IUST v7n4p249 En
-
Upload
joydeepd3232 -
Category
Documents
-
view
215 -
download
0
Transcript of IUST v7n4p249 En
-
8/22/2019 IUST v7n4p249 En
1/11
Iranian Journal of Electrical & Electronic Engineering, Vol. 7, No. 4, Dec. 2011 249
Different Methods of Long-Term Electric Load Demand
Forecasting; A Comprehensive Review
L. Ghods* and M. Kalantar*
Abstract: Long-term demand forecasting presents the first step in planning and developingfuture generation, transmission and distribution facilities. One of the primary tasks of an
electric utility accurately predicts load demand requirements at all times, especially forlong-term. Based on the outcome of such forecasts, utilities coordinate their resources tomeet the forecasted demand using a least-cost plan. In general, resource planning isperformed subject to numerous uncertainties. Expert opinion indicates that a major sourceof uncertainty in planning for future capacity resource needs and operation of existing
generation resources is the forecasted load demand. This paper presents an overview of thepast and current practice in long- term demand forecasting. It introduces methods, whichconsists of some traditional methods, neural networks, genetic algorithms, fuzzy rules,
support vector machines, wavelet networks and expert systems.
Keywords: Long-term, Demand Forecasting, Neural Networks, Genetic Algorithms, FuzzyRules.
1 Introduction1
A power system serves one function and that is tosupply customers, both large and small, with electricalenergy as economically and as reliability as possible.Another responsibility of power utilities is to recognizethe needs of their customers (Demand) and supply the
necessary energies. Limitations of energy resources inaddition to environmental factors, requires the electricenergy to be used more efficiently and more efficientpower plants and transmission lines to be constructed[1]. Long-term demand forecasts span from eight yearsahead up to fifteen years. They have an important rolein the context of generation, transmission anddistribution network planning in a power system. Theobjective of power system planning is to determine an
economical expansion of the equipment and facilities tomeet the customers' future electric demand with anacceptable level of reliability and power quality [2].
Accurate long-term demand forecasting plays anessential role foe electric power system planning. Itcorresponds to load demand forecasting with lead times
enough to plan for long-term maintenance, constructionscheduling for developing new generation facilities,
purchasing of generating units, developing transmissionand distribution systems. The accuracy of the long-term
Iranian Journal of Electrical & Electronic Engineering, 2011.Paper first received 18 Jan. 2011 and in revised form 13 Jun. 2011.
* The Authors are with the Department of Electrical Engineering, Iran
University of science and technology (IUST), Tehran, Iran.E-mails: [email protected], [email protected].
load forecast has significant effect on developing futuregeneration and distribution planning. An expensiveoverestimation of load demand will result in substantialinvestment for the construction of excess powerfacilities, while underestimation will result in customerdiscontentment. Unfortunately, it is difficult to forecast
load demand accurately over a planning period ofseveral years. This fact is due to the uncertain nature ofthe forecasting process. There are a large number ofinfluential that characterize and directly or indirectlyaffect the underlying forecasting process; all of themuncertain and uncontrollable [3]. However, neither theaccurate amount of needed power nor the preparationfor such amounts of power is as easy as it looks,because: (1) long-term load forecasting is always
inaccurate (2) peak demand is very much dependant ontemperature (3) some of the necessary data for long-term forecasting including weather condition andeconomic data are not available, (4) it is very difficult tostore electric power with the present technology, (5) ittakes several years and requires a great amount of
investment to construct new power generation stationsand transmission facilities [4]. Therefore, any long-term
load demand forecasting, by nature, is inaccurate!Generally, long-term load demand forecasting
methods can be classified in to two broad categories:parametric methods and artificial intelligence basedmethods. The artificial intelligence methods are furtherclassified in to neural networks [1], [2], [4], [8], [10],
support vector machines [15], genetic algorithms [14],
-
8/22/2019 IUST v7n4p249 En
2/11
250 Iranian Journal of Electrical & Electronic Engineering, Vol. 7, No. 4, Dec. 2011
wavelet networks [12] [13], fuzzy logics [16] and expertsystem [17] methods. The parametric methods are basedon relating load demand to its affecting factors by amathematical model. The model parameters areestimated using statistical techniques on historical data
of load and it's affecting factors. Parametric loadforecasting methods can be generally categorized under
three approaches: regression methods, time seriesprediction methods [3]. Traditional statistical loaddemand forecasting techniques or parametric methodshave been used in practice for a long time. Thesetraditional methods can be combined using weightedmulti-model forecasting techniques, showing adequateresults in practical system. However, these methods
cannot properly present the complex nonlinearrelationships that exist between the load and a series offactors that influence on it [2].
In this paper, we introduce a brief overview in long-
term forecasting methods. This paper is organized asfollows. Next section briefly describes parametric
models. Section III describes different artificialintelligence based methods and section IV is the
conclusions of paper.
2 Parametric Methods
The three types of well-known parametric methodsare as, trend analysis, end-use modeling andeconometric modeling.
2.1 Trend Analysis
Trend analysis extends past rates of electricity
demand in to the future, using techniques that rangefrom hand-drawn straight lines to complex computer-produced curves. These extensions constitute theforecast. Trend analysis focuses on past changes ormovements in electricity demand and uses them topredict future changes in electricity demand. Usually,
there is not much explanation of why demand acts as itdoes, in the past or in the future. Trending is frequently
modified by informed judgment, wherein utilityforecasters modify their forecasts based on theirknowledge of future developments which might makefuture electricity demand behave differently than it hasin the past [5].
The advantage of trend analysis is that, it is simple,quick and inexpensive to perform [5].
The disadvantage of a trend forecast is that it
produces only one result, future electricity demand. Itdoes not help analyze why electricity demand behavesthe way it does, and it provides no means to accuratelymeasure how changes in energy prices or governmentpolities influence electricity demand [5].
