Energy 70 (2014) 165e171

Contents lists available at ScienceDirect

Energy
journal homepage: www.elsevier.com/locate/energy

Forecasting the annual electricity consumption of Turkey using an

optimized grey model
Coskun Hamzacebi*, Huseyin Avni Es
Karadeniz Technical University, Department of Industrial Engineering, Trabzon, Turkey

a r t i c l e i n f o a b s t r a c t

Article history: Energy demand forecasting is an important issue for governments, energy sector investors and other
Received 20 November 2013 related corporations. Although there are several forecasting techniques, selection of the most appropriate
Received in revised form technique is of vital importance. One of the forecasting techniques which has proved successful in
17 March 2014
prediction is Grey Modeling (1,1). Grey Modeling (1,1) does not need any prior knowledge and it can be
Accepted 24 March 2014
used when the amount of input data is limited. However, the basic form of Grey Modeling (1,1) still needs
Available online 17 April 2014
to be improved to obtain better forecasts. In this study, total electric energy demand of Turkey is pre-
dicted for the 2013e2025 period by using an optimized Grey Modeling (1,1) forecasting technique called
Keywords:
Energy demand forecasting
Optimized Grey Modeling (1,1). The Optimized Grey Modeling (1,1) technique is implemented both in
Grey modeling (1,1) direct and iterative manners. The results show the superiority of Optimized Grey Modeling (1,1) when
Direct forecasting compared with the results from literature. Another finding of the study is that the direct forecasting
approach results in better predictions than the iterative forecasting approach in forecasting Turkey’s
electricity consumption. The supply values of primary energy resources in order to produce electricity
have calculated for 2015, 2020 and 2025 by using the outputs of Optimized Grey Modeling (1,1).
Ó 2014 Elsevier Ltd. All rights reserved.

1. Introduction second in the world countries in terms of growth rates of electrical

energy consumption in the last ten years [2].
Turkey, a transit point between Asia and Europe, is important A final report by the Ministry of Energy and Natural Resources
country in terms of energy policy owing to its consumption (MENR), electrical energy demand for 2021 is estimated to reach
amounts and strategic location. The population and the GDP of approximately 467 billion kWh and 424 billion kWh according to
Turkey reached 735 billion people and $73.7 million in 2010 with the higher demand and lower demand scenarios, respectively [3].
an average increase of 8.7% and 177% compared to 2000, respec- Accurate and reliable forecasting of the electrical energy con-
tively. Energy consumption rates in Turkey have been rapidly sumption is of great importance in order to meet the increasing
increasing in parallel with GDP and population. Thus, Turkey has demand for electrical energy as well as sustaining industrialization
become a country that has the most swift increase in energy de- and long-term stable energy policies. Since storing the alternating
mand among the Organization for Economic Co-operation and current electricity is impossible, accurate and reliable forecasting is
Development (OECD) member countries in the last decade [1]. also necessary for planning the right tools which provide the
Turkey has been fast growing in terms of both its economy and amount of required electrical energy when it is needed.
population. As a result, the demand for electrical energy has In the literature, many different forecasting techniques were
increased quickly in the last decade. used in order to predict the electrical energy demand. Saab et al. [4]
While the electricity consumption of Turkey was 98 billion kWh studied univariate autoregressive models to forecast the monthly
in 2000, it reached 172 billion kWh in 2010 with an average of 7.5 electrical energy consumption in Lebanon. By using artificial neural
percent increase per year. Similarly, electricity generation reached networks (ANNs); An et al. [5] investigated the prediction of half-
211 billion kWh in 2010 with an average of 7 percent increase per hour electricity demand in Australia, Pao [6] determined the con-
year. Thus, Turkey was the first in European countries and the sumption of electrical energy in Taiwan, and Gonzalez et al. [7]
predicted the monthly electricity demand of Spain. By using
regression models; Mohamed and Bodger [8] predicted the elec-
* Corresponding author. Tel.: þ90 462 377 2950; fax: þ90 462 325 6482.
E-mail addresses: [email protected] (C. Hamzacebi), [email protected] (H. tricity consumption under the effect of selected economic and
A. Es). demographic variables, Mirasgedis et al. [9] predicted the demand

http://dx.doi.org/10.1016/j.energy.2014.03.105
0360-5442/Ó 2014 Elsevier Ltd. All rights reserved.
166 C. Hamzacebi, H.A. Es / Energy 70 (2014) 165e171

