Reliable Automatic Decision Making for Washing the Polluted

Insulators in Coastal Substations

Satoru G O T O t , Masatoshi NAKAMURAt , Nishantha NANAYAKKARAt and Takashi TANIGUCHIX
tDepartment of Electrical Engineering, Saga University, Honjomachi, Saga 840, .Japan
$Saga Branch, Kyushu Electric Power Co., Inc., 2-3-6, Konohigashi, Saga 840, Japan

Abstract deposits and weather information. T h e substation

Insulators in coastal substations are polluted by engineers make decision t o wash the insulators based
wind from seashore with much content of salt. T h e on the prediction of the pollution deposits accord-
insulator pollution causes flash over of the substa- ing t o the weather information and their experience.
tions equipments and results power failures. Hence, Hence, in this paper, mathematical model of the pol-
the polluted insulators are washed according t o the lution deposits of substation insulators is constructed
decision making of engineers in substations. and the pollution deposits are predicted by using the
This study is aimed t o develop a method of reli- mathematical model. T h e automatic decision of in-
able automatic decision making for washing the pol- sulator washing is made according to the predicted
luted insulators in substations near coastal areas. pollution.
T h e proposed method consists of the modeling of 2. Problem Statement
the pollution deposits and the discriminant analysis
by using the multi-dimensional normal distribution. 2.1 Insulator Pollution in Coastal Substa-
Effectiveness of the proposed method is assured by tions
applying it t o the actual d a t a of substations. Insulators of substations near the coastal areas are
subjected t o rapid pollution because the salt parti-
1. Introduction cles are brought t o the island by the wind from sea,
Since Japan is a country of islands and is sur- and attach t o the insulators. Strong wind velocity
rounded by the sea, most of the substations are sit- causes the rapid pollution of the insulators. I t has
uated near coastal areas and the insulators in these been experienced that the pollution deposit comes t o
substations are frequently subjected t o pollution by hazardous levels within two or three hours, especially
the wind with much content of salt. Insulator pollu- a t the times of high winds and typhoons. Wind di-
tion is a big problem which causes flash overs of the rection is also related to the pollution beca.use the
substation equipments and results long power fail- wind from the seashore contains much salt. T h e rain
ures. Not only in Japan, this occurs in all coastal has an effect for the natural washing of the insulators
substations in the world. T h e polluted insulators are and, in turn decreases of the pollution.
washed by pure water before dielectric breakdown 2.2 Right State of Decision Making of Insu-
occurs, following the decision made by engineers lator Washing
in the substations. T h e insulator washing requires Presently, the insulators in substations are fre-
much m a n power and pure water, hence, the inef- quently washed for safety operation. When the pol-
fective insulator washing must be avoided. On the lution deposit reaches its hazardous level, tjhe substa-
other hand, the late decisions might lead t o big dam- tion has t o be shut down and insulators to be wa.shed
ages to the substations. T h e problem of the insulator because hot-line wa.shing m a y ignite flash over. Such
pollution is investigated[l, 2, 3, 41 and the modeling situation causes the long power failures a.nd it must,
of the insulator pollution has been proposed[5, 6, 71. be avoided. On the other hand, the number of wash-
In this study, a method of automatic decision mak- ing times are necessarily reduced t o lower the over
ing of washing the polluted insulators in substations all cost of the insulator washing, and the accidents in
is developed. We have already proposed the au- the substations which sometimes ocsur during wash-
tomatic decision making of insulator washing[El, 91. ing. These lead t o a requirement of an appropriate
T h e previous proposed methods were based on the decision malting.
discriminant analysis whose factors were pollution Frequent measurements of pollution deposits are
408
extremely important and depending on the measure- negative pollution caused by rainfall and the nega.tive
ments, substa.t,ionengineers predict the insulator pol- pollution deposit is described as
lution. When the predicted pollution becomes the
hazardous level, the engineers ma.lte decision of insu-
lator washing.
3. Automatic Decision Making of Insulator
p - ( l ~ )= exp
(II i
- w , . ~ ( i ) p~( k - 1)

where n,. is measuring times of rainfall from IC - 1

to IC, p ( k - 1) is the pollution deposit at IC - 1 and
(2)

