Energy Reports 7 (2021) 5467–5474

Contents lists available at ScienceDirect

Energy Reports
journal homepage: www.elsevier.com/locate/egyr

Review article

A comprehensive review: Machine learning and its application in

integrated power system
∗
Aanand Kumbhar, Pravin G. Dhawale , Shobha Kumbhar, Uday Patil, Pravin Magdum
School of Electrical Engineering, Sanjay Ghodawat University, Kolhapur, India

article info a b s t r a c t

Article history: A comprehensive review about machine learning application in power system especially in smart grid,
Received 31 May 2021 renewable energy sector etc. is summarized in this paper. In the power sector, the power consumption
Received in revised form 31 July 2021 is increased day by day very tremendously. So, it is very important that we have to generate the
Accepted 19 August 2021
more power without disturbing the environment and whatever the generated power must be utilize
Available online 7 September 2021
effectively with minimum losses and higher efficiency. This will be possible with effective way of using
Keywords: the modern technology like machine learning (ML), artificial intelligence etc. This paper also describes
Smart grid the different types of machine learning techniques with diagram which will be very useful for many
Renewable energy researchers who want to understand the basic fundamentals of machine learnings.
Power system © 2021 Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
Machine learning (ML) (http://creativecommons.org/licenses/by-nc-nd/4.0/).
Environment
Demand response
Dynamic thermal rating
Reliability
Advanced metring infrastructure

Contents

1. Introduction..................................................................................................................................................................................................................... 5467
2. An introduction to machine learning........................................................................................................................................................................... 5468
2.1. Learnings that are supervised .......................................................................................................................................................................... 5469
2.2. Learning without supervision........................................................................................................................................................................... 5469
2.3. Learning that is reinforced ............................................................................................................................................................................... 5469
2.4. Outfit strategies.................................................................................................................................................................................................. 5469
3. Application of machine learning in the integrated power system .......................................................................................................................... 5469
3.1. Generation forecasting in renewable energy sector ...................................................................................................................................... 5471
3.2. Application in smart grid of machine learnings ............................................................................................................................................ 5471
3.3. Techniques for detecting and diagnosing faults using machine learning................................................................................................... 5471
3.4. (ML) Machine Learning structure/framework in an integrated power system.......................................................................................... 5473
4. Conclusion ....................................................................................................................................................................................................................... 5473
Declaration of competing interest................................................................................................................................................................................ 5474
Acknowledgements ........................................................................................................................................................................................................ 5474
References ....................................................................................................................................................................................................................... 5474

1. Introduction friendly power sources, the internet security, request side ad-
ministration, and dynamic of framework arranging and activ-
The current force frameworks are going through a quick ity (Ibrahim et al., 2020). The satisfaction of cutting-edge func-
progress towards their more dynamic, adaptable, and canny part- tionalities in the brilliant framework solidly depends on the
ner keen network, which achieves gigantic difficulties in nu- basic data and correspondence foundation, and the productive
merous spaces, e.g., mix of different conveyed environmentally treatment of a monstrous measure of information produced from
different origin such as phasor estimation units, Smart metres and
∗ Corresponding author. many types of sensors (Ibrahim et al., 2020; Gigoni et al., 2018).
E-mail addresses: [email protected] This paper, far-reaching study more than two hundred late
(A. Kumbhar), [email protected] (S. Kumbhar). distributions are led for survey cutting-edge approaches and

https://doi.org/10.1016/j.egyr.2021.08.133
2352-4847/© 2021 Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
A. Kumbhar, P.G. Dhawale, S. Kumbhar et al. Energy Reports 7 (2021) 5467–5474

