Electric Power Systems Research 123 (2015) 192–204

Contents lists available at ScienceDirect

Electric Power Systems Research

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

Evaluating the effectiveness of normal boundary intersection method

for short-term environmental/economic hydrothermal
self-scheduling
Abdollah Ahmadi a , Amirabbas Kaymanesh b , Pierluigi Siano c,∗ ,
Mohammadreza Janghorbani d , Ali Esmaeel Nezhad e , Debora Sarno c
a
School of Electrical Engineering and Telecommunications, The University of New South Wales, Sydney, Australia
b
Electrical Engineering Department, South Tehran Branch, Islamic Azad University, Tehran, Iran
c
Department of Industrial Engineering, University of Salerno, Fisciano, Italy
d
Young Researchers and Elite Club, Central Tehran Branch, Islamic Azad University, Tehran, Iran
e
Department of Electrical Engineering, Marvdasht Branch, Islamic Azad University, Marvdasht, Iran

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

Article history: The problem of the optimal scheduling of available hydro and thermal generating units considering a
Received 20 September 2014 short scheduling period (one day-one week) in order to maximize the total profit is denoted as short-term
Received in revised form 12 January 2015 hydro thermal self-scheduling (SHTSS). Mixed-integer linear programming (MILP) method is proposed
Accepted 14 February 2015
to model the SHTSS problem in the day-ahead energy and reserve markets. MILP formulation allows
Available online 9 March 2015
for considering a precise model for the prohibited working zones, dynamic ramp rate constraints and
operating services of thermal generating units, as well as the characteristics of multi-head power dis-
Keywords:
charge for hydro generating units and reservoirs’ spillage. This problem is modelled as a multi-objective
Short-term hydrothermal self-scheduling
Normal boundary intersection method
(MO) optimization one, having two objectives, i.e. maximization of the profit of Generation Company’s
Fuzzy decision maker (GENCO’s) and minimization of emissions from thermal units. In order to solve the problem and generate
Precise modelling the non-inferior solutions, normal boundary intersection (NBI) method is applied. The main advantage is
Emission the provision of set of uniformly distributed non-dominated solutions regardless of the scales of objective
Multi-Objective Optimization functions values. Then, a fuzzy based decision maker is employed in order to select a non-inferior solu-
tion. In order to demonstrate the effectiveness of the presented method, several numerical simulations
are presented. Furthermore, the obtained results are compared with those obtained considering different
methods for obtaining non-inferior solutions, such as weighted sum method, evolutionary programming-
based interactive fuzzy satisfying method, differential evolution, particle swarm optimization and hybrid
multi-objective cultural algorithm.
© 2015 Elsevier B.V. All rights reserved.

1. Introduction In this regard, the application of the mixed-integer linear pro-

Determining the optimal schedule of available generating units
(hydro and thermal generators) over a scheduling period (one
day–one week) that minimizes the total operating costs is defined
as short-term hydro thermal scheduling (SHTS) [1,2]. On the con-
trary, in short-term hydro thermal self-scheduling (SHTSS), hydro
and thermal generating units are used so that the total profit is max-
imized [3,4]. Different optimization methods for the SHTS problem
have been presented in literature.
∗ Corresponding author. Tel.: +39 089964294.
E-mail address: psiano@unisa.it (P. Siano).
gramming (MILP) method for solving SHTSS problems in day-ahead
market has been proposed in [3,4] in which a deterministic opti-
mization framework is used to model the problem and the expected
profit is maximized employing 0/1 MILP technique. In order to pre-
cisely model hydro units, the characteristics of multi-head power
discharge relating to hydro generating units have been presented in
[4,5] while MILP method has been utilized to maximize the profit.
Mandal and Chakraborty [6] has studied the control parameters for
optimal hydro thermal scheduling based on differential evolution.
The SHTS problem has been solved in [7] using teaching–learning
based optimization method while nonlinearities of hydro units’
reservoirs have been considered. In addition, Liao et al. [8] has
solved the SHTS problem employing adaptive bee colony method
and chaotic exploration to efficiently avoid a local optimum. A

http://dx.doi.org/10.1016/j.epsr.2015.02.007
0378-7796/© 2015 Elsevier B.V. All rights reserved.
 A. Ahmadi et al. / Electric Power Systems Research 123 (2015) 192–204 193

Nomenclature
pn,j minimum power relating to the jth hydro generat-
Indices -
ing unit for the nth performance curve (MW)
i thermal generating units index p̄j jth hydro generating unit’s capacity (MW)
j hydro generating unit index
pdn,i lower bound of the prohibited operating zone n per-
t time interval (hour) index
taining to the ith thermal generating unit (MW)
Constants pun−1,i upper bound of the prohibited operating zone n − 1
tb bilateral contract price at time t ($/MWh) pertaining to the ith thermal generating unit (MW)
tns market price of non-spinning reserve ($/MWh) Qj minimum water discharge relating to the jth hydro
-
ts market price for energy at time t ($/MWh) generating unit, provided that it is on (m3 /s)
tsr market price of spinning reserve at time t ($/MWh) Q̄n,j maximum water discharge relating to the nth block
 conversion factor which is 3.6*10−3 (Hm3 s/m3 h) of the jth hydro generating unit (m3 /s)
 number of time steps of the scheduling period RDLn,i ramp down limit relating to the nth block (MW)
- j,t lower bound of water discharge pertaining to the RULn,i ramp up limit relating to the nth block (MW)
hydro generating unit j at time t (m3 /s) s0 i number of time periods during which the ith ther-
¯ j,t upper bound of water discharge pertaining to the mal generating unit was shut down at the start of
hydro generating unit j at time t (m3 /s) the planning horizon (h)
 ij time interval among reservoir of hydro generating s̄j maximum spillage relating to the jth hydro gener-
unit i and hydro generating unit j (h) ating unit (m3 /s)
Ai cost of shut down pertaining to the thermal gener- smax,i maximum hours that the ith thermal generating
ating unit i ($) unit can be off (h)
Aj cost of start-up pertaining to the hydro generating SUEi emission generation of the ith thermal generating
unit j ($) unit when started-up (lb)
bn,i nth block’s slope pertaining to the cost curve of the SDEi emission generation of the ith thermal generating
thermal generating unit i ($/MWh) unit when shut down (lb)
bn ,j slope of the block n pertaining to the reservoir of the SDi limit of shut down ramp rate relating to the ith ther-
hydro generating unit j (m3 /s/Hm3 ) mal generating unit (MW/h)
k
bn,j nth block’s slope relating to the performance curve SUi limit of start-up ramp rate relating to the ith thermal
k of hydro generating unit j(MW/m3 /s) generating unit (MW/h)
ben,i nth segment’s slope relating to the emission curve UTi minimum up time relating to the ith thermal gener-
of the thermal generating unit i (lb/MWh) ating unit (h)
D objective function of the NBI method U0 i periods during which the ith thermal generating
emin,i minimum emission generated by thermal generat- unit has been on-line at the start of the scheduling
ing unit i (lb) period (h)
EGR emission group including SO2 and NOx v0,j minimum storage volume of the reservoir pertain-
E(pun−1,i ) emission generation relating to the n − 1th supe- ing to the jth hydro generating unit (Hm3 )
v0j reservoir at the beginning of the scheduling period
rior limit in emission curve belonging to the thermal
generating unit i (lb) (Hm3 )
F(pun−1,i ) generation cost relating to the n − 1th superior limit v
j
reservoir at the conclusion of the scheduling period
in cost curve belonging to the thermal generating (Hm3 )
unit i ($/h) vn,j maximum storage volume of the jth reservoir per-
Fj,t the predicted value for the natural water inflow taining to the variable head n (Hm3 )
relating to the reservoir of the hydro generating unit
j at time t (Hm3 /h) Variables
K i cost relating to the discrete interval  of the start- ˇ weighting factor in NBI method
up cost pertaining to the thermal generating unit i ˇn,i,t binary variable which is equal to 1 when the nth
($/h) block of cost curve pertaining to the ith thermal
I0 i ith thermal generating unit’s initial status (0/1) generating unit at time t is chosen
L number of variable heads ˇn,j,t binary variable which is equal to 1, provided that
M number of not-allowed operating zones variable head n + 1 of the jth hydro generating unit
N number of blocks relating to the piecewise lin- at time t is chosen
earized hydro generating unit’s performance curve ın,i,t generation of the nth block relating to the cost curve
NB number of bilateral contracts of the ith thermal generating unit at time t (MW)
NE number of blocks relating to the piecewise lin- ˚ payoff table
earized thermal generating unit’s emission curve r aggregate membership function of the rth Pareto
NL number of blocks relating to the piecewise lin- optimal solution
earized variable cost function rn individual membership function for the function n
pbt power capacity relating to the bilateral contract at in the Pareto best solution r
time t (MW) n,j,t the volume of the nth block of the reservoir per-
pmin,i minimum power relating to the ith thermal gener- taining to the jth hydro generating unit at time t
ating unit (MW) (Hm3 )
pmax,i maximum power relating to the ith thermal gener- feasible region
ating unit (MW)
194 A. Ahmadi et al. / Electric Power Systems Research 123 (2015) 192–204

