2021 IEEE Green Technologies Conference (GreenTech)

Time Series Analysis of Electricity Consumption Forecasting Using

ARIMA Model
Meftah Elsaraiti#1, Gama Ali#*2, Hmeda Musbah#*3, Adel Merabet#4, Timothy Little#*5
# Saint Mary’s University, Halifax, NS, Canada
#* Dalhousie University, Halifax, NS, Canada
#1
[email protected], #*[email protected], #*[email protected]
#4
[email protected], #*[email protected]

Abstract -- Power consumption is a very important factor in there are many factors that have an impact on electricity
smart grids for load management process. Forecasting energy consumption, which require information integration
consumption is the first step in dealing with load management. technologies for a comprehensive understanding. For this
For forecasting time series, the ARIMA models are one of the purpose, the researchers studied methodologies to collect
widely used models which showing encouraging results. In this
information related to electricity consumption and multi-
2021 IEEE Green Technologies Conference (GreenTech) | 978-1-7281-9139-3/21/$31.00 ©2021 IEEE | DOI: 10.1109/GREENTECH48523.2021.00049

study, ARIMA models were proposed to predict future

electricity consumption. The ACF and PACF plots were used as
factor energy consumption prediction models. In addition,
well as stationarity of the data to identify (p,d,q) values. The conducting a comprehensive analysis and obtaining an
results showed the accuracy and efficiency of the models and accurate assessment of energy consumption is the basis for
their ability to compete with current techniques for forecasting designing and transforming a more robust and efficient power
electricity consumption based on the use of the Mean Absolute grid [4]. Forecasting electricity consumption is an effective
Percentage Error (MAPE) to measure the accuracy of the measure that helps power grid designers and planners build
prediction, as the model was able to predict with an error of robust, adaptive, efficient, and economic smart grids. It is
4.332%. aimed at modeling electricity consumption under various
constraints along with environmental factors and the
Keywords--— ARIMA; Electricity consumption; Forecast,
regulations. A pre-estimated and calculated electricity
Time series
demand can be obtained based on the history data including
dates, economic, climate and so on [5]. This study aims to
I. INTRODUCTION
forecasted the electricity consumption in southern Tripoli
Libya is among the developing countries that strive to Libya for two weeks based on data from Jan 1st, 2016 to Mar
develop, advance and develop its economy, due to its direct 24, 2016 (12 Weeks) using the autoregressive integrated
dependence on oil, which represents about 97% of export moving average (ARIMA), mean absolute percentage error
earnings as national income. Energy is also one of the most (MAPE) is used to select the best model performance can be
important drivers of the economy, and among its sources is used for future forecasting.
electricity, and it is considered a vital commodity. It is not
possible to imagine improving the living conditions of the
population and economic and industrial development except
with electricity, and it has become as an economic standard II. LITERATURE REVIEW
that explains the progress or backwardness of a country. Many models have been developed to improve forecasting
Today, electricity consumption is one of the most important accuracy of electricity consumption. Studying the ability of
indicators for developing governments and making good the ARIMA model to compete with current technologies for
accuracy of long-term forecasting of electricity consumption forecasting electricity consumption using Mean Absolute
is vital to increasing energy productivity and avoiding costly Percentage Error (MAPE) to measure forecast accuracy [6].
mistakes [1]. Power forecasting is important for good ARIM models are used to predict the most appropriate period
planning of the electricity consumption as well as for the for individual household electrical energy consumption [7].
implementation of decision support systems which guide the The ARIMA model was used to forecast wind speed using
decision-making process of the power system [2]. Accurate common error ratio measures for model prediction accuracy
forecasting of electricity consumption plays a crucial role for [8]. Six time series methods namely, simple moving average
policy makers to formulate electricity supply policies. (SMA), weighted moving average (WMA), simple
However, generally limited data and variables cannot provide exponential smoothing (SES), Holt linear trend (HL), Holt-
beneficial information to obtain a satisfactory prediction Winters (HW) and centered moving average (CMA) were
accuracy [3]. With the recent developments in the field of applied and compared the accuracy of each to forecast
modern information technology, the smart grid has become monthly electricity consumption [9]. A comparative analysis
one of the main components of smart cities. In order to take of the SARIMA and ARIMA model was performed on the
full advantage benefits of the smart grid, smart planning and same data set forecasting the electricity consumption, and the
power planning is crucial. From a practical point of view, SARIMA model demonstrated better performance than the

