Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

Generation availability assessment of wind farms

F. Castro Sayas
R.N.Allan

Indexing terms: Wind farms, Reliability, Probabilistic model, Analytical technique, Practical application

-
wind power plants witlh higher installed capacities are
Abstract: Wind generation is one of the most incorporated into existing power systems, it is becoming
successful sources of renewable energy for the important to study the reliability of these generation
production of electrical energy. As the nurnber of systems and assess the effects that they will have on the
installations connected into existing networks entire system and its reliability.
grows rapidly worldwide there is a need to study Earlier studies of the reliability impact of unconven-
the reliability of these energy sources and assess tional energy sources have dealt with the problem only
their effects on the system. However, the technical at the generation level (hierarchical level 1, HL1 [3]).
characteristics of wind generation make existing The first attempts to include unconventional energy
conventional generation models not (directly sources in reliability analysis of the generating system
applicable. A new probabilistic model of a wind used the loss-of-load approach. A method which
farm to be used in reliability studies is prcsented. included frequency and duration concepts is presented
This model considers the stochastic nal ure of in [4] for wind energy conversion systems. The method
wind, the failure and repair processes of the wind presented in [4] is enhanced to consider different types
turbines under different wind regimes, and the of units within the same wind farm 15, 61 and average
output cliaracteri,stics of the wind turbine in wind velocity spatial correlation [6]. Another approach
detail. Spatial winid speed correlation and waking using the cumulant method including the failure and
effects can also be considered. The model can be repair characteristics of the wind turbines is illustrated
used with one or several types of wind turbines. in [7].Correlation between load and the power output
Analytical techniques are used. A practical of unconventional units is taken into account in [SI.
application of the models is also included. The latest publications use a loss-of-energy approach
- (LOEE). Methods based on the load modification tech-
nique were developed in [2, 91 with economic assess-
ment included in [:’]. None of these previous
1 lntroduiction approaches consider exlplicit rnodelling of all the factors
affecting the wind farm, and instead treat the farm as a
The technologies for the generation of electrilcal energy single entity. The effect of the individual components
from renewable energy sources have evolved in recent and factors are therefore masked.
years from small experimental plants to a real option This paper overcomes this weakness by introducing a
for the utilities to prloduce significant amounts of elec- wind farm model for reliability studies that can take
tric power. The high cost of energy derived from fossil into account the stochastic nature of wind, the failure
and nuclear fuels, environmental issues and a protec- and repair processes of the wind turbines, their output
tive legal framework encouraging the use of renewable curves, wind speed spatial correlation and wake effects
sources of energy in many countries have resulted in using reliability analysis based on analytical techniques.
the number of installations commissioned every year This model produces re:sults that can not only be easily
worldwide increasing very rapidly. Wind has proved to integrated with available models of the distribution net-
be one of the most successful of all available sources of work but can also be used for standard generation level
renewable energy (wind, solar, wave, tidal, ...) offering (HLI) studies allowing and .including a full reliability
relatively high capacities, with generation costs that are assessment of the system.
becoming Icompetit ive with conventional energy Wind farms are generally connected to medium-low
sources. In the UK the installed wind capacity grew voltage distribution networks and it is not necessary for
from 8MW in 1990 to 170.5MW in 1994 [J.J. During the wind farm owner to be the owner of the interfacing
the same period in Germany it soared from 60MW to network. The transmisslion network and conventional
632.2MW. The estimated installed wind capacity generators will be affected as the combined installed
worldwide is presently about 4000 MW. However, capacity of wind farms in the distribution network
because of their intermittent nature they cannot replace increases to significant levels. The load profiles seen at
conventional generation and are really energy-replace- the bulk supply points will change and so will the
ment rather than capacity-replacement sources 121. As power flows in both the transmission and distribution
- networks. Conventional generators will sell less energy
0IEE, 1996 and might be forced to operate outside optimum points
ZEE Proceedings online no. 19960488 because of the necessary load1 regulation action needed
Paper received 2nd October 1995 to compensate for the rmdom wind speed fluctuations.
The authors are with the Manchester Centre for Electrical Energy, Therefore up to five different parties can be identified
URIZIST, Mdnch1:ster M60 lQD, UK as being affected by the inclusion of wind generation in
IEE Proc.-Gener. Trunsm. nistrih., Vol. 143, No. 5 . September IY96 507
distribution networks; the wind farm owner, the inter- conventional units, cannot occur in the case of wind
facing distribution network, the customers, the trans- turbines because it would require different wind speeds
mission network and the conventional generators. driving the two turbines. However, this state would
Their perspectives are different and so can their inter- appear if statistical independence is assumed and the
ests and the benefits they derive from wind energy. individual capacity output tables are convolved. This is
Hence the information required by each of the parties a very simple example. In practice there are other
involved is different. effects that should be considered such as wake effects,
This paper focuses on the wind farms perspective spatial wind speed correlation, etc. In reality, identical
looking from inside the wind farm into the network. wind turbines can produce slightly different outputs
The results obtained are easily integrated into existing and these effects are considered in Section 5.
reliability assessment techniques allowing further stud-
ies to include the network. In this paper, however, the Table 1 Comparison between the convolution of capac-
network and load are not included: study results are ity outage tables for t w o conventional and wind gener-
valid up to the wind farm tee-point connection. Secu- ated machines
rity domain aspects such as voltage dips, voltage flicker
Combined output states example
and induction engine self-excita tion are not considered.
It is important to recognise that even when the Conventional generation Wind generation
impact at the HL1 or hierarchical level 2 (HL2) [3] are
Output States (9) Output States ( 12)
required, the input data needed will be that derived
from studies at the distribution level. Therefore the 2.0 uu 2.0 UUF
results given by the models described in this paper form 1.5 UR, RU -
the primary set of indices needed by all the parties 1.o UD, DU, RR 1.0 UDF, DUF, UUM
involved in order to identify the effect of wind genera-
0.5 RD, DR 0.5 UDM, DUM
tion on their activities.
0.0 DD 0.0 DDF, DDM, UUN,
2 Wind generation reliability modelling UDN, DUN, DDN
U: u p U: u p M: half rated speed
The total output of a wind farm is obtained by aggre- R: derated D:d o w n N: unsuitable speed
gating the individual output of each turbine. The usual D: d o w n F: full rated speed
approach to find the combined effect of many genera-
tors operating in parallel using probabilistic analytical Determination of the statistical output characteristics
techniques is the convolution of their individual capac- of a wind farm for reliability analysis requires the
ity outage probability tables [ 10, 31. However. this tech- simultaneous consideration of all wind turbines. There-
nique is not valid when wind generators are considered. fore a new wind farm generation model is needed
The problem is that the output of each wind turbine is which represents the stochastic characteristics of all
dependent on a common source: the wind. The statisti- processes involved. This can be accomplished by merg-
cal independence between the different generation out- ing a wind model and a wind turbine model.
put states is an underlying assumption in the
convolution of capacity outage probability tables. 3 Wind model
Therefore it is not possible to calculate the wind farm
capacity table by convolving the output table of each 3. I Modelling concepts
wind turbine. The wind [arm has to be considered in its Wind speed is a continuous physical phenomenon that