Bi,t start-up cost of the ith thermal generator at time t Yi,t binary variable which is equal to 1, provided that
($) the ith thermal generator is started-up at the start
Ci,t cost of valve loading influence relating to the ith of time t
thermal generating unit at time t ($) Yj,t binary variable which is equal to 1, provided that
Fi,t the fuel cost of the ith thermal generating unit at the jth hydro generator is started-up at the start of
time t ($) time t
fU , fN and fSN Utopia, nadir point and pseudo nadir point, Zi,t binary variable which is equal to 1, provided that
respectively. the ith thermal generator is shut down at the start
fnU , fnN and fnSN Utopia nadir point and pseudo nadir point in of time t
the nth objective function, respectively.
hn,j,t binary variable equal to 1 when the water discharge Sets
of the jth hydro generator at time t has exceeded the I thermal units’ set
nth block J hydro plants’ set
Ii,t binary variable equal to 1 when the ith thermal gen- T set of indices related to the periods of the market
erator is on-line at time t time horizon
Ij,t binary variable equal to 1 when the jth hydro gen-
discrete intervals’ set related to start-up cost func-
erator is on-line at time t tion of thermal units
Id,i,t binary variable equal to 1 when the ith thermal gen- ˝j upstream reservoirs’ set
erator at time t generates non-spinning reserve.
n̂ unit normal to the CHIM simplex in the direction of
the origin
Nd,i,t non-spinning reserve of the ith thermal generator differential real-coded quantum-inspired evolutionary based SHTS
at time t for bid on day-ahead market when unit is has been proposed in [9], while some improvements have been
off (MW) achieved with adaptive mechanisms to enhance the ability of the
Nd,j,t non-spinning reserve of the jth hydro generator at method in the case of global search over a continuous space.
time t in the day-ahead market when unit is off Furthermore, benders decomposition [10], an enhanced adaptive
(MW) optimization based on a particle swarm method [11], particle
Nu,i,t non-spinning reserve of the ith thermal generator swarm mixed-binary optimization [12], particle swarm method
at time t in the day-ahead market when unit is on based on a small population [13], quantum-inspired method with
(MW) clonal real-coded [14], and Lagrangian relaxation [15] have been
p number of competing objective functions also used for solving hydrothermal scheduling problem.
pi,t power generated by the ith thermal generator at Due to the recent international agreements, such as the Kyoto
time t (MW) protocol and similar European and US acts, the emission of gaseous
p̄i,t maximum power output of the ith thermal genera- pollutants have been become a global concern that should be taken
tor at time t (MW) into account [16,17].
pj,t power output of the jth hydro generator at time t Ahmadi et al. [17], Vahidinasab and Jadid [18] have used
(MW) MO optimization framework to model the optimal usage of
pst power considered to bid in day-ahead market at available resources. In this regard, hybrid augmented-weighted ε-
time t (MW) constraint technique has been employed to generate the Pareto
Qj,t water discharge of the jth hydro generator at time t set while Generation Company’s (GENCO’s) profit maximization
(m3 /s) and emission minimization represent the objective functions. The
qn,j,t water discharge of the nth block relating to the jth AC-constrained SHTS problem is solved in [19] based on benders
hydro generator at time t (m3 /s) decomposition technique, enhanced by bacterial foraging based
Ri,t spinning reserve of the ith thermal generator at time on Particle Swarm Optimization (PSO) algorithm. In this problem,
t in the day-ahead market (MW) the objective functions to minimize are the generation costs, as
RDL(pi,t ) ramp down limit relating to the ith thermal gener- well as emission costs of the power system, taking into account
ator at time t (MW) the valve point effects and banned zones of thermal units. A civi-
RUL(pi,t ) ramp up limit relating to the ith thermal generator lized swarm optimization algorithm is proposed in [20] to solve the
at time t (MW) multi-objective (MO) SHTS problem, which is an hybrid algorithm
si,t periods during which the ith thermal generator has between society-civilization method and PSO. The economic emis-
been shut down at time t (h) sion problem has been solved in [21] using a MO hybrid differential
sj,t reservoir’s spillage pertaining to the jth hydro gen- evolution based on culture belief based, while MO cultural differen-
erator at time t (m3 /s) tial evolution based on simulated annealing has been used in [22].
vj,t storage volume of the reservoir pertaining to the jth Zhang [23] has utilized MO differential evolution based on three
hydro generator at time t (Hm3 ) chaotic sequences to solve SHTS with economic/emission problem,
wn weighting factor of the nth objective function while the elitist archive method has been adopted to derive the
wi,t binary variable which is equal to 1, provided that non-dominated solutions. In [24], predator–prey optimization has
the ith thermal generator is started-up at the start been employed for the main level search while for the solution
of time t and it was off-line for  hours of the economic/emission hydrothermal scheduling problem, the
x̄i∗ decision variables’ vector for fi objective function approach of Powell has been used. Other methods for the solu-
tion of the hydrothermal scheduling problem are self-organizing
hierarchical PSO technique associated with varying acceleration
parameters [25], MO cultural algorithm based on PSO [26], evo-
lutionary programming based interactive fuzzy satisfying method
 A. Ahmadi et al. / Electric Power Systems Research 123 (2015) 192–204 195

technique [27], differential evolution [28], quantum-behaved PSO and start-up costs. The last term of the main objective function
[29] and hybrid MO cultural algorithm [30]. stands for the start-up cost pertaining to hydro generating units.
It is worth mentioning that in [19–30], problems have been for- 
mulated so that costs and emissions are simultaneously minimized, 
F EMISSION = emin,i Id,i,t + SUEi Yi,t + SDEi Zi,t
while in [17,18], a MO framework has been proposed to simul-
i∈EMG t∈T
taneously maximize profit as well as minimize emissions for an

hydrothermal system [17] and thermal generating units [18]. Solu- 
NE
tion methods proposed for SHTS and SHTSS problems have been + [ˇn,i,t E(pun−1,i ) + ben,i ın,i,t] (2)
categorized into analytical and heuristic approaches [2,31]. n=1
In the presented paper, the normal boundary intersection (NBI)
method is used to solve the MO SHTSS optimization problem. To where EMG = {SO2 , NOx }.
the best of our knowledge, no study has investigated on MO SHTSS The above function indicates the other objective function, i.e.
problem. The significant contributions of the paper can be high- emission generation that GENCOs want to minimize. As SO2 and
lighted as follows: NOx have an environmental impact, they are considered as the most
relevant for the electrical sector [17,18]. In order to model the prob-
lem accurately, the emission function also consists of emissions due
(a) Proposing a MO framework for SHTSS problem taking into
to the start-up and shut down of thermal generators. Note that the
account the profit maximization and emission minimization as
first term represents the emission caused by turned-off thermal
objective functions;
units once providing non-spinning reserve [17].
(b) Employing NBI method to concurrently maximize profit and
minimize emissions, as well as generating evenly distributed
Pareto optimal solutions; 2.2. Power balance
(c) Implementing the decision making process by the fuzzy deci-
sion making method and assessing bilateral contracts by means The power balance is known as one of the most significant con-
of performance indicators; straints which guarantees that the total generated power is equal
(d) Modelling of prohibited operating zones and dynamic ramp rate to the power sold for bilateral contracts and the day-ahead market
limits of thermal generating units, as well as hydro units’ multi- at each time of scheduling period:
performance curves;  
pi,t + pj,t = pst + pbt ∀t ∈ T (3)
(e) Transforming mixed-integer non-linear programming (MINLP)
i∈I j∈J
of the SHTSS problem into a MILP problem.

2.3. Thermal units’ model

The remaining paper is organized as follows. The hydrothermal
model is presented in Section 2 and performance indices are
The linear model of thermal units can be illustrated through the
defined in Section 3. The proposed NBI method is presented in Sec-
following expressions:
tion 4. Section 5 deals with the simulation results, while conclusions
are given in Section 6. 
M+1
Fi,t = [ˇn,i,t F(pun−1,t ) + bn,i ın,i,t ] ∀i ∈ I, ∀t ∈ T (4)
2. MILP formulation for SHTSS n=1

In the following paragraphs, the MILP formulation for SHTSS is 

M+1
Pi,t = [pun−1,i ˇn,i,t + ın,i,t ] ∀i ∈ I, ∀t ∈ T (5)
reported.
n=1

2.1. Objective function ın,i,t ≥ 0 n = 1, 2, . . ., M + 1, ∀i ∈ I, ∀t ∈ T (6)

The emissions of gaseous pollutants are a global concern that ın,i,t ≤ [pdn,i − pun−1,i ]ˇn,i,t n = 1, 2, . . ., M + 1, ∀i ∈ I, ∀t ∈ T
should be taken into account by GENCOs in electric energy produc- (7)
tion section. In this regard, this paper considers minimization of
emissions in addition to the profit maximization. Thus, two objec- where pu0,i = pmin,i and pdM+1,i = pmax,i .
tive functions are considered as follows.