Washing through Mathematical Modeling of w,. is the coefficient of the negative pollution. If the
Pollution Deposits rainfall is O[mm], then p - ( k ) = p ( k - 1) because of
3.1 Data Acquisition no natural washing. T h e pollution deposit p ( k ) is
described as
Actual d a t a of the pollution deposit, the decision
malting of t,he washing and t,he weather conditions +
P ( k ) = P + ( k ) P- (IC) + e ( k ) (3)
were acquired during the period of September 1st where e ( k ) is modeling error at, k .
1993 to October 31st 1993 at Karatsu substation of 3.3 Automatic Decision Making of Insulator
Saga prefecture in Japan. Karatsu substation is just, Washing
one kilometer range from the sea and it, is one of the The discriminant analysis is adopted to the auto-
badly hit substations from salty winds. T h e records matic decision making of the insulator washing[9].
of pollution deposits were taken from the instrument Probability of the insulator washing is used for
called 'salt meter' which had a pilot insulator and the discriminant function of the decision making.
a special chamber t o measure the leakage current T h e probability is determined by use of the multi-
of the insulators at a constant humidity and damp dimensional normal distribution.
condition. T h e available d a t a are, average wind ve- Two groups for discrimination are set as G A and
locity [m/s], wind direction (16 directions), accumu- Gg and the factors are z = ( X I ,z2, . . . , z ~ ) The~ .
lated rainfa.11 [mm/h], mean temperature ["C] and group Gi ( i = A , B ) are cha.racterized by generation
humidity [%I in hourly basis, and poliution depositss rate pi and probability density funct,ion fi(z) whose
[mg/cm2] c.ollect,ed every 2 hours for day and every 3 population mean is pi and popuhtion variance ma.-
hours for night a.t normal winds a.nd every 1 hour at trix is &. T h e probability ~ A ( z ) where the sample
the time of strong winds. Figure 1 shows the pollu- z belongs to G A ,is in proportional t o ~ A ~ A ( z then
) ,
tion deposit) (bar), decision making (+), rainfall and the probability ~ A ( z ) is calculated as
wind velocit,y during t8heperiod of September 1993
at Karatsu substa.tion.
3.2 Mathematical Modeling of Pollution De- T h e discriminant, func.tion is defined by the proba.-
posits bilit,y ~ A ( z ) , then t,he decision is made by
The pollut,ioii deposit p ( k ) was modeled as the E (;A, g A ( z ) >
addition of t,wo sepa.ra.tely identified components of
positive pollution deposit p+( k ) and nega.tive pollu-
{ :E E G B , otherwise (5)
where X is a. threshold va.lue. The multi-dimensional
tjion deposit, p - ( k ) [ 6 , 71. As the pollution at, rainy norma.1 dist,ribut,ion is a.dopted for t,he proba.bility
conditions a.re a combina.t,ion of t,he negative effect, densit,y function fi(z) as
due to ra.infal1 and t8hepositrive effect caused by wind,
it, wa.s understood that t8he phenomenon of positive
pollut,ion has t,o be fully identified first,, and then the T
(z- p i ) , i = A , B .
-1
dat,a. with the effect of ra.infal1 should be dealt, with. Qi (z- p i )
T h e positjive polliit,ion deposit, is proportional to In calculation of (4), the generation ra.t,epi, t,he pop-
t,he summation of the t8hird power of wind velocity ulat,ion mean pi and the population variance matrix
as Ci a.re estima.t,ed by tshese sa.mple values. Then, the
a.ut,oma.tic decision is made according t o (5).

