Content uploaded by Nadeem Javaid
Author content
All content in this area was uploaded by Nadeem Javaid on Feb 17, 2020
Content may be subject to copyright.
Electricity Price Forecasting based on Enhanced
Convolutional Neural Network in Smart Grid
Nazia Daood, Zahoor Ali Khan, Ashrafullah, Muhammad Jaffar Khan, Muhammad
Adil and Nadeem Javaid
Abstract Electricity price forecasting is significant component of smart grid. Elec-
tricity systems are managed by the electricity market. The market operators perform
electricity price forecasting for an efficient energy management. This paper deals
with the electricity price forecasting based on deep learning. The fluctuations in
electricity prices are due to the increase in fuel prices, demand of electricity and so-
cial variables such as weather conditions, peak hours, weekdays, weekends and sea-
sons. Therefore, there is a need to maintain equilibrium between shortage and over-
flow of the electricity. Deep learning is most widely used for classification, image
recognition and forecasting. The proposed work is categorized into two stages: first
stage is feature engineering, in which features selection is performed by Xgboost
technique, while features extraction is done through Linear Discriminant Analysis
(LDA). These techniques reduce the dimensionality of data and forward important
data to classifier for electricity price forecasting. Second stage is price forecasting,
which is based on Enhanced Convolutional Neural Network (ECNN) classifier. For
validation of proposed work, three performance metrics (i.e., Mean Absolute Er-
ror (MAE), Root Mean Square Error (RMSE) and Mean Absolute Percentage Error
(MAPE)) are used. Simulation results show that our proposed scheme outperforms
existing benchmark techniques in terms of price forecasting.
Nazia Daood, Ashrafullah, Muhammad Jaffar Khan, Muhammad Adil, and Nadeem Javaid (Cor-
responding Author),
COMSATS University Islamabad; email: nadeemjavaidqau@gmail.com
Zahoor Ali Khan
Computer Information Science, Higher Colleges of Technology, Fujairah 4114, United Arab Emi-
rates; email: zkhan1@hct.ac.ae
1
2 Nazia et al.
Table 1: List of Abbreviations
Acronym Representation
Convolution Neural Network CNN
Enhanced Convolution Neural Network ECNN
Feature Selection and Extraction FSE
Independent System Operator New Eng-
land
ISO-NE
Linear Discriminant Analysis LDA
Long Short Term Memory LSTM
Mean Absolute Error MAE
Mean Absolute Percentage Error MAPE
Multilayer Perceptron MLP
Naive Base NB
Pennsylvania, Jersey, Maryland PJM
Root Mean Square Error RMSE
Smart Grid SG
Stochastic Gradient Descent SGD
Traditional Grid TG
1 Introduction
The demand for electricity increases due to exponential use of electricity by differ-
ent sectors such as industrial [1], commercial and residential [2]. Traditional grid is
used as electromechanical technology, which faces several difficulties like one-way
communication, central distribution, manual monitoring, few sensors and manual
restoration [3] [4]. That is why, there is a need for such system which saves us from
energy crises and has an efficient transmission. Smart grid is used as digital tech-
nology and distributed system, in which sensors are placed, self-monitoring is done,
quick restoration takes place and two-way communication from utility to consumer
is ensured. It also reduces the cost of maintenance and operation, and improves se-
curity system. Cost is an important factor which is involved on both sides, i.e., utility
and consumer [5]. Millions of smart meters are deployed in different places to col-
lect the records of electricity consumption from which information is transferred to
utilities [6]. Electricity price forecasting is a crucial task that is based on smart grid
data.
The shortage of electricity problem is mostly occurred due to high demand and to
control this problem, an accurate prediction of load and price of electricity must be
done [7]. The motive behind accurate load and price forecasting is to save electricity
cost and energy resources because electricity demand fluctuates on various weather
conditions [8]. Furthermore, handling and analyzing data is complicated task for
researchers. Thus, the large data is passed through features selection and extraction
Electricity Price Forecasting based on ECNN in Smart Grid 3
techniques in order to remove irrelevant and redundant information. The optimal
action plan made for reducing the load of electricity also decreases its price [9].
