Impact of Interconnection Photo Voltaic Wind System ANFIS

ARTICLE IN PRESS
Renewable Energy 31 (2006) 2475–2491

www.elsevier.com/locate/renene
Impact of interconnection photovoltaic/wind system

with utility on their reliability using a fuzzy scheme
Hassan Hussein El-Tamaly, Adel A. Elbaset Mohammed
Faculty of Engineering, Elminia University, Elminia, Egypt
Received 30 January 2005; accepted 29 November 2005
Available online 2 February 2006
Abstract
Reliability analysis has been considered as an important step in any system design process.
A reliable electrical power system means a system which has sufficient power to feed the load demand
during a certain period or, in other words, has small Loss of Load Probability (LOLP). LOLP is
defined as an expected fraction of load not met by its power needs from electrical power system
during its lifetime. Photovoltaic (PV)/Wind Energy System (WES) Hybrid Electric Power System
(PV/WES HEPS) differs considerably from the Utility Grid (UG) in its performance and operating
characteristics. With the interconnection of PV/WES as a HEPS into the UG, the fluctuating nature
of the energy produced by these systems has a different effect on the overall system reliability than
that of the fluctuating nature of energy produced by UG. Therefore, this paper presents a complete
study, from reliability point of view, to determine the impact of interconnecting PV/WES HEPS into
UG. Four different configurations of PV/WES/UG have been investigated and a comparative study
between these four different configurations has been carried out. The overall system is divided into
three subsystems, containing the UG, PV and WES. The generation capacity outage table has been
built for each configuration of these subsystems. These capacity outage tables of UG, PV/UG, WES/
UG and PV/WES/UG are calculated and updated to incorporate their fluctuating energy production.
This paper also presents a fuzzy logic technique to calculate and assess the reliability index for each
HEPS configuration under study.
r 2006 Elsevier Ltd. All rights reserved.
Keywords: Design of PV/WES hybrid; Reliability; Probabilistic model; Fuzzy logic
Corresponding author. Tel.: +20 86 2362083; fax: +20 86 2346674.

E-mail addresses: dr_h_tamaly@yahoo.com (H.H. El-Tamaly), Ayman087@yahoo.com
(A.A.E. Mohammed).
0960-1481/$ - see front matter r 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.renene.2005.11.012
ARTICLE IN PRESS
2476 H.H. El-Tamaly, A.A.E. Mohammed / Renewable Energy 31 (2006) 2475–2491
Nomenclature
LOLP loss of load probability

PV photovoltaic
WES wind energy system
HEPS hybrid electric power system
UG utility grid
g forced outage rate
ANFIS adaptive-network-based fuzzy inference system
FL fuzzy logic
FIS fuzzy inference system
EFORpv effective forced outage rate for photovoltaic system
EFORw effective forced outage rate for wind system
EFORpvw effective forced outage rate for photovoltaic/wind system
1. Introduction
As energy demands around the world increase, the need for a renewable energy source
that will not harm the environment has increased. Some projections indicate that the
global energy demand will almost triple by 2050 [1,2]. Renewable energy sources currently
supply somewhere between 15% and 20% of the total world energy demand. PV and WES
are the most promising as future energy technologies. A 30% contribution to world energy
supply from renewable energy sources by year 2020 as proposed in Ref. [2] would reduce
the energy related CO2 emission by 25%. The index used to measure generation reliability
is probabilistic estimates of the ability of a particular generation configuration to supply
the load demand. Generator units may be unavailable due to failures, this is called ‘‘forced
unavailability’’, or due to preventive maintenance, this is called ‘‘planned unavailability’’.
Both types of unavailability may be treated stochastically, but the planned unavailability is
often treated deterministically. The uncertainty in the generation capacity is due to failures
(outages) of generator units [3]. The reliability analysis for renewable system uses a
capacity outage probability table, which is an array of capacity levels and the associated
probabilities of existence. This is obtained by combining availability and unavailability of
the generating units using basic probability concepts. From the individual probability
table, we prepared a cumulative probability table. The probability tables for PV, WES and
PV/WES HEPS are modified hour by hour to include the effect of fluctuation in generation
capacity. Finally, these tables are combined to evaluate the loss of load probability using
Fuzzy algorithm. Also, the impact of the interconnection of PV/WES HEPS on their
reliability has been analyzed.
2. Methodology
2.1. Probabilistic modeling and reliability index
Probabilistic reliability index serves as an accurate and consistent basis for assessing
reliability of power systems, where components outage and load demand are of stochastic
ARTICLE IN PRESS
H.H. El-Tamaly, A.A.E. Mohammed / Renewable Energy 31 (2006) 2475–2491 2477
nature [4]. The reliability analysis for renewable system uses a capacity outage probability
table, which is an array of capacity levels and the associate probabilities of existence. This
is obtained by combining the generating units availability and unavailability using basic
probability concepts. From the individual probability table, we prepare cumulative
probability table. The basic elements used to evaluate generation adequacy are shown in
Fig. 1. The system is deemed to operate successfully as long as there is sufficient generation
capacity to supply the load [5].
The cumulative probability of a particular capacity outage state of X MW after adding a
two-state unit of capacity C MW with forced outage rate, g is given [6,7] as
PðX Þ ¼ ð1 gÞ P0 ðX Þ þ g P0 ðX CÞ, (1)
0
where P (X) and P(X) denote the cumulative probabilities of the capacity outage state of
X MW before and after the unit is added. P(X) is also the probability of capacity outage
being XX.
The above expression is initialized by setting
P0 ðX Þ ¼ 1:0 for X p0 and P0 ðX Þ ¼ 0:0 otherwise.
Forced outage rate, g, is given by the following equation [5]:

