Effects of wind energy supplied by independent power producers

on the generation dispatch of electric power utilities

Chia-Liang Lu
a
, Chun-Lung Chen
b,1,
*
, Daw-Shang Hwang
b
, Yuang-Tung Cheng
b
a
Department of Electrical Engineering, St. Johns University, Taiwan, ROC
b
Department of Marine Engineering, National Taiwan Ocean University, 2 Pei-Ning Road, Keelung 20224, Taiwan, ROC
a r t i c l e i n f o
Article history:
Received 11 September 2007
Received in revised form 5 August 2008
Accepted 15 August 2008
Keywords:
Wind-thermal coordination dispatch
Wind power penetration
Independent power producers
Avoided generation cost
Evolutionary simulated annealing algorithm
Direct search algorithm
a b s t r a c t
Integration of wind power plants into the existing isolated system presents challenges to power system
operators. The problem is further complicated to the generation dispatch imposed by the presence of
wind turbine generators (WTGs), owned by independent power producers (IPP) with a large penetration
of dispatchable wind energy sources. An evolutionary simulated annealing (ESA) algorithm combined
with the direct search method (DSM) is developed in this paper for the solution of wind-thermal coordi-
nation dispatch (WTCD) problem without restrictions on the shape of cost functions. The proposed hybrid
algorithm can be used to maximize the contribution of non-utility wind farms to reducing the cost of
thermal dispatch while maintaining an adequate level of supply reliability. The developed WTCD soft-
ware is also a useful tool for the evaluation of avoided generation cost (AGC) due to purchase of IPP wind
energy by the utility. Numerical experiments are included to understand the wind generator output
variations in system operating cost analysis and to assess the impact and economic benets of the
installation of non-utility wind farms.
2008 Elsevier Ltd. All rights reserved.
1. Introduction
The rise of environmental protection and the progressive
exhaustion of traditional fossil energy sources have increased the
interests in integrating renewable energy sources into existing
power systems. The government in Taiwan has commissioned re-
search on renewable energy applications under the consideration
of diversifying energy sources [1]. Among various renewable en-
ergy sources in Taiwan, wind energy could be, in the short term,
one of the most promising renewable energy sources [2]. Since
the cost of wind turbine generators (WTGs) has been reduced rap-
idly, installation of WTGs as fuel savers is economically and envi-
ronmentally attractive in windy regions. It would be benecial to
increase the power supply capacity in Taiwan by the installation
of non-utility (or public utility) power plants using wind energy
sources. However, it is widely believed that large wind penetra-
tions would put an increased burden on system operations due
to the uncertain nature of wind power [310]. One of the most
important future challenges seems to be the management of the
integration of uctuations in the electricity production from wind
energy sources [11]. Another important issue regarding the inte-
gration of non-utility wind generation energy into a public utility
is the evaluation of the appropriate avoided generation cost
(AGC) due to purchase of IPP energy by the utility. These complex
conditions make it very important to solve the wind-thermal coor-
dination dispatch (WTCD) problem for achieving optimal utiliza-
tion of non-utility wind energy sources in the existing isolated
system.
Integration of non-utility wind power plants into the existing
isolated system presents challenges to power system operators.
Both intermittency and unpredictability of wind generator output
require different operation strategies to be developed. This devel-
opment will cause a severe change in the demands on the power
reserve due to the high uctuations and the limited predictability
of wind power generation. To maintain an adequate level of supply
reliability, the system operator will require the scheduling of addi-
tional emergency reserves that signicantly increase operating
costs. However, it is also found that the utility also tries to maxi-
mize its economic saving by utilizing IPP wind energy in the short
term planning. The key integration question then becomes how
variations in wind plant outputs affect the operation of the isolated
0142-0615/$ - see front matter 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijepes.2008.08.002
* Corresponding author. Tel.: +886 2 24622192x7107; fax: +886 2 24633765.
E-mail address: cclung@mail.ntou.edu.tw (C.-L. Chen).
1
Chun-Lung Chen was born in Taiwan, Republic of China, 1965. He received his
BSEE degree from National Taiwan Institute of Technology in 1990, MSEE from
National Taiwan University in 1992, and Ph.D. from National Taiwan University of
Science and Technology in 2001. From 2001 to 2008, he has been with the
Department of Electrical Engineering, Minghsin University of Science and Technology.
Since 2008, he came to National Taiwan Ocean University, where he is assistant
professor of marine engineering. His areas of major interests are demand side
management, unit commitment, economic dispatch, and wind energy systems.
Electrical Power and Energy Systems 30 (2008) 553561
Contents lists available at ScienceDirect
Electrical Power and Energy Systems
j our nal homepage: www. el sevi er . com/ l ocat e/ i j epes
system on a day-to-day basis and what the associated costs are. In
general, the wind generation owned by an IPP will have a cost that
must be based on the special contractual agreements (i.e., xed
schedule or economic schedule) for buying or selling energy be-
tween each IPP and the public utility. The system operator may
be obliged to dispatch all available power produced by wind en-
ergy resources through guaranteed contracts. However, the wind
generator outputs may be curtailed by the system operator
through interruptible contracts when security constraints are
violated. The goal of this study is to develop a systematic planning
process for the system operator to purchase wind generation from
an IPP and how much to pay for this service.
The evaluation of AGC is based on the optimal generation
dispatch both before and after the inclusion of non-utility wind en-
ergy. The importance of WTCD problem is, thus, likely to increase,
and more advanced algorithms are worth developing to operate an
isolated hybrid power system reliably and efciently. However,
incorporating additional physical and economic operation con-
straints into the generation scheduling problem adds further
complexity to the solution methodology. Several optimization
algorithms based on classical calculus based techniques [12] or
stochastic searching techniques [1317] could be used to solve
the extended economic dispatch problem. In this paper, an evolu-
tionary simulated annealing (ESA) algorithm combined with the
direct search method (DSM) is developed for the solution of WTCD
problem. The proposed ESA algorithm incorporates the parallel
searching mechanism of evaluation programming into the simple
simulated annealing to enhance its search capacity that leads to
a higher probability of obtaining the near global optimal solution.
An effective strategy based on the two levels of competition is used
for global exploitation. To deal effectively with a number of
inequality and equality constraints, a constraint satisfaction tech-
nique is also developed to improve the ESA solution process. The
DSM algorithm is used as a nely tuning to determine the global
optimal solution at the nal with a reduced computing time. Test
results are provided to illustrate the merits of the proposed meth-
od and to evaluate the appropriate price that it has to pay to the IPP
for the exchange of electrical energy.
2. Mathematical formulation of WTCD problem
2.1. Notation
The following notation is used throughout the paper.
i index for thermal units
j index for wind units
NT number of thermal units in system
NW number of wind units in system
F
i
() operation cost function of thermal unit i
a
i
, b
i
, c
i
cost coefcients of thermal unit i
e
j
, f
j
constants from the valve-point loading effect of thermal
units
F
T
total operation cost
P
L
system load demand
P
i
current output power of thermal unit i
P
0
i
previous output power of thermal unit i
P
max
i;r
upper generation limit of thermal unit i
P
min
i;r
lower generation limit of thermal unit i
P
max
i
modied maximum output power of thermal unit i
P
min
i
modied minimum output power of thermal unit i
UR
i
upramp limit of the ith thermal generation (MW/time-per-
iod)
DR
i
downramp limit of the ith thermal generation (MW/time-
period)
TUR system ramp-up capacity
TDR system ramp-down capacity
USR
B
system up reserve requirement not considering wind
power generation
US
i
up reserve contribution of thermal unit i
DS
i
down reserve contribution of thermal unit i
US
max
i
maximum up reserve contribution of thermal unit i
DS
max
i
maximum down reserve contribution of thermal unit i
d% percentage of maximum unit capacity
r% percentage of total actual wind power generation
wu% percentage of system load demand
P
max
Wj
upper generation limit of wind unit j
P