2.2 End-Use Models
The end-use approach directly estimates energyconsumption by using extensive information on endusers, such as applications, the customer use, their age,
sizes of houses, and so on. Statistical information about
customers along with dynamics of change is the basisfor the forecast [5].
End-use models focus on the various uses ofelectricity in the residential, commercial, and industrialsector. These models are based on the principle that
electricity demand is derived from customer's demandfor light, cooling, heating, refrigeration, etc. Thus, end-
use models explain energy demand as a function of thenumber of applications in the market [5].
Ideally, this approach is very accurate. However, itis sensitive to the amount and quality of end-use data.For example, in this method the distribution ofequipment age is important for particular types ofappliances. End-use forecast requires less historical data
but more information about customers and theirequipments [5].
This method predicts the energy consumptions. Ifwe want to calculate the load, we have to have the load
factor in each sections and different types of energyconsumptions and then by load factor we can calculatethe load in each section.
The system load factor is defined as follows
equation:
Average - load demandLoadFactor =
Peak - load demand
annual KWh energy=
peak - load demand 8760 hours/year
(1)
The disadvantage of end-use analysis is that most
end-use models assume a constant relationship between
electricity and end-use (electricity per appliance). Thismight hold for over a few years, but over 10 or 20-yearperiod, energy saving technology or energy prices willundoubtedly change, and the relationships will notremain constant [6].
2.3 Econometric Models
The econometric approach combines economictheory and statistical techniques for forecasting
electricity demand. The approach estimates therelationship between energy consumption (dependentvariables) and factors influencing consumption. Therelationships are estimated by the least-square method
or time series methods. One of the options in thisframework is to aggregate the econometric approach,when consumption in different sectors (residential,
commercial, industrial, etc.) is calculated as a functionof weather, economic and other variables, and thenestimates are assembled using recent historical data.Integration of the econometric approach in to the end-use approach introduces behavioral components in tothe end-use equations [5].
The advantage of econometrics are that it providesdetailed information on future levels of electricity
demand, why future electricity demand increases, andhow electricity demand is affected by all the various
factors [6], [7], [29].
-
8/22/2019 IUST v7n4p249 En
3/11
Ghods & Kalantar: Different Methods of Long-Term Electric Load Demand Forecasting 251
A disadvantage of econometric forecasting is that inorder for an econometric forecast to be accurate, thechanges in electricity remain the same in the forecastperiod as in the past. This assumption, which is calledconstant elasticity, may be hard to justify especially
where very large electricity prices changes, makecustomers more sensitive to electricity prices.
2.4 Differences Between These Traditional
MentionedMethodAs mentioned in the trend analysis, just past changes
or movements in electricity demand and uses them topredict future changes in electricity demand, there isn'tany process on why those movements happened. In this
method, end users and their behavior aren't important.But in end use method, statistical information aboutcustomers along with dynamics of change is the basisfor the forecast.
In Economical methods, the results estimate therelationship between dependent variables and factors
influencing consumption. The relationships areestimated by the least-square method or time series
methods.In comparison, trend analysis can't be trustworthy; in
this method we need a wise and knowing judge torecognize unreal date and omit them from previousinformation.
Up to this part we describe the old methods for longterm load forecasting. They are also useful today. Butthe new following methods can use for their accuracy
and fast possessing system. Special the new methods are
used for different economical inputs in forecasting. Bythese new methods, we can have a model from the pastdata and correct its inaccurate date. After that we canpredict the following peak load.
3 Artificial Intelligence Based Methods
3.1 Artificial Neural Networks
Artificial neural networks (ANNs) have succeeded
in several power system problems, such as planning,control, analysis, protection, design, load forecasting,security analysis, and fault diagnosis. The last three arethe most popular. The ANNs ability in mappingcomplex non-linear relationships is responsible for the
growing number of its application to load forecasting[8], [9]. Most of the ANNs have been applied to short-time load forecasting. Only a few studies are carries out
for long-term load demand forecasting [10], [22], [24],[28].
In developing a long-term load forecast, thefollowing are some of the degrees of freedom whichmust be iterated upon with the objective to increase thepotential for an accurate load forecast: (1) fraction of
the database that will be used for training and testingpurpose, (2) transformations to be performed on thehistorical database, (3) ANNs architecturespecifications, (4) optimal stopping point during ANNs
training, and, (5) relative weights for use in forecastcombination [8].
The design of neural network architecture involvesdecision making on type, size and number of neuralbeing used [11].
The result of Output ANN is in (2).n
Y W Xi 1i i i
= = (2)
where i 1, 2, ..., n= , Xi is input, and Wi is weight ofnetwork and Yi is one of the ANN's outputs.
The first question to be asked is if an ANN can learn
to perform the desire application, and if so what wouldbe the most suitable form of the network. In this section,various aspects of ANNs are analyzed to determine asuitable model. These aspects include the networkarchitecture and method of training. There are threetypes which can be useful for long-term load demandforecasting: Recurrent neural network (RNN) for
forecasting the peak load, feed-forward backpropagation (FFBP) for forecasting the annual peak load[10] and radial basis function network.(RBFN) forfasting training and better following the peaks andvalleys [4].
1) Recurrent neural network: Recurrent neuralnetwork contains feedback connections, which enablethem to encode temporal context internally. Thisfeedback can be external or internal. RNN has be ability
to learn patterns from the past records and also togeneralize and project the future load patterns for anunseen data [10]. We have different types of RNNs,such as Jordan RNN, Elman RNN and others. Feedback
connections in these RNNs are different from networkto network. For instance, Jordan RNN has feedbackconnections from output to input while the Elman RNN
has feedback connections from its hidden layer neuronsback to its inputs. Additional neurons in input layer,which accept these feedback connections, are calledstate or context neurons. The role of context neurons inRNN is to get inputs from the upper layer, and afterprocessing send their outputs to the hidden layertogether with other plan units. In long-term loaddemand forecasting, there is strong relationship between
the present and next year loads. For this type ofproblem, Jordan's model of RNN proved to be suitable.