Nomenclature RBFNN Radial basis functions neural network

RBM Restricted Boltzmann machine
AC Alternating current
RL Reinforcement learning
ACE Average coverage error
RMSE Root mean squared error
AMI Advanced metring infrastructure
RNN Recurrent neural network
ANN Artificial neural network
SAE Stacked auto encoder
BPNN Back propagation neural network
SVM Support vector machine
CNN Convolutional neural network
SVR Support vector regression
CRBM Conditional restricted Boltzmann ma-
chine
DBM Deep Boltzmann Machine
DBN Deep belief network of energy frameworks, ultra-lightweight Machine Learning based
DC Direct current arrangements, etc. Additionally, the future viewpoints of using
DG Distribution generation progressed processing and correspondence innovations, e.g., edge
DR Demand response figuring, pervasive web of things and 5G remote organizations, in
DSM Demand side management the savvy framework are additionally featured. Apparently, this is
ELM Extreme learning machine the primary audit of AI driven arrangements covering practically
EV Electrical vehicle all the brilliant framework application spaces. ML will be one
FCRBM Factored conditional restricted Boltz- of the most significant tool in electrical system frameworks,
mann machine this investigation will give a starter establishment for additional
investigation and also advancement of the concern information
FDI False data injection
including experiences (Yang et al., 2021).
FFNN Feed forward neural network
The history of the electric stockpile dates back approximately
GAN Generative adversarial network two hundred years. The main public utilization of power traces all
GBM Gradient boosting machine the way back to 1881 at Godalming, Surrey, United Kingdom. The
GBRT Gradient boosting regression tree capacity of AC to be changed into different level of voltage was a
GRU Gated recurrent unit critical component in its control over DC frameworks, permitting
HMM Hidden Markov model AC to convey at high voltages with lower misfortunes. Electri-
IoT Internet of things cal force frameworks have progressed from traditional energy
KNN K-nearest neighbour frameworks to a more productive cutting-edge brilliant network
framework over the previous decade. A schematic outline of the
LASSO Least absolute shrinkage and selection
regular force lattice is appeared in Fig. 1(a). The regular power
operator
frameworks depend on a couple of unified and huge force age
LDA Linear discriminant analysis
sources, essentially hydropower or petroleum derivative-based
LSM Liquid state machine force age frameworks, possessing a massive transmission net-
LSTM Long short term memory work that delivers electricity to customers using a conveyance
MAE Mean absolute error system (Ferrag et al., 2020). There has been a surge in inter-
MAPE Mean absolute percentage error est and trends in employing machine learning-based techniques
MARS Multivariate adaptive regression splines to resolve lingering difficulties in smart grids in recent years.
MDA Multiple discriminant analysis We conducted a comprehensive evaluation of suggestions from
ML Machine learning academia and engineering practises for applications using ma-
ML-KNN Multi-label k-nearest neighbour chine learning (Gigoni et al., 2018). The literature presents and
discusses a complete picture of a number of issues and current
MLP Multilayer perception
achievements in machine learning applications.
MSE Mean squared error
NIALM Non-intrusive appliance load monitor- 2. An introduction to machine learning
ing
NILM Non-intrusive load monitoring Machine Learning is an information investigation innovation
NRMSE Normalized root mean squared error that uses a learning interaction to instruct PCs to do exercises
PICP Prediction interval nominal confidence like people and creatures. Machine Learnings calculations utilize
PSO Particle swarm optimization PC strategies to ‘‘learn’’ data straightforwardly from informa-
PV Photovoltaic tion as opposed to relying upon set conditions, and they can
QDA Quadratic discriminant analysis change their exhibition as information turns out to be more
plentiful. Machine Learning requires careful arranging and pro-
gramming of calculations to accomplish different usefulness, like
characterization, grouping, and relapse (Yang and Wu, 2021).
proposition of ML strategies and also examine the pattern have Machine Learnings and profound learning as a specific field have
big scope of keen framework application areas This examination made expected commitments to an assortment of exploration
illustrates the growing concern and fast accelerated improvement and designing fields in the earlier decade, including information
while using machine learning procedures to effectively address mining, clinical imaging, and advanced mechanics. One of the
the shrewd framework’s particular issues from several perspec- primary stages towards a savvy lattice is the coordination of
tives (Ibrahim et al., 2020). It is additionally uncovered that a refined data and correspondence advances, especially IoT, into
few issues actually stay open and worth further examination the force matrix foundation. Since the capacity of IoT gadgets
endeavours, like the elite information preparing and examination to impart and pass on information to different gadgets in a
for astute dynamic in enormous scope convoluted multi-source more far and wide way is their fundamental ability, a lot of
5468
A. Kumbhar, P.G. Dhawale, S. Kumbhar et al. Energy Reports 7 (2021) 5467–5474

