Volume 8, Issue 9, September – 2023 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Application of Hybrid Ensemble Machine Learning

Approach For Prediction of Residential Natural Gas
Demand and Consumption
Agabus Aminu Fatima Umar Zambuk
Department of Computer Science Department of Mathematical Science
Nigerian Army University Biu Abubakar Tafawa Balewa University
Borno, Nigeria Bauchi, Nigeria

Abdulsalam Ya’u Gital Mustapha Abdulrahman Lawal

Department of Mathematical Science Department of Mathematical Science
Abubakar Tafawa Balewa University Abubakar Tafawa Balewa University
Bauchi, Nigeria Bauchi, Nigeria

Yusuf Pyelshak Ismail Zahraddeen Yakubu

Department of Health Information Management Department of Computing Technologies
College of Health Technology Zawan SRM Institute of Science and Technology
Plateau State, Nigeria Chennai, India

Abstract:- The only byproducts of burning natural gas Keywords:- Ensemble, Hybrid, Machine Learning, Natural
are carbon dioxide, water vapor, and very little amounts Gas, Prediction.
of nitrogen oxide, making it the cleanest fossil fuel on the
planet. A wide range of consumer products, such as I. INTRODUCTION
stoves, dryers, fireplaces, and furnaces, are also powered
by natural gas. At least one of your appliances Natural gas has long been used in all aspects of human
undoubtedly runs on natural gas. In this work, the endeavor, but particularly in home construction. Natural gas is
demand for residential natural gas was forecasted using a the cleanest fossil fuel on earth because the only byproducts
hybrid ensemble regression machine learning approach. of burning it are carbon dioxide, water vapor, and very little
Accurate forecasting of the demand for natural gas is amounts of nitrogen oxide. Natural gas is also used to power a
crucial for effective energy management and resource sizable variety of consumer goods, including stoves, dryers,
allocation. The hybrid ensemble approach mixes a fireplaces, and furnaces. At least one of your appliances
number of regression algorithms, including linear undoubtedly runs on natural gas. Like almost all other energy
regression (LR), decision tree regression (DTR), support sources, natural gas can be dangerous if used improperly. A
vector regression (SVR), and K-nearest neighbor (KNN), few simple safety measures and knowledge on what to do in
to take advantage of the benefits of each unique model the event of a gas leak can help you protect yourself and those
and improve prediction performance. The hybrid you love. Therefore, a well-designed forecasting model is
ensemble regression model's process has two steps. In the crucial to managing energy policy successfully by providing
first stage, distinct regression models are trained on the energy diversity and energy requirements that adapt to the
dataset. The second stage involves evaluating each dynamic structure of a country, region, or the entire world, in
model's predictions. To evaluate the effectiveness of the line with previously unheard-of increases in energy demand
hybrid ensemble model, a range of measures, including (Faruk et al., 2019).
mean absolute error (MAE), mean squared error (MSE),
coefficient of determination (R-squared), and accuracy, As a fossil fuel, natural gas is one of them. It is
are generated and compared to those of individual environmentally friendly to use natural gas to supply energy
regression models. The anticipated accuracy of the model requirements for industry, transportation, and other uses.
is further assessed using cross-validation techniques to Between a fifth and a third of the energy used worldwide is
ensure resilience. The results of the experiment consumed by buildings. Natural gas accounts for more than a
demonstrated that the hybrid ensemble regression third of the energy used in European residential buildings
technique routinely outperformed individual regression (Then et al., 2020). The strategic growth of nations'
models in terms of prediction accuracy. Combining economies and societies depends on energy. The basic
numerous models enables the collection of the various purpose of data mining is to create models using preprocessed
correlations and patterns contained in the data, or existing data to find patterns that the data set's attributes
enhancing the model's overall performance. have shown. Energy plays a crucial role in the strategic
development of society and economies. To uncover patterns
that the data set's features have revealed, data mining's