Forced outage in hours
g¼ . (2)
Forced outage in hours þ In services hours
Eq. (1) can be modified as follows to include multi-state unit representations [6,7]:
X
n
PðX Þ ¼ pi P0 ðX C i Þ, (3)
i¼1
where n: the total number of units states; Ci: the capacity outage of state i for the unit being
added; pi: the probability of existence of the unit state i and is defined as follows [6,7]:
Xn
n
pi ¼ gr ð1 gÞnr ; (4)
r¼k
r
Generation
Load Model
Model
Risk of generation
< Load
Reliability Index
LOLP
Fig. 1. Elements of generation reliability evaluation [5].

ARTICLE IN PRESS
where k: minimum number of units required to the system succeed. The term ðnrÞ is defined
as follows:

n n!
¼ . (5)
r ðn rÞ! ðrÞ!
The overall probability that the load demand will not be met is called LOLP is computed
as
X
n
LOLP ¼ pi PðLj 4C i Þ; (6)
r¼k
where Lj: forecast peak load at hour j; P(Lj4Ci): probability of loss of load at hour j;
LOLP: loss of load probability at hour j for state i.
2.2. Fuzzy logic application on reliability study of renewable energy
There are many methods for sizing PV system, WES and PV/WES/UG HEPS. The first
group of these methods is intuitive methods. They are used as a first approach, but they are
very inaccurate. The second group of these methods is denominated numerical method and
uses system simulations. They are more accurate than the intuitive methods. Lastly, there
are methods which use equations to describe the system under study as a function of
reliability. These are called analytical methods. FL techniques have superseded
conventional technologies in many scientific applications and engineering systems. Fuzzy
techniques are applicable in various areas such as control, pattern recognition, quantitative
analysis, planning, and prediction. The applications of fuzzy technique are increasing so
rapidly that it is not possible to offer a limited list of them. An efficient reliability scheme
based on FL technique, is suitable for sizing the PV system, WES and PV/WES/UG
HEPS. The reliability proposed makes use of an ANFIS. In order to use the FL technique,
the input parameter of each configuration should be determined precisely. The input data
for each configuration under study can be obtained from power for one module for PV
system, power for one wind turbine generator and load power at a given site. FIS employs
the theory of fuzzy sets and fuzzy if-then rules to derive an output of LOLP. Typically, an
FIS scheme performs its action in several steps including:
Fuzzification (comparing the input values with membership functions to obtain

