Next Article in Journal
Resilience of Natural Gas Pipeline System: A Review and Outlook
Previous Article in Journal
Analysis and Comparative Study of Signalized and Unsignalized Intersection Operations and Energy-Emission Characteristics Based on Real Vehicle Data
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:

A Review on Sustainable Energy Sources Using Machine Learning and Deep Learning Models

Ashok Bhansali
Namala Narasimhulu
Rocío Pérez de Prado
Parameshachari Bidare Divakarachari
4 and
Dayanand Lal Narayan
Department of Computer Engineering and Applications, GLA University, Mathura 281406, India
Department of Electrical and Electronics Engineering, Srinivasa Ramanujan Institute of Technology (Autonomous), Ananthapuramu 515701, India
Telecommunication Engineering Department, University of Jaén, 23700 Jaén, Spain
Department of Electronics and Communication Engineering, Nitte Meenakshi Institute of Technology, Bengaluru 560064, India
Department of Computer Science Engineering, GITAM School of Technology, GITAM University, Bengaluru 561203, India
Author to whom correspondence should be addressed.
Current address: Department of Electrical and Electronics Engineering, GATES Institute of Technology (Autonomous), Ananthapuramu 515401, India.
Energies 2023, 16(17), 6236;
Submission received: 9 July 2023 / Revised: 11 August 2023 / Accepted: 22 August 2023 / Published: 28 August 2023
(This article belongs to the Section F: Electrical Engineering)


Today, methodologies based on learning models are utilized to generate precise conversion techniques for renewable sources. The methods based on Computational Intelligence (CI) are considered an effective way to generate renewable instruments. The energy-related complexities of developing such methods are dependent on the vastness of the data sets and number of parameters needed to be covered, both of which need to be carefully examined. The most recent and significant researchers in the field of learning-based approaches for renewable challenges are addressed in this article. There are several different Deep Learning (DL) and Machine Learning (ML) approaches that are utilized in solar, wind, hydro, and tidal energy sources. A new taxonomy is formed in the process of evaluating the effectiveness of the strategies that are described in the literature. This survey evaluates the advantages and the drawbacks of the existing methodologies and helps to find an effective approach to overcome the issues in the existing methods. In this study, various methods based on energy conversion systems in renewable source of energies like solar, wind, hydro power, and tidal energies are evaluated using ML and DL approaches.

1. Introduction

In recent days, the electricity generation system based on renewable energies has been used in various advanced technologies and in economically emerging nations [1,2]. The application of renewable energies is increasing due to major reasons such as energy instability, change in environmental conditions, and global warming [3]. Moreover, by considering factors such as energy objectivity, reduction in green gas, and improvement in the quality of air, advancements in renewable systems are incorporated using various techniques based on Artificial Intelligence (AI) [4,5]. Conventional or non-renewable source of energies like coal, oil, and natural gas are referred to as fossil fuels, which are used in the process of power generation [6]. The aforementioned non-renewable sources of energy are known for their high energy density, thus requiring less time than renewable sources to create electric energy. However, renewable energy sources are emission-free and environmentally friendly as they do not create any negative impact on the environment [7]. Moreover, they can be regenerated to fulfil the day-to-day requirement of the current generation without affecting the survivability of future generations. The usage of energy at the global level is estimated to increase by 56% in the year 2040 [8]. To minimize carbon emissions and fight against global warming, the usage of non-renewable energies must be minimized, and usage of renewable