However, it should be noted that as the period of targetforecast loads becomes longer, the forecast errors mightincreases relatively [10]. This is why the feed-forwardback propagation is used for forecasting loads of longerthan 1 year. The Jordan RNN used in most of case studyis shown in Fig. 1.
2) Feed-forward back propagation: Feed-forwardback propagation is one of is one of the most widelyused neural network paradigms, which have beenapplied successfully in application studies. FFBP can beapplied to any problem that requires pattern mapping.Given an input pattern, the network produces anassociated output pattern. Its learning and update
procedure is intuitively appealing, because it is based on
-
8/22/2019 IUST v7n4p249 En
4/11
252
a relatively sboth a set o
response fowrong answ
error is less
network areusing such atoo. Firstlyknowledgethey have ha
recognize pLastly, oncepatterns, thaccomplisheadvantagesdesigned totraining patt
[10]. The F
Fig. 2.3) Redial
function netnetwork, whprocessing esymmetric [layer, hidde
each layer ashown in Fi
For a hihave the foll
A cand
vect A di
inpu
the s
A trsinglof toutp
funcstro
othefuncvect
Fig. 1 Jordan
imple conceppatterns to
each patterr, then the w
ned and as a
ore likely tonetwork cent, they autnabling thed seen. Seco
tterns, whichthey have beeir recognitid very quickfor FFBP, Bminimize therns and it ha
BP used in
basis functiork (RBFN)
ich has an intlements (patt4]. It consis
layer and o
e fully conn. 3.
den unit to bowing three c
nter, which i
which is ty
r from the instance meast vector is fro
tandard Eucli
ansfer functie variable, ane processin
ut of the di
tion is a Gauger value wh
word, thetion of onlyr and the sto
ecurrent Neu
t: the networe learned an
n. If the neeights are cor
result future
be correct.er on some oomatically
to recognizdly, they are
have been oen trained onon of simily [10]. TheP training ismean squarsupervised t
ost of case s
on network:in most gen
ernal represern units), w
ts of three ltput layer.
cted to the p
e radically syonstituents:
s a vector i
ically store
put layer to tre, to deter
m the center.
dean distance
n, which isd which dete
elementsstance functi
ssian functioen the distan
utput of ahe distanceed center.
al Networks.
ranian Journ
is suppliedd desired sys
twork givesrected so that
responses of
he advantagef their properteneralize tpatterns, w
robust enoug
scured by nothe initial selar patternse are twomathematicerror across
aining techni
tudy is show
A redial beral terms istation of hidich are radicayers; the ihe nodes wi
revious layer
mmetric, it
the input sp
in the we
e pattern unitine how far
Typically, thi
measure.
a function omines the ouy mappingon. A com
, which oute is small. In
attern unit ietween an i
al of Electric
ithem
thethe
the
s ofies,eirichto
ise.t of
isoreally
allque
in
asisanydenallyputhin
, as
ust
ace
ght
ans is
f aputtheon
utsthe
s aput
Fi
nefu
paintrefoanfreth
foex
res
wfonefo
al & Electron
Fig. 2 Feed-networks.
. 3 Redial Basi
The key toworks is toction [10].
This section
ket in powroduced toeived wideecasting met
flexible tooquency com
characterist
ecasting accension of wa
olution [12],Several pa
velet analysiecasting acwork mustecast well:
Selectforecastimportano genimporta
ic Engineeri
orward back
s Function Net
a successfulfind suitable
3.2 Wavelet
investigates
r load forecpower loadttention. Coods, wavele
l to decompoonents, makics of each
racy. Wavelelet analysis
[23].ers investigs in poweruracy washave the fo
proper wang. The selet for wavelral rule tillt thing of p
g, Vol. 7, No.
propagation n
work.
implementacenters for
Networks
the applicati
asting. Wavforecastingparing to trtheory provi
se load dataing it possibcomponent
et packet aand it has be
ated the aload forecasimproved.
llowing mea
elet functiotion of wavet applicationnow .Generawer load for
4, Dec. 2011
eural
ion of thesethe Gaussian
n of wavelet
let theory isrecently andditional loaddes powerfulinto differene to analyzeand improve
alysis is theter frequency
plication oing, and thehe wavelet
surements to
n for loadet function is, but there islly, the mostecasting is to
-
8/22/2019 IUST v7n4p249 En
5/11
Ghods & Ka
imprdisto
recoothe
Amo
bioranddistorecoof
fore
Avotransdata
probdeco
excefore
loadcoefforegive
foredeal
simpsevesymAlthforsym
bior
The struFig. 4.
This strunetwork.
The mosspanned inp
than multiadvantages
resolution amuch poweraccuracy [13
A newnetwork mo
neural callovercomes trules. It can
and speed [2Another
model for lfunction isthe shortcoseen that tforecast acc
Managintask. The m
lantar: Differ
ove forecastirtion during t
struction prapplication
ng various
hogonal wahas linear prtion durinstruction pr
biorghogonal
asting is suit
d borderform. Becauin wavelet
lem arisesmposition pr
pt the methoasting appli
data, borderficients of theasting basedaccurate fore
ast without twith the pr
le and effecral types of
etrical extough it is ha
selectingetrical bor
hogonal wav
cture of the
cture is very
advantage fts although t
layer neuro apply to l
alysis capabin function
].ethod is thedel for long
unction is be shortcomiimprove eff
6].new methodng term loaasis of nonliing of singlis methodracy and spe
3.3 Gene
g electricalst important
ent Methods
g accuracy,he wavelet d
cess, whichs such as
wavelet
elet functioase, so it w
the deccess. Theref
wavelet f
ble [12].
distortionse of the lipplications,
duringcess, but use
d proposed iation with
distortion mlatest data ison these coecasting resul
he help of thblem, borde
tive solutionborder ext
ension andrd to give g
proper exter extension
elet which is
avelet netw
similar to
ctor of wavele accuracy
l networks.ng-term for
lity of wavepproximatio
uzzy rules aterm load
sis of nonlig of single tctively the
is the waveleforecasting.