membership values of each linguistic term).
Fuzzy reasoning (firing the rules and generating their fuzzy or crisp consequents).
Defuzzification (aggregating rule consequents to produce a crisp output) [8,9].
The investigations described here have been carried out for a Sugeno-type FIS structure as
shown in Fig. 2, where the output of each rule (y1 ; y2 ; . . . ; yn ) is a linear combination of
input variables (x1 ; x2 ; x3 ; x4 ) plus a constant term, the inner nodes (R1,R2,y,RN)
represent the rules and the final output LOLP is the weighted average of each rule’s output:
w1 y1 þ w2 y2 þ þ wn yn
LOLP ¼ , (7)
w1 þ w2 þ þ wn
yk ¼ ak x1 þ bk x2 þ ck x3 þ d k x4 þ f k , (8)
ARTICLE IN PRESS
[µi(Xi)]
Inputs Output
R1 LOLP
X1 [yi]
MF R2
X2 R3
MF ANFIS
R4
X3
MF R5 Output
X4 R143
MF
RN
Set of if-then rules
Fig. 2. FIS under study a general structure.
wk ¼ m1k ðx1 Þm2k ðx2 Þm3k ðx3 Þm4k ðx4 Þ, (9)

where mik ðxi Þ 2 fmiLow ; miMedium ; miHigh g membership functions for linguistic terms low,
medium, high associated with the ith input signal, wi—weighting factor for the ith rule
consequent [9]. The membership function maps each input element of x1, x2, x3 and x4 to a
membership value between 0 and 1.
The membership functions mik ðxi Þ and yik represent the fuzzy sets which describe the antece-
dent’s consequents. There are many membership function types. The selection of the member-
ship functions and their boundaries should express the performance of the FL algorithm.
3. Applications of probabilistic modeling and results
A new computer program has been designed to calculate the LOLP index for
ElZafaranna site, located on the western coast of the Suez Gulf, latitude 29.071N and
longitude 31.361E, Egypt. The flowchart of this program is shown in Fig. 3. The inputs
data of this program are:
(1) hourly radiation, kW/m2;

(2) hourly wind speed, m/s;
(3) characteristic of PV module;
(4) characteristic of wind turbine;
(5) hourly load demand, W.
It is assumed here that the load demand varies monthly. Therefore, there are 12 daily load
curves through the year i.e. 12 24 ¼ 288 h through the year. Fig. 4 shows the load
demand for January, April, July and October [1]. The outputs of this program are:
(1) hourly power from PV system, W;

(2) hourly power from WES, W;
(3) LOLP values for each hour for UG, PV/UG HEPS, WES/UG HEPS and PV/WES/
UG HEPS at different penetration values under condition of using ASE-300 DGF/17
solar cell module type with rating 300 W and T600-48 Wind turbine generator type
with rating 600 kW.
ARTICLE IN PRESS
Start
Read Radiation, Wind speed, PV module, wind type

parameters and load demand.
Modification of radiation on surfaces tilted by monthly best tilt angle and

wind speed at hub height. Set Npv, Nw
For pent ratio = 0 : 1 : 0.1
Calculate maximum power for one module based on

Maximum power point
Energy balance for PV to determine number of PV

module
Calculate maximum power for one turbine based on

Maximum power point
Calculate generated power from WES.
Calculate generated power from PV sysetem
Calculate Capacity Outage table.

Eq. (2) and Eq. (3)
Calculate LOLP
Eq. (5)
Take a decision to select optimum number of PV and

WTG for minimum LOLP
End
Fig. 3. Flowchart of the proposed computer program.
3.1. Capacity outage table for the configuration of UG only
One of the most commonly used methods of determining the required generation for
LOLP calculations is the generation capacity outage table [7]. The generation capacity
outage table is based on the independent behavior of different units where each has its own
unavailability. The power system in this study consisted of 20 generating units each with
ARTICLE IN PRESS
100
90
July
80
October
Power, MW
70
60 April
50 January
40
30
20
0 5 10 15 20 25
Time, Hour
Fig. 4. Hourly load demand for the months of January, April, July and October [1].
Table 1
Twenty units system capacity outage table
State i MW output MW input Ci Probability pi Commutative

