Professional Documents
Culture Documents
Renewable Energy
journal homepage: www.elsevier.com/locate/renene
a r t i c l e i n f o a b s t r a c t
Article history: The amount of electricity generation and its availability to the residents of a country reflects its level of
Received 26 April 2019 development and economic condition. Water being one of the cheapest and renewable sources of energy,
Received in revised form is being used to produce one-quarter of the total electricity production in Pakistan. This article presents a
12 February 2020
forecasting study of hydroelectricity consumption in Pakistan based on the historical data of past 53
Accepted 28 February 2020
Available online 4 March 2020
years using Auto-Regressive Integrated Moving-Average (ARIMA) modeling. Based on the developed
forecasting equation, the hydroelectricity consumption was predicted up to the year 2030. For validating
the reliability of the forecasted data, the results were compared to the actual values which showed good
Keywords:
Hydroelectricity consumption
fit with minimum deviation. The forecasted values of hydroelectricity consumption revealed an average
ARIMA model annual increment of 1.65% with a cumulative increase of 23.4% up to the year 2030. The results were
Time series forecasting compared with the hydroelectricity generation plans of the Government of Pakistan for its effectiveness.
Hydropower A sensitivity analysis was also performed to study the relation of hydroelectricity consumption to the
Renewable energy annual population and GDP growth rate of the country. The research shall significantly prove to be useful
Pakistan in better planning and management of water resources of Pakistan for future.
© 2020 Elsevier Ltd. All rights reserved.
https://doi.org/10.1016/j.renene.2020.02.117
0960-1481/© 2020 Elsevier Ltd. All rights reserved.
2 R. Jamil / Renewable Energy 154 (2020) 1e10
Table 1
Existing installed capacity of major hydroelectricity plants in Pakistan [39].
2. Methodology of analysis [27]. Both of these conditions are fulfilled for this research and the
data is made eligible to be tested by using this method.
The data used for forecasting of hydroelectricity consumption in A typical ARIMA model is said to have the format of ARIMA (p, d,
Pakistan is obtained from the website of BP Public Limited Com- q), where p is the Auto-Regressive term, d is the order of differ-
pany, based in United Kingdom which is related with oil and energy encing required for the data to make it stationary and q is the
[40]. BP has expertise in exploration, extraction and production of Moving-Average term. Although this model is quite complicated
renewable and non-renewable energy resources and related works. but Box and Jenkins presented a comprehensive stepwise approach
Besides this, the company maintains a comprehensive record of for analysis and forecasting of a time series by using ARIMA models
statistics for energy production by various means and its con- in 1976. Their methodology became so popular that the ARIMA
sumption by countries around the world. The data is available for models are very often called as Box-Jenkins models.
download at no cost for the purpose of study and research. The The time series forecasting by using ARIMA models can be
hydroelectricity consumption data is available in Terawatt Hours performed in four basic steps namely, Identification, Estimation,
(TWh) units of measurement which ranges from 1965 to 2017, Diagnosis and Forecasting [28,29]. The whole process of Box-
considerably a good span of 53 years. The descriptive statistics of Jenkins methodology used in the research is shown as a flow
the data shows an increasing trend in hydroelectricity consumption chart in Fig. 2. Before the data is put into the Identification stage, it
with a minimum value of 5.78 TWh in 1965 and a maximum value is necessary to check the stationarity of the data. To check the
of 94.69 TWh happening in 2017. The mean of the data is found to stationarity of the data, unit root test is performed for the time
be 46.36 TWh/year with a standard deviation of 27.89 TWh. The series. The unit root test regresses the time series yt on its lagged
annual GDP and population growth rates of Pakistan for the same value yt-1 [10] and the results of this test show the nature of the
span of time are obtained from the official website of World Bank data in terms of stationarity.
[26] to perform sensitivity analysis. Statistical package EViews 10 is If the data is non-stationary, the estimation cannot be per-
used to execute all the statistical analysis for this research. formed and there is a need to differentiate the data repeatedly till it
becomes stationary. The number of times the data is differenced for
2.1. Time series analysis and ARIMA modeling its stationarity defines the order of differencing, d, of the ARIMA
model. Many types of unit root tests are found in the literature
A time series can be defined as a variable that is observed including Augmented Dickey-Fuller (ADF) test [30], Philips-Perron
repeatedly after a specified duration of time. The duration of test [31], Levin-Lin-Chu test [32], Breitung test [33], Im-Pesaran-
measurement could be as short as a fraction of second and could be Shin test [34] etc. However, ADF unit root test is used for the
as long as the years. The variable may show a single upward or research which is best suited for large samples given by the
downward trend or it could consist of multitrending structure over following regression equation [29].
the recorded period of time. The process of measuring the trend
and internal statistical structure of the time series is often termed X
p
as time series analysis. This process includes various statistical Dyt ¼ ayt1 þ dx’t þ bi Dyti þ εt (1)
i¼1
testing and calculations.
