See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/3268561

System Reliability Assessment Method for Wind Power Integration

Article  in  IEEE Transactions on Power Systems · September 2008

DOI: 10.1109/TPWRS.2008.926090 · Source: IEEE Xplore

CITATIONS READS
88 121

3 authors:

François Vallee Jacques Lobry

Université de Mons Université de Mons
126 PUBLICATIONS   757 CITATIONS    151 PUBLICATIONS   856 CITATIONS   

SEE PROFILE SEE PROFILE

Olivier Deblecker
Université de Mons
140 PUBLICATIONS   689 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

ADABEL - Advanced Data Analytics ensuring cost-efficient security of supply and balancing throughout the BELgian energy transition View project

Energies_MDPI - Guest Editor of Special Issue "Steady-State Operation, Disturbed Operation and Protection of Power Networks" View project

All content following this page was uploaded by François Vallee on 11 December 2015.

The user has requested enhancement of the downloaded file.

1288 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 3, AUGUST 2008

System Reliability Assessment Method

for Wind Power Integration
François Vallée, Jacques Lobry, and Olivier Deblecker

Abstract—Wind power penetration is expected to be largely in- are problematic as those units are generally required to cover
creased in the near future. Nevertheless, due to the high variability the peak load and present long delay when starting up. So, this
of wind, the increase of the installed capacity in wind energy could research focuses on elaborating a new method to estimate the
sometimes lead to a stopping of nonadjustable power plants if the
planning of those units was still decided without considering a re-
wind capacity of a given country (here, Belgium). The pursued
alistic estimation of the available wind potential. In that way, and objective is not to traditionally define a wind capacity equiva-
as it is illusory to imagine that all the installed wind capacity will lent to some particular technology (thermal or nuclear produc-
always be produced, this article proposes a new method, based on tion) but rather to relate the calculated wind capacity to the “av-
a discrete convolution process, to compute a two-state probabilistic erage” wind generation for the considered country. By doing
model for wind generation and to define, so, an equivalent capacity so, a wind capacity not depending on the equivalent considered
for global wind production in order to introduce it in the predic-
tive peak load covering process. This global model is, in fact, based
classical technology can be established and be used as first esti-
on the convolution between each single wind park multistates his- mation for wind production in the optimal dispatch. Obviously,
tograms and permits to compute accurate equivalent capacity for hourly wind production will fluctuate around this fixed capacity
the wind production of an entire country. but this last one has the advantage to give a first reliable estima-
Finally, note that, here, the proposed method is applied to the tion of the expected mean hourly wind potential for the consid-
Belgian production park. ered country.
Index Terms—Power control, power system modeling, Weibull The article will thus be organized as follows. In the first sec-
distributions, wind energy. tion, the probabilistic distribution for the Belgian wind park is
calculated, based on the convolution between the different wind
units established in Belgium. Thanks to this statistical distribu-
I. INTRODUCTION tion, a forced outage rate is determined for wind gen-
eration in Belgium. Then, in a second part, in order to calcu-
A S the prospects for the future years predict, for number
of countries, an increase of the wind penetration in the
electrical networks, it will become an absolute requirement for
late the equivalent wind capacity to be included in the proposed
two-state model, a nonsequential Monte Carlo simulation [3],
the market parties to take that kind of energy into account when [4] is presented and leads to an equivalent installed capacity for
planning the operation of nonadjustable nuclear units and big wind generation. Finally, a conclusion points out the interest of
thermal plants. Indeed, actually, the only solution for variability introducing a realistic classical two-state model for wind gen-
and limited predictability1 of wind energy is to use conven- eration in place of the former -state one usually dedicated to
tional generation units to compensate. In that way, technical that kind of fluctuating energy (lower computation time and re-
capabilities of thermal generation, mainly consisting of com- inforced accuracy in the production park management) [5].
bined heat and power , have already been investigated
in [1] in order to determine wind power’s impact on thermal unit II. PROBABILISTIC DISTRIBUTION OF THE
commitment. So, it has been shown that wind unpredictability BELGIAN WIND PARK
had few effects on operation cost and reliability of the consid- A. Wind Speed Distributions in Belgium
ered system. However, in case of increased wind inputs, the
large-scale integration of wind power into the energy system As shown in Fig. 1, the Belgian wind speed repartition can
can involve important excess of electricity [2] leading to the ne- be divided in three great regions including: the Polders near the
cessity of stopping highly powered thermal or nuclear classical sea, the central part of the country and, finally, the Ardennes in
units. Those situations of big classical units’ untimely stopping the South [6].
Based on the previous wind repartition, the need in wind
speed measurements for a site located in each of the three ge-
Manuscript received February 16, 2007; revised April 2, 2008. Paper no. ographical regions of interest becomes obvious. In that way,
TPWRS-00098-2007.
The authors are with the Electrical Engineering Department, Fac- thanks to the confidential data base collected by the Royal In-
ulté Polytechnique de Mons, B-7000 Mons, Belgium. (e-mail: Fran- stitute of Belgian Meteorology, the access to wind speed mea-
cois.Vallee@fpms.ac.be; lobry@magnetism.fpms.ac.be; Olivier.De- surements for Ostende (North), Uccle (Centre) and Saint-Hu-
blecker@fpms.ac.be).
Color versions of one or more of the figures in this paper are available online bert (South) was possible. The first treatment on these data con-
at http://ieeexplore.ieee.org. sisted in the calculation of a probabilistic histogram for each of
Digital Object Identifier 10.1109/TPWRS.2008.926090 the three available sites.
1Wind prediction errors can involve, beyond 36 h, large uncertainties on the
As the precision for the collected wind speed measurements
expected wind production (50% of the installed wind capacity for some cases) is only 0.1 m/s, it was decided to consider wind intervals of 1
[3]. m/s for the calculation of the desired histograms. Also note that
0885-8950/$25.00 © 2008 IEEE
VALLÉE et al.: SYSTEM RELIABILITY ASSESSMENT METHOD 1289

