A Computing Model of Artificial Intelligent Approaches to Mid-term Load Forecasting : a state-of-the-art- survey for the researcher

IACSIT International Journal of Engineering and Technology Vol. 2, No.1, February, 2010
ISSN: 1793-8236
A Computing Model of Artificial Intelligent
Approaches to Mid-term Load Forecasting
: a state-of-the-art- survey for the researcher

Pituk Bunnoon, Kusumal Chalermyanont, and Chusak Limsakul


high, the exceeding investment is obtained and it will push
Abstract—This article presents the review of the
this expenses to consumers. However, if the forecasted
computing models applied for solving problems of mid-
values are too low, the inadequate investment is occurred
term load forecasting. The load forecasting results can
and it will cause electricity deficiency in the country.
be used in electricity generation such as energy
The forecasted electricity demands are defined as two
reservation and maintenance scheduling. Principle,
values: the peak value (Maximum load demand) and energy
1
strategy and results of short term, midterm, and long
value (Electric energy demand).
term load forecasting using statistic methods and
The peak value is used in planning for new electricity
artificial intelligence technology (AI) are summaried,
plants while the energy value is used in planning for fuel
Which, comparison between each method and the
providing. In the past, each electricity authority used the
articles have difference feature input and strategy. The
different methods to find these values.
last, will get the idea or literature review conclusion to
Forecasting of the energy value
solve the problem of mid term load forecasting (MTLF).
- The MEA and PEA uses the econometric model with

Error Correction Model of Engle-Granger method or the
Index
Terms-Artificial
Intelligent,
Mid-term
load,
econometric method with auto-Regressive Distributed Lag
Forecasting, state-of-the-arts.
(ARDL) for monthly forecasting. The forecasting variables
effected to electric load demands are electricity bill rate,

GDP and temperature. Since, there are no monthly GDP
data, they used the money quantity that circulate in people
I. INTRODUCTION
hand and the deposit reserve call of people in bank system
The electricity is the necessity in the daily life and it is
instead adding with the specific losses in the distribution
one of the main driving factors for country economic. In
systems. The loss values in the distribution systems of the
order to provide sufficient electricity and make the
MEA and PEA are respectively 3.64% and 5.20% of the
economic grown continuously, the load forecasting is
electricity demand taking from EGAT.
required for the related electricity producers. Since, the
- Direct customers of EGAT use the direct inquires
construction of a power plant must take 5-10 years from
method from electricity consumers.
planning,
designing,
environmental
admitting
to
- The EGAT uses combining energy values from the
constructing step and there are few electric networks of
MEA, the PEA and direct customers adding with the loss
Thailand and neighbor countries, the midterm load
values in the generation system and transmission system to
forecasting (MTLF) and the long term load forecasting
be the energy value of the system. The specific loss is about
(LTFL) are very important for building up the energy
5.10 % of energy value of the system.
stability in Thailand [1,2].
Electricity load forecasting is not only significant for
investment planning of three electricity authorities
(Electricity Generating Authority of Thailand (EGAT),
Metropolitan Electricity (MEA), and the Provincial
Electricity Authority (PEA)) but also it is useful for
estimating the financial statement of three electricity
institutes. Figure1 shows the peak load profiles of the MEA
and PEA classified by types of electricity consumers. The
accuracy forecasted values make the proper investment for
three electricity authorities. If the forecasted values are too

Figure.1: The use of Load Profile to find peak values of the MEA and the

PEA[2].
Manuscript received February 22, 2010.

P. Bunnoon is with Department of Electrical Engineering, Prince of
Forecasting of the peak value
Songkla University,Thailand (e-mail:add2002k@hotmail.com) Ph.D. Prog.
K. Chalermyanont is with Department of Electrical Engineering, Prince

of Songkla University, Thailand (e-mail: kusumal.c@psu.ac.th).
C. Limsakul is with Department of Electrical Engineering, Prince of

Songkla University, Thailand (e-mail: chusak.l@psu.ac.th).
94

---

IACSIT International Journal of Engineering and Technology Vol. 2, No.1, February, 2010
ISSN: 1793-8236
- The MEA and PEA use the character of load profile of
each customer to calculate energy value and adjust this
value to be equal to the forecasted energy value of each
type of customers. The over all peak value can be obtained
by adding every load profiles of all customers as illustrated
in Figure1.
- The direct customers of the EGAT use the principle of