A novel model for day-ahead price forecasting is used, named as Dynamic Choice
Artificial Neural Network (DCANN) [10]. ARMAX model is an efficient, simple
and reliable model to achieve high accuracy [11]. A competent model CNN that is
more mature and efficient is used as motivation to perform price forecasting [12].
However, the price of electricity is increased due to affecting factors such as price
of fuel, renewable energy and consumption of electricity that need to be considered.
Problem Statement: Electricity price is very high in the market due to increas-
ing demand and less supply. Without electricity storage, variations in demand cause
high price fluctuations. Accurate price forecasting is, therefore, necessary for de-
cision making mechanisms by energy companies at the corporate level. In [13],
authors proposed Multilayer Perceptron (MLP) for electricity price forecasting,
through reasonable accuracy is achieved, however, it traps in local minima. As it
has characteristics of fully connected layers, it takes too many parameters which
result in high dimensionality due to this high dimensionality of MLP network, it
falls into overfitting problem. Only lags of load and price variables are used as input
features that result in low forecasting accuracy. In [14, 15, 16], authors proposed
deep LSTM model for price forecasting. However, this model has high computa-
tional time. LSTM may loss some information during training process. It requires
four linear layers per cell to run for each time-step. So, it takes large amount of
memory for computation. In [12], EPNet integrates CNN and LSTM for electricity
price forecasting. EPNet model improves price forecasting shown in experimental
results. Both of the models are deep neural network, due to which its computational
time is very high. In [17], Vector Auto Regression (VAR) is a statistical model that
deals with time-series data. VAR model forecasts demand and supply based on three
specific variables, i.e., solar radiation, temperature and wind speed. It finds correla-
tion among three different types of variables. Lag is composed of Akaike (AIC) and
Bayesian (BIC). It captures strong connection between various locations in same
period. This model is evaluated on the basis of different case studies from several
cities of United States. The model is categorized into three parts: first part is based
on hourly weather data in applied linear regression. Secondly, fourier series is ana-
lyzed to find seasonality pattern. In third part, VAR is used for stochastic time series
data.
In [18], a novel model is proposed for electricity price forecasting, which is de-
ployed on huge data in smart grid. Hybrid Selection, Extraction and Classification
(HSEC) model is introduced for price forecasting which consists of three modules.
ReliefF and Random Forest (RF) are used for feature selection based on Gray Co-
relation Analysis (GCA). Kernel Principal Component Analysis (KPCA) is used for
features extraction which reduces the high dimensionality of features. The data is
fetched from GCA and KPCA that is input data for Support Vector Machine (SVM).
SVM draws the hyper plane between two classes. To improve the performance of
SVM classifier, heuristic technique Differential Evolution (DE) is used to tune the
hyper parameter of SVM. It provides better result as compared to existing schemes
and reduces computational time.
4 Nazia et al.
In [19], EMD-Mixed-ELM model is used for short term electricity load forecast-
ing by half hourly data from three places of Australia. Empirical Mode Decompo-
sition (EMD) removes noisy features from the dataset and makes de-noised data.
Extreme learning machine (ELM) is applied for load forecasting. ELM performs
same like ANN however, low computational time of ELM. ELM is made up of sin-
gle hidden layer, bias and weight. Bias is randomly assigned to input layer. Least
square method is used to verify the weights of output layer. ELM does not provide
up to the mark error minimization. However, EMD-Mixed-ELM has achieved high
accuracy as compared to Radial Based Function (RBF), UKF kernel and Mixed-
ELM approaches.
In [20], experiment conducted on waveform data from unlabeled ferroresonance
data and Softmax regression has achieved favorite result. Sparse Autoencoder (SA)
is used for features extraction. However, the model has higher computational capa-
bility and less time consumption. Similar works are also done in [21]-[27].
