of society members: The remaining parti- swarm.
cles form a set of society members, i.e. SMðkÞ ¼ 9
s
XSL s
s ¼ SL ðkÞ þ Y ðk þ 1Þ; s ¼ 1; 2; . . . ; Ns =
fSM1 ; SM2 ; . . . ; SMr ; . . . ; SMNr g and the society members set n o ð24Þ
SL SL SL
has to satisfy the following condition: SM(k)  Swarm(k) such Swarmðk þ 1Þ Swarmðk þ 1Þ [ X1 ; X2 ; . . . ; XNs ;
that SM(k) \ SL(k) = £.
 Step6. Formation of societies: For each SMr e SM(k), r = 9
r
XSM r
r ¼ SM ðkÞ þ Y ðk þ 1Þ; r ¼ 1; 2;. .. ;N r
=
1, 2, . . . , Nr, the Euclidean distance between the society member n o ð25Þ
and the society leaders are determined using. Swarmðk þ 1Þ Swarmðk þ 1Þ [ XSM SM SM ;
1 ;X2 ; .. .; XN r

N hy þNth T
!12
X X
2
Ds ¼ SLsit ðkÞ  SMrit ðkÞ for s  Step11. Termination: If k – kmax then the procedure is repeated
i¼1 t¼1
from Step3 with k k + 1 else, the optimization process is ter-
¼ 1; 2; . . . ; Ns ð17Þ minated. The minimum objective function value is objective(k)
r and the corresponding hydrothermal schedule is given by L(k).
The society member SM is assigned to society ‘s’, if it is closer to
4. Test results and analysis
SLs.

For demonstrating the effectiveness of the proposed CSO algo-


 Step7. Identification of leader: The leader of the swarm is the
rithm in solving MSTHTS, two test systems with nonlinear and
society leader with minimum w value and belongs to the set
non-smooth characteristics are considered. The CSO algorithm
of society leaders, i.e. L(k) e SL(k). The objective function value
developed in Section 3 has been implemented using MATLAB soft-
is assigned as w(L(k)) if k = 1. For k – 1, it is updated by
ware package and the programs are executed on a Pentium IV,
3 GHz, 2-GB RAM computer. The control parameters of CSO, which
objectiv eðkÞ ¼ wðLðkÞÞ if wðLðkÞÞ < objectiv eðk  1Þ
are tuned experimentally, are listed in Table 1. In applying the
objectiv eðkÞ ¼ objectiv eðk  1Þ if wðLðkÞÞ P objectiv eðk  1Þ
CSO-based algorithm for the two test systems, the maximum num-
ð18Þ ber of iterations is set to 200. The Pareto-optimal front of the
 Step8. Updating particle velocities: The new velocities for differ- MSTHTS is determined using the proposed ideal guide based meth-
ent particles are calculated as below. od as outlined in Section 2.3.
For leader:
  4.1. Test system-1
L
Y Lit ðk þ 1Þ ¼ wY Lit ðkÞ þ C l r1 Pbestit ðkÞ  Lit ðkÞ ;
for 8i 2 ðNhy þ Nth Þ; 8t 2 T ð19Þ The test system-1, which has been adopted from [19], consists
of a multi-chain cascade of four hydro units and three thermal
For society leaders:
s

Y sit ðk þ 1Þ ¼ wY sit ðkÞ þ C sl1 r 2 Pbestit ðkÞ  SLsit ðkÞ Table 1


þ C sl2 r 3 Lit ðkÞ  SLsit ðkÞ Control parameters of CSO algorithm for the two test systems.

for 8i 2 ðNhy þ Nth Þ; 8t 2 T; 8s 2 N s ð20Þ Parameter Value


Swarm size (M) 40
For society members:
Number of societies (Ns) 4
Inertia weight wmax = 0.9, wmin = 0.4
Acceleration coefficients Cl = 2, Csl1 = 0.5, Csl2 = 1.5, Csm1 = 0.25, Csm2 = 0.75
184 A. Immanuel Selvakumar / Electrical Power and Energy Systems 51 (2013) 178–189

Table 2
Hourly plant discharge and hydrothermal generation schedule obtained by CSO for test system-1 (economic dispatch).