Because of the difference of time period, forecasted
load profile by adjusting the load profile of each customer
values and aims of each load forecasting type, researchers in
to be equal to the forecasted value and adding all load
the past proposed many different algorithms and methods in
profiles to get the over all peak value.
order to obtain the precise load forecasting values. Next,
- The EGAT adds all the load profile from The MEA and
relative papers and research topics of each load forecasting
PEA as well as the direct customers to find the peak value.
type are briefly concluded.
At the same time, this method can determine the peak value
In 1987, [9] described about short-term load forecasting
of the MEA, PEA and the direct customers in the same
survey and comparing load forecasting in short-term,, mid-
system (Coincident Peak) [2].
term and long-term. In this paper, each research article has
Beside the mentioned methods, the expert systems
used differential techniques for determining the accurate
[3,4,5,6,7] such as the artificial intelligent (AI) are
output value. In [10-16], neural network for short-term load
frequently used for the system that required training and
forecasting are used based on historical load and
making decision based on the massive data. In the past,
temperature input data. Moreover, some paper use
many artificial intelligence (AI) and Expert system (Es)
additional input data from day types, humidity, wind speeds
methods such as artificial neural network (ANN), Fuzzy
and seasons. This method is performed in compared with
logic (Fs) and Genetic algorithm (GA) are proposed for the
conventional method. Training network is achieved by
electricity load forecasting in short-term, mid-term and even
supervise learning and back propagation algorithm. Another
long-term forecasting.
technique for short-term load forecasting is using fuzzy
This paper presents a survey for a state of the art of load
logic and neural network [17]. This paper presents that the
forecasting
methods including input classifications,
neuro-fuzzy method that gives more accuracy results
algorithm approaches and output determinations. The
compared to one of the neural network method. In [18-19],
reviewed papers are covered in short-term, mid-term and
types of input data using in fuzzy logic and neural network
long-term load forecasting. However, in last few sections,
algorithms are historical load and weather. The case study is
there will be emphasized on mid-term load forecasting.
Electric company in China (Hang zhou Electric Power
Load forecasting classification and paper reviewed are
Company) In this paper, the principle of fuzzy rough sets
presented in section II. In section III, the mid-term load
is used to help neural network in forecasting. In [20], fuzzy
forecasting concepts are explained in details. The
logic with back propagation algorithm (BP) is used for
experimental and results analysis are described in section IV.
short-term load forecasting in the uncertainty of the data
Finally, the conclusion with some comparison are in section
input case. In this paper, the network composes of 51 inputs
V.
and 24 outputs and it is simulated by MATLAB . [21]

presents short-term load forecasting by combining neural
network and genetic algorithm with the case study in
II. LOAD FORECASTING CLASSIFICATION AND PAPER
Taiwan while [22] presents the implementation of genetic
REVIEWED
algorithm method for fastening computation and increasing
Load forecasting results have been used for operation
forecasting accuracy. The time period of this load forecast
planning of electric system as well as maintenance and fuel
value is in 24 hours. In year 2001, [23] presents load
reserved planning [8].
forecasting model using the principle of wavelet
The load forecasting can be classified into 3 different
decompositions to bring to more accuracy in electric load
types according to the forecast period.
forecasting. In year 2006, [24] presents short-term load
1. Short-term load forecasting,
forecasting using fuzzy logic algorithm and input data of
2. Mid-term load forecasting,
time and temperature. The input variable ‗time‘ has been
3. Long-term load forecasting
divided into eight triangular membership functions. The
In each load forecasting, period of time, forecasted values
membership functions are Mid Night, Dawn, Morning, Fore
and aims of forecasting are noticeably different and they are
Noon, After Noon, Evening, Dusk and Night. Another input
comparably described in Table 1.
variable ‗temperature‘ has been divided into four triangle

membership functions. They are Below Normal, Normal,
TABLE I. TYPES OF LOAD FORECASTING [8]
Above Normal and High. The ‗forecasted load‘ as output
has been divided into eight triangular membership functions.
They are Very Low, Low, Sub Normal, Moderate Normal,
Normal, Above Normal, High and Very High. The case
studies have been carried out for the Neyveli Thermal
Power Station Unit-II (NTPS-II) in India. In 2004, [25]
95