entirety. evolves randomly in time and space. A random varia-
This characteristic is shown in the following simple ble can be associated with each value of time. A sto-
example. The output states obtained from the convolu- chastic process is considered to be a model of a system
tion of capacity tables from two identical conventional which develops randomly in time according to probabi-
generators are compared with those obtained when two listic laws [ I I]. Thus, wind speed is a stochastic process
identical wind turbines are considered. In order to per- with a continuous state space (wind speed values) and
form the comparison, the conventional and wind gener- continuous parameter space (time). This type of proc-
ators are assumed to have the same individual output ess can be modelled approximately as a discrete state
levels. In the case of conventional generation, three space and continuous parameter state process.
generation levels are considered for each unit: full, der-
ated and out-of-service, corresponding to 1. 0.5 and
0p.u. output, respectively. For wind generation the tur-
bines are considered to be operational or out of service.
The wind has three speed levels each creating the same iJ-1 112 p3 PJ 111'1
individual wind turbine output levels ( I , 0.5 and Opu) Fig, 1 Birth ond &utI/ ~Mrirhoc.tlruiri
h , : transition ratc from state / to statc i+ I ; p, : transition from state 1 to
as for the conventional generators. The resulting com- state 1- 1
bined output states and their corresponding generation
states are shown in Table 1.
Five output states result from convolution of the Many studies have reported statistical tests on wind
conventional generators' capacity tables. However speeds using different probability distributions (Wei-
there are only four combined output states in the bull, Rayleigh, x2,...) It is generally accepted that the
equivalent wind generation case. The I .5pu combined Weibull distribution adequately represents the wind
output state with conventional generation is not possi- speed probability distribution [12-141 for most sam-
ble when wind generation is considered since it would pling times. This is all that is required to estimate the
require one unit generating at full output and the other wind farms' expected output. For reliability analysis,
at half output. This situation, perfectly possible for however, the probability and frequency distributions of
508 IE% Proc.-Gener. Trcinsrri. Dhtrih., Vol. 143, No 5, Septrmhet 1996
the wind speed stochastic process are also required. data is norinally values of wind speed observations
Therefore a new wind model is needed to reflect not averaged over ten minute periods. The data that needs
only the probability but also the frequency and dura- to be extracted from the sample record to calculate the
tion characteristics OF wind speed. model pararnctcrs are: (i) the number of transitions
'The proposed model to represent wind speed is a from state i to i -c 1; and (ii) the duration of the resi-
birth and death Markov chain (Fig. 1) [15, 161 with a dence tiinc In a stale before going to a different state.
finite number of states. The following assumptions are The noncoi-rtinuoits sampling process and the average
made in this model: values obtained for each observation mean that some
( 0 ) The annual wind speed values in a period (e.g. transitions From recorded data could occur between
annual, sea~~onal, etc.) are represented by a set of N nonadjacent wind speed states. This is an important
wind speed levels or wind speed states. State 1 corre- consideration because wind speeds do not increase or
sponds to tlhe lowest speed range and state N to the decrease instantaneously, but change continuously
highest. albeit over very short periods of time. Therefore the
(h) The wind speed model is statistically stationary. speeds will encounter all! intermediate states even if not
The stochasl.ic behaviour of the wind speed must be the monitored. If this is the case., the duration of the inter-
same at all points of time irrespective of the point of mediate states is estimated by a linear proportion of
time being considered, i.e. the probability of making a the sampling time.
transition from one given stale to another is the same The wind model parameters can then be calculated
(slationary) a t all times in the past and future (mean using two main methods depending on the adequacy of
and variance do noi~depend on the choice of a time the exponential distribution to represent the residence
origin). Wind speed is not strictly a stationary process t iines .
but the effect of the nonstationarity can often be
neglected as long as there is no trend in the data over 3.2.1 Wind speed duration follows an expo-
the period of time considered so that the process c m be nential distribution: The states are recurrent in a
treated as wide-sense stationary [I 71. finite state. irreducible Markov chain. The probability
(c) The distribution of residence times in a given state that the process returnis to any state an infinite number
of the birth and death process is exponential. I t is not of times is unity. Frorn the law of large numbers, the
necessary, however, for the wind speed discretised proportion of time spent in a state when the number of
model to follow this distribution. Analytical techniques returns to that state tends to infinity approaches the
only calculate long term average values. It can be reciprocal of the mean residence time. This time is
shown [18, 191 that the exponential distribution pro- equal to the parameter of the exponential distribution.
duces these values regardless of the residence time dis- The transition rate between any two states can then be
tribution th,at the system is actually follownig. Using calculated as
the actual statistical distribution of duration is only
important m7hen evaluating the time-dependent charac-
teristics of the stochastic process. where N , = number of transitions from state i to state
(d) The probability of a transition from a given wind , j 3and Dji = duration of state i before going to state .j.
speed state to another state is directly proportional to Therefore to make a good parameter estimation it is
the long-term average (i.e. quiescent or stmdy state) necessary that the numlber of interstate transitions is as
probability of existence of the new state. large as possible.
(e) Transitions between wind speed states occur inde- The exponential distribution gives a rough approxi-
pendently of transitions between wind turbines states. mation to the duratioii of wind speed runs when sam-
It should be noted that the techniques used to account pling intervals of one hour are used [20]. Studies were
for failure bunching do not imply dependence between carried out to check the validity of exponential resi-
the failures of wind turbines. Although the wind t UT- dence times in a given discretised wind speed range
bines may reside within a common environment which with real data extracte'd from real wind farm sites with
affects the failure rates of the wind turbines, the actual dirferent sampling times. Assuming an exponential resi-
failure process of overlapping outages sti I1 assumes dence time in each of the states, the parameters of the
wind turbine failures to be independent. There is no birth and death process were calculated. The model
suggestion therefore that the process is a common was solved to obtain the predicted probability and fre-
mode or dependent failure, only that the independent quency distributions. These were compared with the
failure rates are enhanced because of the common envi- ones obtained from the sample which, by expanding for
ronment. several years, is considered representative of the long-
v> From a given wind speed state, only the case of term wind behaviour. This comparison is shown in
transitions to immediately adjacent (lower and higher Table 2 from which ii can be seen that the statistical
speed) states are considered although the mlodels pro- distributions predicted from the model are quite differ-
posed later could treat equally the case of transitions ent from those obtained from the wind sample. Hence
from one state to any other state. this calculation method cannot be used since the wind
speed runs show a poor fit .with sample data when an
3.2 Wind mode/ parameter estimation exponential residence time is used.
The wind speed range of interest is divided into a finite
number of states that do not necessarily have to be 3.2.2 Wind speed duration follows an
equally spa'ced. This is an advantage because of the unknown distribution. Sample adjustment:
nonlinear characteristic of the turbine output curve The second calculation method does not require that
described in Section 4. The parameters of the wind the residence time of the stochastic process follows an
model are c:alculatetl from a wind speed rec:ord. Wind exponential distribution. It generates a finite state
speed is usually sampled at regular intervals. Available Markov chain with exponentially distributed residence
Table 2 Wind speed model statistical distributions assuming exponential residence times

