Integrating Pakistan’s Electricity Demand with

Demographic and Energy indicators: Analysis

and Forecast
Amjad Ali1* Abdul Hameed Memon 2
1Centerof Research Excellence in 2Facultyof Engineering Sciences and
Renewable Energy, Research Institute, King Technology (FEST), Hamdard University
Fahd University of Petroleum & Minerals, Karachi 74600, Pakistan
Dhahran 31261, Saudi Arabia
[email protected]
[email protected]
Kashif Irshad1 Syed Hassan Arif 2
1Centerof Research Excellence in 2Facultyof Engineering Sciences and
Renewable Energy, Research Institute, King Technology (FEST), Hamdard University
Fahd University of Petroleum & Minerals, Karachi 74600, Pakistan
Dhahran 31261, Saudi Arabia
[email protected]
[email protected]

is driven by several economic indicators as Rafindadi and

Abstract—Pakistan is facing significant electricity outages for the Ozturk analyzed in [11]. Urbanization rate in Pakistan is also
last couple of decades. The main reasons of this issue are increasing rapidly. In 2016, the urban population has reached
increasing supply and demand gap and lack of planning in power up to 40% of the total population which also a reason of extreme
sector considering effective indicators. Analysis and forecasting of
EC. It has created a wide gap in supply and demand. It is
electricity demand of a country play a key role for its electricity
sector’s supply planning and development. Statistically, electricity estimated by 2025 that it nearly half of the total population will
consumption is directly proportional to different economic, be living in [12].
demographic and energy indicators. In this paper, original model According to the National Electric Power and Regulatory
based on Multiple Linear Regression has been proposed to Authority (NEPRA), in 2016 as fiscal year ending off on 30th
investigate the relationship between electricity demand of June, the demand and supply gap has increased up to 5,633
Pakistan with three indicators including population, energy use
MW. Table I shows Pakistan’s combined electricity generation
per capita and fossil fuel energy consumption on data from 1975
to 2010. Also, the proposed model is used to project the estimated and peak demand for both National Transmission & Dispatch
electricity demand with precision for future years from 2011 to Company (NTDC) and K-Electric (KE) sector for the last five
2025. The aim of this quantitative analytical research is to provide years [13].
useful information and insight to the planning regime,
policymakers, and decision-making authorities for short- and TABLE I. DEMAND AND SUPPLY DURING PEAK HOURS
long-term growth of electricity sector of Pakistan.
Generation Capability Peak Demand Deficit
Year
(MW) (MW) (MW)
Keywords—Electricity demand, electricity consumption, load 2012 14,483 21,536 -7,053
forecasting, demographic indicator, economic indicator, 2013 16,846 21,605 -4,759
regression, Pakistan electricity market 2014 18,771 23,505 -4,734
2015 19,132 24,757 -5,625
I. INTRODUCTION 2016 20,121 25,754 -5,633

Utilities, operators and policy-makers are in need to have

I n any country, the planning and forecasting processes for
electricity demand (ED) are extremely important for the
development of its power sector and affects the energy policy-
clear projections of data for future years to make decisions for
managing the grid condition.
The structure of this paper in the later part summarizes as;
makers and decision-making regime [1], [2]. Electricity
Section II defines different demographic and energy variables
consumption (EC) is explicitly related to the social,
taken for correlation with ED, while the Section III documents
demographic, economic and technological growth as studied
related studies conducted to forecast electricity demand with
and surveyed by various researchers [3]–[5]. Also, different
regression method. The proposed model is discussed in Section
methods and approaches for ED forecasting has been applied,
IV confirming the relationship to the selected independent
which includes econometric, traditional, modified,
variable. Section V describes the estimation based on the
computational techniques and other artificial intelligence (AI)
previous data and proposed model. Conclusion and the
based models [6]–[10].
limitations of this study are discussed in Section VI.
With increasing ED worldwide, which is caused by extreme
growth in population, standards of living and urbanization; EC
978-1-7281-5404-6 ©2019 IEEE

