You are on page 1of 8

2014 Ninth International Conference on Ecological Vehicles and Renewable Energies (EVER)

Optimal Design of a Hybrid Renewable Energy


System
Carmen Delgado1 and Jos A. Domnguez-Navarro2
University of Zaragoza
Mara de Luna 3
Zaragoza, Spain, 50018
Email: carmend@hotmail.com1 , jadona@unizar.es2

Abstract In a micro-generation system that includes


renewable energies, the reliability assessment might
consider the random behavior of the renewable resources.
When a planning of this kind of generation systems are
carried out, the objectives of interest must be optimized
based on outcome of each one, since join them in a single
function is not always possible or recommend.

In this paper a multi-objective optimization of a


generation system is done, using as objectives the cost of
energy, two different reliability indexes and the percentage
of renewable energy used. The power generation system
combines solar and wind energies, diesel generators as
conventional source and the possibility to take energy from
the grid. Each energy source and the load was modeled as a
multi-state system (two or more performing levels), in order
to represent in a closest form their variable nature. This
multi-state representation of the generators and loads
combined with the Universal Generating Function enables
the estimation of the reliability indexes of the system,
reducing the computation time, while maintaining adequate
results.

Universal Generating Function


I. INTRODUCTION

In recent years the contribution of renewable energies


in electricity supply has increased. The integration of
renewable energy has an important impact in the planning
of generation systems. However, in this kind of planning
the elements of power systems are considered commonly
as elements with only one or two performance states. The
renewable energy systems have attributes that cannot be
modeled with only binary states, defined by, perfectly
working and out of operation. Other performance levels
should be introduced in order to consider the randomness
of renewable energy resources. The multi-state system
(MSS) model is an alternative to represent this
randomness, since it describes the elements behavior by
using their performance states and the probability of these
states.

ACRONYM

In the field of reliability assessment of power systems,


the methodologies have to be able to incorporate the
variability of the renewable resources. This situation has
been not explored enough, as is exposed in [1] by Zhou.
The Universal Generating Function (UGF) [2]-[6] is
among others like Monte Carlo Simulation (MCS) [7][10] and Markov chains [11]-[12], one of the tools that
can be used on the study of the reliability in power
generation systems, for the case where such systems
exhibit probabilistic performance. In case of the MCS and
Markov chains, their main disadvantage lies in the huge
computational requirements and the time needed to
achieve an acceptable solution. The UGF is an analytical
method based on algebraic operations and MSS modeling,
that can reduce the computational time and delivers
accurate solutions.

Availability
Expected Energy Not Supplied
Force Outage Rate
Genetic Algorithms
Non-dominated Sorting Genetic Algorithm
Monte Carlo Simulation
Multi-State System
Non-dominated Sorting Genetic Algorithm
Performance Distribution
Percentage of Renewable Energy

In the planning of generation systems, the cost has


been the objective most frequently optimized. Nowadays,
with the interest of introducing renewable energies in the
power generation systems, the reliability as objective has
a more important role than before. If there are in the
planning process a similar interest in both objectives, cost
and reliability, it should be considered that reliability
estimation is a most consuming process. Then the time
and resources used in the whole optimization could be
minimized if it is improved the reliability evaluation.

The proposed system was compared with Monte


Carlo Simulation in the reliability assessment objective to
evaluate the performance of the Universal Generating
Function in this case. Finally the proposed system gives us
as a result a variety of utilities that can serve as a basis for
further evaluation in the whole planning process and it
allows us to observe the changes produced on the solutions
planning due to the impact of the renewable energies on
reliability.
Keywords:

power systems reliability, renewable


generation, universal generating function, artificial
intelligence.
A
EENS
FOR
GA
NSGA
MCS
MSS
NSGA
PD
PRE