M+1
 ˇn,i,t = Ii,t ∀i ∈ I, ∀t ∈ T (8)
 
F PROFIT
= tb pbt + ts pst + {tsr Ri,t + tns {Nu,i,t + Nd,i,t }} n=1

t∈T i∈I ˇn,i,t ∈ {0, 1} n = 1, 2, . . ., M + 1, ∀i ∈ I, ∀t ∈ T (9)


   
NL
+ {tsr Rj,t + tns {Nu,j,t + Nd,j,t }} − Ai Zi,t + Fi,t + Bi,t − Aj Yj,t Bi,t = Ki wi,t

∀i ∈ I, ∀t ∈ T (10)
j∈J i∈I j∈J
=1
(1)

NL

wi,t = Yi,t ∀t ∈ T (11)
where FPROFIT is the profit objective function that GENCOs are
=1
willing to maximize. Generally, this function consists of the rev-
enue coming from bilateral contracts and the energy sale in the 
NL−1

day-ahead market plus the revenue from selling reserve to the day- si,t−1 = 
wi,t + i,t ∀i ∈ I, ∀t ∈ T (12)
ahead market by thermal and hydro generators. Furthermore, this =1
profit function includes the cost of thermal and hydro generating NL
NLwi,t NL
≤ i,t ≤ s̄i {wi,t − Yi,t + 1} ∀i ∈ I, ∀t ∈ T (13)
units. The cost of thermal units includes fuel costs, shut down cost
196 A. Ahmadi et al. / Electric Power Systems Research 123 (2015) 192–204


wi,t ∈ {0, 1} ∀i ∈ I, ∀ ∈
, ∀t ∈ T (14) thermal generators. Further details about Eqs. (1)–(20) can be found
in previous papers published by the authors in this area [3,17].
pmin,i Ii,t ≤ pi,t ≤ p̄i,t ∀i ∈ I, ∀t ∈ T (15)
2.4. Hydro plants’ model
p̄i,t ≤ pmax,i {Ii,t − Zi,t+1 } + SDi Zi,t+1 ∀i ∈ I, ∀t ∈ T (16)

pi,t−1 − pi,t ≤ SDi Zi,t + RDL(pi,t ) ∀i ∈ I, ∀t ∈ T (17) The adopted model considers the head-dependent reservoirs
in the MILP formulations. The reason beyond taking into account
pi,t+1 − pi,t ≤ SUi Yi,t+1 + RUL(pi,t ) ∀i ∈ I, ∀t ∈ T (18) multi-head reservoirs is that the water discharge from a small stor-
age capacity of reservoirs influences the plant head and the power

M+1 output. Therefore, the number of heads related to hydro generat-
RDL(pi,t ) = RDLn,i ˇn,i,t ∀i ∈ I, ∀t ∈ T (19) ing units is assumed to be L through the formulations presented as
n=1 follows. It is worth mentioning that the higher value assigned to L
means a more accurate model.

M+1
RUL(pi,t ) = RULn,i ˇn,i,t ∀i ∈ I, ∀t ∈ T (20) vj,0 = v0j ∀j ∈ J (21)
n=1
vj, = v
j
∀j ∈ J (22)
The piecewise linearized model of the fuel cost function, includ-
ing M prohibited operating zones (POZs), is presented through Eqs. vj,t ≥ v0,j ∀j ∈ J, ∀t ∈ T (23)
(4)–(9) [3,17].
The fuel cost function Fi,t (4) is expressed as function of ˇn,i,t (a 
L

binary variable equal to 1 if the power block n of thermal unit i of vj,t ≤ vL,j ˇL−1,j,t + vn−1,j [ˇn−2,j,t − ˇn−1,,t ] ∀j ∈ J, ∀t ∈ T
the piecewise fuel cost curve is selected), F(pun−1,i ) (generation cost n=2 (24)
of pun−1,i ) and the slope and generation of power block n. Constraint
(5) regards with the power output of thermal unit i: the power 
L

of the thermal unit i at hour t is the sum of the “power lengths”. vj,t ≥ vL−1,j ˇL−1,j,t + vn−2,j [ˇn−2,j,t − ˇn−1,j,t ] ∀j ∈ J, ∀t ∈ T
Constraint (6) assures that the positive power output of each block n=3
(25)
and (7) limits it to the upper bound of the block. Expressions (8)
and (9) guarantee that when the thermal unit i is online, it operates
ˇ1,j,t ≥ ˇ2,j,t ≥ ˇ3,j,t ≥ . . . ∀j ∈ J, ∀t ∈ T (26)
in one operating zone.
The exponential function of the hours of thermal unit switching- where ˇ0,j,t = 1.
off is adopted for the cold start-up cost. Its piecewise linear model  
can be expressed by Eqs. (10)–(14), [3,17]. In particular, constraint  
k−1 
L−1

(10) is a time varying linear function of the start-up cost in which pj,t − pk,j Ij,t − qn,j,t bkn,j − p̄j (k − 1) − ˇn,j,t + ˇn,j,t
-
Ki is the start-up cost of the -th discrete interval of thermal n∈N n=1 n=k
 is a binary variable equal to 1 if thermal unit i
unit i while wi,t ≤ 0 ∀j ∈ J, ∀t ∈ T, 1 ≤ k ≤ L (27)
is switched on at the beginning of hour t, having been off for 
hours. Constraint (11) assures that only one of the binary variables  
 
k−1

L−1
is equal to 1. A time counter is related to the variables by means pj,t − pk,j Ij,t − qn,j,t bkn,j + p̄j (k − 1) − ˇn,j,t + ˇn,j,t
of the dummy variable  i,t (12). It is used when the thermal unit i, -
n∈N n=1 n=k
which has been turned-off for NL hours or longer, is started-up at
hour t or is off at hour t (13). ≥ 0 ∀j ∈ J, ∀t ∈ T, 1 ≤ k ≤ L (28)
Constraint (15) denotes the power generation limit of thermal
units and constraint (16) indicates the upper limit of power gen-
pj,t + Rj,t + Nu,j,t ≤ p̄j Ij,t ∀j ∈ J, ∀t ∈ T (29)
eration by thermal units at each time. The ramp rate of shut down
and ramp-down limit (RDL) are shown in constraint (17), while Nd,j,t ≤ QSCj (1 − Ij,t ) ∀j ∈ J, ∀t ∈ T (30)
constraint (18) states the ramp rate of start-up and ramp-up limit 
(RUL). It is noted that unlike the majority of the previously pub- Qj,t = Q j Ij,t + qn,j,t ∀j ∈ J, ∀t ∈ T (31)
-
lished papers, the presented paper applies the dynamic ramp rate n∈N
to obtain an accurate model. The presented dynamic ramp rate is
a function of power output and considers POZs. Dynamic RDL and
- j,t ≤ Qj,t + sj,t ≤ ¯ j,t ∀j ∈ J, ∀t ∈ T (32)
RUL are represented through constraints (19)–(20), and the binary 0 ≤ sj,t ≤ s̄j ∀j ∈ J, ∀t ∈ T (33)
variables allow to define the selected operating zone.
In order to cover sudden events, such as outages of transmis- 
L

sion lines or generating, operating services are utilized [3]. These sj,t = bn,j n,j,t ∀j ∈ J, ∀t ∈ T (34)
services are generally categorized into spinning and non-spinning n=1
reserves. The first type, i.e. spinning reserve is defined as unloaded

L
units pertaining to the synchronized generation that can be in vj,t = v0,j + ∀j ∈ J, ∀t ∈ T (35)
n,j,t
service in 10 min and the second one, i.e. non-spinning reserve,
n=1
relates to the unsynchronized generators that can be in service
in 10 min. Due to the minimum up-time limit, when a unit turns [v1,j − v0,j ]ˇ1,j,t ≤ 1,j,t ≤ v1,j − v0,j ∀j ∈ J, ∀t ∈ T (36)
on, it should not be switched off for at least several hours. On the
[vn,j − vn−1,j ]ˇn,j,t ≤ n,j,t n = 2, 3, . . ., L, ∀j ∈ J, ∀t ∈ T (37)
other hand, when a unit is turned off, it should not be commit-
ted for at least several hours. In order to eliminate the conflicting n,j,t ≤ [vn,j − vn−1,j ]ˇn−1,j,t n = 2, 3, . . ., L, ∀j ∈ J, ∀t ∈ T (38)
states between start-up and shut down states at the same time,
the formulation of [3] are implemented for both hydro and q1,j,t ≤ Q̄1,j Ij,t ∀j ∈ J, ∀t ∈ T (39)
 A. Ahmadi et al. / Electric Power Systems Research 123 (2015) 192–204 197

q1,j,t ≥ Q̄1,j h1,j,t ∀j ∈ J, ∀t ∈ T (40) calculated from the following relation:

qn,j,t ≤ Q̄n,j hn−1,j,t ∀j ∈ J, ∀t ∈ T, ∀n ∈ N
qn,j,t ≥ Q̄n,j hn,j,t ∀j ∈ J, ∀t ∈ T, ∀n ∈ N
vj,t = vj,t−1 + Fj,t −  · (qj,t + sj,t )

+ [qj,t−ij + sj,t−ij ] ∀j ∈ J, ∀t ∈ T
i∈˝j
The initial and the final volume of reservoir at the beginning
and the end of the case study are illustrated through constraints
(21) and (22). In addition, the minimum volume of each hydro
unit at each time of scheduling period is indicated in constraint
(23) while constraints (24) and (25) relates to the right head cor-
responding to the volume. Meanwhile, constraint (26) ensures the
prevention from the combination of 0/1 for these binary variables.
Also, the linear relationship between the generated power, the dis-
charged water and the variable head is stated in constraints (27) and
(28) and the relationship between generation and ancillary services
is indicated in constraints (29) and (30) [3]. In particular, con-
straint (29) assures that the capacity of hydro plant is higher than
the power generated and spinning and non-spinning reserves of a
hydro plant. The inequality (30) bounds the non-spinning reserve
to the hydro plant quick start capacity.
The piecewise linear model of hydro unit’s water discharge
is proposed in constraint (31). Furthermore, constraint (32) is
applied to prevent flooding and also for irrigation requirements.
The spillage is restricted by constraint (33) and it is defined as a
function of the content of the reservoir in (34). Besides, the relation-
ship between the content of the reservoir and the volume blocks
is represented in (35). The first block of the reservoir’s volume
is limited by constraint (36), while other blocks of the volume
of the reservoir at each time are limited by inequalities (37) and
(38). Similarly, the first block of the water discharge is restricted
by constraints (39) and (40), while other blocks of the water dis-
charge are limited by constraints (41) and (42). In case of cascaded
hydro plants, the upstream hydro plant water drained away is the
inflow of the following downstream plant. Constraint (43) denotes
the continuity equation of the jth plant at time t. Further details
on constraints (21)–(43) can be found in [3] and [17]: the mod-
elling of hydro and thermal units is presented in [3], wherein the
expected profit is maximized utilizing 0/1 MILP technique; [17]
proposes the augmented-weighted ε-constraint method to concur-
rently maximize GENCO’s profit and minimize emissions while this
paper employs the NBI technique to solve the proposed MO-SHTSS
problem.
 