energies must be maximized. The usage of new energy storage systems is enhanced due to the improved damage rate created by nuclear and fossil power sources [9]. Renewable energy sources such as wind and solar are among the common sustainable energy sources, widely utilized all over the world, even for small-household applications. The rise of the industrial revolution led to an increase in the amount of greenhouse gases and became a major cause for various kinds of pollution [10,11]. Power plants using natural gas emit 50% less CO 2 than power plants that use fossil fuels. Thus, renewable energy is utilized to diminish the effect of harmful emission by factories and industries, which in turn helps in the betterment of the environment [12,13,14].
Machine learning and deep learning approaches play a significant role in power prediction, energy conversion, and forecasting sustainable energy sources. The major reason for power predictions is to offer a cost-effective approach for the incorporation of renewable power resources to electrical networks. The forecasting of sustainable energy sources is growing day by day and is frequently utilized in various applications related to power grids which help both small-scale as well as large-scale producers [15]. With respect to wind energy, the quality of offshore wind is superior to the quality of onshore wind. Moreover, there is a huge increase in the installation of wind turbines in offshore locations, which is expected to reach its maximum in the following decades. Based on the wind power output, the prediction of accurate speed of the wind is considered as the significant factor in the majority of applications related to wind energy [16,17,18,19,20]. Likewise, energy prediction in hydropower is based on the electrical energy obtained from the energy potential from higher elevations to lower elevations. A larger number of countries prefer hydropower as the major source of renewable energy to generate electrical energy, with an efficiency of 90%. Moreover, hydropower is known for its cost-effectiveness, making it more economical when compared with other renewable energy sources [21,22]. In addition, geothermal energies are implemented by drilling and injecting wells with either water or anti-freeze materials. A large number of geothermal plants are undeveloped due to insufficient availability of technologies and inadequate capital intensives. Geothermal power plants emit low greenhouse gases when compared to fossil-fuel-based power plants. Geothermal energy sources are known for their lesser dependence on energy imports and minimize the emission of greenhouse gases [23]. In a similar way, the power prediction in tidal or ocean energy is improvised using various technologies such as deep learning, machine learning, and so on. The drawbacks existing in energy storage systems based on solar, wind, and tidal energies face problems related to the basis of electricity generation for varying meteorological conditions. This variation leads to forecasting power generation with the existing grid. The power prediction in tidal and wind energy is accomplished using tidal and wind turbines, respectively. In tidal turbines, the water is pushed against the generator which forces it to move, and the wind turbine rotates based on the wind strength. Finally, the power prediction in biomass takes place by comprising various biomass resources such as agriculture, forestry, and urban waste materials [24,25]. For instance, when the biomass is used as an alternative energy source, it produces a heating value of 3 × 10 6 kcal Mg 1 . The prediction of energy from various sources such as wind, solar, tidal, biomass, and geothermal is a challenging task unless properly trained ML and DL models are utilized in the process of energy conversion, forecasting, and power prediction. In this survey, the evaluation based on real-time metrics and comprehensive data is performed for the aforementioned sustainable sources of energy using various ML and DL approaches. Moreover, the challenges faced by the machine learning and deep learning approaches are discussed in an effective manner.