Atak [42] used grey rolling prediction model in order to forecast

Nomenclature Turkey’s electricity demand of industrial sector and total electricity
demand. Huang et al. [43] combined grey model and Markov model
AGO Accumulated generating operation to forecast China’s electricity consumption. Bianco et al. [44]
q Horizontal adjustment coefficient compared the performance of grey rolling trigonometric model
A Parameter in GM (1,1) and exponential smoothing technique for prediction of Romania’s
a Development coefficient electrical energy consumption. Kumara and Jain [45] used grey
b Grey action quantity rolling model to forecast electrical energy and coal consumptions,
C Change ratio of forecasting error GreyeMarkov model to forecast petrol consumption, single spec-
ei The i th value of forecasting error trum analysis to forecast natural gas. The usage of grey prediction
h Length of forecasting horizon model in energy demand forecasting is expanding rapidly.
IAGO Inverse accumulated generating operation As seen in the literature, the different forecasting methods were
k Length of subsequence in GM(1,1) used to predict electricity demand. One of these is basic GM (1,1)
MAE Mean absolute error which is as a prediction tool in grey system theory. In the studies
MAPE Mean absolute percentage error performed with basic GM (1,1), the recommended parameter
s2 Standard deviation of forecasting errors values were used (39e45). In the proposed GM (1,1), parameter
s1 Standard deviation of original data values were optimized by an improved algorithm in order to in-
X0 Original sequence crease forecasting accuracy. In addition, direct and iterative fore-
Xt0 the t th entry of original sequence casting approaches for grey model were presented as a novelty.
X1 accumulated sequence In this study, electricity consumption of Turkey is predicted by
Xt1 the t th entry of accumulated sequence using GM (1,1) technique. In addition to basic GM (1,1), some of the
b
X the t þ 1 th entry of forecasting value parameters of GM (1,1) are optimized to obtain a better forecasting
tþ1
performance. Using the optimized GM (1,1), the electricity demand
of Turkey by 2025 is predicted. The remainder of the study is
organized as follows. The basic GM (1,1) and the optimized GM (1,1)
are given in Section 2. Section 3 deals with the application of the
of electrical energy via climatic variables, Al-Shobaki and Mohsen optimized GM (1,1) on Turkey’s electricity demand prediction and
[10] forecasted the electrical energy demand in Jordan, Bianco et al. have been calculated electricity supply of Turkey. Finally, the con-
[11] forecasted electricity consumption in Italy, and Meng and Niu clusions are given.
[12] predicted the annual electrical energy demand of China.
Amarawickma and Hunt [13] predicted electricity demand for Sri 2. Grey prediction
Lanka and Zahedi et al. [14] predicted the electricity demand of
Canada via neuro-fuzzy model. An interdisciplinary approach, the grey theory (GT) was put
Similarly, several methods have been used to predict Turkey’s forward as an alternative method to digitize uncertainty in early
electricity demand. The following studies were selected from the 1980s by Deng [29]. GT has been successfully applied in many fields
literature in order to present the variety of the methods. Hamzacebi such as industry, social systems, economics, traffic, management,
and Kutay [15] used ANN model to predict the long-term con- education, environment, engineering etc. GT deals with systems
sumption, Yumurtaci and Asmaz [16] predicted the demand via owned poor information and can find solutions to the problems
regression model. Using genetic algorithms, total and industrial with small samples and incomplete information [46]. The main
sector demand were predicted by Ozturk et al. [17]. Tunc et al. [18] purpose of the theory is to predict the behavior of systems which
predicted the demand by using regression analysis in order to cannot be detected with stochastic or fuzzy methods with limited
optimize power plant investments. Hamzacebi [19] predicted the data [47]. Although there are many different grey prediction
sectoral demands by using ANN. Erdogdu [20] and Sumer et al. [21] models, the most common grey prediction model is GM (1,1).
used ARIMA models, Kavaklioglu et al. [22] used ANN models,
Toksari [23] used ant colony optimization methods, Kucukali and 2.1. Basic GM (1,1)
Baris [24] used fuzzy logic method, Demirel et al. [25] used ANFIS
and ARIMA models, Kavaklioglu [26] used support vector regres- GM (1,1) is a time series forecasting model which implies a first
sion method, Dilaver and Hunt [27] used time series techniques, order single variable prediction model. GM (1,1) does not need any
and Arisoy and Ozturk [28] used time varying parameters model to prior knowledge such as probability distribution of the input data
predict demand. and it can be used when the amount of input data is limited (at least
Grey forecasting models are amongst the recent prediction 4 data is necessary). This is a facilitator characteristic of the model
methods. The most widely used Grey prediction model is GM (1,1), among the prediction methods under the limited data conditions.
which is a part of grey system theory developed by Deng in 1980s GM (1,1) is based on three essential steps [48]:
[29]. GM (1,1) is used as a prediction tool in several different areas
such as the prediction of the severity of the earthquake [30], i. Accumulated generating operation (AGO)
vegetable production amounts [31], El Nino events [32], amount of ii. Grey modeling
airline passengers [33], stock prices [34], risks of fatal accidents [35], iii. Inverse accumulated generating operation (IAGO)
carbon emissions, energy consumption and economic growth [36],
education expenditure [37], wave energy [38]. The literature is wide The steps of GM (1,1) are shown below:
and continues to expand. Besides these prediction areas, energy
consumption forecasting is also one of the vital usage areas of GM. Step 1: For the initial time sequence
For example; Hsu and Chen [39] predicted power demand and Yao
et al. [40] predicted electricity demand of Taiwan. Zhou et al. [41]
n o  
developed a trigonometric grey forecasting model and they used
X0 ¼ X10 ; X20 ; X30 ; ::::::; Xn0 ¼ Xt0 ; t ¼ 1; 2; 3; :::; n; n  4 (1)
this model to forecast electricity consumption of China. Akay and
 C. Hamzacebi, H.A. Es / Energy 70 (2014) 165e171 167

