Content uploaded by Nadeem Javaid

Author content

All content in this area was uploaded by Nadeem Javaid on Jul 13, 2019

Content may be subject to copyright.

Electricity Load and Price Forecasting

Using Enhanced Machine Learning

Techniques

Hamida Bano1, Aroosa Tahir2, Ishtiaq Ali1, Raja Jalees ul Hussen Khan1,

Abdul Haseeb1, and Nadeem Javaid1(B

)

1COMSATS University Islamabad, Islamabad 44000, Pakistan

nadeemjavaidqau@gmail.com

http://www.njavaid.com

2Sardar Bhadur Khan Women University Quetta,

Quetta 87300, Pakistan

Abstract. The exponential increase in electricity generation and

consumption pattern are the two main issues in the wholesale markets.

To handle these issues diﬀerent machine learning techniques are used for

load and price prediction in the research ﬁeld. The wholesale utilities

provide real-time data of load and price for the better prediction of elec-

tricity generation purposes. The New York Independent System Operator

(NY-ISO) is one of the utility which provide electricity to diﬀerent coun-

ties like United States, Canada and Israel. In this paper, hourly data of

2016–2017 is used for the forecasting process of load and price of New

York City. Feature selection and extraction are used to achieve important

features. The feature selection is done by two techniques Classiﬁcation

and Regression Tree (CART) and Recursive Feature Elimination (RFE)

and Feature extraction by using Singular Value Decomposition (SVD).

The Multiple Layer Perceptron (MLP), Support Vector Machine (SVM)

and Logistic Regression (LR) classiﬁers are separately used for forecast-

ing purposes of electricity load and price. Further enhance these three

techniques EMLP, ESVM and ELR to take more accurate results for

electricity load and price forecasting.

1 Introduction

Electricity is the basic need of smart environment (smart city, smart industry,

smart homes, smart devices etc.). Due to the exponential increase in the demand

of electricity from diﬀerent sectors (residential, commercial and industrial), it is

important to proposed a management scheme or smart system which betterly

handle the generation and consumption rate of both utility and consumption

side.

Smart Grid (SG) is the network of transmission lines from where electricity is

eﬃciently transferred from the power plants to residential, commercial or indus-

trial sectors. It has a two way communication between the utility and consumers

c

Springer Nature Switzerland AG 2020

L. Barolli et al. (Eds.): IMIS 2019, AISC 994, pp. 255–267, 2020.

https://doi.org/10.1007/978-3-030-22263-5_25

256 H. Bano et al.

Table 1. List of abbreviations

ARIMA AutoRegressive Integrated Moving Average

CART Classiﬁcation and Regression Tree

DAs Data Analytics

DSM Demand Side Management

EMLP Enhanced Multilayer Perceptron

ELR Enhanced Logistic Regression

IWNN Improved Wavelet Neural Network

LR Logistic Regression

LSTM Long Short Term Memory

MLP Multilayer Perceptron

NN Neural Network

NY-ISO New York Independent System Operator

NLSSVM Normal Least Square Support Vector Machine

NSW New South Wales (Australia market)

PJM Pennsylvania, Jersey, Maryland

PV Photovoltaic

RNN Recurrent Neural Network

RFE Recursive Feature Elimination

STLF Short Term Load Forecasting

SG Smart Grid

SVM Support Vector Machine

SVD Singular Value Decomposition

however, the traditional grids have one way communication. The communica-

tion capability of SG eﬃciently increases its usage in the management schemes

of load and price. The SG is beneﬁcial in many aspects quick restoration after

power outage, cost management of both utilities and consumers, reduced peak

load and improve security. Due to these beneﬁcial factors utilities move to smart

grid to reduce the demand load in the peak hours.

Data Analytics (DA) is the broad ﬁeld of qualitative and quantitative exam-

ining of large data sets. The techniques and technologies of DA are used for the

better decision making in business industries which increase the business revenue

and improve the operational eﬃciency.