Hour Hourly plant discharge (104 m3) Hydro generation (MW) Thermal generation (MW)
q1 q2 q3 q4 PH1 PH2 PH3 PH4 PS1 PS2 PS3
1 5.5185 10.4295 22.0521 12.5277 57.8760 72.9450 30.4392 226.3467 20.0000 205.7161 136.6769
2 13.7927 11.5057 23.4628 6.0000 97.2571 74.8637 15.8585 149.5921 98.2464 204.4306 139.7515
3 6.3555 6.8719 16.1657 15.8695 63.9837 52.6083 45.8575 231.1398 36.8848 40.0042 229.5217
4 8.7687 6.6884 21.4940 15.5389 79.5580 52.8386 26.9286 207.9010 103.0133 40.0007 139.7598
5 5.6410 14.3268 20.0085 7.0206 57.9505 81.4827 34.6938 147.4234 102.6191 106.0709 139.7596
6 5.7029 8.2449 20.4032 10.3811 58.7539 57.4431 32.5544 199.4585 102.4398 209.7285 139.6218
7 9.9339 7.4102 16.0859 11.6810 85.4366 51.5736 47.4196 217.4380 20.0000 209.7731 318.3590
8 6.1753 10.7346 20.4217 11.5329 62.7619 65.6885 32.7972 226.5985 174.9119 40.0000 407.2420
9 10.7489 9.4060 17.3998 15.1890 89.7227 58.7141 45.1610 264.6799 102.6660 209.8062 319.2500
10 6.0716 11.1481 17.3788 15.2146 62.8053 64.4841 44.2599 270.5371 102.6763 126.1981 409.0392
11 8.4176 9.2724 13.0113 16.3897 80.3454 56.1918 54.1555 279.2464 101.0624 209.7313 319.2671
12 10.5638 7.2293 21.4546 16.9394 91.6048 45.5755 29.2974 287.0520 102.5562 185.1856 408.7284
13 9.8612 8.0811 17.1227 17.6223 88.6128 50.4949 48.9021 291.4915 102.2759 209.8153 318.4075
14 9.1091 6.2904 20.1175 10.7593 85.2667 41.4786 39.1729 237.3261 102.5300 294.7079 229.5178
15 8.0610 7.9367 10.9789 13.8898 79.3018 52.4840 56.3364 268.1179 20.0000 124.8670 408.8929
16 11.6410 7.3707 17.3197 18.6566 97.5077 49.5932 52.3147 307.6330 102.6740 124.9103 325.3670
17 6.7329 6.8866 17.9603 10.7725 69.7172 46.6580 50.0857 244.6842 20.0000 209.8158 409.0392
18 7.0802 7.7864 13.2998 17.9715 72.5097 50.8536 59.2068 312.1706 101.0580 294.6820 229.5194
19 9.1462 7.0776 12.2858 20.0000 85.9764 46.5291 59.4629 314.8571 106.9368 222.7147 233.5230
20 7.4158 6.1907 20.8033 13.9189 74.4825 42.0374 40.7809 273.8865 102.6704 209.4918 306.6504
21 5.8598 7.9407 19.1011 16.5182 62.3915 53.7662 47.7439 297.3124 102.3490 209.8114 136.6255
22 7.8638 6.9610 16.7173 18.3584 77.8410 49.2455 54.5890 304.9360 20.0000 124.1193 229.2692
23 6.7401 7.1310 12.5167 14.6240 69.8775 50.9135 58.5644 274.3621 57.4160 124.9079 213.9586
24 7.7983 9.0791 14.0289 18.6508 78.0354 61.2553 58.7647 306.5771 21.2275 44.6140 229.5259

Table 3
Hourly plant discharge and hydrothermal generation schedule obtained by CSO for test system-1 (emission dispatch).