NB
(41) Tot Revenuet = Ancillary sale + ts pst + pbm,t
m=1
(42)

NB
b
+ (m,t − ts ) pbm,t ∀t ∈ T (45)
m=1
(43) The first term of (45) stands for the revenue from providing
ancillary services, which is assumed to be almost equal to that
of (44) because of lower amount and lower price of such service
compared to power sold in day-ahead market and through bilat-
eral contracts. The second term indicates the revenue that would
be obtained, provided that the total power generated by GENCO is
sold in day-ahead market. The GENCO’s performance indicator is
included in the last term which is possible to be either positive or
negative. Once the selling price under bilateral contracts is more
than the price in day-ahead market, this performance indicator is
positive, otherwise not. In the case of negative indicator, the future
bilateral contracts must be revised in terms of selling price and
amount of power [3].
4. Multi-objective optimization
In general, the MO optimization is stated as below:
Min{f1 (x), f2 (x), . . ., fp (x)}
(46)
s.t : g(x) ≤ 0, h(x) = 0
where x indicates the decision variables and f1 (x), f2 (x), . . ., fp (x) are
p objectives while p ≥ 2. h(x) and g(x) denote the problem’s equality
and inequality constraints, respectively.
MO optimization problems deal with two or more objective
functions and finally do not lead to any single optimal solution
concurrently optimizing all objective functions. In such conditions,
the “most desired” or the “most preferred” solution is intended to
be found by decision makers (DMs). In these optimization prob-
lems, the concept of optimality no longer means, and a new concept
named Pareto optimality is used. The non-dominated solutions (i.e.
Pareto optimal solutions also known as Pareto set) are those which
are impossible to get better in each of objective functions except
declining their performances through at least another objective
function [17,32].
4.1. Weighted sum method
One of the techniques most used to solve MO optimization prob-
lems is the weighted sum method. This method minimizes a convex
combination of different objectives [33].


p

3. Performance indicators Min wi fi (x)

(47)
i=1
Different combinations of trades in bilateral and day-ahead mar- s.t : g(x) ≤ 0, h(x) = 0
kets can be used by GENCO [3]. The following equation denotes the
GENCO’s revenue at each time: 
p

where wi ≥ 0, wi = 1.

NB i=1
Revenuet = Ancillary sale + ts pst + b
m,t pbm,t ∀t ∈ T (44) In spite of becoming a popular method, the results obtained
m=1 by weighted methods on the Pareto front have serious problems
with changing the coefficients. It has been noticed that changing
As it can be observed from (44) that the overall revenue of w by small values may lead to remarkable changes in the objec-
GENCO at each time of the scheduling horizon comprises three tive vectors, while changing w by large values may cause almost
terms, i.e. revenue obtained from providing spinning and non- negligible changes in the objective vectors. This phenomenon is
spinning reserves, revenue from selling power through day-ahead mainly because of the fact that this method is not a Lipschitzian
market and bilateral contracts. Thus, the overall revenue can be function of w [34]. Obviously, this illustrates the sophisticated and
198 A. Ahmadi et al. / Electric Power Systems Research 123 (2015) 192–204

non-intuitive relation between the weights and solutions’ perfor- Therefore, the MO mathematical programming problem can be
mance. In such conditions, deriving a proper approximation of resolved in a non-dimensional, unit-less criterion space [36,38].
Pareto optimal front using uniform sampling of w may be ineffi- The normalized values through this method are used to form the
cient to a large extent as it may result in uneven sampling of the ¯
normalized payoff table ˚.
Pareto front [27,34]. It is worth mentioning that RP , that is the set of points, consists
of p real-valued objective functions and RP is called objective space
4.2. NBI method and RP are convex combinations of each row of payoff table and
known as Convex Hull of Individual Minima (CHIM). Any point P(ˇ1 ,
NBI technique is considered an efficient tool for numerical com- ˇ2 , . . ., ˇp ) in the normalized space on this line can be stated as
putation of evenly distributed solutions on the Pareto set [35,36]. below:
The first step in NBI technique is to form the so-called payoff
¯
P(ˇ1 , ˇ2 , . . ., ˇp ) = {˚ˇ, ˇ ∈ RP } (55)
table ˚. For each of the objective functions fi , the optimal solu-
tions are firstly considered to form the payoff table for a MO
mathematical programming problem which has p competing, non-

P
where the parameters ˇi = 1 and 0 ≤ ˇ1 , ˇ2 , . . ., ˇp ≤ 1.
commensurable objective functions [17,37]. fi∗ (x̄i∗ ) denotes the
i=1
optimum value obtained for fi , in which the decision variable vector
Assume that ˇ is a convex weighting, then ˚ˇ refers to a point
optimizing objective functions fi is indicated by x̄i∗ . The next step
in the CHIM. If the unit normal to the CHIM simplex towards the
is to calculate the value pertaining to other objective functions f1 ,
origin is indicated by n̂, the set of points on that normal is denoted
f2 ,. . ., fi−1 , fi+1 ,. . .,fp , using the solution to optimize the objective
by ˚ˇ + Dn̂.
function fi . These calculated values are indicated by f1 (x̄i∗ ), f2 (x̄i∗ ),. . .,
Note that the point of intersection of the normal and the bound-
fi−1 (x̄i∗ ), fi+1 (x̄i∗ ),. . ., fp (x̄i∗ ). Note that the ith row in the payoff table
ary of F̄(x̄) closest to the origin is the global solution of the following
represents f1 (x̄i∗ ), f2 (x̄i∗ ),. . ., fi∗ (x̄i∗ ),. . ., fp (x̄i∗ ). Similarly, all other rows
sub-problem:
of the payoff table can be calculated as below:
⎛ ⎞ Max D
f1∗ (x̄1∗ ) ··· fi (x̄1∗ ) ··· fp (x̄1∗ )
⎜ ⎟ s.t. ˚ˇ + Dn̂ = F̄(x̄) (56)
⎜ .. .. . ⎟
⎜ . . .. ⎟ g(x) ≤ 0, h(x) = 0
⎜ ⎟
˚=⎜ ∗ ∗ ∗ ∗ ⎟
⎜ f1 (x̄i ) · · · fi (x̄i ) · · · fp (x̄i ) ⎟ (48) ˚ˇ + Dn̂ = F̄(x̄) ensures that the point x is actually mapped onto
⎜ ⎟
⎜ .. .. .. ⎟ a point on the normal while other constraints guarantee the feasi-
⎝ . . . ⎠ bility of x considering the original problem [36,38]. A point-wise
f1 (x̄p∗ ) ··· fi (x̄p∗ ) ··· fp∗ (x̄p∗ ) approximation of the Pareto front would be derived by solving (56)
for different values of ˇ.
As it can be observed from (48), there are p rows and columns
while the jth column consists of the values obtained for the objec-
tive function fj , where the maximum and minimum values define 4.3. Fuzzy decision maker
the limits of the objective function fj .
However, in order to better state the performance of the NBI After that Pareto set is derived, the most desired solution per-
method, it is necessary to introduce some few concepts. In this taining to the MO SHTSS problem must be chosen by means of a
regard, Utopia point defines a precise point, mostly referring to the fuzzy DM front. The basics of Fuzzy decision making is to introduce
region outside the feasible region which matches to all objectives a membership function which is a linear function to every objective
concurrently taking their best possible values. This point can be function in every Pareto optimal solution determining the relative
stated in mathematical form as follows: distance of the objective function’s value in each Pareto optimal
solution and its values in the corresponding Utopia and Pseudo Nadir
f U = [f1U , . . ., fiU , . . ., fpU ] = [f1∗ (x̄1∗ ), . . ., fi∗ (x̄i∗ ), . . ., fp∗ (x̄p∗ )] (49) points. The membership value obtained for each solution would
be higher, provided that the resulted value is closer to its respec-
The next concept relates to a point in the objective space,
tive Utopia value, i.e. far from its corresponding Pseudo Nadir value.
wherein all objectives take their possible worst values at the same
The higher membership value means the better and more desired
time. This point is known as Nadir point which can be indicated as
the solution would be. These linear membership functions can be
below:
mathematically stated as follows:
f N = [f1N , . . ., f1N , . . ., fpN ] (50) ⎧
⎪ 1 fnr ≤ fnU
where ⎪
⎨
fnSN − fnr
rn = fnU ≤ fnr ≤ fnSN (Minimization) (57)
fiN = Maxfi (x̄), subject tox̄ ∈ ˝ (51) ⎪
⎪ SN U
x̄ ⎩ fn − fn
0 fnr ≥ fnSN
In (51), ˝ refers to the feasible region. The Pseudo Nadir Point is
another point having a concept close to the Nadir point which can ⎧
⎪ 0 fnr ≤ fnSN
be stated as follows: ⎪
⎨
fnr − fnSN
f SN = [f1SN , . . ., fiSN , . . ., fpSN ] (52) rn = fnSN ≤ fnr ≤ fnU (Maximization) (58)
⎪
⎪ U SN
⎩ fn − fn
fiSN = Max{fi (x̄1∗ ), . . ., fi∗ (x̄i∗ ), . . ., fi (x̄p∗ )} (53) 1 fnr ≥ fnU