2166-5478/21/$31.00 ©2021 IEEE 259

DOI 10.1109/GreenTech48523.2021.00049

Authorized licensed use limited to: Dalhousie University. Downloaded on September 01,2021 at 00:27:07 UTC from IEEE Xplore. Restrictions apply.
ARIMA model [10]. Classic 'ad hoc' statistical models, the electricity consumption, the electricity consumption data
advanced ensemble techniques, and neural networks were of in southern Tripoli from National Electricity Corporation
used to predict electricity power demand for a wholesale in Libya, collected from the electricity distribution unit of it.
power transmission company [11]. Combining seasonal and This data set contains the electricity consumption in unit (in
trend decomposition using Loess (STL) and ARIMA models MWh) from Jan 1st, 2016 to Mar 24, 2016 (12 Weeks).
were proposed to monthly electricity consumption forecasting
B. ARIMA Model Building Process
[12]. ARIMA model was used to predict electricity
consumption in the Philippines, and (p, d, q) values were Figure 1. Shows schematic representation of the Box-
determined using ACF and PACF plots as well as stationarity Jenkins for ARIMA approach.
of the data [13].

III. OVERVIEW OF ARIMA MODEL

The ARIMA model, also known as Box-Jenkins has been
widely applied with time series data of variables measured
over time [14]. ARIMA model includes three non-negative
integers variables (p, d, q), respectively are refer to the
autoregressive (AR) model shows that the value of a variable
in one period is related to the values of the same variable in
previous periods, integrated (I) shows the number of times
that the data value have to be “differenced” to obtain a
stationary data and moving average (MA) indicates to the
number of lagged values forecast errors in the prediction
equation [15]. Seasonal ARIMA models are usually
represented as ARIMA(p,d,q)(P,D,Q)m, where m refers to the
number of periods in each season, and the uppercase P,D,Q
Figure 1. Schematic Representation of the Box-Jenkins Methodology for
depict the autoregressive, differencing, and moving average ARIMA approach
terms for the seasonal part of the ARIMA model [16].
To estimate ARIMA model various terms values, the steps V. RESULTS AND DISCUSSION
involving in finding autocorrelation and partial
autocorrelation between the values of the data [17]. The electricity consumption in southern Tripoli - Libya during
Autocorrelation is the correlation of a time series with a the period from Jan 1st, 2016 to Mar 24, 2016 (12 Weeks) is
delayed copy of itself and is defined as ACF = corr (Xt, Xt+k). depicted in Figure. 2 represents the constituents of the time
Here Xt and Xt+k are the current observation and the series viz. are trend, seasonality and residual values. It can be
observation after k period respectively. Partial Auto- seen that the electricity consumption data has a slow
Correlation (PACF) [18] is the partial correlation of Xt+k with downward trend and has seasonality. Therefore, seasonal
Xt i.e. it controls the values of the time series at all shorter ARIMA was used for the forecasting.
lags which ACF does not. It is defined for positive lag only
with values lying between −1 and +1. Table 1 gives the idea
as how to make the estimation for initial values of ARIMA (p, Time series of data
4000
d, q) [19]. 3500
3000
Observation

TABLE I. MAIN CHARACTERISTICS OF ARMA (p, q) MODELS 2500

. 2000
AR (P) MA (q) ARMA (p, q)
1500
ACF Tails off Cuts off after Tails off
1000
lag q
500
PACF Cuts off after Tails off Tails off
0
Week10
Week11
Week1
Week2
Week3
Week4
Week5
Week6
Week7
Week8
Week9

Wee12

lag p

Time (Weekly)
IV. METHODOLOGY Figure.2. Electricity consumption data for 12 Weeks

