ALL Metrics
-
Views
-
Downloads
Get PDF
Get XML
Cite
Export
Track
Research Article

Forecasting electricity consumption of Malaysia’s residential sector: Evidence from an exponential smoothing model

[version 1; peer review: 1 approved with reservations]
PUBLISHED 17 Jan 2022
Author details Author details
OPEN PEER REVIEW
REVIEWER STATUS

This article is included in the Energy gateway.

Abstract

Background: In Malaysia, residential electricity consumption has shown a steady upward trend year by year. Due to this increase in energy consumption, it is important to forecast the value of electricity consumption until the year 2032 to accommodate the electricity demand.
Methods: Three exponential smoothing models were compared to identify the most appropriate model in forecasting electricity consumption. The three exponential smoothing models are Simple, Holt, and Brown exponential smoothing. To identify the most appropriate model, a mean absolute percentage (MAPE) was chosen.
Results: The results show that Holt’s exponential smoothing has the best performance with the lowest MAPE score of 2.299.
Conclusions: Consequently, it was found that electricity consumption would substantially increase from 2647 ktoe (kilotonne of oil equivalent) to 3873 ktoe within the period of 2019 to 2032.

Keywords

forecasting, electricity consumption, exponential smoothing, time series

Introduction

In Malaysia, energy is no longer viewed as a luxury as it is becoming a necessity in our everyday activities such as in transportation, production, commercial activities, and residential anthropogenic activities.1 Energy consumption in residential sectors grew up to 7% within the period 1978 to 2015.2 As one of the countries with the highest recorded energy per capita and energy intensity over the years, electricity is considered as an important form of energy which drives Malaysia’s economic development.3,4 The Malaysia Energy Information Hub (MEIH) revealed that electricity consumption for the residential sector in 2018 was 2553 ktoe (kilotonne of oil equivalent), which is a slight decrease from 2017 with 2610 ktoe. However, in 2017 the World Energy Markets Observatory (WEMO) reported that electricity consumption is projected to increase by 4.8% annually up to 2030.

Thus, it is important to conduct a forecast of electricity consumption as it is significant for economic development as well as policy improvement.5,6 As a reliable and an important tool for making decisions, there are several forecasting techniques that can be used. In fact, the accuracy of forecasting can be observed to obtain significant results in the projection of electricity consumption.5 In this study, three different exponential smoothing models were utilised to forecast electricity consumption: simple, Holt, and Brown’s exponential smoothing.

This aim of this research was to conduct a study in order to forecast electricity consumption in Malaysia until the year 2032. This research also attempts to identify the most appropriate exponential smoothing model in forecasting electricity consumption.

Literature review

Lee et al.,7 conducted a study involving six different forecasting methods, which were used to predict electricity consumption in Universiti Tun Hussein Onn Malaysia (UTHM), Malaysia. The study selected a mean absolute percentage error (MAPE) as the measurement of error. Historical data were obtained monthly from 2011 until 2017 which generated a projection up until December 2018. From the six forecasting models, Holt-Winters’ exponential smoothing was found to be the best technique implemented due to having the lowest MAPE. Similarly, Nazim and Afthanorhan8 also found that Holt’s exponential smoothing was the best method to predict Malaysia’s population from 2004 to 2020. The study selected a mean square error (MSE) as a criterion to determine the best model. There were four different techniques used which were single exponential smoothing (SES), double exponential smoothing (DES), Holt’s exponential smoothing, and adaptive response rate exponential smoothing (ARRES). Lima et al.,9 claimed that economic data could be forecasted using Holt-Winters’ exponential smoothing, which involves comparing additive and multiplicative practices. The results showed that multiplicative exhibited the best forecasting performance. Maçaira et al.,10 forecasted the best yearly projection for residential electricity consumption in Brazil by applying exponential smoothing. All the studies mentioned above provided economic projection values together with consumption growth and were validated by the accuracy of the forecast.