Hour Hourly plant discharge (104 m3) Hydro generation (MW) Thermal generation (MW)
q1 q2 q3 q4 PH1 PH2 PH3 PH4 PS1 PS2 PS3
1 8.3456 7.8226 26.1623 6.0000 77.7423 61.0530 0.8328 159.0269 158.5617 162.3919 130.3914
2 11.1045 6.7621 22.4437 6.0000 90.3904 55.0796 20.9857 155.7437 158.9141 174.4488 124.4378
3 6.5237 6.1876 22.8197 6.0000 65.1912 52.7564 14.0052 151.6253 143.2680 157.8994 115.2545
4 9.3463 7.7708 24.5786 6.0000 82.3570 63.8772 0.0020 145.8221 139.3635 121.2474 97.3308
5 7.5531 6.1891 24.0475 6.0000 71.3713 54.4290 0.0004 164.4665 133.9841 144.5717 101.1771
6 6.8455 6.3289 21.1905 6.0000 66.5576 55.7741 17.6353 177.8664 170.0777 186.2262 125.8627
7 7.2860 8.1670 18.9713 9.1531 69.8466 66.4366 28.3292 230.1327 175.0000 217.6869 162.5680
8 9.6518 7.3354 17.0714 12.0021 83.3892 61.1448 34.5300 273.2693 174.9999 212.3749 170.2919
9 7.4578 8.4603 18.6341 14.8733 71.5593 67.6726 26.9729 309.9106 175.0000 249.4537 189.4308
10 7.0526 6.9225 13.8473 13.9584 69.7728 59.3550 42.0388 306.3582 175.0000 250.0052 177.4700
11 9.0936 8.3947 17.4162 14.8622 83.5379 68.8412 33.1032 318.4056 175.0000 239.2539 181.8582
12 8.0725 12.4356 15.4749 17.0236 77.8766 84.4391 39.4035 338.2019 174.9986 241.8956 193.1847
13 8.7035 6.7483 15.4413 18.5016 82.3458 56.7553 41.2971 350.1573 175.0000 232.8786 171.5659
14 8.4917 9.2823 17.9812 15.7110 81.7817 71.6822 34.6040 325.2441 174.9972 202.7859 138.9048
15 7.4316 9.8615 12.7287 14.3989 75.1522 74.0574 49.1231 314.7376 172.3223 195.1148 129.4927
16 12.0503 8.5687 13.7052 18.8417 99.3318 66.9645 49.6598 350.8395 169.8173 185.5877 137.7995
17 7.2277 7.5618 15.4354 19.5754 73.6112 60.6878 48.3825 352.3990 174.7676 190.8808 149.2711
18 9.9311 8.4086 14.1253 19.9915 90.2951 64.2303 53.1508 353.3218 174.9972 225.0567 158.9481
19 9.6571 11.4355 12.1896 19.9973 88.2514 75.5243 54.9250 346.5713 172.6679 175.3437 156.7163
20 10.3545 9.9766 12.9866 19.7789 90.5199 68.0139 56.2909 339.3213 170.8971 188.7974 136.1594
21 5.6464 9.6158 16.5421 19.9993 59.9116 65.8791 53.1290 336.0782 145.6408 145.4990 103.8622
22 5.5506 7.1943 17.8339 19.9991 59.4084 53.6330 51.4616 329.9856 138.4817 140.7667 86.2631
23 5.0167 11.0602 15.7827 19.5287 54.9838 71.3040 56.2839 319.3960 130.4670 125.0270 92.5382
24 6.6058 9.5100 12.5037 19.8547 69.0313 63.3985 58.8864 313.6117 111.6150 104.8571 78.6000

units. The schedule horizon is 1 day and there are twenty-four 1-h [23], PSO [24], QADEVT [25] and MODE [29] in three cases: eco-
intervals. Water transport delays of 1–2 h between hydro plants nomic dispatch, emission dispatch, and combined economic and
are considered. The fuel cost functions of thermal units are non- emission dispatch (CEED). Tables 2–4 present the optimal water
smooth due to valve-point effect, and the emission release discharge schedules, optimal power output of hydro generators
functions are expressed as the sum of quadratic and exponential and thermal generators for different time intervals of scheduling
functions. Complete data of this system can be found in [19] and horizon obtained by the proposed CSO for three dispatch cases.
it is not repeated here. The optimal storage volume levels are depicted in Figs. 2–4.
The performance of the proposed CSO is verified on the above- The optimal fuel cost and emission release results obtained for
mentioned system with water discharge rates and thermal gener- test system-1 using fuzzy EP, MDE, QPSO, PSO, QADEVT, MODE and
ations as decision variables. The electric losses are not considered the proposed CSO are presented in Table 5. The results are tabu-
in this case. CSO is compared with fuzzy EP [19], MDE [22], QPSO lated for three dispatch cases. In economic dispatch, the objective
A. Immanuel Selvakumar / Electrical Power and Energy Systems 51 (2013) 178–189 185

Table 4
Hourly plant discharge and hydrothermal generation schedule obtained by CSO for test system-1 (combined economic & emission dispatch).