probability P(Lj 4Ci)
0 0 100 3.584E01 1.000E+00

1 5 95 3.773E01 6.41E01
2 10 90 1.886E01 2.64E01
3 15 85 5.958E02 7.548E02
4 20 80 1.332E02 1.590E02
5 25 75 2.244E03 2.573E03
6 30 70 2.953E04 3.292E04
7 35 65 3.108E05 3.394E05
8 40 60 2.659E06 2.856E06
9 45 55 1.865E07 1.979E07
10 50 50 1.080E08 1.134E08
11 55 45 5.168E10 5.379E10
12 60 40 2.040E11 2.108E11
13 65 35 6.608E13 6.794E13
14 70 30 1.739E14 1.859E14
15 75 25 3.661E16 1.202E15
16 80 20 6.021E18 8.367E16
17 85 15 7.457E20 8.306E16
18 90 10 6.541E22 8.306E16
19 95 5 3.624E24 8.306E16
20 100 0 9:536E 27 8.306E16
¼1
5 MW of capacity to feed the load demand. The forced outage rate of each unit is 0.05,
i.e. it has an availability of 0.95 [10]. The nominal installed capacity of the system is
100 MW based on maximum load. The capacity outage distribution of the system is shown
in Table 1 which indicates the amount of capacity out of service (column 2) and available
ARTICLE IN PRESS
0.8
LOLP for UG only
0.7
LOLP for WES / UG only
0.6
LOLP, Hour
0.5
0.4 LOLP for PV/ UG only
0.3
0.2
0.1
0
0 1 2 3 4 5 6 7 8 9 10 11 12
Time, Months
Fig. 5. Total hourly LOLP values for all months during the year.
(column 3) for each state, the probability of each state (column 4) and the cumulative
probability (column 5) that is the probability that more than that capacity is out of
service. An example, the probability that 25 MW or more is out of service is 2.573E3.
By applying Eqs. (2) and (5) combined with load demand shown in Fig. 4. From
methodology of probabilistic modeling, we found the LOLP each hour per month. From
summation of LOLP for each hour during each month, we can get total LOLP for each
month for UG alone as shown in Fig. 5 which indicates the total LOLP for UG alone
during year is equal to 2.181018301377 h with percentage reliability equal to (2882.1810)/
288 100 ¼ 99.242%.
3.2. Capacity outage table for the configuration of PV/UG HEPS
The fluctuating nature of the energy produced by PV system has a different effect on
the overall system reliability than that of the energy produced by UG. Therefore, the
EFORpv has been calculated every month. The EFORpv equal to the maximum PV energy
output for any period to the total PV energy during this period is shown in the following
equation [11]:
EFORpv ¼ 1 ðMaximum of PV energy=Total of PV energyÞ. (10)
The sliding window approach has been applied here which allows for variability in PV
power output throughout time. From the computer program the EFORpv for all months of
the year is shown in Fig. 6. On the other hand, the hourly LOLP for each hour of year has
been found by probabilistic modeling. Figure 5 displays the total LOLP for each month for
PV/UG HEPS. From this figure it can be seen that the total LOLP during each month was
improved when load demand was fed from PV/UG HEPS. For example, LOLP during
April was 0.0968992 h when load was fed from UG, on the other hand during April LOLP
was 0.073406492 h when load was fed from PV/UG HEPS. The yearly LOLP when the
load feed form PV/UG HEPS is equal to 1.64259304 h with percentage reliability equal to
(2881.64259)/288 100 ¼ 99.4296%.
ARTICLE IN PRESS
0.9
EFOR for PV/ UG
0.8
0.7
0.6
EFOR
0.5
0.4
EFOR for WES / UG
0.9 EFOR for PV / WES / UG
When penetration level=0.28
0.3
0.2
0 1 2 3 4 5 6 7 8 9 10 11 12
Time, Months
Fig. 6. Monthly effective force outage rate values during the year.
3.3. Capacity outage table for the configuration of WES/UG HEPS
In order to estimate the LOLP of WES/UG HEPS, the EFORw was calculated as shown
in the following equation [11]:
EFORw ¼ 1 ðMaximum energy of WES=Total energy of WESÞ: (11)
From designed computer program, monthly value of EFORw of the year is shown in Fig.
6. Figure 5 displays the total hourly LOLP value for each month of the year for UG alone
and WES/UG HEPS. From this figure it can be seen that the total LOLP for each month
improved when load demand was fed from WES/UG HEPS. For example, the value of
LOLP during April was 0.0968992 h when load was fed from UG, on the other hand the
value of LOLP during April was 0.0283574 h when load was fed from WES/UG HEPS.
The total yearly LOLP value for WES/UG HEPS was equal to 1.504018 h with percentage
reliability equal to (2881.5040)/288 100 ¼ 99.477%.
3.4. Capacity outage table for the configuration of PV/WES/UG HEPS
The techno-economical design of this configuration has been carried out in Ref. [1]. This
design was based on the following equation:
Pgtotal ðtÞ ¼ a Ppv ðtÞ þ ð1 aÞ PWES ðtÞ, (12)
where Pgtotal: the total generated power from PV/WES HEPS; Ppv(t): the generated power
from PV system, W; PWES(t): the generated power from WES, W; a: the penetration level
of PV system, 0, 0.1, 0.2, 0.3,y,1.
If Pgtotal4load demand, then the surplus energy is fed to the UG. If Pgtotaloload
demand, then the deficit energy will be taken from the UG. From this study it was found
that the most economic penetration level is equal to 0.28. The EFORpvw can be calculated
as follows:
EFORpvw ¼ 1 ðMaximum of Pgtotal energy=Total of Pgtotal energyÞ. (13)
ARTICLE IN PRESS
2.5
2
LOLP, Hour
1.5
0.5
0
UG alone
Penet. ratio= 0.28