near wavelettrain set of
can improved [30].
ic Algorith
nergy supplart of electri
f Long-Term
o it requirescomposition
is different ffault detect
functions,
is symmetrnt bring simpositionre, the selec
or power l
during wavited numberorder distor
the wavrs often igno
n [13]. For linite number
eans the wavdistorted andfficients coul. It is too ba
e latest data.r extension i
, and thereension, such
zero paddneral guidelinsion metis suitable
symmetrical.
ork is show
ulti layer ne
et network isf model is be
It hascast. The m
let functionsto obtain be
d wavelet neforecasting.
ear waveletsrain set of fuorecast accur
t neural netThe neural
s. We overcANN. It caneffectively
s
y is a comc utility reso
Electric Loa
lessand
omon.
the
icalnalandionoad
eletof
ion
elete it
oadof
eletthentto
Tos a
areas
ng.nesod,for
in
ral
nottter
orelti-
hastter
ralhe
. Itzzyacy
orkcallmebethe
lexrce
plre
acin
da
foewi
m
fothwraevbeflo
ec
de
fomnohige
opth
inalthof
tecpr
ge
strexc
asen
na
rosu
Fi
Demand Fo
nning is foreional or n
ieved byormation, su
a. Genetic
ecast long tepirical resulth a low errorGenetic Algch attention
various probmechanism
ich combinedom andluation of thn successfull
w problems
nomic disp
and forecasConsidering
ecasting aredel that is ab-linear or ihly robustneeric neural
nions of expt a compreh
luence the porithm is ablneural netw
which are ch
Genetic alghnique. Morcedures bas
etics.They combi
ategy withhange amromosomesThis techniq
a robust optiineering, sci
GAs accomely uncert
ustness. Somarized as
. 4 Schematic
ecasting
casting of thetional servi
constructinh as climate
rogramming
rm electricals demonstratrate [19].orithms (Gas robust st
lems. This clf natural sel
s the notionet structur
e points in ty applied in, fault det
tch, power
ing [14].the featur
complicated,le to adapt toprecise rul
ss. For avoidnetwork itse
erts are intrensive effec
wer load cato search pr
rk is able tosen together
orithms arespecifically,
ed upon th
ne a Darwirandom, y
ong a p.ue has gaine
ization toolnce, econom
odate all thinty, impre
e of theollowing [14
f Wavelet Net
future loade area. Thi
modelsand previous
approach is
power conse successful
s) have recechastic sear
ass of methoction and natof survivald, searche search spaarious areas
ction, stabil
system cont
es of longa generic nand learn frs, so it ising the inflelf, many exp
duced durinof differen
be reflectedcisely at glo
fit well at loc[20].
a numericalthey are pare mechanic
ian survivalet structuredopulation
popularity i
or a variety oics, finance,
facets of soision, non-l
ttractive fea:
work.
253
emand in thes is usually
on relativeload demand
proposed to
mption. Theload forecas
ntly receivedh algorithms
s is based onural genetics,f the fittest,
and parallele. GAs havesuch as, loadity analysis,
ol and load
term loadural networkm amount omodel with
ibility of theeriences and
the use, sofactors that
. The generical scope, andl scope, both
optimizationmeter search
of natural
-of-the-fittestinformation
f artificial
recent years
f problems intc.
ft computing,nearity, and
ures can be
-
8/22/2019 IUST v7n4p249 En
6/11
254 Iranian Journal of Electrical & Electronic Engineering, Vol. 7, No. 4, Dec. 2011
Learning: GAs are the best known and widelyused global search techniques with an ability to
explore and exploit a given operating spaceusing available performance (or learning)
measures.
Generic Code Structure: GA operates on anencoded parameter string and not directly on the
parameters. This enables the user to treat anyaspect of the problem as an optimizablevariable.
Optimality of the Solutions: In many problems,there is no guarantee of smoothness. Traditionalsearch techniques often fail miserably on such
search spaces. GA is known to be capable offinding near optimal solutions in complexsearch spaces.
Advanced Operators: This includes techniquessuch as nicking (for discovering multiple
solutions), combinations of Neural, Fuzzy, andchaos theory, and multiple-objectiveoptimization.
The GAs approach presented in this work isemployed to find the optimum values of the state vectorX that minimizes the absolute summation of the
forecasting error r(t). In order to emphasize the beststring and speed up convergence of the iterationprocedure, fitness is normalized into range between 0and 1. The fitness function (ff) adopted is [14]:
1ff
m1 k r(t)k 1
=+ =
(3)
where k is a scaling constant (for example, k=0.0001)
Like other stochastic methods, the GA has a number ofparameters that must be selected, size of population,probability of crossover and probability of mutation. r(t) is the error vector associated. GA tries to keep the r(t) in the allowed limitation. If the r (t) is kept in theallowed limitation, the fitness function has the bestvalue for load demand forecasting [14].
Forecasting results using GA were found to be thebest. This indicates that the GA approaches is quitepromising and deserves serious attention because of itsrobustness and suitability for parallel implementation[14].
With r (t), we can calculate the load demand
forecasting by the following equation:
inP(t) a a t r(t)
i 10 i= + + = (4)
where P(t) is the peak load demand at time t, a0
,ia
are
the regression coefficients relating the load demand P(t)to the time t. r(t) is the residual load at year (t).
3.4 Support Vector Machine (SVM)
SVM (Support Vector Machine) is a usefultechnique for data classification. Even though peopleconsider that it is easier to use than Neural Networks,
however, users who are not familiar with SVM often getunsatisfactory results at first [15].