The major objective of this research survey is listed as follows:
  • The recent machine learning and deep learning approaches based on popular renewable energy sources such as wind, solar, tidal, and hydropower are reviewed for their usage in techniques like power prediction, energy conversion, and forecasting.
  • A summary is included to exhibit the main contributions and the ideologies of researchers who worked with various sources of renewable energy.
  • The power prediction, energy conversion, and forecasting based on wind, solar, tidal, and hydropower energy sources using ML and DL techniques are examined thoroughly with their advantages and drawbacks.
  • Thus, this survey acts as a key tool for future researchers to overcome the challenges in existing research in order to build a more robust model with advanced technologies.
The remaining portion of this survey paper is organized as follows: Section 2 describes the related works of this research; the methodology is presented in Section 3, and the summary of this research survey is presented in Section 4.

2. Literature Survey

In this section, recent examples of research involving machine learning and deep learning approaches are examined for their utility in different renewable energy sources such as solar, wind, hydro, and tidal energies.

2.1. Applications of Solar Energy Using Machine Learning and Deep Learning Approaches

Ikram et al. [26] introduced an Improved version of the Multi-Verse Optimizer (IMVO) algorithm that was combined with the Least Square Support Vector Machine (LSSVM) for modeling solar radiations. Initially, the data were gathered from the stations and the optimization was performed to evaluate the optimistic limits of the LSSVM approach. The optimization performed using IMVO effectively minimized the issues related to stacking and helped in modeling the solar radiations. However, the convergence occurred at a premature stage, which affected the balance among the stages of exploration and exploitation. Abdelghany et al. [27] introduced an Improved Bonobo Optimizer (IBO) to determine characteristics of Photovoltaic (PV) systems to obtain the different techniques of solar cells. The performance of the IBO is enhanced by usage of the Levy flight and Sine-cosine function, respectively in the phases of exploration and exploitation. The usage of Levy flight diminished the problems related to randomness, which were existent in the conventional BO. However, slight fluctuations were seen when the value of sine-cosine function showed too much variance.
Liu et al. [28] introduced a Distributed Energy System (DES) that is an integration of energy usage technology and a hybridized system to store the energy. The hybridized energy storage system had the ability to store heat, ice, and electricity. The configuration of the suggested DES system was optimized based on its operational characteristics and usage. Moreover, the DES with the hybrid energy system reliably minimized carbon emissions and the cost of electricity and equipment. However, regular monitoring and maintenance of the hybrid energy storage system is a must since a small leakage of coolant leads to severe impact in the surroundings. Tercan et al. [29] introduced an efficient approach to enhance the self-consumption of PV energy storage systems with various penetrating rates. Initially, optimistic allocations in the cloud storage platform using the most suitable newly improvised algorithms combined with genetic algorithm and the residual energy was utilized in the process of evaluating the feasibility of PV storage system. The suggested approach offered an improved power quality with self-consumption rates and significantly enhanced the renewable penetration levels in the power grid. However, the suggested approach had a limitation of fluctuating power supply, which drastically minimized the performance of the storage systems.
Mostafa et al. [30] suggested a framework based on the employment of big data analytics in smart grids and renewable source utilities by using three machine learning approaches. The Convolutional Neural Network (CNN) model was used to provide stability for optimal linkage to the source of power and the energy consumption outlet. While implementing the smart grids, the error occurrence was minimalized based on the data size by using the penalized linear regression. However, the demand was not fulfilled while storing large power sources due to limited storage capacity. Mehrpooya et al. [31] introduced a brainchild integration method to split the hydrogen production cycle and a molten cell along a turbine. The thermochemical cycle uses a higher solar reactor to minimize the requirement of non-renewable sources of energy. The molten fuel cells feed on the gas created due to the gasification process and the heat produced from the fuel cell is oppressed by the gas turbine. However, pressure loss from the equipment leads to inaccuracy while modeling the integration process.
Almeshaiei et al. [32] suggested a novel approach on the basis of short-term data and neural networks which were used in the process of assessing the efficacy of micro-scale photovoltaic panels. The rapid utilization in the suggested approach considers the success rate and helps in estimating the micro-scale of grid panels. However, the external factors such as humidity, dust, and varying environmental conditions affected the performance of the approach. Koo et al. [33] introduced a methodology to estimate the Monthly Average Daily Solar Radiation (MADSR) which had complex spatial patterns. The MADSR utilized k-means clustering and an Advanced Case Based Reasoning (A-CBR) model to estimate solar radiation at the global level in the horizontal surface. Additionally, the suggested approach exhibited enhanced accuracy while estimating for five solar radiation zones. However, it was incapable of evaluating the suitable location and size while implementing the solar energy storage system. Munawar and Wang [34] introduced a framework that incorporated different feature selection methods involved in short-term solar power forecasting applications. ML approaches like Extreme Gradient Boosting (XGBoost), random forest, and artificial neural network were used, and the feature selection was performed using Principle Component Analysis (PCA). The suggested approach was reliable for the process involved in solar power forecasting and proved to be effective for various applications related to smart grids. However, the suggested approach lacked in feature extraction. AlKandari and Ahmad [35] introduced an ML and Statistical Hybrid Model (MLSHM) which integrates the ML approach with statistical approach for predicting the solar power in renewable solar plants. The ML approach included long short-term memory (LSTM), gate recurrent unit (GRU), Auto Encoder LSTM (Auto-LSTM), and Auto-GRU. The accuracy of MLSHM was enhanced using two techniques based on diversities of structure and data. The forecasting results were evaluated by ensembling the aforementioned approaches using simple averaging, weighted averaging, and combinational variance. However, diversities occurred among the training sets which led to improper validation of the suggested approach. Nejati and Amjady [36] introduced a new solar power prediction method using feature clustering and hybridized classification regression. The proposed feature selection approach filtered the irrelevant features into two individual subsets which minimized the redundancy of the features. Each subset was trained individually based on two forecasts and was assigned to a single regression model. However, the prediction performance of the suggested approach degraded due to the poor learnability of the regression models.

