A generalized regression model based on hybrid empirical mode decomposition and support vector regression with back‐propagation neural network for mid‐short‐term load forecasting

• Author: Jia‐Mei Zheng,Wei‐Chiang Hong,Guo‐Feng Fan,Yan‐Hui Guo
• Date: 01 August 2020
• Published date: 01 August 2020
• DOI: http://doi.org/10.1002/for.2655

RESEARCH ARTICLE

A generalized regression model based on hybrid empirical

mode decomposition and support vector regression with

back-propagation neural network for mid-short-term load

forecasting

Guo-Feng Fan

1

| Yan-Hui Guo

1

| Jia-Mei Zheng

1

| Wei-Chiang Hong

2

1

School of Mathematics and Statistics

Science, Ping Ding Shan University, Ping

Ding Shan, Henan, China

2

Department of Information

Management, Oriental Institute of

Technology, New Taipei, 220, Taiwan

Correspondence

Wei-Chiang Hong, Department of

Information Management, Oriental

Institute of Technology, New Taipei, 220,

Taiwan.

Email: samuelsonhong@gmail.com

Funding information

Ministry of Science and Technology,

Taiwan, Grant/Award Number: MOST

108-2410-H-161-004; Science and

Technology of Henan Province of China,

Grant/Award Number: 182400410419;

Startup Foundation for Doctors, Grant/

Award Number: PXY-BSQD-2014001; The

Foundation for Fostering the National

Foundation of Pingdingshan University,

Grant/Award Number: PXY-PYJJ-2016006

Abstract

Since load forecasting plays a decisive role in the safe and stable operation of

power systems, it is particularly important to explore forecasting methods

accurately. In this article, the hybrid empirical mode decomposition (EMD)

and support vector regression (SVR) with back-propagation neural network

(BPNN), namely the EMDHR-SVR-BPNN model, is proposed. Information the-

ory is mainly used to solve the data tendency problem, and the EMD method

is used to solve the data volatility problem. There is no interaction between

these two methods; thus these two models can complement each other

through generalized regression of orthogonal decomposition. Taking the load

data from the New South Wales (NSW, Australia) market as an example, the

obtained simulation results are compared with other models. It is concluded

that the proposed EMDHR-SVR-BPNN model not only improves the forecast-

ing accuracy but also has good fitting ability. It can reflect the changing ten-

dency of data in a timely manner, providing a strong basis for the electricity

generation of the power sector in the future, thus reducing electricity waste.

The proposed EMDHR-SVR-BPNN model has potential for employment in

mid-short term load forecasting.

KEYWORDS

back-propagation neural network (BPNN), empirical mode decomposition (EMD), general

regression, information theory, mid-short-term load forecasting, support vector regression (SVR)

1|INTRODUCTION

Mid-short-term load forecasting usually refers to forecast-

ing electrical load in the coming months and weeks,

which is more accurate than long-term load forecasting.

Thus mid-short-term load refers to load data within a few

months or weeks, mainly including monthly data.

Monthly data are cyclical and contain extreme

value points, reflecting people's sleep and rest schedule,

which is mainly affected by work plans and holidays

(Abu-Shikhah & Elkarmi, 2011; Amjady & Daraeepour,

2011). Improving the technological level of electrical load

forecasting is conducive to rationally arranging power

grid operation modes and unit maintenance plans, which

in turn is conducive to reducing the economic costs of

operators in the electricity market (Andini, Cabral, &

Santos, 2019; Eladl & ElDesouky, 2019; Kuboth, Heberle,

König-Haagen, & Brüggemann, 2019). It is also condu-

cive to formulating a reasonable electricity supply con-

struction plan to ensure the stability and safety of the

Received: 12 July 2019 Revised: 8 January 2020 Accepted: 11 January 2020

DOI: 10.1002/for.2655

Journal of Forecasting. 2020;39:737–756. wileyonlinelibrary.com/journal/for © 2020 John Wiley & Sons, Ltd. 737

power supply system (Wang, Xiong, Hu, & Lu, 2019).

Especially in the electricity market, improving the accu-

racy of load forecasting is an important means to improve

the management of power systems.