Fig. 1. Overview of machine learning (Ibrahim et al., 2020).

information is made accessible at control focuses. Because of 2.3. Learning that is reinforced
the expanded framework mindfulness and information accessi-
bility, Machine Learnings based arrangements and devices are This is learning worldview in which a specialist communi-
needed to perform successful information handling and inves- cates with its current circumstance and changes its behaviour
tigation to improve framework operational administration and in response to stimuli. Unlike supervised learning, RL doesn’t
dynamic (Farhoumandi et al., 2021). need labelled yield/input pairings and alternatively rewards or
A keen framework is a convoluted digital actual framework punishes the agent based on its behaviour in the environment As
that joins a huge electrical force network with a fundamental data a result, RL allows the expert to choose techniques that cannot
and correspondence foundation. Ongoing operational condition be performed by administered or solo study this motivate to the
checking and compelling information examination can exten- reinforced learning to progress in the same path that people,
sively improve framework the board in an assortment of ways, creatures act in the real world (Shi et al., 2020).
for instance. Evaluation of framework resources, operational se-
curity confirmation, framework inconsistency identification, is- 2.4. Outfit strategies
sue determination, load interest and force age control, etc. AI
approaches have earned a ton of consideration over the most This method which is utilize a few calculations related to ma-
recent couple of years because of the acknowledgement that chine learning to enhance the exhibition looked at utilization of a
conventional registering strategies can presently don’t adapt to solitary calculation. Outfit knowledge builds a groups of theories
the enormous measure of information produced by the savvy produced through a few base students to be utilized together to
framework. Much exploration has gone into analysing various tackle a solitary issue, which gives preferred generalizability over
AI and profound learning-based strategies to meet issues in dif- the individual base students
ferent savvy network application areas (Ibrahim et al., 2020). AI
approaches are information driven in that they utilize existing
3. Application of machine learning in the integrated power
information to do different errands, and they can be generally
system
gathered into four gatherings, as follows:
The vast array of electrical components that are heteroge-
2.1. Learnings that are supervised
neous and spatially dispersed parts in customary force frame-
This is a sort of machine learning in which using a labelled col- works prompts basic changes of uninvolved and unidirectional
lection of input/output pairs, the goal is to take in a planning from force frameworks into dynamic, conveyed, and trans active force
contributions to yields and large number of training instances in frameworks. There are generous measures of information which
a training set. are being produced and traded among various substances in these
cutting-edge power frameworks (Haq et al., 2020).
2.2. Learning without supervision The productive activity and the executives of such force frame-
works rely upon quick and solid correspondence and computa-
This is a machine learning paradigm in which an algorithm tional organizations for successfully preparing estimations and
is trained on unlabelled and unclassified data, allowing the com- moving control signs to important electrical segments for speedy
puter to group data in light of likeness or contrast. Unaided activities. Fig. 2 indicate descriptive of machine learning structure
learning calculations can accomplish more intricate handling un- in energy frameworks including four phases which are infor-
dertakings than regulated learning calculations, however they mation assortment, highlight eradication, grouping, information
are additionally more erratic (Shi et al., 2020). When contrasted motivated arrangement yields (Yin et al., 2020).
with other characteristic learning approaches, widely utilized The critical assignment for AI utilization in electrical power
exploratory data analysis (EDA) is a sort of data analysis that uses or energy frameworks is to understand speculation of the system
cluster analysis, an unsupervised learning method, to hunt for model which is based on gathered information. Along these lines,
hidden patterns or groups of data. the AI methods can give quick and exact information driven
5469
A. Kumbhar, P.G. Dhawale, S. Kumbhar et al. Energy Reports 7 (2021) 5467–5474

Fig. 2. Application of machine learning (Ibrahim et al., 2020).