Many researchers around the world have used various types of
where Dyt is the first difference operator of the given time series, yt-
time series analysis techniques for forecasting as discussed in
1 is the lagged value, a and d are the parameters to be estimated, xt
Section 1. Artificial Neural Network (ANN) is quite common among
is the optional regressor which may consist of a constant and εt is
them which is based on machine learning approach and it is mostly
the white noise error term.
used to define a trend in the a time series data. Long-range Energy
The ADF test builds a parametric correction for higher-order
Alternatives Planning (LEAP) model is used for energy policy
correlation by supposing that the y series follows an auto-
analysis and climate change assessment. Grey Model (GM) is used
regressive process and adding p lagged difference terms of the
for forecasting when there are monotonous type of processes
dependent variable y to the test regression. This improved speci-
involved in a data but in this type of a model all the values are
fication is then used to test the hypothesis using the t-ratio.
assumed to be positive. Auto-Regressive Integrated Moving-
The total data used in this research is spread over 53 years which
Average (ARIMA) method has also become one of the most
is transformed to stationary nature. After differencing the data, to
important tool during the last decade for forecasting due to its wide
make it stationary, it remained at 52 which is well above the
range of applicability and flexibility of use. It is the combination of
requirement of minimum sample size, hence satisfying both con-
two individual statistical processes namely Auto-Regression and
ditions of ARIMA modeling.
Moving-Average techniques with an addition of Integrative mod-
ule. Although the above-mentioned methods have their own ad-
vantages over one another but on the other hand each model has
few limitations as well. A univariate ARIMA modeling technique is 2.1.1. Identification of ARIMA model
used for forecasting in this research due to the fact that it has In the first step of ARIMA modeling by using Box-Jenkins
proven to be good and reliable in forecasting a value based on its approach, the number of significant lags are observed from the
historic observations and it does not require any additional vari- correlogram of the stationary time series data. A correlogram
ables. Besides its flexibility of use, ARIMA modeling methodology shows the autocorrelation (ACF) and partial autocorrelation func-
also has a few limitations. One of those being that this method is tions (PACF) in the graphical and numerical form. These functions
appropriate only for a time series that is stationary in nature. A describe the pattern of temporal dependence in the time series data
series is said to be stationary when the autocovariance and mean of [35]. Both columns of ACF and PACF are studied and the number of
the series are independent of time and also it shows that there is no lags which are found significant, either beyond or very close to the
significant autocorrelation existing between the data points. Hence range of 95% confidence interval. The sequential number of sig-
the data is must to be transformed to a stationary nature before it is nificant lags shown by PACF gives the AR term, p, and the same
put to test. Another important limitation is the minimum sample shown by ACF gives the MA term, q, of the ARIMA model. The ACF is
size which is required to perform ARIMA modeling for better and the linear correlation between observations k time lags separately
reliable results. The minimum sample size is recommended as 50 of a time series Y and it is estimated by the following equation.
4 R. Jamil / Renewable Energy 154 (2020) 1e10
8 t1 k¼1
>
>
< Xk1
∅k ¼ tk ∅k1; j tkj (3)
>
> j¼1
k>1
: Xk1
P 1 ∅k1; j tkj
t¼kþ1 ðYt YÞðYtk YÞ j¼1
tk ¼ PT 2
(2)
t¼1 ðYt YÞ where tk is the autocorrelation estimated at lag k and t1 depicts the
value of ACF at first lag when k ¼ 1.
where Y is the sample mean of Y, T is the number of observations
and k show the number of lags. The PACF of a series measures the
correlation of Y values that are k periods apart after removing the 2.1.2. Estimation
correlation from the superseding lag. The PACF is measured by After the significant lags have been identified, the tentative
using the following algorithm. ARIMA (p, d, q) models are prepared and equations are developed
R. Jamil / Renewable Energy 154 (2020) 1e10 5
Table 2
ADF test results for statistical hypothesis.