Utopia and Pseudo Nadir points can be a normalized in the inter- Note that, Eq. (57) is defined for objective functions which are
val [0 1] as follows [36,38]. minimized while for objective functions intended to be maximized,
fi (x̄) − fiU Eq. (58) is used. Finally, the total membership function is defined
f̄i (x̄) = i = 1, 2, . . ., p (54) to determine the total optimality degree relating to each solu-
fiSN − fiU tion according to the individual membership functions and the
 A. Ahmadi et al. / Electric Power Systems Research 123 (2015) 192–204 199

Table 1 0.65
Payoff table for SHTSS problem.

Total membership
  0.6
6035627.63 332798.12
˚=
4629566.28 4581.16
0.55

significance of objective functions with respect to others (wn val-
ues). The total membership can be stated as:
p
w · rn
n=1 n
 = 
r
(59)
p
wn
n=1
By defining such function as (59), the most preferred solution
is selected as the one with the highest value of the total member-
ship r . It is worth noting that this solution mainly optimizes the
MO optimization problem’s objective functions taking into account
their relative significance compared to the other Pareto optimal
solutions.
5. Case study
5.1. Description of the system
In order to verify the effectiveness of the suggested framework,
the model is applied to IEEE 118-bus test system which is comprised
of 54 thermal generating units, as well as 8 hydro generating units.
It is assumed that the bilateral contract at each time of scheduling
period is 1000 MWh with 40 $/MWh. However, since the informa-
tion data is not enough, without loss of generality, fixed ramp rate
is used. The data required for hydro units can be found in [5] while
data required for thermal units, such as market price, POZ data and
emission coefficients are reported in [39]. Furthermore, the com-
puter used to solve the presented MO SHTSS problem is a 2.6 GHz
Pentium IV PC equipped with 4 GB RAM employing CPLEX solver
in GAMS [40] software.
5.2. Pareto solutions
Table 1 represents the obtained payoff table. As it can be
observed from Table 1, if the emission generation is discarded,
GENCO can get a profit of 6,035,627.63 ($) while, in this case, the
emissions are of 332,798.12 (lbs). In addition, the emission can be
reduced to 4581.16 (lbs) by GENCO, if only the second objective
0.5
0.45
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
Pareto soluon number
NBI method weighted sum method
Fig. 1. Total membership of Pareto solutions obtained by the NBI method and
weighted sum method for case 1.
function (FEMISSION ) is taken into account. However, in this case, the
profit is 4,629,566.28 ($). This means that, considering the emis-
sion function leads to a profit loss of 1,406,061.35 ($), even if the
emissions decrease of 328,216.96 (lbs).
It can be observed that the time required to form the payoff table
˚ is 111.27 s while there are 49,477 single constraints, 49,473 vari-
ables and 25,110 discrete variables. Also, the number of constraints,
variables and discrete variables pertaining to the weighted sum
method are 49,468, 49,473 and 25,110, respectively.
The Pareto optimal solutions obtained by using the NBI tech-
nique and weighted sum methods are represented in Table 2
with the information on profit, emissions and computational time
related to each Pareto solution. Although, the computational time
needed by the weighted sum method is less than that required by
NBI technique. It is important that the time needed to solve the
proposed method is still reasonable; so GENCO can use it for SHTSS.
5.3. Best optimal solution
In order to compare the Pareto solutions obtained using the
NBI method with those obtained by the weighted sum method, a
Fuzzy decision maker method proposed in Section 4.3 and a pay-
off table have been utilized. In this case weighting factors relating
to these two objective functions in the fuzzy decision making pro-
cedure are considered to be the same in order to determine the
most desired solution of the Pareto set [36,37]. Fig. 1 illustrates the
results obtained by considering the same weighting factors.

Table 2
Pareto solutions by NBI method and weighted sum method.

Pareto No. ˇ1 NBI method Weighted sum method

Profit ($) Emission (lb) Time (S) Profit ($) Emission (lb) Time (S)

1 0.00 4,629,566.28 4581.16 16.54 1,229,726.10 4581.16 3.73

2 0.05 4,741,901.08 11,180.54 157.98 4,629,259.86 4581.16 5.71
3 0.10 4,835,718.03 22,102.54 114.70 4,766,108.86 18,572.20 5.69
4 0.15 4,927,831.93 33,422.08 121.89 5,267,867.92 92,567.43 5.69
5 0.20 5,017,528.91 45,305.81 87.47 5,601,244.70 161,971.08 5.72
6 0.25 5,103,344.65 58,095.53 79.07 5,772,692.54 213,692.62 5.68
7 0.30 5,187,502.78 71,272.19 62.55 5,865,359.97 247,483.01 5.76
8 0.35 5,268,269.06 85,240.62 61.41 5,926,293.57 277,428.90 5.63
9 0.40 5,347,587.09 99,547.10 62.64 5,973,876.14 306,786.99 5.61
10 0.45 5,425,145.24 114,264.39 56.56 5,992,505.65 320,723.88 5.59
11 0.50 5,502,145.92 129,111.82 66.37 6,016,326.86 341,070.25 5.64
12 0.55 5,571,560.30 145,730.11 57.41 6,016,096.38 345,362.56 5.75
13 0.60 5,640,842.56 162,379.25 73.64 6,026,858.26 354,266.43 5.62
14 0.65 5,704,234.18 180,403.43 63.41 6,030,528.46 360,727.07 5.80
15 0.70 5,769,595.01 197,967.95 78.53 6,031,901.71 366,867.84 5.61
16 0.75 5,827,484.73 217,276.44 101.10 6,035,561.57 371,327.73 5.60
17 0.80 5,883,371.61 237,052.46 78.94 6,034,616.74 374,910.26 5.61
18 0.85 5,931,955.34 258,533.26 65.08 6,036,900.31 376,363.79 5.66
19 0.90 5,974,607.16 281,398.74 207.19 6,037,370.79 379,538.19 5.62
20 0.95 6,013,423.81 305,159.46 129.59 6,037,542.55 381,905.30 5.63
21 1.00 6,035,627.63 332,798.12 94.73 6,035,627.63 384,247.87 5.53
200 A. Ahmadi et al. / Electric Power Systems Research 123 (2015) 192–204
70
Eergy price ($)
6000
Power (MW)
50
4000
30
2000 10
0 -10
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour
Total thermal power Total hydro power Energy price
Fig. 2. Scheduling results of thermal and hydro units and prices for Pareto solution
11.
The total membership pertaining to all Pareto solutions obtained
by NBI technique and weighted sum method are shown in Fig. 1. The
Pareto optimal front derived from uniform sampling of weighting
factors (ˇ1 , ˇ2 ) obtained by using the weighted sum method is not
efficient due to very uneven sampling of the Pareto optimal front.
Also, Fig. 1 and Table 2 clarify that the total membership values
of Pareto optimal solutions from 14 to 21 are very similar. Fur-
thermore, Table 2 shows that the profit variation of Pareto optimal
solutions from 14 to 21 obtained by the weighted sum method is
very small. Besides, also the emissions’ variation of Pareto optimal
solutions from 14 to 21 obtained by weighted sum method is very
small. The weighted sum method is, therefore, not able to well coor-
dinate objectives, since they have different units of measurement
causing problems in determining weighting factors to produce the
correct Pareto optimal solutions [27]. Unlike the weighting sum
method, the effectiveness of the NBI method for numerical compu-
tation of the evenly distributed points on the Pareto optimal front
is verified through Table 2 and Fig. 1. As Fig. 1 shows, the most pre-
ferred solution is the Pareto solution 11, while when employing the
weighted sum method; Pareto solution 5 is the most preferred solu-
tion. Table 3 gives detailed information of the most desired Pareto
solutions obtained by NBI technique and weighted sum method. It
is worth mentioning that the value in parentheses in Table 3 indi-
cates the membership value (rn ) of each objective function that
can be considered as its optimality degree. As it can be observed
from Table 3, the NBI method leads to a more favoured solution if
compared to the weighted sum method (0.621 versus 0.606).
Further details on Pareto solution 11 obtained by using the NBI
method are reported in the following. The energy produced by
the GENCO in the day-ahead market, as well as the correspond-
ing day-ahead market prices, are illustrated in Fig. 2. Hydro units
generate 16,056.34 MWh over the entire scheduling period with
an average value of 669.01 MWh. In addition, the minimum power
generated by hydro generating units is 23.95 MW at hour 24, while
the maximum power of 2261.47 MW occurs at hour 17, i.e. the hour
with the highest energy price at day-ahead market. As it can be
Table 3
The most preferred solution obtained by the proposed NBI method and weighted sum method for case 1.
Method Total membership
NBI 0.621
Weighted sum method 0.606
Table 4
Optimum solution of SHTSS with different weighting factors.
Objective function Weighting factor
FPROFIT ($) 2 5,931,955.34
FEMISSION (lb) 1 258,533.26
Total membership of all objective functions
30000
Performance indicator ($)
20000
10000
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
-10000
-20000
-30000
Hour
Fig. 3. GENCO performance index for Pareto solution 11.
observed from Fig. 2, the variation of the power generated by hydro
units is remarkable (i.e. around 238.03%) and follows the variations
of the energy price at the day-ahead market. The highest power
generated by hydro generating units relates to the higher market
prices of electrical energy (hours 13–19). During the initial periods,
reservoirs store water as the energy price is low while during the
middle periods, due to higher energy prices, hydro units gener-
ate more power. However, at the end of the scheduling period, the
power produced by hydro units decreases since the energy price is
low. Therefore, they store water to meet the final reservoir volume
[41,42].
On the other hand, thermal units produced 141,933.74 MWh
over the entire scheduling period with an average value of
5913.91 MWh. It is worth mentioning that the minimum power
generation of thermal generating units is 4854.00 MW occurring
at hour 4, i.e. the hour with the lowest energy price in the day-
ahead market while the maximum power generation occurs at
hours 16 and 17 with 6794.00 MW, i.e. at hours with the highest
energy prices in the day-ahead market. Thermal units generated
the nominal power and their power variation is relatively small (i.e.
of around 17.92%). It is worth noting that, since thermal generators
are not able to track the changes of load, their power generation’s
variation is minimal. Furthermore, the great difference in thermal
units’ generation may cause serious harms to the boilers [4,43,44].
The performance indicator of GENCO for the Pareto solution 11
over the scheduling period is illustrated in Fig. 3. A positive index
can be noted up to 7 a.m., while it becomes negative in the range
8–21, and once again positive from hour 22 to the end of the con-
sidered period. The performance indicator of GENCO is negative
in most hours and the sum of this performance indicator is nega-
tive (−78,944.00) over the entire scheduling period. Thus, it can be
stated that this GENCO has not a profitable bilateral trade.
The membership value represented in Table 3 demonstrates
the optimality degree. By considering the same weighting factors,
the membership value of the most desired solution is only 0.621
which is not acceptable for the GENCO. However, the decision
FPROFIT ($) FEMISSION (lb)
5,502,145.92 (0.621) 129,111.82 (0.621)
5,601,244.70 (0.691) 161,971.08 (0.520)
Objective function value Membership value
0.926
0.226
0.693
 A. Ahmadi et al. / Electric Power Systems Research 123 (2015) 192–204 201