---

IACSIT International Journal of Engineering and Technology Vol. 2, No.1, February, 2010
ISSN: 1793-8236
proposes a short term load forecasting using autoregressive
which combines the fuzzy linear regression method and the
integrated moving average (ARIMA) and artificial neural
general exponential smoothing method with the analysis of
network (ANN) method based on non-linear load. It is
temperature sensitivities. [32], in 2006, proposes the
concluded that using both methods can help each other in
development of load forecasting which combines the fuzzy
short-term load forecasting of the system. In 2007, [26]
logic, neural network and chaos and another algorithm as
proposes a novel method approach to load forecasting using
shown in Figure 2. The proposed algorithm shows good
regressive model and artificial neural network (ANN model)
accuracy or better than conventional method. [33] proposes
with the case study carried out for Turkey. In this research,
an approach based on combined regression method and
two methods are separately performed and compared. It
fuzzy inference system that developed for load forecasting.
shows that both methods give high accuracy results. In [27-
In addition, the fuzzy inference system makes a load
29], combination of artificial neural network (ANN),
correction inference from historical information and past
Genetic algorithm and Fuzzy logic (Fs) method are propsed
forecast load errors from a multi linear regression model to
for adjusting short-term load forecasting of electric system.
infer a forecast load error. The effectiveness of the proposed
Genetic algorithm is used for selecting better rules and back
approach to the short term load forecasting problem is
propagation algorithm is also for this network. The papers
demonstrated by practical data from the Taiwan Power
show that they give more accuracy results and faster
Company. Paper [34] presents the development of a neuro-
processing than other forecasting methods. In 2005, [30]
expert system for medium term load forecasting, back
proposes short-term load forecasting for holiday by using
propagation algorithm is slightly modified and is used to
fuzzy linear regression method. The proposed algorithm
train the artificial neural network. The proposed algorithm
shows good accuracy and the average maximum percentage
is tested on the practical 66/11 kV primary distribution
error of 3.57 % in the load forecasting of the holidays. [31],
system of Mysore, Karnataka State, South India.
in 2006, proposes a novel hybrid load forecasting algorithm,

TABLE II. SUMMARY OF SHORT TERM LOAD FORECASTING CLASSIFIED BY ALGORITHM.

Reference

Algorithm


[10],[11],[12],[13],[14],[15],[16],[23],[

Artificial Neural Network


60]

[17],[18],[19],[20]

Artificial Neural Network + Fuzzy


logic

[21],[22]

Artificial Neural Network + Genetic


[24]

Fuzzy logic


[25],[59]

ARIMA+ Artificial Neural Network


[26]

Regression+ Artificial Neural


Network

[27],[28],[29]

ANN + GAs + Fuzzy


[30],[33]

Fuzzy logic +Regression


[31],[32]

Hybrid


[55]

Support Vector Machine (SVM)


*ARIMA = Autoregressive Integrated Moving Average [25]



TABLE III. SHORT TERM LOAD FORECASTING CLASSIFIED BY TYPE OF INPUT.

Reference

Input


[10],[11],[13],[17],[18],[19],

Historical load


[20],[23],[24],[25]

Temperature







[16]

Humidity




Rainfall


[12],[14],[15],[27],[28],[29],[3

Wind speed


0],

[31]

Season




Weekday-mon-friday




Weekend- sat-sunday




Special day

Form Table 2 ,it can see that the methods of load The pre-processing methods are used in some articles for

forecasting are mainly classified by algorithms. However,
selecting inputs before forecasting process. Many and

preceding article interests do not absolutely give the
different input data are chosen or grouped before the

importance with algorithms. yet, they will also study in
forecasting [17], [55], [42] The preprocessing methods are

different kinds of the input variables in short term load
summarized in Table 4.

forecasting such as temperature and load in the past etc.
The previous summaries are the guidelines for midterm

Other input variables of short term load forecasting are
load forecasting that will be described in Section III.

s
ummarized in Table 3.