Fig. 1. Belgian wind speeds repartition [6].

the wind data are given on an hourly basis, which implies a total
number of 8760 cases over one year; the belonging probability
to interval being then defined as

(1)

where is the annual number2 of wind speeds comprised be-

tween wind speeds and .
Once the three desired probabilistic histograms are estab-
lished, each of them can then be approached by a classical
Weibull law [7], [8]

(2)

(3)

with the scale parameter (information on the mean wind

speed), the shape parameter (distribution of the wind speed
around the mean value), the wind speed, the
density of probability and the cumulative distribu-
tion function.
Finally, the obtained wind profiles and statistical Weibull laws
are presented for each of the three considered sites in Fig. 2.
The calculated Weibull parameters are summarized in Table I
for each of the three sites under scope and clearly confirm the
Belgian wind distribution depicted in Fig. 1. Indeed, the hourly
mean speed calculated for the Ostende site well draws near the
expected values alongside the North Sea (7–8 m/s, cf. Fig. 1)
while the obtained speeds for Uccle and Saint-Hubert confirm
the observed tendencies respectively for the Centre ( 5 m/s)
and the South of the country ( 4 m/s).
Based on the previous observations, the choice of the Weibull
law to be associated with a wind park will thus depend on its ge-
ographical situation. For example, the wind speed distribution
of a site located in the South of Belgium will be calculated by
allowing the Saint-Hubert Weibull parameters to that site. Iden- Fig. 2. Probabilistic histograms and calculated Weibull laws for: (a) Uccle, (b)
Saint-Hubert, and (c) Ostende.
tically, the calculated parameters for Ostende and Uccle will be
2The data given by the Royal Institute of Meteorology are based on one typ-
ical year. By definition, a typical year is a year that has never existed but that
correctly fit the observed weather trends for a larger period of time (the 1990s affected to wind parks respectively established in the North and
for our case). the Centre of the country.
1290 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 3, AUGUST 2008

TABLE I
CALCULATED WEIBULL PARAMETERS FOR THREE BELGIAN SITES

TABLE II
BELGIAN WIND PARK IN 2004 + C-POWER PROJECT

Fig. 3. P-W characteristic established for 1.5-MW doubly-fed (continuous) and

1.5-MW fixed speed asynchronous wind generators (dotted).