IJISRT23SEP126 www.ijisrt.com 158

Volume 8, Issue 9, September – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
fundamental goal is to build models utilizing preprocessed or most frequently generated on a daily and annual level on a
existing data. Because they show how the attributes are nation area level.
related to one another, some of the patterns act as
explanations. According to the most recent statistics, other Natural gas is one of the main sources of energy in the
modes forecast prospective values for specific attributes. In entire world, according to a study by (Laib et al., 2019). By
data mining, four different types of patterns are commonly creating a Multi Layered Perceptron (MLP) neural network as
sought after association patterns, prediction patterns, a nonlinear predicting monitor, it presents a novel hybrid
clustering patterns, and sequencing patterns (Chen & Zhang, prediction model that addresses the weakness of the two-stage
2022). method. The approach offers a fresh, useful functionality. The
results of the forecast performance were significantly
The need to predict how much natural gas demand and improved by its estimates of the next day's consumption
consumption is required for residential addresses is of profile, especially during seasons with strong client demand
paramount important as it will help in planning and and consumption.
scheduling activities including making a sound decision
(Driven et al., 2019). In some countries, the demand for In their study (Bala and Shuaibu, 2022), the authors
natural gas depend on seasons, like in Algeria for example the employed machine learning and hybrid techniques to forecast
demand is higher during holidays and special days (Laib et al., the energy consumption of the United Kingdom. However,
2019). Likewise in Nigeria the demand for natural gas machine learning techniques like neural networks and support
particularly premium motor spirit is higher most often during vector regression outperformed dynamic regression models,
the last quarter of the year. while the seasonal hybrid model outperformed time series and
machine learning models, even though combination forecasts
Many researchers have gone in to it (Šebalj, 2019), such underperformed other models.
as (Liu et al., 2022) who acknowledged that the impact of
climate change may be lessened, people's living conditions Energy consumption modeling for the residential and
could be improved and reasonable and appropriate private and commercial sectors was also measured using machine
governmental policies could be influenced with accurate learning. These models are developed based on a number of
predictions of natural gas demand and consumption in both factors, including population, natural gas and electricity costs,
residential and commercial structures. The reminder of the gross domestic product, and the share of renewable energy
paper is organized as follows: Section II present the existing sources in total energy consumption. The results of the three
relevant literature review, in section III the methodology machine learning algorithms show that by 2040, Iranian
employed was stated. Section IV provides the results obtained residential and commercial energy consumption will be 76.97,
and discussion. Finally, conclusion was drawn in section V. 96.42, and 128.09 Mtoe, respectively. According to the
findings, Iran must develop and implement new policies to
II. RELATED WORK increase the share of renewable energy sources in overall
energy consumption (Nabavi et al., 2020).
The recent research, which ranges from the last four
years, were examined in this study, and is reviewed in this Using certain hybrid algorithms inspired by nature,
section. There are a lot of studies in the literature. Machine (Qiao et al., 2020) predicted the monthly natural gas usage.
learning algorithms were employed in the work by (Faruk et According to the study, the invasive weed optimization (IWO)
al., 2019) to forecast the consumption of natural gas in the can be a more accurate predictive network when compared to
province of Istanbul. The results demonstrate that the support other hybridization strategies for enhancing the MLP method's
vector regression (SVR) technique is much superior to the error performance. According to the anticipated outcomes,
artificial neural network (ANN) technique for time series combining optimization techniques could increase prediction
forecasting of natural gas consumption, providing more precision and boost the multi-layer perceptron's (MLP)
reliable and accurate results in terms of fewer prediction productivity.
errors. This study may very likely be used as a useful
benchmarking study for many developing nations because of Artificial neural networks (ANN) were also used to
the data format, customer consumption frequency, and forecast ultimate natural gas usage. Speculate about how
consumption behavior over multiple time periods. natural gas will be used in the future. For Myanmar, actual
recorded consumption data from 1990 to 2015 are used. The
(Šebalj et al., 2019) suggests analyzing methodologies most recent five years' worth of data (2011-2015) are
and procedures for predicting the consumption of natural gas. combined with the prior years' data for testing. The years
The paper offers a comprehensive synthesis and analysis of (1990–2010) are used for instruction. Data on the population,
the most recent research on natural gas consumption GDP, and other factors that influence the final use of natural
projections for residential and commercial use. The results gas are used to develop the model. The training model is
show that while neural networks are the most popular, genetic reliable (MSE) with a minimum error rate of 0.005 Mean
algorithms, support vector machines, and Adaptive Network- Squared Error. Myanmar's yearly natural gas consumption is
based Fuzzy Inference System (ANFIS) are the best accurate projected using the created ANN model (Yin & Htay, 2021).
approaches for estimating natural gas demand. The two most
frequently used input variables are past natural gas
consumption data and meteorological data, and predictions are