In contrast, Popeangă and Lungu11 found that the double moving average was the best technique in forecasting energy consumption in Romania from quarter one until quarter four of 2014. The study used two different moving average practices. Suresh12 used an autoregressive integrated moving average model (ARIMA) to analyse and forecast electricity consumption in India. The proposed ARIMA with the neural network (NN) model provided the best future time-series forecasting. A data set of 40 years of electricity consumption was analysed by Mahia et al.,13 using ARIMA methods. Several steps were conducted, and the best model was selected based on the lowest Akaike Information Criterion (AIC). Notably, the literature has specific reasons to conduct a moving average when the data set performs seasonally, particularly throughout the study period.

From a business perspective, exponential smoothing is used in projecting sales. For example, Sidqi and Sumitra14 applied the single and double exponential smoothing in a study with MAPE as the criterion to determine the accuracy of the forecast model. As a result, the single exponential smoothing exhibited the lowest MAPE. In the agricultural sector, Talwar and Goyal15 analysed and compared coriander prices in India using several exponential smoothing methods. The results showed that the Holt-Winters’ trend adjusted model provided the best model with the lowest error measurement in MSE. A study by Booranawong and Booranawong16 in Thailand showed that the double exponential smoothing method provided better performance in predicting Thai chilli and lemongrass prices. The study utilised MAPE to determine the minimum error measurement from other methods such as multiplicative Holt-Winters’ (MHW) and additive Holt-Winters’ (AHW). The forecasted prices of the agricultural products were from October 2016 to December 2019. Regardless of the sector, exponential smoothing is very suitable to conduct forecasting together with the least error of measurement, such as MAPE.

Based on previous literature, exponential smoothing has been widely used, and this technique captures the time series that may change its behaviour, and the model parameters should adapt to that change in behaviour as well.17 Currently, studies that provide exponential smoothing as a mechanism to predict the electricity consumption of the residential sector are limited. Therefore, this study selected this method as it is considered an appropriate technique to conduct a forecasting of the residential electricity consumption in Malaysia.

Methods

Three exponential smoothing models were used to forecast electricity consumption in this study: simple exponential smoothing, Holt’s exponential smoothing, and Brown’s exponential smoothing. Yearly data from 1997 to 2018 was obtained from the Malaysia Energy Information Hub (MEIH). The forecasting period started from the year 2019 until 2032. Details on the function and formula of the simple exponential smoothing, Holt’s exponential smoothing, and Brown’s exponential smoothing are explained in the following sections.

Simple exponential smoothing

Simple exponential smoothing is the most widely used model in forecasting if there are no cyclic variation patterns or consistent growing patterns involved.14 The formula for this method is as follows18:

(1)
Ft+m=αyt+1αFt

where:

Ft+m: simple exponentially forecast value in period m, for m=1,2,3,t; yt: actual value in time t;

α: unknown smoothing constant to be determined for value between 0 and 1;

Ft: forecast value at period t.

Holt’s exponential smoothing

Holt’s exponential smoothing can be used in forecasting when there is a linear trend in the historical data of the forecast. According to Alias,19 this method requires three equations which are exponential smoothed series, trend estimate, and forecast.

Exponential smoothed series:

(2)
St=αyt+1αSt1+Tt1

Trend estimate:

(3)
Tt=βStSt1+1βTt1

Forecast for m period:

(4)
Ft+m=St+Tt×m

where:

St: exponential smoothed series;

Tt: trend estimates;

Ft+m: forecast for m-step-ahead period.

Brown’s exponential smoothing

This method is also known as the double exponential smoothing and can be used if there is a linear trend in the data. The trend is a smoothed estimate of average growth at the end of each period.17 The formula for Brown’s exponential smoothing is as follows:

(5)
Ft+m=αyt+1αFt
(6)
at=2StSt
(7)
bt=α1αStSt
(8)
St=αyt+1αSt1
(9)
St=αSt+1αSt1
where:

St: exponentially smoothed value of yt at time t;

St: double exponentially smoothed value of yt at time t;

at: computes the difference between the exponentially smoothed values;

bt: computes the adjustment factor;

Ft+m: forecast for m-step-ahead period.