C. Belgian Wind Power Distribution

To establish the total wind power distribution, a production
B. Considered Belgian Wind Park histogram is firstly drawn for each of the ten considered wind
sites (cf. Table II).
For the need of the present study, the Belgian wind park ex- Here, in a first approximation, seven states are considered to
isting at the end of the year 2004 is considered and is summa- represent the wind turbines in our classical probabilistic relia-
rized in Table II [9]. At that time, the installed wind power ca- bility computations. So, the operating wind speed range [0, 20
pacity reached 46.6 MW distributed all over the country. Note m/s]3 is divided into seven equal intervals of 2.85 m/s; the proba-
that, in order to introduce a significant amount of wind pro- bility associated to a power interval being
duction in our Monte Carlo simulation (compared to the nearly obtained by the use of the following:
15000 MW of Belgian electrical production park), the actual
“being built” 200 MW C-Power off-shore park was also con-
sidered in our calculations. Moreover, as the Royal Institute (4)
of Belgian Meteorology does not dispose of any measurement
point in the North Sea, the Weibull parameters given in [10] were where is the th wind speed and the Weibull cumulative
taken for the added off-shore site ( and ). distribution function associated to the considered wind park (see
Concerning the operation of the wind turbines, mainly fixed Fig. 4 for the different Belgian cumulative distribution func-
speed squirrel cage and variable speed doubly-fed asynchronous tions).
structures can be found in Belgium; the actual trend being 60% However, the validity of (4) can be limited depending on
of installed doubly-fed wind generators for only 40% of fixed the considered P-W characteristic. Indeed, for a variable speed
speed ones [11]. So, in order to calculate the power distribution turbine, is always superior or equal to when
of each wind park presented in Table II, two power-wind char- (cf. Fig. 3) which is not necessarily the case
acteristics are considered and established in Fig. 3. The first for a fixed speed structure, especially after the wind speed
one is based on the speed control characteristic of a classical corresponding to the maximum of the P-W curve (cf. Fig. 3).
doubly-fed wind generator like the one installed in Schelle and In that case, after the “maximal wind speed,” the considered
is made up of the four usual control areas (starting area, max- 2.85 m/s intervals (with at
imal power extraction area, constant speed area and nominal that time) must be divided in smaller ones and the
power limitation area) [12]. Next to that first P-W characteristic, associated belonging probabilities added to the already calcu-
a second one is also established for an asynchronous fixed speed lated interval (with ) in
generator. Note that those two characteristics will be considered which and are entirely contained.
for each of the Belgian wind parks even if the control strategy Examples of power histograms obtained, thanks to the de-
could be slightly different from one doubly-fed structure to the picted method, for a 1.5 MW variable speed (in Schelle) wind
other. However, those control adaptations would not modify in turbine and a 1.5 MW fixed speed one (in Wondelgem) are pre-
any case the philosophy and the method followed to establish sented in Fig. 5.
our equivalent two-state model for wind production. As the two considered sites are geographically subject to the
Finally, Fig. 3 points out the importance to consider two char- same wind speed regime (Centre), it is not surprising to observe
acteristics for reliability studies as it can be seen that the variable 3Wind speed measurements greater than 20 m/s never happen during all “typ-
speed structure starts producing power at a lower cut-in wind ical year” data bases. It is thus sufficient to take that wind speed as an upper
speed compared to the fixed speed case. margin for our calculations.
VALLÉE et al.: SYSTEM RELIABILITY ASSESSMENT METHOD 1291

Fig. 4. Distribution function Uccle (dashed), Ostende (continuous), and Saint- Fig. 6. Global wind power distribution after convolution of each wind park for
Hubert (dotted). totally independent wind sites over the same wind region.

wind region will undergo the same hourly wind speed) and the
best conditions (smoothing of the global hourly wind power due
to total geographical independency between all wind sites) in
terms of fluctuating unconventional production. In that way, by
limiting the reliability evaluation to those opposite cases, the
best and the worst indices related to the introduction of wind
production can be obtained.
In a first time, we develop our convolution product for totally
independent wind sites. In that case, all the wind turbines must
be defined to give

(5)

with the intermediate power of a power interval defined

in the th wind park ( as ten wind parks are con-
Fig. 5. Probabilistic power histograms for 1.5-MW fixed (Wondelgem) and
sidered in the present case), the number of power intervals
variable (Schelle) speed wind turbines. considered for wind park and the total number of calcu-
lated global powers when the iteration is made on each inter-
mediate power of each defined wind park
a higher probability of zero power production for the fixed rotor
speed case compared to the variable speed one as the power
production is allowed on a larger wind speed range with the (6)
variable rotor speed operation.
Once a power histogram has been dedicated to each estab- The probability associated to each calculated global
lished wind park, the next step consists in a convolution of the power is then obtained by multiplying all the probabilities
several production histograms to obtain a global wind power previously established (cf. Fig. 5) for the
distribution for Belgium. Following the geographical correla- intermediate powers constituting all the
tion level between sites located in the same wind region (four considered wind park histograms
regions for Belgium: North, South, Centre, and off-shore), two
cases are, here, investigated, respectively considering entirely
correlated or totally independent wind sites over the same wind (7)
region. Indeed, contrary to [13] where the exact correlation be-
tween wind sites was taken into account (clustering), the advan- Finally, the obtained wind power distribution for totally inde-
tage of restricting the reliability evaluation to both totally op- pendent wind sites is depicted in Fig. 6. This last one has been
posite correlation levels for wind production is to quantify the established via a discrete convolution4 and it is thus not sur-
ability of the electrical system to cover the load when facing prising to observe areas of zero probability outside the allowed
the worst (100% correlated wind sites will involve highly fluc- 4Only a finite number of states is considered for each power histogram related
tuating hourly global wind production as all sites of the same to a wind park.
1292 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 3, AUGUST 2008