Hour Hourly plant discharge (104m3) Hydro generation (MW) Thermal generation (MW)
q1 q2 q3 q4 PH1 PH2 PH3 PH4 PS1 PS2 PS3
1 9.4973 7.0547 25.0838 6.0388 83.7540 56.7239 9.6337 159.5125 175.0000 125.6158 139.7601
2 7.0437 14.4070 21.0656 6.1138 69.3720 83.4559 29.7215 157.1155 102.8938 124.9079 212.5333
3 10.7495 6.5699 22.3590 7.5534 89.1002 51.0975 18.7241 169.7096 106.6834 124.9249 139.7603
4 7.1339 7.8494 20.9307 6.0000 69.2487 59.9286 24.7113 144.1327 87.6975 124.7367 139.5445
5 9.1998 7.2179 20.3885 8.3767 80.7773 56.5287 30.7399 190.1670 103.7940 124.9092 83.0839
6 10.5387 12.1270 22.4564 13.7110 85.4462 76.8392 17.3170 253.2412 175.0000 137.2457 54.9108
7 6.8908 6.9192 19.5099 7.4840 65.8688 50.7543 33.0862 198.5274 102.6736 269.8115 229.2782
8 6.8472 6.9356 15.6447 14.2300 66.2229 50.9025 46.9806 279.5260 127.9641 209.8150 228.5888
9 11.3149 6.3381 17.1238 17.2867 89.1246 48.0243 44.3826 307.7196 164.6550 209.7612 226.3327
10 8.4884 6.8070 15.8674 14.5045 77.2053 52.5506 47.1281 292.1589 174.9993 209.8148 226.1430
11 5.7128 10.4851 15.5332 18.2454 59.3345 71.6240 49.0163 323.7314 159.1503 209.8136 227.3299
12 9.9867 14.1192 14.4275 15.6347 87.1775 78.6920 51.4346 303.3989 175.0000 224.8341 229.4629
13 9.4640 8.7871 15.7903 16.1834 85.0684 58.5782 49.5156 308.8905 170.6947 209.7619 227.4907
14 9.2705 7.2435 12.3692 17.3093 84.8193 50.9744 55.5226 316.4542 173.3577 209.1123 139.7596
15 9.1125 8.2782 19.7380 17.4985 84.4378 57.5799 43.1579 315.9714 170.9421 198.1513 139.7597
16 6.7383 9.1170 15.0900 11.4711 69.2277 61.2649 56.8878 262.6024 174.9548 209.6775 225.3849
17 8.7499 7.0981 12.8056 15.6518 83.1312 49.7268 59.1982 304.1155 115.1471 209.8158 228.8655
18 7.4477 8.2612 15.4016 18.9292 74.6369 55.0130 58.0790 322.3524 174.9994 209.8148 225.1046
19 5.8954 6.3130 15.8641 15.8921 62.6417 43.5304 58.0469 303.6710 173.7288 288.6393 139.7419
20 9.3602 7.7646 11.4425 15.3197 86.2630 52.8395 59.6427 298.5605 175.0000 209.8157 167.8786
21 6.4506 7.9329 18.0999 14.8624 66.8350 54.5740 53.5793 292.6363 102.6735 204.5843 135.1177
22 6.2682 8.5169 20.8541 18.7193 65.6128 58.1910 42.5225 319.1812 108.2333 126.4995 139.7598
23 5.6561 8.5224 19.7479 19.0062 60.7996 57.8492 45.9620 317.7020 102.9226 125.0048 139.7598
24 7.1831 7.3350 14.4931 19.6694 73.5393 51.4436 58.4507 312.5878 95.6918 124.8853 83.4016

200
200 Reservoir 1
Reservoir 1 Reservoir 2
180 Reservoir 3
180 Reservoir 2
Hydro Storage (x 104 m3)

Reservoir 4
Hydro Storage (x 104 m3)

Reservoir 3
Reservoir 4 160
160
140
140
120
120
100
100
80
80
60
60 0 2 4 6 8 10 12 14 16 18 20 22 24
0 2 4 6 8 10 12 14 16 18 20 22 24
Hour
Hour
Fig. 3. Hydro reservoir storage trajectories obtained by CSO for test system-1
Fig. 2. Hydro reservoir storage trajectories obtained by CSO for test system-1 (emission dispatch).
(economic dispatch).

is to minimize only the cost without considering emission release. 200


For the economic dispatch, fuzzy EP is inefficient as it is able to Reservoir 1
180 Reservoir 2
achieve the hydrothermal schedule with the highest cost. Among
Hydro Storage (x 104 m3)

Reservoir 3
the previous methods, MODE [29] is able to obtain cheaper sche- Reservoir 4
160
dule than Fuzzy EP, MDE, QPSO, PSO and QADEVT. Nevertheless,
the proposed CSO method finds the cheaper hydrothermal sche- 140
dule than MODE. Among the various methods, CSO is the best per-
former; it is able to find the hydrothermal schedule with minimum 120
thermal generation cost. For emission dispatch case, all the algo-
rithms try to find the hydrothermal schedule that minimizes the 100
emission release. As seen from the result, MDE [22] is able to find
80
the schedule with minimum emission release when compared to
all the previous methods. However, the proposed CSO outperforms 60
MDE and it is able to schedule the hydrothermal system with the 0 2 4 6 8 10 12 14 16 18 20 22 24
minimum emission release than that of MDE. Hour
To find out the Pareto-optimal front in the case of CEED, a set of
eight ideal guides, i.e., points (more ideal guides can be used to get Fig. 4. Hydro reservoir storage trajectories obtained by CSO for test system-1
(combined economic & emission dispatch).
more Pareto-optimal points) are generated along the FCmin and
186 A. Immanuel Selvakumar / Electrical Power and Energy Systems 51 (2013) 178–189

Table 5
Comparison of optimal fuel cost and emission release for test sytem-1.