2 Proposed Model
In order to resolve aforementioned problems, ECNN model for electricity price
forecasting is proposed. Being motivated from CNN proposed in paper [12], we
use CNN in our model with further enhancement. Traditional CNN is composed
of three layers that are convolution, pooling and fully connected layers. Each layer
has various parameters and these parameters perform different tasks. Min-max and
standard scales are prosecuted for the preparation of data. Normalized data is then
passed to feature engineering method, in which irrelevant and redundant features are
removed. The important features extracted is then fed into corresponding ECNN for
price forecasting. ECNN consists of network layers which pass the information from
one layer to another. These layers are briefly described as follows: Convolutional:
Convolutional layer takes normalized data and applies filters to input metrics. Pool-
ing: Pooling is the second layer that takes information from convolution layer and
calculates average and maximum value. It also reduces the dimensionality. There
are two type of pooling, one is max pooling and other is average pooling. Max
pooling takes greater values in certain regions. Fully connected or Dense Layer:
Each neuron is connected to the neuron of next layer. This means each layer is fully
connected to specific region and makes a layer which is known as receptive field.
Its structure is similar to multiple layer perceptron. Dense layers use data in matrix
form and converts it into vector’s form. The square area is used for specific region
(3 by 3). Filters: Filter is represented by a vector of weights with which convolves
the input. Each neuron has different weight and bias. Zero padding, stride and filter
size are the parameter values. Activation Function: There are various activation
functions available to calculate the output value such as Sigmoid, ReLU and Tan
etc.
Electricity Price Forecasting based on ECNN in Smart Grid 5
Foreccasting
Model
Features Features
Engineering Output
Input
Layer
Convolution
Layer
Pooling
Layer
Fully
Connected
Layer
Output
Layer
Fig. 1: System Model
2.1 Techniques Description
The first step in the proposed model is features selection using Xgboost techniques.
The second step is features extraction by using LDA. In third step, forecasting is
done by ECNN. Brief description of these techniques are given below.
XGboost: XGboost technique is imported through XGboost library. It deals with
both regression and classification problems. XGboost is used for features selection
which finds the importance of features. Best features are chosen on the bases of
grade, rank, weight or threshold values. The irrelevant features are removed. It is
straight forward way to calculate the score of each feature and compare than with
other features. First scaling the data (between 0 and 1) and set threshold value then
find important features in training set. The boosted trees are built in gradient boost-
ing.
Features Extraction: Feature extraction removes redundant features, compress
large dataset and reduces the dimensionality. Filtered data is provided to the classi-
fier in order to forecast electricity price, efficiently. In our proposed scheme, LDA
is used to extract most suitable features. PCA is unsupervised machine learning
technique, mainly use for dimensionality reduction. it ignores class label whereas
LDA is supervised. In contrast to PCA, LDA attempts to find feature subspace that
maximizes class disjointness.
Enhanced Convolutional Neural Network Electricity price is predicted by us-
ing different approaches; however, all have certain limitations. CNN works similar
to human brain in which one neuron connects to another neuron in the same region.
Each neuron passes the visual cortex information in the respected field. ECNN con-
sists of two parts: features learning and classification. The convolutional layer is first
layer that holds parameters, i.e. input, kernel or filters. Convolutional layer takes
normalized data and applies filter to input parameters. Kernel size is most important
parameter in convolutional layer. A kernel size of 3-by-3 is used in proposed model.
The objective of convolutional operation is to extract high order features.
6 Nazia et al.
Max pooling layer is added as second layer and It extracts dominant features in
convolved features. Max pooling layer trains model effectively. The pooling process
reduces time consumption, memory and dimensionality. Dense layer is next layer in
which neurons are fully connected to hidden layers neurons. And important advan-
tage of this layer is its parameter sharing property. Leaky Relu is update version
of Relu. Leaky Relu activation function handles dying neurons and adjusts training
loss value accordingly. It provides slightly better result than Relu.
Dropout is a regularization technique, which aims to reduce the complexity of the
model with the goal to prevent over fitting. Dropout layer is added that reduces the
computational time, converts high dimension matrix into low dimensions. Adam
optimizer is used that performs better than other optimization techniques such as
SGD, Adagrade and Adadelta. Adam looks for global solution with the help of
default parameters such as momentum, learning rate and decay. Flatten layer trans-
forms matrix (3 by 3) into vector (9 by 1) form and passes it to softmax classifier
for forecasting. From result it can be seen that our proposed model achieves higher
accuracy and less computational time as compared to benchmark models.