Fig. 5. The correlogram of the differenced hydroelectricity consumption data. Fig. 6. The correlogram of the residuals for selected ARIMA model.
Table 3
Results of tentative ARIMA models showing comparison of parameters.
Sr. No. ARIMA Model No. of Significant Coefficients Variance (s2) Adjusted R2 AIC Value SBIC Value
Table 4
Results of adjusted ARIMA models showing comparison of parameters.
Sr. No. ARIMA Model No. of Significant Coefficients Variance (s2) Adjusted R2 AIC Value SBIC Value
models, model no. 11 ARIMA (9, 1, 7 & 19) comes out to be the Hence the adjusted ARIMA model (9, 1, 7 & 19) was finally
optimum model. selected for forecasting the hydroelectricity consumption for
For further investigation, the correlogram of the residuals for Pakistan up to the year 2030. After getting the results from EViews
this model was observed and it was found to be flat showing that all 10 by going through the above process the general forecasting
the required data and information has been captured in it and all equation (Eq. (9)) can now be written as follows.
lags were well within the limit of confidence interval, as shown in
Fig. 6. yt ¼ 1:696 þ 0:435 yt9 0:417 εt7 0:381 εt19 (11)
8 R. Jamil / Renewable Energy 154 (2020) 1e10
Fig. 7. Graph showing forecasted hydroelectricity consumption along with historical data.
R. Jamil / Renewable Energy 154 (2020) 1e10 9
Table 6
Sensitivity analysis for hydroelectricity consumption of Pakistan against annual GDP
and population growth rate.
the population growth rate remains the same over time. Fig. 10
Fig. 9. Graph showing forecasted hydroelectricity consumption along with historical shows the graphical representation of the effect on the hydro-
data.
electricity consumption caused by the variation in GDP growth rate
for a constant population growth rate of 2%.
produced at the end of year 2018 [2]. By keeping a capacity factor of
55%, as calculated for hydroelectricity plants in Pakistan [37], the 4. Applicability of presented methodology for other countries
annual hydroelectricity generation would increase from 39.5 TWh
to 120 TWh by 2030. Fig. 9 shows the comparison of the forecasted Forecasting of hydroelectricity consumption can be performed
hydroelectricity consumption by using forecasting equation ob- for other developing countries by following the similar methodol-
tained through ARIMA modeling and generation by the year 2030. ogy presented in this research. The historical data of the country
From the figure it is clear that Pakistan is currently in deficit of under study can be obtained through the sources mentioned in
hydroelectricity generation. The lag is being covered by the Inde- Section 2 or any other governing authority. For a reliable statistical
pendent Power Producers (IPPs) which have either been estab- analysis of a time series by using ARIMA method, it is recom-
lished indigenously or have been hired on rent from external mended to have a minimum sample size of 50, which means that
sources. With all the projects streamlined for future construction, the past data for at least 50 years should be obtained [27]. The four
only in the year 2030, Pakistan would be able to catch up to the major steps of Box-Jenkins methodology for ARIMA modeling
hydroelectricity demand without depending on IPPs and producing (Fig. 2) can be followed to develop the forecasting equation and
the same through hydropower plants only. Major hydroelectricity further estimation of hydroelectricity consumption in future years
projects planned to be constructed up to the year 2030 include for any country.
Diamer-Bhasha dam, Dasu dam, Suki Kanari and Karot hydropower
plant with production capacity of 4,500 MW, 4,320 MW, 870 MW 5. Conclusion
and 720 MW respectively [38]. But as mentioned by Nicholas and
Buckley [2]; huge challenges are being faced for these projects to be Considering the electricity supply and demand condition in the
implemented as planned including financial and political issues. In country, this research was taken up to study the historical and
case of any delay in any of the projects, as mentioned by Abbasi future trend of electricity consumption produced by hydropower
et al. [38]; the country would be spending money on IPPs in future plants in Pakistan. Forecasting studies hold an important signifi-
as well and it would have to rely on other means of thermal and cance in planning and managing the resources for the future. The
nuclear power which would affect the country’s economy badly. obtained historical data was put to statistical analysis and ARIMA
modeling was performed to predict the consumption of hydro-
3.5. Effect of GDP and population on hydroelectricity consumption electricity up to the year 2030 for Pakistan. The results obtained by
the developed forecasting equation through ARIMA modeling were
After performing multivariate MLR analysis, the general form of validated and found to be matching with the actual available data,
regression equation (Eq. (10)) can now be written as follows.
showing reliability of the equation for forecasting. The forecast of hydropower consumption forecasting in China, Energy 36 (2011) 6542e6554.