IJISRT23SEP126 www.ijisrt.com 159

Volume 8, Issue 9, September – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Then et al., (2020) modeled a decline in gas demand in A consistent intraday load forecasting technique for the
57 metropolitan German distribution grids to examine the natural gas flow state space model is presented in the work by
effects of various distribution network operator (DNO) Chen and Zhang (2022). The trials' findings show that the
strategies and grid-specific factors on grid charges. By model's load forecasting accuracy and relative error both
analyzing the entire operating costs and related grid charges reached 0.026 and 98.5%, respectively, solving the problem of
for various scenarios and strategies, the implications on DNO processing the long-term collected historical data of gas intra-
business models are calculated. Depending on the DNO's day load. The calculation's data input was also reduced by
policy and the pattern of consumers leaving the grid, grid 33.6%, which promoted qualitative research and the
prices could increase as a result of the substitution of gas- quantification of factors influencing intraday load.
bound equipment, which could generate a self-reinforcing
feedback loop and lead to grid defection. Machine learning was picked after looking through the
models that were available. In particular, a genetic algorithm
Data mining techniques are first used to uncover and (GA) and artificial neural networks (ANN) were coupled.
gather the hidden patterns of electricity usage in the data. The Their suggested models were put to use in a real-world
cluster centers are used to divide the level of electricity usage testbed. They made advantage of CompactRIO to put ANNs
after the clustering analysis has been submitted to the particle into practice. To train and validate the given models, real-
swarm optimization-K-means algorithm. Finally, a workable, world data from a PV system and SB electrical appliances is
effective classification model is proposed (Cai et al., 2019), used. Despite the model's modest prediction accuracy, which
which uses a support vector machine as the core framework is brought on by the small size of the data set (Bourhnane et
for optimization. The results demonstrate that the accuracy al., 2020), the model is highly advised as a blueprint for
and F-measure of the novel model, which both significantly researchers willing to deploy real-world SB testbeds and
outperform conventional methods, reach 96.8% and 97.4%, investigate machine learning as a promising venue for energy
respectively. consumption prediction and scheduling.

This study aims to address the problems raised by (Shapi Investigate data-driven prediction models for predicting
et al., 2020), who built a prediction model for energy natural gas prices based on well-known machine learning
utilization in the cloud-based machine learning platform of techniques, such as Gaussian process regression, support
Microsoft Azure. A case study involving two tenants of a vector machines (SVM), and artificial neural networks (ANN)
business building is utilized to highlight real-world (GPR) (Driven et al., 2019). For evaluation and the cross-
applications in Malaysia. The obtained data is evaluated and validation method of model training, we use monthly Henry
prepared prior to being used for training and testing models. Hub natural gas spot pricing data from January 2001 to
To compare the potency of each strategy, RMSE, NRMSE, October 2018. The results show that these four machine
and MAPE metrics are used. The experiment's findings show learning methods perform differently when attempting to
that the distribution of energy use differs depending on the forecast natural gas prices. SVM, GBM, and GPR all perform
renter. predictions less accurately than ANN does generally. This has
proven that machine learning classifiers can accurately predict
It was possible to accurately predict the consumption of outcomes for many tasks.
natural gas in Istanbul's Bahçeşehir by using a range of
powerful machine learning techniques. Final performance The research could be used to considerably enhance
evaluations showed that XGBoost outperformed MLP, Forest natural gas consumption forecast systems, despite its
Regression, and Linear Regression by 0.04 Mean Absolute limitations due to the small number of articles it looked at. For
Error each. Because it is highly scalable, efficiently cuts down supply-demand equilibrium and investment purposes, accurate
on compute time, and makes best use of memory, XGBoost predictions of natural gas consumption are crucial. Accurate
performs better at prediction (Ahmed et al., 2021). Accurate and exact forecasts prevent economic loss and keep supply
forecasts prevent economic loss and keep supply and demand and demand in equilibrium.
in balance.
III. METHODOLOGY
The amount of natural gas used by data centers was
calculated by (Liu et al., 2022). Under the premise that The suggested model system, which is represented in the
electricity and gas are employed as energy suppliers of energy accompanying picture, will forecast natural gas demand and
supply and energy consumption from two viewpoints, the data consumption from both residential and commercial
center energy scheduling model is built by taking into account components using two separate datasets and will be
the service level of the data center. The lowest model is the implemented using a machine learning technique-based data
scheduling calculation model used in the data center. The best mining approach. The model would attempt to fix the issue
model is the data center energy supply scheduling model. The identified with the current system. The addition of a strategy
particle swarm approach is used to mimic the timetable. The for noise reduction in the dataset is due to the fact that noisy
results show that, while accounting for the degree of data data will lead to biased predictions, which could lead to
center service delay, using natural gas as an additional energy inaccurate performance accuracy. Two datasets would be
source can significantly reduce the data center's overall energy developed from different sources in order to further ensure
consumption. good and reliable performance.