Predictive DA is one of the types of DA which analyze the current or historical

facts to make future prediction. The utilities need to enhance the generation

and consumption rate of electricity, so that accurate prediction of load and price

of electricity is necessary. Electricity load and price forecasting techniques are

used in demand response and load management systems. Demand response is

the balance of supply side (utility) and demand side (consumer) to eﬃciently

control the load in peak hours (to minimize load and price).

Electricity Load and Price Forecasting Using Enhanced Machine ... 257

The utilities provide incentive programs to customers [1] to manage the

demand response. Electricity load forecasting techniques are classiﬁed into short

term, medium term and long term methods. Short term forecasting in smart

grids shows the continuous supply of electricity by saving cost. In short term

forecasting, prediction of a few minutes, hours or days ahead while in medium

term forecasting a week or month ahead and in long term forecasting prediction

of 1–10 years.

Recurrent Neural Network (RNN) is a type of Neural Network (NN) used for

the processing of sequential data. In [2], authors describe the training procedure

of LSTM network useful for the short term load forecasting. LSTM sequential

models are eﬃciently used for data to memorize by its memory unit for long

sequence. In paper [3], the LSTM technique improves the accuracy of model by

presenting the optimum minimization of gradient vanishing problem.

1.1 Motivation

Deep Neural network (DNN) [4] , Gated Recurrent Unit (GRU) [7], Hybrid

method + MLP [8], Convolution Neural Network (CNN), Shallow Neural Net-

work (SNN) and LSTM [9] models are eﬃciently used in the forecasting methods

which increase the prediction accuracy. Classiﬁer based models are mostly used

in forecasting such as Sperm Whale algorithm + LSSVM [10]. The NN type MLP

has traditionally have one hidden layer network however the simpler models of

NN have potential to predict the accurate accuracy of price [12]. Other energy

related applications shows excellent results obtained in time series prediction

[13,14], the electricity price prediction is possible by using DL architectures.

These prediction techniques motivate the worth of the techniques to enhance

the prediction accuracy of load and price of power markets.

1.2 Problem Statement

Traditional techniques are diﬃcult to handle big data in SGs. However, the

accurate load and price forecasting using huge amount of data from the smart

grid is the main challenges in the data analytic ﬁeld. The exponential increase of

the electricity demand and the weather conditions also aﬀect on the generation

side in terms of load and price. The high demand (on peak hours) of electricity

eﬀect on the pricing schemes. The high complex computational time during

training process directly aﬀect on accuracy. However, load forecasting is in [2]

ignored the computational time.

In [7], tune the hyperparameters (cost penalty, insensitive loss function and

kernal) of SVM are addressed for price forecasting however, the over ﬁtting

problem is not handle.

1.3 Contributions

•The main contribution of this paper is the time series hourly prediction of

load and price of electricity of NY-ISO market.

258 H. Bano et al.

•First step to normalize the dataset in which selection and extraction techniques

are involved.

•In feature selection method, two steps are used to ﬁnd the relevance and

validity of data from the dataset which eﬀect on the forecasting accuracy.

•Feature Selection: CART technique is used to investigate the relationship

between a dependent variables (target) and independent variables (predictor).

It shows the strength or important impact of multiple independent variables on

dependent variables.

RFE technique is the second step of feature selection to further reduce the

dimensionality of dataset. To remove the low important features, RFE directly

improve the accuracy of prediction and reduce the model complexity.

•Feature Extraction: Further reduce the high dimensionality of dataset and

computational cost SVD technique is applied.

•Classiﬁer: Three classiﬁers MLP, SVM and LR models are separately used for

the time series hourly prediction of load and price. Furthermore, enhance the

three techniques for more accuracy in forecasting process.

•Simulation results shows that MLP, SVM and LR gives best accuracy in

electricity load and price forecasting process.