A. Dataset Description
Day by day demand of electricity is increasing because of
uses of electrical and electronics instruments. To forecasting

260

Authorized licensed use limited to: Dalhousie University. Downloaded on September 01,2021 at 00:27:07 UTC from IEEE Xplore. Restrictions apply.
 Figure 3. Decomposition of additive time series

Observing 4 graphs closely in figure 3, we can find out if the Figure.6. The PACF graph for the first seasonal difference.
data satisfies all the assumptions of ARIMA modeling,
mainly, stationarity and seasonality we note that there is A function was created using all possible combinations of
seasonal, as well trend, so it is not surprising that the time parameters for fitting the models, the outcome was predicted
series is not stationary. These patterns revealed that the series using the models, and the model with the smallest MAPE was
is not stationary and therefore must be transformed before selected. The results showed that the best model that can be
attempting to form a stationary model. Therefore, the first used to forecast future electricity consumption is the seasonal
difference of the series is analyzed, the plot of the first ARIMA model (0, 1, 6) (1, 1, 6)7 with a MAPE of 4.332% as
difference data conclude that the data is stationary after the shown in Table 2.
seasonal first difference was taken as shown in Figure 4.
TABLE II. FORECAST ACCURACY OF THE MODEL USING MAPE

Model MAPE
SARIMA (6,1,0) (2,1,1) 7.193
SARIMA (6,1,0) (1,1,2) 6.872
SARIMA (6,1,0) (1,1,6) 4.332
Figure.4. The first difference of data
SARIMA (6,1,0) (1,1,5) 5.206
The ACF and PACF correlogram was plotted as shown in SARIMA (6,1,0) (1,1,3) 6.460
Figures 5 and 6, to select the suitable terms for the model. SARIMA (6,1,0) (1,1,1) 7.187
From the correlograms, it was observed that both the ACF and
PACF cuts the upper confidence level for the first time at lag Figure 7 shows the forecasted electricity consumption and the
value 0 and 6 respectively hence, the coefficients of both AR actual data for two weeks from Mar 25, 2016 to Apr 07, 2016.
and MA terms would 6 and zero i.e. p = 6 and q = 0.
4000 Actual
Electric consumption (MWH)

3500 Forecast
3000
2500
2000
1500
1000
500
0
Week10
Week11

Week13
Week14
Week1
Week2
Week3
Week4
Week5
Week6
Week7
Week8
Week9

Wee12

Weekly time series forecasting

Figure. 5. The ACF graph for the first seasonal difference
Figure.7. Actual data vs forecasted data.

