TEHNICI DE CLUSTERIZARE IN ANALIZA SARCINII

CLUSTERING TECHNIQUES IN LOAD ANALYSE

Gheorghe CRIN Gheorghe GRIGORA
Universitatea Tehnic Gh. Asachi Iai, Facultatea de Electrotehnic, Romnia
Bd. Dimitrie Mangeron, 51-53, Iai 6600, Tel.: 232 27 86 83; Fax: 232 23 76 27; E-mail: ggrigor@ee.tuiasi.ro, gcartina@ee.tuiasi.ro

Elena-Crenguta BOBRIC
Universitatea Stefan cel Mare Suceava, Facultatea de Inginerie Electrica, Romania
Strada Universitatii 13, 720229 - Suceava, Tel/Fax: 230 52 48 01 E-mail: crengutab@eed.usv.ro

Rezumat: In reelele de distribuie exista puine informaii despre

gradul de ncrcare a transformatoarelor din posturile de
transformare. Feederii si sarcinile nu sunt in mod obinuit
monitorizate. Deci, in fiecare moment exista un grad de incertitudine asupra sarcinilor din noduri si prin urmare si asupra
nivelului de ncrcare al reelei, nivelului de tensiune din noduri
si al pierderilor de putere. Prin urmare, in reelele electrice de
distribuie este importanta estimarea profilurilor de sarcina din
noduri ca o alternativa la determinarea consumului de energie pe
baza msurtorilor efectuate in nodurile reelei. Lucrarea prezint
o analiza a pofilelor graficelor de sarcina a unei staii electrice
de distribuie utiliznd tehnici de grupare spaiala.

Abstract: In distributions network existing little information

about loading level of transformer of substations. Feeders and
loads are not monitoring usually. Therefore, in each moment
existing one incertitude degree about buses loads and accordingly
and about load level of network, voltage level by buses and power
loses. Accordingly, in distribution electric network is important to
estimate the load profiles of buses, as one alternative at determination of energy consumption from effectuated measurement
in busses network. The paper presents one analyse at load profile
for a distribution substation utilising clustering techniques.

Keywords: profil de sarcina, tehnici de grupare, zile caracteristice,

tipuri de consumatori

Keywords: load profile, clustering techniques, characteristic day,

consumer tips

1. Introducere
Sarcinile din nodurile reelelor de distribuie sunt, in
general, intr-o continu cretere in timp, ca urmare a utilizrii
unui numr din ce in ce mai mare de receptoare electrice,
cat si datorita creterii serviciilor pentru populaie.
Modelele pentru determinarea sarcinilor electrice sunt
diferite, in funcie de tipul de reea: urban, rural sau
industrial.
Obinerea de informaii referitoare la sarcinile din
nodurile reelelor de distribuie se realizeaz prin:
- teletransmisia sarcinilor din nodurile reelei;
- utilizarea unor dispozitive semiautomate locale de msurare a sarcinilor electrice;
- utilizarea de sisteme electronice moderne de achiziie
de date;
- utilizarea informaiilor primare privind sarcinile electrice
din noduri.
Obinerea ultimei categorii de informaii necesita efectuarea de msurtori ale sarcinilor electrice ale consumatorilor in anumite perioade caracteristice: iarna, vara, zi
lucratoare, zi de repaus, dup care se prelucreaz rezultatele
msurtorilor efectuate.
Variaia in timp a sarcinii electrice se reflect n graficul de
sarcina zilnic, sezonier si anual, care trebuie sa prelucreze un
numr ct mai mare de date reale de consum de energie.
Reelele electrice de distribuie au un numr mare de
noduri de sarcin chiar dac se iau n considere numai nodurile in care sunt amplasate posturi de transformare. Consuma-