where Xt0 is a non-negative sequence and n is the number of the iterative or direct forecasting approaches can be applied. The next
data. sections give a brief knowledge about both approaches.

Step 2: When this sequence is subjected to AGO, the following

2.1.1. Grey iterative forecasting method
sequence X1 is obtained. It is obvious that Xk1 is monotonically
Iterative forecasting approaches are based on a rolling mecha-
increasing.
nism. When the number of periods to be predicted is equal to h
n o   (h ¼ 1, 2, 3, ..., h), the iterative forecasting approach can be applied
X 1 ¼ X11 ;X21 ;X31 ;::::::;Xn1 ¼ Xt1 ; t ¼ 1;2;3;:::;n;n  4 (2) as follows:

where i. Predict the value of next first period,

ii. By using the obtained prediction value of the next first
( )
X
k period, predict the value of next second period,
Xk1 ¼ Xt0 ; t ¼ 1; 2; :::; n; k ¼ 1; 2; :::; n (3) iii. Repeat the operation stated in the second step until the
t¼1 prediction value of hth period is obtained.

Step 3: Form grey modeling by establishing a first order grey When the number of period to be predicted is denoted by h, the
differential equation length of subsequence in GM (1,1) is denoted by k and the predic-
b tþ1 , then the two situations that are
tion value for t þ 1th period is X
encountered according to h < k or h > k are given as follows:
Xt0 þ aZt1 ¼ b; t ¼ 2; :::; n (4)
i) for h < k;
where
b
X tþ1 ¼ GMð1; 1Þ½Xt ; Xt1 ; .; Xtkþ1  (12)
Zt1 ¼ qXt1 þ ð1  qÞXt1
1
; t ¼ 2; ::; n (5)
h i
b b
q denotes a horizontal adjustment coefficient, and 0 < q < 1. The X tþ2 ¼ GMð1; 1Þ X tþ1 ; Xt ; Xt1 ; .; Xtkþ2 (13)
selecting criterion of q value is to yield the smallest forecasting
error rate. In Eq. (4), a is called the development coefficient and b is h i
b b b
called driving coefficient. Applying least square method co-
X tþ3 ¼ GMð1; 1Þ X tþ2 ; X tþ1 ; Xt ; .; Xtkþ3 (14)
................
efficients, [a b]T can be estimated as
   1 h i
b b b b
tþh ¼ GMð1; 1Þ X tþh1 ; X tþh2 ; .; X tþ1 ; ; Xt ; .; Xtkþh
a X (15)
A ¼ ¼ BT B BT Yn (6)
b

where ii) for h  k;