wind power combinations. This observation does not matter in but must rather be seen as an “average” estimation of the wind
the present case as only the probability associated to a simulta- generation in Belgium. So, the capacity credit to be associated
neous zero total wind power production will be interesting here. with the calculated wind capacity must not be defined as “the
Indeed, that calculated probability will then be considered as the amount of conventional power plant capacity that can be re-
forced outage rate (FOR) during the establishment procedure of placed by wind power, without decreasing the level of security
a classical two-state equivalent model for the global wind pro- of supply for the power system” but rather considered as “the
duction in Belgium. equivalent wind energy that, if available almost all of the time,
So, to determine our two-state model for wind production, would replace all the wind parks without decreasing the level
the FOR for the independent case will be taken as 0.00141 of security of supply for the power system.” In the goal of de-
following Fig. 6 (probability of zero wind power production) termining this equivalent wind capacity, an equivalent two-state
and the equivalent production capacity will be determined on model is established via two cases treated with the proposed
basis of fixed reliability criteria thanks to a nonsequential Monte Monte Carlo simulation:
Carlo simulation (MCS) [3], [4]. • First case: Introduction of the complete detailed Belgian
In case of entirely correlated wind sites over the same wind wind park production and calculation of several reliability
region, (5)–(7) can be extended to indices in that first configuration;
• Second case: Replacement of the complete detailed wind
park by a two-state equivalent model, which FOR equals
(8) the pre-determined values calculated in Section II-C,
and increase of the equivalent production capacity until
with the intermediate power of a power interval defined reaching the same reliability indexes as the ones estab-
in the th wind region ( with in the lished with the first configuration. That calculated capacity
Belgian case as four wind regions are considered: North, South, may then be considered as an equivalent wind production
Centre and off-shore), the number of power intervals con- power in the peak load covering predictive models.
sidered for the wind parks of region (as all the wind parks Also note that the reliability analysis proposed in this article
of a single region are supposed entirely correlated, they are si- will be based on a well-being approach [4], [14] and limited to
multaneously in the same state) and the total number of hierarchical level I as the electrical network will be considered
calculated global powers when the iteration is made on each in- as always available (outages only drawn on generation units)
termediate power of each defined wind [15].
region
A. Nonsequential Monte Carlo Simulation

(9) References [13], [16], and [17] have shown that reliability
evaluations could efficiently be realized by taking into account
The probability associated to each calculated global an accurate hourly correlation level not only between wind sites
power is again obtained by multiplying all the probabilities but also between load and wind production. However, via the
previously established for the inter- analytical way, only global reliability indices (loss of load ex-
mediate powers constituting all the pectation or expected unserved energy) can be evaluated and
considered wind regions histograms the risk level associated to each problematic situation can not
be separately evaluated. So, to simultaneously quantify the loss
of load situations frequency and the severity of those system
states, a Monte Carlo approach is here considered; even if the
(10) nonsequential behavior of the chosen simulation will not permit
to correlate the hourly wind production and the load.
Using a discrete convolution similar to the one presented for the Practically, Monte Carlo simulations can be used to estimate
independent case, a second FOR is established and represents reliability indices by simulating the actual process and random
thus the probability that all the wind sites do simultaneously not behavior of the considered electrical system. In theory, those
produce any power. The so-obtained value reaches 0.0132 and simulations can include system effects which may not be pos-
is greater than the one calculated in the independent case. This sible without excessive approximation in a direct analytical ap-
observation is quite logical as the wind sites are here supposed proach and can generate a wide range of indices within a single
entirely correlated over the same wind region, what involves a study [4]. In fact, there are two basic techniques used when
pronounced “all or nothing” behavior in the present case. Monte Carlo methods are applied to power system reliability
evaluation, these methods being known as the sequential and
nonsequential techniques [3], [4].
III. DETERMINATION OF THE EQUIVALENT
In the present study, a nonsequential Monte Carlo method
WIND PRODUCTION CAPACITY
has been developed under Matlab to evaluate the reliability
In the way, we define it here, the calculated wind capacities indices of interest. This Monte Carlo simulation theoretically
are not equivalent to one particular technology (as, for example, could incorporate any number of system parameters and states
nuclear units have a different FOR from the one calculated for but it has been assumed, in our calculations, that a generation
thermal units; leading, thus, given the associated classical tech- unit was only able to reside in one of the following two states:
nology, to two different calculated equivalent wind capacities) fully available and unavailable. Moreover, in the established
VALLÉE et al.: SYSTEM RELIABILITY ASSESSMENT METHOD 1293