Exponential Unknown distribution Sample Prediction


Speed distribution
Wind range
state
(m/s) % >Pi A, F, Probability Freq’ Probability Freq‘
(occ./year)
(occ./year) (occ./yea r) (occ./yea r) (occ./yea r) (occ./year)
1 0-5 4 162 - 3 805 - 0.305800 1372 0.029538 1104
2 5-6 30 635 27 871 13 474 13224 0.1 13300 5715 0.044123 1104
3 6-7 31 271 29 231 13620 13980 0.108500 5865 0.046255 2675
4 7-8 32 646 30 809 13776 15 159 0.100300 5500 0.046962 2922
6 9-10 35 367 33 959 14711 16939 0.070540 4363 0.047691 3272
7 10-11 35 922 35019 14 872 17 519 0.0581 30 3665 0.048181 3405
8 11-12 36 97 1 36 797 15 578 18 147 0.047260 3040 0.047051 3522
9 12-13 35 009 37 440 15072 18 123 0.0351 50 2433 0.046479 3681
10 13-25 39 066 71 783 139 6 303 0.025960 1774 0.226770 372 1
11 25 - 27 51 195 32 969 10229 23 868 0.049 100 500.7 0.026897 2447
12 27-30 18724 28 932 7 165 15533 0.000188 8.395 0.047626 1156
13 30-35 - 31 312 - 28 608 0.000023 0.7901 0.028510 543.9

Table 3 Results for the IEEE-RTS load model using a Markov chain

Individual State Cumulative State

Frequency Frequency
Frequency Duration (occ./year)
State Load Probability Probability (occ./year)
(occ./year) (year)
to lower states t o higher states
~

1 100.0 2.175 x 7.000 3.107 x 2.175 x 7.000 0.000


2 90.0 4.327 x IO-’ 8.800 x 10’ 4.917 x 4.544 x 10-2 8.100 x IO’ 7.000
3 80.0 1.567 x IO-’ 2.980 x IO2 5.257 x IO4 2.021 x IO-’ 2.170 x IO2 8.100 x IO’
4 70.0 2.178 x IO-’ 5.360 x IO2 4.064 x IO“ 4.199 x IO-’ 3.190 x IO2 2.170 x IO2
5 60.0 2.020 x IO-’ 7.090 x IO2 2.849 10-4 6.219 x IO-’ 3.900 x IO2 3.190 x IO2
6 50.0 2.335 x IO-’ 6.110 x IO2 3.819 x 8.553 x IO-’ 2.210 x 102 3.900 x IO2
7 40.0 1.357 x 10-1 2.480 x IO2 5.472 x 10-4 9.910 x 10-1 2.700 x IO’ 2.210 x 102
8 30.0 9.043 10-3 2.700 x IO’ 3.349 x 10-4 1.000 0.000 2.700 x IO’
9 20.0 0.000 0.000 - 1.000 0.000 0.000
10 10.0 0.000 0.000 - 1.000 0.000 0.000

times that replicate perfectly the probability and fre- be calculated. However, they should be generated from
quency distributions found in the sample. The parame- a wind speed record that is long enough to ensure that
ters are calculated from the principle of frequency the generated wind model is indicative of the long term
balance between any two states. The transition rates conditions found at the particular site of study. It
are calculated as should be noted that this model can also be used for
the same conditions to reproduce the statistical charac-
teristics of the load. The results obtained for the load
as specified in the 1EEE-RTS using ten load states are
where, Pi = state probability given by shown in Table 3.