?=I 4. Application to Actual Data

where . u ( i ) is wind velocitsy from sea. at, t,ime i , U,, is T h e proposed metshod is applied to the actua.1 da.ta
mea.suring times of wind velocit,y from k - 1 to k , a.t. I<a.ratsu substat,ion during the period of Sept,em-
T is measurement, int,erva.l of the wea.ther conditions ber 1st 1993 to October 31st, 1993. T h e measurement,
a.nd 'w,, is the coefficient,of t,he positfive pollution[7]. int,erval of weather conditions T wa.s 1 hour and to
T h e ra.te of tlhe na.tura.1wa.shing is int,roduced for the est,imat,e the coefficients tu,, and 7 4 , t2he sampling
409
 Karatsu: September, 1993
N! 0.02
p
0
0.015
v
0.01
8
0.005
0

- 20
2.$ 15
-52
0 10
5
g o 911 5 10 20 25 30
l5 [date]

Fig. 1. Pollution deposit (bar), decision making (+), rainfall and wind velocity during the period of Septem-
ber 1993 at Karatsu substation of Saga prefecture in Japan.

time of the measurements of the pollution deposits ahead prediction of the pollution deposits. T h e n ,
was set a.s 1 day because the measurement interval for the decision making d a t a (19 times), the mean
of 2 or 3 hours was too short to observe the changes was 4 . 6 1 6 9 ~ 10-3[mg/cm2] and the sta.nda.rd devi-
of the pollution deposits especially at normal con- ation was 2.2287x10-'[mg/cm2], for the no de-
ditions. T h e pollution deposits at 23 o'clock were cision making d a t a (598 t,imes), the mean was
used a.s the day's mea.surement, and the wind veloc- 2 . 6 6 1 2 ~10-3[mg/cm2] a.nd t8he standard deviation
it8ies were c.onsidered only from seashore (upper di- was 1 . 3 0 2 8 10-3[mg/c~n2]l.
~ T h e results shows t h a t
rections from NW t o SE). First, we extract the d a t a the decision makings were done when the predic-
of no decision making (44 days) a n d the coefficient tions of the pollution deposits were rat,her high
,wTjwa.s calculated by using the d a t a of no rainfall compared with those of no decision making. T h e
(31days). T h e n , the coefficient w,.was calculated by mean and the standa.rd deviation of the pollution de-
using the d a t a of rainfall. To estimate the coeffi- posits of the decision making time were calculated as
cient 'IU,., the d a t a of which accumulated rainfall of a 4 . 2 6 3 2 ~10-'[mg/cm2] and 2 . 1 1 7 2 ~ 1 0 - ~ [ m g / c m ~ ] ,
c1a.y is great,er t,han l[mm] are used (13 days) because respectively. As a. result,, the decision rna.ltings were
t,here are few effect for natural washing when the ac- done when the pollution deposits became increa.sed
cumulated rainfall in a day is I["]. T h e estimated and the dangerous situa.tions were predictmed.
coefficients were w, = 1.2649 and TU,. = 0.2077. Finally, automatic decision making wa.s ca.rried
To assure the effectiveness of the proposed model, out by use of the proposed method. T h e fa.ctJorfor
t8he prediction of the pollution deposits were car- the automatic decision malting in (5) was z ( k ) =
ried out at every measurement d a t a of the pollu- (jj(k+2),exp(-1/1--141)/14). Here, @ ( k + 2 ) is the 2
t,ion deposits. To predict the pollution deposits, hours ahead prediction of t,he pollution deposits by
we a.ssumed that, the weather condition did not using the model (3) where the wind velocity, wind
change until next measurement time. We com- direction and rainfall were assumed t o be kept as 2
pared the proposed model with another simple model hours and the factor exp(-lk - 141)/14) wa.s intro-
whose prediction kept the pollution deposit until duced because almost a.11 decision were ma.& at 14
next measurement. T h e root mean squares error o'clock (13 times), 3 times of decision malting were
were 1 . 8 0 l0-'[mg/cm2]
~ for the proposed model a.t 17 o'clock and the last times of 3 decision mak-
and 1 . 9 3 1O-'[mg/cm2]
~ for the simple predictmion ings were 0, 7 a.nd 21 o'clock. T h e proba.bility of the
method a.nd the results shows the effectiveness of the decision making g A ( z ) was calcuhted by using (4).
proposed model. Figure 2 shows the proba.bility and the predictions
Next,, t,he decision making d a t a and no deci- of the pollution deposits. T h e result, of the a . u t o
sion making data. were compared by the 2 hours inatic decision malting (5) when the threshold value