[13] C. Yuan, S. Liu, Z. Fang, Comparison of China’s primary energy consumption
data showed an average annual increment of 1.65% in hydroelec-
forecasting by using ARIMA (the autoregressive integrated moving average)
tricity consumption with a cumulative increase of 23.4% by the year model and GM(1,1) model, Energy 100 (2016) 384e390.
2030. A brief comparison of this forecasted consumption values and [14] K.M. Cassiano, L.A.T. Junior, R.M. De Souza, M.L. De Menezes, J.F.M. Pessanha,
the planned hydroelectricity generation by the Government of R.C. Souza, Hydroelectric energy forecast, Int. J. Energy Statist. 1 (3) (2013)
205e214.
Pakistan was also presented. The comparison clearly showed that [15] U. Perwez, A. Sohail, S.F. Hassan, U. Zia, The long-term forecast of Pakistan’s
currently the country is in deficit of hydroelectricity and the gap is electricity supply and demand: an application of long range energy alterna-
being filled by using independent power producers (IPPs). If the tives planning, Energy 93 (2015) 2423e2435.
[16] A. Hussain, M. Rahman, J.A. Memon, Forecasting electricity consumption in
planned hydropower projects are constructed in time as per Pakistan: the way forward, Energy Pol. 90 (2016) 73e80.
schedule, only then the country shall be able to meet the demand of [17] M. Asif, Sustainable energy options for Pakistan, Renew. Sustain. Energy Rev.
hydroelectricity by the year 2030. A sensitivity analysis was also 13 (2009) 903e909.
[18] M.M. Rafique, S. Rehman, National energy scenario of Pakistan - current sta-
performed to study the relation of hydroelectricity consumption to tus, future alternatives, and institutional infrastructure: an overview, Renew.
the annual population and GDP growth rate of the country. The Sustain. Energy Rev. 69 (2017) 156e167.
results of the sensitivity analysis were found befitting and useful. [19] M.A. Sheikh, Renewable energy resource potential in Pakistan, Renew. Sus-
tain. Energy Rev. 13 (2009) 2696e2702.
[20] M.A. Sheikh, Energy and renewable energy scenario of Pakistan, Renew.
Author contribution section Sustain. Energy Rev. 14 (2010) 354e363.
[21] O. Rauf, S. Wang, P. Yuan, J. Tan, An overview of energy status and develop-
ment in Pakistan, Renew. Sustain. Energy Rev. 48 (2015) 892e931.
Rehan Jamil Conceptualization, Investigation, Methodology,
[22] M. Wakeel, B. Chen, S. Jahangir, Overview of energy portfolio in Pakistan,
Statistical Analysis, Original Draft Preparation, Reviewing and Applied Energy Symposium and Summit 2015: low carbon cities and urban
Editing, Revisions. energy systems e CUE2015, Energy Procedia 88 (2016) 71e75.
[23] G.D. Valasai, M.A. Uqaili, H.R. Memon, S.R. Samoo, N.H. Mirjat, K. Harijan,
Overcoming electricity crisis in Pakistan: a review of sustainable electricity
Declaration of competing interest options, Renew. Sustain. Energy Rev. 72 (2017) 734e745.
[24] M. Kamran, Current status and future success of renewable energy in Pakistan,
The authors declare that they have no known competing Renew. Sustain. Energy Rev. 82 (2018) 609e617.
[25] U. Zafar, T. Rashid, A.A. Khosa, M.S. Khalil, M. Rashid, An overview of imple-
financial interests or personal relationships that could have mented renewable energy policy of Pakistan, Renew. Sustain. Energy Rev. 82
appeared to influence the work reported in this paper. (2018) 654e665.
[26] WB, The World Bank. https://data.worldbank.org/country/pakistan, 2019.
(Accessed 12 November 2019).
References [27] G.E.P. Box, G.C. Tiao, Intervention analysis with applications to economic and
environmental problems, J. Am. Stat. Assoc. 70 (349) (1975) 70e79.