1. Introduction
Generally, the load of busses distribution network are in a
continuous development in time, pursuant to of a utilisation
number more and more big of electric receivers, quotients and
by reason of the development of logistics for population.
The models for electric loads determinations are different, depending on the networking tips: urban, rural and
industrial.
Obtain of information referring to load from busses
networks is achieved through:
- load teletransmission from busses networks;
- use the half-automatic local devices of measure the
electric loads;
- use the electronic modern systems for data acquisitions;
- use primary information concerning electric loads from
busses.
Obtain last categories of informations require the
effectuation of measurements of the electric load of the
consumers of certain in characteristic periods: winter, the
summer, working day and weekend day, thereto is remake
result of the effectuated measurements.
Temporally variation of the electric load reflected the
graphic of daily, seasonal and annual, that carry to remake
a how much number of real dates of consumption of electric
energy.
The electric distribution network have a big number of
loads busses even if taken the in consider merely the busses
which in place the substation. The consumers connect in

124

th

The 6 International Power Systems Conference

torii racordai in nodurile reelei sunt si ei foarte numeroi,

eterogeni ca puteri absorbite, tehnologii utilizate sau comportament social, impunnd cicluri particulare de lucru.
Aceasta dificultate este nlturata daca in studiile de
regim ale acestor reele se utilizeaz graficele de sarcina
zilnice ale fiecrui nod, in regimurile caracteristice (iarna
si vara, zi lucratoare si zi repaus). Acestea se constituie pe
baza graficelor tip de sarcina ale consumatorilor.

the network busses are and they very numerous, heterogeneous as the absorbed powers, using his technologists
the social behaviors, enforcing the particular loops of thing.
This difficulty is eliminate if in under consideration of
regimes ale these network is utilized the graph of daily
for each bus, in the characteristic regimes (winter and the
summer, working day and weekend day). These constitute
on the strength of load graph tip of the consumers.

2. Curbe de sarcina
Seriile de timp care descriu curbele zilnice de sarcin
sunt procese nestaionare, afectate de condiii atmosferice,
n special temperatur, dar i ali parametri care ar putea
s fie luai n consideraie sunt umiditatea sau viteza
vntului, evenimente speciale (greve, evenimentele de
sport) i caracterul aleatoriu al consumului industrial. Ca
regul, grupul cel mai stabil de consumatori, ale crei
variaii de sarcin urmeaz un model comun care poate fi
evaluat, este grupul consumatorilor casnici.
Modelele curbelor de sarcin sunt caracterizate de o
periodicitate zilnic sau sptmnal. Oricum, curba de
sarcin pentru mine sau sptmna urmtoare nu este
tocmai o copie a curbei de sarcin de azi sau aceast
sptmn. n schimb, curba de sarcin este puin modificat de la zi la zi i sptmn la sptmn, reflectnd
schimbri n comportarea de consumatorilor sau condiiile
de vreme. Sunt autorii care consider analize separate
pentru fiecare zi de sfrit de sptmn, n timp ce alii
propun analize separate pentru 3 tipuri de zilele din
sptmn: luni, mari pn joi i vineri. n ultimul caz,
curbele de sarcin sunt similare pentru toate zilele cu
excepia dimineii de luni i serii de vineri. Un tip special
de zi este vacana. Civa autori grupeaz zilele de vacan
cu cele de sfrit de sptmn.
Curbele de sarcina pot fi astfel folosite pentru urmtoarele aplicaii:
- calculul tensiunii din nodurile reelelor de distribuie;
- determinarea vrfului de sarcina pentru diverse tipuri
de consumatori;
- calcularea pierderilor de putere si energie
- determinarea sistemelor de tarifare
Profilul consumului de putere pentru diferite tipuri de
consumatori pot fi reunite pentru a gsi vrful de sarcin a
sistemului. Cu estimarea ratei de cretere a sarcinii pentru
fiecare tip de consumator, se poate face o mai buna activitate
de planificare a reelelor de distribuie i poate fi realizat
o calitate superioar a serviciilor. O dat cu utilizarea de
noi receptoare specifice diferitelor anotimpuri (aerul condiionat de exemplu), vrful de sarcin crete semnificativ
astfel nct se pune problema controlului rezervei de putere
necesar pentru a asigura sigurana n funcionare. De asemenea, se poate stabili structura tarifelor energiei electrice
innd cont de consumul de putere a diferitelor clase de
consumatori n ansamblul graficului de sarcina a sistemului.
Reprezentarea curbelor de sarcina precum i curbele
pentru medie i dispersie sunt utilizate pentru calculele
inginereti si pentru analiza statistic. Se poate stabili astfel
un criteriu de performan bazat pe valorile determinate
probabilistic. In reelele de distribuie urbane i rurale,
sarcinile activ i reactiv se supun, n orice moment legii
normale de distribuie. Expresiile de calcul pentru cele dou
mrimi caracteristice: valoarea medie (1) i dispersia (2).