410
 11 + +t+ ' + + ++ + + t + + + + + Japanese).
Electric Joint Research, vol. 35, no. 3 (1979) (in
Japanese).
Y. Higashimori: Insuhtor Contamination due
to Typhoons and Shehering Effects of Topo-
graphical Environments, Trans. of IEE Japan,
vol. 110-B, no. 11 (1990) (in Japanese).
Y. Higashimori: Rapid Salt Contamination of
3 ' dog tjhe Insulator Surfa.ce by Typhoons, Trans. of
'0 100 200 300 400, 500 600 IEE Japan, vol. 113-B, no. 2 (1993) (in
Number of measurement [tunes]
.Japanese).
Fig. 2. Probability of the decision malting, actual N . Nanaya.kkara., M. Nakamura, S. Goto and
decision making (+) and the 2 ahead prediction of T. Tmiguchi: Modeling and Prediction of Rapid
the pollution deposits at Karatsu substation Pollution of Insulators in Substations Ba.sed on
Weather Information, Proceedings of the 1994
Korean Automatic. Control Conference (1994).
T. Sensyu, et al: Analysis of the Relation be-
tween Salt Pollution for Insulators and trhe Me-
teorological Conditions, Power Research Insti-
tute Report, no. 73020 (1973) (in .Japanese).
Total Wash Det. Corr. Fault Miss T. Sensyu, et al: On the Estimation of Max-
617 19 25 10 9 15 imum Density of Sea Salt Deposition on the
52.6% 47.4% 2.4% Insulators, Power Research Institut,e Report,
no. 73056 (1974) (in Japanese).
S. Gotso, M. Na.ltamura., N . Nanayakkara.,
X = 0.165 is shown in Table 1. T h e percentage of the T . Mat,sunaga. and T. Ta.niguchi: Automatic
correct decision making (52.6 %) is not, so high and Decision Making of Washing the Ra.pid Pol-
it caused by the safety side of the human decision lutjed Insulators in Substations, Proceedings of
ma.king. T h e pollution deposits of the 9 fault detec- t,he 26th ISCIE Intjernationa.l Symposium on
tion data. are equal or less than 0.0031[mg/cm2] and Sttoc,ha.stic Systems Theory a.nd It,s Applica-
it is not, fatal error for t,he decision making of the t8ions,pp. 167-171 (1995)
insulator washing. S. Got,o, M. Naltamura., N . Nanaya.kkara and
T. Ta.niguchi: Automat,ization of Human Deci-
5 . Conclusion sion Ma.lting for Washing Time of Polluted In-
A niet,hod of automatic decision making for wasli- s u h t o r s in Substat,ions, Proceedings of the 34th
ing the polluted insulators in substations near coa.sta1 SICE Annual Conference (1995)
area.s was developed. The automatization was based T. Oltuno, H. Kume, T. Haga and T. Yoshizawa,
on t,he modeling of the polliit,ion depositss and the Mult8ivaria.ble Analysis Method, pp. 273-302,
discriminant a.nalysis by using the multi-dimensional Nikka.giren (1971) (in .Japanese).
norma.1 distribution. T h e proposed method was ap-
plied t,o the actual da.ta of t8he 1Cara.tsu substation
and over SO[%] of human decisions were correctly de-
tec,ted with the proposed automatic decision making
process. T h e future problem is t,o increa.se the per-
centage of t8hecorrect, detection a.nd to decrea.se t,he
miss det,ect,ion.
Acknowledgnent
T h e authors a.re grat,eful t,o engineers in Sa.ga,
bra.nch of Kyushu Elect,ric Power Compa.ny for t,heir
cominent,s and dat>a.a.cquisit8ion.
References
[l] Elect>ric.Joint, Resea.rch, vol. 20, no. 2 (1964) (in

411