UGF

978-1-4799-3787-5/14/$31.00 2014 IEEE

There are some studies concerning to the optimal


planning of power generation systems that have the cost
or reliability as objectives to be optimized, as it is the case
of this work. Some of these are: [13] in it the UGF and
GA are used to determine the optimal reserve structure for
a restructured generic power generating system that is
composed by hydropower and conventional generators. In
[14], it was proposed a hydro/thermal reserve allocation
method to determine the reserve required for power
systems with high wind power penetration, using the UGF
to model the generation system and the load. In [15], they
proposed a reserve expansion method to decide the
conventional reserve required by power systems with high
wind power penetration, this proposal is based on
reliability. They used the UGF and MSS model to
evaluate the wind and conventional (diesel) energies as
resources.
If it is considered the huge computational requirement
that reliability indexes calculation could take. The
combination of methodologies such as those used in
previous works may be adapted to reduce the computation
time needed for the power generation systems planning,
when renewable energy is included. This was done in the
present work, using not only wind and conventionaldiesel energy, as commonly is done. The solar
photovoltaic energy was added too, and the grid was
proposed as additional power source. All of them were
combined in a multi-objective optimization algorithm
looking for a group of better power generation solutions
considering a load that varying on time. All elements
were considered as multi-state systems, being the
conventional generation and the reserve modeled with
two states. Renewable energy and load were modeled
with a variable number of states. This multi-objective
optimization included objectives such as reliability,
energy cost and the percentage of renewable energy
(PRE) used in the installation. As the SMC is frequently
used to obtain some reliability indexes, a comparison
between it and some of the result obtained with UGF was
done, for validation purposes.
The main contributions of this paper are:
- The proposal of a multi state model that includes
different types of energies: solar, wind, diesel and grid
power, considering the random behavior of some sources
and including the differentiation between hours with and
without solar radiation for the solar energy.
- Unify the reliability assessment of all energy systems
mentioned before.
- Demonstrate
for this case the
performance
improvement of UGF against SMC regarding the
computational time required.
- Test the impact of different multi-states numbers for
the representation of the elements.

- Depict through multi-objective optimization, the


impact of combining different energies on the cost and
reliability of the resultant systems.
This paper is organized as follows. In Section II, it is
introduced to the UGF. In Section III, it is established the
use of the UGF for the reliability evaluation using MSS,
specifying each of the elements considered in the
generation system. Section IV presents the algorithm used
in multi-objective optimization. Section V shows the
study cases and results, and lastly the conclusions are
drawn.
II. UGF DEFINITIONS
A.

Multi-State System(MSS)

An element that has different performance levels can


be represented as a MSS. That means, the element is
modeled as a discrete random variable X, which has a
finite number of possible values represented by a vector
x=(x0,,xi), and each value in x is associated to some
probability in the vector p=(p0, ,pi). The resulting
mapping from the xipi is usually defined as
probability mass function (pmf) or performance
distribution (PD) of
X. The application of the z
transform to the discrete random variable allows to
represent the PD as a polynomial, which is referred as the
u-function, u(z). Where kX represents the possible states
of performance of the variable X [16].
(1)

( )=
B.

MSS and the UGF

If two independent elements (G and W) are represented


with its respective u(z) and them are related through the
composition operator
, the U-function of the entire
system, denoted by U(z) is obtained with:
( )=

(2)

If the number of multi-state independent elements in


the system is n, such that:
( )=

( )=

,,

Then the U-function of the entire system will be:


( )=

,,

(3)

In case of multi-state elements are connected in


parallel, the composition operator is
. This technique

based in the z transform and the composition


operator
, applied to a function with arbitrary
structure, is called UGF (Universal Generating Function)
[17].
III. ESTIMATION OF THE MSS RELIABILITY

The estimation of reliability for a power system supply


is achieved through the UGF, based on two indexes,
Expected Energy Not Supplied (EENS) and Availability
(A). Firstly, it is presented the way to obtain the PD of
each element that compose the system, and later, it is
explained the combination of these u-functions to get the
reliability of the whole system.
A.