IJISRT23SEP126 www.ijisrt.com 160

Volume 8, Issue 9, September – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Fig 1. Architecture of the Proposed Model

A. Natural Gas Dataset to train and change the model to boost its effectiveness and
Dataset will be collected from kaggle.com which is the accuracy.
largest community of data scientist
(https://www.kaggle.com/datasets). These are the stages involved in fine tuning
 Loading the pre-trained model
B. Data Preprocessing  Freeze most if not all layers in the model to prevent them
Data modifications done on it before feeding it to the from further training
algorithm are referred to as pre-processing. Data  Swap final layer or layers of the model with a new one
preprocessing is a method for transforming unclean data into that are specific to the tack
clean data sets. In other words, anytime data is acquired from  Train the model on dataset using a lower learning rate than
various sources, it is done so in a raw manner that makes in the pre-trained phase
analysis impossible. Data cleaning, dimensionality reduction,  Analyze the performance of the improved model and make
feature engineering, data sampling, data transformation, and any necessary hyper-parameter adjustments.
imbalanced data are a few of the crucial data preprocessing
approaches. E. Prediction Tools and Methods
Throughout this study, the Python programming
This crucial stage is utilized to improve the data's quality language would be used. To assist with the experiment, Sci-
in order to encourage the extraction of valuable insights. To kit Learn libraries would be used. Prediction model-specific
prevent overfitting or underfitting the suggested developed libraries and extensions are abundant in the Python
model, it is also important to make sure that there isn't too programming language. One of the finest places to get
much noise in the dataset. Boost the model performance's machine learning algorithms is Sci-kit Learn (https://scikit-
accuracy and computational effectiveness to get the dataset learn.org), where almost every form of algorithm is easily
ready for the right prediction. accessible and can be evaluated quickly and easily. The data
will be handled using Numpy and Pandas. Jupyter Notebooks
C. Feature Extraction would be utilized for debugging and for its ability to present
In order to decrease dataset overfitting, increase code elegantly.
prediction accuracy, and shorten model training time, feature
extraction is crucial. The properties of the dataset that would F. System Specification/Configuration
be utilized to train the machine learning models have a A laptop computer with the following specifications
significant impact on the algorithm's performance. Model would be used to perform and/or implement these research
performance may be negatively impacted by irrelevant, experiments: Intel(R) Core(TM) i5-2520M CPU @ 2.50GHz,
unsuitable, or only partially relevant features. In order to RAM 4.00 GB, 64-bit operating system, x64-based processor.
automatically choose the features in the dataset that contribute
the most to the output or prediction variable that interests us, G. Performance Evaluation Technique
feature selection will be applied to the data. Unrelated Building a prediction model using machine learning
elements in the data might make models less accurate, techniques has as its main objective the generality of the
especially those that use linear techniques like logistic and training datasets. On real data, machine learning models
linear regression. should be able to perform fairly effectively. Training data and
testing data will be separated into two groups. Testing data
D. Fine Tuning will be used to put machine learning classifiers to the test,
In order to enhance the performance of a machine whilst training data will be used to train classifiers.
learning model that has already been trained, fine-tuning often
refers to the process of further training the model on a H. Choice of Evaluation Metrics
particular task or dataset. This is done to make use of the data- Evaluation metrics are employed to measure the
driven knowledge and information the model has previously magnitude of mistakes in the performance of the prediction
gathered during the pre-training phase. It enables you to apply models. It aids in properly determining which of the results
prior knowledge to a new assignment where you can continue acquired is more accurate and reliable for application and