96

---

IACSIT International Journal of Engineering and Technology Vol. 2, No.1, February, 2010
ISSN: 1793-8236

TABLE IV. SHORT TERM LOAD FORECASTING CLASSIFIED BY INPUT SELECTION METHODS

Reference

Pre-processing


[17],[55]

Self-Organizing map


[13]

Graphical modeling for selection input variable


[57]

Input dimension reduction


[42]

Fuzzy clustering



forecasting in order to get the best answer. [17-19],[28-29].
III. THE MID-TERM LOAD FORECASTING CONCEPTS
4) Genetic algorithms

The articles of mid term load forecasting are summarized
In [48-49], genetic algorithms are used for electric load

into 3 classification topics .
demand forecasting. In [48], it is used together with neural

network for load forecasting. This article is hourly load

A. Data Inputs
forecasting,. It is used as a base for monthly load forecasting,
1) I nput classification
which duration time is 45 day and 49 day ahead by using

Historical load inputs and meteorological data such as
historical data in 2005. In this method genetic algorithm is

monthly maximum temperature, minimum temperature are
used to seek the initial weight of the neural network without

used in [35-37] and [39, 40]. Economic variables are also
random initial weight. It will give the network getting the

included in [38]. The historical load data are collected from
results faster.

El ectricity Generating Authority of Thailand (EGAT).
5) Support Vector Machine (SVM)

Temperature data are provided from Meteorological
Support vector machine (SVM) is presented in [50] for
D epartment and economic data are presented by the electric load forecasting. This method used historical data
go vernment of Thailand.
in the past , present and the future in the weather and load,
2) Input improvement
in 2001 to forecast load in 31 days ahead. This method is

Articles [13, 17, and 42] proposed the input improvement
similar to neural network unless the support vector machine

or input selection for preprocessing of load forecasting.
(SVM) can be used to separate input data before going to
M any researchers used grouping data technique for
forecasting process.
de creasing complication of data and calculation time of the
6) Statistics
ne twork. The preprocessing methods are for example Self-
[51-53] proposes the methods of load forecasting by
O
rganizing Map and Data mining.
using the principle of mathematic or statistic such as

Physical series algorithm [52] Autoregressive [51] and Non-
B. Artificial Intelligence technology methods and other
linear regression [53]. Statistics method is suitable for
method for load forecasting
linear type of data. for example humidity, heat or
1) Expert systems (Es)
temperature or meteorological parameter and historical
Article [34], proposes neuro-expert for electric mid term
monthly load [51].
load forecasting. The principle of Heuristic Rules is used for
sorting out complex data can decreased the time and
C. The adaptation of output forecasting
memory in training network.. Back propagation algorithm
In 1995, L.D.Voss, M.M.A.Salama, and J.Reeva [56]
neural network is used for forecasting.
developed the forecasting technique to load forecasting by
2) Artificial Neural Network (ANN)
using neural network and output filter correction as shown
[34-38] used Artificial Neural Network (ANN) approach
in Figure 3 . In this article, MA Filtering is used to improve
to forecast electrical demand load, by using the data
the output.
supporting from the government. The forecasting can be
performed the results in yearly (to 15 years), weekly (to 3
years) and hourly (to 24 hours). ,Many groups of researchers
used this forecasting approach for electricity mid term load
forecasting. Also this method can be used for electricity
peak load forecasting of distribution system [37]. By using

the relationship of learning data in the past, present and the
Figure.2: Load forecasting adaptation, MA Filtering [56]
future of temperature, the network can forecast a daily

peak load, total load of day and electricity monthly load
consumption. However, in [38], historical load, economic
IV. THE EXPERIMENTAL RESULTS AND ANALYSIS
and population variables are added for demand load
The articles mentioned previously can be comparably
forecasting by using neural network, time lagged feed
summarized in different points of view based on MAPE in
forward network (TLFN).
Table 5-8. However, the comparison can not tell that which
3) Fuzzy logic (Fs)
one of the method gives the best accuracy or which one is
In [42], principle of fuzzy logic is integrated with another
the best method. It is because each research technique is
method, The fuzzy logic method is used to manage the data-
performed in different objectives and using different data.
tolerance. In several articles used it for separating the input
The forecasting will focus on the output error that can be
data [42]. Some articles use it in repeating process of
determined as
97

---

IACSIT International Journal of Engineering and Technology Vol. 2, No.1, February, 2010
ISSN: 1793-8236
1 N
R
F
X  X
N is number of test set
k
k
MAPE 

.100 (1)

R
N 
X
k 1
k

Where

R
X is actual load of monthly load in k-th year
k

F
X is parameter of forecasted in same year
k
TABLE V. ACCURACY FOR EACH TYPE OF INPUT
Reference
Type of input
Accuracy
[24]
Historical load , temperature
MAPE < 2%
[38]
Historical load, temperature, GDP, CPI, HIS,
Can decrease
population
error
Historical load , temperature, humidity, wind speed,
MAPE < 2%
[40]
Rainfall
* GDP = Gross Domestic Product, CPI =Current Price Index, HIS = Housing