Accuracy of forecast

According to Kalekar,17 the forecasting model should be validated. Here, an error measurement such as mean absolute percentage error (MAPE), relative mean square error (RMSE), or mean absolute error (MAE) is needed. The selection of an error measurement has a significant effect in determining the most accurate forecasting method. In this study, MAPE was selected to verify the model because it is the most suitable measurement to compare the accuracy of the forecasting methods, as it measures relative performance.7,14,20 A low MAPE score indicates that the forecasting model has a good performance.21 The range of the MAPE score is shown in Table 1.

Table 1. Significance of MAPE score.

MAPESignificance
<10%Excellent forecasting ability
10-20%Good forecasting ability
20-50%Reasonable forecasting ability
>50%Bad forecasting ability

Results

An analysis of the historical data was conducted to examine the pattern of electricity consumption trend (kilotonne of oil equivalent, ktoe) from 1997 until 2018 for the residential sector in Malaysia.

Based on Figure 1, the trend pattern in this time series is linear and no seasonality is involved. In addition, the graph shows a constant increase in the pattern. In 2016, a total of 2679 ktoe of electricity was consumed as the highest value. The consumption value gradually decreased from 2017 to 2018 with values of 2610 ktoe and 2553 ktoe respectively as shown in Table 2. Thus, all three exponential smoothing methods, namely simple, Holt’s and Brown exponential smoothing were capable of forecasting electricity consumption. The lowest MAPE was compared to identify the most appropriate exponential smoothing model.

962a4f4f-7f6c-491d-a922-d011acf530a7_figure1.gif

Figure 1. Electricity consumption (ktoe) residential increases over time.

(Source: Malaysia Energy Information Hub, https://meih.st.gov.my/statistics.)

Table 2. Electricity consumption (ktoe) residential increases over time.

YearFinal electricity consumption (ktoe)
Residential
1997770
1998874
1999885
2000975
20011,081
20021,161
20031,248
20041,319
20051,395
20061,514
20071,598
20081,668
20091,792
20101,937
20111,974
20122,126
20132,262
20142,346
20152,471
20162,679
20172,610
20182,553

Simple exponential smoothing

The simple exponential smoothing was analysed using SPSS version 26 (IBM SPSS Statistics, RRID:SCR_019096) starting from the year 2019. Referring to Figure 2, the three lines represent the limit of the forecast values for the forecast year: yellow indicates the upper limit of electricity consumption, purple indicates the best prediction of electricity consumption, and brown indicates the lower limit of electricity consumption at a particular year. This means that the expected electricity consumption could fall between the green and brown dotted lines.

962a4f4f-7f6c-491d-a922-d011acf530a7_figure2.gif

Figure 2. Simple exponential smoothing.

UCL: upper confidence level (upper limit); LCL: lower confidence level (lower limit).

Based on Table 3, the year 2019 recorded the best forecasted electricity consumption, which is 2553 ktoe; the consumption maintains this value until 2032. However, this method also predicted a wide range of possible values until 2032. For example, the electricity consumption in 2032 is predicted to be between 3345 ktoe as the upper limit and 1760 ktoe as the lower limit, but the best value generated is 2553 ktoe.

Table 3. Forecasting result: simple exponential smoothing.

YearForecast value (ktoe)Upper limit (ktoe)Lower limit (ktoe)
2019255327732333
2020255328642242
2021255329342172
2022255329932113
2023255330442062
2024255330912015
2025255331341972
2026255331751931
2027255332121894
2028255332481858
2029255332821824
2030255333141792
2031255333451761
2032255333451761

The accuracy of the forecasting value was measured via model fit statistics. As shown in Table 4, there are three error measurement scores generated by SPSS version 26: relative mean square error (RMSE), mean absolute percentage error (MAPE), and mean absolute error (MAE). For simple exponential MAPE, the score is 5.678.