nonsequential simulation, only hourly uncorrelated states where , and are, respectively, the total number
are considered as it is supposed that a generation unit outage of hourly healthy, marginal and risky simulated states; being
state does not condition or influence its state during the next the total amount of simulated years.
or previous hours of simulation (and inversely). Finally, at the Note that a nonnegligible advantage of Monte Carlo methods
start of each hour, a uniformly distributed random number can be found in the fact that they only require to simulate a lim-
on the interval [0, 1] is drawn for each generation unit in order ited amount of system states before having a convergence on the
to decide its operation state, using the following procedure: calculated indices. In that way, if, in [4], a solution was found
• If , the unit is decided to be unavailable; after 734 simulated years (6 429 840 hours), a convergence is
• If , the unit is decided to be fully available. here established after 228 simulated years (1 997 280 hours) for
Finally, note that, even if sequential techniques present the our system (Table V); the difference between these two simu-
advantage of allowing the possibility to take into account the lation lengths coming from a higher system complexity (much
temporal correlation between hours of peak load and wind pro- more constitutive elements) in the case of [4] and from a con-
duction, nonsequential simulations can also lead to interesting vergence tolerance for the reliability indices fixed, here, to
results if men are not focussed on hourly transitions but rather in order to quantify the impact of very low wind penetrations.
on a larger time-scale horizon. Indeed, in this last case, it can
be considered that calculated reliability indices and, thus, es- C. Belgian Production Park and Load Consumption
tablished wind equivalent capacities are based on a large set of Beside the introduced wind park (Table II), the other produc-
possible correlation states between wind production and load tion units considered in this study are the ones that were op-
(even if those situations are not sequentially simulated). erating in Belgium by the year 20006 [18]. At that time, the
installed capacity was reaching 14 758 MW distributed among
nuclear production (Table III(a)), classical thermal units ((b)),
B. Well-Being Analysis gas-vapor turbines ((c)), hydraulic production ((d)), diesel cen-
trals ((e)), cogeneration ((f)) and, finally, 67 MW of energy re-
In the Monte Carlo simulation, the power system is thus mod-
generation.
eled by specifying a set of “events,” where an event is a random
Finally, next to this production park, the base and peak values
occurrence that changes the “state” of the system [4]. In the
considered to represent the load evolution are extracted from
present study, the events recognized by the established program
[6] and model the 2004 Belgian electrical consumption (cf.
are the changes in load5 and the possible failure of a generating
Table IV).
unit. Each simulated system state is then defined in terms of
available margin, which is the difference between the available
D. Simulations and Wind Equivalent Capacity: Belgian Case
production capacity and the load. Indeed, in order to proceed to
a well-being analysis, the total available system capacity must To determine the equivalent production capacity related to the
be superimposed on the load during each simulated hour to de- Belgian wind park, the two cases described in Section III have
fine [4], [14]: thus been simulated. In a first time, the implemented nonsequen-
• Healthy state: The total available capacity minus the tial Monte Carlo simulation has been launched for the entire
largest production unit remains greater than the corre- Belgian production park with and without explicitly defining
sponding hourly load; all the existing wind turbines by the year 2004. The simulation
• Marginal state: The total available capacity is greater than results obtained in that first case are summarized in Table V and
the corresponding hourly load but becomes less than that also compared with the indices analytically determined
same load value when the capacity of the largest unit is via the simulation tool developed in [18]. Note that, in Table V,
set back of the available production park. In this study, to the wind sites are supposed totally independent over the same
point out the effects of the expected nuclear stopping in wind region (one drawn wind speed for each park).
Belgium, the marginal state indices have been calculated, The first observation to be drawn from those calculated
based on the loss of the greater nuclear unit (Tihange; see well-being indices is derived from the fact that the addition of
Table III(a); wind production decreases the risk of load nonrecovery (from
• At risk state: The total available capacity is directly less without wind production to
than the corresponding hourly load value. with that same renewable energy), which is rather evident as
Three well-being indices are then defined as [4] the total energy production is increased while the load stays
unchanged for both investigated cases.
Moreover, the reliability indices analytically calculated
thanks to the tool developed in [18] are greater than the ones
(11) obtained with the established nonsequential Monte Carlo
simulation (from with simulation to
with the analytical tool developed in [18]).
(12)
This observation can be explained by the fact that every unit of
a production park is taken into account in the simulation
(13) while the analytical approach considers all the production parks
5The load is determined as a base value (required 100% of the time) to which 6As the Belgian production park has not undergone significant changes since
a random normal distributed contribution, comprised between the base and peak 2000, it is reasonable to mix this entire classical production park with the 2004
load values, is added. wind park for the desired calculations.
1294 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 3, AUGUST 2008

TABLE III TABLE III CONT’D

(a) NUCLEAR PARK CONSIDERED IN THE MONTE CARLO SIMULATION [18]. (b) (d) DESCRIPTION OF THE CONSIDERED HYDRAULIC PARK [18]. (e) DIESEL
CLASSICAL THERMAL PARK IN THE MONTE CARLO SIMULATION [18]. (c) PARK CONSIDERED IN THE MONTE CARLO SIMULATION [18]. (f) DESCRIPTION
GAS-VAPOUR TURBINES FOR THE MONTE CARLO SIMULATION [18] OF THE COGENERATION PARK CONSIDERED IN THE MONTE CARLO SIMULATION
[18]

(a) (d)

(e)

(b)

(f)

TABLE IV
PROPERTIES OF THE LOAD EVOLUTION CONSIDERED IN THIS PAPER [18]

(c)
Forced outage rates considered here are the ones given in [18] and [19].