2.2. Applications of Wind Energy Based on Machine Learning and Deep Learning Approaches

Zhang and Chen [37] introduced a hybrid model on the basis of Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), Singular Value Decomposition (SVD), and Particle Swarm Optimization (PSO) to predict the speed of winds. CEEMDAN was integrated with SVD to decompose and de-noise the data and the PSO was used in the process of optimization. At last, an autoregressive integrated model was utilized to predict the speed of winds. The suggested hybrid approach effectively stabilized the process of the windmill and the connectivity of grid-based power plants. However, the Elman neural network provided unstable output, which resulted in randomization of the initial weight and threshold value. Zhao and You [38] introduced a ML-based two-stage adaptive robust optimization framework with sets of volatile renewable generation. The two-stage adaptive approach addressed the uncertainties caused due to disjunctive renewable energies and the operations related to reliable power systems. The data which were in the state of uncertainty were combined using K-means and a method known as Density-Based Spatial Clustering of Applications with Noise (DBSCAN). The sustainable power solution was obtained by means of a tailored decomposition algorithm. However, lack of operational capabilities during the period of low wind flow gave rise to forecast error.
Fathy et al. [39] introduced an effective neural network based on an ensemble approach for Wind Energy Conversion (WEC) and Fault Detection and Diagnostic (FDD) systems. Initially, the combination of various neural networks such as artificial neural network, cascade forward neural network and feed-forward neural network were ensembled into a single optimal model. Moreover, this ensemble neural network was used to minimize computational time and storage cost by ensembling the best output out of the three models using a bagging technique. However, misclassification occurred due to false alarm rates which impacted the overall performance during the wind energy conversion. Wang et al. [40] introduced the Self-Adjusted Triboelectric Nano Generator (SA-TENG) for active and randomized wind energy. The SA-TENG could adjust to the external wind speed and achieve high output power. The SA-TENG effectively created an area by using centrifugal force with improved performance. However, the centrifugal mechanism of SA-TENG did not work when the rpm was lower than 231. Angadi et al. [41] introduced the hill-climbing Maximum Power Point Tracking (MPPT) algorithm for a Self-Excited Induction Generator (SEIG) in the systems used for wind energy conversion. A single voltage convertor was used in the process of conditioning the power and for mining maximal power with minimized requirement of sensor. Moreover, the SEIG helped in the operation of constant flux delivery and the induction motor pump was used in enlarging the region of operational range.
Anisa et al. [42] introduced a large-scale Gravity Energy Storage (GES) system in a hybrid PV-wind plant to diminish the cost used for construction by developing a simple system structure. Moreover, the genetic algorithm was used to determine the optimistic dimensionalities of the components which were based on GES. The renewable energy was dispatched by incorporating an effective storage system to prevent charging and discharging of GES. However, increased demand was experienced due to coupling of the GES with a grid system. Yang et al. [43] developed a deep reinforcement learning approach to manage uncertainties that occur in a wind farm by using energy storage system. The suggested approach utilized a data-driven controller which maps input interpretations such as direction of the wind. Moreover, a type of deep reinforcement algorithm was used to enhance the efficacy of the controller. At the time of optimization, the influence of uncertainties related to wind was considered to control the wind current and the revenue of the wind power producers. However, the suggested approach was applicable for measuring the forecasted wind power in a limited area. Rushdi et al. [44] introduced a methodology to predict the power of Airborne Wind Energy (AWE) systems by using the multivariate ML techniques. Performances of various Machine Learning (ML) algorithms namely, neural networks and ensemble approaches, were evaluated for their predictions on tether force and traction power in applications of AWE. However, the measurement errors occurred during the stage of data collection, leading to improper accuracy, which affected the process of power generation.
Aksoy and Selbas [45] introduced a novel approach using machine learning techniques to estimate the production of energy obtained from wind turbine. There were two stages involved in the conversion of wind energy to electrical energy. In the first stage, the wind energy was transformed to kinetic energy and this kinetic energy was then converted to electrical energy. A machine learning model based on Multiple Linear Regression (MLR) was used to evaluate the density of the wind energy. Before implementation of wind turbines, the proposed approach was useful to determine the wind direction and a suitable place. However, due to varying environmental conditions, the success rate of the evaluated values derived from the machine learning approach was diminished. Wang et al. [46] introduced a deep learning network stacked by an Independent Recurrent AutoEncoder (IRAE) which was designed based on ultra-short term wind power data referred to as Staked Independently Recurrent AutoEncoder (SIRAE). Initially, the variation mode decomposition technique was used to degrade the sub-sequences after which IRAE was used to extract the structural features in its initial stage. The outcome of the suggested approach varied based on climate, temperature, or humidity, and any change in the environment led to poor accuracy. Meka et al. [47] introduced a deep learning model using temporal convolutional networks to perform short-term forecasts of wind turbines used for power generation in windfarms. The temporal convolutional network was used to capture the temporal dynamics of wind power along with some meteorological variables. The short-term predictions of the power created from the wind turbine was computed by numerical models with minimalized error values. However, the error rate was higher when wind speed was in the range of 8 to 12 m/s.