2 Related Work

Literature reviews shows that diﬀerent heuristic and meta heuristic techniques

are used in electricity markets to predict load and price. The accurate forecasting

techniques are beneﬁcial for utility companies to enhance the stability in the

market. Management of supply and demand side by reducing the cost is also

depend on the forecasting models. Long term hourly load prediction for the

period of ﬁve years by LSTM-RNN is introduced in [16]. Abedinia et al. [17]

shows the hourly solar energy load prediction of PV plant generation. In [18]

proposed a novel hybrid algorithm (ARIMA and NLSSVM) for simultaneous

prediction of load and price. The DSM schedule the connected moments of all

the shiftable devices in a smart grid to bring the load consumption curve close

to the objective load consumption curve.

In [19] hybrid model LSSVM with the combined kernal function is used for

price forecasting of Australian electricity market. A hybridize model of Phase

Space Reconstruction (PSR) with Bi-Square Kernel (BSK) regression model

called PSR-BSK model [10] . The PSR extract the evolutionary trends of power

system and the valuable features improve the forecasting procedure. The main

goal of electricity price and load prediction is to increase and decrease the power

generation of utility companies. The precise prediction of cost and load of elec-

tricity increase the competency and stability of utilities in the power markets

[20]. The medium term load and price forecasting by Multi-block NN (Elman

NN) has high capability to reduce error and improve the training mechanism of

forecast process. It also shows that the hybrid topology present an accurate pre-

diction as compared to the prediction models [21]. A day ahead price prediction

using Wavelet Packet Transform, Generalized Mutual Information for normal-

ization of data, Least Square Support Vector Machine for forecast engine and

Electricity Load and Price Forecasting Using Enhanced Machine ... 259

Artiﬁcial Bee Colony for optimization. (WRT, GMI, modiﬁed LSSVM, ABC)

in [22] Hybrid algorithm is applied on diﬀerent datasets of NY-ISO, PJM and

NSW for simultaneous and accurate peak hour’s prediction of load and price and

optimizes DSM in [24]. The electricity load is predicted in microgrid scenario by

hybrid evolutionary fuzzy technique in [25]. In [26], STLF method is eﬃcient

in managing the scheduling pattern and reducing cost of utility companies. In

[27], Wavelet Least Square Support Vector Machine (WLSSVM) model is used

to optimize by fruit ﬂy algorithm. The summary of related work is mentioned

in Table 2.

Table 2. Summary of related work

References Fore cast

method

Dataset Techniques Objectives Limitations

[5]Short

term

Nord pool

spot market

Univariate and

multivariate

models

Average

hourly

price

prediction

Over ﬁtting

problem

[6]Short

term

Australian

electricity

markets

Multilayer

neural

network

Hourly

price

forecasting

Less capability

to memorize

the previous

sequential

data

[11]Medium

term

Ontario and

Australian

electricity

markets

IWNN Price

forecasting

Over ﬁtting

problem

[16]Long

term

ISO New

England

market

LSTM-RNN Hourly load

forecasting

No

appropriate

normalization

technique

[17]Short

term

PV plant

generation

Hybrid (neural

network and

metaheuristic

algorithm)

Load

prediction

Less precise

prediction

[18]Short

term

PJM CNN, LSTM Price

forecasting

Negative

impact of

feature

reduction on

forecasting

accuracy

[10]Short

term

NSW and

NY-ISO

PSR-BSK

regression

model

Load

prediction

Not applied

on natural gas

forecasting

[27]Long

term

ZheJiang

province

FOA Load

forecasting

Less

prediction

accuracy

260 H. Bano et al.

3SystemModel

The proposed scheme shows the ﬂow of the techniques which involve in the fore-

casting procedure of load and cost of electricity market. The electricity genera-

tion and consumption rate are greatly increasing by smart grids. These increasing

rates create serious problems for utility markets to manage the load and cost of

electricity changes. In the research ﬁeld diﬀerent machine learning techniques are

used to ﬁnd the nearer predictive cost and load of electricity provider companies.