TABLE V
as single stand alone big units with the same as the one MCS AND ANALYTICAL SIMULATION RESULTS WITH AND WITHOUT WIND
associated to a smaller constitutive production unit. However,
it is quite logical to think that the simultaneous nonproduction
probability of an entire production park will be lower than the
one related to a single constitutive unit and, therefore, from
the reliability point of view, not surprising to measure over-
estimated indices with the analytical approach (as each kind
of production park is there considered as a single stand alone
big unit). Finally, the margin probabilities calculated in
Table V also emphasize on the fact that sufficient alternatives
to the nuclear dismantling must absolutely be found in order
to avoid not only frequent load nonrecovery situations but also is increased until reaching the same reliability indices7 as those
critical power fluxes at the boundaries. obtained earlier with the complete wind park. The so-calculated
The well-being indices of the Belgian production park estab- equivalent capacity is evaluated to be 50 MW in the present case
lished in Table V with a classical complete definition of each and represents thus 20.28% of the total installed wind capacity
wind generator will now be used to determine the equivalent (cf. Table VI).
two-state wind capacity. Indeed, in order to replace the com- By definition, if talking about reliability, the calculated equiv-
plete wind park, the equivalent two-state wind model depicted alent wind capacities will thus have the same behavior as the
in Fig. 7 for independent wind parks over the same wind region total wind park and can therefore be considered as an expected
(with ) is now introduced in the Monte Carlo 7Where [18] only determines an equivalent wind production only on basis of
Simulation and the production capacity of that last model the same LOLP value.
VALLÉE et al.: SYSTEM RELIABILITY ASSESSMENT METHOD 1295

TABLE VII
EQUIVALENT WIND PRODUCTION FOR DIFFERENT INSTALLED
WIND CAPACITIES. CASE OF TOTALLY INDEPENDENT
WIND SITES IN THE SAME WIND REGION

Fig. 7. Equivalent two-state wind model for independent wind sites over the
same wind region.
TABLE VIII
EQUIVALENT WIND CAPACITIES CALCULATED WITH
TABLE VI THE ANALYTICAL TOOL DEVELOPED IN [18]
RELIABILITY INDICES WITH THE TWO-STATE WIND EQUIVALENT MODEL
(INDEPENDENT WIND SITES OVER THE SAME WIND REGION)

our equivalent two-state model as the one calculated with 200

MW offshore wind energy (Section II-C).
wind production in the predictive management of the load and The equivalent capacities calculated in Table VII stay approx-
the derived power allocation of nuclear and big thermal units. imately constant around 20% of the total installed capacity. That
Indeed, by taking into account an accurate zero wind production observation is quite logical and tends to validate the imagined
probability and a complete definition of the production park, method of equivalent wind capacity calculation. Indeed, as all
the calculated equivalent wind model should globally permit to the added wind capacity was located at the same offshore site
limit inopportune closing down of classical big thermal or nu- (Thornton bank) for our calculations, it had no impact neither
clear units, which could appear if the wind penetration continues on the zero wind production probabilities calculated earlier with
to increase in the next future and that this kind of fluctuating en- lower offshore wind capacities (as the total number of wind sites
ergy is still ignored in the predictive management of the week is maintained unchanged) nor on the global probabilistic be-
peak load. Also note, in Table VI, that the reliability indices ob- havior of the installed Belgian wind park (as, for the three study
tained with a capacity taken in the two-state equivalent model cases, the offshore installed capacity is far away bigger than the
as equal to the total wind installed production (246.6 MW) are on-shore one which is constantly maintained at 46.6 MW); the
characterized by lower values of and and con- calculated equivalent capacities representing thus essentially the
firm thus that the wind can not be considered as always entirely behavior of the offshore park.
available in predictive models. Those calculated wind equivalent capacities have then been
To evaluate the impact of the considered correlation level (be- compared to the ones obtained thanks to the analytical tool de-
tween wind parks) on the calculated equivalent capacities, the veloped in [18] (see Table VIII). Even though there are sev-
same simulations as the ones depicted in Tables V and VI have eral assumptions made in [18] (only two single big offshore and
been launched, taking this time into account the (0.0132) on-shore parks are considered and only one Weibull law is taken
established when wind sites are supposed to be entirely corre- for the entire on-shore park), those calculated values are close
lated over the same wind region. We so obtain an equivalent to the ones obtained in Table VII and comes nearer as the in-
capacity of 51.5 MW, which is very close to the one calculated troduced offshore capacity is increased. That result can be ex-
in Table VI for the independent case. Those results are easily plained by the fact that, in our simulation like in [18], all the
explained by the very low calculated values of whatever offshore capacity is supposed to be introduced at the same ge-
the geographical correlation level is. ographical point involving quasi identical calculated equivalent
Another interesting result is also proposed in Table VII where capacities when the offshore installed wind power is far away
the offshore capacity has been successively increased from 100 bigger than the on-shore one. Note that, if the installed wind ca-
to 800 MW. The objective of that study was to evaluate the im- pacity had been increased by geographically spreading the wind
pact of wind power repartition on the installed capacity per- parks (involving, thus, different wind regimes for each added
centage that has to be introduced in the equivalent two-state wind park), it would have normally led to a reduction of the
model in order to fit the reliability indices calculated with the percentage (of the installed wind capacity) represented by the
entire wind park. To simplify the proposed study, all the offshore calculated wind equivalent capacities as wind power capacity
production is supposed to be entirely correlated in the North Sea credit normally decreases with the installed wind power level
and, thus, to be simultaneously subject to the same wind speed. [20] (being known that we do not consider, here, a wind ca-
Following that realistic assumption, it is therefore convenient to pacity equivalent to one particular technology but rather to an
consider, for each offshore wind power level, the same in “average” wind generation capacity for Belgium).
1296 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 3, AUGUST 2008