Table 4. Model fit statistics for simple exponential smoothing.

Error measurementScore
RMSE105.677
MAPE5.678
MAE92.504

Holt’s exponential smoothing

Figure 3 shows Holt’s exponential smoothing method, which was conducted. It is expected that the electricity consumption value would fall between the pink and brown dotted lines for the forecast year, but the width is narrower compared to the previous method. It can be observed that the forecast has a considerable upward trend from 2019 until 2032.

962a4f4f-7f6c-491d-a922-d011acf530a7_figure3.gif

Figure 3. Holt's exponential smoothing.

UCL: upper confidence level (upper limit); LCL lower confidence level (lower limit).

The value of the best forecasted electricity consumption was 2647 ktoe for 2019 and 3873 ktoe for 2032 as shown in Table 5. In 2032, the predicted electricity consumption value would be between 3360 and 4386 ktoe whereas the best projected value is 3873 ktoe. Table 6 shows the accuracy of the forecasting where the score measured by MAPE is 2.299.

Table 5. Forecasting result: Holt’s exponential smoothing.

YearForecast value (ktoe)Upper limit (ktoe)Lower limit (ktoe)
2019264727832511
2020274229342549
2021283630722600
2022293032032658
2023302433292720
2024311934532785
2025321335742852
2026330736942921
2027340238122992
2028349639293063
2029359040443136
2030368541593210
2031377942733285
2032387343863360

Table 6. Model fit statistics for Holt’s exponential smoothing.

Error measurementScore
RMSE65.235
MAPE2.299
MAE42.664

Brown’s exponential smoothing

The last forecasting method is Brown’s exponential smoothing (Figure 4). This method enjoys a very wide range of forecasting values where the upper and lower limit values experienced a steep trend. In 2019, the forecasted electricity consumption value is 2402 ktoe. However, the value gradually decreases year by year until 2032 with a consumption value of 2399 ktoe as presented in Table 7. In 2032, the predicted electricity consumption value would be within 4967 ktoe as the maximum value and 90 ktoe as the minimum value; the best projected value is 2399 ktoe. Table 8 shows the MAPE score for Brown’s exponential smoothing, which is 3.125. The lower the MAPE score, the higher the accuracy of the forecast.

962a4f4f-7f6c-491d-a922-d011acf530a7_figure4.gif

Figure 4. Brown’s exponential smoothing.

UCL: upper confidence level (upper limit); LCL: lower confidence level (lower limit).

Table 7. Forecasting result: Brown exponential smoothing.

YearForecast value (ktoe)Upper limit (ktoe)Lower limit (ktoe)
2019255327022403
2020254127992282
2021252929142144
2022251730431991
2023250531871824
2024249433431645
2025248235111453
2026247036901251
2027245838791038
202824474079815
202924354288582
203024234506341
20312411473390
203224004968−169

Table 8. Model fit statistics for Brown’s exponential smoothing.

Error measurementScore
RMSE72.041
MAPE3.125
MAE51.085

Discussion

Table 9 shows the summary of all smoothing methods using MAPE as the measurement. Holt’s exponential smoothing is the most suitable method to predict the electricity consumption of the residential sector in Malaysia. This is because this method has the lowest MAPE score which is 2.299 compared to the other two methods. The result is consistent with Omer et al.22 where they used MAPE to identify the most appropriate prediction model. Therefore, in this study, Holt’s exponential smoothing is the best method to forecast future electricity consumption in Malaysia.

Table 9. Summary of the results of the exponential smoothing models.

MethodMAPE
Simple exponential smoothing5.678
Holt’s exponential smoothing2.299
Brown exponential smoothing3.125

Table 5 lists the forecasting values from 2019 until 2032 generated by Holt’s exponential smoothing. In 2032, the maximum electricity consumption is 4386 ktoe and the minimum consumption is 3360 ktoe. However, the best predicted value is 3873 ktoe.