Wj
available generation of wind unit j
P
Wj
actual generation of wind unit j
P

WT1
available wind generation supplied by public utility
P
WT1
actual wind generation supplied by public utility
P

WT2
available wind generation supplied by IPP
P
WT2
actual wind generation supplied by IPP
P

WT
total available wind generation
P
WT
total actual wind generation
P
0
WT
previous output power of total actual wind generation
v wind speed
v
Ij
cut in wind speed of wind unit j
v
Rj
rated wind speed of wind unit j
v
Oj
cut out wind speed of wind unit j
u
j
() wind power curve of wind unit j
S predened calculation step
K reduced factor
FC1 system fuel cost considering the inclusion of IPP wind en-
ergy
FC2 system fuel cost neglecting the inclusion of IPP wind en-
ergy
HSS hybrid stochastic search
EPSQP evolutionary programmingsequential quadratic program-
ming
PSOSQP particle swarm optimizationsequential quadratic pro-
gramming
2.2. Formulation
The optimal dispatch of generation in a wind-thermal system
involves the allocation of generation among wind plants and
thermal plants so as to minimize the total production costs while
satisfying various constraints. Generally, the fuel cost of a ther-
mal generation unit will be a second order polynomial function
[12]
F
i
P
i
a
i
b
i
P
i
c
i
P
2
i
1
However, the fuel cost functions of units may be much more com-
plicated due to the physical operation limitations, which actually
exist in a practical optimization problem. The inputoutput perfor-
mance curve for a typical thermal unit with many valve-points was
presented in [14]. The fuel cost functions take into account the
valve-point effects were expressed as
F
i
P
i
a
i
b
i
P
i
c
i
P
2
i
e
i
sinf
i
P
min
i
P
i

2
The generation cost of WTGs from the public utility is the cheapest
because it needs no fuel, but in a privately owned non-utility, the
basic requirement of WTGs is to achieve the maximum fuel saving
of thermal generation and to determine the appropriate price that
the utility should pay to the IPP for this dispatchable wind energy
sources. The available output of a wind generator can be obtained
from the wind speed by applying the wind power curve. The total
available wind power generation can be easily calculated using
the following equation
554 C.-L. Lu et al. / Electrical Power and Energy Systems 30 (2008) 553561
P

WT
P

WT1
P

WT2
3
The reduction of a wind generator output can also be achieved
through blade pitch control. The total actual wind generation can
also be calculated as following:
P
WT
P
WT1
P
WT2
4
Because of the unpredictable rise and fall in wind generator output,
the system operator will require the scheduling of additional emer-
gency reserves to maintain adequate frequency control. In this
study, a simple linear model relating additional up/down spinning
reserve requirements and total actual wind power generation is
considered. The increased up/down spinning reserve requirements
are calculated as a simple fraction of the predicted wind generation.
The mathematical model of the WTCD problem can be stated as
follow.
Objective function:
Minimize F
T

X
NT
i1
F
i
P
i
5
subject to the following constraints.
2.3. System constraints
Power balance constraint
X
NT
i1
P
i
P
WT
P
L
6
System up/down spinning reserve requirements
X
NT
i1
US
i
PUSR
B
r% P
WT
7
X
NT
i1
DS
i
Pr% P
WT
8
Minimum/maximum thermal plant output constraints
P
L
P1 r% P
WT