2. Load curves
The time series that describe the daily load profiles in
a system are nonstationary processes, affected by weather
conditions, especially temperature, but other parameters
that could be considered are humidity or wind speed),
special events (strikes, sport events) and the randomness
of the industrial consumption. As a rule, the most stable
group of consumers, whose load changes following a
common pattern that can be assessed, is the group of
residential loads.
The shape of the load profiles describes usually a daily
and weekly periodicity. However, the load profile for
tomorrow or the next week is not just a simple copy of the
load profile from today or this week. Instead, the load
profile is slightly modified from day to day and week to
week, to reflect changes in consumers behavior or weather
conditions. Typically, daily load profiles are classified as
week days and weekend days. Some authors consider
separate analysis for each weekend day, while others deals
with separate analysis for 3 types of week days, Monday,
Tuesday to Thursday and Friday. In the last case the shapes
of the load profiles are similar for all week days except
the morning of Monday and the evening of Friday. A special
type of day is the holiday. Some authors group the holidays
with the weekend days.
The curves in this form can already been used for
following applications:
- voltage calculation for busses of distribution network;
- determination the peak load for diversity tips of consumer;
- energy and power losses calculation;
- proper tariff design
The power consumption profiles of various customer
types can be integrated to find the system peak loading.
With the prediction of load growth rate for each
customer type, more accurate load forecast can be obtained
and better distribution network expansion planning can be
made to achieve good service quality. With the usage of
more and receivers, differently season, air conditioners for
examples, the peak loading is increased so significantly,
so that to put the problems for reserve capacity control
that is necessary for system operation criterion. As well,
the rate tariff structure can therefore be designed according
to the contribution of power consumption of different
customer classes to the whole system load profile.
The loading curve representation by their mean and
standard deviation curves is useful for engineering calculation and statistical analysis. A performance criterion can
also be established based on probabilistic value. In urban
and rural distribution network, the active and reactive loads
are submitted, in every moment of normally distribution
law. The calculus expressions for there two characteristic
sizes: mean (1) and standard deviation (2).

03-04.11.2005, Timioara, Romania

125
n

P=
n

P =

P
i =1

P
i =1

,Q=

i =1

(1)