3 Simulation Results
Python platform is used for experimental analysis. The system specification used
for this purpose is window10, core i5, 4GB RAM and 500GB drive. Two datasets,
PJM2017 and ISO-NE2017 are used for experiments. Dataset Description In this
part, detailed information is provided about electricity prices, PJM dataset and ISO-
NE, which are used in proposed model for experiments. PJM and ISO-NE dataset
are collected from PJM and ISO-NE websites for the period of 1st January 2017
to 31st December. Preprocessing Initial dataset has inconsistency in data samples.
Table 2: Features of Dataset
Dataset Features
PJM dataset Date, Hour, Zone PTID, TWI zonal losses, DAM zonal locational based
marginal price, DAM Zonal losses, DAM zonal congestion, DAM constraint
cost, TWI zonal congestion, TWI zonal price version, TWI zonal locational
based marginal price
ISO-NE dataset Date, Hour, Day-ahead Demand, Dynamic demand , Day-ahead energy part,
Day-ahead price of congestion, Day-ahead marginal loss part, Marginal
price of dynamic location, Dynamic price of the energy part, Dynamic price
of the congestion part, Dynamic price of the marginal loss part, Dry bulb,
Dew point, System load, Region capacity price, Region service price
Data normalization and standardization techniques are applied as preprocessing in
PJM dataset which remove outliers, missing values and noisy data.
•First raw data is normalized.
Electricity Price Forecasting based on ECNN in Smart Grid 7
•Best feature is selected
•After selection, reduce dimensions of features
•Tuning the hyper parameter of classifier
•Minimized computational time
Fig. 2: Features selection based on XGboost from PJM dataset
Fig. 3: Features selection based on XGboost from ISO-NE dataset
Results Figure 2 shows all features of the PJM dataset. Dam Zonal LBMP shows
the higher value of all features. Dam construction cost and TWI Zonal Congestion
shows the equal value in plot, which means that both are of equal importance. How-
ever, others features show less values. In addition, figure 3 shows all features of the
ISO-NE dataset. Most of the feature’s value is above 0.5. We drop the feature with
less value
Figure 4 shows the features extraction through LDA and its comparison with
PCA using PJM dataset. Figure 5 shows the features extraction through LDA and
comparison with PCA from the ISO-NE dataset, accumulative contribution com-
parison among LDA with PCA is shown. When the accumulative contribution rate
8 Nazia et al.
Fig. 4: Features extraction PJM dataset
Fig. 5: Features extraction from ISO-NE dataset
reaches 95% LDA extract most of the component thus we select LDA to guarantee
at the accuracy of forecasting ISO-NE and PJM dataset are used for the evaluation
of proposed model. Figure 6 shows the performance of NB model and actual price.
Figure 7 shows the performance of SGD model and actual value. Whereas, figure
8 shows the performance of MLP model and actual value. In figure 9 the perfor-
mance of CNN model and actual value is shown. Figure 10 shows the performance
of ECNN model and actual value from the ISO-NE dataset. Figure 11 shows the
comparison of proposed schemes and existing schemes.
3.1 Evaluation Matrix
Three evaluation metrics are used to measure the error rate of proposed model and
existing approaches i.e., MAE, RMSE and MAPE.Figure 12 shows the performance
Electricity Price Forecasting based on ECNN in Smart Grid 9
Fig. 6: Comparison between actual and forecasted value using NB for one week
Fig. 7: Comparison between actual and forecasted value using SGD for one week.
Fig. 8: Comparison between actual and forecasted value using MLP for one week.
10 Nazia et al.
Fig. 9: Comparison between actual and forecasted value using CNN for one week.
Fig. 10: Comparison between actual and forecasted value using ECNN for one week.
Fig. 11: Comparison between proposed and existing schemes.