X
NT
i1
P
min
i
9
X
NT
i1
P
max
i
1 r% P
WT
PP
L
USR
B
10
2.4. Thermal unit constraints
Unit capacity constraints
P
min
i;r
6 P
i
6 P
max
i;r
11
Units maximum up/down reserve contribution constraints
US
max
i
d% P
max
i
12
DS
max
i
d% P
max
i
13
Units up/down spinning reserve contribution constraints
US
i
min fUS
max
i
; P
max
i
P
i
g 14
DS
i
min fDS
max
i
; P
i
P
min
i
g 15
Unit ramp-rate limit constraints
P
min
i
6 P
i
6 P
max
i
16
where
P
min
i
max fP
min
i;r
; P
0
i
DR
i
g 17
P
max
i
min fP
max
i;r
; P
0
i
UR
i
g 18
2.5. Wind unit constraints
Wind power curve constraints
P

Wj

0 v 6 v
Ij
or v > v
Oj
u
j
v v
Ij
6 v 6 v
Rj
P
max
Wj
v
Rj
6 v 6 v
Oj
8
>
<
>
:
19
Total actual wind generation limit
0 6 P
WT
6 P

WT
20
System maximum up/down spinning reserve contribution
constraints
P
WT
6
P
NT
i1
US
i
USR
B
r%
21
P
WT
6
P
NT
i1
DS
i
r%
22
Wind power penetration limit
P
WT
6 wu% P
L
23
Wind generation uctuation constraints
P
WT
P
0
WT
6 TDR if P
0
WT
6 P
WT
24
P
0
WT
P
WT
6 TUR if P
0
WT
P P
WT
25
3. Development of the WTCD software
Incorporating non-utility wind generators into the existing util-
ity economic dispatch problem adds further complexity to the
solution methodology and system reliability evaluation. In this pa-
per, a hybrid algorithm based on a combination of evolutionary
simulated annealing (ESA) algorithm and direct search method
(DSM) is developed to solve the non-smooth economic dispatch
problem considering wind power generation. The possibility of
occurrence of nding the near global optimal solution for the ESA
can be greatly increased by using the two levels of competition
strategy. Many nonlinear characteristics of units could be handled
C.-L. Lu et al. / Electrical Power and Energy Systems 30 (2008) 553561 555
properly by using a constraint satisfaction technique. The DSM
algorithm is used as a nely tuning to determine the global optimal
solution at the nal with a reduced computing time. The outline of
the proposed algorithm is shown in the owchart in Fig. 1. The pro-
posed algorithm can be stated in detail as follows:
3.1. Determine the feasible region of total wind power generation
The system ramping capacity, the load demand and the size of
the wind-thermal energy system are important factors in deter-
mining the proportion of wind generator capacity that can be inte-
grated into the isolated system. In this study, several technique
constraints are applied to decide the feasible region of total wind
power generation. The allowable penetration region of total wind
power generation is determined from the wind speed and is
restricted by the system maximum ramping up/down capabilities
of the thermal units, by the thermal plant upper/lower generation
limits, by the system load demand and by the special contractual
agreements between each IPP and the public utility. In this study,
the actual power output of total wind generation is limited by
applying the following equation.
P
D
WT
6 P
WT
6 P
U
WT
26
where
P
D
WT
min max
P
L
USR
B

P
NT
i1
P
max
i
1 r%
; 0
( )
; P

WT
( )
27
P
U
WT
min
P
L

P
NT
i1
P
min
i
1r%
;
P
NT
i1
US
i
USR
B
r%
;
P
NT
i1
DS
i
r%
; wu% P
L
; P

WT
( )
28
However, the wind generation of system will be curtailed if the
WTGs power output increase is greater than the system ramping
down capacity. That is
P
U
WT
P
0
WT
TDR 29
Since there is no other means of increasing the output of WTGs, the
infeasible state will be found when the system ramping up capacity
cannot absorb the WTGs power output decrease.
3.2. Initialization using a constraint satisfaction technique
A good initial solution could enhance the possibility to obtain a
better solution. However, it is easily trapped in local minima since,
with a single initial solution, it is hard to explore the search space,
where the global optimal solution exists. In this paper, the stochas-
tic technique is applied for the proposed ESA to generate a popula-
tion of NP initial candidate solutions at random and nds solution
in parallel using a simulated annealing procedure. Let rand() be
uniform random value in the range [0, 1]. The initial power outputs
of generating units, without violating unit capacity constraints, are
generated randomly by
P
WT
P
D
WT
rand P
U
WT
P
D
WT
30
P
i
P
min
i
rand P
max
i
P
min
i
31
To satisfy the power balance equation, a dependent thermal gener-
ating unit is arbitrarily selected among the committed NT units, and
the output of the dependent generating unit P
d
is determined by
P
d
P
L
P
WT