4 Wind turbine model

The wind turbine model defines the output characteris-


Fy = frequency of transitions between state i and s t a t e j tics of the wind turbine as well as its failure and repair
and N = total number of states. processes.
Therefore the main advantage of this model is that it The wind turbine output curve determines the power
represents exactly the real characteristics of the wind output of the machine for corresponding wind speed
speed found at a particular site with a representation levels. The wind turbines can be divided into two major
suitable for conventional reliability analysis. Hence types according to the type of output control: pitch or
regardless of the sampling time, if the wind speed range stall regulated wind turbines (Fig. 2). Both can be rep-
to be considered is discretised into a finite number of resented by the proposed wind turbine model. The
states, it is always possible to find a finite birth and reactive power characteristic of the generator can also
death process with the same number of states that be defined in the same way although it is frequently
reproduces exactly the probability and frequency distri- specified as a percentage of the output active power
butions of the sample. since in most cases the power factor is corrected by
This model requires a small number of parameters to means of capacitor banks or SVAR compensators.
510 IEE Proc-Gener. Trnnsm. Distrib., Vol. 143, No. 5. September 1996
on the other hand are relatively simple machines and
so the maintenance time is much less than that of con-
ventional units. The most common machine used for
wind generation is the induction generator which is a
very reliable and robust electrical machine. As a result
wind units are usually left unattended for long periods
of time. It is also possible to further minimise the
impact of maintenance by making it coincident with
periods of low winds. With these considerations it
seems that preventive maintenance when using wind
generation has a significantly smaller impact than when
using conventional generation and therefore it is not
~~~

I I - included in the models.


cut-in rated cut-out wind
speed
5 Wind farm model
Fig.2 Typical output curve for pitch and Atall controlled wind turbines
-~ pitch control
- _ _ stall control The wind farm model comprises the wind model and
the wind turbine model. For each of the wind states
The failure and repair rates of the individual wind considered, the associated wind turbines’ failure and
turbines should be taken into account to callculate the repair processes are represented. States with the same
expected wind farm output. The structural safety of the wind turbines contingency are connected by the wind
tower and blades is specified by the manufacturer up to model transition rates. The resulting diagram can be
a certain wind speed value called the survival wind viewed as a set of layers (wind turbines states) con-
speed. The failure rate of the turbines as wind speed nected by the wind model transition rates (Fig. 4 for
increases also increases [21-23]. Failures under extreme two wind states, I and 11). For each wind farm state,
wind speeds are norrnally of a catastrophic nature such the number of wind turbines in service is known, as is
as cracking of the tower and blades. These ,Ire expen- the corresponding wind speed level. Therefore it is
sive, lengthy and costly failures with a low probability straightforward to determine the output of each wind
but whose expected Jfrequency should be carefully con- turbine by using the output curves. The total wind
sidered. Therefore two major wind speed ranges can be farm output for eachi specified wind farm state is
distinguished. A range within the design wind speeds obtained by adding the output of all wind turbines
and another above it comprising extreme wind speeds. while in that state. This; solves the problem of statistical
Consequently, different sets of failure and repair rates dependence to calculate the wind farm output.
are considered within each range.
The wind turbine is modelled as a binary ‘state com-
ponent, i.e. the machine is considered fully capable of
giving any output within its power curve limits (up) or
it is out of service (down). The failures of different
wind turbines are considered independent. Therefore
the resulting state transition diagram for a given down /
number of lurbines is equal to that obtained when a , /

down .I
system with the same number of independent compo- down down
nents is considered (Fig. 3 shows the diagram for two down
/’
,
UP ,
independent components). down

W’
Fig.4 Multi-layered windfirm state diagram for two wind turbinesfor
two wind stutes I und II
ll..

In general the transition rates for all states in each


wind speed layer could be different although it seems
realistic to consider only two sets of values, one for
normal and the other fix extreme wind speeds as stated
before. The impact of failures in the elements connect-
ing the different wind turbines inside the wind farm is
neglected.
The main advantages of this model are:
The addition of more wind state levels to increase the
precision of power out put and energy calculations is
simple.
More turbines can be easily added to the wind farm.
U It is always theoretically possible to have different
Fig.3 State riiugrum j b r twu hinury wiird lurbirres 9

sets of transition rates for each wind state.


Preventive maintenance in conventional ,generation States with different but non-zero output are not
units usually tend to be lengthy because of thie inherent directly connected, so that the cumulation of identical
complexity ‘of the systems involved. Wind generators output capacity states is simple.
IEE Proc.-Gener. Trunsm. DiAtrib., Vol. 143, No. 5, September 1996 511
The model can be combined with existing reliability The extended model also prevents unrealistic traiisi-
models of the distribution network to evaluate the tions from the ininor damage states to the severe dam-
overall impact of the wind farm. aged states which would create pessimistic results since
The effect of increased repair time necessary to it would have the effect of increasing the probability of
restore damaged parts under extreme wind conditions the severe damage states.
can be masked by the rapid transition to normal wind Some cases were evaluated to compare the above
speed levels with normal repair times. Consider Mods1 models. Consider an example of one wind turbine and
A shown in Fig. 5 for a one-turbine system and four four wind levels. The input parameters for the wind
wind levels. This is the same model as the one shown in speed transition rates and the wind turbines failure
Fig. 4 using a flat representation. The states belonging rates are shown in Tables 4 and 5, respectively. The