Table 5
Result of different method.

Method [27] [28] [29] [30] Proposed method

ELS EES CEES ELS EES CEES ELS EES CEES ELS EES CEES ELS EES CEES

Cost ($) NR NR 47,906 43,500 51,449 44,914 41,909 45,392 43,507 41,805 48,191 43,593 41,549.99 48,921.93 43,501.896
Emission (lb) NR NR 26,234 21,092 18,257 19,615 30,724 17,659 18,183 16,841 15,746 16,204 17,076.52 15,666.61 16,146.542
Solution time (s) NR NR 4582 72.96 72.74 74.97 NR NR NR NR NR NR 1.92 1.61 40.95

NR = not reported.

maker is able to easily achieve its desired solution by chang-
ing the weighting factors, considering that GENCO tends towards
profit maximization. Hence, the presented problem is solved again
considering different weighting factors for profit and emissions
objective functions. A Pareto solution having a higher membership
value for the profit and a lower membership value for emissions
has been obtained, as depicted in Table 4. According to Table 4,
the profit membership value considerably enhanced from 0.621 to
0.926 when different weights are used. The profit represented in
Table 4 is 5,931,955.34 ($) which is indeed very close to its ideal
value (6,035,627.63) in the Payoff table ˚. To the contrary, the
emission increases, since the membership value of emissions is low
(0.226).
5.4. Discussion and comparison
This paper proposes an effective MO tool which is suitable for
GENCO for day-ahead scheduling of electricity generating units
even for power systems of large scale, taking into account many
variables and constraints. However, the authors believe that there
is not any published paper proposing the SHTSS problem for such
a large system (IEEE 118-bus) including 54 thermal units and 8
hydro generating units where two competing objectives as profit
and emissions are concurrently optimized. Besides, this problem
comprises many practical constraints, such as dynamic ramp rate
limit, multi POZs relating to thermal generating units as well as
multi performance curves relating hydro units. Accordingly, there
is no published paper to be compared with.
In order to show the performances of the proposed approach
with respect to other techniques, the presented NBI method has
been tested on different power systems ([27–30,45–49] and [50])
looking at the simultaneous minimization of both costs and emis-
sions.
5.4.1. Case 1
This test system [27–30] includes 4 hydro units and 3 thermal
units. Only economic issues are considered in the economic load
scheduling (ELS). Furthermore, the economic emission scheduling
(EES) strategy tries to minimize the emission issues while com-
bined economic emission scheduling (CEES) strategy optimizes the
two objective functions comprehensively by transforming the orig-
inal MO optimization problem into a unified function using a set of
weights. The non-linear formulation presented in [27–30] is used
to make a fair comparison with the results obtained in these works.
Table 5 expresses the optimal solution derived from NBI method,
as well as the one obtained in [27–30]. Based on this table, the
method proposed in this paper leads in all cases to a better solution
compared to those reported in [27–30], both from the obtained val-
ues viewpoint and computational time. Besides, the solution time
required by the proposed method is reasonable and better com-
pared to other methods. Table 6 represents the detailed information
on the best Pareto solution. The obtained results demonstrate that

Table 6
The details of the best compromise solution.

Hour Water discharge rates (*104 m3 ) Hydro generations in MW Thermal generations in MW Total in MW
 
Qh1t Qh2t Qh3t Qh4t Ph1t Ph2t Ph3t Ph4t Ps1t Ps2t Ps3t Ph + Ps

1 8.55 6.00 23.44 6.00 78.87 50.16 21.66 129.03 175.00 155.52 139.76 750
2 7.27 6.00 24.16 6.00 71.19 51.30 10.16 125.74 175.00 206.85 139.76 780
3 7.39 6.00 22.96 6.00 72.17 52.93 13.60 121.63 175.00 124.91 139.76 700
4 6.72 6.00 21.56 6.00 67.39 54.50 19.27 115.82 128.35 124.91 139.76 650
5 6.54 6.00 19.41 6.00 65.92 55.50 28.85 132.09 122.97 124.91 139.76 670
6 5.82 6.00 17.30 6.00 60.44 55.99 36.74 147.09 175.00 184.97 139.76 800
7 9.92 7.22 16.46 12.19 86.60 63.80 38.85 231.71 175.00 209.82 144.22 950
8 9.90 7.61 16.43 14.29 86.20 66.02 37.87 258.93 175.00 211.68 174.30 1010
9 9.92 7.95 17.03 16.10 86.32 68.17 36.07 277.97 175.00 216.95 229.52 1090
10 9.62 7.74 17.37 17.05 85.30 67.51 35.29 285.85 175.00 209.82 221.24 1080
11 9.87 8.38 17.32 16.46 87.20 71.72 35.99 281.17 175.00 219.40 229.52 1100
12 9.51 8.17 17.95 16.43 85.55 70.37 34.48 280.94 175.00 280.08 223.59 1150
13 9.69 8.80 18.30 17.03 86.87 73.62 34.57 285.69 175.00 224.73 229.52 1110
14 9.55 8.91 18.69 17.37 86.80 74.25 33.98 288.30 175.00 214.56 157.10 1030
15 9.19 8.81 18.43 17.32 85.33 73.83 36.01 287.87 175.00 209.82 142.14 1010
16 9.24 9.32 17.25 17.95 85.76 75.86 41.24 292.58 175.00 209.82 179.75 1060
17 9.24 9.80 16.14 18.30 85.73 76.74 45.66 295.06 175.00 214.57 157.25 1050
18 9.05 10.11 15.80 18.69 84.41 75.77 47.84 297.71 175.00 212.13 227.15 1120
19 8.99 10.89 15.47 19.98 83.54 76.73 49.78 304.28 175.00 214.91 165.76 1070
20 8.94 11.64 15.19 20.00 82.47 77.20 51.68 301.55 175.00 211.40 150.70 1050
21 5.08 9.42 12.13 19.03 54.60 67.29 55.78 292.66 175.00 124.91 139.76 910
22 5.00 9.75 12.67 19.26 54.30 68.38 57.88 290.45 124.32 124.91 139.76 860
23 5.00 10.64 13.12 20.00 54.71 70.57 58.83 289.82 111.41 124.91 139.76 850
24 5.00 10.83 13.51 20.00 55.02 69.26 58.96 284.40 102.67 124.91 104.77 800
Cost ($) 43,501.896 Emission (lb) 16,146.542
202 A. Ahmadi et al. / Electric Power Systems Research 123 (2015) 192–204

applying NBI technique results in better solutions for the daily 800000
scheduling of hydrothermal energy systems with rational solution 780000
time. 760000
740000
720000