X
NT
i1
id
P
i
32
If P
d
violates Eq. (16), the invalid solution is discarded. However,
this method of initialization will lead to a large number of infeasible
solutions being generated due to the power balance constraint, gen-
erator operation constraints and system up/down reserve require-
ments. To restore the feasibility of these constraints, a repairing
strategy is applied to pick one unit at random to increase (or de-
crease) its output by the random or predened step (e.g., 10 MW),
one by one, until it can satisfy all the constraints.
3.3. Solution quality enhancement using two levels of competition
To increase the possibility of exploring the search space where
the global optimal solution exists, there are two levels of competi-
tion in the ESA algorithm. One is the local competition: If the
neighboring solution is better than its current solution, then it is
accepted as the current solution for next generation. If the
neighboring solution is worse than its current solution, we have
START
Read system data
Initialize a population of
candidate solutions NP
Perform evolutionary simulated
annealing procedure for
candidates
Apply DSM for local
optimization
Termination criteria
reached?
END
Obtain solution
YES
NO
Fig. 1. Flow chart for the proposed ESA
*
algorithm.
556 C.-L. Lu et al. / Electrical Power and Energy Systems 30 (2008) 553561
a Boltzmann probability that the neighboring solution be accepted.
The second level competition is the competition among the current
candidate solutions. In the ESA algorithm, we have NP candidate
solutions. That means, we have NP parallel simulated annealing
machines. These NP machines compete with each other. The better
the machine is, the more trials it will have. The selection scheme is
known as roulette wheel selection [13,14]. The calculation pro-
cesses of the proposed ESA are shown as follows:
Step 1: Initialize a population of candidate solutions NP.
Step 2: Set the trial counter C = 1 and the initial temperature
TM
0
.
Step 3: Let TM = TM
0
.
Step 4: Choose a candidate solution using the roulette wheel
selection mechanism, which is assigned as the current solution
X
1
and evaluate its cost function TC
1
.
Step 5: Generate a neighboring solution X
2
through a random
perturbation of the current solution and evaluate its cost func-
tion TC
2
.
Step 6: Perform the acceptance test: If TC
2
6 TC
1
, accept TC
2
as
the current best solution, set X
1
= X
2
and TC
1
= TC
2
, and go to
step 7.If TC
2
> TC
1
, generate a random number q from the uni-
form distribution in the range [0, 1]. If
expTC
1
TC
2
=TM P q 33
accept the X
2
as current best solution. Set X
1
= X
2
and TC
1
= TC
2
.
Otherwise, reject the trial solution X
2
.
Step 7: If C < NP, set C = C + 1 and go to step 4.
Step 8: If the equilibrium condition is satised, go to step 4.
Otherwise, go to step 9.
Step 9: Check the stopping criterion. If satised, go to stop;
otherwise, decrease the temperature according to some per-
ceived rules. A simple cooling schedule is used in this paper as
follows. Go to step 4.
TM a TM 34
where a is a constant smaller than but close to 1.
To preserve the solution optimization, the nal solution must
be the best one of all the states visited since the present solu-
tion may be worse than some of the solutions searched. It can
be terminated when the maximum number of iterations is
reached.
3.4. Non-uniform mutation operator for generating trial solutions
Exploration on current solutions begins with nding the best
trial solutions for improvement. In forming the new trial solution,
the current solution is usually perturbed according to the Gaussian
probability distribution function (GPDF) [13]. After adding a Gauss-
ian random number to current solution, the new trial solution may
violate constraint given by (16). In this paper, we follow a different
approach to generate a trial solution. The proposed random-per-
turbation scheme with a non-uniform mutation operator is used
to satisfy the unit capacity constraints. The operator is described
below
P
mut
WT

P
WT
DTM; P
U
WT
P
WT
if z 0
P
WT
DTM; P
WT
P
D
WT
if z 1
(
35
P
mut
i

P
i
DTM; P
max
i
P
i
if z 0
P
i
DTM; P
i
P
min
i
if z 1
(
36
DTM; YY YY 1 n
TM=TM
0