TABLE VI. PRE-PROCESSING
Reference
Pre-processing
Algorithm
Accuracy
[41]
Knowledge based
ANN+ Knowledge
MAPE 2.29 %
base
[48]
Fuzzy clustering
Fuzzy+ANN
MAPE 1.568 %
[55]
Self-Organizing map
SVM
MAPE (w) 1.65% (s)
2.42%
* SVM = Support Vector Machine

TABLE VII. ALGORITHMS
Reference
Input
Algorithm
Accuracy
[24]
Historical load , temperature, day
Hybrid
MAPE 1.60%
type
[37]
Historical load , temperature
ANN+hidden
2 hidden best
adj
results
[46]
Historical load , day type
Hybrid+GA
MAPE 2.80%
* GA= Genetic algorithm, adj = adjust, Hybrid = more than a algorithm

TABLE VIII. FILTERING OUTPUT
Reference
Input
Algorithm
Output
Accuracy
Historical
Avg. error decreased from
load
3.24% to 1.26%
Peak
Weather
ANN+MA
Peak error decreased from
[56]
Monthly
Economic
Filtering
9.55% to 4.81%
Load
Populatio
n
* MA = Moving Average

Table 5-8 show accuracy for each type of input, pre-
part in mid term load forecasting. It has ability to work with
processing, algorithm, and filtering output. The next,
non-linear data. Moreover, it can be effectively performed in
presents the article preceding of the forecasting from
complicate forecasting model for continuous data or signal.
international research.
This technology will help the conventional method

(statistical) in the complexity problems based on the value
of the variable between the variable input and nonlinear
V. CONCLUSION
correlation by training of data, learning process.
Mid term load forecasting (MTLF) becomes an essential
ACKNOWLEDGE
tool for today power systems, mainly in those countries
whose power systems operate in a deregulated environment.
This work was granted (funded) by Office of the Higher
This kind of load forecast is useful for many applications
Education Commission. Pituk Bunnoon was supported by
such as maintenance scheduling and development of cost
CHE510382 Ph.D. Scholarship. I would like to thank the
efficient fuel purchasing strategies. In the past load
Thai meteorological department, Ministry of commerce
forecasting are performed using the principle of artificial
Thailand, the office of the nation economic and social
intelligence technology. Each method uses difference input
development board, and Organization of electricity
and gives difference accuracy depending on the complexity
generating authority of Thailand (EGAT) for data
of input.
information and thank assistance Professor Dr. Kusumal
The artificial intelligence technology is used as a decision
Chalermyanont and associate Professor Dr.Chusak Limsakul
98