Penet. ratio= 0.1
Penet. ratio= 0.2
Penet. ratio= 0.3
Penet. ratio= 0.4
Penet. ratio= 0.5
Penet. ratio= 0.6
Penet. ratio= 0.7
Penet. ratio= 0.8
Penet. ratio= 0.9

PV/UG alone Pent. ratio= 1.0
WES/UG alone, Pent=0.0
Fig. 7. Total yearly LOLP value for each penetration ratio.
Using a proposed computer program the monthly EFORpvw values through the year has
been estimated at penetration level equal to 0.28. Figure 6 shows the relation between
EFORpvw and year months at penetration ratio equal to 0.28. The LOLP values for PV/
WES/UG HEPS at different penetration levels have been estimated as shown in Fig. 7,
which indicates the minimum LOLP value for this configuration occurs at penetration level
equal to 0.28. i.e. the PV will feed 28% of the load demand; and WES will feed the
remaining 72% of the load demand. The yearly LOLP value for PV/WES/UG HEPS was
equal to 1.223 h with percentage reliability equal to (2881.223)/288 100 ¼ 99.575%.
4. Application of FL and results
4.1. Configuration of PV/UG
In order to use the ANFIS technique for this configuration, the input parameters limit
should be determined precisely. The input parameters are hourly load demand and hourly
power generated from PV system. The output is LOLP. The ANFIS consists of five
layers of nodes performing different operations on incoming signals. The nodes in
particular layers are responsible for determination of membership grades for each
linguistic term, executing the rules and generating the weighted output. The membership
function mik ðxi Þ with its grade of membership in fuzzy sets associates for this configura-
tion are shown in Figs. 8 and 9. The parameters of the ANFIS have been adjusted
via training (similar as for neural network schemes). The training set of input–output
patterns and testing set have generated a new computer program as shown in Fig. 3. Each
row of training data is a desired input/output pair of the target system to be modeled.
Each row starts with an input vector and is followed by an output value. The training data
ARTICLE IN PRESS
in1mf3 in1mf5 in1mf7 in1mf9

1
in1mf1 in1mf2 in1mf4 in1mf6 in1mf8 in1mf10
Degree of membership
0.8
0.6
0.4
0.2
30 40 50 60 70 80 90
input1, Load
Fig. 8. Membership function for input 1 (hourly load).
in2mf1 in2mf2 in2mf3 in2mf4 in2mf5 in2mf6 in2mf7 in2mf8 in2m9