The support vector machines (SVMs) are based onthe principle of structural risk minimization (SRM)rather than the principle of empirical risk minimization,
which conducted by most of traditional neural networkmodels. SVMs have been extended to solve nonlinear
regression estimation problems [16]. Recurrent neuralnetwork (RNN) is one kind of SVM which is based onthe main concept in which every unit is considered as anoutput of the network and the provision of adjustedinformation as input in a training process. RNNs areextensively applied in long-term load time seriesforecasting and can be classified in three types, Jordan
networks, Elman networks, and Williams and Zipsernetworks. Both Jordan and Elman networks use mainlypast information to capture detailed information.Williams and Zipser networks take much more
information from the hidden layer and back intothemselves. Therefore, Williams and Zipser networks
are sensitive when models are implemented. Jordan andElman networks are suited to time series forecasting.
Traditionally, RNNs are trained by back-propagationalgorithms. SVMs with genetic algorithms are used to
determine the weights between nodes [16].The basic concept of the SVM regression is to map
nonlinearly the original data x into a higher dimensionalfeature space. Hence, given a set of data
NG {(x , a )}
i i i 1= = (where xi is the input vector, ia the
actual value and N is the total number of data patterns),
the SVM regression function is:(x )f g(x) w b
i i= = + (5)
where (x )i
is the feature of inputs, and both iw and b
are coefficients. The coefficients (iw and b) are
estimated by minimizing the following regularized riskfunction:
( )N1 1 2
r(C) C a , f || ||i ii 1N 2
= + =
(6)
where,
0 if a f
(a ,f) a f otherwise
=
(7)and C and are prescribed parameters. In (6), (a, f) iscalled the -insensitive loss function. The loss equals
zero if the forecasted value is within the -tube (7). Thesecond term, 1/2 ||w||, measures the flatness of thefunction. Therefore, C is considered to specify the trade-off between the empirical risk and the model flatness.Both C and are user-determined parameters.
The architecture of SVMs with genetic algorithm
(SVMG) is shown in Fig. 5.The superior performance of the RSVMG model has
several causes. First, the RSVMG model has nonlinear
-
8/22/2019 IUST v7n4p249 En
7/11
Ghods & Kalantar: Different Methods of Long-Term Electric Load Demand Forecasting 255
Fig. 5 Architecture of SVMG.
mapping capabilities and thus can more easily captureelectricity load data patterns than can the ANN andregression models. Second, improper determining ofthese three parameters will cause either over-fitting or
under-fitting of a SVM model. In this section, the Gascan determine suitable parameters to forecast electricity
load. Third, the RSVMG model performs structural riskminimization rather than minimizing the training errors.Minimizing the upper bound on the generalization errorimproves the generalization performance compared tothe ANN and regression models.
3.5 Fuzzy Logic Model
Fuzzy control systems are rule-based systems in
which a set of so-called fizzy rules represents a controldecision mechanism to adjust the effects of certainstimulus. The aim of fuzzy control systems is normally
to replace a skilled human operator with a fuzzy rule-based system. The fuzzy logic model provides analgorithm, which can convert the linguistic strategybased on expert knowledge into an automatic strategy.
Fig. 6 represents the basic configuration of a fuzzy logicsystem, which consists of a fuzzification, knowledgebase, fuzzy interface and a defuzzification (IO). Thefuzzy logic method is applied for scoring. Theapplication of fuzzy rules will improve the modelaccuracy by avoiding arbitrariness for the purpose of the
stud. The fuzzy rule base is composed of some rulesgenerated from the analysis of the historical load data
[16], [21].
One of the applications of the fuzzy rules is tocombine them with neural network to train ANN andhave a better load demand forecasting. The trainingpatterns for the ANN models were collected from thehistorical load data. The number of training cycles hasbeen determined through a trial process, to avoidovertraining [16].
The benefit of the proposed hybrid structure was toutilize the advantages of both, i.e., the generalizationcapability of ANN and the ability of fuzzy inference forhandling and formalizing the experience and knowledgeof the forecasters. It has been demonstrated that themethod give relatively accurate load forecasts for the
actual data. The test results showed that this method of
Fig. 6 Block diagram of the fuzzy logic system.
Fig. 7 Structure of ANN and Fuzzy based used.
forecasting could provide a considerable improvementof the forecasting accuracy. This indicates that the fuzzyrules and the training patterns for the ANN is quitepromising and deserve serious attention of its robustnessand suitability for implementation. It can be concluded
that the outcome of the study clearly indicates that theproposed composite model can be used as an attractiveand effective means for the industrial load forecasting.The improvement of forecast accuracy and theadaptation to the change of customers would fulfill theforecasting needs [16]. Fig. 7 shows the structure of
ANN and Fuzzy based used in forecasting.We can also combine two different methods to
achieve better result. These two methods can be ANNand Fuzzy control. Long term load forecasting of powersystem is affected by various uncertain factors. Usingclustering method numerous relative factors can besynthesized for the forecasting model so that theaccuracy of the load forecasting would be improved
significantly. A clustering neural network consisting oflogic operators is quoted, which can be used in longterm load forecasting Applying logic operators and inthe fuzzy theory, the algorithm speed of the clusteringnetwork will be increased. Although competitivelearning algorithm is used here for the network, it solvesthe dead unit problem and gives more room to select theinitial values of the clustering center in the clusteringanalysis of the history data. The proposed model
considers the influences of both history and futureuncertain factors [25], [27], [31], [32].
Previous data
Fuzzification
Combination of the
results
Forecasted of selected
load
Selection of ANN
Load data
Fuzzy Interface
Defuzzification
Load forecasted
Forecasted load
-
8/22/2019 IUST v7n4p249 En
8/11
256 Iranian Journal of Electrical & Electronic Engineering, Vol. 7, No. 4, Dec. 2011
3.6 Expert System
The confidence levels associated with classicalforecasting techniques, when applied to forecastingproblem in mature and stable utilities are unlikely to besimilar to those of dynamic and fast growing utilities.
This is attributed to the differences in the nature ofgrowth, socio-economic conditions, occurrence of
special events, extreme climatic conditions, and thecompetition in generation due to the deregulation of theelectricity sector with possible changes in tariffstructures. Under such conditions, these forecastingtechniques are insufficient to establish demand forecastfor long-term load demand. Consequently, this caserequires separate consideration either by pursuing the
search for more improvement in the existing forecastingtechniques or establishing another approach to addressthe forecasting problem of such systems [17].
