See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/341869222

ARIMAX Model for Short-Term Electrical Load Forecasting

Article · June 2020

DOI: 10.35940/ijrte.D7950.118419

CITATIONS READS

3 334

1 author:

Shilpa Natraj Gn
Sri Siddhartha Institute of Technology
4 PUBLICATIONS   32 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Electrical load forecasting View project

All content following this page was uploaded by Shilpa Natraj Gn on 17 June 2020.

The user has requested enhancement of the downloaded file.

 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-8 Issue-4, November 2019

ARIMAX Model for Short-Term Electrical Load

Forecasting
Shilpa G N, G S Sheshadri

Abstract: The scope for ARIMAX approach to forecast short Several electrical utilities have adopted the conventional
term load has gained a lot of significant importance.In this methods of forecasting the future loads. In conventional
paper, ARIMAXmodel which is an extension of ARIMA model techniques, the models are mainly designed with respect to
with exogenous variables is used for STLF on a time series data the relationship between the load and factors that influence
of Karnataka State Demand pattern. The forecasting accuracy
the load power [5]. Recently, many similarity based
of ARIMA model is enhanced by taking into consideration hour
of the day and day of the week as exogenous variables for approaches have been studied and developed for load
ARIMAX model. Forecasting performance is thus improved by forecasting. These approaches rely on the load
considering these significant load dependent factors. The behaviourinformation of the similar days that improves the
forecasted results indicate that the proposed model is more forecasting accuracy. Such methodologies have the ability
accurate according to mean absolute percentage error (MAPE) to deal with the nonlinear nature of the load along withits
obtained during the testing period of the model. As the historical behavioural pattern in weekdays, weekends and special
load data are available on the databases of the utility, researches days/holidays [6-7]. Many researches have been carried
in the areas of time series modelling are ongoing for electrical out in the field of short term load forecasting [8]. Time
load forecasting. In the proposed paper real time demand data
Series Models and Artificial neural networks (ANN)are the
available on Karnataka Power Transmission Corporation Ltd.
(KPTCL) website is taken to develop and test the proposedload main methodologies [9]used for load forecasting.
forecasting model.The power utility system operational costs and Autoregressive integrated moving average (ARIMA)
its securitydepend on the load forecasting for next few hours. modelis the best adopted procedure among various time
Regional load forecasting helps in the accurate management series methods [10].The modified form of ARIMA model
performance of generation of power plant. Today’s deregulated is autoregressive integrated moving average (ARIMAX)
markets have great demand for prediction of electrical loads, model with exogenous variables. This method is
required for generating companies. There has been tremendous implemented by applying the difference operation to get
growth in electric power demand and hence it is very much stationary time series. Then, the preliminary model order is
essentialfor the utility sectors to have theirdemand information
identified and confirmed by analyzing the plots of the
in advance.
autocorrelationand partial autocorrelation[11]. The
Index Terms:Artificial neural networks (ANN), proposed model parameter estimationis done by applying
autocorrelation, autoregressive integrated moving average with maximum likelihood or weighted least-square methods
exogenous variables (ARIMAX),mean absolute percentage [12].As a result, electrical future load is forecasted based
error, partial autocorrelation, short term load forecasting
on intelligent projection of previous load and present
(STLF).
loadwith greater reliability.The past load pattern of the
power system and its constitution along with the
I. INTRODUCTION
distribution of region wiseload is essential to get an overall
Among the different categories of load forecasting, good behaviourof load pattern. This paper presents time
short-term load prediction has gained importance andhas series ARIMAX modeling for STLF which takes into
undoubtedly become an integral part in operational account hour of the day and day of the week along with the
performance of any electric utility [1]. Some of the load as the inputs to the model. Practically, load depends on
operational decisions namely economic scheduling of the hour of the day, day of the week; these significant
power station, fuel purchase schedule, energy factors should be taken into account for further
transactionsplanningand power system security improvement of the forecasting accuracy.Similarly other
assessment[2]. In addition to power utilities other various exogenous variables (special days, temperature etc.) that
firms likeelectricity marketers, traders and independent affect the load pattern can be included in the proposed
load system operators also need accurate load forecasts for ARIMAX modeling. This approach is analysed and
their operations. Hence, accurate and reliable electrical load simulated on a set of real time load data of Karnataka State
forecasting can result in better financial rewards for all Demand (2016).
these entities.Accurate load predictions leads to great
savings, as it improves economic load dispatch, unit II. ARIMAXMODEL
commitment etc[3].Many nonlinear factors like weather
variables, daily, weekly and seasonal etc. practically affect The forecasting accuracy of ARIMA model is improved by
STLF[4]. addition of weekdays and weekends pattern that leads to
ARIMAX model. Due to dynamic nature of the load,there
Revised Manuscript Received on November 15, 2019 exists a very strong correlation with the input
Shilpa G N, Assistant Professor, Department of Electrical and variablesselected.As a result,
Electronics Engineering, Sri Siddhartha Institute of Technology,
Tumakuru, Karnataka, India.
Dr. G S Sheshadri, Professor, Department of Electrical and
Electronics Engineering, Sri Siddhartha Institute of Technology,
Tumakuru, Karnataka, India.