to each wind layer can be identified by having the same results are shown in Table 6 and are typical of the
capital Roman number. If the system is in state 4 (Up- results obtained in other cases analysed. The differ-
ences between the two models are small (for example,
IV) and severe damage occurs, it goes to state 8
the frequency, of severe damage, 0.23Xflyr given by the
(Down-IV). The large restoration time needed to repair
simpler model and 0.236flyr given by the extended
severe damage would not be taken into account if the
model). Model A tends to give slightly more pessimistic
system transits to state 7 (Down-111) due to a wind
results due to the prevailing effect of transitions from
change. This is a minor damage zone and the repair
minor damage states to severe damage states over the
time in such states corresponds to minor damage.
opposite direction transitions.
Therefore the severe damage effect is not reflected
correctly. Table 4 Wind speed transition rates in occ/yr

Wind speed model transition rates

= 200 ~ 1 1 1=
11400
1 ~ l l I / l V= 10
hll/l = 20 ~ 1= 300
11,11 ~ l v , l l l= 500
I

down Table 5 Wind turbine model transition rates


__..-

6 I I Wind speed regime


I I
I / Normal Extreme
I
\ minor damage
_ _ ___---____-_--
1 '\ I
I
_ _ _ _ _ /'
\ - -/ \ Failure rate (occ./year) 4 24
minor&severe Repair rate (occ./year) 90 24
damage ~~~~ ~~

hg.5 Model A for a 1vind.furm c?f m e turbine and j&r wind levels
Another important aspect as the number of wind tur-
bines is increased is the number of states that result.
This shortcoming can be overcome at the expense of The gain in theoretical precision from Model B may be
increased complexity and an enormous number of a disadvantage in practical situations. If the number of
states by using the alternative Model B shown in states is very large, many states will have probabilities
Fig. 6. In this model any failure under extreme wind that could differ by several orders of magnitude. The
speed conditions is considered to create a separate ill-conditioning of the problem will be degraded and
damage state. Separating these severe damage states in the computed values will be subject to large precision
this way prevents a change in the wind making the sys- errors. Therefore the increased complexity and small
tem transit to minor damage state. Only the repair of differences expected in real cases indicates that Model
the severe fault allows the system to return to the nor- A (Fig. 5) should be used for these studies.
mal damage states region. Therefore a new transition If a wind turbine is exposed to the turbulent air wake
rate for the repair of severe damage is needed. Differ- of another turbine, the amount of power that can be
ent repair rates for severe damage can also be consid- extracted is less than that if it was exposed to free air
ered. However, correlation between the repair time and flow. This effect is minimised for a carefully laid out
the duration times of severe wind speeds and its recog- wind farm . However if the wind has significant varia-
nition would be very difficult. Therefore it is more rea- tions in direction. this effect might be significant for
sonable to consider a repair time for minor and one for those occasions when the hubs turn to face particular
severe damage, irrespective of the wind speed regimes. wind directions. To consider this effect, a correction

I
1
I
I I I

-.

...

I
1 1 1
minor damage I I severe damage I
'\.-----__________-____
/' '\ ,
,
/
\

Fig. 6 Model B j b r U wind,furm of one turbine andfour ~iitidlevels

512 IEE Pro(.-GenPr Tvanrm. Distrih , Vol. 143, No. 5, September 1996
Tablle 6 Results obtained with the two models
Model A Model B
-
State Probability Frequency Duration State Probability Frequency Duration
(occ./year) (years) (occ./year) (years)
___-
I 0.038863 7.928 0.004902 1 0.038608 7.882 0.004902
2 0.388366 164.667 0.002358 2 0.385703 1163.538 0.002358
3 0.517541 162.508 0.003185 3 0.513962 1161.384 (0.003185
4 0.009919 5.198 0.001908 4 0.009854 5.163 0.001908
5 0.00 1787 0.518 0.003448 5 0.001712 0.496 0.003448
6 0.018138 9.250 0.001961 6 0.017095 8.719 0.001961
7 0.024465 9.786 0.002500 7 0.022778 9.1 11 0.002500
8 0.000921 0.483 0.001908 8 0.000435 0.228 0.001908
9 0.000331 0.0741 0.004464
10 0.003706 1.646 0.002252
11 0.005265 1.758 0.002994
12 0.000552 0.289 0.001908

Zero output states Zero output states


1,4,5,6,7,8 0.094093 16.566 0.005680 1,4,5,6,7,8,9, 0.100345 16.452 0.006099
10,11,12
Nonzero output states Nonzero output states
23 0.905907 16.566 0.054684 23 0.899655 16.452 0.054681
Severe damage frequency Severe damage frequency
443 0.009919 x 24 = 0.238 9,10,11,12 0.236
Minor damage Minor damage
0.041585
5,6,7 0.044389 4.240 0.010470
--3.970 0.010474

factor in the power generated is applied to I hose wind 6 Solution of the wind farm model
turbines more likely to experience this effect. The possi-
bility that ,I waked turbine can extract more power Solution of the wind farm model involves calculation
when the waking turbine is out of service i:; neglected of the probability ancl frequency of each state in the
for the same reasons given above regarding mainte- system.
nance and repair times. The limiting state probabilities are given by the equa-
tion a P = a, where P is the stochastic transitional
Different types of wind turbines having different out-
probability (STP) matrix 1191 of the system and cx the
put curves and/or failure and repair characteristics are
state probability vector. It follows that a(P-I) = 0
treated as different components in the layers. In this where I is the identity matrix. The P-I matrix is not
way it is simple to add the effects of spatial (correlation linearly independent, another equation is needed saying
or waked effects as new turbines (components) with that the sum of problabilities is one (PZ'matrix).Two
different output curves and/or failure and repair problems arise in solving the proposed wind farm
parameters. model as the number of wind turbines and/or wind
I 11 Ill IV states increases. The model :shown in Fig. 7 is for two
turbines and four wind states, and can be compared
with the one shown in Fig. 4 and the one shown in
Fig. 5 , Model A, for one wind turbine. The problems
are: (i) reduction of the number of states resulting from
the wind farm model; (ii) construction of the STP
matrix representing the connections between the differ-
ent states (or alternat.ively the PI' matrix to calculate
directly the states probaibilities).
The number of states that result from a full Markov
analysis of the wind farm m.ode1 with I I turbines, each
machine having two states, is 2n times the number of
wind states. As the number of wind turbines increases,
the number of states in the combined wind farm model
reaches enormous proportions. For example, in a wind