n
n

2
2

P , Q =

unde: Pi , Qi - sarcina activ, respectiv reactiv msurat

la momentul i

P, Q - sarcina activ, respectiv reactiv medie

3. Metode de clusterizare
Analiza datelor st la baza multor aplicaii, fie n faza
de proiectare sau ca parte a operaiilor lor on-line. Analiza
gruprii reprezint organizarea unei colecii de tipare (de
obicei reprezentate ca un vector de msurri, sau un punct
ntr-un spaiu multidimensional) n grupri bazate pe
similitudini. n mod intuitiv, tiparele dintr-o grupare valid
sunt mai asemntoare cu fiecare dect cu un tipar ce
aparine unei grupri diferite.
Varietatea de tehnici de reprezentare a datelor, msurarea apropierii (similaritii) dintre elementele de date, i
gruparea elementelor de date a produs o confuzie mare i
adesea a sortrii metodelor de clusterizare.
Este important s nelegem diferena dintre clusterizare
(clasificare nesupravegheat) i analiza discriminant
(clasificarea supravegheat). n cadrul nvrii supravegheate, ne sunt asigurate cu o colecie de tipare etichetate
(preclasificate); problema const n a eticheta un nou tipar
ntlnit dar nc neetichetat. n mod tipic, tiparele etichetate
date sunt folosite pentru a nva descrierea claselor care
n schimb sunt folosite pentru a eticheta un nou tipar. n
cazul clusterizrii, problema const n a grupa o colecie
dat de tipare neetichetate n grupri semnificative. ntr-un
fel, etichetele sunt asociate de asemenea cu clusterii, dar
aceste categorii de etichete sunt antrenate de date; asta dac,
sunt obinui exclusiv numai din date.
Clusterizarea este folositoare n cteva analize de proba
a tiparelor, gruparea, luarea deciziilor, i situaii de nvare
a mainilor, inclusiv explorarea datelor, recuperarea documentelor, segmentarea imaginilor, i clasificarea tiparelor.
Totui, n multe asemenea probleme, exist puin informaie
important (de ex. modelele statistice) disponibile despre
date, i cel care ia decizia trebuie s fac cteva presupuneri
pe ct posibil despre date. Sub aceste restricii metodologia
de clusterizare este n special apropiat de cercetarea
raporturilor dintre date pentru a face o evaluare (probabil
preliminar) a structurii lor. Termenul clusterizare
(clustering) este folosit n mai multe comuniti de cercetare
pentru a descrie metode pentru gruparea datelor neetichetate.
Aceste comuniti au terminologii i presupuneri diferite
pentru componentele procesului de clusterizare i contextul
n care clusterizarea este folosit.
Clusterizarea este tehnica de grupare a irurilor, care
atribuie valori similare peste un numr de variabile. Este o
tehnic de cercetare deosebit care ajut pentru a nelege
organizarea pe grupe a datelor. Sunt dou metode majore
de clusterizare: clusterizare ierarhic i clusterizare k medii.

2
i

i =1

(2)

where: Pi , Qi - active, respectively reactive load, measurement in i moment

P, Q - active, respectively reactive mean load

3. Clustering methods
Data analysis underlies many computing applications,
either in a design phase or as part of their on-line operations.
Cluster analysis is the organization of a collection of patterns
(usually represented as a vector of measurements or a point
in a multidimensional space) into clusters based on similarity. Intuitively, patterns within a valid cluster are more
similar to each other than they are to a pattern belonging
to a different cluster.
The variety of techniques for representing data, measuring proximity (similarity) between data elements, and
grouping data elements has produced a rich and often
confusing assortment of clustering methods.
It is important to understand the difference between
clustering (unsupervised classification) and discriminant
analysis (supervised classification). In supervised classification, we are provided with a collection of labeled
(preclassified) patterns; the problem is to label a newly
encountered, yet unlabeled, pattern. Typically, the givenlabeled (training) patterns are used to learn the descriptions of classes that in turn are used to label a new pattern.
In the case of clustering, the problem is to group a given
collection of unlabeled patterns into meaningful clusters.
In a sense, labels are associated with clusters also, but
these category labels are data driven; that is, they are
obtained solely from the data.
Clustering is useful in several exploratory patternanalysis, grouping, decision-making, and machine-learning
situations, including data mining, document retrieval, image
segmentation, and pattern classification. However, in many
such problems, there is little prior information (e.g.,
statistical models) available about the data, and the decision-maker must make as few assumptions about the data
as possible. It is under these restrictions that clustering
methodology is particularly appropriate for the exploration
of interrelationships among the data points to make an
assessment (perhaps preliminary) of their structure. The
term clustering is used in several research communities
to describe methods for grouping of unlabeled data. These
communities have different terminologies and assumptions
for the components of the clustering process and the contexts
in which clustering are used.
Clustering is the technique of grouping rows together
that share similar values across a number of variables. It is a
wonderful exploratory technique to help you understand the
clumping structure of your data. There are two major methods
of clustering: hierarchical clustering and k-means clustering.