TABLE IX equivalent model. Then, the equivalent capacities were calcu-

RELIABILITY INDICES AND EQUIVALENT CAPACITY WITH THREE ADDED lated on basis of a “well-being” analysis worked out via a non-
500-MW WIND TURBINES. CASE OF TOTALLY INDEPENDENT WIND SITES sequential Monte Carlo simulation. In that way, given the very
IN THE SAME WIND REGION
low values of , it could be emphasized on the very close
calculated equivalent capacities, whatever the considered geo-
graphical correlation level was.

ACKNOWLEDGMENT
The authors would like to thank M. Stubbe, K. Karoui, and
Finally, a further last study consisted in the introduction of S. Rapoport from Tractebel Eengineering (Suez) for the support
three additional onshore 500 MW wind parks (each of them given to this work.
being located in the North, in the Centre and in the South of
the country) in order to quantify how the geographical repar- REFERENCES
tition of wind installations affects the percentage of equivalent [1] B. C. Ummels, M. Gibescu, E. Pelgrum, W. L. Kling, and A. J. Brand,
wind production capacity. In Table IX, the equivalent capacity “Impacts of wind power on thermal generation unit commitment and
dispatch,” IEEE Trans. Energy Convers., vol. 22, no. 1, pp. 44–51, Mar.
(with a new value of 0.0000159 determined, with the three 2007.
additional wind sites, thanks to the convolution procedure de- [2] H. Lund, “Large-scale integration of wind power into different energy
systems,” Energy, vol. 30, pp. 2402–2412, 2005.
scribed, for totally independent sites over the same wind re- [3] R. Billinton and R. Karki, “Application of Monte Carlo simulation to
gion, in Section II-C) calculated with the introduction of 1.5 generating system well-being analysis,” IEEE Trans. Power Syst., vol.
GW additional wind production (for a total installed capacity 14, no. 3, pp. 1172–1177, Aug. 1999.
[4] R. Billinton and A. Sankarakrishnan, “Adequacy assessment of com-
of 1.746 GW) reaches 14.8% of the installed wind capacity. posite power systems with HVDC links using Monte Carlo simulation,”
From this last observation, it can thus be concluded that the IEEE Trans. Power Syst., vol. 9, no. 3, pp. 1626–1633, Aug. 1994.
[5] R. Karki and P. Hu, “Wind power simulation model for reliability eval-
geographical spreading of wind turbines not only tends to de- uation,” in Proc. IEEE CCECE/CCGEI, Saskatoon, SK, Canada, May
crease the probability of zero wind power production (which is 2005, pp. 541–544.
quite logical given the fact that the probability to have simulta- [6] H. Buyse, “Production d’énergie électrique,” Documentation Elec-
trabel, Univ. Catholique de Louvain, 2004.
neously weak or, on the contrary, very strong winds everywhere [7] B. Raison, M. Crappe, and J. Trécat, “Effets de la production d’élec-
is decreased as the wind sites are geographically spread), but tricité d’origine éolienne sur les indices de fiabilité des réseaux élec-
triques,” in Proc. Colloque Electrotechnique du Futur (EF’2001), Nov.
also slightly reduces the equivalent percentage of wind capacity 2001.
compared to Table VII. That decrease in the computed equiv- [8] J. G. Slootweg, “Wind power modelling and impact on power systems
dynamics,” Ph.D. dissertation, Tech. Univ. Delft, Delft, The Nether-
alent percentages is here explained by the modification of the lands, Dec. 2003.
wind power repartition between the four considered geograph- [9] Electrabel, “Parcs éoliens belges en MW (fin 2004),” Documentation
ical regions (the onshore installed capacity is now higher than Electrabel, 2005.
[10] Belgian Public Planning Service Science Policy, Optimal Offshore
the offshore one thanks to the addition of the three supplemen- Wind Development in Belgium, pt. 1, CP-21, 2003, pp. 19–20.
tary 0.5 GW onshore wind parks). Indeed, by the lower wind [11] B. Robyns, “Vers une meilleure intégration de l’éolien dans le réseau
électrique grâce à l’électronique de puissance,” GREPES Study Day,
speeds blowing in onshore sites, it is obvious that the wind pro- Mar. 2006.
duction availability will be less at those onshore sites (compared [12] S. Al Aimani, “Modélisation de différentes technologies d’éoliennes
to offshore parks) involving thus a decrease in the global wind intégrées à un réseau de distribution moyenne tension,” Thèse de
l’Ecole Centrale de Lille, 2004, ch. 2, pp. 24–25.