Authorized licensed use limited to: INTERNATIONAL ISLAMIC UNIVERSITY. Downloaded on July 29,2020 at 09:57:20 UTC from IEEE Xplore. Restrictions apply.
 2

II. SELECTED VARIABLES AND THEIR ANALYSIS

Year-wise data was collected from different official
authorities and databases for the period of 1975-2010. Pakistan
Ministry of Finance’s annual report; Economic Survey of
Pakistan and NEPRA’ annual report; State of Industry Report
were used for extraction of data of EC [13], [14]. The World
Bank database was used for the data of population, total energy
use per capita (EUpc) in terms of kilograms of oil equivalent
and fossil fuel energy consumption (FFEC) in percentages of
total EC [15]. See Appendix I for the complete data set.
A. Economic Indicators as Variables
A variety of forecasting models have been developed on Fig. 2. Growth trend of energy use per capita
different economic, demographic and energy variables [16]–
[18]. For our study, the variables we have selected are
population, EUpc, and FFEC of Pakistan. As variables these are
highly correlated with the EC of the country, computed by the
Pearson correlation coefficient as shown in Table II, it is more
likely to ensure the dependency of ED on selected indicators
[19].

TABLE II. CORRELATION MATRIX FOR SELECTED VARIABLES

Energy Fossil fuel

Electricity Total
Use per energy
Demand Population
capita consumption
Electricity
1 0.9879 0.9925 0.9770
Demand Fig. 3. Growth trend of fossil fuel energy consumption
Total Population 1 0.9956 0.9918
Energy Use per
Like Pakistan’s population, the trend of energy use of the
1 0.9943 country is also a straight line with a positive slope. It is also
capita
Fossil fuel energy observed that there are several spikes occurred during the
1
consumption selected period of years, in which 1987, 1992, 1999 and 2004
are noticeable with 6.4%, 3.7%, 3.1% and 4.9% of immediate
B. Trends of Selected Parameters growth, respectively.
Another economic indicator that is selected as our third
The population of Pakistan has increased rapidly in the
variable is FFEC and historical data shows consistent linear
selected range of years. Dramatic changes in social behavior
growth. In [20], it is examined that the relationship between
have brought urbanization and which has also increased from
FFEC and economic growth is uniform that leads to uniformity
26.34% to 36.60% of the total populations [15] by 2010. This
with the EC of the country.
change has a great impact on the total EC of the country.
Figure 1, Figure 2 and Figure 3 are the depiction of historical
data to overview the linear trends of our chosen variables.

III. STUDIES ON ED FORECASTING

A new modeling approach for ED forecasting is introduced
in a long-term scenario through ordinal regression analysis by
Angelopoulos et al. in [6]. Historical data have collected for the
period from 1999 to 2015 from Greek interconnected power
systems. Proposed two models have analyzed and verified with
different statistical accuracy indicators. It was observed that the
GDP significantly influences ED. Estimations have been
carried out for total net ED for the years from 2016 to 2025 and
found that ED forecast model with energy efficiency (EE)
Fig. 1. Growth trend of population actions reduce the future ED growth. R2 values are 0.9618 and
0.970 for model with EE and model without EE, respectively.
Kialashaki et al. [8], have investigated energy demand in the
residential sector of United States using Artificial Neural
Networks (ANN) and MLR. Several factors that effects the