Methods Economic dispatch Emission dispatch Combined economic & emission dispatch
FC ($) ER (lb) FC ($) ER (lb) FC ($) ER (lb)
Fuzzy EP [19] 45063.0000 48797.0000 59288.0000 16554.0000 47906.0000 26234.0000
MDE [22] 42611.0000 33323.0000 48714.0000 15730.0000 43198.0000 20385.0000
QPSO [23] 42359.0000 31298.0000 45271.0000 17767.0000 44259.0000 18229.0000
PSO [24] 42474.0000 28132.0000 48263.0000 16928.0000 43280.0000 17889.0000
QADEVT [25] 42587.0000 30786.0000 46100.0000 17535.0000 43395.0000 18324.0000
MODE [29] 42051.0000 17861.0000 45040.0000 16091.0000 43279.0000 16603.0000
CSO 41323.5462 28069.9261 46375.6942 13617.8373 41515.4360 16180.0037

4
3
x 10 Pareto-optimal fronts obtained by the proposed CSO and MODE
MODE [29] algorithms are depicted in Fig. 5. There are two goals in a mul-
2.8 CSO
tiobjective optimization: (1) To find a set of solutions as close as
2.6 possible to the true Pareto optimal-front. (2) To find a set of solu-
Emission Release (lb)

tions as diverse as possible along the true Pareto optimal-front.


2.4 Many performance metrics are available to check whether an opti-
2.2 mization algorithm finds a Pareto-optimal front that satisfies the
abovementioned goals [35]. Following three metrics are used in
2 this paper to measure the quality of the Pareto-optimal fronts ob-
1.8 tained by various algorithms:

1.6
1. Spread: Measure of equal spacing of Pareto-optimal solu-
1.4 tions along the Pareto-optimal front. As long as the spread
is uniform this metric produces a small value.
1.2
4.1 4.2 4.3 4.4 4.5 4.6 4.7 2. Maximum Spread: For two objective problems, this metric
Fuel Cost ($) x 10
4 refers to the normalized Euclidean distance between the
two extreme solutions in the objective space. A high value
Fig. 5. Pareto-optimal fronts obtained with the Proposed CSO and MODE for test of this metric indicates that the Pareto front has the maxi-
system-1. mum spread, i.e., maximum diversity.
3. Hypervolume Ratio: This metric calculates the normalized
volume (area in the case of two objectives) in the objective
Table 6 space covered by the members of the Pareto-optimal front
Comparison of Pareto optimal front quality. and it captures in one scalar both the closeness of the solu-
Test system Algorithm Spread Maximum Hypervolume tions to the true Pareto set and their spread across objective
spread ratio space. It is said that whenever a set of Pareto-optimal solu-
1 MODE [29] 0.8482 0.4272 0.6664 tions is better than another set, its hypervolume ratio value
CSO 0.4227 1.0000 0.9015 is higher than the latter one. In this paper, the ideal Pareto
2 IGAMU [21] 0.5335 0.7764 0.8759 front (FCmin and ERmin axes) is taken as the reference.
CSO 0.1551 1.0000 0.9087 Therefore, the area of the rectangle with the opposite cor-
ners ideal objective vector (FCmin, ERmin) and Nadir point
FCmax, ERmax is considered for normalization.
ERmin axes. By minimizing the GCF (Eq. (3)) with different ideal
guides by the CSO, the entire Pareto-optimal points have been Table 6 gives the quality measure of the Pareto fronts obtained
determined one by one. To find out the best compromise solution, by the proposed CSO and MODE [29] algorithms. For the Pareto-
the ideal objective vector ({FCmin, ERmin}) is the ideal guide. The optimal front obtained by CSO, the spread is less than that of

Table 7
Hydrothermal generation schedule obtained by CSO for test system-2.

Dispatch case Subinterval Hydro generation (MW) Thermal generation (MW) PLoss (MW)
PH1 PH2 PS1 PS2 PS3 PS4
Economic dispatch 0–12.0 h 213.6726 307.3253 20.2780 112.6808 124.8358 139.7621 18.5547
12–24.0 h 202.0079 439.3017 20.1306 112.6722 124.9088 229.5261 28.5474
24–36.0 h 228.0377 389.9071 22.1596 113.9869 40.1854 229.5230 23.7998
36–48.0 h 234.8313 448.3498 100.5895 113.4436 210.1807 230.5112 37.9062
Emission dispatch 0–12.0 h 169.8881 312.5052 73.1042 134.5989 136.0926 91.9459 18.1349
12–24.0 h 247.3758 408.1579 80.3410 143.1200 148.1845 101.0486 28.2278
24–36.0 h 208.4863 360.1620 76.7099 138.8790 142.1583 96.4810 22.8763
36–48.0 h 250.0000 500.0000 104.0032 169.1280 185.5542 130.8783 39.5638
Combined economic & emission dispatch 0–12.0 h 156.5514 364.2755 20.8507 112.6736 124.9079 139.7598 19.0189
12–24.0 h 233.9132 417.9972 98.5398 112.8357 124.9159 139.7598 27.9615
24–36.0 h 234.4996 311.3782 98.5398 112.8970 124.9780 139.7598 22.0525
36–48.0 h 249.9905 488.4033 98.5398 152.7930 209.8158 139.7598 39.3022
A. Immanuel Selvakumar / Electrical Power and Energy Systems 51 (2013) 178–189 187

Table 8
Comparison of optimal fuel cost and emission release for test sytem-2.