1
Degree of membership
0.8
0.6
0.4
0.2
0 50 100 150 200

input2 (PV Power)
Fig. 9. Membership function for input 2 (hourly PV power).
set consisted of 180 samples and the testing set 108 samples. A hybrid training algorithm
being a combination of the least squares method and back-propagation gradient descent
method was used here to prepare the FIS for the LOLP calculation. ANFIS from Matlab
converts the trained data points to rules and fuzzy sets [12]. The membership functions are
Gaussian membership functions and rules are design tools that give opportunity to
calculate LOLP. The Max–Min inference method and average weight defuzzification
strategy are used.
There are 90 rules which are sufficient to assign a LOLP using ANFIS. Some of these
rules are as follows:
1. If (input1 is in1mf1) and (input2 is in2mf1) then (output is out1mf1) (1)
ARTICLE IN PRESS

k
89. if (input1 is in1mf10) and (input2 is in2mf8) then (output is out1mf89) (1)
The output of LOLP value predicted using ANFIS is shown in Fig. 10. The difference
between exact solution by probabilistic model and FIS predicted is very small as shown in
Fig. 11. Thus we only see one curve as shown in Fig. 10. Once the yearly LOLP value has
been obtained, it is very easy to design the configuration of PV/UG.
0.06
0.05
0.04
LOLP, Hour
0.03
0.02
0.01
-0.01
180 200 220 240 260 280 300
Samples
Fig. 10. Output of LOLP value using ANFIS for the configuration PV/UG HEPS.
x10-4
10
5
Error
-5
180 200 220 240 260 280 300
Samples
Fig. 11. The difference between ANFIS and probabilistic model for the configuration PV/UG HEPS.
ARTICLE IN PRESS
The yearly value of LOLP obtained by ANFIS was 1.6426 h. To get this value for the
configuration of PV/UG HEPS we need 709770 modules from ASE-300 DGF/17 solar cell
module type to feed the load demand with maximum demand 100 MW.
The FIS-based reasoning unit itself has the following design parameters:
Type—Sugeno.
Gaussian membership functions.
Nine and 10 linguistic terms for each input membership function.
90 linear terms for output membership functions.
90 rules (resulting from number of inputs and membership function terms).
Fuzzy operators: product (and), maximum (or), product (implication), maximum
(aggregation), average weight (defuzzification).
4.2. Configuration of WES/UG
The input parameters in this configuration are hourly load demand and hourly power
generated from WES. The output is LOLP. The membership function mik ðxi Þ with its grade
of membership in fuzzy sets associates for this configuration are similar to Figs. 8 and 9 but
with different number of membership function. The parameters of the ANFIS for this
configuration were also adjusted via training (similar to as for neural network schemes).
Each row of training data is a desired input/output pair of the target system to be modeled.
Each row starts with an input vector and is followed by an output value. The training data
set consisted of 180 samples and the testing set 108 samples. A hybrid training algorithm
being a combination of the least squares method and back-propagation gradient descent
method was used here to prepare the ANFIS for the LOLP calculation. ANFIS from
Matlab converts the trained data points to rules and fuzzy sets [12]. The membership
functions are Gaussian membership functions and rules are design tools that give
opportunity to calculate LOLP. The Max–Min inference method and average weight
defuzzification strategy are used. There are 120 rules which are sufficient to assign a LOLP
using ANFIS. Some of these rules are as follows:
8. If (input1 is in1mf1) and (input2 is in2mf8) men (output is out1mf8) (1)
12. If (input1 is inlmf2) and (input2 is in2mf12) then (output is out1mf12) (1)
13. If (input1 is inlmf2) and (input2 is in2mf1) then (output is out1mf13) (1)
k
ARTICLE IN PRESS
0.06
0.05
0.04
LOLP, Hour
0.03
0.02
0.01
-0.01
180 200 220 240 260 280 300
Samples
Fig. 12. Output of LOLP value using ANFIS for the configuration WES/UG HEPS.
x10-3
1.5
1
0.5
0
-0.5
Error
-1
-1.5
-2
-2.5
-3
180 200 220 240 260 280 300
Samples
Fig. 13. The difference between ANFIS and probabilistic model for the configuration WES/UG HEPS.
The output of LOLP value predicted using ANFIS is shown in Fig. 12. The difference
between exact solution by probabilistic model and FIS predicted is very small as shown in
Fig. 13. Once the yearly LOLP value was obtained, it is very easy to design any
configuration from WES/UG. The yearly value of LOLP obtained by ANFIS was
1.65041 h. To get this value for the configuration of the WES/UG HEPS we need 300 wind
turbines from T600-48 Wind turbine generator type with rating 600 kW to feed the load
demand with maximum demand 100 MW.
Type—Sugeno.
ARTICLE IN PRESS
Ten and 12 linguistic terms for each input membership function.