Conclusion

The aim of the study was to forecast the total electricity consumption in Malaysia until the year 2032 and to identify the most appropriate exponential smoothing model to predict this consumption. Based on the results and discussion, Holt’s exponential smoothing was found to be the most appropriate model in forecasting electricity consumption in Malaysia. Generally, electricity consumption in residential sectors comes from electrical appliances. Hence, based on the predicted consumption growth, energy efficiency regulations by electrical manufacturers should be revised. Campaigns for less energy demand in households should also be implemented. This can benefit both parties by maximising energy efficiency and minimising electricity consumption. As a suggestion for further study, moving average methods can also be applied to predict electricity consumption. There are various types of moving average method including the integrated model such as autoregressive–moving-average (ARMA), autoregressive integrated moving average (ARIMA), autoregressive integrated moving average- neural network (ARIMA-NN), or the hybrid model. These methods can be used by future researchers to further expand on this study.

Data availability

All data underlying the results are available as part of the article and no additional source data are required.

Comments on this article Comments (0)

Version 1
VERSION 1 PUBLISHED 17 Jan 2022
Comment
Author details Author details
Competing interests
Grant information
Copyright
Download
 
Export To
metrics
Views Downloads
F1000Research - -
PubMed Central
Data from PMC are received and updated monthly.
- -
Citations
CITE
how to cite this article
Ishak I, Othman NS and Harun NH. Forecasting electricity consumption of Malaysia’s residential sector: Evidence from an exponential smoothing model [version 1; peer review: 1 approved with reservations]. F1000Research 2022, 11:54 (https://doi.org/10.12688/f1000research.74877.1)
NOTE: If applicable, it is important to ensure the information in square brackets after the title is included in all citations of this article.
track
receive updates on this article
Track an article to receive email alerts on any updates to this article.

Peer review discontinued

Peer review at F1000Research is author-driven. Currently no reviewers are being invited. What does this mean?
Version 1
VERSION 1
PUBLISHED 17 Jan 2022
Views
28
Cite
Reviewer Report 27 Jul 2022
Azer Dilanchiev, Faculty of Business and Technologies, International Black Sea University, Tbilisi, Georgia 
Approved with Reservations
VIEWS 28
To begin with, it was a pleasure for me to review this engaging paper. In my perspective, the paper is valid. Throughout the investigation, the authors created a well-written, well-structured manuscript while adhering to rigorous research procedures. I'd like to ... Continue reading
CITE
CITE
HOW TO CITE THIS REPORT
Dilanchiev A. Reviewer Report For: Forecasting electricity consumption of Malaysia’s residential sector: Evidence from an exponential smoothing model [version 1; peer review: 1 approved with reservations]. F1000Research 2022, 11:54 (https://doi.org/10.5256/f1000research.78678.r143421)
NOTE: it is important to ensure the information in square brackets after the title is included in all citations of this article.

Comments on this article Comments (0)

Version 1
VERSION 1 PUBLISHED 17 Jan 2022
Comment
Alongside their report, reviewers assign a status to the article:
Approved - the paper is scientifically sound in its current form and only minor, if any, improvements are suggested
Approved with reservations - A number of small changes, sometimes more significant revisions are required to address specific details and improve the papers academic merit.
Not approved - fundamental flaws in the paper seriously undermine the findings and conclusions
Sign In
If you've forgotten your password, please enter your email address below and we'll send you instructions on how to reset your password.

The email address should be the one you originally registered with F1000.

Email address not valid, please try again

You registered with F1000 via Google, so we cannot reset your password.

To sign in, please click here.

If you still need help with your Google account password, please click here.

You registered with F1000 via Facebook, so we cannot reset your password.

To sign in, please click here.

If you still need help with your Facebook account password, please click here.

Code not correct, please try again
Email us for further assistance.
Server error, please try again.