th

The 6 International Power Systems Conference

126

n clusterizarea ierarhic datele nu sunt atribuite unui

cluster anumit ntrun singur pas. n schimb, au loc o serie de
mpriri care pot varia de la un singur cluster ce conine
toate obiectele, la n clustere fiecare coninnd un singur
obiect. Clusterizarea ierarhic este submparit n procedee
de aglomerare, care constau ntro serie de fuziuni a celor
n obiecte n grupe, i procedee de mprire, care separ
cele n obiecte, succesiv, n grupe finite. Tehnicile de aglomerare sunt, n mod obinuit, mult mai utilizate. Gruparea
ierarhic poate fi reprezentat printro diagram bidimensional denumit dendogram, care ilustreaz fuziunea sau
mprirea fcut la fiecare stadiu succesiv al analizei. Clusterizarea ierarhic este adecvat pentru tabele mici, pn la
cteva sute de linii. Se poate alege numrul de clusteri
dorii dup ce este construit arborele. Cteva procedee de
aglomerare sunt: clusterizare prin legtur simpl, clusterizare prin legtur complet, clusterizarea prin legtur medie, metoda centrului de greutate i metoda de clusterizare
ierarhic a lui Ward. Diferena ntre metode apare prin ci
diferite de a defini distana dintre clustere.
n metoda centrului de greutate, metoda utilizat n
analiz, distana dintre doi clusteri este definit ca distana
ptratic euclidian dintre mediile lor. Metoda centroidului
este mult mai robust, prin abaterea mare de la medie,
dect alte metode ierarhice dar n alte privine poate s nu
calculeze aa de bine ca i metoda Ward sau lungime medie:

DKL = X K X L

In hierarchical clustering the data are not partitioned

into a particular cluster in a single step. Instead, a series
of partitions takes place, which may run from a single cluster
containing all objects to n clusters each containing a single
object. Hierarchical clustering is subdivided into agglomerative methods, which proceed by series of fusions
of the n objects into groups, and divisive methods, which
separate n objects successively into finer groupings. Agglomerative techniques are more commonly used. Hierarchical clustering may be represented by a two dimensional
diagram known as dendrogram which illustrates the fusion
or divisions made at each successive stage of analysis.
Hierachical clustering is appropriate for small tables. Up
to several hundred rows. You can choose the number of
clusters you like after the tree is built. Several agglomerative
techniques are single linkage clustering, complete linkage
clustering, average linkage clustering, centroid method and
Wards hierarchical clustering method. Differences between
methods arise because of the different ways of defining
distance (or similarity) between clusters.
In the centroid method, method utilizing in this analysis,
the distance between two clusters is defined as the squared
Euclidean distance between their means. The centroid method
is more robust to outliers than most other hierachical
methods but in other respects may not perform as well as
Wards method or average linkage:

DKL = X K X L

Kmedii este unul din cei mai simplii algoritmi de nvare nesupravegheat care rezolv bine cunoscuta problem
de clusterizare. Procedeul urmeaz o cale simpl i uoar
pentru a clasifica setul datelor de intrare printr-un numr de
clusteri (presupunem K clusteri) fixai a priori. Ideea de baz
este a definii K centrii de greutate, unul pentru fiecare cluster.
Aceste centre de greutate trebuie fixate inteligent deoarece
locaii diferite conduc la rezultate diferite. Deci, alegerea cea
mai bun este s le fixm att ct este posibil mai ndeprtate
unele de altele. Urmtorul pas este s lum fiecare punct de-a
lungul setului datelor de intrare i s-i asociem celui mai apropiat centru de greutate. Cnd nu exist puncte nedecise, primul
pas este complet, iar o grupare iniial s-a fcut. n acest punct
trebuie s recalculm K, noi centre de greutate ale clusterilor
rezultai din pasul anterior. Dup ce avem aceti K noi centriozi,
o nou legtur a fost fcut dintre acelai set de date i cel
mai apropiat centru de greutate nou. A fost generat un nod.
Ca rezultat al acestui nod, notm c, K centrii de greutate i
modific locaia pas cu pas cnd nu se mai fac modificri.
Cu alte cuvinte nu se mai mic centroizii. n final scopul
acestui algoritm este minimizarea unei funcii obiectiv, n
acest caz o funcie eroare ptratic. Funcia obiectiv este:
k