4.3. Configuration of PV/WES/UG
Here the penetration ratio for PV system is equal to 0.28 and the WES equal to 0.72. The
LOLP under this condition was the lowest LOLP. The input parameters in this
configuration are hourly load demand and hourly power generated from PV system,
hourly power generated from WES. The output is a LOLP. The membership function
mik ðxi Þ with its grade of membership in fuzzy sets associates for this configuration are
similar to Figs. 8 and 9 but with different number of membership functions. The training
data set consisted of 180 samples and the testing set 108 samples. ANFIS from Matlab
converts the trained data points to rules and fuzzy sets [12]. The membership functions are
Gaussian membership functions and rules are design tools that give opportunity to
calculate LOLP. The Max–Min inference method and average weight defuzzification
strategy are used.
There are 210 rules which are sufficient to assign a LOLP using ANFIS. Some of these
rules are as follows:
1. If (input1 is in1mf1) and (input2 is in2mf1) and (input3 is in3mf1) then (output is
out1mf1) (1)
out1mf2) (1)
out1mf3) (1)
out1mf4) (1)
out1mf5) (1)
out1mf6) (1)
out1mf7) (1)
out1mf8) (1)
out1mf9) (1)
out1mf10) (1)
k
out1mf209) (1)
out1mf210) (1)
ARTICLE IN PRESS
0.16
0.14
0.12
0.1
LOLP, Hour
0.08
0.06
0.04
0
-0.02
180 200 220 240 260 280 300

Samples
Fig. 14. Output of LOLP value using FL.
x10-5
6
0
Error
-2
-4
-6
-8
180 200 220 240 260 280 300
Samples
Fig. 15. The difference between FL and probabilistic model.
The FL algorithm predicted LOLP for any value of penetration ratio. Overall, the
output for 108 samples from FL is shown in Fig. 14. The difference between fuzzy
technique and probabilistic model as shown in Fig. 15. From Figs. 14 and 15 it can be seen
that, the results indicate good performance of the FL as well as the probabilistic model.
The yearly value of LOLP obtained by ANFIS in this configuration was 1.233 h. To get
this value for the configuration of the PV/WES/UG HEPS we need 183085 modules from
ASE-300 DGF/17 solar cell type and 218 wind turbines from T600-48 Wind turbine
generator type with rating 600 kW to feed the load demand with maximum demand
100 MW.
Type—Sugeno.
ARTICLE IN PRESS
Six, five and seven linguistic terms for each input membership function.
5. Conclusions
This paper presents a complete reliability study to determine the impact of

interconnecting PV/WES HEPS into UG. From results detailed above, the following
salient conclusions can be drawn:
1. Introducing PV with penetration ratio 0.28 into WES/UG combination decreases

LOLP from 1.63 to 1.223 h.
2. A new proposed method to evaluate the reliability index (LOLP) for each configuration
under study using FL algorithm has been proposed. The results obtained show that the
proposed method gives good estimations.
3. Simulations by FL show that the LOLP curves generated by the proposed method fit
the curves generated by the probabilistic model.
4. The proposed methodology for generating LOLP curves based on FL can be used for
sizing PV/UG power system, WES/UG power system or PV/WES/UG HEPS.
References
[1] El-Tamaly HH, Mohamed AAE. Design and control strategy of utility interfaced PV/WTG hybrid system.
The ninth international middle east power system conference, MEPCON’2003, December 16–18, vol. 2.
Egypt: Faculty of Engineering, Shebin El-Kom, Minoufiya University; 2003. p. 699–74.
[2] International energy agency report. Key issues in developing renewables. 1997.
[3] Ehnberg J. Generation reliability for isolated power systems with solar, wind and hydro generation. MSc
thesis, Department of Electric Power Engineering, Chalmers University of Technology, Gäoteborg, Sweden;
2003.
[4] Yu DC, et al. Bayesian network model for reliability assessment of power system. Proc IEEE Trans Power
Syst 1999;14(2):426–32.
[5] Prada JF. The value of reliability in power systems pricing operating reserves, MIT EL99-005 WP. MIT:
Energy Laboratory, Massachusetts Institute of Technology; July 1999.
[6] Kim J-O, Singh C. Including uncertainty in LOLE calculation using fuzzy set theory. IEEE Trans Power Syst
2002;17(1):19–25.
[7] Billinton R, Allan R. Reliability evaluation of power systems. New York: Plenum Press; 1996.
[8] Fuzzy logic control systems: fuzzy logic controller (into two parts). IEEE Trans Syst Man Cyber B 1990;
20(2).
[9] Drainkov D. An introduction to fuzzy control. Berlin: Springer; 1993.
[10] Preston EG. Reliability of electric generation with transmission constraints. PhD thesis, Faculty of the
Graduate School, University of Texas, Austin; May 1997.
[11] Milligan MR. A sliding window technique for calculating system LOLP contributions of wind power plants.
Proceedings on the wind power 2001 conference, NREL/CP-500-30363. Washington, DC: American Wind
Energy Association; 2001.
[12] Matlab–fuzzy user’s guide. The Mathworks, Inc., 2000.