37
where z is a random bit; function D(TM, YY) returns a value in the
range [0, YY]; and n is a random oating-point number in the inter-
val [0, 1]. At the beginning of the temperature, the mutation oper-
ator makes a uniform search to explore the search space where
the global optimal solution exists. The probability of the value re-
turned (D(TM, YY)) being close to 0 increases with TM. The muta-
tion operator in later stages narrows the search around the local
area resembling a hill-climbing operator. However, it is quite likely
that a randomly generated solution is infeasible owing to the large
set of operational constraints inherent in the WTCD problem. In this
paper, a repairing strategy is also used to generate a feasible candi-
date solution as shown in Section 3.2.
3.5. Local optimization using the direct search method
Usually, the stochastic search technique can identify a near glo-
bal region but slows in a nely tuning local search. In contrast, the
local searching technique can climb hills rapidly but is easily
trapped in local minima. In this paper, the ESA algorithm was
responsible for global exploitation and the DSM algorithm was
used to local optimization with the current solutions of the ESA
as the starting points. Like many local search techniques, the
DSM is more sensitive to the initial starting points. To further
weaken the dependence of nding the global optimal solution on
the initial starting solutions, the selection of calculation step S
in the direct search procedure is vital to the success of DSM to nd
the global optimal solution. In this study, the DSM with large initial
calculation step S
1
and small reduced factor K is usually com-
mended to enhance its search capacity that leads to a higher prob-
ability of obtaining the global optimal solution. Many nonlinear
characteristics of the generators can be handled properly in the di-
rect search procedure with light computational expenses. The
main attractive feature of multi-level convergence level is to re-
duce the step size gradually for improving the solution quality
and performance. The details for solving the extended economic
dispatch problem are the same as that in [18].
3.6. Evaluation of AGC using the proposed WTCD software
Integration of non-utility wind power plants into the existing
isolated system is subject to signicant restrictions, affecting their
expected energy yield and hence the feasibility of the investments.
It appears that a large integration of non-utility wind capacity
would require further changes to power system operation that sig-
nicantly increase operating costs. However, it is also found that a
public utility can reduce its energy cost by purchasing the electrical
wind energy froman IPP. The effects of total wind power generation
and additional up/down reserve requirements are consistently re-
ected in the total production cost. The price for purchase of wind
energy from an IPP is based on the energy cost that a utility can
thereby avoid. The proposed algorithm can be used for the evalua-
tion of AGC to maximize the contribution of wind farms owned by
the IPP to reducing the cost of thermal dispatchinthe existing utility
system. Therefore, the assessment of the AGC can be readily quanti-
ed by using the developed WTCD software to simulate the system
operationwithand without the inclusionof non-utility wind energy
for the dispatch period. The overall procedure of the proposed ap-
proach for determining the AGC can be stated as follows:
Step 1: Simulate the system operation considering the inclusion
of IPP wind energy. Compute the system fuel cost (FC1).
Step 2: Simulate the system operation neglecting the inclusion of
IPP wind energy. Compute the system fuel cost (FC2).
C.-L. Lu et al. / Electrical Power and Energy Systems 30 (2008) 553561 557
Step 3: The avoided generation cost (AGC) is
AGC FC2 FC1 38
The goal of this study is to develop a systematic planning process
for the system operator to assess the impact and economic ben-
ets of the installation of non-utility wind farms. The developed
WTCD software is also a useful tool for appropriate assessment
of the added cost to cover the unpredictable wind generator out-
put variations. Knowing the AGC can help system operator to
purchase wind generation from an IPP and how much to pay
for this service.
4. Numerical experiments
To verify the feasibility and effectiveness of the proposed algo-
rithm, several test systems were simulated. The proper value of the
wind power penetration level in the study is always set to be P
U
WT
without adjusting the wind generation since a large wind power
penetration will usually achieve maximum fuel cost saving of ther-
mal generation. The reserve constrained economic dispatch of
modied system load due to wind generation can, thus, be com-
puted using the proposed algorithm. After testing and evaluating
different parameter combinations, parameters of the proposed
algorithm used in the three examples are listed in Table 1 for clar-
ity. All the computations are performed on a PC Pentium IV/
3.0 GHz and two computer programs were developed in FORTRAN:
ESA: Evolutionary simulated annealing algorithm.
ESA
*
: Evolutionary simulated annealing algorithm with local
optimization.
The studied cases are stated in detail as follows:
4.1. Example 1: test for a 13-unit thermal system
In the rst example, the thirteen-generator example is given to
illustrate the solution quality and performance of the proposed ap-
proach. The system unit data is given in Table 2 and the load de-
mand is 2520 MW [13,17]. Network losses and reserve
requirements of the system are neglected for comparison. The tra-
ditional approaches, such as lambda-iteration dispatch method
cannot be used to solve the problem due to its non-smooth fuel
cost function. The same multiple minimum problem has been
solved by the HSS in [16], the EPSQP in [17] and the PSOSQP
in [17]. Table 3 depicts the numerical results of various methods.
From these results, although multiple local minimum solutions ex-
ist in this studied case, the proposed ESA can still nd a better solu-
tion than many existing techniques. To investigate effects of initial
trail solutions on the nal results, different initial random solutions
were given to the ESA approach. Table 4 shows the dispatch results
under various population sizes for 100 trial runs. In these test
cases, the proposed ESA can easily obtain the satisfactory solutions
using two levels of competition. However, only the near global
optimal solution ($ 24174.17) can be obtained by the proposed
ESA approach. To examine the merits of the proposed ESA
*
algo-
rithm, Table 5 shows the worst cost, average cost, and best cost
achieved. The solution quality could be enhanced as the population
Table 1
Best parameter setting of the ESA
*
Parameter Example 1 Example 2 Example 3
Initial temperature (TM
0
) 900 900 600
Final temperature (TM
f
) 1.0 1.0 1.0
Cooling schedule (a) 0.96 0.98 0.9
Maximum number of iterations
at each temperature (ITER
max
)
30 80 200
Initial calculation step (S
1
) 80 MW 80 MW 80 MW
Reduced factor (K) 1.5 1.5 1.5
Predetermined resolution (e) 0.0001 MW 0.0001 MW 0.0001 MW
Table 2
Parameters for the 13 thermal units
Unit no. P
max
i
P
min
i
a
i
b
i
c
i
e
i
f
i
1 680 0 550 8.1 0.00028 300 0.035
2 360 0 309 8.1 0.00056 200 0.042
3 360 0 307 8.1 0.00056 200 0.042
4 180 60 240 7.74 0.00324 150 0.063
5 180 60 240 7.74 0.00324 150 0.063
6 180 60 240 7.74 0.00324 150 0.063
7 180 60 240 7.74 0.00324 150 0.063
8 180 60 240 7.74 0.00324 150 0.063
9 180 60 240 7.74 0.00324 150 0.063
10 120 40 126 8.6 0.00284 100 0.084
11 120 40 126 8.6 0.00284 100 0.084
12 120 55 126 8.6 0.00284 100 0.084
13 120 55 126 8.6 0.00284 100 0.084
Table 3
Comparison of dispatch results for the load of 2520 MW in the system Example 1
Unit HSS [16] EPSQP[17] PSOSQP[17] ESA ESA
*
1 628.23 628.3136 628.3205 628.3068 628.3185
2 299.22 299.1715 299.0524 298.8529 299.1990
3 299.17 299.0474 298.9681 298.7195 299.1990
4 159.12 159.6399 159.4680 159.7211 159.7330
5 159.95 159.6560 159.1429 159.5390 159.7330
6 158.85 158.4831 159.2724 159.6340 159.7328
7 157.26 159.6749 159.5371 159.0156 159.7328
8 159.93 159.7265 158.8522 159.6087 159.7329
9 159.86 159.6653 159.7845 159.0345 159.7329
10 110.78 114.0334 110.9618 76.3879 77.3996
11 75.00 75.0000 75.0000 77.1473 77.3996
12 60.00 60.0000 60.0000 92.2443 92.3998
13 92.62 87.5884 91.6401 91.7883 87.6868
Cost ($/h) 24275.71 24266.44 24261.05 24174.17 24169.92
Table 4
Comparison of results with 100 trail tests for the load of 2520 MW in the system
Example 1 using the ESA algorithm
NP 1 10 50 100
Worse 24804.58 24300.80 24231.84 24229.53
Average 24494.21 24221.69 24197.24 24184.71
Best 24245.04 24174.75 24174.54 24174.17
NTS 0 0 0 0
ACT (sec.) 0.023 0.24 1.23 2.53
NP: number of initial candidate solutions.
NTS: number of times reached satisfactory solution ($ 24169.92).
ACT: average computation time for 100 trail tests.
Table 5
Comparison of results with 100 trail tests for the load of 2520 MW in the system
Example 1 using the ESA
*
algorithm
NP 1 10 50 100
Worse 24358.83 24225.62 24174.08 24169.92
Average 24201.82 24179.85 24169.96 24169.92
Best 24169.92 24169.92 24169.92 24169.92
NTS 35 62 99 100
ACT (sec.) 0.025 0.26 1.35 2.72
NP: number of initial candidate solutions.
NTS: number of times reached satisfactory solution ($ 24169.92).
ACT: average computation time for 100 trail tests.
558 C.-L. Lu et al. / Electrical Power and Energy Systems 30 (2008) 553561
size NP increased. The relative frequency of convergence for the
proposed algorithm is also listed in Table 6 for each cost range
among 100 randomly initiated trials. Table 6 reveals that the pro-
posed ESA
*
algorithm has provided the satisfactory solution ($
24169.92) with a high probability to demonstrate its effectiveness
and efciency. This test case study converges within 2.72 s for 100
trail tests when the value of NP is chosen to be 100.
4.2. Example 2: test for a 13-unit with an equivalent wind generator
In this example, the same thirteen-unit thermal system with an
equivalent wind generator owned by an IPP is considered. For sim-
plicity, all of wind generators are represented an equivalent wind
plant and the available wind power generation of the equivalent
wind generator, which can be calculated from the wind power
curve, is assumed to be known in the studied cases. The system
up-spinning reserve requirement without considering wind power
generation is assumed to be 250 MW. Table 7 shows the dispatch
results to illustrate the effects of non-utility wind power genera-
tion on the system up/down reserve requirements. Ignoring the
wind power generation, only the system up reserve requirement
(250 MW) is required to respond to unit forced outages and load
forecasting errors. As shown in the case 2.1, it is obvious that both
up and down spinning reserves meet the requirements without
changing the dispatch pattern. In the case 2.2, it can be seen that
1000 MW of non-utility wind power generation causes a
200 MW increase in the need for up/down reserves. In the case
2.3, signicant amount of wind power generation (1500 MW) will
be curtailed since several technique constraints must be observed.
As shown in the forth columns of Table 7, the maximum wind
power penetration is restricted to be 1008 MW due to the wind
power penetration limit, which is decided by the system load
demand (wu% = 40%). Table 8 shows details of the best solutions
obtained by the proposed ESA
*
algorithm. Table 9 shows the
dispatch results under various population sizes for 100 trial runs
in case 2.2. The proposed ESA
*
algorithm has proven that it is a
reliable solution procedure by producing the satisfactory solution
($ 15628.77) 100 times as shown in the fth columns of Table 9.
Table 10 depicts the numerical experiment results under various
load demands to illustrate the effect of wind generator capacity
on AGC in the system. It can be observed that the AGC increases
with an increase in the wind energy supplied by the IPP to the util-
ity. However, the amount of non-utility wind power generation
would need to be curtailed since several technique constraints
must be observed. For example, the maximum wind power pene-
tration is restricted to be 928 MW at the load of 2320 MW. The
AGC increase will also saturate ($ 7858.30) when P
WT2
is greater
than 1000 MW. The suitableness of the algorithm presented in this
paper to the solution of the optimal wind-thermal coordination
dispatch is, thus, conrmed.
4.3. Example 3: test for the practical Taiwan power system
In the last example, the proposed ESA
*
approach was numeri-
cally tested on the Taiwan power system with 52 thermal units
Table 6
Comparison of the relative frequency of convergence in the ranges of cost under
various NP using the ESA
*
algorithm
Method:
ESA
*
Range of cost ($)
24360
24300
24300
24200
24200
24180
24180
24175
24175
24170
24169.92
NP = 1 7 22 20 10 6 35
NP = 10 0 8 14 9 7 62
NP = 50 0 0 0 0 1 99
NP = 100 0 0 0 0 0 100
Table 7
Comparison of system reserves and production cost for three different cases in the
system Example 2
Case (MW) 2.1 2.2 2.3
P