In this section, the classical forecasting methods are
firstly applied to obtain the long-term load demandforecasts, for a typical fast growing utility as well as
normal developing system [17].A poor performance is observed when such methods
are applied to fast developing system, whereas most ofthese models are valid when used to produce the
forecasts of the normal developing system.Consequently, an extended logistic model is developedto reflect the critical forecasting problem in fast growingareas. Although the developed model gives an accurateload demand forecast compared with the classicalmodels, it is hardly difficult to dependent on single
method for producing the demand of such fast growing
and dynamic system [17]. This is because severalimportant factors related to the cyclic and dynamicevents that contribute significantly to the system loadare difficult to involve it into the existing forecastingmodels.
Thus, there is a need to develop a computational tool
which allows one to store the knowledge associatedwith this problem along with the mathematical models
to support the choice of the most suitable loadforecasting model, for long-term power systemplanning. Therefore, the implementation of long-termforecasting strategies using a knowledge-based expertsystem (ES) is then presented in this section. In the
expert system, key system variables which have majoreffects on system load are identified based on pastplanners experiences. A set of decision rules relating
these variables are then established and stored in theknowledge base to select the recommended forecasting[17].
The main components of the proposed expert systemare shown in Fig. 8.
With the knowledge base at hand (rules and facts),
an inference engine can be used to search through thisknowledge base according to the solution strategy. Thedetailed procedures of the solution strategy to ascertainthe accuracy and credibility of selecting forecasting
method. In addition to knowledge base, inference
engine, solution strategy, a user interface is alsodeveloped in the expert system to facilitate thenavigation between the expert system and the user.
The variables of the formulated problem can begrouped into Static and Dynamic Facts as follows:
Static Facts: This kind of knowledge is developedbefore starting the planning process. A sample of these
facts is: system conditions to identify the currentsituation of the system, i.e., mature, or underdeveloping, isolated or interconnected with othersystem, etc.
Forecasting horizon to define the load forecastingperiod, i.e., long-term.
Load pattern to describe the load behavior(stable, or dynamic pattern, cyclic, or seasonalpattern, or combination of all, time of systempeak, load types, etc.)
Historical peak load to indicate annual andseasonal growth, coincidence factor, area peak,etc.
Historical energy to describe the informationrelated to number of consumers of each sector,consumption rate, tariff rate, etc.
Major factors affecting the system peak, i.e.,weather variables, economic variables,demographic variables, special event,suppressed demand, bulk loads to be connectedinto the network, coincidence factor of thesystem peak, etc.
Dynamic Facts: These facts are developed andautomatically updated during the inference process to
represent the planning attributes needed for evaluating adecision making process. Samples of these facts includethe following:
Fig. 8 Structure of expert system for long-term loadforecasting.
-
8/22/2019 IUST v7n4p249 En
9/11
Ghods & Kalantar: Different Methods of Long-Term Electric Load Demand Forecasting 257
load and energy attribute for the estimated loadand energy forecast;
System losses attribute for the estimated systemlosses;
Error attribute related to the forecasting model.In this section, a long-term load forecasting is
developed and classified according to the forecastingproblem using a knowledge-based expert system (ES).The proposed methodology is applied successfully toforecast yearly peak load for normal and fast developingpower systems. Since the expert system is very flexible
in updating the forecasting methods and heuristic rules,it is expected that the expert system can serve as a
valuable assistant to system planners in performing theirannual load forecasting duties. Finally, it can beexpected to serve as a valuable assistant also for trainingpurposes [17].
4 Contrasting New Forecasting MethodsRecurrent neural network (RNN) has be ability to
learn patterns from the past records and also to
generalize and project the future load patterns for anunseen data. In this type, some additional neurons areavailable. Additional neurons in input layer, whichaccept these feedback connections, are called state orcontext neurons. The role of context neurons in RNN isto get inputs from the upper layer, and after processing
send their outputs to the hidden layer together withother plan units.
In the other neural network method, feed-forwardback propagation an input pattern is given, the networkproduces an associated output pattern. Its learning andupdate procedure is intuitively appealing, because it isbased on a relatively simple concept: the network is
supplied with both a set of patterns to be learned anddesired system response for each pattern. This method is
much better than the RNN method. Because if thenetwork gives the wrong answer, then the weights arecorrected so that the error is lessened and as a resultfuture responses of the network are more likely to becorrect. It can have a trustworthy result.
Another method for long term load forecasting is
Wavelet Network. The most advantage factor of waveletnetwork is not spanned inputs although the accuracy of
model is better than multi layer neural networks. This isone reason which can be ended up in this choice. It hasmore advantages to apply to long-term forecast. Themulti-resolution analysis capability of wavelet functionshas much power in function approximation to obtainbetter accuracy. This accuracy can make a better resultin future forecasting.
For one method, we can call Genetic Algorithm forlong-term load demand forecasting, Genetic algorithmsare a numerical optimization technique. Morespecifically, they are parameter search procedures basedupon the mechanics of natural genetics. Forecastingresults using GA were found to be the best. Thisindicates that the GA approaches is quite promising and
deserves serious attention because of its robustness andsuitability for parallel implementation.
Other most commonly method is SVM. This methodis much more comparable with ANN. First, the RSVMGmodel has nonlinear mapping capabilities and thus can
more easily capture electricity load data patterns thancan the ANN and regression models. Second, improper
determining of these three parameters will cause eitherover-fitting or under-fitting of a SVM model. Third, theRSVMG model performs structural risk minimizationrather than minimizing the training errors.
Fuzzy system as another method is normally toreplace a skilled human operator with a fuzzy rule-basedsystem. One of the applications of the fuzzy rules is to
combine them with neural network to train ANN andhave a better load demand forecasting.