Authorized licensed use limited to: INTERNATIONAL ISLAMIC UNIVERSITY. Downloaded on July 29,2020 at 09:57:20 UTC from IEEE Xplore. Restrictions apply.
 3

energy consumptions are considered as predictor variables for PT = C11 + C12 · Y (3)
analysis such as space and water heating, cooling, lightings,
dwelling size, number of occupants, etc. Three models for MLR EU = C21 + C22 · Y (4)
and ANN are suggested and analyzed. Historical data from
1984 to 2010 is collected. Results show that the selected FFE = C31 + C32 · Y (5)
variables are strongly correlated with energy consumption of
the residential sector with correlation coefficient (%) values of Where ÊD is estimated annual electricity demand
95.69%, 97.06% and 97.16% for MLR models respectively. PT is estimated total population
Similarly, ANN models have percentage values of R2 as EU is estimated energy use and
98.22%, 98.48%, and 98.95% respectively. Estimations carried FFE is the estimated fossil fuel energy consumption.
out for period of 2011-2030 and observed that there is only 2% C11, C12, C21, C22, C31, and C32 are the regression coefficients
change in residential energy demand for next 20 years. and ɛi represents the error of estimation.
In [10] Bianco et al., have investigated the influence of
economic and demographic indicators in residential EC in Italy. Eq. (1) is the proposed function of regression and Eq. (2) is
Data from 1970-2007 is considered for the regression analysis. the proposed model for examining the EC correlation with
The predictor variables are GDP, population and GDP per population, EUpc, and FFEC of the country. Eqs. (3), (4) and (5)
capita. Two objectives are defined: (1) to find the short- and are the proposed unitary linear regression models over the past
long-run price elasticity of the selected indicators for both years.
domestic and non-domestic sector and (2) to forecast the The quality of fit of all three performed linear regressions can
consumption up to 2030 with regression model and compare the be validated with the coefficient of determination R2, which
outcomes. Results have shown that GDP and GDP per capita measures the proportion of variability described by the fitted
changes are quite relevant to the change in EC. proposed model [21]. As shown in Table III, which is the output
Gul et al., [19] have performed ED forecasting through two summary, provides the illustration of a good fit (R2 ≈ 1.0).
models i.e. (i) Uni-variate Time Series and (ii) MLR. An
econometric model is developed based on three demographic TABLE III. RESULTS OF PROPOSED UNITARY LINEAR REGRESSIONS
variables of population, gross domestic product (GDP) and per Fossil fuel
Energy Use per
capita income of Pakistan. Data from the year 1992 to 2010 is Population
capita
energy
consumption
used to regress the proposed model. Results show the adjusted
R2 0.9988 0.9785 0.9498
R2 value of 0.98 and error deviation range of ±2.5. Estimations
Adjusted R2 0.9988 0.9778 0.9483
have been made with the proposed model until 2025. 1.0774 10.4691 1.9339
Standard Error
Ndiaye and Gabriel have performed regression models of
EC for housing units in Ontario Canada, by collecting data from
Successively, the regression coefficients (RC) of Eqs. (3), (4)
surveys and smart meters in [21]. Out of 59 predictors from the
and (5) are obtained and used as follows:
database, variable selection technique of Hawkins used to
reduce up to 85% of the variables, resulting in only 9 to be
PT = (-5854.5723) + (2.9964) · Y (6)
selected. Their analysis with selected 9 predictor variables
shows R2 values of 0.79 and adjusted R2 valued of 0.75 with all
EU = (-12760.0256) + (6.6092) · Y (7)
of the variables are statistically significant at 95% confidence
level.
FFE = (-1516.6474) + (0.7873) · Y (8)
IV. PROPOSED MODEL
After unitary regressions, MLR is performed where the
Among all other forecasting approaches, our proposed response variable is ED, while three predictor variables are
model is based on the MLR method. Our model has one chosen indicators. The proposed model of ED is highly
dependent variable, that is, annual electricity demand and a set satisfactory as the R2 value is 0.9966, as shown in Table IV. This
of three independent variables, on which our projections are is representing that the change of ED is 99.6% related with the
based. The aim of this proposed model is to justify the fact that combination of our three selected indicators. With such R2 value,
the ED is correlated with our selected indicators and to predict highly accurate estimations can be performed.
the data with precision for future.
Model of MLR is developed to explore the relationship TABLE IV. RESULTS OF PROPOSED EC MULTIPLE LINEAR REGRESSIONS
between the EC and the three chosen independent variables and R2 Adjusted R2 Standard Error
to predict the ED for the future years.
Electricity Demand 0. 9966 9962 1439.29
The following is our proposed model:
Subsequently, the all RCs of the performed MLR are
ÊD = f (PT, EU, FFE) (1)
obtained and used in Eq. (5) as shown below:
ÊD = C0 + C1 · PT + C2 · EU + C3 · FFE + ɛi (2)
ÊD = (-55771.9088) + (512.4716) · PT + (360.1083) · EE