WT1
=P

WT2
0/0 0/1000 0/1500
P
WT1
/P
WT2
0/0 0/1000 0/1008
Up reserve requirement 250 450 451.6
Down reserve requirement 0 200 201.6
Actual up reserve 390.88 592 592
Actual down reserve 592 388 316
Production cost ($/h) 24169.92 15628.77 15553.75
Table 8
Comparison of dispatch results for three different cases in the system Example 2
Unit (MW) Case 2.1 Case 2.2 Case 2.3
1 628.3185 448.7989 538.5587
2 299.1990 222.0018 223.9751
3 299.1990 149.5996 149.5996
4 159.7330 109.8665 60.0000
5 159.7330 60.0000 60.0000
6 159.7328 109.8665 60.0000
7 159.7328 60.0000 60.0000
8 159.7329 60.0000 60.0000
9 159.7329 109.8665 109.8665
10 77.3996 40.0000 40.0000
11 77.3996 40.0000 40.0000
12 92.3998 55.0000 55.0000
13 87.6868 55.0000 55.0000
P
WT1
/P
WT2
0/0 0/1000 0/1008
Table 9
Comparison of results with 100 trail tests for the load of 2520 MW in the system
Example 2 considering the wind power generation of 1000 MW
NP 1 10 50 100
Worse 15920.37 15678.96 15633.63 15628.77
Average 15793.60 15641.58 15628.89 15628.77
Best 15729.68 15628.77 15628.77 15628.77
NTS 0 41 94 100
ACT (sec.) 0.026 0.25 1.24 2.59
Parameter setting: r% = 20%; d% = 20%; wu% = 40%.
NP: number of initial candidate solutions.
NTS: number of times reached satisfactory solution ($ 15628.77).
ACT: average computation time for 100 trail tests.
Table 10
Comparison of AGC under various total available wind power generation levels for the
system Example 2
P
L
(MW) P