The real time electricity dataset 2016–2017 per hour are collect from NY-

ISO which is known as one of the electricity provider company and manage the

wholesale energy markets of diﬀerent areas. The New York City hourly load and

cost are predict and store in a database for pre-processing. Preprocessing is a

machine learning technique which purify the dataset from redundancy, missing

values and quality of data. Data preprocessing is done by both the selection and

extraction techniques. The selection and extraction techniques are generally used

for preprocessing in machine learning techniques. The data set are divided into

training set and test set. 70% data is used for training and 30% for testing. In

the feature selection, two steps are involve by using CART and RFE techniques

to recognize the relevant and important features for forecasting purpose. SVD

technique is used for feature extraction to select the good quality of data however,

bagging is also used in extraction technique [23] . The main importance of feature

extraction is to reduce the dimensionality of dataset which directly aﬀect the

computational time of the system. The hourly data of NY-ISO are rearranged

to compatible with the classiﬁers (MLP, LR and SVM) to predict the electricity

load and price. After the completion of the forecasting engine the predicted

values of load and price are evaluated by four performance metrics i.e RMSE,

MSE, MAE and MAPE. The proposed system shows the ﬂow of forecasting

procedure of electricity load and price in Fig. 1.

Fig. 1. Proposed system model

Electricity Load and Price Forecasting Using Enhanced Machine ... 261

4 Techniques Description

In this section, all the techniques are described which are used to proposed

forecasting models for electricity load and price of NY-ISO New York market.

4.1 Data Preprocessing

•CART

The selection technique is mainly used to overcome the redundancy and

dimensionality of data which have low importance in the performance of pre-

diction.

The CART is a binary tree and information based learning algorithm. The

nodes split and the terminal leaf node contains an output which is used to

make prediction. This technique also investigate the relationship between

a dependent (target) and independent variables (predictor). It shows the

strength or importance impact of multiple independent variables on depen-

dent variables.

•RFE

RFE commonly used to reduce the redundancy and discard the weak fea-

tures of data which has no eﬀect on the electricity prediction performance.

The RFE technique also overcomes the uninformative data from the dataset

and ﬁnd the prediction errors. In the algorithm the training set is used for

three purposes for predictor selection, model ﬁtting and performance evalua-

tion.

For further extracting the interested features and to reduce the high-

dimensional data into low-dimensional data, extraction process is done by

SVD.

•SVD

SVD is a numerical technique based on simple linear algebra. This technique is

a factorization or decomposing a complex matrix to reduce the dimensionality

of dataset features. Classiﬁcation and selection techniques improve forecasting

accuracy and simplify the classiﬁer complexity.

4.2 Classiﬁers

•MLP Technique

MLP is a deep learning technique and is the class of Feed-Forward Artiﬁcial

Neural Network (FFANN). It consist of three layer; input, hidden and output

layers. Each neuron has its own activation function. The input layer has no

activation function just introduced the input values. Hidden layers perform

the classiﬁcation of features obtained from the input layer and the output

layer produce output through an activation function. MLP is fully connected,

each node in one layer connected with certain weights to every node in the

following layer.

The enhancement in MLP is done by tuning the parameters. Three types

of solvers ‘lbfgs, sgd and adam’ are used for weight optimization in MLP

classiﬁer. In EMLP lbfgs solver gives better results than other two optimizers.

262 H. Bano et al.

•SVM Technique

SVM is a supervised machine learning algorithm which can be used for both

classiﬁcation and regression methods. The SVM linearly separable binary

sets of data. The objective to design a hyperplane in between the number

of features to classiﬁes all training vectors in two classes. Many hyperplanes

could be chosen to separate the two classes of data points. The plane has

the maximum margin distance provides future data points can be classiﬁed.

The eﬀectiveness of SVM depend on the high dimentional spaces of support

vectors (data points near to hyperplane) to the hyperplanes.

•LR Technique

LR is a statistical method which is used for predictive analysis. When one

or more independent variables exist in the dataset the logistic function (sig-

moid function) analyze the outcomes. The sigmoid function was developed to

describe the properties of population growth. In the plots the sigmoid func-

tion is represented by S-shape curve which take any real number and map it

into a value 0 and 1. This machine learning technique can be used for both

binary or multivariate classiﬁcation tasks.