IJISRT23SEP126 www.ijisrt.com 161

Volume 8, Issue 9, September – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
further research. Regression, in contrast to classification, sometimes known as the root mean square error (RMSE), is a
entails making predictions about a continuous or quantitative measure of error. The units are returned to the output
variable. Mean Absolute Error (MAE), Mean Squared Error variable's original units when the mean squared error is taken
(MSE), and R Square (R2) Metric are the three (3) primary into account, which might be significant for explanation and
metrics for assessing predictions on regression machine presentation. The calculation of mean squared error is
learning issues. Although there are many of these tools illustrated in the formula below.
available, this study will examine the best and most widely
used measures for assessing each model's success. These
measurements or indices aid in determining how accurate the
models being utilized are (Qiao et al., 2020). The values of the
parameters used in the equations are calculated as follows:  R Square (R2)
= Actual value Usually, regression models employ this metric. It
= Predicted value provides a measurement of how well a collection of
predictions fits the data. The coefficient of determination is
= Number of data points/rows the statistical name for this measurement. For no-fit and
= Mean of all actual values perfect fit, it is a value between zero (0) and one (1). This
formula is used to determine the mean R2 given a group of
 Mean Absolute Error (MAE) predictions.
The sum of the absolute differences between the
expected and actual values is known as the mean absolute
error, or MAE. It illustrates just how inaccurate the
projections were. The metric provides a notion of the error's
size but not its direction, such as over prediction or under IV. RESULT AND DISCUSSION
prediction.
The longest period dataset in the literature was gathered
from the Kaggle machine learning repository. It contains
monthly figures of natural gas usage from January 1997 to
August 2020, as presented in Figure 2 below.
 Mean Square Error (MSE)
Similar to the mean absolute error in that it gives a broad
idea of the extent of error, the mean square error (MSE),

Fig 2: Natural Gas Demand History

TABLE I. COMPARISON OF PREDICTION MODELS PERFORMANCE

Model MSE MAE R^2 score Accuracy
LR 5.517869605 1.692801814 0.017428471 95.38447525
DTR 0.570523256 0.487209302 0.89840646 59.0135889
SVR 2.448486661 0.995936355 0.89840646 60.49370316
KNN 0.489458884 0.444046512 0.912841658 92.4553162
HE 1.164821453 0.721612403 0.792579296 97.48707913

IJISRT23SEP126 www.ijisrt.com 162

Volume 8, Issue 9, September – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Fig 3: Mean Squared Error (MSE) Fig 6: Accuracy Score

The mean square error (MSE), also known as the root The accuracy performance measures are displayed in
mean square error (RMSE), is shown in Figure 3. It is similar Figure 6 and show that the hybrid ensemble machine learning
to the mean absolute error in that it gives a general notion of model has the highest accuracy, followed by linear regression
the size of the error. The LR has a significant amount of and K-nearest neighbor.
prediction error, as depicted in the figure. SVR and Hybrid
Ensemble (HE) come next.

Fig 7: Comparison of Models Used

The figure 7 shows the comparison of all the various

Fig 4: Mean Absolute Error (MAE) machine learning model used in the study as also shown in
Table 1 above. As displayed in the figure above the hybrid
Figure 4 above depicts the mean absolute error (MAE), ensemble have the highest prediction accuracy and minimum
which is used to illustrate how inaccurate the forecasts were. error errors. Therefore, for efficient energy management and
The MAE results produce similar performance for the various decision-making, a precise forecast of residential natural gas
machine learning models as the MSE results did. demand is essential. For energy management organizations
and regulators, the hybrid ensemble regression approach
offers a potent tool for making well-informed decisions about
resource allocation, infrastructure development, and energy-
saving strategies. By assuring reliable and efficient natural gas
transportation, energy companies may streamline their supply
chains, reduce waste, and boost customer satisfaction.