2.3. Applications of Hydro and Tidal Energies Based on Machine Learning and Deep Learning Approaches

El-Aziz [48] introduced a hybridized ML approach to forecast the energy level of renewable resources. The suggested approach was the integration of Multilayer Perceptron (MLP), Support Vector Regression (SVR), and cat boost algorithm. The renewable and non-renewable energy sources were integrated and conceded through the energy supplier to evaluate the total energy load. The suggested hybrid approach was capable in forecasting the energy obtained from the renewable sources with minimal error rate. However, the count of observations exceeded the number of features which resulted in high dimensionality issues.
Zhao and Foong [49] introduced a soft computing approach to predict the output of power capability of a Combined Cycle Power Plant (CCPP). Hybridization was performed with an Electrostatic Discharge Algorithm (ESDA) and Artificial Neural Network (ANN). The suggested ESDA approach was used with an MLP predictor to provide an optimistic efficiency during the model’s training. However, the feasibility of the proposed approach was diminished while evaluating it with external toolset.
Naik et al. [50] introduced the Supervised Power Management Scheme (SPMS) for sustained flow of power in a DC micro-grid. The suggested SPMS operated based on the Micro Hydro Power Plant (MHHP), which managed the flow between load transients in a loop experimental system. However, the suggested approach was not suitable for power management in adverse environmental conditions.
Fonseca et al. [51] introduced an optimization approach for multiple criteria to evaluate and operate a system based on distributed energy sources on sustainable dimensionalities. The suggested method considered the time variations based on converting the energy that responds to the electric and hydrogen demands of storage systems. The introduced model considered the grid-connected system to fulfill the electricity demands. The developed approach had the ability to restore energy for a small area with higher efficiency. However, the process of energy conversion at higher levels led to higher emissions of CO 2 .
Ma et al. [52] introduced a new design called the Hybrid Thermal Energy Storage System (HTESS) to enhance the utility factor of Packed Bed Thermal Storage (PBTES), which was filled with Encapsulated Phase Change Material (EPCM) to minimize the effect of outlet temperature. The hybrid approach is a combination of Packed Bed Thermal Storage (PBTES) and Two-Tank Thermal Energy Storage System (TTES). Afterwards, cut-off temperature at the time of charging and discharging were evaluated on the basis of thermal capacity ratio of TES in HTESS. However, the variation was seen in electricity generation when exegetic efficiency was decreased. Sahu et al. [53] introduced a Deep Q-Network (DQN) algorithm with a fuzzy controller for stabilizing the frequency of tidal power on the basis of an Alternating Current (AC) micro grid. The suggested micro grid was modelled with penetrating renewable energies that were based on micro generating units. The suggested approach utilized a tilt-based fuzzy cascade controller which controlled the frequency of the model. However, the power created using the permanent magnet linear synchronous generator could not be fed to the grid in a direct way, which resulted in time complexity.
Chen et al. [54] introduced the Artificial Intelligence-based useful Evaluation Model (AIEM) with the ultimate goal of forecasting renewable energy and energy efficiency in an economical way. The AIEM had the ability to deploy the cost of renewable energy operations and evaluate the metrics related to the forecast of renewable energy. However, the obtained results of AIEM did not take into account the issues related to environmental conditions and financial efficiency.
Mostafa et al. [55] introduced an energy storage model which considered the applications of energy storage systems based on long-term, medium-term, and short-term forecasts. Metrics such as the Life Cycle Cost of Storage (LCCOS) and Leveled Cost of Energy (LCOE) were utilized for assessing the operational performance. This developed energy storage model acted as a decision-making support which helped to review the cost of various energy storage techniques. However, the energy storage system for storing hydrogen was not cost-effective and was highly sensitive in its implementation.
The overall results of the above-stated papers describe the analysis and findings of various machine learning and deep learning approaches based on energy conversion, power prediction, and forecasting. Though the usage of a Machine Learning (ML) and Deep Learning (DL) approach provides better output, it faces some issues based on power management, improper balancing of load at the time of energy conversion, and poor prediction accuracy. By considering the issues in the existing methodologies, a robust and an efficient methodology can be introduced to overcome the drawbacks in existing models.

3. Methodology

In this research, the energy source is classified based on long-term availability of resources categorized as two types—termed as renewable and non-renewable resources. The focus of this research is mainly on popular renewable energy sources (i.e., sustainable energy) such as solar, wind, tidal, and hydropower energy. To offer a sustainable environment, this study performs a valid survey on different techniques used for energy conversion, power prediction, and forecasting of renewable energy sources. The ML and DL approaches play an important role in various applications related to conversion of energy, prediction, and forecasting. For wind energy, the energy prediction takes place using wind turbines and the energy is created based on the speed of winds. Similarly, the power prediction using solar energy takes place using the power grid and photovoltaic cells. The solar energy is converted into electrical energy with the help of charge controllers and stored in batteries. The process of forecasting the energy from the renewable sources is a complex process. In hydro energy, the Direct Current (DC) micro grid acts as a dispatchable power source and it can be effectively regulated with the help of ML and DL algorithms. In this way, the ML and DL based approaches are used to precisely estimate the power consumption and the demands regarding the supply and production of renewable energy sources such as solar, wind, tidal, and hydro energy.
The following sections provide a brief discussion about various applications that incorporate machine learning and deep learning techniques. Figure 1 depicted below presents the diagrammatical representation for the taxonomy. As per Figure 1, energy source is separated into various categories of renewable energy sources. This study is based on sustainable energy sources, the usage of machine learning and deep learning techniques in energy conversion, power prediction, and forecasting of familiar renewable energy sources such as solar energy, wind energy, and hydro energy.