In expert system, we can use traditional methods toforecast the peak load forecasting. The expert system is
very flexible in updating the forecasting methods andheuristic rules, it is expected that the expert system canserve as a valuable assistant to system planners inperforming their annual load forecasting duties.
5 Conclusions
Load forecasting plays a dominant part in theeconomic optimization and secure operation of electricpower systems.
Long-term load forecasting represents the first stepin developing future generation, transmission, anddistribution facilities. Any substantial deviation in the
forecast, particularly under the new market structure,
will result in either overbuilding of supply facilities, orcurtailment of customer demand. The confidence levelsassociated with classical forecasting techniques, whenapplied to forecasting problem in mature and stableutilities are unlikely to be similar to those of dynamicand fast growing utilities. This is attributed to the
differences in the nature of growth, socio-economicconditions, occurrence of special events, extremeclimatic conditions, and the competition in generationdue to the deregulation of the electricity sector withpossible changes in tariff structures. Under suchconditions, these forecasting techniques are insufficientto establish demand forecast for long-term power
system planning. Consequently, this case requiresseparate consideration either by pursuing the search formore improvement in the existing forecasting
techniques or establishing another approach to addressthe forecasting problem of such systems.
Different methods of long-term load demandforecasting are defined in this paper. All of thesemethods can forecast the load of the power system, butthe amount of previous data and such variables which
they need to forecast, make them different in accuracyfrom area to area.
Finally, for long-term load forecasting, we shouldknow the power system in details, and after that we can
select the best method for the specified power system.
-
8/22/2019 IUST v7n4p249 En
10/11
258 Iranian Journal of Electrical & Electronic Engineering, Vol. 7, No. 4, Dec. 2011
Sometimes we can combine different methods andcompare the accuracy of them together.
Traditional methods, such as time series, regressionmodels and etc. are used in most of the countries,because of their reliable result.
Neural networks can solve nonlinear problems, andbecause of nonlinear behavior of load, so they can be
useful for long-term load forecasting.Genetic algorithm can forecast long-term load
forecasting, when we have a lot amount of differentvariables and we want to find the best solution to followthe future load. Also it can be useful to estimate thesupport vector machine parameters.
Wavelet can estimate peak and valley of load
behavior better than Furious series. It can combine withANN have a better forecast.
References
[1] Al Mamun M., and Negasaka K., Artificialneural networks applied to long-term electricity
demand forecasting, Proceedings of the FourthInternational Conference on Hybrid Intelligent
Systems (HIS'04), pp. 204-209, Dec. 2004.[2] Dang Khoa T. Q. and Oanh P. T., Application of
Elman and neural wavelet network to long-termload forecasting, ISEE Journal, track 3, sec. B,No. 20, pp. 1-6, 2005.
[3] Al-Hamidi H. M. and Soliman S. A., Long-term/mid-term electric load forecasting based onshort-term correlation and annual growth,
Electric Power System Research (Elsevier), Vol.
74, No. 3, pp. 353-361, June 2005.[4] Negasaka K. and Al Mamun M., Long-term
peak demand prediction of 9 Japanese powerutilities using radial basis function networks,IEEE Power Eengineering Society GeneralMeeting, Vol. 1, pp. 315-322, 6-10 June 2004.
[5] Engineering and design hydropower proponent,Load forecasting methods, in EM 110-2-1701,
appendix B, Dec 1985.[6] Genethliou D. and Feinberg E. A., Load
forecasting, Applied mathematics for restructuredelectric power system: optimization, control andcomputational intelligence (J. H. Chow, F.F. Wu,
and J. J. Momoh, eds.), chapter 12, pp. 269-285,2005.
[7] Fu C. W. and Nguyen T. T., Models for long-term energy forecasting, IEEE PowerEngineering Society General Meeting, Vol. 1, pp.235-239, 13-17 July 2003.
[8] Taradar Heque M. and Kashtiban A. M.,Application of neural networks in powersystems; A review, Transaction of Engineering,
Computing and Technology, Vol. 6, No. 1, ISSN1305-5313, pp. 53-57, June 2005.
[9] Atiya A. F., Development of an intelligent long-term electric load forecasting system,
Proceedings of the International Conference,ISAP apos, pp. 288-292, 1996.
[10] Kermanshahi B. S., and Iwamiya H., Up to year2020 load forecasting using neural nets, ElectricPower System Research (Elsevier), Vol. 24, No.
9, pp. 789-797, 2002.[11] Phimphachanh S., Chamnongthai K., Kumhom
P., and Sangswang A., Using neural network forlong term peak load forecasting in Vientianemunicipality, IEEE Region 10 Conference,TENCON 2004, Vol. 3, pp. 319-322, 2004.
[12] Khoa T. Q. D., Phuong L. M., Binh P. T. T. andLien N. T. H., Application of wavelet and neuralnetwork to long-term load forecasting,
International Conference on Power SystemTechnology (POWERCON 2004), pp. 840-844,Singapore, 21-24 November 2004.
[13] Khoa T. Q. D., Phuong L. M., Binh P. T. T., LienN. T. H., Power load forecasting algorithmbased on wavelet packet analysis, International
Conference on Power System Technology(POWERCON 2004), pp. 987-990, Singapore, 21-
24 November 2004.[14] EL_Naggar K. M. and AL-Rumaih K. A.,
Electric load forecasting using genetic basedalgorithm, optimal filter estimator and least errorsquare technique: Comparative study,Transaction of Engineering, Computing andTechnology, Vol. 6, pp. 138-142, ISSN 1305-5313, June 2005.
[15] Pai P.-F., and Hong W.-C., Forecasting regionalelectricity load based on recurrent support vectormachines with genetic algorithms, ElectricPower System Research (Elsevier), Vol. 74, No.3, pp. 417-425, 2005.
[16] Faraht M. A., Long-term industrial loadforecasting and planning using neural networks
technique and fuzzy interface method, 39thInternational Universities Power EngineeringConference, UPEC 2004, Vol. 1, pp. 368-372,2004.