 Further Discussions
A hybrid ensemble regression machine learning
approach has some notable benefits and consequences for
predicting residential natural gas demand and consumption.
The discussion examines the approach's main principles and
Fig 5: R^2 score any potential field repercussions. When compared to
individual regression models, the hybrid ensemble regression
The outcomes of the r-square evaluation model for all approach showed greater prediction accuracy. The method
models were reported, as seen in figure 5. It gives a hint as to captures a wider range of patterns and correlations contained
how well a collection of predictions fits the data in terms of in the data by merging different algorithms, such as LR, SVR,
actual values. It is employed to further examine and guarantee DTR, and KNN. As a result, estimates of residential natural
the accuracy of the prediction performance results obtained gas demand become more precise and trustworthy, which is
from the created model. very advantageous for energy management firms, decision-

IJISRT23SEP126 www.ijisrt.com 163

Volume 8, Issue 9, September – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
makers, and energy providers when planning and maximizing ACKNOWLEDGMENT
resource allocation.
We wish to thank our supervisors Dr. F.U Zambuk and
The usage of ensemble techniques improves the Prof. A.Y Gital for their expert support and guidance towards
prediction model's resilience and generalization abilities. The the success of this programmed.
hybrid ensemble strategy decreases the possibility of over-
fitting and improves model stability by including a variety of REFERENCES
regression models. This is crucial when predicting residential
natural gas demand because the addition of numerous [1]. Abubakar, A., and Ugail, H., (2019), "Discrimination of
variables, such as socioeconomic and meteorological human skin burns using machine learning," in
variables, might result in complex interactions. The hybrid Intelligent Computing-Proceedings of the Computing
ensemble approach makes sure the model can handle various Conference, pp. 641-647: Springer.
situations and adjust to changes in the data. [2]. Ahmed, S., Madanian, S., & Mirza, F. (2021).
Prediction of Natural Gas Consumption in Bahçe ş ehir
The hybrid ensemble regression approach is flexible and Using Machine Learning Models. February.
able to be adjusted to a variety of prediction scenarios. The [3]. Abba, S., Boukari, S., Ajuji, M., and Muhammad, A. N.
ensemble process' fusion techniques, including stacking or (2022). Performance Evaluation and Analysis of
weighted averaging, allow for many combinations of separate Supervised Machine Learning Algorithms for Bitcoin
models. Moreover, the hybrid ensemble approach can be Cryptcurrency Price Forecast. International Journal of
easily extended to incorporate additional regression models or Computer Science and Security (IJCSS), 16(3), 28-42.
incorporate domain-specific knowledge, further enhancing its [4]. Bala, D. A., & Shuaibu, M. (2022). Forecasting United
adaptability. Kingdom ’ s energy consumption using machine
learning and hybrid approaches.
V. CONCLUSION https://doi.org/10.1177/0958305X221140569
[5]. Barik, S. S., & Nayak, S. (2021). Electricity
The study concludes that the hybrid ensemble regression Consumption & Prediction using Machine Learning
machine learning methodology is an effective technique for Models Electricity Consumption & Prediction using
precisely forecasting residential natural gas demand and Machine Learning Models Tapas Ranjan Jena Regn .
consumption. The hybrid ensemble approach captures various No . -180303110001 , Department of Computer Science
patterns and correlations within the data and improves and Engineering , Centurion University of Technology
predictive performance by merging numerous regression and Management , Odisha , India
models. The evaluation metrics repeatedly showed that the [6]. Bourhnane, S., Riduan, M., Rachid, A., Khalid, L.,
hybrid ensemble model was superior to individual regression Dine, Z., & Elkamoun, N. (2024). Machine learning for
models, including MAE, RMSE, and R-squared. The energy consumption prediction and scheduling in smart
outcomes showed how effective this strategy could be at buildings. 2020.
improving energy planning and resource management in the [7]. Cai, H., Shen, S., Lin, Q., Li, X., & Xiao, H. U. I.
residential natural gas market. The results point to the (2019). Predicting the Energy Consumption of
potential value of the suggested hybrid ensemble technique as Residential Buildings for Regional Electricity Supply-
a tool for precise and dependable residential natural gas Side and Demand-Side Management. IEEE Access, 7,
demand prediction, enabling more effective energy planning 30386–30397.
and resource management. https://doi.org/10.1109/ACCESS.2019.2901257
[8]. Chen, L., & Zhang, J. (2022). A Forecast Model of City
 Future Research Natural Gas Daily Load Based on Data Mining.
Though the hybrid ensemble regression method Scientific Programming, 2022.
produces encouraging findings, there are still areas that could https://doi.org/10.1155/2022/1562544
be applied. The model's forecasting skills can be improved by [9]. Driven, D., Gas, N., Price, S., Using, M., & Learning,
enlarging the dataset to incorporate a wider range of variables M. (2019). Data Driven Natural Gas Spot Price
impacting natural gas demand, such as building attributes, Prediction Models Using Machine Learning Methods.
energy saving measures, and customer behavior. To [10]. Faruk, O., Cayir, B., Tatoglu, E., Gokcin, P., & Zaim, S.
summarize, there are many benefits to forecasting residential (2019). Using machine learning tools for forecasting
natural gas consumption utilizing a hybrid ensemble natural gas consumption in the province of Istanbul ☆.
regression machine learning approach, including improved Energy Economics, 80, 937–949.
prediction accuracy, robustness, flexibility, and practical https://doi.org/10.1016/j.eneco.2019.03.006
application. The residential natural gas market could benefit [11]. Gao, F., & Shao, X. (2021). Forecasting annual natural
from this strategy by optimizing resource allocation, gas consumption via the application of a novel hybrid
supporting energy management techniques, and informing model. 21411–21424.
policy decisions. The subject can advance and goals for [12]. J. P. D., García-martín, E., Faviola, C., Riley, G., &
sustainable energy management can be attained with more Grahn, H. (2019). Estimation of energy consumption in
study and application of the hybrid ensemble approach. machine learning. Journal of Parallel and Distributed
Computing, 134, 75–88.
https://doi.org/10.1016/j.jpdc.2019.07.007