Algorithm 1 MLP

1: Set MLP Network (mlp signals, weights and activation function)

2: Summation of signals with weights

3: Load and normalized the dataset

4: Splitting the dataset (training and testing)

5: Select training data

6: FOR nepochs and batchsize

7: Train the Network

8: ENDFOR

9: Run Prediction using Network

10: Calculate the loss function

Algorithm 2 SVM

1: Require: X and y loaded with training labeled data, a = 0

2: partially trained SVM

3: C = some value (for example 10)

4: repeat

5: FOR (xi, yi),(xj, yj )do

6: Optimize ai and aj

7: ENDFOR

8: until no change in a or other resource constraint criteria met

9: Ensure: Retain only the support vectors (ai > 0)

Electricity Load and Price Forecasting Using Enhanced Machine ... 263

5 Simulation and Results

In this paper the prediction of electricity load and price by using MLP, LR and

SVM Classiﬁers. The system type x64-based processor, 4.00 GB RAM and Intel

(R) core i5 processor. The implementation environment is in anaconda (spyder)

by using python language.

The four months (ﬁrst two months of 2016 and 2017) hourly electricity data

of New York are taken as input from NY-ISO wholesale electricity provider. The

diﬀerent features for example humidity, temperature, pressure, wind speed and

wind direction impact on the target load (TWIActualLoad) and target price

TWIZonalLBMP.

Figure 2shows the importance of features of a dataset in forecasting process.

Here we can see that the zonal transmission losses has more important role than

zonal price version in load forecasting. The features play an important role in

forecasting process, after the pre-processing of data, the normalized data of four

months is shown in Fig. 3.

Fig. 2. Feature importance of load Fig. 3. Normalized load data

Prediction of load and price of electricity market by using MLP, SVM, LR

with EMLP, ESVM and ELR techniques is shown in the following plots. The

hourly dayahead load with MLP and EMLP model and price prediction with

LR and ELR are shown in Figs. 4and 5.

Fig. 4. Dayahead load prediction Fig. 5. Dayahead price prediction

264 H. Bano et al.

The electricity load and price of ﬁrst week, January 2016 and four months

of electricity load with MLP and EMLP techniques are shown in Fig. 6.

Fig. 6. a Load prediction of 1st-week, bload prediction of January, celectricity load

of four months

Electricity price prediction of week ahead and month based are shown in

Fig. 7.

Fig. 7. a Price prediction of 1st-week, bprice prediction of January, cprice prediction

of four month

5.1 Performance Evaluation

Diﬀerent performance evaluators are used in this section for evaluation of the

simulation results. In particular four metrics are used, which include, Mean

Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Square Error

(RMSE), and Mean Absolute Percentage Error (MAPE). Figures 8and 9show

the performance graph of electricity load and price data.

Fig. 8. Performance matrix of load Fig. 9. Performance matrix of price

Electricity Load and Price Forecasting Using Enhanced Machine ... 265

MAE =1

n

n

t=1

|et|(1)

MSE =1

n

n

t=1

e2

t(2)

RMSE =

1

n

n

t=1

e2

t(3)

MAPE =100%

n

n

t=1

et

yt

(4)

Table 3shows the performance evaluation comparison of diﬀerent techniques

which are used for electricity load and price prediction of New York market by

using Eqs. 1–4.