The annual solar radiation data, which consist of


monthly averages for each hour of the year is used to
obtain the model of the MSS. These radiation values are
separated into two groups of radiation schedules. In the
first group all hours with zero radiation in the year (H0)
and the other group, with the remaining hours (H1). Each
group has its own performance distribution. The H0-group
has a unique performance level, zero, 0, with a
probability of 1. For the H1-group, the performance
distribution is obtained through a frequency table
(containing all H1 radiation values), with so many classes
as performance levels are required to model the MSS. The
averages of the radiation values belonging to each class
are considered the states xi, and the relative frequencies
are the probabilities, pi.
( ) = (1)

(4)

( )=

Since uH0(z) and uH1(z) describe only the solar


radiation source, then each solar installation to assess is
considered as a binary element, with performance levels,
operation and out of operation, and the probability to be
out of operation is described as FOR (Force Outage Rate).
The u-function for the solar installation is then:
(1

B.

( )=

( )

( )

( )=

( )

( )

(6)

Wind system

The wind power of the installations is roughly


estimated basing on a series of wind speeds values that

(7)

)]

The power related with each wind speed (V) is


approximated using the equations in (8), where Vci is the
cut in speed, Vr is the nominal speed, Vco is the cut out
speed and Pr is the nominal power of the wind turbine.
0

(8)

Pr
0

The estimated wind power is used to create a


frequency table, with so many classes as performance
levels are required for the MSS model of the wind turbine
power. The average values belonging to each class are
considered the states xi, and the relative frequencies are
the probabilities, pi. The values 0 and Pr are two fixed
states in xi. Then, the performance distribution for the
wind power generated by the wind turbine WTGn is:
(9)
( )=

As in the solar system model, the u-function uPWn(z),


only describes the interaction between the wind and the
turbine. Then, each wind turbine or farm to assess is
considered as a binary element, with performance levels
operation and out of operation and the probability to be
out of operation is described as Force Outage Rate (FOR).
The u-function for the wind installation is then:
( )=

(1

(10)

Again, to obtain the whole wind installation u-function


considering its available wind resource, uWTGn(z) must
be combined with uPEn(z) , as in (11):
(11)

( )=

(5)

To obtain the whole solar installation u-function,


considering its available radiation, uH0(z) and uH1(z) ,
they must be combined with uSn(z) , through the
multiplication operator as follows:

(1

( )=

Photovoltaic solar system

( )=

are analyzed to obtain the Weibull parameters -shape


factor (k) and scale factor (c)-. Later, these parameters are
used in the simulations to generate the wind speed values
(V) using the approximation in (7), where x are random
uniform distribution values [18].

C.

Diesel system and grid

Diesel generators Dn are considered as a two-state


MSS. The states are operation (PDn) and out of operation
(zero power). The u-function is:
( )=

(1

(12)

Meanwhile, the grid (Rn) is considered in this


simulations as the last reserve source. It is established as a
two-state MSS being zero and full availability (PRn power)
these states. The u-function for this MSS is then:

( )=
D.

(1

(13)

Integrated generation system

The u-functions of multi-state elements such as solar,


wind, diesel or reserve type elements are combined in a
parallel structure to form the whole generation system
(14). Assuming that for each type of multi-state element j,
it can be used until mW, mS, mD or mR times for wind,
solar, diesel or reserve element types respectively. If the
system could use NW, NS, ND or NR different type of
elements, the whole generation system is represented
( ) and
( ) functions:
through the following
( )=

( )

( )=

( )

( )=

( )

( )=

( ) , repeted

times

( ) , repeted

( ), ,
( ), ,

( )=

( )

( ), ,

( )

(17)
,

, , )

, , )

]
( )

(15)

Demand model

( )=

( ))

= (

Since load (L) is variable in time, it is also modeled as


a MSS. Its performance distribution is obtained from the
average hourly demand curve for a whole year. Again it is
generated a frequency table with three classes
corresponding to: peak (p), flat (f) or valley(v) states, and
probabilities are the relative frequencies for each class.
Therefore, the u-function for a load with PP peak value is:
F.