IJISRT23SEP126 www.ijisrt.com 164

Volume 8, Issue 9, September – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
[13]. Laib, O., Tarek, M., & Mihaylova, L. (2019). Toward ef
fi cient energy systems based on natural gas
consumption prediction with LSTM Recurrent Neural
Networks. Energy, 177, 530–542.
https://doi.org/10.1016/j.energy.2019.04.075
[14]. Li, J., Guo, Y., Zhang, X., & Fu, Z. (2021). Using
Hybrid Machine Learning Methods to Predict and
Improve the Energy Consumption Efficiency in Oil and
Gas Fields. Mobile Information Systems, 2021.
https://doi.org/10.1155/2021/5729630
[15]. Liu, X., Hou, G., & Yang, L. (2022). Dynamic
Combined Optimal Scheduling of Electric Energy and
Natural Gas Energy Consumption in Data Center.
Discrete Dynamics in Nature and Society, 2022.
https://doi.org/10.1155/2022/3917170
[16]. Mosavi, A., & Bahmani, A. (2019). Energy
Consumption Prediction Using Machine Learning ; A
Review Energy consumption prediction using machine
learning ; a review. March.
[17]. Nabavi, S. A., Aslani, A., Zaidan, M. A., Zandi, M.,
Mohammadi, S., & Motlagh, N. H. (n.d.). Machine
Learning Modeling for Energy Consumption of
Residential and Commercial Sectors. 1–22.
https://doi.org/10.3390/en13195171
[18]. Qiao, W., Moayedi, H., & Original, W.-. (2020). Energy
& Buildings Nature-inspired hybrid techniques of IWO
, DA , ES , GA , and ICA , validated through a k-fold
validation process predicting monthly natural gas
consumption. 217.
https://doi.org/10.1016/j.enbuild.2020.110023
[19]. Rajaraman, S., Jaeger, S., and Antani, S. K. J. P. (2019).
"Performance evaluation of deep neural ensembles
toward malaria parasite detection in thin-blood smear
images," vol. 7, p. e6977.
[20]. Šebalj, D. (2019). Analysis of Methods and Techniques
for Prediction of Natural Gas Consumption : A
Literature Review Analysis of Methods and Techniques
for Prediction of Natural Gas Consumption : A
Literature Review. June.
https://doi.org/10.31341/jios.43.1.6
[21]. Shapi, M. K. M., Ramli, N. A., & Awalin, L. J. (2020).
Journa l Pre of Developments in the Built Environment,
100037. https://doi.org/10.1016/j.dibe.2020.100037
[22]. Then, D., Spaltho, C., Bauer, J., Kneiske, T. M., &
Braun, M. (2020). Impact of Naatural Gas Distribution
Network Structure and Operator Strategies on Grid
Economy in Face of Decreasing Demand.
[23]. Yin, K. S., & Htay, S. S. (2011). Prediction of Natural
Gas Final Consumption using Artificial Neural
Networks. 224–229.

IJISRT23SEP126 www.ijisrt.com 165