---

IACSIT International Journal of Engineering and Technology Vol. 2, No.1, February, 2010
ISSN: 1793-8236
advisors for clarifying several points in my research.
[24] S.Chenthur Pandian, K.Duraiswamy, C.Christer Asir Rajan,
N.Kanagaraj, ―Fuzzy approach for short term load forcasting,‖
Electric Power Systems Research 76, 2006, pp:541-548.
REFERENCE
[25] Jian-Chang Lu, Dong-xiao niu, zheng-yuan jia, ―A study of short term
[1] Energy to Energy Policy and Planning Office, Ministry of Energy,
load forecasting based on arima-ann,‖ Machine Learning and
Royal Thai Government, Thailand.
Cybernetics, vol.5, 2004, pp:3183 - 3187.
[2] Energy to Energy Policy and Planning Office, ―Load forecasting
[26] Ummuhan basaran filik, Mehmet kurban, ―A new approach for the
reports 2547,‖ Ministry of Energy, Royal Thai Government,
short term load forecasting with autoregressive and artificial neural
network models,‖
Thailand.
International journal of computational intelligence
[3] Peter Vas.,―Artificial-Intelligence-Based Electrical Machines and
research , vol.3, no.1, pp:66-71.
Drives,‖ Oxford science publication, 1999.
[27] P.K.Dash, S.Mishra, S.Dash, A.C.Liew, ―Genetic optimization of a
[4] Lefteri H.Tsoukalas, Robert E.Uhrig, ―Fuzzy and Neural Approaches
self organizing fuzzy-neural network for load forecasting,‖ IEEE
in Engineering,‖ A wiley-Interscience publication, JohnWilley&Sons
2000.
INC, 1997.
[28] Gwo-ching liao, Ta-peng tsao, ―Integrated genetic algorithm/Tabu
[5] Peter M.ills, Albert Y.Zomaya, and Moses O.Tade, ―Neuro-Adaptive
search and neural fuzzy networks for short-term load forecasting,‖
Process Control A Practical Approach,‖ John Willey&Sons Ltd, 1996.
Power Engineering Society General Meeting, 2004, pp:1082 – 1087.
[6] R.A. Aliev, R.R.Aliev,―Soft Computing and its Application,‖ World
[29] Gwo-ching liao, Ta-peng tsao, ―Novel GA-Based Approach and
scientific publication, 2001.
Neural fuzzy networks application in short-term load forecasting,‖
[7] Stamatios V.Kartalopoulos.Understanding, ―Neural Network and
Power Engineering Society General Meeting, 2004, pp:589-594.
Fuzzy logic Basic Concepts and Applications,‖ AT&T, 1996.
[30] Kyung-bin song, young-sik baek, dug hun hong and gilsoo jang,
―Short
[8] K. Metaxiotis, A. Kagiannas, D. Askounis, J. Psarras, ―Artificial
-term load forecasting for the holidays using fuzzy linear
intelligence in short term electric load forecasting :a state-of-the-art
regression method‖ IEEE Trans. power syst., vol.20, no.1, 2005,
survey for the research,‖ Energy Conversion and Management 44,
pp:96-101.
2003, pp.1525-1534.
[31] Kyung-bin song, seong-kwan ha, jung-wook park, dong-jin kweon,
[9] Georgr gross,franscisco D.galiana, ―Short-term load forecasting,‖
kyu-ho kim, ―Hybrid load forecasting method with analysis of
temperature sensitivities,‖
Preceeding of the IEEE, vol.75,no.12, December 1987, pp.1558-1573.
IEEE Trans. power syst., vol.21, no.2,
[10] Tomonoobu senjyu, hitoshi takara,katsumi ueezato,toshihisa
2006, pp:869-876.
funabashi, ―One-hour ahead load forecasting using neural network,‖
[32] Gwo-ching liao, Ta-peng tsao, ―Application of a fuzzy neural network
IEEE Trans. power syst., vol.17, no.1, February 2002, pp.113-118.
combined with a chaos genetic algorithm and simulated annealing to
short term load forecasting,‖
[11] Zhiyong Wang,Yijia Cao, ―Mutual Information and Non-fixed ANNs
Evolutionary Computation, vol.10, issue
for Daily Peak load forecasting,‖ Power Systems Conference and
3, 2006, pp:330 – 340.
Exposition, IEEE PES ,2006, pp:1523 – 1527.
[33] R.H.Liang and C.-C.Cheng, ―Combined regression-fuzzy approach
[12] wenjin Dai, ping Wang, ―Application of pattern recognition and
for short-term load forecasiting,‖ Generation, Transmission and
artificial neural network to load forecasting in electric power system,‖
Distribution, vol.147, issue 4, 2000, pp:261 – 266.
Third international conference on natural computation, ICNC 2007.
[34] Adiga S. Chandrashekara, T. Ananthapadmanabha, A. D. Kulkarni. A
[13] Hiroyuki mori,eitaro kurata, ―Graphical Modeling for Selecting Input
neuro-expert system for planning and load forecasting of distribution
Variables of Short-term Load Forecasting,‖ Power Tech, 2007 IEEE
systems. International Journal of Electrical Power & Energy Systems,
Lausanne, July 2007, pp:1084 – 1089.
vol.2, issue 5, 1999, pp:309-314.