Methods Economic dispatch Emission dispatch Combined economic & emission dispatch
FC ($) ER (lb) FC ($) ER (lb) FC ($) ER (lb)
SAGA [20] 70718.4000 37778.0000 89094.0000 23200.8000 73612.8000 26080.8000
IGAMU [21] 66539.5567 29503.0270 89316.9316 23223.2653 68492.9082 26080.8000
CSO 66084.7800 36016.9600 89275.9908 23222.6815 67136.2886 25550.2527

Table 9
Comparison among SAGA, IGAMU and CSO for constraint violation in emission dispatch (test sytem-2).

Methods Load balance constraint violation Water availability constraint violation SCV
Subinterval 1 Subinterval 2 Subinterval 3 Subinterval 4 Hydro plant1 Hydro plant2
SAGA [20] 0.2 0.3 0.2 0.4 4.9 5.0 11
IGAMU [21] 0.0 0.0 0.0 0.0 0.0 0.0 0
CSO 0.0 0.0 0.0 0.0 0.0 0.0 0

4
x 10 Table 12
3.8
IGAMU Savings in fuel cost and reduction in emission release obtained by the proposed CSO.
CSO
3.6 Test Economic Emission Combined economic
system dispatch dispatch & emission dispatch
Emission Release (lb)

3.4
Savings in Reduction in Savings in Reduction in
3.2 fuel cost ($) emission (lb) fuel cost ($) emission (lb)
1 727.4500 2112.1627 1682.5639 423.0000
3 2 454.7767 0.5838 1356.6196 530.5473

2.8

2.6
observed from Table 5, that the best compromise solution obtained
2.4 by the proposed CSO algorithm dominates that of all the other
methods. It is clear that the proposed CSO method is efficient in
2.2 locating the cheapest hydrothermal schedule with less emission
6.5 7 7.5 8 8.5 9
x 10
4 in the multi-minima environment and gives the best Pareto
Fuel Cost ($)
front.
Fig. 6. Pareto-optimal fronts obtained with the Proposed CSO and IGAMU for test
system-2. 4.2. Test system-2

To further demonstrate the robustness of the proposed CSO, it is


Table 10 applied to another test system comprising two hydro plants and
Comparison between IGAMU and CSO for combined economic and emission dispatch four thermal plants [20]. The scheduling horizon of this system is
after 50-trials (test sytem-2). divided into four subintervals and involves four subinterval de-
Methods FC ($) CPU time (s) mands; the duration of each subinterval is 12 h. B-matrix loss for-
mula is utilized to find the network losses in each interval. To make
Minimum Mean Maximum Mean Equalized
mean
the thermal model more practical, the fuel-cost function is in-
cluded with valve-point effect and the emission function is in-
IGAMU [21] 68492.9082 70949.1614 72389.6900 51.72 12.93
CSO 67136.2886 68041.3673 69958.8524 9.02 9.02
cluded with exponential component.
The proposed CSO has been applied to this system with hydro
and thermal generations as decision variables for economic dis-
patch, emission dispatch and CEED cases. The constraints included
are demand–supply balance, bounds on hydro generations, bounds
Table 11
Relative frequencies of convergence for combined economic & emission dispatch on thermal generations and water availability. Simulated anneal-
using IGAMU and CSO (test sytem-2). ing based goal-attainment (SAGA) method [20] and improved GA
with multiplier updating (IGAMU) [21] have already been used
Methods Range of cost ($)
for solving this problem.
67000– 69500– 70500– 71500–
Table 7 summarizes the best generation outputs of the hydro
69500 70500 71500 72500
and thermal plants obtained by CSO. The results are presented
IGAMU [21] 5 9 18 18
for three dispatch cases along with the transmission losses of each
CSO 47 3 0 0
subinterval. The minimum fuel costs and the emissions deter-
mined by SAGA method [20], IGAMU [21], and CSO are compared
in Table 8.
MODE; the maximum spread and the Hypervolume Ratio are For assessing the feasibility of these three methods, the sum of
greater than those of MODE. It indicates that the Pareto-optimal equality constraint violations (SCVs) is calculated and the results
front of CSO is better than that of MODE in satisfying the two goals are given in Table 9. Although the SAGA provides minimum fuel
of multiobjective optimization (closeness and diversity). It can be cost and emission in the emission dispatch case, the solution is
188 A. Immanuel Selvakumar / Electrical Power and Energy Systems 51 (2013) 178–189