[17] Kandil M. S., El-Debeiky S. M. and Hasanien N.E., The implementation of long-term forecastingstrategies using a knowledge-based expert
system: part-II,Electric Power System Research(Elsevier), Vol. 58, No. 1, pp. 19-25, 2001.
[18] Carmona D., Jaramillo M. A., Gonzalez E. andAlvarez A. J., Electric energy demandforecasting with neural networks, IEEE, 28thAnnual Conference of the Industrial ElectronicsSociety, Vol .3, pp. 1860-1865, 2002.
[19] Karabuluta K., Alkanb A. and Yilmaz A. S.,Long term energy consumption forecasting
genetic programming, Association for ScientificResearch, Mathematical And ComputationalApplications, Vol. 13. No. 2, pp. 71-80, 2008.
[20] Yingling, Hongsong S., Yawei Y. and NanshengD., Research on long term load forecasting
-
8/22/2019 IUST v7n4p249 En
11/11
Ghods & Ka
basedIEEE,
[21] Zhangterm
rules,
(ICIMConfe
[22] GhanbSabagregres
Iran'sENG.
[23] Ji Z.,Wavelto FLoadScienc
Works
[24] HobbsTermNeuraSysteISAPObjec
[25] Yuefuzzyfor mi
2007.967, 2
[26] ZhangWavel
term2010446, 2
[27] GhanbMohaSeriesNeuroPOW
[28] ShrivApproBased
ICPS[29] Yingy
Principower
lantar: Differ
on ImprovePCAIIA, pp.
Q. and Liuelectric load
Information
E), 2010 Tence, pp. 461ari A., Nagian M., A
sion approac
annual elecConference,Zhang P. a
et Neutral Nrecast Mid-
, Educatione, 2009. E
hop on , Vol.
N. J., KimLoad Forec
Networks Applicatio2007. InternIdentifier, p., Zhang Y.,
logic clusterddle and lon
IEEE Intern007.
Q. and Liet Neural N
lectric Loadecond Inter
010.
arian M.,madi A. andRegression
-Fuzzy TechR ENG., pp.stava V. an
ach of InputLoad Fore
, pp. 1-5, 200ing L. and
al Componload forec
ent Methods
d Genetic N80-84, 2008.
., Researchforecasting
Management
e 2nd IE-463, 2010.avi A., Gh
tificial Neures compariso
ricity load,p 675-679, 2d Zhao Z.,twork and RLong-TermTechnology
CS '09. Fi
1, pp 1104-1
. H. and Leasting UsinArchitecture
ns to Powertional Confe. 1-7, 2007.Xie H. and
ing neural nterm load fo
tional Con
T., A Ftwork Meth
Forecastingational Con
avehnia F.,Keivani H.,to Load F
niques, I769-773, 200d Misra R.
Variable Seasting , I
8.ongxiao N.,
ent Regressisting for m
f Long-Term
eural Netwo
on the mid-lbased on fu
and Enginee
E Internati
aderi S. F.l Networksn for forecas
IEEE PO009.Application
ough Set TheElectric Poand Comp
st Internati
108, 2009.
e K. Y. , LoSystem T, Intelli
Systems, 2rence on Di
Zhong Q.,
etwork approrecasting,
rence, pp. 9
zzy Rulesd for Mid-lo
, ICCNT, Ierence, pp 4Askari M.Applying Tirecasting UEE Confere7.
B., A N
ection for AEE Confere
Applicatio
on Analysisdium and l
Electric Loa
k,
ongzzy
ing
nal
andanding
ER
ofory
erternal
ng-ypeent07.ital
he
achSIS
63-
andng-
EE42-
R.,e-
ingnce
vel
NNce,
of
inong
[3
[3
[3
[3
int
aropainc
Ge
sta
Demand Fo
term, IEV3-203, 2
] Zhang Q.Term Lo
Network
217-220,] Maralloo
Kalhor A.via a nCSICC20
] DalvandLong-tergrid usinIEEE Co
] Ghods L.Load DeFunction
Electrical
Vol. 6, N
rested in pow
und 30 Journaers at Internalude wind a
nerations, po
ility and opti
ecasting
EE Conferen010.
. and Liu T.d Forecastin
, IEEE Con
2010.M. N., Kou
., Long termurofuzzy m09, pp 35-40,
M. M., Azam load fore
fuzzy andference 2008
and Kalantand Forecas
Neural Net
& Electro
. 3, pp. 175-
Ladan
Iran indegree
Tehran.engineeri
universit(IUST) ineural n
forecastr installation e
Mohsen
in Iran.Indian I
Delhi, InAssociate
of ElectUniversit
Tehran.Center ofAutomati
l publicationstional Conferd solar po
er system dy
ization.
ce, ICACTE,
, ResearchBase on W
erence, ICC
shki A. R.,electrical lo
odel, IEEE2009.
i S. and Tcasting of Irtificial neur, pp 1-4, 2008ar M., Loning by Usingorks, Irania
ic Engineer
82, 2010.
hods was b980. She rec
from Azad
She graduatng in M.S
of science a2008. She i
etwork and i
he future factspecially in ind
alantar, was
e received hstitute of Tec
ia in 1991. HeProfessor in t
rical Engineeof Science a
e is also founExcellence for
on and Oper
and has presences. His fieer generatio
namics and c
259
pp. V3-201-
on Mid-longavelet Neural
A 2010, pp,
ucas C. andd forecastingConference,
rimoradi H.,anian poweal networks,.g-Term PeakRadial Basis
n Journal o
ing (IJEEE),
rn in Tehran,ived her B.S.University o
d in powefrom Ira
nd technologs interested its abilities to
rs. She is alsoustries.
born on 1961
is Ph.D. frohnology, Ne
is currently ahe Departmen
ring at Irad Technology,
ing member oPower Systetion. He has
ted about 100ds of interess, Distribute
ontrol, syste