Authorized licensed use limited to: INTERNATIONAL ISLAMIC UNIVERSITY. Downloaded on July 29,2020 at 09:57:20 UTC from IEEE Xplore. Restrictions apply.
 4

+ (-2158.9564) · FFE (9) necessary demand of the country and electricity supply has to
meet it. Also, avoiding the upper bound is recommended [22].
The validation of the proposed model can be checked and Figure 5 shows comparative and cumulative view of the actual
seen with the scatter plots of predicted ED and standard residuals figures and quite accurate estimated ED of Pakistan till 2025.
as shown in Figure 4. It is observed that the fit is significantly Figure 6 shows the exploded view of the proposed model.
best with the range of ±2.2 with no such particular or recurring
pattern.

Fig. 4. Standard Residual plot of the proposed model

V. ESTIMATIONS
As observed in the previous section that with the proposed Fig. 5. A comparative view of the proposed model
model, R2 value is very close to 1; showing that the fitted model
is significantly good and the independent variable and dependent
variables are highly correlated. Through the proposed model,
estimations with precision can be carried out for deriving overall
ED of Pakistan in the future.
Table V shows the predicted values that have been obtained
using the Eqs. (6) to (9). It can be seen that in 2025, the ED of
Pakistan will be almost 140% as of ED in 2011.

TABLE V. ESTIMATIONS FROM PROPOSED MODELS

Year PT EU FFE ÊD
(Y) (Millions) (Kgoe) (% of total) (GWh)
2011 171.20 531.09 66.72 79,166
2012 174.20 537.70 67.51 81,382
2013 177.19 544.31 68.29 83,597
2014 180.19 550.92 69.08 85,813
2015 183.18 557.52 69.87 88,029
2016 186.18 564.13 70.66 90,245
2017 189.18 570.74 71.44 92,460
Fig. 6. Exploded view of the proposed model
2018 192.17 577.35 72.23 94,676
2019 195.17 583.96 73.02 96,892
VI. CONCLUSION AND LIMITATIONS
2020 198.17 590.57 73.81 99,108
2021 201.16 597.18 74.59 101,323 In this paper, the proposed original model provides more
2022 204.16 603.79 75.38 103,539 accurate estimations of ED of Pakistan that aim to help utilities,
2023 207.16 610.40 76.17 105,755 operators, policy- and decision-makers for the growth of power
2024 210.15 617.01 76.96 107,971 sector of Pakistan. Data of 36 years have collected from several
2025 213.15 623.62 77.74 110,186 authentic sources for the analysis and forecasting. Proposed
model based on MLR is fitted with 1 response variable that is,
Later, accumulated projection has been made. Upper and ED, and 3 predictor variables, that are, population, EUpc, and
lower bounds of error have been constructed for accurate FFEC, over time. It is observed that the ED is highly correlated
predictions and to avoid variations obtained by regression with chosen indicators, resulting in R2 value equals to 0.9966
through proposed model. Focusing on attaining lower bound is (96.6%) that support the proposed models to project future
essential for electricity utilities as it will be the minimum and estimations with accuracy. Pakistan ED has been estimated from

Authorized licensed use limited to: INTERNATIONAL ISLAMIC UNIVERSITY. Downloaded on July 29,2020 at 09:57:20 UTC from IEEE Xplore. Restrictions apply.
 5