Published By:
Retrieval Number: D7950118419/2019©BEIESP 2786 Blue Eyes Intelligence Engineering
DOI:10.35940/ijrte.D7950.118419 & Sciences Publication
 ARIMAX Model for Short-Term Electrical Load Forecasting

the forecasting performance of the ARIMAX model Autocorrelation Function (ACF)

1
strongly depends on the input variablesselected and the
0.8
input variables are to be chosen in such a way that the
0.6
forecasting error of the developed model is minimized.

Autocorrelation
0.4
Hourly load changes from one day to another day of the
0.2
week affects the load pattern and hence becomes an
0
importantfactorto be considered and included in the
-0.2
proposedARIMAX model. -0.4
The ARIMAX model equation is given by, -0.6
A(q)y(t) = B(q)u(t) + C(q)e(t)-----------------(1) 0 24 48
Lag
72 96

Fig.2. Plot of ACF for Differenced January load data

where,
A(q), B(q)&C(q)– Delay Polynomials Partial Autocorrelation Function (PACF)
u(t) - Exogenous Input Variables 1

e(t) - Noise 0.8

y(t) -Output Load

Partial Autocorrelation
0.5

Autoregressive integrated moving average model with

exogenous variables (ARIMAX) model development is 0.2

mainly classified into three steps:

1. Determination of model order -0.1

2. Estimation of model parameters

3. Output load forecasting -0.4
0 10 20 30 40 50 60 70 80 90 100
Lag
The present paper attempts to simulate and execute all
the three phases of ARIMAX approach in STLF using real Fig.3. Plot of PACF for Differenced January load
time load data of Karnataka Demand (2016). The proposed data
work is carried out using Mat lab software.Power The ACF & PACFplots as shown in Fig.2 and Fig.3
griddemands accurate load forecasting that is essential for suggesta model of order 2 i.e.autoregressive AR (2) and
economic, stable and secure operation. It also helps in moving average MA (2).As a result,
making proper necessary arrangements for planning the thedevelopedARIMAX (p,d, q)model takesthe form
maintenance. ARIMAX (2, 1, 2)24.After determining the orders of the
model, the next step is to estimate the model parameters.
2.Estimation of model Parameters:In this phase, the
III. SIMULATION RESULTS
proposedARIMAX (2, 1, 2)24model estimates the values
Statistical study of Karnataka State demand pattern is ofcoefficients of delay polynomialsstated in equation (1) in
carried out for short term load forecasting. Analysis of order to minimize the energy of the noise term e(t).
simulation results & model development for accurate Parameters are estimated by an efficient gradient based
STLFcomprises of three major computational steps as method known as theleast square method. After all the
explained below. parameters of ARIMAX (2, 1, 2)24model are estimated, the
1.Determination of model order:Model is developed using next step is to forecast the future valuesof the output load
January load data and tested on February load data. Fig.1 using this proposed model.
shows the plot of ACF for the month of January 2016. ACF Figs.8-9 show load behaviour on weekdays and weekends.
plotindicates that thenature of the load is oscillatory Monday through Saturday is taken as weekdays and
andnon-decaying.As a result, differencing of the load data Sunday is taken as weekend.The load pattern for weekdays
is appliedin order to make the set of considered load series and weekends of January/February month is analysed and it
stationary. Fig.2 shows the resulting auto correlation and is observed that weekdays have similar pattern and
partial autocorrelationplots for differenced set of time weekends have similar pattern. Hence ARIMAX (2, 1,
series load data. The initial order of the ARIMAX model is 2)24model is developed that takes into consideration day
determined by studying and analyzing autocorrelation and code, hour code, weekdays and weekends as exogenous
partial autocorrelation plots for differenced time series load variables along with January (2016) load data of Karnataka
data. State.
Autocorrelation Function (ACF) 3. Output load forecasting:January load data is used to
1
develop ARIMAX ( 2, 1, 2)24model and is applied to
0.8
forecastthe next month i.e. February hourly load. In order
to check the ARIMAX model forecasting performance, it is
0.5
essential to calculate the output load forecasting errors.
Autocorrelation

Fig.10 shows the proposed and developed model

0.2
withMean Absolute Percentage Error (MAPE) of 2.86%.
The model outputs encouragingforecasting results with
-0.1
least forecasting errors.The forecastedresultsshown in
Figs.4-7indicate that the model is most suitablefor short
-0.4
0 20 40 60
Lag
80 100 120 140 150 term load forecasting. Table I gives the weekly forecasted
Fig.1. Plot of ACF for January load data. results (MAPE) for the
month of February.