K-means is one of the simplest unsupervised learning

algorithms that solve the well known clustering problem.
The procedure follows a simple and easy way to classify a
given data set through a certain number of clusters (assume k
clusters) fixed a priori. The main idea is to define k
centroids, one for each cluster. These centroids shoud be
placed in a cunning way because of different location causes
different result. So, the better choice is to place them as
much as possible far away from each other. The next step
is to take each point belonging to a given data set and
associate it to the nearest centroid. When no point is
pending, the first step is completed and an early groupage
is done. At this point we need to re-calculate k new centroids
as barycenters of the clusters resulting from the previous
step. After we have these k new centroids, a new binding
has to be done between the same data set points and the
nearest new centroid. A loop has been generated. As a result
of this loop we may notice that the k centroids change their
location step by step until no more changes are done. In
other words centroids do not move any more. Finally, these
algorithms aim at minimizing an objective function, in this
case a squared error function. The objective function:

J = xi( j ) c j

j =1 i =1

( j)

unde: xi

cj

este o distan aleas, msurat ntre

( j)

punctul xi i centru clusterului cj

J este un indicator de distan a n puncte din centrele
respective clusterilor proprii. Clusterizarea Kmedii
este convenabil pentru tabele mari de la sute de mii
de linii.

( j)

where: xi

cj

is a chosen distance measure between

( j)

a data point xi

and the cluster center cj.

J is an indicator of the distance of the n data points

from their respective cluster centres. K-means
clustering is suitable for larger tables, up to hundreds of thousands of rows.

03-04.11.2005, Timioara, Romania

127

4. Studiu de caz
Lucrarea i propune s analizeze curbele de sarcin,
nregistrate pe o perioad de dou luni, pe partea de 20 kV a
transformatorului de 16 MVA dintr-o staie de distribuie de
110/20 kV. Curbele de sarcin sunt ridicate utiliznd nregistrrile contorului de energie activ. Intervalul de timp de
preluare a datelor este de 15 minute. Astfel, curba de sarcin
este reprezentat cu 96 valori ale sarcinii pentru o zi.
Consumul este format preponderent din consumatori casnici rurali. Pe lng acetia, exist mici consumatori industriali
i consumatori edilitari, dar a cror consum nu reprezint
mai mult de 10% din consumul lunar de energie electric.
Analiza este efectuat pe 58 de zile, nregistrri din
perioada 20 februarie 20 aprilie.

4. Case study
This paper proposed the analysis of load profile recording for two mounts period, of part from 20 kV of transformer
by 16 MVA of the distribution substation 110/20 kV. The
load diagram is constructed using the register of watt-hour
meter. The time interval of sampling load curve data is
15 minutes. Thus, the load profile is represented by 96 load
values throughout of the day.
The consumption is preponderant formatted by the rural
consumers. Addition this, exist the little industrial consumers
and the town consumers, but these consumption not representing moreover 10% by the monthly electric energy consumption.
The analyses are effectuated on 58 days, registration
by 20 February 20 April period.

Fig. 1. Curba de sarcina pentru zile caracteristice

zi lucrtoare
zi sfrit de sptmn

Fig. 1. Load profiles for characteristic days

week day
weekend day

n figura 1 sunt reprezentate curbele de sarcin medii

pentru zile caracteristice din luna martie respectiv aprilie.
Toate msurtorile efectuate trebuie prelucrate, prin
aranjarea i normalizarea acestora. Ca factor de normalizare
s-a utilizat consumul de energie.

zij =

In figure 1 are represented the mean load profile for

characteristic days from March respectively April.
Every measurement effectuated must be elaborated
through these arranged and normalized. As normalized
factor is using the energy consumption.

xij
24

x
i =1

unde : z ij - valoarea normalizat a puterii active

xij - valoarea msurat a puterii active orare

24

xij - consumul de energie electric zilnic

i =1

Gruparea curbelor de sarcin prin comparare vizual

este subiectiv i greu de aplicat. Pentru rezolvarea problemei s-a aplicat metoda tehnicilor de grupare. Gruparea
se face progresiv n grupe coerente i reprezentative.
Ca metod de grupare a fost folosit metoda centrului
de greutate. Aceasta a fost aplicat pentru baza de date
format din graficele de sarcin a puterii active pentru cele
58 de zile.