WT1
=P

WT2
(MW)
P
WT1
/P
WT2
(MW)
Production cost
($/h)
AGC
($/h)
2320 0/0 0/0 22410.23
0/500 0/500 18134.33 4275.90
0/1000 0/928 14551.93 7858.30
0/1500 0/928 14551.93 7858.30
2520 0/0 0/0 24169.92
0/500 0/500 19836.34 4333.58
0/1000 0/1000 15628.77 8541.15
0/1500 0/1008 15553.75 8616.17
2620 0/0 0/0 25136.93
0/500 0/500 20688.69 4448.24
0/1000 0/1000 16455.59 8681.34
0/1500 0/1048 16042.11 9094.82
C.-L. Lu et al. / Electrical Power and Energy Systems 30 (2008) 553561 559
considering valve-point effects [11,18]. The ramp-rate constraints
of the thermal units are set at 50% of its rated capacity. Since the
wind generators can change their output rapidly, the ramp-rate
constraints of the WTGs are neglected in the studied cases. How-
ever, the wind power uctuations have to be absorbed by the sys-
tem ramping capacity. Owing to the limits of the space, the
parameters of the units are not listed. Ignoring the wind power
generation, only the system up reserve requirement (900 MW) is
required to cover the potential loss of the largest unit in the sys-
tem. However, both the up and down spinning reserve require-
ments in the system must increase as the wind power generation
increases to operate the isolated Taiwan power system reliably
and efciently. The equivalent wind generator owned by the public
utility is considered and the total available wind power generation
is assumed to be 2000 MW. The problem is further complicated
when the equivalent wind plant owned by an IPP is present in
the system. The algorithm presented is to achieve the optimal
exploitation of non-utility wind energy sources for the optimal
operation of the system in this scenario.
To illustrate the impacts of incorporating non-utility wind gen-
erators into the system on the existing utility generation schedul-
ing problem, Table 11 gives an analysis of the added costs under
various available wind power generation levels at the load level
of 18,000 MW by using the proposed ESA
*
algorithm. The assess-
ment of the added cost to cover the unpredictable wind generator
output variations can be readily quantied by using the developed
WTCD software to simulate the system operation with and without
the additional up/down reserve requirements for the dispatch per-
iod. In the case 3.1, only 2000 MW of wind power generation
owned by the utility is considered and the added cost is about
95,200 NT $/h to cover the unpredictable wind generator output
variations. In the case 3.2, power from the WTGs increases and
from the thermal generators decreases as the amount of wind
power generation owned by the IPP is chosen to be 2000 MW. It
can be seen that the added cost drops (88,918 NT $/h) because of
the increasing ability of thermal generators to provide reserve with
changing the dispatch pattern slightly. However, in the case 3.3,
the added costs increase rapidly since the thermal units have to
change their generation to satisfy increased up/down reserves
when the amount of wind power generation owned by the IPP is
chosen to be 3000 MW. In this case study, the added cost value
is about NT $ 163,606 for providing this additional reliability and
the corresponding AGC value is NT $ 1642,103 due to purchase
of IPP wind energy by the utility. As a result, the added cost of car-
rying additional up/down reserve requirements can be determined
properly by the algorithm presented.
To understand the integration of non-utility wind generator
capacity in the AGC analysis, wind parks with installed capacities
ranging from 0 MW to 3000 MW will be in operation. In the stud-
ied case, the wind power generation owned by the utility is still as-
sumed to be 2000 MW. Figs. 24 depict the numerical experiment
results under various load demands to illustrate the effect of non-
utility wind generator capacity on avoided generation costs in the
system. The effects of load demand have to be consistently re-
ected in the marginal cost of the utility, which is important factor
in determining the AGC. The economic penetration limit of non-
utility wind generation into an existing utility system can be deter-
mined properly by using the proposed technique. It is also found
that the AGC depend on the degree of wind power penetration le-
vel. As shown in Fig. 2, the AGC increases with P
WT2
at the lightly
load demand (12000 MW). However, the avoided generation costs
are the same when P
WT2
is greater than 500 MW. The low value of
wind power penetration is due to the minimum permissible power
of thermal generators. Fig. 3 shows a variation of AGC with a var-
iation of P
WT2
at the load demand of 16000 MW. A signicant pro-
portion of non-utility wind power generation can be integrated
into the systemand the AGC increases with P
WT2
to reect the mar-
ginal cost of the utility. However, in a heavy load demand, such as
18,000 MW, the AGC becomes more signicant since the displace-
ment of expensive fuels in the thermal units system can, therefore,
be obtained by a high wind power penetration level. The results
show that it could provide a much greater proportion of non-utility
wind generator capacity that can be integrated into the existing
Taiwan power system. Numerical results also give a good indicator
to provide valuable information for IPP to achieve the optimal
exploitation of wind energy sources.
Table 11
Comparison of added cost and AGC under various total available wind power
generation levels for the Taiwan power system
Case 3.1 3.2 3.3
P

WT1
=P

WT2
(MW) 2000/0 2000/2000 2000/3000
P
WT1
/P
WT2
(MW) 2000/0 2000/2000 2000/3000
r% 0% 20% 0% 20% 0% 20%
Fuel cost (10
6
NT$/h) 10.436 10.531 9.262 9.351 8.725 8.889
Added cost (NT$/h) 95200 88918 163606
AGC (NT$/h) 0 1179842 1642103
Parameter setting: NP = 100; r% = 20%; d% = 10%; wu% = 40%.
0
0.02
0.04
0.06
0.08
0 500 1000 1500 2000 2500 3000
Available wind generation owned by the IPP (P
WT2
)
(MW)
A
G
C

(
1
0
6

N
T
$
)
Fig. 2. Comparison of AGC under various P
WT2
for the load of 12,000 MW.
0
0. 5
1
1. 5
0 500 1000 1500 2000 2500 3000
Available wind generation owned by the IPP ( P
WT2
)
(MW)
A
G
C