2011 to 2015 in which selected variable are integrated and also [2] S. Al-Shobaki and M. Mohsen, “Modeling and forecasting of electrical
power demands for capacity planning,” Energy Conversion and
verified with residual plot. This original study supports power Management, vol. 49, no. 11, pp. 3367–3375, Nov. 2008.
utilities, operators and governing authorities to address issues [3] S. Tiba and A. Omri, “Literature survey on the relationships between
related to ED and helps government, investors and market energy, environment and economic growth,” Renewable and
Sustainable Energy Reviews, vol. 69, pp. 1129–1146, Mar. 2017.
players in the development of energy management systems,
[4] J. E. Payne, “A survey of the electricity consumption-growth
Demand-side management and demand response of Smart grid. literature,” Applied Energy, vol. 87, no. 3, pp. 723–731, Mar. 2010.
Different approaches are available for estimation and [5] N. Antonakakis, I. Chatziantoniou, and G. Filis, “Energy consumption,
forecasting of electricity i.e. statistical, stochastic and AI [1]. In CO 2 emissions, and economic growth: An ethical dilemma,”
Renewable and Sustainable Energy Reviews, vol. 68, pp. 808–824, Feb.
this paper, statistical-based method MLR is used to forecast ED 2017.
based on historical data. It is suggested that estimations could [6] D. Angelopoulos, J. Psarras, and Y. Siskos, “Long-term electricity
also be made with other different stochastic and AI-based demand forecasting via ordinal regression analysis: The case of
Greece,” 2017, pp. 1–6.
techniques with accuracy. [7] L. G. B. Ruiz, R. Rueda, M. P. Cuéllar, and M. C. Pegalajar, “Energy
consumption forecasting based on Elman neural networks with
VII. APPENDIX I evolutive optimization,” Expert Systems with Applications, vol. 92, pp.
380–389, Feb. 2017.
[8] A. Kialashaki and J. R. Reisel, “Modeling of the energy demand of the
TABLE VI. DATASET USED FOR THE PROPOSED MODEL
residential sector in the United States using regression models and
artificial neural networks,” Applied Energy, vol. 108, pp. 271–280,
Fossil fuel Estimated
Population Energy Use Aug. 2013.
Year energy Electricity [9] N. Fumo and M. A. Rafe Biswas, “Regression analysis for prediction
per capita
(Y) consumption Demand of residential energy consumption,” Renewable and Sustainable Energy
(Millions) (Kgoe) (% of total) (GWh) Reviews, vol. 47, pp. 332–343, Jul. 2015.
1975 66.79 304.67 38.09 7,430 [10] V. Bianco, O. Manca, and S. Nardini, “Electricity consumption
1976 68.82 299.17 36.52 7,428 forecasting in Italy using linear regression models,” Energy, vol. 34, no.
1977 70.95 302.56 37.68 7,820 9, pp. 1413–1421, Sep. 2009.
1978 73.20 305.43 37.78 8,913 [11] A. A. Rafindadi and I. Ozturk, “Effects of financial development,
economic growth and trade on electricity consumption: Evidence from
1979 75.58 312.86 39.32 10,566
post-Fukushima Japan,” Renewable and Sustainable Energy Reviews,
1980 78.07 317.14 40.32 10,619 vol. 54, pp. 1073–1084, Feb. 2016.
1981 80.69 328.89 42.57 11,919 [12] Department of Economic and Social Affairs Population Division,
1982 83.43 336.92 44.11 13,290 United Nations, “An overview of urbanization, internal migration,