2 3 2 03
Z21 1 X2
6 Z 1 b
B ¼ 6 17 6 X0 7
7Y ¼ 6 3 7
X tþ1 ¼ GMð1; 1Þ½Xt ; Xt1 ; .; Xtkþ1  (16)
4 ::::3 ::: 5 4 ::::: 5 (7)
h i
Zn1 1 Xn0 b b
X tþ2 ¼ GMð1; 1Þ X tþ1 ; Xt ; Xt1 ; .; Xtkþ2 (17)
Y ¼ BA
h i
b b b
Step 4: The whitenization processes is defined as follows X tþ3 ¼ GMð1; 1Þ X tþ2 ; X tþ1 ; Xt ; .; Xtkþ3 (18)
................
h i
dXt1 b b b b
þ aXt1 ¼ b (8) X tþh ¼ GMð1; 1Þ X tþh1 ; X tþh2 ; .; X tþhk (19)
dt
Hence, using a and b coefficients the AGO grey prediction model
One of the major disadvantages of this approach is that it uses
can be obtained.
only the predicted values to obtain the prediction value of the next
  period when the size of forecasting horizon is large. When equa-
b 1 ¼ X 0  b eat þ b;
X t ¼ 0; 1; 2; :::: (9) tions are examined for h > k, new prediction values cannot be
tþ1 1
a a
obtained by using the past observed values after a certain period. As
Step 5: Obtain the prediction value of original sequence by IAGO a result, the new prediction values are obtained from the previous
operation. IAGO operation can be defined as: prediction values. Therefore, it increases the prediction error in the
model. This is an undesirable situation. Direct forecasting approach
0
Xtþ1 1
¼ Xtþ1  Xt1 (10) can be preferred to avoid the deficiency of iterative forecasting
approaches when the number of periods to be predicted is suffi-
If Eq. (9) is substituted in Eq. (10) then Eq. (11) is obtained as ciently large.
 
b 0 ¼ ð1  ea Þ X 0  b eat
X (11) 2.1.2. Grey direct forecasting method
tþ1 1
a
The prediction value of the period h can be forecasted using only
This prediction process can be applied only for one future the past observation values of k periods in the Grey direct fore-
period. If the number of periods to be predicted is greater than one, casting approach. The following equations illustrate how the grey
168 C. Hamzacebi, H.A. Es / Energy 70 (2014) 165e171

direct forecasting method is applied to predict the value of h future

periods:
h i
b
X tþ1 ¼ GMð1; 1Þ Xt ; Xt1 ; .; Xtðk1Þ (20)

h i
b
X tþ2 ¼ GMð1; 1Þ Xt ; Xt2 ; Xt4 ; .; Xt2ðk1Þ (21)

h i
b
X tþ3 ¼ GMð1; 1Þ Xt ; Xt3 ; Xt6 ; .; Xt3ðk1Þ (22)
................

h i
b
X tþh ¼ GMð1; 1Þ Xt ; Xth ; Xt2h ; .; Xthðk1Þ (23)

As it is seen in Eqs. (20)e(23), the grey direct forecasting

approach is based on the usage of the past real observations. It
selects the past observation values according to the value of h.
For example, if h ¼ 2, the approach uses [Xt, Xt  2, Xt  4, .,
Xt  2(k  1)] subsequence to predict the next 2 periods. Generally Fig. 2. Turkey’s total electricity consumption between 1945 and 2010.
speaking, in order to forecast h future periods ahead the
approach selects the past observation values from the present
time to k periods backwards by in every h periods. Hence, [Xt, Xt  h, Xt  2h, ., Xt  h(k  1)] subsequence is obtained to forecast
b
X tþh .

2.2. Optimized grey forecasting model

The q and k parameters affect the prediction performance of GM

(1,1). In the literature, these parameters are generally used as
q ¼ 0.5 and k ¼ 4. However, different values of these parameters
may increase the prediction performance of GM (1,1). For this
reason the best value of these parameters are investigated by ex-
periments. This process is called optimized grey forecasting model
(OGM (1,1)). Fig. 1 shows the flow chart of OGM (1,1).