Cost ($)
5.4.2. Case 2
700000
This case, used in [45–49], comprises 3 plants and 6 thermal
680000
units, and the only emission taken into account is NOx , while CO2
660000
and SO2 are neglected. The problem objective is to find the best
640000
dispatch with minimum cost and emissions. The total demand is 620000
900 MW. 600000
The Pareto optimal solutions (including 13 variables and 9 equa- 0 20000 40000 60000 80000 100000 120000
tions) obtained applying the NBI method are shown in Table 7; all Emission (lbs)
of them are feasible [50–53]. Other information reported in Table 7
is, for each Pareto solution total loss, iteration and solution time, NBI Epsilon constraint
which ranges from 0.082 to 0.197 s (this upper limit is exceptional).
Then, the fuzzy decision making technique is used. DM can Fig. 4. Case 3 – pareto optimal solutions of cost versus emission with epsilon-
constraint and NBI methods.
choose a set of combinations of weighting factors. For example,
if DM is looking for a lower cost, the weighting factor of the cor-
respondent cost function would be higher while the other factor
functions are, as before, costs and emissions minimization. A com-
would decrease. In case the weighting factor of the cost is 3 and the
parison is made between this method and the application of the
one of emission is 1, the Pareto solution 15 will be selected, being
epsilon-constraint method [54–56], whose ratio is to optimize the
characterized by the highest membership value.
main objective function and consider the other objective functions
In Table 8 the results of other optimization techniques are pre-
as constraints. The results are illustrated in Fig. 4.
sented [45–49], demonstrating that the presented technique is
Fig. 5 shows the different results obtainable by using the two
clearly better than the others [52].
methods in terms Pareto optimal solutions versus total, cost and
Indeed, in the case study the proposed method is characterized
emission membership values, having assigned the same weighting
by the least costs and emissions with respect to the others.
factors.
Here, it is clear that the total membership of NBI method’s Pareto
5.4.3. Case 3 solutions is always higher than the ones achieved by the method.
With the objective of verifying the effectiveness of the model, Moreover, the emission membership with the NBI method are bet-
the case 3 deals with the implementation of the NBI technique on ter than the epsilon-constraint ones. Figs. 4 and 5 [50] demonstrate
IEEE 118-bus test system described in Section 5.1. The two objective the effectiveness of the presented method.

Table 7
Case 2 – pareto solutions.

Pareto number ˇ1 Cost ($) Emission (kg) Total loss (MW) Iteration Time (s)

1 0.05 49,978.274 701.820 40.869 25 0.096

2 0.10 49,704.654 702.922 40.628 25 0.197
3 0.15 49,444.778 704.782 40.396 27 0.101
4 0.20 49,198.963 707.417 40.174 27 0.093
5 0.25 48,967.513 710.843 39.964 26 0.093
6 0.30 48,750.710 715.077 39.765 25 0.095
7 0.35 48,548.820 720.133 39.577 27 0.098
8 0.40 48,362.084 726.023 39.401 25 0.094
9 0.45 48,190.719 732.761 39.238 24 0.096
10 0.50 48,034.915 740.356 39.087 21 0.095
11 0.55 47,894.822 748.818 38.949 24 0.093
12 0.60 47,770.548 758.150 38.824 33 0.095
13 0.65 47,662.142 768.358 38.712 24 0.091
14 0.70 47,569.580 779.438 38.614 26 0.092
15 0.75 47,492.740 791.385 38.530 29 0.082
16 0.80 47,431.371 804.185 38.459 23 0.102
17 0.85 47,385.056 817.814 38.403 28 0.095
18 0.90 47,353.176 832.239 38.361 28 0.083
19 0.95 47,334.870 84/7.412 38.335 21 0.188

Table 8
Case 2 – results achieved applying different optimization methods.