Fig. 3. Outline of how to use ML in fault analysis (Yang et al., 2021).

answers for a wide scope of force framework applications, zero- only a couple of the machine learning applications in the keen
ing in on anticipating and control, booking and power markets, framework. Fig. 3 portrays the scope of AI applications in the bril-
client investment and disseminated request reactions, deficiency
liant lattice, from summed up impression of hidden frameworks
recognition and assurance, and network protection (Yin et al.,
to clever and versatile dynamic, lastly to constant or close to
2020). Expectations of burden and estimating, course disappoint-
ment forecast, power age and control, deficiency discovery and ongoing tasks with regards to keen networks (Paturi and Cheruku,
conclusion DSM, and recognition of the internet dangers are 2020).
5470
A. Kumbhar, P.G. Dhawale, S. Kumbhar et al. Energy Reports 7 (2021) 5467–5474

Fig. 4. In power systems, a machine learning framework that is illustrative (Farhoumandi et al., 2021).

3.1. Generation forecasting in renewable energy sector (2) ML based digital assault identification techniques have
shown promising execution because of their adaptability
Present day power frameworks reliably coordinate variable in- towards versatility and their astounding characterization
exhaustible energy age assets, e.g., wind power and sun-oriented exactness. Moreover, the current examination in this space
PV age to adapt to an unnatural weather change concern. In any shows the strength of managed learning techniques in
case, the age patterns of inexhaustible age assets are profoundly assault location (Haq et al., 2020).
delicate to environment also, occasional variables (Yang et al., (3) The vast majority of the writing tending to energy and
2021). At that point, the multiplication of these variable energy monetary dispatch essentially use multi-specialist frame-
age assets has presented numerous difficulties to the planning work hypothesis to resolve the issue of dispatch in brilliant
and activity of force frameworks (Gigoni et al., 2018). The current lattices which requires a precise numerical model of the
model-based strategies are inadequately fit for accomplishing expense work, though, as of late a couple of examination
wanted exhibitions for sustainable age figures. Accordingly, se- works utilize AI, for example, RL calculations to deal with
the issue of energy and financial eviction issue.
cure and dependable activities of force frameworks call for more
precise determining procedures as we send variable environmen-
3.3. Techniques for detecting and diagnosing faults using machine
tally friendly power age in exceptionally huge scopes (Zhao and
learning
You, 2021). Sun powered PV is perhaps the most conceivably
available and broadly conveyed sustainable power assets on the Contrasted and the models with a model-driven foundation,
planet. Furthermore, the cost of installation and operation of information besetting models embracing machine learning pro-
solar-powered PV units has decreased dramatically over the last cedures is more similar to a ‘‘black-box’’, and nothing express
decade, making the sun-oriented PV age a fierce competitor with scientific depiction of characterizing connection between input
conventional power ages, particularly in topographical locations also, yield (Haq et al., 2020). It depends on the deficiency high-
with abundant sunshine (Akhter et al., 2019) (see Table 1). light extricated from enormous arrangement of exploratory infor-
mation informational collection and to frame a capacity relating
3.2. Application in smart grid of machine learnings inputs and yields (Yang et al., 2021). An overall system de-
lineating the method of the ML based shortcoming analysis is
There are number of applications related of machine learning appeared in Fig. 3, where it comprises 4 stages: information
in the Smart Grid. Table 2 will give the detail information about assortment, FE (feature extraction), ML models, outputs as shown
the ML application in the Smart Grid and also techniques which in the below figure. Generally in information assortment, observ-
are used are as follows. ing gadgets disseminated inside the force framework continually
gather information, like shaking , audial discharge and also pro-
(1) ML based NILM strategies have likewise dominated with tection including current information and feeder position and
the utilization of profound learning and progressed multi- status (Nikolaidis and Chatzis, 2021). In these amorphous system,
mark grouping strategies like ML-KNN and SVM, where the ranch-organized information gathered in terms of picture, video,
requirement for earlier element extraction is decreased be- text etc. Likewise, in view of the sorts of gathered informa-
cause of such techniques’ programmed highlight extraction tion and application areas, scientists utilize diverse ML methods
capacity, which was one of the difficulties for conventional to understand the component model learning and extraction
AI methods (Ibrahim et al., 2020). particular (Farhoumandi et al., 2021).
5471
A. Kumbhar, P.G. Dhawale, S. Kumbhar et al. Energy Reports 7 (2021) 5467–5474