Electricity Price Forecasting based on ECNN in Smart Grid 11
Table 3: Hourly MAPE and RMSE calculation for three days
Hours Actual Forecast MAPE RMSE Actual Forecast MAPE RMSE Actual Forecast MAPE RMSE
0 7.00 29.51 321.59 22.51 11.00 29.96 17.33 18.96 26.00 28.21 8.51 2.21
1 40.00 27.56 31.09 12.44 22.00 29.83 35.57 7.83 37.00 28.59 22.72 8.41
2 20.00 28.84 44.18 8.84 16.00 29.12 81.99 13.12 67.00 29.33 56.22 37.67
3 22.00 28.50 29.52 6.50 16.00 29.10 81.88 13.10 13.00 28.08 11.03 15.08
4 25.00 27.76 11.04 2.76 16.00 27.74 73.37 11.74 34.00 28.01 17.62 5.99
5 51.00 29.36 42.43 21.64 11.00 28.49 15.96 17.49 57.00 28.87 49.36 28.13
6 18.00 28.69 84.41 155.31 27.00 28.25 4.63 1.25 34.00 28.40 16.47 5.60
7 15.00 28.99 93.27 13.99 31.00 28.74 7.30 2.26 24.00 28.53 18.87 4.53
8 28.00 28.14 0.52 0.14 24.00 29.04 21.02 5.04 9.00 28.34 214.92 19.34
9 13.00 28.17 11.66 15.17 26.00 28.18 8.40 2.18 17.00 30.49 79.34 13.49
10 12.00 28.32 13.99 16.32 23.00 29.63 28.82 6.63 23.00 28.99 26.03 5.99
11 20.00 28.77 43.83 8.77 10.00 28.63 18.28 18.63 24.00 27.86 16.09 3.86
12 13.00 29.89 12.90 16.89 18.00 28.61 58.94 10.61 20.00 28.69 43.44 8.69
13 22.00 29.25 32.97 7.25 13.00 28.58 11.86 15.58 14.00 27.91 99.33 13.91
14 52.00 32.68 37.16 19.32 13.00 28.94 12.63 15.94 13.00 28.29 11.65 15.29
15 13.00 29.18 12.48 16.18 48.00 29.01 39.57 18.99 24.00 29.26 21.90 5.26
16 32.00 27.94 12.68 4.06 20.00 28.79 43.93 8.79 20.00 28.85 44.24 8.85
17 22.00 29.78 35.37 7.78 19.00 28.28 48.86 9.28 21.00 29.87 42.22 8.87
18 66.00 29.14 55.85 36.86 50.00 28.41 43.18 21.59 37.00 29.00 21.62 8.00
19 19.00 28.98 52.51 9.98 15.00 28.53 90.17 13.53 26.00 29.14 12.07 3.14
20 39.00 27.54 29.38 11.46 19.00 28.95 52.38 9.95 26.00 28.99 11.50 2.99
21 21.00 29.77 41.76 8.77 32.00 28.24 11.74 3.76 8.00 30.88 28.95 22.88
22 25.00 28.74 14.95 3.74 3.00 30.31 91.45 27.31 25.00 28.12 12.47 3.12
23 1.00 29.17 28.89 28.17 14.00 29.18 10.42 15.18 3.00 30.48 91.08 27.48
of proposed model and existing approaches from the PJM dataset. From this figure,
it can be clearly seen that our proposed model outperforms with other models: NB,
SGD, MLP and CNN. Figure 13 shows the performance of proposed model and
existing approaches from the ISO-NE dataset.
MAPE =
1
n
n
∑
p=1
ep
yp
100% (1)
RMSE =s1
n
n
∑
p=1
e2
p(2)
MAE =
1
n
n
∑
p=1
|ep|(3)
4 Conclusion
In this paper, FSE based ECNN has been proposed for electricity price forecast in
smart grid. To ensure minimum maintenance and less operational cost, smart grid
performs vital role. It provides better way to generate the electricity according to the
demand of consumer. In recent years, lot of work has been done on load forecasting
however, price factor has not brought under consideration in-spite of the continuous
12 Nazia et al.
Fig. 12: Error comparison between existing schemes from PJM dataset.
Fig. 13: Error comparison between existing schemes from ISO-NE.