1983 86.26 343.66 44.98 14,757 population distribution and development in the world,” United Nations
1984 89.18 343.53 44.76 16,384 Expert Group Meeting on Population Distribution, Urbanization,
1985 92.17 350.10 46.28 18,336 Internal Migration and Development, 14-Jan-2008. [Online].
1986 95.21 353.68 46.60 20,398 Available:
1987 98.30 376.34 49.67 22,454 http://www.un.org/esa/population/meetings/EGM_PopDist/P01_UNP
1988 101.42 382.91 50.60 25,919 opDiv.pdf. [Accessed: 10-Dec-2017].
[13] “State of Industry Report 2016,” National Electric Power Regulatory
1989 104.53 392.44 52.02 27,652
Authority, 2016.
1990 107.61 398.67 52.68 29,865 [14] “Pakistan Economic Survey 2015-16,” Ministry of Finance,
1991 110.63 397.61 52.31 32,922 Government of Pakistan, 2016.
1992 113.62 412.41 53.55 37,979 [15] The World Bank, “Pakistan Data,” 2017. [Online]. Available:
1993 116.58 422.34 54.91 39,099 https://data.worldbank.org/country/pakistan. [Accessed: 01-Nov-
1994 119.56 428.72 55.00 41,364 2017].
1995 122.60 436.69 55.85 43,964 [16] A. Vaghefi, M. A. Jafari, E. Bisse, Y. Lu, and J. Brouwer, “Modeling
1996 125.70 448.25 57.69 45,271 and forecasting of cooling and electricity load demand,” Applied
1997 128.85 448.93 57.83 46,858 Energy, vol. 136, pp. 186–196, Dec. 2014.
[17] C. Hamzacebi and H. A. Es, “Forecasting the annual electricity
1998 132.01 447.64 57.87 45,529
consumption of Turkey using an optimized grey model,” Energy, vol.
1999 135.16 461.65 59.57 48,227 70, pp. 165–171, Jun. 2014.
2000 138.25 463.43 59.41 51,586 [18] A. Anand and L. Suganthi, “Forecasting of Electricity Demand by
2001 141.28 460.73 58.93 53,552 Hybrid ANN-PSO Models:,” International Journal of Energy
2002 144.27 455.93 58.41 55,661 Optimization and Engineering, vol. 6, no. 4, pp. 66–83, Oct. 2017.
2003 147.25 466.58 58.93 60,532 [19] M. Gul, S. A. Qazi, and W. A. Qureshi, “Incorporating economic and
2004 150.27 489.77 60.81 64,701 demographic variablesfor forecasting electricity consumption in
2005 153.36 497.24 61.08 71,269 Pakistan,” 2011, pp. 1–5.
2006 156.52 507.20 61.94 76,472 [20] J. Asafu-Adjaye, D. Byrne, and M. Alvarez, “Economic growth, fossil
2007 159.77 523.27 63.34 77,118 fuel and non-fossil consumption: A Pooled Mean Group analysis using
proxies for capital,” Energy Economics, vol. 60, pp. 345–356, Nov.
2008 163.10 503.55 62.62 72,447
2016.
2009 166.52 500.18 62.09 76,649 [21] D. Ndiaye and K. Gabriel, “Principal component analysis of the
2010 170.04 496.34 61.44 79,338 electricity consumption in residential dwellings,” Energy and
Buildings, vol. 43, no. 2–3, pp. 446–453, Feb. 2011.
[22] R. E. Walpole, R. H. Myers, S. L. Myers, and K. Ye, Eds., Probability
& statistics for engineers & scientists, Ninth edition. Boston: Pearson,
VIII. REFERENCES 2017.
[1] A. R. Khan, A. Mahmood, A. Safdar, Z. A. Khan, and N. A. Khan, [23] M. Gilliland, U. Sglavo, and L. Tashman, Eds., Business forecasting:
“Load forecasting, dynamic pricing and DSM in smart grid: A review,” practical problems and solutions. Hoboken, New Jersey: Wiley, 2015.
Renewable and Sustainable Energy Reviews, vol. 54, pp. 1311–1322,
Feb. 2016.

Authorized licensed use limited to: INTERNATIONAL ISLAMIC UNIVERSITY. Downloaded on July 29,2020 at 09:57:20 UTC from IEEE Xplore. Restrictions apply.