ij

where : z ij - normalised value for active power

xij - measured value for active power

24

x
i =1

ij

- daily energy consumption

Classification of the load profiles through the visual

comparison is subjective and impractical. The cluster
analysis method was applied to solve that problem. The
clustered process is making progressive into coherent and
representative cluster.
As clustering method to be using the centroid method.
This method to be applied for the dates formatted by load
profile for active power for these 58 days.

th

The 6 International Power Systems Conference

128

29

36

27

55

34

56

47

33

58

57

43

51

46

45

44

54

52

40

39

37

32

31

48

41

49

42

35

In figure 2 is represented the dendogram for load profile

clustering.

21

50

28

20

14

24

25

38

23

22

17

16

53

11

26

15

19

18

10

13

12

30

n figura 2 este reprezentat dendograma pentru gruparea

curbelor de sarcin.

Fig. 2. Dendograma pentru gruparea curbelor de sarcin

Fig. 2. The dendogram for cluster the load profiles

Dac se analizeaz aceast dendogram se observ c

au rezultat trei grupe, care, cu cteva excepii, au fost
realizate din curbe de sarcin corespunztoare a trei perioade de timp: 21-25 februarie, 1 20 martie i respectiv
21 martie 19 aprilie.
Fiecare grup este format din dou subgrupe corespunztoare zilelor caracteristice: zile lucrtoare i zilelor
de repaus. De asemenea exist cteva zile ale cror grafice
de sarcin nu s-au alturat nici uneia din cele trei grupe.
Se poate face observaia c o influen major asupra
structurii consumului o au condiiile atmosferice, in special,
temperatura mediului ambiant.
Dac se analizeaz curbele de sarcin ale zilelor care
au rmas n afara grupelor, se poate observa c abaterea
lor fa de curba de sarcin medie este mare, fa de cea a
unei zile a crei curb de sarcin s-a grupat. Acest lucru
poate fi observat n figura 3.

If we analyse this dendogram it observed that have

result three cluster, who, several excepting, have be realized
by profiles load accordingly whit three time periods:
21-25 February, 1 20 Mars and respectively 21 Mars
19 April.
Every cluster is formatted by two subgroups corresponding the characteristic days: week days and weekend
days. As well, existing several days they load profile for
active power not enclosed in any by three clusters.
It possible to make observation that a majored influence about consumption structure to have the atmospheric
conditions, specially, the environment temperature.
If is analyzing the load profile for neclustering days, it
is observed that standard deviation against mean load profile
is big, beside that for one clustering day. This thing can be
observed in figure 3.

Fig. 3. Curba de sarcin medie n comparaie cu un profil negrupat respectiv un profil din grup
Fig. 3. Mean profile and load profile out of cluster respectively one load profile from cluster
curba de sarcin medie
mean load profile
load profile out cluster
load profile from cluster

03-04.11.2005, Timioara, Romania

129

30

21

37

36

26

32

29

28

27

35

33

25

24

22

20

19

31

16

If it eliminated the load profile for stains day without

clusters and it effectuated a new grouping. It observes that
for each forenamed period is clustering separate in weekdays and weekend days (fig. 4).

17

23

15

14

11

10

34

18

13

12

Dac se elimin curbele de sarcin pentru zilele care au

rmas n afara grupelor i se va face o nou grupare. Se observ c pentru fiecare perioad amintit se grupeaz separat
zilele lucrtoare si cele de sfrit de sptmn (fig. 4).