[1] P.R. Blood, Pakistan: A Country Study, sixth ed., Federal Research Division, [28] G.E.P. Box, G.M. Jenkins, G.C. Reinsel, G.M. Ljung, Time Series Analysis: Fore-
Library of Congress, Washington DC, 1995. ISBN: 0-8444-0834-4. casting and Control, John Wiley & Sons, Inc. Publication, 2015, ISBN 978-1-
[2] S. Nicholas, T. Buckley, Pakistan’s Power Future, Institute for Energy Eco- 118-67502-1.
nomics and Financial Analysis e IEEFA, USA, 2018. http://ieefa.org/wp- [29] EViews, EViews 10 User’s Guide II, IHS Global Inc., 2017b, ISBN 978-1-880411-
content/uploads/2018/11/Pakistans-Power-Future_December-2018.pdf. 38-4.
[3] M.E. Shaawat, R. Jamil, M.M. Al-Enezi, Analysis of challenges in sustainable [30] D.A. Dickey, W.A. Fuller, Distribution of the estimators for autoregressive time
construction industry by using analytic hierarchy process: a case study of series with a unit root, J. Am. Stat. Assoc. 74 (366a) (1979) 427e431.
Jubail industrial city, Saudi arabia, Int. J. Sustain. Real Estate Construct. Econ. 1 [31] P.C.B. Phillips, P. Perron, Testing for a unit root in time series regression,
(2) (2018) 109e122. Biometrika 75 (2) (1988) 335e346.
[4] U.K. Mirza, N. Ahmad, T. Majeed, K. Harijan, Hydropower use in Pakistan: past, [32] A. Levin, C.F. Lin, C.S.J. Chu, Unit root tests in panel data: asymptotic and finite-
present and future, Renew. Sustain. Energy Rev. 12 (2008) 1641e1651. sample properties, J. Econom. 108 (2002) 1e24.
[5] A. Siddiqi, J.L. Wescoat Jr., S. Humair, K. Afridi, An empirical analysis of the [33] J. Breitung, The local power of some unit root tests for panel data, Nonstation.
hydropower portfolio in Pakistan, Energy Pol. 50 (2012) 228e241. Panels Panel Cointegr. Dyn. Panels 15 (2000) 161e177.
[6] N.H. Mirjat, M.A. Uqaili, K. Harijan, G.D. Valasai, F. Shaikh, M. Waris, A review [34] K.S. Im, M.H. Pesaran, Y. Shin, Testing for unit roots in heterogeneous panels,
of energy and power planning and policies of Pakistan, Renew. Sustain. Energy J. Econom. 115 (2003) 53e74.
Rev. 79 (2017) 110e127. [35] EViews, EViews 10 User’s Guide I, IHS Global Inc., 2017a, ISBN 978-1-880411-
[7] R. Jamil, Estimation of sediment deposition at nausehri reservoir by Multiple 37-7.
linear regression and assessment of its effects on reservoir life and power [36] H. Caswell, N.S. Gassen, The sensitivity analysis of population projections,
generation capacity, Int. J. Adv. Thermofluid Res. 1 (2) (2015) 46e59. Demogr. Res. 33 (28) (2015) 801e840.
[8] K.B. Debnath, M. Mourshed, Forecasting methods in energy planning models, [37] SNC-Lavalin, NESPAK, National Power System Expansion Plan 2011-2030,
Renew. Sustain. Energy Rev. 88 (2018) 297e325. National Transmission and Despatch Company Ltd., Water and Power
[9] A.A. El-Desouky, M.M. El-Kateb, Hybrid adaptive techniques for electric-load Development Authority, Pakistan, 2011.
forecast using ANN and ARIMA, IEE Proc. Generat. Transm. Distrib. 147 (4) [38] A.H. Abbasi, F. Mahmood, M. Kamal, A. Baig, Pakistan Energy Vision 2035,
(2000) 213e217. Sustainable Development Policy Institute, Pakistan, 2014.
[10] E. Erdogdu, Electricity demand analysis using cointegration and ARIMA [39] http://www.wapda.gov.pk/index.php/projects/hydropower-development-in-
modelling: a case study of Turkey, Energy Pol. 35 (2007) 1129e1146. pakistan, 2016. (Accessed 20 March 2019).
[11] V.S. Ediger, S. Akar, ARIMA forecasting of primary energy demand by fuel in [40] https://www.bp.com/content/dam/bp/business-sites/en/global/corporate/
Turkey, Energy Pol. 35 (2007) 1701e1708. pdfs/energy-economics/statistical-review/bp-stats-review-2019-full-report.
[12] S. Wang, L. Yu, L. Tang, S. Wang, A novel seasonal decomposition based least pdf, 2018. (Accessed 15 February 2019).
squares support vector regression ensemble learning approach for