farm with 30 turbines (11 = 30), the resulting number of
states per wind state is; 1073 741 824. This clearly shows
that the solution of the proposed model in terms of the
full Markov matrix becomes practically impossible
when the number of turbines in the wind farm is
increased to commercial levels.
This difficulty has been overcome by reducing the
number of equations created for the generalised system
IEE Proc -Gene? Trrmsm Distrib., Vol. 143, No. 5, S~p.ptrmherI996 513
by considering that the probabilities of those states can be represented as divided into small submatrices as
with the same contingency level are the same. The shown in eqn. 4. The submatrices A and D (eqns. 5
resulting number of equations becomes realistic and the and 8) along the major diagonal correspond to subma-
results obtained are the same as those which would be trices WT, and WT2,respectively, in eqn. 3, and sub-
obtained from a full Markov analysis. The main prob- matrices B and C (eqns. 6 and 7) on the minor
lem is to describe an algorithm which directly translates diagonal correspond to W1+2 and W2+1, respectively,
the complete matrix into its smallest possible form in eqn. 3. The wind transition rates in the elements of
avoiding storage of the initial number of states. This the main diagonal in submatrices A and D have not
matrix is called PItCLR (contingency level reduced PI been included for clarity purposes. In the case of two
matrix). wind states there are no submatrices of type 2.The
The problem associated with a large number of states submatrices along the diagonal can also be subdivided
becomes worse when several types of turbine are con- into smaller boxes. For each contingency level of each
sidered within the same wind farm. In this case, the type of wind turbine present in the wind farm all con-
resulting reduced matrix using the above method shows tingencies of other types are considered recursively. In
an irregular pattern that has to be corrected by means this way the failure and repair rates of the wind tur-
of further reordering of the equations. The matrix then bines occupy diagonals which can be identified and
presents a structure that is retained as the number and1 described in an algorithm. It should be noted that this
or types of turbines increases and which is suitable for matrix is not a STP matrix, the elements in the main
a computer algorithm. Each row in the matrix repre- diagonal do not add up to the bum of elements in each
sents a unique contingency state of the different types row. Therefore it is not directly possible to derive a
of turbines in the wind farm. For each contingency corresponding simplified state diagram:
level of a particular type of wind turbine, all possible
combinations of contingencies are considered recur-
sively. The diagonal elements contain the negative sum (4)
J
of the transition rates departing from each state. 0 ...
Although the size of the system of equations is reduced ...
2x2
by assuming those states with the same contingency to
(2&1+2P2) ’ ’ ’
have the same probability, the states at this level are ...
0
not lumped but treated individually. Therefore calcula-
tion of the individual state frequency is straightforward A= 0 ...
...
when the state probabilities have been calculated. 21’1

...
Finally, after considerable manipulation of the system (1

equations, a completely general algorithm was created 0 ...


for writing this matrix. The solution algorithm can 0 ...
directly create the smallest possible transition matrix
without losing any information for a system made up ... ...
XI 0 0
of n binary components with different interstate transi- ... ...
0 A1 0
tion rates in a multilayered environment. ... ...
0 0 X I
The general structure of the PICLRmatrix is shown in ... ...
0
eqn. 3. It can be subdivided into as many groups of ...
-(2hz+I’l+Xl) A2
...
rows and columns as the number of states used in the 2112 - 2+PZ
1 +/’I + 1 2x2
... 112 ...
wind model. Three different types of submatrices can 0 -(x1+1’1+2U2)
... ...
be identified: WLI 0 0
... 0 I’ I 0 ...
The WT submatrix. This submatrix contains the ... ...
0 0 PI
resulting transition rates derived from the analysis of
the wind turbines failure and repair processes in each
wind state. ... U

The Wj+j submatrix. This submatrix contains the ... 0

information about the transition between wind speed ... n


states i and j . ... 2xx1

The Z null submatrix. All elements in this submatrix


*.. 0
... n
are zero.
... -izv,+2x21
All these submatrices have the same dimension and
have the same number of rows and columns:

1 : :!
WTl

i:
WIi2
w211 wT2 w2+3 z - A n 0 0 0 0 0 0 0 1
z w3+2 wT3 w31-4 ‘.’
O X A 0 0 0 0 0 0 2
: : (3) 0 0 0 0 0 1
As an example of the resulting PI’CLR matrix con- 0 0 0 X A O 0 0 0 2
sider a wind farm with two types of wind turbines and B= 0 0 0 O A A 0 0 0 4
two wind speed states. The wind speed transition rates 0 0 0 0 O X A O 0 2
are hA and h,. The transition rates for the wind tur- 0 0 0 0 0 O X A 0 1
bines are A,, p1, A2, p2for normal wind speeds and A’,, 0 0 0 0 0 0 O X A 2
p‘,, At2, pt2for extreme wind speeds. The PI’,,, matrix .o 0 0 0 0 0 0 0 1
514 IEE Proc.-Cener. Transm. Distrib., Vol. 143, No. 5, Septemhrr 1996
A B 0 0 0 0 0 0 0 0 where
O A R 0 0 0 0 0 0 0 &y = equivalent transition rate from subset a to sub-
0 o x g o 0 0 0 0 0 set6,
0 0 0 A B 0 0 0 0 0
A, = transition rate from state i of subset a to statej of
C= 0 0 0 o x g o 0 0 0
subset 0,
0 0 0 0 o x g o 0 0
0 0 0 0 0 o x g o 0 PI= probability of state i of subset a.
0 0 0 0 0 0 0 x g 0 Sparsity techniques for the storage of the Markov
0 0 0 0 0 0 0 O A R transition matrix can also be used to take advantage of
(7) the sparse nature of the matrix resulting from the state
0 ... diagram to solve the Markov model.
2x; ...
(Zx;+2WL;) ... 7 Wind generation indices
0 ...
... Wind generation indices have been defined to reflect
,D = 11
the specific aspects affecting the wind farm perform-
2P.i
...
...
ance [24].
0

0 ... Installed wind power (IWP). Sum of the rated power


0 ... of all the wind generators in the wind farm.
Installed wind energy (IWE). Installed wind power
... multiplied by a year, tlhis is the energy that could be
A ;
... extracted if the units could be operated continuously
0
... (IWP x 8760).
0

... -(2x;--,1; +A;)


Expected available wind energy (EAWE). Expected
... 211:
amount of energy that would be generated in a year, if
... 0 there were no wind turbine generator (WTG) outages.
... P ; Expected generated wind energy (EGWE). Expected
... 0 maximum amount of energy that would be generated
... 0 in a year by the real WTGs considering their outage
rates and the real wind to wlhich they are exposed.
... Wind generation ratios:
wind generation availability factor WGAF: EGWEi
IWE
wind generation utiilisation factor WGUF: EWEUi