3.1. Renewable Energy Sources

Energy resources are classified into two types known as renewable and non-renewable energy sources. Renewable sources are described as natural energy which can be replenished after its depletion. Renewable sources of energy are also known as sustainable energies that create zero emissions of carbon and provide energy in a natural way [56]. In this survey, renewable energy (sustainable energy) such as solar, wind, hydro, and tidal energy are discussed with regard to their applications, scope, and limitations. Sustainable energy sources are considered to be ideal for building a sustainable environment as they do not emit greenhouse gases and minimize the effect of air pollution. The following subsections describe the usage of different renewable (sustainable) energy sources pertaining to various applications implemented for power prediction, forecasting, and energy conversion [57].

3.1.1. Wind Energy

Wind energy prediction is generally performed with the help of wind turbines that help to create wind energy based on the speed of winds. The wind exemplar iterates for longer period of time and leads to long-term fluctuations which are obscure and cannot be seen. In the same way, short-term fluctuations in the velocity of wind are described with the help of a distribution function. The speed of the wind is detected using the parametric of Weibull frequency, which is represented in Equation (1) as follows:
f v = k c   v c k 1 e v c k
where the pattern of the curve is defined as k and the variation which occurs in the velocity of the wind is represented as c . The wind energy flows at the rate of kinetic power per second, and it changes in accordance with wind kilocycle denoted as ρ . The cleared rotor region is represented as A and the variation in the velocity of air is represented as v .
Actual energy attained by rotor knives is used to differentiate between upstream and downstream air pressures. The power created by the rotor blades is a fraction of upstream air pressure and is referred to as the effective co-efficient of the rotor and helps in energy forecasting. The value of power degree is denoted as the tip speed of the rotor λ and the pitch side of the rotor blade is represented as β . The rate of the tip speed is evaluated by means of velocity of the rotor to the velocity of the air Ω . The power generation of the rotor blade is represented in Equation (2) as follows:
p T = 1 2 ρ A v 3 C p λ , β = 1 2 ρ π R 2 v 3 C p λ , β
where the value of λ is presented in Equation (3) as follows:
λ = R . Ω v
The value of power coefficients is evaluated in two different classes which is represented in Equations (4) and (5) as follows:
C p λ = i = 1 n C p i λ
C p λ , β = C 1 C 2 λ i C 3 β C 4 β 5 C 6 e C 7 λ i
The rotational axis of wind turbines is categorized into two types known as the Vertical Axis Wind Turbine (VAWT) and Horizontal Axis Wind Turbine (HAWT), which are utilized in the process of power prediction. The HAWT is known for its maximized power capacity and the VAWT is known for its ability to capture wind from any direction. The effectiveness of VAWT is based on fixed pitch or variable pitch and the VAWT variable pitch exhibits higher efficiency than the fixed pitch. Due to the larger structure of HAWT, the blades present in the wind turbine are attached to the center of gravity of the turbine. The yaw mechanism of the HAWT tends to consistently face against the wind flow direction which leads to unwanted noises. However, the HAWT is not suitable for urban areas due to the high cost of implementation and maintenance. The energy storage system is placed at the bottom portion of the windmill to store the energy and the transformer is used to transform the wind energy to electrical energy. The diagrammatical representation of the VAWT and HAWT is presented in Figure 2.
The recent research techniques based on energy conversion, power prediction, and forecast of wind energy using deep learning and machine learning techniques are represented in Table 1 as follows:
Table 1 describes methodologies based on various machine learning and deep learning approaches used in the process of energy conversion, power prediction, and forecasting of wind energy sources. Furthermore, the advantages and drawbacks of the existing approaches are also surveyed in Table 1.
During the time of wind generation, the power loss occurs due to the heat produced by the generator and the climatic changes relies as an important parameter which affects the durability of the windmill. So, during those critical stages, the power generation capability will be reduced and affects the windmill’s reliability. Therefore, an appropriate usage of a machine learning and deep learning model has the capability to enhance the power generation ability with acute forecasting and durability for generating electricity and pumping water applications.