Table 1
Literature on machine learning.
Sr. No. Author name Year of Main findings
publication
1. Jiahashen Teh 2021 In this paper authors proposed a technique that ideally sent BESSs and decided their ability in a
et al. two-section system to limit sun based energy abridgement, by considering network geography and
force stream imperatives. Results show that proposed technique is more proficient than the regular
arrangement procedure in that it figures out how to store more excess sun based force i.e. solar
power.
2. Khoo et al. (2020). 2020 In this paper, Integration of wind energy and demand response is done with existing system for
optimum generation, reliability, and cost and for low carbon emission. Integration of wind energy
with other is done with satisfying the other objectives (Khoo et al., 2020).
Jimada-Ojuolape 2020 This work thoroughly surveys examines that go past segment based unwavering quality evaluation
and Teh (2020b) and records for the effect of ICT combinations on framework wide dependability while expressly
considering the impacts of glitches of the digital structure. Also, presented quantitative and
subjective data about the effect of ICT sent with different smart structure advancements and
applications on the unwavering quality of current force frameworks (Jimada-Ojuolape and Teh,
2020b).
3. Jimada-Ojuolape 2020 Jiahashen Teh et al. done the detail review on the Impact of integration of information and
and Teh (2020a) communication technology on power system reliability. In this review paper authors investigate the
impact of ICTs on power system in term of reliability (Jimada-Ojuolape and Teh, 2020a).
4. Yang et al. (2021). 2020 For day-ahead anticipating of PV power creation, a joined profound conviction organization and dim
hypothesis based information pre-preparing model is proposed, in which significant level reflections
in chronicled PV yield information are made by utilizing progressive structures, which would
diminish the required dataset.
5. Eom et al. (2020). 2020 A group model, which joins AI strategies and conventional time-arrangement expectations, is
proposed for highlight particular long haul forecasts of PV power age in which the fluctuations of
climate, hardware accessibility, monetary activity, and government approaches are completely
thought of
6. Metwaly and Teh 2020 In this paper, authors proposed two methods first one is probabilistic new method to find out the
(2020b). various combination of battery energy storage system power rating, energy capacities and check the
impact on the transmission network. Peak demand is fulfilled by BESS. Second proposed method
focus to check the effect of deploying battery energy storage system with demand response and
dynamic thermal ratings.
7. Metwaly and Teh 2020 In this paper, authors proposed a multi objective framework for enhanced wind penetration and
(2020a) reliability by optimizing the network ageing and battery storage (Metwaly and Teh, 2020a).
8. Mohamad et al. 2019 In this paper, authors divide the proposed method in the two parts first one is focused on the
(2019). optimization of penetration of solar and wind energy and second one is focused on the optimization
of energy storage capacity. The optimum solution is find out (Mohamad et al., 2019).
9. Teh and Lai 2019 This paper present the reliability modelling of two technologies namely Battery energy storage
(2019a) system and dynamic thermal rating. This paper focused on the reliability effect of the dynamic
thermal rating and battery energy storage system on wind integrated power system network (Teh
and Lai, 2019a).
10. Akhter et al. 2019 Give an intensive evaluation of machine learning and metaheuristic techniques to estimating
(2019). sun-based photo voltaic age, featuring the advantages and disadvantages of every strategy as far as
anticipating skylines and info boundaries.
11. Teh and Lai 2019 In this paper, new model for exploring the unwavering quality effects of the DTR framework on
(2019b) power lattices considering the remote correspondence network is proposed. The powerlessness of
the DTR framework to perform, either because of its inaccessibility or correspondence
disappointment, is taken as a component of line appraisals that influence the bandwidth and
unwavering quality of force matrices. At the point when the DTR framework administration is
inaccessible, the line rating is either returned to the static warm evaluating or assessed by means of
the multi-straight relapse (MLR) model dependent on data from adjoining areas. The correlation of
these two models on the IEEE 24-transport unwavering quality test framework uncovers that the
MLR model is more useful in relieving DTR and correspondence frameworks disappointments. This
paper additionally investigates different correspondence framework disappointment situations and
the unwavering quality effects of excess plans. The discoveries uncover that the correspondence
framework with moderate dependability levels yield the most advantages when the excess number
increments from 1 to 4. Generally speaking, the correspondence framework altogether influences
the accessibility of the DTR framework (Teh and Lai, 2019b).
11. Gigoni et al. 2018 Throughout a year, specialists evaluate and contrast many existing information-based PV power
(2018). creation determining approaches, including k-closest neighbours, neural organizations, helping
vector relapse, quantile irregular woods, on thirty-two photo voltaic plants of different shapes and
size innovations. Force standardization (for example voltage glimmer), guideline cost could all be
influenced by transient variances in photo voltaic yields, on the grounds that sending guideline
stores to make up for PV motions is typically costly.
12. Golestaneh et al. 2016 A limit learning machine is used to give a quantile relapse system for making very transient
(2016). nonparametric thickness expectations of PV power yields, dodging obliging constraints on the
prescient disseminations.
13. Yin et al. (2020). 2017 A special factual methodology coordinating limit learning machine and quantile relapse is
recommended to produce quality expectation timespans power yields, which has fundamentally
high precision and handling proficiency gauge.