3. Prediction of Turkey’s electricity demand by using OGM

(1,1)

In order to make satisfactory and successful prediction, a sys-

tematical forecasting approach is followed which includes data
collection, performance measure selection, computer coding of the
model and analyzing the outputs of computer modeling. Following
next sections give brief information about these steps.

3.1. Data

Although there are several data sources on the energy sta-

tistics of Turkey, the data used in this study are taken from

Table 1
Adequacy levels for performance measures.

Threshold MAPE C

Accuracy scale
1st level (perfect) 1% 0.35
2nd level (well) 5% 0.50
3rd level (acceptable) 10% 0.65
4th level (incapable) >10% >0.65

Table 2
The performance values of forecasting approaches for the test set.

Forecasting approaches MAPE MAE C

Direct OGM (1,1) 3.28 5239.31 0.18

Iterative OGM (1,1) 5.36 8662.97 0.81
Fig. 1. The flow chart of OGM (1,1).
 C. Hamzacebi, H.A. Es / Energy 70 (2014) 165e171 169

Table 3
Comparison results between OGM (1,1) and studies in literature.

Akay and Atak [42] Pi. Liu and Qin [53] OGM (1,1) MENR of Turkey (MAED)

Direct approach Iterative approach High scenario (HS) Low scenario (LS)

The test set data include the years 1994e2004 1996e2006 2006e2010 2006e2010 2006e2010 2006e2010
MAPE 3.43 4.6 3.28 5.36 8.15 5.14

post error ratio (C) has been used in order to determine the accu-
racy of OGM (1,1). The following equations show the calculation of
the performance measures.

1X n
Mean Absolute ErrorðMAEÞ ¼ je j (24)
n i¼1 i

n  
1X  ei *100
Mean AbsolutePercentageErrorðMAPEÞ ¼ (25)
n i¼1 Yi 

s2
Post Error RatioðCÞ ¼ (26)
s1

where ei and Yi denote the error value and the observation value of
the ith period, s1 and s2 represent the standard deviation of original
Fig. 3. Graphs of the test sets between MAED and OGM (1,1). data and standard deviation of the errors, respectively. Lower
values of all performance measures indicate the better prediction.
Turkish Statistical Institute (TUIK) which reports complete, cor- According to Lin and Hsu [49], if MAPE is less than 10%, it is
rect and official data. The data includes the amount of total accepted as a successful prediction. Recommendations of C mea-
electricity consumption from 1945 to 2010. Fig. 2 depicts the sure are also available in the literature for successful prediction
used data graphically. As it is seen in Fig. 2, there is a rapid [50,51]. Adequacy levels for both performance measures are pre-
growing trend in the electricity consumption series beginning sented in Table 1 [52].
from 1960s.
The data divided into two parts called modeling data set and test 3.3. Obtained results and comparison
data set. The modeling data set includes the amount of electricity
consumptions during the years 1945e2005 while the test data set Applying the both direct and iterative grey forecasting ap-
includes the consumption rates for the years 2006e2010. The proaches, the most appropriate grey prediction model was deter-
modeling data set is used to determine the best GM (1,1) model mined by using the data reserved for model building stage and then
which is called OGM (1,1) previously. The test data set is used to the validity of the selected model was tested with the test set. The
generalize OGM (1,1) predictions. Finally, using OGM (1,1) the best value of q and the value of k were found as 0.45 and 4,
amount of electricity consumption of Turkey are forecasted for the respectively. The performance values of OGM (1,1) for the test data
period until 2025. set is given in Table 2.
It can be deduced from Table 3 that the OGM (1,1) direct fore-
3.2. Performance measures casting approach gives better results than the OGM (1,1) iterative
forecasting approach. The results obtained by OGM(1,1) were
In the literature, different performance measures have been compared with the results of Akay and Atak [42], and Pi et al. [53] as
proposed in order to determine the accuracy of the prediction well as the official results of Turkey’s MENR in order to make a fair
models. In this study, three performance measures such as mean
absolute error (MAE), mean absolute percentage error (MAPE) and
Table 4
Prediction values of Turkey’s total electricity energy demand until 2025.