(
1
0
6

N
T
$
)
Fig. 3. Comparison of AGC under various P
WT2
for the load of 16,000 MW.
0
0.5
1
1.5
2
0 500 1000 1500 2000 2500 3000
Available wind generation owned by the IPP ( P
WT2
)
(MW)
A
G
C

(
1
0
6

N
T
$
)
Fig. 4. Comparison of AGC under various P
WT2
for the load of 18,000 MW.
560 C.-L. Lu et al. / Electrical Power and Energy Systems 30 (2008) 553561
5. Conclusions
Integration of non-utility wind power plants into the existing
isolated system would require further changes to power system
operation due to the uncertainty posed by wind power. To illus-
trate the impacts of incorporating non-utility wind generators into
the system on the existing utility generation scheduling problem, a
hybrid algorithm based on a combination of evolutionary simu-
lated annealing (ESA) algorithm and direct search method (DSM)
is developed to solve the non-smooth economic dispatch problem
considering wind power generation. Using the parallel searching
mechanism with two levels of competition strategy, the ESA algo-
rithm can give a good direction to identify the near global region. A
local optimization technique, which utilizes the DSM approach, is
also used as a nely tuning to determine the global optimal solu-
tion at the nal with light computational expenses. Furthermore,
a systematic planning process is developed for appropriate assess-
ment of the added cost to cover the unpredictable wind generator
output variations. The developed WTCD software is also a useful
tool for the system operator to evaluate the appropriate price that
it has to pay to the IPP for the exchange of electrical energy. The
economic penetration limit of non-utility wind generation into
an existing utility system can be determined by using the proposed
technique. The results show that the algorithm is reliable, and the
solution is reasonable. The results also show that it could provide a
much greater proportion of non-utility wind generator capacity
that can be integrated into the existing Taiwan power system.
However, the utilization factor of non-utility wind energy sources
will be dropped at the lightly load demand. The avoided generation
costs are also saturated due to the minimum permissible power of
thermal generators. Numerical experiments are included to pro-
vide valuable information in both operational and planning prob-
lems and to give a good indicator to invest in new power plants
using non-utility wind energy sources for the Taiwan power sys-
tem in the future.
Acknowledgements
The authors deeply appreciate the support of Taiwan Power
Company for providing the system data and test cases. Financial
supports under the Grant NO. NSC96-2221-E-159-018 from the
National Science Council, Taiwan, ROC are acknowledged.
References
[1] Yue CD, Liu CM, Liou Eric ML. A transition toward a sustainable energy future:
feasibility assessment and development strategies of wind power in Taiwan.
Energy Policy 2001;29:95163.
[2] Chang TJ, Wu YT, Hsu HY, Chu CR, Liao CM. Assessment of wind characteristics
and wind turbine characteristics in Taiwan. Renew Energy 2003;28:85171.
[3] Nogaret E, Stavrakakis G, Kariniotakis G. An advanced control system for the
optimal operation and management of medium size power systems with a
large penetration from renewable power sources. Renew Energy
1997;12(2):13749.
[4] Tripathy SC, Lakshmi SR, Balasubramanian R. Production costing and economic
analysis of power systems containing wind energy conversion systems. Energy
Convers Manage 1998;39(7):64959.
[5] Hongwei Z, Xuanfu L, Yuzhao F. Optimal operation of large scale wind-diesel
generation systems. Renew Energy 1999;16:8514.
[6] Dokopoulos PS, Saramourtsis AC, Bakirtzis AG. Prediction and evaluation of the
performance of wind-diesel energy systems. IEEE Trans Energy Convers
1996;11(2):38593.
[7] El-Tamaly HH, Elbaset MAA. Impact of interconnection photovoltaic/wind
system with utility on their reliability using a fuzzy scheme. Renew Energy
2006;31(15):247591.
[8] Bouzguenda M, Rahman S. Value analysis of intermittent generation sources
from the system operations perspective. IEEE Trans Energy Convers
1993;8(3):48490.
[9] Rahman S, Bouzguenda M. A model to determine the degree of penetration and
energy cost of large scale utility interactive photovoltaic systems. IEEE Trans
Energy Convers 1994;9(2):22430.
[10] Doherty R, OMalley M. A new approach to quantify reserve demand in
systems with signicant installed wind capacity. IEEE Trans 2005;PWRS-
20(2):58795.
[11] Chen CL. Simulated annealing-based optimal wind-thermal coordination
scheduling. IET Gener Transm Distrib 2007;1(3):44755.
[12] Wood AJ, Wollenberg BF. Power generation operation and control. 2nd
ed. New York: Wiley; 1996.
[13] Wong KP, Wong YW. Genetic and genetic/simulated annealing approach to
economic dispatch. IEEE Proc 1994;141(5):50713.
[14] Walters DC, Sheble GB. Genetic algorithm solution of economic dispatch with
valve-point loading. IEEE Trans 1993;PWRS-8(3):3251331.
[15] Yang HT, Yang PC, Huang CL. Evolutionary programming based economic
dispatch for units with non-smooth fuel cost functions. IEEE Trans 1996;
PWRS-11(1):1128.
[16] Das DB, Patvardhan C. Solution of economic load dispatch using real coded
hybrid stochastic search. Electr Pow Energy Sys 1999;21:16570.
[17] Victoire TAA, Jeyakumar AE. Hybrid PSOSQP for economic dispatch with
valve-point effect. Electr Pow Syst Res 2004;71:519.
[18] Chen CL, Lee TY, Jan RM. Optimal wind-thermal coordination dispatch in
isolated power systems with large integration of wind capacity. Energy
Convers Manage 2006;47(1819):345672.
C.-L. Lu et al. / Electrical Power and Energy Systems 30 (2008) 553561 561