3.1.2. Solar Energy

Solar energy is one of the purest forms of renewable energy that is free from emission of carbon components. The earth’s surface receives 140 PW of power from the sun in which 36 PW is only suitable for utilization. The process of harvesting energy from the sun takes place by means of two techniques which are Photovoltaic (PV) and Concentrated Solar Power (CSP). However, solar energy also proves to be disadvantageous due to its absence during the night and in cloudy weather conditions. Solar radiation that is received by the PV panels is in constant motion in accordance with the panels. The radiation from the solar panel comprises three types of radiation: ground reflected radiation, diffuse radiation, and direct radiation. The major count of solar radiation includes direct and diffused radiations. The equations that are used in evaluating solar radiation are represented in Equations (6)–(9). The direct radiation, which is typical of solar rays, is evaluated using the Equation (6) as follows:
I b r = G s c P M
where the solar constant is represented as G s c , and the transparency factor and the mass of the air are represented as P and M , respectively. The value of M is evaluated using Equation (7) as follows:
M = 1 s i n α s
where the angle of solar altitude is represented as α s . The direct radiation of horizontal and the tilted plane is represented in Equations (8) and (9) as follows:
I b H = I b r sin α s
I b β = I b r sin α s cos β + cos α s cos γ s sin β
where the tilt angle is represented as β and the azimuth angle is represented as γ s .
The intensity of the horizontal plane caused due to diffused radiation is represented as I D H which is evaluated using Equation (10) as follows:
I D H = 0.5 G s c 1 P M 1 1.4 ln P sin α s
The value of the diffused radiation which is tilted with the angle β is evaluated using Equation (11) as follows:
I d β = cos 2 β 2 I d H
The value of ground reflected radiation is evaluated using the Equation (12) as follows:
I ρ = H ρ 1 c o s β 2
where the value of H is evaluated using Equation (13) as follows:
H = I b H + I d H
where the diffused ground reflectance is denoted as ρ . The total solar radiation is computed as the sum of direct, ground reflected, and diffused radiation from the sun and is represented in Equation (14) as follows:
I R = I b β + I d β + I ρ
The conversion of solar energy into electrical energy undergoes the following steps:
The radiation from the sun is absorbed by the solar cells which is the primary step involved in converting the solar energy into electrical energy.
Every individual solar cell is comprising a thin layer of semiconductor material with two silicon layers. The usage of silicon-based semiconductors can act as both conductors and insulators.
Among the two silicon layers, one layer is positively charged and another one is negatively charged. The positively charged material is represented as P-type and the negatively charged material is represented as N-type.
Generally, the energy from the sun strikes the ground surface in the form of smaller packets known as photons. When these photons strike the photovoltaic material, it creates an electric current.
After the stage of conversion of solar energy into electric energy, it needs to undergo further conversion to be fit for usage. The energy obtained from the solar energy is in the form of Direct Current, so it is not suitable for direct use in buildings, home appliances, and households. So, the DC current must be converted into AC current. The inverters play a crucial role in the conversion of DC current to AC current and the inverters are capable of converting DC current to 120 volts of AC. The solar energy produced is passed into an electric panel and then through an electric grid in order to make it suitable for household applications. The recent research conducted on solar energy conversion applications that use machine learning and deep learning approaches is discussed in Table 2 as follows:
Table 2 describes the methodologies based on various machine learning and deep learning approaches used in the process of energy conversion, power prediction, and forecasting of solar energy. Moreover, the advantages and drawbacks of the existing approaches are also surveyed in Table 2.
Solar energy is mostly utilized in household applications rather than industrial applications. However, the prediction of solar energy on the basis of electricity generation for varying meteorological conditions is an issue that can occur while forecasting generation of solar power plants into the existing grid. The employment of ML is one technique to solving this complex problem. With the right training model, such algorithms can anticipate the quantity of electricity generated for the day ahead with great accuracy (up to 95%).

3.1.3. Hydro Energy and Tidal Energy

The hydro power plant is combined in a DC micro grid as a dispatchable power source and is encompassed with a Kaplan turbine, permanent magnet synchronous generator, and a diode rectifier with DC-DC converter. The flow rate allotted by the Kaplan turbine offers the required mechanical power to the permanent magnet synchronous generator. This generator tends to dispatch output power to load via AC/DC and DC/DC converters. Then, the tidal energy system was utilized in the process of producing energy from the current produced by tide movement. The total energy generated from the tidal power plant is given using the Equation (15) as follows:
T P = i = 1 m ( n × 1 2 ρ C A V )
where the total energy created from power plant is represented as T P and the thrust co-efficient is represented as C . The flow facing area is represented as A , velocity vector is represented as v , and the number of turbines is represented as n .
The recent research based on the application of hydro and tidal energy conversion techniques using machine learning and deep learning approaches are discussed in Table 3 as follows:
Table 3 describes the methodologies based on various machine learning and deep learning approaches used in the process of energy conversion, power prediction, and forecasting tidal and hydro energy sources.
Hydropower plants vary based on the day-to-day variation in daily discharge of water so a precise self-organizing method can be utilized to overcome these issues. The aforementioned problem can be overcome with the help of machine learning and deep learning approaches. The usage of ML and DL approaches effectively predicts the daily variation in water level and helps in energy forecasting and power prediction. The electrical power of ocean waves forecasting models became a stable and credible approach. However, the models require a large amount of data to train and take more time for small-scale forecasts. To overcome these general issues related to power prediction, the machine learning and deep learning approaches have the capability to enhance the accuracy and promptness of power prediction which is used grain mills and hydro-electric dams.
The taxonomy exhibits the process involved in power prediction, energy forecasting, and conversion of popular sustainable energy sources—wind, solar, hydro power, and tidal energy. The overview presented in Table 1, Table 2 and Table 3 will serve as a helpful guide for future researchers to know about the advantages and drawbacks in the existing approaches. By knowing these pros and cons, a robust approach can be introduced to overcome the existing issues. Thus, this survey will remain as a tool for future researchers and will aid in developing an effective approach that helps build a sustainable environment.