Year Direct OGM (1,1) (106 kWh) Iterative OGM (1,1) (106 kWh)

2011 174201 174201

2012 189821 185498
2013 206182 191225
2014 221098 201202
2015 224723 208916
2016 236078 218725
2017 252829 227751
2018 270920 237933
2019 282629 248005
2020 294867 258834
2021 308973 269870
2022 327054 281506
2023 339575 293507
2024 349834 306059
2025 354306 319069
Fig. 4. Graphical representation of the prediction values.
170 C. Hamzacebi, H.A. Es / Energy 70 (2014) 165e171

Table 5
The utilization ratios of the primary energy sources (%).

Gas Hydro Coal Lignite Nuclear Geothermal Oil Wind Others Total

2015 32.5 38.9 8.4 13.7 0 0.3 2 3.7 0.5 100

2020 30.8 37.2 10.2 10.9 2.6 0.3 1.5 6.0 0.5 100
2025 28.1 37.3 11.9 8.3 5.5 0.5 0.8 8.2 0.2 100

Table 6
Supply values for Turkey electricity production (103x tones of equivalent).

Gas Hydro Coal Lignite Nuclear Geo-thermal Oil Wind Others Total

2015 6281.0 7517.9 1623.4 2647.7 0 58.0 386.5 715.1 96.6 19326.2
2020 7810.4 9433.4 2586.6 2764.1 659.3 76.1 380.4 1521.5 126.8 25358.6
2025 8562.2 11365.4 3626.0 2529.0 1675.9 152.4 243.8 2498.6 60.9 30470.3

comparison. MENR prepares energy demand forecasts by using the methods and select the best ones in terms of prediction perfor-
“Model for Analysis of Energy Demand” (MAED) simulation tech- mance. This study also approves that there is still a room for
nique considering the high and low scenarios. The comparison improving the performance of the existing prediction methods as
results are given in Table 3 and Fig. 3, and it is seen that the Direct OGM (1,1). These results may guide Turkey’s MENR and other in-
OGM (1,1) performs more accurate predictions than the selected stitutions related to energy production in determining the future
models. energy policies.
We then predicted Turkey’s total electricity demand by 2025 by In order to ensure security of electricity supply, it needs to
using both direct and iterative OGM (1,1) methods. Fig. 4 shows the determine energy policies, take measures and make investments.
prediction values graphically while the prediction values are given Using the predicted values in this study, installed capacity which is
Table 4. required according to the resources type has been calculated for the
future. These obtained values for the security of electricity supply
3.4. Electricity supply of Turkey have been presented to the attention of all relevant institutions
such as public institutions and the private sector.
As the other countries, Turkey produces the electricity from As a future work, grey forecasting model based on causeeeffect
primary sources such as coal, hydro, gas etc. Although there are relationship as well as time series grey modeling will be studied in
several built plants, these are not sufficient. In addition, Turkey is a order to further improve the prediction accuracy in energy
country lacking in natural gas and oil, so it wants to reduce the rates demand.
of electricity generation on these resources. Turkey started a new
energy program in 2000s. According to this program, several hy- References
droelectric and two nuclear power plants will be built. Turkey plans
to reduce the rate of imported gas and increase the share of do- [1] MENR Report. The Republic of Turkey Ministry of Energy and Natural Re-
sources: 2011 annual report; 2011 [in Turkish].
mestic and clean energy resources in the production of electrical [2] EUAS sector report of electric generation; 2011 [in Turkish].
energy [54]. In order to obtain supply values of primary energy [3] TEIAS projection of Turkey electricity capacity for 2011-2020 [in Turkish].