5472
A. Kumbhar, P.G. Dhawale, S. Kumbhar et al. Energy Reports 7 (2021) 5467–5474

Table 2
Application of ML in smart grid (Ibrahim et al., 2020).
Technique Key benefits Drawbacks Application in smart grid Algorithm
Simple architecture and fast Complex calculations for DSM, Cyberspace Security ELM, MLP, Lavenberg
Artificial neural network convergence rate. finding optimal number of Marquardt algorithm
hidden layer neurons
Wide application spectrum. Problems of under fitting Forecasting BPNN
and overfitting
High ability of addressing Computationally expensive Fault and Failure Analysis RBFNN
complex non-linear training due to high number
relationships. of hidden layer neurons
Automatic feature Outputs are usually Forecasting SAE, CNN
Deep learning extraction and ability of comprehended using
learning natural variation in classifiers resulting in
data automatically. increased computational
cost
Flexible tool for application Most deep learning DSM, Energy and Economic DBN
in new problems algorithms require Dispatch
dedicated computing
resources
Wide application spectrum. Most algorithms need large Cyberspace Security DBM
amounts of training data for
efficient performance
Inherently non-linear Need large amount of data Energy and Economic LSTM
RNN models that learn Dispatch
non-linearity from data
Can perform exceptionally Limited choice of training DSM LSM
well with dynamic time algorithms
series
Can be used with Limited robustness of the Forecasting GRU
convolutional layers common activation
functions
Does not require any Requires large memory to Fault and Failure Analysis, Q-learning
RL
training data record the states Power System Operations
The model can eradicate The curse of dimensionality Energy and Economic Deep Q-learning
errors during training Dispatch, Cyberspace
Security
Can achieve high Memory intensive Forecasting, NILM, Fault and SVM, KNN
Classification
classification accuracy Failure Analysis
Flexible kernel function Cannot perform well with Cyberspace Security, Decision Trees, C4.5
selection problems having many Planning and Operation
training examples
Better prediction accuracy Reduced interpretability due Forecasting Random forest
Ensemble methods than a single machine to complexity resulting
learning mode from using various models
Improved performance by Not suitable for real-time Fault and Failure Analysis Adaboost, Bagging, Boosting
combined decisions from applications due to high
different models computational time
Can capture the linear as Selection of appropriate Cyberspace Security GBRT
well as non-linear algorithms to create an
relationships in data efficient ensemble model is
a challenging task
Easy to understand and Algorithms like linear DSM, Forecasting Linear Regression, Logistic
Regression
implement regression and logistic Regression
regression cannot work well
with non-linear data
Limited number of Overfitting problems Power System Operations SVR
hyper-parameters
Clustering Fast, simple, and flexible Requires pre-processing NILM, Forecasting, k-means Clustering,
Cyberspace Security C-means, Fuzzy C-means