Table 3. The diﬀerent evaluation criterion of price and load prediction

Model Criterion Price error values Load error values

MLP MSE

MAE

RMSE

MAPE

69

15.4

101.7

18.9

320.07

109.3

146.07

45.41

EMLP MSE

MAE

RMSE

MAPE

68.3

14.3

100.4

12.2

236.7

95.08

125.6

34.98

SVM MSE

MAE

RMSE

MAPE

69.08

15.5

101.7

19.14

207

83.03

117.7

31.28

ESVM MSE

MAE

RMSE

MAPE

68.3

14.3

100.4

12.2

206.7

81.12

117.3

29.5

LR MSE

MAE

RMSE

MAPE

67.14

14.7

100.35

17.37

294.7

112.9

140

32.52

ELR MSE

MAE

RMSE

MAPE

53.58

14.49

89.65

15.47

291.23

112.7

139

32.21

266 H. Bano et al.

6 Conclusion

In this paper, MLP, EMLP, SVM, ESVM, LR and ELR techniques are used

for the time series hourly prediction of load and price of NY-ISO electricity

market. Feature selection and extraction techniques (CART, RFE and SVD)

are used for the normalization of the dataset. These normalization techniques

remove the redundancy and irrelevant features which have less impact on the

forecasting process. The high dimensionality reduction of the dataset improved

the computation of the system and reduces cost. The LSTM and MLP models

presents clear representation of load and price forecast. In our scenario LSTM

model performs better prediction than MLP. Due to this problem, we proposed

enhanced MLP model for load and price prediction. The three techniques MLP,

SVM and LR gives 65, 64.85 and 82.62% accuracy for price prediction while

54.58, 68.71 and 67.47% for load prediction. The EMLP, ESVM and ELR models

present 76.30, 70.85, 84.52 and 79.40% and 65.01, 70.40 and 67.78% accuracy

for price and load prediction. These accurate prediction performance models

maximize the proﬁts and formulate the long term strategies for utility companies.

References

1. Jindal, A., Singh, M., Kumar, N.: Consumption-aware data analytical demand

response scheme for peak load reduction in smart grid. IEEE Trans. Ind. Electron.

65, 8993–9004 (2018)

2. Liu, C., Jin, Z., Gu, J., Qiu, C.: Short-term load forecasting using a long short-

term memory network. In: Innovative Smart Grid Technologies Conference Europe

(ISGT-Europe), 2017 IEEE PES, pp. 1–6. IEEE (2017)

3. Zheng, J., Xu, C., Zhang, Z., Li, X.: Electric load forecasting in smart grids using

long-short-term-memory based recurrent neural network. In: 2017 51st Annual

Conference on Information Sciences and Systems (CISS), pp. 1–6. IEEE (2017)

4. Wang, F., Li, ., Zhou, L., Ren, H., Contreras, J., Shaﬁe-Khah, M., Catal˜ao, J.P.:

Daily pattern prediction based classiﬁcation modeling approach for day-ahead elec-

tricity price forecasting. Int. J. Electr. Power Energy Syst. 105, 529–540 (2019)

5. Raviv, E., Bouwman, K.E., van Dijk, D.: Forecasting day-ahead electricity prices:

utilizing hourly prices. Energy Econ. 50, 227–239 (2015)

6. Mosbah, H., El-Hawary, M.: Hourly electricity price forecasting for the next month

using multilayer neural network. Can. J. Electr. Comput. Eng. 39(4), 283–291

(2016)

7. Wang, K., Xu, C., Zhang, Y., Guo, S., Zomaya, A.: Robust big data analytics for

electricity price forecasting in the smart grid. IEEE Trans. Big Data (2017)

8. Chahkoutahi, F., Khashei, M.: A seasonal direct optimal hybrid model of compu-

tational intelligence and soft computing techniques for electricity load forecasting.

Energy 140, 988–1004 (2017)

9. Ahmad, A., Javaid, N., Guizani, M., Alrajeh, N., Khan, Z.A.:. An accurate and

fast converging short-term load forecasting model for industrial applications in a

smart grid. IEEE Trans. Ind. Inf. 13(5), 2587–2596 (2017)

10. Liu, J.P., Li, C.L.: The short-term power load forecasting based on sperm whale

algorithm and wavelet least square support vector machine with DWT-IR for fea-

ture selection. Sustainability 9(7), 1188 (2017)