Published By:
Retrieval Number: D7950118419/2019©BEIESP 2787 Blue Eyes Intelligence Engineering
DOI:10.35940/ijrte.D7950.118419 & Sciences Publication
 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-8 Issue-4, November 2019

TABLE I ARIMAX Model Forecasted Results for Actual Load Vs Forecasted Load
10000
February Month
WEEK ARIMAX Model (Day
8000
code, hour code,
weekends and weekdays)

Load in MW
MAPE 6000

1 3.48% Actual Load

4000 Forecasted Load
2 2.17%
2000
3 2.38%
4 3.43% 0
0 20 40 60 80 100 120 140 160 180
Time in Hours
Overall 2.86%
(February) Fig.6 Forecasted result for 3rd week (15/2 – 21/2
2016)

Actual Load Vs Forecasted Load Actual load Vs Forecasted load

10000 10000

8000

8000
Load in MW

6000

Load in MW
Actual Load
4000 Forecasted Load 6000

2000
Actual Load
4000
Forecasted Load
0
0 20 40 60 80 100 120 140 160 180
Time in Hours
st 2000
Fig.4 Forecasted result for 1 week (1/2 – 7/2 2016)

Actual Load Vs Forecasted Load 0

0 20 40 60 80 100 120 140 160 180
10000
Time in Hours
Fig.7 Forecasted result for 4th week (22/2 – 28/2
8000
2016)
Load in MW

6000

4000 Actual Load

Forecasted Load

2000

0
0 20 40 60 80 100 120 140 160 180
Time in Hours

Fig.5 Forecasted result for 2nd week (8/2 – 14/2

2016)
Load behaviour on weekends(Sundays)
9000

8000

6000
Load in MW

7th February( Sunday )

14th February( Sunday )
4000

2000

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Hour

Fig.8 Load behaviour on weekends (Sundays) (07/2 &14/2 2016)

Published By:
Retrieval Number: D7950118419/2019©BEIESP 2788 Blue Eyes Intelligence Engineering
DOI:10.35940/ijrte.D7950.118419 & Sciences Publication
 ARIMAX Model for Short-Term Electrical Load Forecasting

Load behaviour on weekends(Sundays) and weekdays(Monday)

9500

8000

6000
Load in MW

7th February Weekend(Sunday)

4000 15th february Weekday(Monday)

2000

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Hour

Fig.9 Load behaviour on weekends (Sundays) and weekdays (Monday) (07/2 &15/2 2016)

15

MAPE = 3%
10
ERROR (%)

-5

-10
0 24 48 72 96 120 144 168 192 216 240 264 288 312 336 360 384 408 432 456 480 504 528 552 576 600
HOUR

Fig.10 Mean Absolute Percentage Error (MAPE)