[14] G.A.Adepoju, S.O.A.ogaunjuyigbe, and K.O.Alawode, ―Application
[35] J.V.Ringwood, D.Bofelli and F.T.Murray, ―Forecasting electricity
of neural network to load forecasting in nigerian electrical power
demand on short, medium and long time scales using neural
system,‖
networks,‖
The pacific journal of science and technology, vol.8, no.1,
Journal of Intelligent and Robotic System 31, 2001,
May 2007, pp.68-72.
pp:129-147.
[15] Mohsen Hayati, Yazdan Shirvany, ―Artificial neural network
[36] Ioan Borlea, Adrian Buts, ―Some aspects concerning mid term
approach for short term load forecasting for Illam region,‖
monthly load forecasting Using ANN,‖ EUROCON , 22-24 Nov.2005.
International journal of electrical, computer, and system engineering,
[37] T.Yalcinoz, U.Eminoglu, ―Short term and medium term power
distribution load forecasting by neural networks,‖
vol.1, no.2, pp.121-125.
Energy Conversion
[16] H.A.Salama, A.F.AbdElGawad, H.M.Mahmond, E.A.Mohamed,
and Management 46, 2005; pp:1393-1405.
S.M.Saker, ―Short term load forecasting investigations of eqyptian
[38] Danilo Bassi, Oscar Olivares, ―Medium term electric load forecasting
electrical network using ANNs,‖ Universities Power Engineering
using TLFN neural networks,‖ IJCCC, vol.I, no.2, 2006, pp:23-32.
Conference, 2007, pp:550 – 555.
[39] M.Ghiassi, David K.Zimbra, H.Saidane, ―Medium term system load
[17] Bhavesh kumar chauha, amit sharma, and m.hanmandlu, ―Neuro-
forecasting with a dynamic artificial neural network model,‖
fuzzy approach based short term electric load forecasting,‖ 2005
Electrical Power system Research 76, 2006, pp:302-316.
IEEE/PES Transmission and Distribution Conference & Exhibition,
[40] M.C.Falvo, R.Lamedica, ―Meteorological parameters influence for
medium term load forecasting,‖ 2006 , pp:1296 –
Asia and Pacific Dalian, Chaina.
1301.
[18] Z.Y.Wang, C.X.Guo , Y.J.Cao, ―A New method for short term load
[41] M.C.Falvo, R.Lamedica, ―A Knowledge based system for medium
term load forecasting,‖
forecasting integrating fuzzy tough sets with artificial neural
IEEE, 2006, pp:1291 – 1295.
network,‖ Power Engineering Conference, IPEC 2005, pp:1 – 173.
[42] Lu Yue, Yao Zhang, Huifan Xie, Qing Zhong, ―The fuzzy logic
[19] Cuiru wang, Zhikun cui , Qi chen, ―Short term load forecasting based
clustering neural network approach for middle and long term load
on fuzzy neural network,‖ Intelligent Information Technology
forecasting, Grey Systems and Intelligent Services(GSIS 2007), 2007,
Application, Workshop on 2-3 Dec. 2007, pp:335 – 338.
pp:963 – 967.
[20] Hari seetha and R.saravanan, ―Short term electric load prediction
[43] Nima Amjady, Farshid Keynia, ―Mid-term load forecasting of power
using fuzzy BP,‖ Journal of computing and information technology-
system by a new prediction method,‖ Energy conversion and
CIT 15, 2007, pp:267-283.
management (science direct),2008, pp.1-10.
[21] Chih-hsien kung, michael J.Devaney, chung-ming huang, chih-ming
[44] E.Gonzalez-Romera,M.A.Jaramillo-Moran,and D.Carmona-
Fernandez, ―Mon
kung. An adaptive power system load forecasting scheme using a
thly electric energy demand forecasting with neural
networks and fourier series,‖
genetic algorithm embedded neural network. IEEE Intrumentation
Energy conversion and management
and Measurement technology conference St.Paul,Minnesota, USA,
(science direct),2008, pp:1-8.
May 18-20.
[45] R.E.Abdel-Aal, ―Univariate modeling and forecasting of monthly
[22] L.L.Lai,H.Subasinghe, N.Rajkumar, E.Vaseekar, B.J.Gwyn,
energy demand time series using abductive and neural networks,‖
V.K.Sood, ―Object oriented genetic algorithm based artificial neural
Computer & Industrial Engineering 54, 2008, pp:903-917.
network for load forecasting,‖ Springer-verlag berlin ,1999, pp:462-
[46] Ronaldo R.B. de Aquino, Aida A.Ferrira, Manoel A.Carvalho Jr,
―Development of a hybrid intelligent system for electrical load
469.
forecasting,‖
[23] Zhao-yang dong, Bai-ling zhang, Qian huang. A daptive neural
IBERAMIA-SBIA, 2006, pp:228-237.
network short term load forecasting with wavelet decompositions.
[47] Ronaldo R.B. de Aquino, Aida A.Ferrira, Manoel A.Carvalho Jr.,
―Combining Artificial neural networks and heuristic rules in a hybrid
IEEE Porto power tech conference, 10-13 september,porto,Portugal.
intelligent load forecast system,‖ ICANN, 2006, pp:757-766.
99