, , )

( ))

The MSS correspond to models developed at the A, B


and C sections, with one exception. For the power grid,
the operation state and out of operation state have always
the same probabilities described before. However the
value of the operation state could change depending on
the consumption. Then (15) is obtained from (13):

E.

, , )

( )

( ),

(1

= (

( )=

UGH0(z) is for zero radiation hours and UGH1(z) is for


the remaining hours of the day . If in one solution some
element is not used, its U(z) does not be used.

times

( ), ,

( ),

times

( ) , repeted

( )=

( )=

times

( ) , repeted

( )

(14)

unavailability state. Both estimations used the results


obtained from equations (14) to (16), and reliability
indexes are estimated through the expectation (E) of:

(16)

Complete Reliability model for the system

The availability is considered as the time for which the


system performs the function for which it is designed,
being then the FOR its complement. The EENS accounted
the energy required by the demand, but could not be
delivered to it from the generation system, because of an

(18)

IV. MULTIOBJECTIVE EVOLUTIONARY ALGORITHM

The genetic algorithms (GA) is an optimization tool


that can search optimal solutions for linear or nonlinear
systems and there are employed for system optimizations
over one or more objectives (multi-objective). For only
one objective, the result is a solution that optimizes the
object of interest and meets restrictions imposed in the
system, while multi-objective option gives as result, a
group of solutions where each one is a better solution to
another, at least in one of the objectives tested. In this
work a multi-objective optimization was made, using
concepts of the evolutionary algorithm NSGA-II (Nondominated Sorting Genetic Algorithm) [21], such as, the
group density operator and the structure of non-dominated
fronts, to let diversity in the entire non-dominated front of
solutions.
The objective functions considered on the multiobjective optimization were:
(1):

(19)

& )

(2): Min (EENS)


(3): Max (A)
(4): Max(PRE)=

Chromosomes are the encoded representation of the


solutions to the problem for the GA. The chromosome
structure shown in Fig. 1 is used for the simulations
presented in the result section. This structure has: G,

generation units groups that can be solar, wind,


conventional or reserve type, where each of these groups
can in turn contained NG elements to evaluate, and each
of these elements may appear until m times (0 <m <9) on
the chromosome, in order to complete the demand.
Wind
generation

Solar
generation
0

0 9

Ns

0 9

Conventional
(e.g. diesel)

Ne

Reserve (grid)

Nc

0 9

...

Nr

0 9

Fig.1.Chromosome structure.

The optimization algorithm performed is shown in the


flow diagram of Fig.2. More information could be found
in [19].
MSS elements u-functions definition*

Accuracy of results and time response of the UGF

It is compared the results obtained for the reliability


indexes in both, its time performance and result accuracy.
This comparison is depicted for MCS and UGF on the
Tables I and II for two representative combinations
(minimum and maximum simulation cases respectively).
These results are an example for a single combination run
and its time results must be multiplied by the number of
different combinations proved in the optimization
process, with the purpose of understanding the difference
between both methodologies.
As it can be seen, the MCS gives results closer as
those delivered with the UGF, as the number of MCS
sample cycles increases, therefore the computational time
using MCS increases too. Then, in these cases the UGF
allowed estimating the reliability indexes, improving the
computation time for similar results.
Methodology
UGF
MCS (100 cycles)
MCS (500 cycles)
MCS (1000 cycles)

Objective functions were evaluated for PB


(Reliability, energy cost and PRE)

yes

Were finished
genetic cycles?
no

Optimal
solutions

New population , PC, was generated (single crossing


point, parent were chosen with binary tournament and
mutation probability was increased on each cycle)

Add one to genetic cycles done

A.

TABLE 1. COMBINATION WITH ONLY ONE RESERVE

Random initial population (PB) was


generated with Np solutions

Objective functions were evaluated for PC

PT were ordered in fn fronts agree


with dominance and PB was emptied
yes

Fig. 2. Optimization algorithm

EENS
16,32
0,00
19,69
17,62

A
0,96
1,00
0,95
0,96

TABLE 2. COMBINATION WITH: 3 PHOTOVOLTAICS, 3


WIND, 3 DIESEL, 1 RESERVE
Methodology
UGF
MCS(1000 cycles)
MCS(5000 cycles)
MCS(10 000 cycles)

B.