3.4. (ML) Machine Learning structure/framework in an integrated learning in the power sector, RE sector. This paper also gives
power system detailed information about the types of machine learning tech-
niques. This paper definitely will help to those people who want
See Fig. 4. to do research work in the field of artificial intelligence, data
analytics and machine learning future application. We summarize
4. Conclusion the uses of ML to blame finding in the force framework, which is
the essential advance of the force framework security and control.
This paper given review of application of ML in integrated We initially sum up the traditional ML based methodologies, like
power system. There are number of applications of machine SVM, ANN and other different methodologies, those are typically
5473
A. Kumbhar, P.G. Dhawale, S. Kumbhar et al. Energy Reports 7 (2021) 5467–5474

directed or semi-regulated learning strategies. Their applications, Jimada-Ojuolape, B., Teh, J., 2020a. Impact of the integration of information
benefits and inconveniences are talked about in each particular and communication technology on power system reliability: A review. IEEE
Access 8, 24600–24615. http://dx.doi.org/10.1109/ACCESS.2020.2970598.
subsection. This audit is required to methodically introduce the
Jimada-Ojuolape, B., Teh, J., 2020b. Surveys on the reliability impacts of power
improvement of ML in applications in deficiency finding of force system cyber–physical layers. Sustain. Cities Soc. 62 (July), 102384. http:
framework and gives a guide to future utilizations. //dx.doi.org/10.1016/j.scs.2020.102384.
Khoo, W.C., Teh, J., Lai, C.M., 2020. Integration of wind and demand response
Declaration of competing interest for optimum generation reliability, cost and carbon emission. IEEE Access 8,
183606–183618. http://dx.doi.org/10.1109/ACCESS.2020.3029273.
Metwaly, M.K., Teh, J., 2020a. Optimum network ageing and battery sizing for
The authors declare that they have no known competing finan- improved wind penetration and reliability. IEEE Access 8, 118603–118611.
cial interests or personal relationships that could have appeared http://dx.doi.org/10.1109/ACCESS.2020.3005676.
to influence the work reported in this paper. Metwaly, M.K., Teh, J., 2020b. Probabilistic peak demand matching by battery
energy storage alongside dynamic thermal ratings and demand response for
enhanced network reliability. IEEE Access 8, 181547–181559. http://dx.doi.
Acknowledgements org/10.1109/ACCESS.2020.3024846.
Mohamad, F., Teh, J., Abunima, H., 2019. Multi-objective optimization of
We are very thankful to School of Technology, Department solar/wind penetration in power generation systems. IEEE Access 7,
of Electrical Engineering, Sanjay Ghodawat University, Kolha- 169094–169106. http://dx.doi.org/10.1109/ACCESS.2019.2955112.
Nikolaidis, P., Chatzis, S., 2021. Gaussian process-based Bayesian optimization
pur for providing necessary research facility. We also thanks to for data-driven unit commitment. Int. J. Electr. Power Energy Syst. 130
Head of Department of Electrical Engineering Dr. R. M. Linus for (February), 106930. http://dx.doi.org/10.1016/j.ijepes.2021.106930.
supporting this work. Paturi, U.M.R., Cheruku, S., 2020. Application and performance of machine
learning techniques in manufacturing sector from the past two decades: A
review. Mater. Today Proc. 38 (xxxx), 2392–2401. http://dx.doi.org/10.1016/