infeasible as seen from Table 9. Yet, IGAMU and CSO provide feasi- In addition, CSO offers considerable savings in fuel cost and
ble solutions and it is found that the proposed CSO provides better reduction in emission release. Hence, the proposed CSO is sug-
fuel cost as well as emission in all the three dispatch cases. For gested as an efficient tool for MSTHTS and can be applied for other
CEED, the Pareto fronts obtained by IGAMU and the proposed optimization problems.
CSO are shown in Fig. 6 and the quality of the Pareto fronts are
compared in Table 6. It is clear that the CSO gives a better Pareto
front than IGAMU. References
To investigate the effects of initial trial solutions on the results
and success rates, this test system is solved with different initial [1] Wood AJ, Wollenberg BF. Power generation, operation, and control. 2nd
ed. New York: Wiley; 1996.
trial solutions by CSO for CEED.
[2] Engles L, Larson RE, Peschon J, Stanton KN. Dynamic programming applied to
Table 10 compares the minimum, mean and maximum fuel hydro and thermal generation scheduling. In: IEEE tutorial course text,
costs of the hydrothermal schedules determined by IGAMU and 76CH1107-PWR, New York, NY; 1976.
[3] Yang JS, Chen N. Short term hydrothermal coordination using multi-pass
CSO in 50 executions. The average fuel cost obtained by CSO is $
dynamic programming. IEEE Trans Power Syst 1989;4(3):1050–6.
68041.3673, which is even smaller than the minimum fuel cost [4] Tang J, Luh PB. Hydrothermal scheduling via extended differential dynamic
of IGAMU. From Table 10, it is evident that the optimality of pro- programming and mixed coordination. IEEE Trans Power Syst
posed CSO is superior to that of the IGAMU in terms of the statis- 1995;10(4):2021–8.
[5] Brannud H, Bubenko JA, Sjelvgren D. Optimal short term operation planning of
tical minimum, mean and maximum fuel costs obtained. To a large hydrothermal power system based on a nonlinear network flow
compare the computational efficiency, the equalized CPU time concept. IEEE Trans Power Syst 1986;1(4):75–82.
(keq  actual CPU time) is taken as the performance measure. The [6] Xia Q, Xiang N, Wang S, Zhang B, Huang M. Optimal daily scheduling of
cascaded plants using a new algorithm of non-linear minimum cost network
equalizing factor (keq) is the ratio of actual CPU speed (GHz) to flow concept. IEEE Trans Power Syst 1988;3(3):929–35.
the speed of CPU (GHz) in which CSO algorithm is executed. The [7] Pereira MVF, Pinto LMVG. Application of decomposition techniques to the mid
advantage of measuring computational effort by the equalized and short term scheduling of hydrothermal systems. IEEE Trans Power
Apparatus Syst 1983;102(11):3611–8.
CPU time is that the added benefit in computational effort due to [8] Wong KP, Wong YW. Short-term hydrothermal scheduling part: I. Simulated
the processor speed is nullified in the comparison. This measure, annealing approach. IEE Proc Gener Transm Distrib 1994;141(5):497–501.
thus, provides a good indication of the relative ranking of the algo- [9] Chen P-H, Chang H-C. Genetic aided scheduling of hydraulically coupled
plants in hydro-thermal coordination. IEEE Trans Power Syst 1996;11(2):
rithms. From the mean CPU time (actual and equalized) compari- 975–81.
son in Table 10, it can be noticed that the CSO method needs less [10] Orero SO, Irving MR. A genetic algorithm modeling framework and solution
computational effort and they are significantly faster than IGAMU. technique for short term optimal hydrothermal scheduling. IEEE Trans Power
Syst 1998;13(2):501–18.
Evidently, the proposed CSO approach is powerful in achieving glo-
[11] Hota PK, Chakrabarti R, Chattopadhyay PK. Short-term hydrothermal
bal optimum solution with less execution time. Out of 50 runs with scheduling through evolutionary programming technique. Electr Power Syst
50 different initial trial solutions, the relative frequencies of attain- Res 1999;52:189–96.
ing a scheduling cost within specific ranges are presented in Table [12] Naresh R, Sharma J. Two-phase neural network based solution technique for
short-term hydrothermal scheduling. IEE Proc Gener Transm Distrib
11. Between the two methods, CSO has the highest success rate of 1999;146(6):657–63.
94% in converging into the lower solution range ($ 67,000–$ [13] Basu M. Hopfield neural networks for optimal scheduling of fixed head
69,500), whereas IGAMU has a success rate of only 10%. This con- hydrothermal power systems. Electr Power Syst Res 2003;64:11–5.
[14] Yang P-C, Yang H-T, Huang C-L. Scheduling short-term hydrothermal
firms that the CSO is the most reliable. generation using evolutionary programming technique. IEE Proc Gener
To highlight the superiority of the proposed CSO in scheduling Transm Distrib 1996;143(4):371–6.
hydrothermal systems, savings in fuel cost (difference between [15] Sinha N, Chakrabarti R, Chattopadhyay PK. Fast evolutionary programming
techniques for short-term hydrothermal scheduling. IEEE Trans Power Syst
previous best cost and the cost obtained by CSO) and reduction 2003;18(1):214–20.
in emission (difference between previous best emission and the [16] Swain RK, Barisal AK, Hota PK, Chakrabarti R. Short-term hydrothermal
emission obtained by CSO) are calculated for three dispatch cases scheduling using clonal selection algorithm. Electr Power Energy Syst
2011;33:647–56.
of the two test systems. From the results summarized in Table [17] Wang Y, Zhou J, Mo L, Ouyang S, Zhang Y. A clonal real-coded quantum-
12, it can be observed that if the CSO approach is employed to inspired evolutionary algorithm with Cauchy mutation for short-term
schedule hydrothermal systems, considerable amount of money hydrothermal scheduling. Electr Power Energy Syst 2012;43:1228–40.
[18] IEEE Current Operating Problems Working Group. Potential impacts of clean
can be saved and significant amount of emission can be reduced.
air regulations on system operations. IEEE Trans Power Syst
1995;10(2):647–56.
5. Conclusion [19] Basu M. An interactive fuzzy satisfying method based on evolutionary
programming technique for multiobjective short-term hydrothermal
scheduling. Electr Power Syst Res 2004;69:277–85.
In this paper, civilized swarm optimization (CSO) technique is [20] Basu M. A simulated annealing-based goal-attainment method for economic
presented to solve multiobjective short-term hydrothermal sched- emission load dispatch of fixed head hydrothermal power systems. Electr
uling problem (MSTHTS) with cost and emission objectives. The pro- Power Energy Syst 2005;27:147–53.
[21] Chiang C-H. Optimal economic emission dispatch of hydrothermal power
posed hybrid swarm intelligence technique CSO has been developed systems. Electr Power Energy Syst 2007;29:462–9.
by embedding the society–civilization concept into the classical par- [22] Lakshminarasimman L, Subramanian S. Short-term scheduling of
ticle swarm optimization (PSO). Two short-term hydrothermal hydrothermal power system with cascaded reservoirs by using modified
differential evolution. IEE Proc Gener Transm Distrib 2006;153(6):693–700.
scheduling problems have been solved with economy, emission, [23] Sun C, Lu S. Short-term combined economic emission hydrothermal
and combined economy–emission objectives by CSO. This paper scheduling using improved quantum-behaved particle swarm optimization.
suggests a new ideal guide based approach, which is used along with Expert Syst Appl 2010;37:4232–41.
[24] Mandal KK, Chakraborty N. Short-term combined economic emission
the CSO to obtain the Pareto-optimal front. Numerical results from scheduling of hydrothermal systems with cascaded reservoirs using particle
the two test systems and comparative analysis with previous ap- swarm optimization technique. Appl Soft Comput J 2011;11:1295–302.
proaches indicate the following advantages of CSO: [25] Lu S, Sun C. Quadratic approximation based differential evolution with
valuable trade off approach for bi-objective short-term hydrothermal
scheduling. Expert Syst Appl 2011;38:13950–60.
 Perfect balance between global and local search. [26] Gjorgiev B, Cepin M. A multi-objective optimization based solution for the
 Ability to produce highly optimal hydrothermal schedule in combined economic–environmental power dispatch problem. Eng Appl Artif
more robust manner with less computational time than the pre- Intell 2013;26:417–29.
[27] Basu M. Economic environmental dispatch of fixed head hydrothermal power
vious approaches. systems using nondominated sorting genetic algorithm-II. Appl Soft Comput J
 Ability to obtain better Pareto-optimal fronts. 2011;11:3046–55.
A. Immanuel Selvakumar / Electrical Power and Energy Systems 51 (2013) 178–189 189