Average time
1,01 ms
1,21 ms
6,51ms
12,7ms

EENS
0,631
0,769
0,717
0,657

A
0,993
0,992
0,993
0,992

Planning optimizations

Cost EENS A, and


CostEENS PRE (solar / wind / together)

no
Solutions in fi were clasified using
the densitity group operator and the
necesary chromosomes to complete
PB were added

Average time
29 s
173 s
279 s
833 s

After it is compared UGF and SMC, here are


presented the multi-objective optimization results
obtained using the algorithm described in the previous
section. Simulations are performed with populations of
500 chromosomes and 20 genetic cycles as stop criterion.
The objectives to optimize were combined in two ways:

PT = PB+ PC

If (fi+ PB) size < Np

V. STUDY CASESAND RESULTS

Front was
added to PB
and i was
incremented

The optimizations results are outlined as follows:


- In Fig.35, it is shown that EENS estimation is not
completely independent from the availability, as it can be
seen in equations (17) and (18) too. This is deducted by
observing that the solutions found by the algorithm are
distributed on a band without occupy most on the surface.

However when the optimization combines EENS with the


percentage of renewable energy, as in the case of Fig.69,
the solutions are located in different zones of the same
surface, showing greater independence among the
objectives.
- When there are some objectives to be optimized,
these could be in conflict, as in the case of the objectives
addressed in these simulations. In Fig.35, it is
appreciated that an increment on reliability usually
increases the cost of energy and vice versa. This can be
noted in each simulation graph as for example, is it
evident comparing the simulation III and IV. Both
simulations used solar and wind energies with the same
data, but in simulation IV was proposed the use of some
diesel units. The Fig.5 is the result to add diesel
generators. It shows that results with lowest energy cost
are very similar to those obtained from the simulation
without diesel generators, and when diesel groups are
added, the cost evidently increases, mainly due to the fuel
costs. While the reliability improved with these diesel
generators.
Another example is found on simulation V (Fig.6). In
this case when the percentage of renewable generation is
optimized, those results of the simulation II with similar
EENS (Fig. 3) incremented their cost. This is mainly due
to randomness of the source that implied greater use of
reserves to maintain the reliability. To evaluate the
individual impact, solar and wind energies have been
optimized by separately. It was found that if it is
optimized the percentage of wind energy (Fig.7)
compared with resulting cost from optimized the solar
energy (Fig.9), the first increased less the cost, for similar
results that those were obtained in simulation II.
- Finally, if the resource potential is increased, like it is
carried out on simulation VII, regard to simulation VI; it
is observed a decrement in cost for a comparable
renewable participation rate. This occurred because wind
resource MSS is modified, and therefore the new
performance states of the wind installation improved its
profitability, while its reliability is maintained and even
enhanced. The simulations above mentioned are shown in
Fig.7 and 8.
The results obtained from some simulations
concerning to the number of states to model an element, it
is shown that the number of states seems to have more
influence on the calculation of A that the EENS. For the
EENS case, the variability does not show a definite
pattern. However, for the availability case, although there
is no identified a particular pattern of behavior, it is
observed that when the number of elements of the
solution or the capacity installed increased, the difference
between models tends to decrease, regardless of using
more or fewer states.