sources, the utilization ratios of energy sources for Turkey elec- [4] Saab S, Badr E, Nasr G. Univariate modeling and forecasting of energy con-
sumption: the case of electricity in Lebanon. Energy 2001;26:1e14.
tricity production in 2015 and 2020 have been taken from TEIAS, [5] An N, Zhao W, Wang Z, Shang D, Zhao E. Using multi-output feedforward
then the ratios of 2025 have generated by energy policies of Turkey neural network with empirical model decomposition based signal filtering for
[3,54]. These ratios are given in Table 5. Using the ratios that given electricity demand forecasting. Energy 2013;49:279e88.
[6] Pao HT. Comparing linear and nonlinear forecasts for Taiwan’s electricity
in Table 5, required values of primary energy resources of Turkey consumption. Energy 2006;31:2129e41.
have calculated according to the electricity demand forecasts ob- [7] Gonzalez RE, Jaramillo MM, Carmona FD. Monthly electric Energy demand
tained by Direct OGM (1,1). These values of primary energy re- forecasting based on trend extraction. IEEE Trans Power Syst 2006;21(4):
1946e53.
sources are given in Table 6.
[8] Mohamed Z, Bodger P. Forecasting electricity consumption in New Zealand
using economic and demographic variables. Energy 2005;30:1833e43.
4. Conclusion [9] Mirasgedis S, Sarafidis Y, Georgopoulou E, Kotroni V, Lagouvardos K, Lalas DP.
Modeling framework for estimating impacts of climate change on electricity
demand at regional level: case of Greece. Energy Convers Manag 2007;48:
In this study, the prediction performance of the basic GM (1,1) 1737e50.
has been improved by using an algorithm which is called optimized [10] Al-Shobaki S, Mohsen M. Modeling and forecasting of electrical power de-
mands for capacity planning. Energy Convers Manag 2008;49:3367e75.
grey forecasting model. The horizontal adjustment coefficient (q) [11] Bianco M, Manca O, Nardini S. Electricity consumption forecasting in Italy
and the length of subsequences (k) which are important factors in using linear regression models. Energy 2009;34:1413e21.
order to develop a successful model have been determined. Then, [12] Meng M, Niu D. Annual electricity consumption analysis and forecasting of
China based on few observations methods. Energy Convers Manag 2011;52:
two forecasting approaches called iterative and direct forecasting 953e7.
have been applied. The results show the superiority of direct [13] Amarawickma HA, Hunt LC. Electricity demand for Sri Lanka: a time series
forecasting approach over the iterative one. This superiority is analysis. Energy 2008;33:724e39.
[14] Zahedi G, Azizi S, Bahadori A, Elkamel A, Alwi SRW. Electricity demand esti-
resulted from the usage of real past data in the long term fore-
mation using an adaptive neuro-fuzzy network: a case study from the Ontario
casting while the iterative forecasting approach uses the predicted province-Canada. Energy 2013;49:323e8.
values. Total electricity demand of Turkey has been predicted by [15] Hamzacebi C, Kutay F. Yapay sinir ag ları ile Türkiye elektrik enerjisi tüketi-
2025 using OGM (1,1). minin 2010 yılına kadar tahmini. Gazi Üniversitesi Mühendislik-Mimarlık
Fakültesi Dergisi 2004;19(3):227e33.
As the predicted values are of great importance in guiding future [16] Yumurtaci Z, Asmaz E. Electric energy demand of Turkey for the year 2050.
investments, decision-makers should consider different prediction Energy Source 2004;26:1157e64.
 C. Hamzacebi, H.A. Es / Energy 70 (2014) 165e171 171