References
j.matpr.2020.07.209.
Shi, Z., et al., 2020. Artificial intelligence techniques for stability analysis
Akhter, M.N., Mekhilef, S., Mokhlis, H., Shah, N.M., 2019. Review on forecasting of and control in smart grids: Methodologies, applications, challenges and
photovoltaic power generation based on machine learning and metaheuristic future directions. Appl. Energy 278 (July), 115733. http://dx.doi.org/10.1016/
techniques. IET Renew. Power Gener. 13 (7), 1009–1023. http://dx.doi.org/ j.apenergy.2020.115733.
10.1049/iet-rpg.2018.5649. Teh, J., Lai, C.M., 2019a. Reliability impacts of the dynamic thermal rating
Eom, H., Son, Y., Choi, S., 2020. Feature-selective ensemble learning-based long- and battery energy storage systems on wind-integrated power networks.
term regional PV generation forecasting. IEEE Access 8, 54620–54630. http: Sustain. Energy Grids Netw. 20, 100268. http://dx.doi.org/10.1016/j.segan.
//dx.doi.org/10.1109/ACCESS.2020.2981819. 2019.100268.
Farhoumandi, M., Zhou, Q., Shahidehpour, M., 2021. A review of machine Teh, J., Lai, C.M., 2019b. Reliability impacts of the dynamic thermal rating
learning applications in IoT-integrated modern power systems. Electr. J. 34 system on smart grids considering wireless communications. IEEE Access
(1), 106879. http://dx.doi.org/10.1016/j.tej.2020.106879. 7, 41625–41635. http://dx.doi.org/10.1109/ACCESS.2019.2907980.
Ferrag, M.A., Babaghayou, M., Yazici, M.A., 2020. Cyber security for fog-based Yang, H., Liu, X., Zhang, D., Chen, T., Li, C., Huang, W., 2021. Machine learning
smart grid SCADA systems: Solutions and challenges. J. Inf. Secur. Appl. 52, for power system protection and control. Electr. J. 34 (1), 106881. http:
http://dx.doi.org/10.1016/j.jisa.2020.102500. //dx.doi.org/10.1016/j.tej.2020.106881.
Gigoni, L., et al., 2018. Day-ahead hourly forecasting of power generation from Yang, Y., Wu, L., 2021. Machine learning approaches to the unit commitment
photovoltaic plants. IEEE Trans. Sustain. Energy 9 (2), 831–842. http://dx.doi. problem: Current trends, emerging challenges, and new strategies. Electr. J.
org/10.1109/TSTE.2017.2762435. 34 (1), 106889. http://dx.doi.org/10.1016/j.tej.2020.106889.
Golestaneh, F., Pinson, P., Gooi, H.B., 2016. Very short-term nonparametric Yin, L., et al., 2020. A review of machine learning for new generation smart
probabilistic forecasting of renewable energy generation - With application dispatch in power systems. Eng. Appl. Artif. Intell. 88 (2019), 103372. http:
to solar energy. IEEE Trans. Power Syst. 31 (5), 3850–3863. http://dx.doi.org/ //dx.doi.org/10.1016/j.engappai.2019.103372.
10.1109/TPWRS.2015.2502423. Zhao, N., You, F., 2021. New York State’s 100% renewable electricity transition
Haq, E.U., Lyu, X., Jia, Y., Hua, M., Ahmad, F., 2020. Forecasting household electric planning under uncertainty using a data-driven multistage adaptive robust
appliances consumption and peak demand based on hybrid machine learning optimization approach with machine-learning. Adv. Appl. Energy 2 (January),
approach. Energy Rep. 6, 1099–1105. http://dx.doi.org/10.1016/j.egyr.2020. 100019. http://dx.doi.org/10.1016/j.adapen.2021.100019.
11.071.
Ibrahim, M.S., Dong, W., Yang, Q., 2020. Machine learning driven smart electric
power systems: Current trends and new perspectives. Appl. Energy 272
(February), 115237. http://dx.doi.org/10.1016/j.apenergy.2020.115237.

5474