EGWE
capacity factor: WCrAF x WGUIF.

... Table 7 Wind speed probability table

Prob,abi,ity Frequency Duration


State Range
Many states in the wind farm model correspond to (4s) (occ./year) (year)
-~
the samc output level. For further network reliability 1 0-5 0.347’1254 1321 2.62828 x IO4
analysis, it is necessary to generate the wind farm out- 2 5-6 0.09!38773 2666 3.74566 x
put table with generation states sorted by output
3 6-7 0.0962636 2657 3.62319 x
power. Therefore the original results are sorted by
power output, cumulated by grouping them with a 4 8-9 0.0864924 2503 3.45608 x
desired step length and the frequencies of going to 5 9 - 10 0.07134044 2286 3.34281 x
higher or lower generation levels calculated The main 6 10 - 11 0.064!5943 2044 3.15963 x IO”
difficulty is the calculation of the frequency of depar- 7 11 - 12 0.0542400 1757 3.08731 x
ture from each new cumulated state to lhigher and 8 12 - 13 0.0444518 1499 2.96522 x
lower generation states. This frequency is calculated as
9 13 - 14 0.03132090 1268 3.01250 x
the product of the icumulated state probability and its
departing transition rates. Additional algcirithms are 10 1 4 - 25 0.09113673 588.6 1.55232 x IO4
therefore necessary to calculate the equivalent transi- 11 25 - 27 0.00~0!5316 18.13 2.93280 x
tion rates between t he subsets of states representing the 12 27 - 30 0.001033542 7.976 4.44101 x
new cumulated stales in the transition matrix. These 13 30 - 35 0.00~00887 2.538 3.49556 x
are given by --
Values i n the sample: 289 ‘152
Valid observations: 269 57:2 (93.23%)
Sampling time: 10 m i n
M a x i m u m speed: 33.724rri/s
(9)
Mean w i n d speed: 7.035m/S
Variance: 18.273 (m/sP
2Ea

IEE ProcGener. Transm. Di.slrib., Vol 143, N o . 5, Seprernher. l Y Y 6 515


301
s of 5.5 years of 10 min average wind
speed values giving a total of 289 152 observations.
25 However, Failure in the anemonieters or the recording
equipment reduced the number of valid values to
269572 (93.2'Xr). The wind speed state space consists of
13 states as shown in Table 7. The wind speed duration
curve i s shown in Fig. 8.
The wind turbine output curve used 1241 in this
example is shown in Fig. 9. This has been discretisecl
into 11 points. The wind turbine failure and repair
characteristics under dif'f'crent wind regimes are shown
in Table 8.
The order or the reduced matrix resulting from tlie
transition diagram is 195. The number of nonzero ele-
ments is 1 11 1 Liild hencc the sparsity ratio is 97.1'Xi.
T h e results For the first ten or the 195 levels are shown
in Table 9. These results are obtained after the wind
r--- - - ~- output states have been cumulated and sorted by
I125
5Ol
/
i
power output. The wind farm output duration curve is
given in Fig. 10.

1
Table 8 Wind turbine transition rates 10

Wind speed regime

Normal Extreme
. - ~ _ _ _ _ _ _ _ _ _
Failure rate (occ./year) 1.oo 12.00 The resulting wind generation indices and ratios are
shown in Table IO. The wind farm capacity table for
Repair rate (occ./year) 50 10.00
network reliability studies. reduccd to 15 states, is
shown in Table 11.
8 Numerical example
9 Discussion
The models described in the previous Sections have
been iinplemented into a Fortran 90 computer pro- The results obtained from the wind farm model are
gram. Thc rollowing example illustrates an application very satisfactory. The wind model adequately repre-
of the iiiodcls to a realistic wind farm containing 14 sents the frequency and probability distributions found
turbines cach rated at 1 501tW. The wind profile in the sample. The total wind fai-rn output found by
obtained fi-om a real wind f;irm site has been used to combining the wind and wind turbine models provide
calculate the parameters of the wind speed model. The very detailed information.

Table 9 Results obtained from the wind farm model after sorting the cumulated states by power
output

Individual Cumu Iated Frequency D u ratio n Wind Turbines Power


Number States
probability probability (occ./year) (year) state UP ikW!
________ -__-___ - ____ ____
1 1 6.9274 x 6.3274 x IO-' 4.4723 .? 102 1.5490 x 1 0 ' 10 14 2100.00
2 14 1.3834 x 1.9367 x 8.9991 1.5373 x I 0-4 10 13 1950.00
3 91 2.7628 x 2.5141 x 1 0 ~ 3 1.8108 x IO-' 1.5257 x 10 10 12 1800.00
4 364 5.5178 x IO-' 2.0085 x 3.6436 x IO-' 1.5144 x IO-' 10 11 1650.00
5 1001 1.1011 x 10-8 1.1032 10 5 7.3317 x 1.5032 x IO-" 10 10 1500.00
6 2002 2.2014 x 10 l o 4.4072 x 10~' 1.4753 x 1.4922 x I 0-4 10 9 1350.00
7 1 2.8970 x 2.8970 x 1 0 ~ 2 9.6207 x IO2 3.01 12 x 9 14 1332.00
8 14 5.7849 x 8.0988 x 1.9239 x I O ' 3.0068 x 9 13 1236.83
9 3003 4.3980 x I O - I 7 1.3207 x IO-* 2.9690 x 1.4813 x 10 8 1200.00
10 91 1.1552 x 1.0512 x 3.8476 x IO-I 3.0024 x 9 12 1141.69
t y turbines IS very high N o t
Hie c ~ \ a i l c ~ b ~ofl ~wind ~pccds.This would increase the generated energy and
s~irpiisingly the ovcr.iIl availability ligui e I S very good alw the capiciiy factor but a1 the cost of ~ncrcasingthe
cd to conveiitiondl generatoi S . I i i this c a ~ ethe freq ucncy of sevci e damage5
I S 0 98041 (Table 10) Phis figure includes the Table 9 contains the detailed iiifcmmtion requiied by
ettect of I ~ i l u i e sin normal and evtienic uind \peed\ [he mind farm owner I-Iowwer, the iietwork planncr
tlowcvei . tlie effect of extieme wind speed^ upon this docs not iieed this detailed information. 7 he most
mliid f<11111it, 11egllglblt: Thc expected WGCJF wltliol~t iinportaiit resulls for t kit. interfacing network are found
coiivdei ing i lie effect of extrenie wind spccd4 f c i i l u i es IS in Table 1 1 ~ v l i i ~includes
h ilre probability and fre-
0 08044 and the fiequcncy of sevcic d‘imaper in t h i ~ quciicy cii\tributioiis ol ihe wind farm a s seen from the
c‘iie is ‘in excellent 0 00796 occiye:rr rhls was expected tee coniiection point This table IS very similar to the
hec‘iuw capacity outage probability table [I01 of a conventional
( C L ) Ilic medii \ m i d spccd IS low [or the lurbine ouiput generator with mulLiple (derated states and hence males
curve used llt IS at the 5ta1t of the nonl1ncar 1x11 t of the the tieatniciit of a wind farm for other iietwork relia-
t ~ i bine
i output tunc vcry necu the c ~ ~ t - i,peed n The bility ~tiidiesgreatly simplified I t can be also observed
stciindard deviation I S 4 24m/s Thercfoi c u ind speed from Table 1 I that. due 10 the grouping process of the
s h y \ well within the iatcd and ~ u i - v ~ vwind ~ i l speed foi cumulated states, somc cumulated states (state 5 ) may
most OJ the t i m e h o w very low pi ohabilities
(/I dtliough the li cqucncy of the extreme wind speed The \pai sity ratio is Imnsiderablc (0 97) justifying the
\tcite 15 3 5:;80cc/ycui. itr piobabilily 15 iiegligihlc dt u x of s p i se ~ i i ~ices
i t i 1111 the calculdtioiis. If inorc wind