Optimization method [45] [46] [47] [48] [49] Proposed

P1 (MW) 51.82 51.82 51.82 51.83 51.82 50.526

P2 (MW) 32.65 32.65 38.64 38.66 32.66 39.400
P3 (MW) 208.78 208.77 248.73 248.74 208.79 157.340
P4 (MW) 128.12 128.12 122.14 122.15 128.12 157.367
P5 (MW) 292.02 292.03 252.02 252.03 291.95 266.279
P6 (MW) 223.57 223.57 223.57 223.58 223.57 267.616
Total cost ($/h) 47,548.96 47,548.97 47,804.55 47,809.03 47,549.87 47,492.740
Net emission (kg/h) 823.35 823.35 843.42 843.53 823.36 791.385
CPU time (s) 14.36 12.54 0.195 0.814 12.03 0.082
 A. Ahmadi et al. / Electric Power Systems Research 123 (2015) 192–204
Fig. 5. Case 3 – variation of total, cost and emission membership functions versus pareto-optimal solutions for equal weighting factors using Epsilon-constraint and NBI
methods.
6. Conclusion
In this paper, the NBI method has been well extended for solving
the short-term combined economic emission hydrothermal self-
scheduling problem in the energy markets of the day-ahead and
reserve, adopting the MILP framework. The key features of the pro-
posed method are related to the nature of the formulation that,
being linear, well adapts to be applied also in cases of real size,
large scale power systems. Moreover, the proposed formulation
also allows considering real-world constraints for thermal units
and hydro plants. It considers, indeed, dynamic ramp rate, mini-
mum up-time and down-time and prohibited operating zones for
thermal units, and multi-performance curve with spillage and delay
time between reservoirs for hydro plants. Simulation results evi-
dence the superiority of the NBI method in terms of solutions’
quality and good convergence properties. In addition, the time
required to solve the problem is rational and shorter in comparison
with other methods. Also, the simulation results for an IEEE 118-bus
system demonstrate the applicability of the NBI technique to large-
scale hydrothermal power systems. The ongoing research activities
are focused on introducing a stochastic framework to take into
account the energy price uncertainty and generating units forced
outage.
References
[1] A.J. Wood, B.F. Wollenberg, Power Generation Operation and Control, John
Wiley & Sons, New York, 1996.
[2] I.A. Farhat, M.E. El-Hawary, Optimization methods applied for solving the short-
term hydrothermal coordination problem, Electr. Power Syst. Res. 79 (9) (2009)
1308–1320.
[3] J. Aghaei, A. Ahmadi, H.A. Shayanfar, A. Rabiee, Mixed integer programming of
generalized hydro-thermal self-scheduling of generating units, Electr. Eng. 95
(2) (2013) 109–125.
[4] S. Bisanovic, M. Hajro, M. Dlakic, Hydrothermal self-scheduling problem
in a day-ahead electricity market, Electr. Power Syst. Res. 78 (9) (2008)
1579–1596.
[5] A.J. Conejo, J.M. Arroyo, J. Contreras, F.A. Villamor, Self-scheduling of a hydro
producer in a pool-based electricity market, IEEE Trans. Power Syst. 17 (4)
(2002) 1265–1272.
[6] K.K. Mandal, N. Chakraborty, Parameter study of differential evolution based
optimal scheduling of hydrothermal systems, J. Hydro-environ. Res. 7 (1) (2013)
72–80.
[7] P.K. Roy, Teaching learning based optimization for short-term hydrothermal
scheduling problem considering valve point effect and prohibited discharge
constraint, Int. J. Electr. Power Energy Syst. 53 (2013) 10–19.
[8] X. Liao, J. Zhou, S. Ouyang, R. Zhang, Y. Zhang, An adaptive chaotic artificial
bee colony algorithm for short-term hydrothermal generation scheduling, Int.
J. Electr. Power Energy Syst. 53 (2013) 34–42.
[9] Y. Wang, J. Zhou, L. Mo, R. Zhang, Y. Zhang, Short-term hydrothermal gener-
ation scheduling using differential real-coded quantum-inspired evolutionary
algorithm, Energy 44 (1) (2012) 657–671.
203
[10] N. Amjady, M.R. Ansari, Hydrothermal unit commitment with AC constraints
by a new solution method based on benders decomposition, Energy Convers.
Manage. 65 (2013) 57–65.
[11] Y. Wang, J. Zhou, C. Zhou, Y. Wang, H. Qin, Y. Lu, An improved self-adaptive
PSO technique for short-term hydrothermal scheduling, Expert Syst. Appl. 39
(3) (2012) 2288–2295.
[12] V.H. Hinojosa, C. Leyton, Short-term hydrothermal generation scheduling
solved with a mixed-binary evolutionary particle swarm optimizer, Electr.
Power Syst. Res. 92 (2012) 162–170.
[13] J. Zhang, J. Wang, C. Yue, Small population-based PSO for short-term hydro-
thermal scheduling, IEEE Trans. Power Syst. 27 (1) (2012) 142–152.
[14] Y. Wang, J. Zhou, L. Mo, S. Ouyang, Y. Zhang, A clonal real-coded quantum-
inspired evolutionary algorithm with Cauchy mutation for short-term
hydrothermal generation scheduling, Int. J. Electr. Power Energy Syst. 43 (1)
(2012) 1228–1240.
[15] F.Y.K. Takigawa, E.L. Silva, E.C. Finardi, R.N. Rodrigues, Solving the hydrothermal
scheduling problem considering network constraints, Electr. Power Syst. Res.
88 (2012) 89–97.
[16] K.D. Le, J.L. Golden, C.J. Stansberry, R.L. Vice, J.T. Wood, J. Balance, J.Y. Amya,
E.K. Nielsen, H. Nakajima, M. Ookubo, I. Iyoda, G.W. Cauley, Potential impacts
of clean air regulations on system operations, IEEE Trans. Power Syst. 10 (2)
(1995) 647–656.
[17] A. Ahmadi, J. Aghaei, H.A. Shayanfar, A. Rabiee, Mixed integer programming
of multi-objective hydro-thermal self-scheduling, Appl. Soft Comput. 12 (8)
(2012) 2137–2146.
[18] V. Vahidinasab, S. Jadid, Multi-objective environmental/techno-economic
approach for strategic bidding in energy markets, Appl. Energy 86 (4) (2009)
496–504.
[19] M. Parastegari, R.A. Hooshmand, A. Khodabakhshian, M. Vatanpour, AC con-
strained hydro-thermal generation scheduling problem: application of Benders
decomposition method improved by BFPSO, Int. J. Electr. Power Energy Syst.
49 (2013) 199–212.
[20] A.I. Selvakumar, Civilized swarm optimization for multi-objective short-term
hydrothermal scheduling, Int. J. Electr. Power Energy Syst. 51 (2013) 178–189.
[21] H. Zhang, J. Zhou, Y. Zhang, Y. Lu, Y. Wang, Culture belief based multi-objective
hybrid differential evolutionary algorithm in short term hydrothermal sched-
uling, Energy Convers. Manage. 65 (2013) 173–184.
[22] H. Zhang, J. Zhou, N. Fang, R. Zhang, Y. Zhang, Daily hydrothermal scheduling
with economic emission using simulated annealing technique based multi-
objective cultural differential evolution approach, Energy 50 (2013) 24–37.
[23] H. Zhang, J. Zhou, Y. Zhang, N. Fang, R. Zhang, Short term hydrothermal sched-
uling using multi-objective differential evolution with three chaotic sequences,
Int. J. Electr. Power Energy Syst. 47 (2013) 85–99.
[24] N. Narang, J.S. Dhillon, D.P. Kothari, Multi-objective fixed head hydrothermal
scheduling using integrated predator-prey optimization and Powell search
method, Energy 47 (1) (2012) 237–252.
[25] K.K. Mandal, N. Chakraborty, Daily combined economic emission scheduling of
hydrothermal systems with cascaded reservoirs using self-organizing hierar-
chical PSO technique, Expert Syst. Appl. 39 (3) (2012) 3438–3445.
[26] R. Zhang, J. Zhou, Y. Wang, Multi-objective optimization of hydrothermal
energy system considering economic and environmental aspects, Int. J. Electr.
Power Energy Syst. 42 (1) (2012) 384–395.
[27] M. Basu, An interactive fuzzy satisfying method based on evolutionary pro-
gramming technique for multi-objective short-term hydrothermal scheduling,
Electr. Power Syst. Res. 69 (2–3) (2004) 277–285.
[28] K.K. Mandal, N. Chakraborty, Short-term combined economic emission
scheduling of hydrothermal power systems with cascaded reservoirs using
differential evolution, Energy Convers. Manage. 50 (1) (2009) 97–104.
[29] S. Lu, C. Sun, Z. Lu, An improved quantum-behaved particle swarm optimization
method for short-term combined economic emission hydrothermal sched-
uling, Energy Convers. Manage. 51 (3) (2010) 561–571.
204 A. Ahmadi et al. / Electric Power Systems Research 123 (2015) 192–204
[30] Y. Lu, J. Zhou, H. Qin, Y. Wang, Y. Zhang, A hybrid multi-objective cultural
algorithm for short-term environmental/economic hydrothermal scheduling,
Energy Convers. Manage. 52 (5) (2011) 2121–2134.
[31] N.P. Padhy, Unit commitment—a bibliographical survey, IEEE Trans. Power Syst.
19 (2) (2004) 1196–2005.
[32] G. Mavrotas, Effective implementation of the ε-constraint method in multi-
objective mathematical programming problems modified augmented, Appl.
Soft Comput. 213 (2) (2009) 455–465.
[33] K. Miettinen, Nonlinear Multiobjective Optimization, Kluwer Academic Pub-
lishers, Boston, 1999.
[34] R.E. Steuer, Multiple Criteria Optimization: Theory, Computation and Applica-
tion, John Wiley & Sons, New York, 1985.
[35] I. Das, J.E. Dennis, Normal boundary intersection: a new method for generating
the Pareto surface in nonlinear multi-criteria optimization problems, SIAM J.
Opt. 8 (3) (1998) 631–657.
[36] V. Vahidinasab, S. Jadid, Normal boundary intersection method for suppliers’
strategic bidding in electricity markets: an environmental/economic approach,
Energy Convers. Manage. 51 (6) (2010) 1111–1119.
[37] J. Aghaei, N. Amjady, H.A. Shayanfar, Multi-objective electricity market
clearing considering dynamic security by lexicographic optimization and
augmented epsilon constraint method, Appl. Soft Comput. 11 (4) (2011)
3846–3858.
[38] C. Roman, W. Rosehart, Evenly distributed Pareto points in multi-
objective optimal power flow, IEEE Trans. Power Syst. 21 (2) (2006)
1011–1012.
[39] http://motor.ece.iit.edu/data
[40] Generalized algebraic modeling systems (GAMS). http://www.gams.com
[41] H.M. Ghadikolaei, A. Ahmadi, J. Aghaei, M. Najafi, Risk constrained self-
scheduling of hydro/wind units for short term electricity markets considering
intermittency and uncertainty, Renew. Sustain. Energy Rev. 16 (7) (2012)
4734–4743.
[42] A. Ahmadi, J. Aghaei, H.A. Shayanfar, Stochastic self-scheduling of hydro units
in joint energy and reserves markets, in: 19th Iranian Con. Electr. Eng. (ICEE),
2011, pp. 1–5.
[43] M. Karami, H.A. Shayanfar, J. Aghaei, A. Ahmadi, Mixed integer programming of
security constrained daily hydrothermal generation scheduling, Sci. Iran. Trans.
D: Comput. Sci. Eng. Electr. Eng. 20 (6) (2013) 2036–2050.
[44] M. Karami, H.A. Shayanfar, J. Aghaei, A. Ahmadi, Scenario-based security-
constrained hydrothermal coordination with volatile wind power generation,
Renew. Sustain. Energy Rev. 28 (2013) 726–737.
[45] M. Motevasel, A.R. Seifi, Expert energy management of a micro-grid considering
wind energy uncertainty, Energy Convers. Manag. 83 (2014) 58–72.
[46] A.A. Moghaddam, A. Seifi, T. Niknam, M.R.A. Pahlavani, Multi-objective
operation management of a renewable MG (micro-grid) with back-up micro-
turbine/fuel cell/battery hybrid power source, Energy 36 (2011) 6490–6507.
[47] C. Palanichamy, N.S. Babu, Analytical solution for combined economic and
emissions dispatch, Electr. Power Syst. Res. 78 (7) (2008) 1129–1139.
[48] C. Palanichamy, K. Srikrishna, Economic thermal power dispatch with emission
constraint, J. Indian Inst. Eng. (India) 72 (11) (1991).
[49] C. Jiejin, M. Xiaoqian, L. Qiong, L. Lixiang, P. Haipeng, A multi-objective chaotic
particle swarm optimization for environmental/economic dispatch, Energy
Convers. Manage. 50 (5) (2009) 1318–1325.
[50] A. Ahmadi, M.S. Masouleh, M. Janghorbani, N.Y.G. Manjili, A.M. Sharaf, A.
Esmaeel Nezhad, Short term multi-objective hydrothermal scheduling, Electr.
Power Syst. Res. (2014) 1–11.
[51] H. Mavalizadeh, A. Ahmadi, A. Heidari, The probabilistic multi-objective gen-
eration and transmission expansion planning problem using normal boundary
intersection, IET Gen. Trans. Distrib. (2014) 1–11.
[52] M. Izadbakhsh, M. Gandomkar, A. Rezvani, A. Ahmadi, Short-term resource
scheduling of a renewable energy based micro grid, Renew. Energy 75 (2015)
598–606.
[53] A. Rezvani, M. Gandomkar, M. Izadbakhsh, A. Ahmadi, Environmen-
tal/economic scheduling of a micro-grid with renewable energy resources, J.
Clean. Prod. 87 (2015) 216–226.
[54] M.R. Norouzi, A. Ahmadi, A.M. Sharaf, A. Esmaeel Nezhad, Short-term environ-
mental/economic hydrothermal scheduling, Electr. Power Syst. Res. 116 (2014)
117–127.
[55] H. Mavalizadeh, A. Ahmadi, Hybrid expansion planning considering security
and emission by augmented epsilon-constraint method, Int. J. Electr. Power
Energy Syst. 61 (2014) 90–100.
[56] A. Ahmadi, M.R. Ahmadi, A. Esmaeel Nezhad, A lexicographic optimization and
augmented ε-constraint technique for short-term environmental/economic
combined heat and power scheduling, Electr. Power Comput. Syst. 42 (9) (2014)
945–958.