1.1 |Traditional load forecasting

Traditional methodologies of electrical load forecasting

include time series methods (autoregressive integrated

moving average (ARIMA); Barak & Sadegh, 2016;

Pereira, Almeida, & Velloso, 2015; Sen, Roy, & Pal,

2016; Yuan, Liu, & Fang, 2016), exponential smoothing

models (D. Yang et al., 2015), regression models (Khan,

Michael, & Stephenson, 2019; Lebotsa, Sigauke, Bere,

Fildes, & Boylan, 2018), and Kalman filter models

(Takeda, Tamura, & Sato, 2016). For example, Barak

and Sadegh (2016) proposed a novel ensemble methodol-

ogy based on hybridization of ARIMA and adaptive

neuro fuzzy inference system (ANFIS) models. In which,

the ARIMA model is implemented to deal with the lin-

ear part of data, where its nonlinear residuals are fore-

casted by ANFIS structures by using grid partitioning,

sub clustering, and fuzzy c means clustering. The results

indicate that the proposed hybrid model not only

improves the accuracy of single ARIMA and ANFIS

models in forecasting energy consumption, but also acts

better than others and the model's MSE criterion. Sen

et al. (2016) used the ARIMA model to forecast energy

demand in India, with favorable results. In the proposed

model, seasonal tendencies can be smoothed by effi-

ciently averaging the method in terms of energy con-

sumption. The authors concluded that correct ARIMA

models provide accurate forecasts, and support better

environmental management practices. Pereira et al.

(2015) proposed the fuzzy inference model and the sea-

sonal autoregressive integrated moving average with

exogenous variables (SARIMAX) model to forecast the

electrical load in Bahia State. Three exogenous variables

(customers, temperature, and rainfall) were combined

with each model. The experimental results demonstrate

that the fuzzy inference forecasting model is superior to

the SARIMAX model. Khan et al. (2019) proposed a

time-segmented regression analysis (TSRA) method to

determine the electricity demand of high-consumption

households at different times of the day to identify that

hot water and heating was the main factor in determin-

ing daily electricity variation. They concluded that the

proposed TSRA method can be used to provide more

targeted electricity demand management strategy for the

relevant departments. The theoretical developments of

these traditional methods are mature, but it can be diffi-

cult to find the optimal solution when dealing with large

amounts of data and complex nonlinear relationships

between data.

1.2 |Modern load forecasting methods

Modern forecasting models such as artificial neural net-

works (ANNs), support vector machine (SVM), and fuzzy

forecasting method overcome the shortcomings of tradi-

tional forecasting models and can conduct nonlinear

mapping relationships among variables to obtain the

optimal solution of the forecasting models. For example,

Jetcheva, Majidpour, and Chen (2014) proposed a novel

building-level neural network-based ensemble model for

day-ahead electrical load forecasting. The experimental

results demonstrate that it outperforms the previously

established best-performing model (SARIMA) by up to

50%. The SVM in machine learning has the advantages of

solid theoretical foundation and strong generalization

ability, and is not easy to be trapped into local minima.

Therefore, it is often used for mid-short-term load fore-

casting, namely the support vector regression (SVR)

model. Barman, Choudhury, and Sutradhar (2018) pro-

posed a regional hybrid short-term load forecasting

model utilizing the SVR model with the grasshopper opti-

mization algorithm, by considering regional

climatic conditions. The results demonstrate better accu-

racy of the proposed model compared with the classical

STLF model incorporating temperature universally as the

only climatic factor. Chen et al. (2017) proposed a new

SVR forecasting model with an ambient temperature

2 hours before demand response (DR) event as input vari-

ables. The empirical results demonstrate that the SVR

model offers a higher degree of forecasting accuracy and

stability in short-term load. However, owing to the

nonlinear and nonstationary characteristics of mid-short-

term load forecasting, the SVR model still has certain

limitations in practical applications. Although the SVR

model is widely applied, there are still some outstanding

problems; for example, the relationships among the load

and the variables affecting the load are difficult to express

by precise mathematical equations and models; the accu-

racy of the forecasting is not satisfied; and the real power

loads cannot be reflected in real time.

Recently, many novel electrical load forecasting

models have been proposed to achieve more accurate per-

formance; Oliveira and Oliveira (2017), for example,

expanded the fields of application of combined bootstrap

aggregating (bagging) and forecasting methods to the

electric energy sector. A comparative out-of-sample anal-

ysis is conducted using monthly electric energy consump-

tion time series from different countries. Al-Musaylh,

Deo, Adamowski, and Li (2017) proposed a short-term

738 FAN ET AL.