Impact of Interconnection Photo Voltaic Wind System ANFIS

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Impact of Interconnection Photo Voltaic Wind System ANFIS

Uploaded by

Copyright:

Available Formats

ARTICLE IN PRESS

Renewable Energy 31 (2006) 2475–2491

Impact of interconnection photovoltaic/wind system

Keywords: Design of PV/WES hybrid; Reliability; Probabilistic model; Fuzzy logic

Corresponding author. Tel.: +20 86 2362083; fax: +20 86 2346674.

LOLP loss of load probability

2.1. Probabilistic modeling and reliability index

Forced outage rate, g, is given by the following equation [5]:

Fig. 1. Elements of generation reliability evaluation [5].

2.2. Fuzzy logic application on reliability study of renewable energy

Fuzziﬁcation (comparing the input values with membership functions to obtain

Set of if-then rules

Fig. 2. FIS under study a general structure.

wk ¼ m1k ðx1 Þm2k ðx2 Þm3k ðx3 Þm4k ðx4 Þ, (9)

3. Applications of probabilistic modeling and results

(1) hourly radiation, kW/m2;

(1) hourly power from PV system, W;

Read Radiation, Wind speed, PV module, wind type

Modification of radiation on surfaces tilted by monthly best tilt angle and

For pent ratio = 0 : 1 : 0.1

Calculate maximum power for one module based on

Energy balance for PV to determine number of PV

Calculate maximum power for one turbine based on

Calculate generated power from WES.

Calculate generated power from PV sysetem

Calculate Capacity Outage table.

Take a decision to select optimum number of PV and

Fig. 3. Flowchart of the proposed computer program.

3.1. Capacity outage table for the configuration of UG only

State i MW output MW input Ci Probability pi Commutative

0 0 100 3.584E01 1.000E+00

0.4 LOLP for PV/ UG only

3.2. Capacity outage table for the configuration of PV/UG HEPS

3.3. Capacity outage table for the configuration of WES/UG HEPS

3.4. Capacity outage table for the configuration of PV/WES/UG HEPS

Penet. ratio= 0.28

Penet. ratio= 0.2

Penet. ratio= 0.3

Penet. ratio= 0.4

Penet. ratio= 0.5

Penet. ratio= 0.6

Penet. ratio= 0.7

Penet. ratio= 0.8

Penet. ratio= 0.9

Fig. 7. Total yearly LOLP value for each penetration ratio.

4. Application of FL and results

4.1. Configuration of PV/UG

in1mf3 in1mf5 in1mf7 in1mf9

Fig. 8. Membership function for input 1 (hourly load).

in2mf1 in2mf2 in2mf3 in2mf4 in2mf5 in2mf6 in2mf7 in2mf8 in2m9

0 50 100 150 200

Fig. 9. Membership function for input 2 (hourly PV power).

6. If (input1 is in1mf1) and (input2 is in2mf6) then (output is out1mf6) (1)

4.2. Configuration of WES/UG

Ten and 12 linguistic terms for each input membership function.

4.3. Configuration of PV/WES/UG

180 200 220 240 260 280 300

Fig. 14. Output of LOLP value using FL.

Fig. 15. The difference between FL and probabilistic model.

This paper presents a complete reliability study to determine the impact of

1. Introducing PV with penetration ratio 0.28 into WES/UG combination decreases

You might also like