[28] Zhang R, Zhou J, Wang Y. Multi-objective optimization of hydrothermal energy [32] Ray T, Liew KM. Society and civilization: an optimization algorithm based on
system considering economic and environmental aspects. Electr Power Energy the simulation of social behavior. IEEE Trans Evol Comput 2003;7(4):386–96.
Syst 2012;42:384–95. [33] Kennedy J, Eberhart R. Particle swarm optimization. In: Proceedings of the IEEE
[29] Zhang H, Zhou J, Zhang Y, Fang N, Zhang R. Short term hydrothermal international conference on neural networks; 1995. p. 1942–8.
scheduling using multi-objective differential evolution with three chaotic [34] AlRashidi MR, El-Hawary ME. A survey of particle swarm optimization
sequences. Electr Power Energy Syst 2013;47:85–99. applications in electric power systems. IEEE Trans Evol Comput
[30] Osyczka A. Multicriterion optimization in engineering with FORTRAN 2009;13(4):913–8.
programs. New York: Halsted Press; 1984. [35] Deb K. Multi-objective optimization using evolutionary algorithms. 1st
[31] Immanuel Selvakumar A, Thanushkodi K. Optimization using civilized swarm: ed. England: John Wiley & Sons, Ltd.; 2001.
solution to economic dispatch with multiple minima. Electr Power Syst Res
2009;79(1):8–16.

You might also like