Fig. 4. Dendograma pentru gruparea curbelor de sarcin

Fig. 4. The dendogram for cluster the load profiles

De asemenea, se observ c exist 3 excepii de la regula

de grupare. Analiznd temperatura, nu se poate spune c
media zilei difer foarte mult de celelalte zile.
Este posibil ca golul de sarcin dintre orele 15-18 a
curbelor de sarcin din ultima perioad analizat s fi fost
acoperit de puterea cerut de consumatori care au alimentarea de rezerv din aceast staie.
Aceast analiz poate continua innd seama de ceilali
factori care influeneaz structura consumului i/sau
inndu-se seama de graficele tip de sarcin ale consumatorilor alimentai din aceast staie de distribuie.

Also, it observes that existing 3 exceptions from

clustering technique. Analyzing the temperature cant say
that average of day is much different from other days.
Is possible that as load emptiness between 15-18 hours
for load profile of the last analyzed period to be was overcast
by demand load by consumer who has the reserved supply
from this substation.
This analysis to be continued if allowance by the other
factors that influenced the consumption structure and/or
allowance by typical load profile for the consumer suppliers
from this distribution substation.

5. Concluzii
In aceast lucrare se propune o metod de analiz
comparativ a profilurilor zilnice de sarcin a unei staii
de distribuie, pe o perioad de dou luni, ntre 21 februarie
19 aprilie.
Gruparea curbelor de sarcin prin comparare vizual
este subiectiv i greu de aplicat. Pentru rezolvarea problemei s-a aplicat metoda tehnicilor de grupare. Gruparea
se face progresiv n grupe coerente i reprezentative.
Pentru aceast analiz s-au folosit tehnici de grupare
ierarhice, mai exact metoda centrului de greutate.
Temperatura este un factor de influen a cererii de
putere a consumatorilor, fapt care se evideniaz din analiza
dendogramelor obinute n urma procesului de grupare.
Dendograma evideniaz gruparea pe zile caracteristice,
zile lucrtoare i zile de repaus, a curbelor de sarcin, pe
cele trei perioade analizate.
Rezultatele demonstreaz abilitatea tehnicilor de grupare
n clasificarea curbelor de sarcin i analiza comparativ a
consumurilor.

5. Conclusion
In this paper is proposed one comparative analyzed
method for daily load profiles for one distribution substation, for a period from two month, between 21February
19 April.
Classification of the load profiles through the visual
comparison is subjective and impractical. The cluster
analysis method was applied to solve that problem. The
clustered process is making progressive into coherent and
representative cluster.
For this analysis was using the hierarchical clustering
techniques, exactly the centroid method.
The temperature is a influenced factor for the demand
power of consumer, fact that is evidentially from the dendogram obtained behind the clustering process.
The dendodram evidenced the clustering of the load
profile by the characteristic days, week days and weekend
days, for there three analyzed periods.
The results demonstrate the abylity of clustering techniques for classification of load profile and for comparative
analyses of consumtion.

130

th

The 6 International Power Systems Conference

Bibliografie (References)
1. Crin, Gh., Grigora, Gh.: Tehnici moderne de optimizare. Aplicaii n energetic (Modern Optimization Techniques. Applications in
Power Systems), Casa de Editur VENUS, Iai, 2002.
2. Crin, Gh., Grigora, Gh.: Tehnici de Inteligen Artificial n Electroenergetic, Editur SETS, Iai, 2004
3. Crin Gh., Song Y.H., Grigora Gh.: Optimal Operation and Planning of Power Systems, Casa de Editur VENUS, Iasi, 2003.
4. JMP Statistics and Graphics Guide: Version 3, SAS Institute Inc., Cary, NC, USA, 1999.
5. Clustering:An ntroduction: http://www.elet.polimi.it/upload/matteucc/Clustering/tutorial_html/
6. Gasperic, S., Gerbek D., Gubina F.: Determination of the Consumers Load Profiles, http://www.telmark.org/2002Sep/2-5_Gasperic.pdf
7. Chicco G., Napoli R., Piglione F., Postolache P., Scutariu M., Toader C.: A Review of Concepts and Techniques for Emergent Customer
Categorisation, http://www.telmark.org/2002Sep/2-4_Chicco.pdf
8. Gavrilas, M.: Neural network based forecasting for electricity markets, http://www.telmark.org/2002Sep/3-6_Gavrilas.pdf.