[17] Ozturk HK, Ceylan H, Canyurt OE, Hepbasli A. Electricity estimation using ge- [36] Pao HT, Fu HC, Tseng CL. Forecasting of CO2 emissions, energy consumption
netic algorithm approach: a case study of Turkey. Energy 2005;30:1003e12. and economic growth in China using an improved grey model. Energy
[18] Tunc M, Camdali U, Parmaksizoglu C. Comparison of Turkey’s electrical energy 2012;40:400e9.
consumption and production with some European countries and optimization [37] Tang HWV, Yin MS. Forecasting performance of grey prediction for education
of future electrical power supply investments in Turkey. Energy Policy expenditure and school enrollment. Econ Educ Rev 2012;31:452e62.
2006;34:50e9. [38] Truong DQ, Ahn KK. Wave prediction based on a modified grey model
[19] Hamzacebi C. Forecasting of Turkey’s net electricity energy consumption on MGM(1,1) for real-time control of wave energy converters in irregular waves.
sectoral bases. Energy Policy 2007;35:2009e16. Energy 2012;43:242e55.
[20] Erdogdu E. Electricity demand analysis using cointegration and ARIMA [39] Hsu CC, Chen CY. Applications of improved grey prediction model for power
modelling: a case study of Turkey. Energy Policy 2007;35:1129e46. demand forecasting. Energy Convers Manag 2003;44:2241e9.
[21] Sumer KK, Goktas O, Hepsag A. The application of seasonal latent variable in [40] Yao AWL, Chi SC, Chen JH. An improved Grey-based approach for electricity
forecasting electricity demand as an alternative method. Energy Policy demand forecasting. Electr Power Syst Res 2003;67:217e24.
2009;37:1317e22. [41] Zhou P, Ang BW, Poh KL. A trigonometric grey prediction approach to fore-
[22] Kavaklioglu K, Ceylan H, Ozturk HK, Canyurt OE. Modeling and prediction of casting electricity demand. Energy 2006;31:2839e47.
Turkey’s electricity consumption using Artificial Neural Networks. Energy [42] Akay D, Atak M. Grey prediction with rolling mechanism for electricity de-
Convers Manag 2009;50:2719e27. mand forecasting of Turkey. Energy 2007;32:1670e5.
[23] Toksari MD. Estimating the net electricity energy generation and demand [43] Huang M, He Y, Cen H. Predictive analysis on electric power supply and de-
using the ant colony optimization approach: case of Turkey. Energy Policy mand in China. Renew Energy 2007;32:1165e74.
2009;37:1181e7. [44] Bianco M, Manca O, Nardini S, Minea AA. Analysis and forecasting of
[24] Kucukali S, Baris K. Turkeys short-term gross annual electricity demand nonresidential electricity consumption in Romania. Appl Energy 2010;87:
forecast by fuzzy logic approach. Energy Policy 2010;38(5):2438e45. 3584e90.
[25] Demirel O, Kakilli A, Tektas M. Anfis ve arma modelleri ile elektrik enerjisi yük [45] Kumara U, Jain VK. Time series models (Grey-Markov, Grey Model with rolling
tahmini. Gazi Üniversitesi Mühendislik-Mimarlık Fakültesi Dergisi mechanism and singular spectrum analysis) to forecast energy consumption
2010;25(3):601e10. in India. Energy 2010;35:1709e16.
[26] Kavaklioglu K. Modeling and prediction of Turkey’s electricity consumption [46] Lin Y, Liu S. A historical introduction to grey systems theory. In: Proceedings
using support vector regression. Appl Energy 2011;88:368e75. of IEEE international conference on system, man and cybernetics; 2004.
[27] Dilaver Z, Hunt LC. Turkish aggregate electricity demand: an outlook to 2020. pp. 2403e8.
Energy 2011;36:6686e96. [47 ] Liu S, Lin Y. Grey information: theory and practical applications. Berlin:
[28] Arisoy I, Ozturk I. Estimating industrial and residential electricity demand in Springer; 2006.
Turkey: a time varying parameter approach. Energy 2014;66:959e64. [48] Deng JL. Introduction to grey system theory. J Grey Syst 1989;1:1e24.
[29] Deng JL. Control problems of grey systems. Syst Control Lett 1982;1:288e94. [49] Lin CT, Hsu PF. Forecast of non-alcoholic beverage sales in Taiwan using the
[30] Li PL. Grey forecasting of earthquake (M greater than or equal to 6.0) fre- grey theory. Asia Pac J Mark Logist 2002;14(4):3e12.
quency in China mainland. J Grey Syst 1990;2(2):133e51. [50] Jiang Y, Yao Y, Deng S, Ma Z. Applying grey forecasting to predicting the
[31] Sun G. Prediction of vegetable yields by grey model GM(1,1). J Grey Syst operating energy performance of air cooled water chillers. Int J Refrig
1991;2:187e97. 2004;27(4):385e92.
[32] Zhu ZX, Cao HX. Prediction of El Nino events using grey model. J Trop [51] Tseng FM, Yu HC, Tzeng GH. Applied hybrid grey model to forecast seasonal
Meteorol 1998;4:359e65. time series. Technol Forecast Soc Change 2001;67:291e302.
[33] Hsu CI, Wen YU. Improved Grey prediction models for trans-Pacific air pas- [52] Liu S, Lin Y. Grey systems: theory and application. Berlin: Springer; 2010.
senger market. Transp Plan Technol 1998;22:87e107. [53] Pi D, Liu J, Qin X. A grey prediction approach to forecasting energy demand in
[34] Wang YF. Predicting stock prices using fuzzy grey prediction system. Expert China. Energy Source 2010;32:1517e28.
Syst Appl 2002;22:33e9. [54] MENR strategy document of electrical energy markets and supply security;
[35] Mao M, Chirwa EC. Application of grey model GM(1, 1) to vehicle fatality risk 2009 [in Turkish].
estimation. Technol Forecast Soc Change 2006;73:588e605.