VI. CONCLUSIONS

The optimization methodologies try to efficiently


explore the optimal solutions within the search space,
looking for the better solutions in the shortest possible
time. The GA demonstrates to be an appropriated tool to
find optimal solutions in the proposed case. Although it is
not the only optimization tool, it result a good option due
to its simple implementation and adaptation to the
problem.
The objectives will play an important role in the
optimization process especially when its evaluation has a
direct influence in the performance of the method. In the
case of the multi-objective optimization that minimizes
costs, maximizes reliability and optionally maximizes
renewable use, the most expensive evaluation in time are
the reliability indexes, and further if each element of the
system has a random behavior as is the case of the
renewable sources and load, then, the use of the UGF to
calculate these indexes reduces the whole computation
time, with accurate results. The UGF accuracy and its
reduction in computation time have been proved with the
comparison of some results obtained with SMC and UGF.
The results given by the system allows observing how
the renewable energies affect the EENS and the cost for
better reliabilities. Since the proposed model is able to
capture the long-term variability of the sources, if the data
introduced to the system are suitable, it will be able to
evaluate and decide when a combination would be
inadequate in the future, as is the case of diesel generators
and the cost of diesel, or if is adequate meanwhile the cost
remains in a lower level as is the case of the grid power.
The structure proposed for the solutions is able to
consider the repetitive use of an element or a mixture of
different elements. Which could be important since
elements of the same type are better for maintenance
planning purposes meanwhile the combination of
different elements could avoid problems of reliability by
common cause failures.

Fig.3.Simulation II.Cost-EENS-A optimization.


All sources were included - Case II.

Fig.4.Simulation III.Cost-EENS-A optimization.

Fig. 7.Simulation VI. Same than simulation V,

Only solar and wind sources was considered Case II.

but only %E optimization with c=5. Case II.

Fig. 5.Simulation IV.Cost-EENS-A optimization.


Fig. 8.Simulation VII. Same than simulation VI,

Solar, wind and diesel groups were considered - Case II.

but only %E optimization with c=7 - Case III.

Fig. 6.Simulation V. Cost-EENS-%S-E optimization.

Fig. 9.Simulation VIII. Same than simulation V,

All sources were included - Case II.

but optimizing only the %S. Case II.

APPENDIX
TABLE 3. INSTALLATIONSPROBABILITIES
Renewables

States and probabilities


Operation
Out of operation
0,96 %
0,04 %
0,98 %
0,02 %

Others

States and probabilities


Operation
Out of operation
Wind
Reserve Case I 1
0,9983 %
0,0017 %
Solar
Reserve Case II1
0,9986 %
0,0014 %
Diesel
0,96 %
0,04 %
1
(R.D. 1955/2000, Case I: tariff 3.0.2, interruption of maximum 15 hours and Case II: tariff 3.1.A, interruption of maximum 12 hours).

TABLE 4. PROBABILITIES Y DEMAND STATES


Peak demand

States
Valley
308,15
34,24

Flat
373,35
41,48

450 kW
50 kW

Peak
429,54
47,73

Flat

Probabilities
Valley

Peak

0,167

0,25

0,583

TABLE 5. VALUESOFTHE PARAMETERSAFFECTINGTHE COST CALCULATION


Both cases
Installations
Power (kW)
Investment (k/kW)

Diesel
32
0,328

80
0,194

11
3,0

Case I
Wind
50
2,5
---

80
2,0

The first author would like to thank CONACyT


(National Council of Science and Technology) for
the financial support.
REFERENCES
[1] W. Zhou, C. Lou, Z. Li, L. Lu, H. Current status of research on
optimum sizing of stand-alone hybrid solar-wind power
generation systems. Applied Energy, vol. 87, pp. 380389, 2010.
[2] A.S. Dobakhshari, M. Fotuhi-Firuzabad. A Reliability Model of
Large Wind Farms for Power System Adequacy Studies, Energy
Conversion, IEEE Transactions on, vol. 24, no. 3, pp.792-801,
2009.
[3] G. Levitin. Redundancy optimization for multi-state system with
fixed resource-requirements and unreliable sources. IEEE
Trans.Reliability, vol. 50, no. 1, pp. 5259, 2001.
[4] J. E. Ramirez, D. W. Coit. A Heuristic for solving the
redundancy allocation problem for multistate series-parallel
systems. Reliability Engineering and System Safety, vol. 83, no.
3, pp. 341349, 2004.