availability when the onshore penetration is higher than the off- [13] C. Singh and Y. Kim, “An efficient technique for reliability analysis of
shore one. In conclusion to that last observation, the geograph- power systems including time dependent sources,” IEEE Trans. Power
Syst., vol. 3, no. 3, pp. 1090–1096, Aug. 1988.
ical spreading of wind turbines does thus not necessarily im- [14] R. Karki and R. Billinton, “Cost-effective wind energy utilization for
prove the expected equivalent wind capacity even if the global reliable power supply,” IEEE Trans. Energy Convers., vol. 19, no. 2,
wind power profile will be, in each case, smoothed. pp. 435–440, Jun. 2004.
[15] R. Allan and R. Billinton, “Probabilistic assessment of power systems,”
Proc. IEEE, vol. 88, no. 2, pp. 140–162, Feb. 2000.
[16] C. Singh and A. L. Gonzalez, “Reliability modelling of generation sys-
IV. CONCLUSION tems including unconventional energy sources,” IEEE Trans. Power
App. Syst., vol. PAS-104, no. 5, pp. 1049–1056, May 1985.
In this paper, a new method for equivalent wind capacity [17] S. Fockens, J. M. Van Wijk, W. C. Turkenburg, and C. Singh, “Reli-
ability analysis of generating systems including intermittent sources,”
calculations has been introduced and is based on Monte Carlo Int. J. Elect. Power Energy Syst., vol. 14, no. 1, pp. 2–8, 1992.
simulations. The objective followed by the introduction of the [18] B. Raison, M. Crappe, and J. Trécat, “Effets de la production décentral-
isée dans les réseaux électriques,” Projet “Connaissance des émissions
equivalent two-state wind models was to determine wind equiv- de CO2,” Sous-projet 5, Faculté Polytechnique de Mons, 2001.
alent capacities (from the supply reliability point of view) in [19] R. Billinton, “A reliability test system for educational purposes—Basic
order to introduce them in load covering predictive models. To data,” IEEE Trans. Power Syst., vol. 4, no. 3, pp. 1238–1244, Aug.
1989.
reach those equivalent wind capacities, a global annual wind [20] R. Billinton, H. Chen, and R. Ghajar, “A sequential simulation tech-
power distribution was firstly established by convolution of the nique for adequacy evaluation of generating systems including wind
energy,” IEEE Trans. Energy Convers., vol. 11, no. 4, pp. 728–734,
constitutive wind parks power histograms (Belgian park). At Dec. 1996.
that step, two cases were investigated respectively considering
totally independent and entirely correlated wind sites aver the François Vallée was born in 1980. He received the civil electrical engineering
same geographical region. The so-obtained probability of zero degree from Faculté Polytechnique de Mons, Monds, Belgium, in 2003.
He is currently an assistant at the Department of Electrical Engineering of
wind production (all the wind turbines simultaneously closed Faculté Polytechnique de Mons. His fields of interest include wind generation
down) were calculated and considered as in the two-state modeling and electrical network reliability in presence of dispersed generation.
VALLÉE et al.: SYSTEM RELIABILITY ASSESSMENT METHOD 1297

Jacques Lobry was born in 1964. He received the electrical engineering and Olivier Deblecker was born in 1971. He received the civil electrical engineering
nuclear engineering degrees and Ph.D. degree in electrical engineering from degree and the Ph.D. degree in electrical engineering from Faculté Polytech-
Faculté Polytechnique de Mons, Mons, Belgium, in 1987, 1990, and 1993, re- nique de Mons, Mons, Belgium, in 1995 and 2001, respectively.
spectively. He is currently an Assistant Professor in the Electrical Engineering Divi-
In 1987, he joined the Department of Power Electrical Engineering of Faculté sion of Faculté Polytechnique de Mons. His fields of interest include magnetic
Polytechnique de Mons, where he is currently a Professor in power systems en- fields calculation using numerical methods, hysteresis modeling, and power
gineering. In 2003, he began the head of this department. His research interests electronics.
mainly include numerical methods in computational electromagnetics. Dr. Deblecker has been a member of the Belgian Royal Society of Electrical
Engineers (SRBE/KBVE) since 1996 and the European Power Electronics and
Drives (EPE) Association since 2004.

View publication stats