Electricity Load and Price Forecasting Using Enhanced Machine ... 267

11. Raﬁei, M., Niknam, T., Khooban, M.-H.: Probabilistic forecasting of hourly elec-

tricity price by generalization of ELM for usage in improved wavelet neural net-

work. IEEE Trans. Ind. Inform. 13(1), 71–79 (2017)

12. Keles, D., Scelle, J., Paraschiv, F., Fichtner, W.: Extended forecast methods for

day-ahead electricity spot prices applying artiﬁcial neural networks. Appl. Energy

162, 218–230 (2016)

13. Feng, C., Cui, M., Hodge, B.-M., Zhang, J.: A data-driven multi-model methodol-

ogy with deep feature selection for short-term wind forecasting. Appl. Energy 190,

1245–1257 (2017)

14. Wang, H.Z., Wang, G.B., Li, G.Q., Peng, J.C., Liu, Y.T.: Deep belief network based

deterministic and probabilistic wind speed forecasting approach. Appl. Energy 182,

80–93 (2016)

15. Wang, K., Xu, C., Zhang, Y., Guo, S., Zomaya, A.: Robust big data analytics for

electricity price forecasting in the smart grid. IEEE Trans. Big Data (2017)

16. Agrawal, R.K., Muchahary, F., Tripathi, M.M.: Long term load forecasting with

hourly predictions based on long-short-term-memory networks. In: Texas Power

and Energy Conference (TPEC), 2018 IEEE, pp. 1–6. IEEE (2018)

17. Abedinia, O., Amjady, N., Ghadimi, N.: Solar energy forecasting based on hybrid

neural network and improved metaheuristic algorithm. Comput. Intell. 34(1), 241–

260 (2018)

18. Kuo, P.-H., Huang, C.-J.: An electricity price forecasting model by hybrid struc-

tured deep neural networks. Sustainability 10(4), 1280 (2018)

19. Chen, Y., Li, M., Yang, Y., Li, C., Li, Y., Li, L.: A hybrid model for electricity

price forecasting based on least square support vector machines with combined

kernel. J. Renew. Sustain. Energy 10(5), 055502 (2018)

20. Liu, Y., Wang, W., Ghadimi, N.: Electricity load forecasting by an improved fore-

cast engine for building level consumers. Energy 139, 18–30 (2017)

21. Gao, W., Darvishan, A., Toghani, M., Mohammadi, M., Abedinia, O., Ghadimi, N.:

Diﬀerent states of multi-block based forecast engine for price and load prediction.

Int. J. Electr. Power Energy Syst. 104, 423–435 (2019)

22. Shayeghi, H., Ghasemi, A., Moradzadeh, M., Nooshyar, M.: Day-ahead electricity

price forecasting using WPT, GMI and modiﬁed LSSVM-based S-OLABC algo-

rithm. Soft Comput. 21(2), 525–541 (2017)

23. Khwaja, A.S., Naeem, M., Anpalagan, A., Venetsanopoulos, A., Venkatesh, B.:

Improved short-term load forecasting using bagged neural networks. Electric Power

Syst. Res. 125, 109–115 (2015)

24. Ghasemi, A., Shayeghi, H., Moradzadeh, M., Nooshyar, M.: A novel hybrid algo-

rithm for electricity price and load forecasting in smart grids with demand-side

management. Appl. Energy 177, 40–59 (2016)

25. Coelho, V.N., Coelho, I.M., Coelho, B.N., Reis, A.J.R., Enayatifar, R., Souza,

M.J.F., Guimar˜aes, F.G.: A self-adaptive evolutionary fuzzy model for load fore-

casting problems on smart grid environment. Appl. Energy 169, 567–584 (2016)

26. Tarsitano, A., Amerise, I.L.: Short-term load forecasting using a two-stage sarimax

model. Energy 133, 108–114 (2017)

27. Dongxiao, N., Tiannan, M., Bingyi, L.: Power load forecasting by wavelet least

squares support vector machine with improved fruit ﬂy optimization algorithm. J.

Comb. Optim. 33(3), 1122–1143 (2017)