Table 4: MAE, MAPE and RMSE calculation for comparison between existing and proposed tech-
niques from ISO-NE dataset
Techniques MAPE RMSE MAE Execution
Time
NB 245 52 29 1.94
SGD 101 33 16 49.45
MLP 195 44 22 35.20
CNN 110 46 16 30.28
ECNN 71 32 18 25.11
fluctuation in price value. The proposed model is integrated using FSE and ECNN
classifier. The price of electricity forecasting is improved using feature engineering
technique. In feature engineering, important features are selected based on Xgboost
and dimensions are reduced from LDA. Price data is highly inconsistence and CNN
Electricity Price Forecasting based on ECNN in Smart Grid 13
Table 5: MAE, MAPE and RMSE calculation to compare existing and proposed schemes from
PJM dataset
Techniques MAPE RMSE MAE Execution
Time
NB 65 47 33 0.98
SGD 69.54 27.95 14.14 12.25
MLP 42.36 27.83 9.88 22.15
CNN 80.54 33.75 7. 39 20.93
ECNN 40 24 11 20.39
may not perform well. ECNN has better ability to learn dynamics features of price
data and as well as better generalization capability. The traditional approaches are
used for price forecasting. Moreover, the performance of proposed model is evalu-
ated and compared with existing techniques, i.e., NB, SGD, MLP, CNN and ECNN.
ECNN result shows that higher accuracy is achieved. Two datasets are used to ver-
ify the effectiveness of proposed model. Furthermore, daily, weekly, monthly and
seasons based price forecasting patterns are present. ECNN has minimized error as
compared to benchmarks. This study has achieved the purpose of proposed model,
i.e., to achieve accuracy with less computational time. In future, we may check the
load of different sector such as residential, public area and commercial.
References
1. Chen, Y., Tan, H., Song, X. (2017). Day-ahead forecasting of non-stationary electric power
demand in commercial buildings: hybrid support vector regression based. Energy Procedia,
105, 2101-2106.
2. Guo, Z., Zhou, K., Zhang, C., Lu, X., Chen, W., Yang, S. (2018). Residential electricity con-
sumption behavior: Influencing factors, related theories and intervention strategies. Renew-
able and Sustainable Energy Reviews, 81, 399-412.
3. Tang, N., Mao, S., Wang, Y., Nelms, R. M. (2018). Solar power generation forecasting with a
LASSO-based approach. IEEE Internet of Things Journal, 5(2), 1090-1099.
4. Vrablecov´
a, P., Ezzeddine, A. B., Rozinajov´
a, V., ˇ
S´
arik, S., Sangaiah, A. K. (2018). Smart grid
load forecasting using online support vector regression. Computers Electrical Engineering,
65, 102-117.
5. Singh, N., Mohanty, S. R., Shukla, R. D. (2017). Short term electricity price forecast based
on environmentally adapted generalized neuron. Energy, 125, 127-139.
6. Agrawal, R. K., Muchahary, F., Tripathi, M. M. (2019). Ensemble of relevance vector ma-
chines and boosted trees for electricity price forecasting. Applied Energy, 250, 540-548.
7. Aryanpur, V., Atabaki, M. S., Marzband, M., Siano, P., Ghayoumi, K. (2019). An overview of
energy planning in Iran and transition pathways towards sustainable electricity supply sector.
Renewable and Sustainable Energy Reviews, 112, 58-74.
8. Wang, F., Li, K., Zhou, L., Ren, H., Contreras, J., Shafie-Khah, M., Catal ˜
ao, J. P. (2019). Daily
pattern prediction based classification modeling approach for day-ahead electricity price fore-
casting. International Journal of Electrical Power Energy Systems, 105, 529-540.
14 Nazia et al.
9. Wang, J., Liu, F., Song, Y., Zhao, J. (2016). A novel model: Dynamic choice artificial neural
network (DCANN) for an electricity price forecasting system. Applied Soft Computing, 48,
281-297.
10. Gonz´
alez, J. P., San Roque, A. M., Perez, E. A. (2017). Forecasting functional time series
with a new Hilbertian ARMAX model: Application to electricity price forecasting. IEEE
Transactions on Power Systems, 33(1), 545-556.
11. Xiao, F., Wang, S., Fan, C. (2017, May). Mining big building operational data for building
cooling load prediction and energy efficiency improvement. In 2017 IEEE International Con-
ference on Smart Computing (SMARTCOMP) (pp. 1-3). IEEE.