Case I

Case II
Reserve

50
--16,982 /kWao y
Operation
9,1 l/h 23,1 l/h
0,0346 /kWh
Maintenance
30 % C.I.
0,007 c/kWh
0,012 c/kWh
--Life time
4 years
20 years
--CASE III.Same as case II, the wind resource varied its scale parameter, from 5 m/s to 7m/s. The shape parameter is considered 2 in all
cases.

ACKNOWLEDGMENT

Case II
Solar
10 10 50
80
4,0
3,5
---

5,56
--22,302 /kWao
y 0,1284 /kWh

[10] A. da Silva, R. Fernandez, C. Singh. Generating Capacity


Reliability Evaluation Based on Monte Carlo Simulation and
Cross-Entropy Methods Power Systems, IEEE Transactions on,
vol. 25, pp. 129-137, 2010.
[11] F. C. Sayas, R. N. Allan. Generation availability assessment
of wind farms.IEE Proc. Gener. Transm. Distrib., vol. 143, pp.
507518, 1996.
[12] A. P. Leite, C. L. T. Borges, D. M. Falcao. Probabilistic wind
farm generation model for reliability studies applied to Brazilian
sites. IEEE Trans. Power Systems, vol. 21, pp. 14931501, 2006.
[13] P. Wang, L. Goel, Y. Ding, L. P. Chang, Andrew M.
Reliability-based long term hydro/thermal reserve allocation of
power systems with high wind power penetration. Power &
Energy Society General Meeting, 2009.PES 09. IEEE, pp. 1-7,
July 2009.
[14] Y. Ding, P. Wang, L. Goel., P. Chiang, et. al. Long term
reserve expansion of power systems with high wind power
penetration using universal generating function methods. IEEE
Trans. Power Systems, vol. 26, no. 2, pp. 766-774, 2010.
[15] Y. Li, E. Zio. A Multi-State Model for the Reliability
Assessment of a Distributed Generation System via Universal
Generating Function.eprint arXiv:1206.6808 , 2012.

[5] H. A. Taboada, J. F. Espiritu, D. W. Coit. MOMS-GA: A MultiObjective Multi-State Genetic Algorithm for System Reliability
optimization Design Problems. IEEE Trans. Reliability, vol. 57,
no. 1, pp. 182-191, 2008.

[16] A. Lisnianski, G. Levitin. Multi-state system reliability


assessment, optimization, applications, World Sci, 2003.

[6] G. Levitin, A. Lisnianski, H.B. Haim, D. Elmakis. Genetic


algorithm and universal generating function technique for
solving problems of power system reliability optimization.
Electric Utility Deregulation and Restructuring and Power
Technologies, 2000.Proceedings. DRPT2000. International
Conference on, pp. 582-586, 2000.

[18] Z. Yu, A. Tuzuner. Wind Speed Modeling and Energy


Production Simulation with Weibull Sampling. Power and
Energy Society General Meeting Conversion and Delivery of
Electrical Energy in the 21st Century, 2008 IEEE, pp 1-6, 2008.

[7] X. Wang, H. Dai, R. J. Thomas. Reliability modeling of large


wind farms and electric utility interface systems. IEEE Trans.
Power App. Syst., vol. PAS-103, no. 3, pp. 569575, 1984.
[8] R. Billinton, H. Chen, R. Ghajar. A sequential simulation
technique for adequacy evaluation of generating systems
including wind energy. IEEE Trans. Energy Conversion, vol. 11,
pp. 728734, 1996.
[9] P. Wang, R. Billinton. Reliability benefits analysis of adding
WTG in a distribution system. IEEE Trans. Energy Conversion,
vol.16, no.2, pp.134-139, 2001.

[17] G. Levitin, The Universal Generating Function in Reliability


Analysis and Optimization, Springer, 2005.

[19] K. Deb, A. Pratap, A. Agarwal, T. Meyarivan. A fast and


elitist multiobjective genetic algorithm: NSGA-II. IEEE Trans.
Evolutionary Computation, vol. 6, no. 2, pp.182-197, 2002.

You might also like