261

Authorized licensed use limited to: Dalhousie University. Downloaded on September 01,2021 at 00:27:07 UTC from IEEE Xplore. Restrictions apply.
 VI. CONCLUSION [15] H. Nguyen and C.K. Hansen. “Short-term electricity load forecasting
with time series analysis.” In: 2017 IEEE International Conference on
Forecasting the energy consumption is playing a crucial role Prognostics and Health Management (ICPHM), Dallas, TX, pp. 214–
in directing plans, programs, load management and policies 221 (2017).
within the institution. The accuracy and the simplicity of the [16] G. Mahalakshmi, S. Sridevi and S. Rajaram. “A survey on forecasting
of time series data.” In: 2016 International Conference on Computing
forecasting model leads to improving proper planning and to Technologies and Intelligent Data Engineering (ICCTIDE 2016),
rational policy regarding production companies. Electricity Kovilpatti, pp. 1–8.
consumption in southern Tripoli - Libya during the period [17] F. Fahmi and H. Sofyan, “Forecasting household electricity
from January 1, 2016 to March 24, 2016 (12 weeks) was consumption in the province of Aceh using combination time series
analyzed to determine the best model for forecasting electrical model.” International Conference on Electrical Engineering and
Informatics (ICELTICs), Banda Aceh, pp. 97–102 (2017)
consumption for two weeks ahead. This study revealed that [18] J. Deng and P. Jirutitijaroen, “Short-term load forecasting using time
the seasonal ARIMA (0, 1, 6) (1, 1, 6)7 model was the best series analysis: a case study for Singapore.” In: 2010 IEEE Conference
and was able to forecast consumption with MAPE of 4.332%. on Cybernetics and Intelligent Systems, Singapore, pp. 231–236 (2010)
[19] Z. Haydari, F. Kavehnia, M. Askari and M. Ganbariyan, “Time-series
load modelling and load forecasting using neuro-fuzzy techniques.” In:
2007 9th International Conference onElectrical Power Quality and
Utilisation, Barcelona, pp. 1–6 (2007)
REFERENCES
[1] B. ŞİŞMAN, "A comparison of grey models for electricity consumption
demand forecasting: The case of Turkey." Kastamonu University
Journal of Economics & Administrative Sciences Faculty 13, 2016.
[2] C. Nichiforov, I. Stamatescu, I. Făgărăşan, and G. Stamatescu, "Energy
consumption forecasting using ARIMA and neural network models." In
2017 5th International Symposium on Electrical and Electronics
Engineering (ISEEE), IEEE, pp. 1-4, 2017.
[3] K. Li and Zhang, T. "Forecasting electricity consumption using an
improved grey prediction model." Information 9, no. 8, 204, 2018.
[4] HX. Zhao and F. Magoulès. "A review on the prediction of building
energy consumption." Renewable and Sustainable Energy Reviews 16,
no. 6, pp. 3586-3592, 2012
[5] H. Song, Y. Chen, N. Zhou and G. Chen. "Electricity consumption
forecasting for smart grid using the multi-factor back-propagation
neural network." In Sensors and Systems for Space Applications XII,
vol. 11017, p. 110170O. International Society for Optics and Photonics,
2019.
[6] PK. Jain W. Quamer and R. Pamula, “Electricity consumption
forecasting using time series analysis.” In International Conference on
Advances in Computing and Data Sciences 2018 Apr 20 (pp. 327-335).
Springer, Singapore.
[7] P. Chujai, N. Kerdprasop and K. Kerdprasop. “Time series analysis of
household electric consumption with ARIMA and ARMA models.” In
Proceedings of the International Multi Conference of Engineers and
Computer Scientists 2013 Mar 13 (Vol. 1, pp. 295-300).
[8] M. Elsaraiti, A. Merabet and A. Al-Durra. “Time Series Analysis and
Forecasting of Wind Speed Data.” In2019 IEEE Industry Applications
Society Annual Meeting 2019 (pp. 1-5). IEEE.
[9] YW. Lee, KG. Tay and YY. Choy. “Forecasting electricity
consumption using time series model.” International Journal of
Enginnering and Technology. 2018;7:218-23.
[10] K. Harveen and A. Sachin, “SARIMA Modelling for forecasting the
electricity consumption of a health care building.” International Journal
of Innovative Technology and Exploring Engineering (IJITEE) ISSN.
2019;8(12): 2278-3075.
[11] M. Kalimoldayev, A. Drozdenko, I. Koplyk, T. Marinich, A.
Abdildayeva and T. Zhukabayeva, “Analysis of modern approaches for
the prediction of electric energy consumption.” Open Engineering.
2020 Apr 28;10(1):350-61.
[12] T. Sun, T. Zhang, Y. Teng, Z. Chen and J. Fang. “Monthly Electricity
Consumption Forecasting Method Based on X12 and STL
Decomposition Model in an Integrated Energy System.” Mathematical
Problems in Engineering. 2019 Oct 3; 2019.
[13] Aj. Delima. “Application of Time Series Analysis in Projecting
Philippines’ Electric Consumption.” International Journal of Machine
Learning and Computing. 2019 Oct;9(5), pp.694-699.
[14] G. Box, G. Jenkins, G. Reinsel and G. Ljung, “Time Series Analysis:
Forecasting and Control.” John Wiley & Sons, Fifth edition, New
Jersey, 2016, 88–339.

262

Authorized licensed use limited to: Dalhousie University. Downloaded on September 01,2021 at 00:27:07 UTC from IEEE Xplore. Restrictions apply.