12. Kuo, P. H., Huang, C. J. (2018). An electricity price forecasting model by hybrid structured
deep neural networks. Sustainability, 10(4), 1280.
13. Gholipour Khajeh, M., Maleki, A., Rosen, M. A., Ahmadi, M. H. (2018). Electricity price
forecasting using neural networks with an improved iterative training algorithm. International
Journal of Ambient Energy, 39(2), 147-158.
14. Jiang, L., Hu, G. (2018, November). Day-ahead price forecasting for electricity market using
long-short term memory recurrent neural network. In 2018 15th International Conference on
Control, Automation, Robotics and Vision (ICARCV) (pp. 949-954). IEEE.
15. Ugurlu, U., Oksuz, I., Tas, O. (2018). Electricity price forecasting using recurrent neural
networks. Energies, 11(5), 1255.
16. Afrasiabi, M., Mohammadi, M., Rastegar, M., Kargarian, A. (2019). Probabilistic deep neural
network price forecasting based on residential load and wind speed predictions. IET Renew-
able Power Generation, 13(11), 1840-1848.
17. Yixian, L. I. U., Roberts, M. C., Sioshansi, R. (2018). A vector autoregression weather model
for electricity supply and demand modeling. Journal of Modern Power Systems and Clean
Energy, 6(4), 763-776.
18. Wang, K., Xu, C., Zhang, Y., Guo, S., Zomaya, A. Y. (2017). Robust big data analytics for
electricity price forecasting in the smart grid. IEEE Transactions on Big Data, 5(1), 34-45.
19. Chen, Y., Kloft, M., Yang, Y., Li, C., Li, L. (2018). Mixed kernel based extreme learning
machine for electric load forecasting. Neurocomputing, 312, 90-106.
20. Chen, K., Hu, J., He, J. (2016). A framework for automatically extracting overvoltage features
based on sparse autoencoder. IEEE Transactions on Smart Grid, 9(2), 594-604.
21. Samuel, O., Alzahrani, F.A., Hussen Khan, R.J.U., Farooq, H., Shafiq, M., Afzal, M.K. and
Javaid, N., 2020. Towards Modified Entropy Mutual Information Feature Selection to Fore-
cast Medium-Term Load Using a Deep Learning Model in Smart Homes. Entropy, 22(1),
2020.
22. Khalid, R., Javaid, N., Al-zahrani, F.A., Aurangzeb, K., Qazi, E.U.H. and Ashfaq, T., 2020.
Electricity Load and Price Forecasting Using Jaya-Long Short Term Memory (JLSTM) in
Smart Grids. Entropy, 22(1), 2020.
23. Mujeeb, S. and Javaid, N., 2019. ESAENARX and DE-RELM: Novel schemes for big data
predictive analytics of electricity load and price. Sustainable Cities and Society, 51.
24. Mujeeb, S., Alghamdi, T.A., Ullah, S., Fatima, A., Javaid, N. and Saba, T., 2019. Exploiting
Deep Learning for Wind Power Forecasting Based on Big Data Analytics. Applied Sciences,
9(20).
25. Naz, A., Javaid, N., Rasheed, M.B., Haseeb, A., Alhussein, M. and Aurangzeb, K., 2019.
Game Theoretical Energy Management with Storage Capacity Optimization and Photo-
Voltaic Cell Generated Power Forecasting in Micro Grid. Sustainability, 11(10).
26. Naz, A., Javed, M.U., Javaid, N., Saba, T., Alhussein, M. and Aurangzeb, K., 2019. Short-
term electric load and price forecasting using enhanced extreme learning machine optimiza-
tion in smart grids. Energies, 12(5).
27. Mujeeb, S., Javaid, N., Ilahi, M., Wadud, Z., Ishmanov, F. and Afzal, M.K., 2019. Deep long
short-term memory: A new price and load forecasting scheme for big data in smart cities.
Sustainability, 11(4), p.987.
28. Guo, Z., Zhou, K., Zhang, X., Yang, S. (2018). A deep learning model for short-term power
load and probability density forecasting. Energy, 160, 1186-1200.