---

IACSIT International Journal of Engineering and Technology Vol. 2, No.1, February, 2010
ISSN: 1793-8236
[48] Ronaldo R.B.de Aquino, Otomi Nobrega Neto, Milde M.S.Lira,Aida
A.Ferreira,Manoel A, ―Development of an artificial neural network by
genetic algorithm to mid term load forecasting,‖ Neural Networks
2007(IJCNN 2007). International Joint Conference, 2007, pp:1726 –
1731.
[49] Ronaldo R.B.de Aquino, Otomi Nobrega Neto, Milde M.S.Lira, Aida
A.Ferreira, Manoel A., ―Using Genetic Algorithm to Develop a
Neural network based load forecasting,‖ ICANN, 2007, pp:738-747.
[50] M Gavrilas, I Ciutea and C Tanasa, ―Load Forecasting Using support
vector machines,‖ IEE 2001,18-21 June 2001, pp:1-7.
[51] S.Mirasgedis, Y.Sarafidis, E.Georgopoulou, D.P. Lalas, ―Models for
mid-term electricity demand forecasting incorporating weather
influences,‖ energy 31,2006, pp:208-227.
[52] Xie Da, Yu Jiangyan, Yu Jilai, ―The physical series algorithm of mid-
long term load forecasting,‖ Electric power research .53, 2000, pp:31-
37.
[53] Alexander Bruhns, Gilles Deurveilher, Jean-sebastien roy, ―A non-
linear regression model for mid-term load forecasting and
improvements in seasonality,‖ 15th PSCC Liege, August 2005.
[54] G.J.Tsekouras, P.B.Kotoulas, C.D.Tsirekis, E.N.Dialynas,
N.D.Hatziargyriou, ―A pattern recognition methodology for
evaluation of load profiles and typical days of large electricity
customers,‖ Electric power systems research78, pp:1494-1510.
[55] Shu fan, Chengxiong Mao, Luonan Chen, ―Electricity Peak Load
Forecasting with Self-Organizing Map and Support Vector
Regression,‖ Transactions on Electrical and Electronic Engineering,
2006, pp:330-336.
[56] L.D. Voss , M.M.A. Salama , J.Reeve, ―A Practical Approach to
Electric Load Forecasting Using Artificial Neural Networks with
Corrective Filtering,‖ IEEE, 1995.
[57] Xu Tao, He Renmu, Wang Peng and Xu Dobgjie, ―Input dimension
reduction for load forecasting based on support vector machines,‖
IEEE International Conference on Electric Utility Deregulation ,April
2004, Hong Kong.
[58] A.A. El Desovley and M.M. ElKateb, ―Hybrid adaptive techniques
for electric load forecasting using ANN and ARIMA,‖ IEE
proceeding, vol.147, no.4, July 2000.
[59] Pituk Bunnoon, Kusumal Chalermyanont, and Chusak Limsakul,
―Mid Term Load Forecasting of the Country Using Statistical
Methodology: Case study in Thailand,‖ 2009 International
Conference on Signal Processing System, pp.924-928, May 2009.
[60] Pituk Bunnoon, Kusumal Chalermyanont, and Chusak Limsakul,
―Peak load demand forecasting using 2-level discrete wavelet
decomposition and neural network algorithm,‖ Proceeding of SPIE,
second international conference on digital image processing, Feb.26-
27, 2010.





Pituk Bunnoon received the B.S. degree from King Mongkut‘s Institute of
techhnology Ladkrabang, Thailand, in 1998, and the M.S. degree in
electrical engineering from Prince of Songkla University, Thailand, in 2004.
His research interest is application of artificial intelligence to power system
planning and operation.

Kusumal Chalermyanont received the B.S. degree from Prince of
Songkla University, Thailand, in 1993, the M.S. degree in electrical
engineering from University of Colorado at Boulder in 1999, and the Ph.D.
in electrical electrical engineering from University of Colorado at Boulder
in 2003. Her research interest are power electronics, magnetic designs for
power electronics, renewable energy system/management.

Chusak Limsakul received the B.S. degree from King Mongkut‘s Institute
of technology Ladkrabang, Thailand, in 1978, and the D.E.A. degree from
INSAT France, in 1982, and Docteur Ingenieur form INSAT France, in
1985. His research interest are digital signal processing, sensors and
instrumentations, and automation.



100

---