4. Ching-Lai Hor, Simon.J.Watson & Shanti Majithia “Analysing The

IV. CONCLUSION
Impact of Weather Variables On Monthly Electricity Demand”, IEEE
The proposed ARIMAX model forecasts theelectric load by Transactions on Power Systems, Vol.20, No.4, pp.2078-2085,
November 2005.
considering the correlation established to the data of the 5. Ibrahim Moghram & Saifur Rahman “Analysis And Evaluation Of
selected similar days.Initially differencing is applied to Five Short-Term Load Forecasting Techniques”, IEEE
January load data to make it stationary as it was observed 6. Transactions on Power Systems, Vol.4, No.4, pp.1484-1490, October
to be oscillatory and non-decaying. Later the differenced 1989.
7. M Nakamura, “Short Term Load Forecasting Using Weekday load
data is used to build the model. The model orders were Models and Bias Models”, Proc. PICA Conference, pp. 37-42, 1984.
determined to be AR(2) and MA(2) by plotting PACF and 8. Hong-Tzer Yang, Chao-Ming & Ching-Lein Huang “Identification of
ACF of differenced data respectively. The load pattern for ARMAX Model For Short Term Load Forecasting: An
weekdays and weekends of January/February month is Evolutionary Programming Approach”, IEEE Transactions on Power
Systems, Vol.11, No.1, pp.403-408, February 1996.
analysed and it is observed that weekdays have similar 9. IEEE Committee Report, “Load Forecasting Bibliography Phase I”,
pattern and weekendshave similar pattern. Hence, the IEEE Trans. on Power Appr. and Sys., Vol. PAS-99, pp. 53-58, 1980.
inputs chosen to build ARIMAX modelare January month 10. Shilpa.G.N, Dr.V.Venkatesh, Nataraja.C “Study & Development of
differenced hourly load data, time of the day, exogenous Short Term Load Forecasting Models Using Stochastic Time Series
Analysis” International Journal of Engineering Research and
variables like weekdays and weekends. The model Development, Vol 9, Issue 11, PP. 31-36, February 2014.
parameters are then estimated and February month load is 11. Shilpa.G.N, Dr.G.S.Sheshadri, “Short Term Load Forecasting Using
forecasted. The forecasted load is compared with the actual ARIMA Model For Karnataka State Electrical Load”, International
load and forecasting errors are calculated. The results Journal of Engineering Research and Development”, Vol 13, Issue 7,
PP. 75-79, July 2017.
forecasted show a mean absolute percentage error (MAPE) 12. Shilpa.G.N “Short Term Load Forecasting Using Time Series
of 2.86%. Theresearch studies have reported that the load Analysis: A Case Study for Karnataka, India”, International Journal
profile is dynamic in nature and hence the loads Of Engineering Science And Innovative Technology, Vol 1, Issue
aresubjected totemporary and seasonal variations. Thus the 2,PP. 45-53, November 2012 .
13. “Short Term Power System Load Forecasting Using the Iteratively
electric load can be more accurately forecasted by Reweighed Least Squares Algorithm”, Electric power system
consideringthese factors like temperature, humidity etc. that research, 19 (1990) pp. 11-12.
has major impact on the load. 14. T Haida and S Muto, “Regression Based Peak Load Forecasting using
a Transformation Technique”, IEEE Trans. Power syst., Vol. 9, pp.
1788-1794, Nov. 1994.
REFERENCES 15. S Rahman and O Hazim, “A generalized Knowledge-Based Short
1. John Wiley and Sons Ltd.“Comparative models for electrical load Term Load Forecasting Technique”, IEEE Trans. Power syst., Vol. 8,
forecasting”, IEEE Transactions on Power Systems 1985. pp. 508-514, May 1993.
2. Martin T Hagan & Suzanne M Bher “The Time Series Approach to 16. Gross G and Galiana F D, “Short-Term Load Forecasting”, Proc.
Short Term Load Forecasting”, IEEE Transactions on Power Systems, IEEE, Vol. 75, No 12, pp. 1558-1573, 1987.
Vol. PWRS-2, No.3, pp.785-790, August 1987. 17. Alex D and Timothy C, “A Regression-Based Approach to Short-
3. Nima Amjady “Short- Term Bus Load Forecasting of Power Systems Term System Load Forecasting”, IEEE Trans. Power syst., Vol. 5, No
by a New Hybrid Method”, IEEE Transactions on Power Systems, 4, pp. 1535-1547, 1990.
Vol.22, No.1, pp.392-401, February 2007.

Published By:
Retrieval Number: D7950118419/2019©BEIESP 2789 Blue Eyes Intelligence Engineering
DOI:10.35940/ijrte.D7950.118419 & Sciences Publication
 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-8 Issue-4, November 2019

AUTHORS PROFILE

Shilpa G N received her B.E. Degree in

Electrical and Electronics Engineering in 2003
from Visvesvaraya Technological University and
M.E. in Power and Energy Systems from
University Visvesvaraya College of Engineering,
Bangalore University, Karnataka, India. From
2007 through 2009, she had been working as
Assistant Professor in Bapuji Institute of
Engineering and Technology, Davangere.
Presently she is serving as Assistant Professor since September 2010 in
the Department of Electrical & Electronics Engineering at Sri Siddhartha
Institute of Technology, Tumakuru, Karnataka, India. She is also working
toward the Ph.D degree in the research area Power and Energy Systems at
Sri Siddhartha Academy of Higher Education, Sri Siddhartha University,
Tumakuru, Karnataka, India. Her research interests include load
forecasting, time series modelling and artificial neural network
applications to power systems.

Dr. G. S. Sheshadri aged 56 years is presently

serving as Professor at Sri Siddhartha Institute of
Technology, Tumakuru, Karnataka. He has
obtained B.E. in Electrical and Electronics
Engineering from Bangalore University in the year
1987. He obtained his M.Tech. Degree from
National Institute of Engineering, Mysore, in the
year 1990, securing first rank. He has been
awarded Ph.D from Visvesvaraya Technological
University, Belagavi in the year 2011. His areas of
interest are power systems, neural networks, Renewable energy sources,
etc.

Published By:
Retrieval Number: D7950118419/2019©BEIESP 2790 Blue Eyes Intelligence Engineering
DOI:10.35940/ijrte.D7950.118419
View publication stats & Sciences Publication