0 0000887 l u r b i n c ~of the wine or Ihfferent types are added to the
wind fcu-m the order of the matrix grows con\idercibly
Table 10 Wind generation indices for the numerical In thLitc a ~ etruncation techniques for those states with
example hig1.1cr conlingencicc 1171gli1 be considered, however, the
icsultj would then iiot be exact.
Wind generation indices

INIP 2100kW 10 Conclusion


IWE 18 396.0 M W h
EAWE 3966.6MWh
A riew probabilistic motki of il wind farm that can be
used in reliability sLudics using analytical techniques
EGWE 3888.9MWh
has been presented. The model c o n d e r s the stochastic
WGAF 0.2140
nature oi‘ wind, the failure and rcpair processes of the
WGUF 0.98041 wind turbiiies under cliffcrciit wind regimes, and tlie
WGAF x WGlJF
- 0.2114
- - wind turbine output cliaracteristics in detail. Spatial
~
wind speed correlation ;and waking effects can also be
I t can be c:onclLided that this machine is overrated for cotisidci-cd. Tlie model c:iiii be used with one or several
thc conditions found 011 the particular wind far111 site. typcs of wind turbines and is easily integrated into
This is coiifirtiied again by the low value of the capac- existing reliability a ment models for distribution
ity I‘ictor (0.21. Tablc 10) which docs not iiicludc any nctworlts. A n~inicricalapplication of the models to a
notwork effkct. 7‘1iis is iiormally about 0.3 for ii wcll realistic wind farm cast: is also presented. The model
(1csigiied wirid f i rm G iven th is infoi-m;t li o 13 L hc wi nd
I
s ~rcces~~fullyprod uces the in for m a t ion required for each
fa 1-111 owner might consider I-cplacing thesc turbines o f tlie partics involved with a detailed representation of
wil h others with lower cut-in. rated arid C L I ~ - O L I Lwind the stochastic processes; iiivolvcd.

Table 11 Wind farm output table for network reliability studies

Probability Frequency
~ _ _ _ _ Duration Power
State To highet To lower (year)
Individual cumulated Individual states (kW)
states
~___ ~____
I 6.927 x I O 1.000 4.472 x I O 2 0.000 4.472 x I O ’ 1.549 x IlP4 2 100.00 1 274 355
2 1.937 x 1 0 ~ 7 9.307 x I O - ’ 1.260 x I O ’ 4.472 x I O ’ 5.713 x I O 2 1.537 x 1950.00 330 834
3 2 . 5 ’ 1 4 ~10-3 9.114 x IO-’ 1.648 x I O ’ 5.713 x I O ’ 5.872 x I O 2 1.526 x 1800.00 39 642
4 2 . 0 0 9 ~I O 4 9.088 x 1 0 ~ ’ 1.326 5.872 x I O 2 5.885 x I O 2 1.514 x 1650.00 2 903
5 1.103 x 1 0 ~ 5 9 086 x IO-’ 7.339 )r IO-’ 5 885 Y 10) 5.886 x I O 2 1.503 x I0-4 1500.00 145
6 3.707 x 1 0 ~ 2 9.086 x 10 ’ 1.231 103 5.886 x 10) 7.018 x I O 2 3.012 x 1311.19 425 780
7 3.476 x IO-’ 8.716 x 10 ’ 1.172 r I O 3 7 018 x 10) 7.920 x I O ’ 2.965 x 10-5 1071.61 326 259
8 1.073 x IO-’ 8.368 x IO-’ 3.642 x I O 2 7.920 107 8.191 x I O ’ 2.962 x 984.74 92 586
9 5.273 x 10~2 8.261 r 10 ’ 1.708 r IO3 8 191 107 9.588 x 10’ 3.087 10-5 820.19 378 822
10 5.059 x IO-’ 7.734 x IO-’ 1.603 r. I O3 9 588 107 1.073 x IO3 3.156 x 625 92 277 382
11 1 . 5 6 2 ~I O ’ 7.228 x 10 ’ 4.949 >: 102 1.073 x I O 3 1.107 x I O 3 3.155 x 572.78 78351
12 7.641 x I O 2 7.071 x 10 ’ 2.286 >. I O 3 1.107 x I O ’ I,204 103 3.343 10-5 430.84 288 391
13 1.595 x I O - ’ 6.307 x 10 ’ 2.530 x: IO3 1.204 x I O 3 1.351 x IO3 6.304 10-5 218.95 305 888
14 1.231 x 1 0 ~ ’ 4.712 x IO’ 2.659 x 10” 1.351 x I O 3 1.333 x I O 3 4.631 x 62.67 67612
15 3.481 x 1 0 ~ ’ 3.481 x 10 ’ 1.333 x I O 3 1.333 x I O 3 0.000 2.611 x 0.00 0
11 References 13 EGGLESTON, D.M., and STODDARD, F.S.: ’Wind turbine
engineering design’ (Van Nostrand Reinhold, New York, 1987)
‘Wind directions’. B W E A Newsletter, 1995, XIV, (4). p. 19 14 JOHNSON, G.L.: ‘Wind energy systems’ (Prentice Hall, Engle-
ALLAN, R.N., and CORREDOR AVELLA, P.: ‘Reliability and wood Cliffs, NJ, 1985)
economic assessment of generating systems containing wind 15 ANDERSON, W.J.: ‘Continuous-time Markov chains’ (Springer
energy sources’, IEE Proc. C, 1985, 132, (I), pp. 8-13 Verlag, New York, 1991)
BILLINTON, R., and ALLAN, R.N.: ‘Reliability assessment of 16 CHUNG, K.L.: ‘Markov chains with stationary transition proba-
large electric power systems’ (Kluwer Academic, Boston 1988) bilities’ (Springer Verlag, New York, 1967)
DESHMUKH. R.G.. and RAMAKUMAR. R.: ‘Reliabilitv anal- 17 THOMANN, G.C., and BARFIELD, M.J.: ‘The time variation
ysis of combined wind-electric and conventional generation sys- of wind speeds and windfarm power output in Kansas’, IEEE
tems’, Solar Energy, 1982, 28, (4), pp. 345-352 Trans. Energy Convers., 1988, 3, (l), pp. 4 U 9
GIORSETTO, P., and UTSUROGI, K.F.: ‘Develoument of a 18 SINGH, C., and BILLINTON, R.: ‘System reliability modelling
new procedure for reliability modeling of wind turbine genera- and evaluation’ (Hutchinson, London, 1977)
tors’, IEEE Trans., 1983, PAS-102, (l), pp. 134143 19 BILLINTON, R., and ALLAN, R.N.: ‘Reliability evaluation of
WANG, X., DAI, H., and THOMAS, R.J.: ‘Reliability modeling engineering systems’ (Plenum Press, New York, 1992, 2nd edn.)
of large wind farms and associated electric utility interface sys- 20 COROTIS, R.B., ARDEN, B.S., and KLEIN, J.: ‘Probability
tems’, IEEE Trans., 1984, PAS-103, (3), pp. 569-575 models of wind velocity magnitude and persistence’, Solar Energy,
SINGH, C., and LAGO-GONZALEZ, A.: ‘Reliability modelling 1978, 20, pp. 483-493
of generation systems including unconventional energy sources’,
IEEE Trans., 1985, PAS-104, (5), pp. 1049-1056 21 CARTER, J.A., JOHNSON, B., SHERWIN, R.W., and RADE-
SINGH, C., and KIM, Y.: ‘An efficient technique for reliability MAKERS, L.W.M.M.: ‘Failure modes and effects analysis for
analysis of power systems including time dependent sources’, the AOC 15/50 wind turbine’. Proceedings of the AWEA Wind-
IEEE Trans. Power Syst., 1988, 3, (3), pp. 1090-1096 power’93 Conference, 1993, pp. 479486
BILLINTON, R., and CHOWDHURY, A.A.: ‘Incorporation of 22 CLARK, R.N., and DAVIS, R.G.: ‘Performance of an Enertech
wind energy conversion svstems in conventional generating caoac- 44 during 1 1 years of operation’. Proceedings of the AWEA
ity adequacy assessment’; IEE Proc. C, 1992, 13G, ( I ) , pp.-47^56 WindPower’93 Conference, 1993, pp. 204-212
10 BILLINTON, R., and ALLAN, R.N.: ‘Reliability evaluation of 23 SCHLUTER L.L. and VACHON W.A.: ‘Optimizing wind tur-
power systems’ (Plenum Publishing, New York, 1984) bine control system pammeters’. Proceedings of the AWEA
11 PAPOULIS, A.: ‘Probability, random variables and stochastic WindPower’93 Conference, 1993, pp. 175-1 82
processes’ (McGraw-Hill, New York, 1991, 3rd edn.) 24 ‘Technical and economic effects of connecting increasing numbers
12 FRERIS, L.L.: ‘Wind energy conversion systems’ (Prentice Hall, of large wind turbines to weak public utility networks’. Report
London, 1990) JOU2-CT92-0095. CEC DG XI1

518 IEE Proc.-Gener. Trunsm. Distrib., Vol. 143, No. 5, September 1996

You might also like