4. Summary

In recent times, AI-based learning models have achieved their ability to overcome real-world problems, especially in applications related to sustainable environments. The generation of electricity using sustainable energy sources suffers from major limitations such as minimal production of electricity and constraints regarding necessary financial investment. This survey makes a valid study for various types of machine learning and deep learning techniques employed in the conversion processes involving sustainable energy sources such as solar, wind, hydro power, and tidal energies. Moreover, the efficiency of the existing approaches presented in the literature is evaluated by a taxonomy. The advantages and limitations of the existing research are tabulated to help future readers. As a result of this survey, a hybridized and improved AI-based approach is introduced to mitigate the existing problems.

Author Contributions

The paper investigation, resources, data curation, writing—original draft preparation, writing—review and editing, and visualization were done by A.B. and N.N. The paper conceptualization and software were conducted by D.L.N. The validation and formal analysis, methodology, supervision, project administration, and funding acquisition of the version to be published were conducted by R.P.d.P. and P.B.D. All authors have read and agreed to the published version of the manuscript.


This research is supported by Spanish Research Projects P18-RT-1994 and PID2020-119082RB-C21.

Data Availability Statement

Not applicable.

Conflicts of Interest

The authors declare no conflict of interest.


  1. Rahman, A.; Aziz, T.; Nahid-Al-Masood; Deeba, S.R. A time of use tariff scheme for demand side management of residential energy consumers in Bangladesh. Energy Rep. 2021, 7, 3189–3198. [Google Scholar] [CrossRef]
  2. Eid, A.; Kamel, S.; Abualigah, L. Marine predators algorithm for optimal allocation of active and reactive power resources in distribution networks. Neural Comput. Appl. 2021, 33, 14327–14355. [Google Scholar] [CrossRef]
  3. Biswas, S.; Roy, P.K.; Chatterjee, K. Development of MADB of P-I controller using LMI technique in a renewable energy based AGC system and study its application in a deregulated environment including energy storage device. Optim. Control Appl. Methods 2023, 44, 426–451. [Google Scholar] [CrossRef]
  4. Connolly, K. The regional economic impacts of offshore wind energy developments in Scotland. Renew. Energy 2020, 160, 148–159. [Google Scholar] [CrossRef]
  5. Alkawsi, G.; Baashar, Y.; Alkahtani, A.A.; Lim, C.W.; Tiong, S.K.; Khudari, M. Viability Assessment of Small-Scale On-Grid Wind Energy Generator for Households in Malaysia. Energies 2021, 14, 3391. [Google Scholar] [CrossRef]
  6. Hassan, M.H.; Kamel, S.; Abualigah, L.; Eid, A. Development and application of slime mould algorithm for optimal economic emission dispatch. Expert Syst. Appl. 2021, 182, 115205. [Google Scholar] [CrossRef]
  7. Yang, Y.; Campana, P.E.; Yan, J. Potential of unsubsidized distributed solar PV to replace coal-fired power plants, and profits classification in Chinese cities. Renew. Sustain. Energy Rev. 2020, 131, 109967. [Google Scholar] [CrossRef]
  8. Maamoun, N.; Kennedy, R.; Jin, X.; Urpelainen, J. Identifying coal-fired power plants for early retirement. Renew. Sustain. Energy Rev. 2020, 126, 109833. [Google Scholar] [CrossRef]
  9. Ganiyu, S.O.; Martínez-Huitle, C.A.; Rodrigo, M.A. Renewable energies driven electrochemical wastewater/soil decontamination technologies: A critical review of fundamental concepts and applications. Appl. Catal. B 2020, 270, 118857. [Google Scholar] [CrossRef]
  10. Gasa, G.; Lopez-Roman, A.; Prieto, C.; Cabeza, L.F. Life cycle assessment (LCA) of a concentrating solar power (CSP) plant in tower configuration with and without thermal energy storage (TES). Sustainability 2021, 13, 3672. [Google Scholar] [CrossRef]
  11. Nazir, M.S.; Ali, N.; Bilal, M.; Iqbal, H.M.N. Potential environmental impacts of wind energy development: A global perspective. Curr. Opin. Environ. Sci. Health 2020, 13, 85–90. [Google Scholar] [CrossRef]
  12. Goh, H.H.; He, R.; Zhang, D.; Lui, H.; Dai, W.; Lim, C.S.; Kurniawan, T.A.; Teo, K.T.K.; Goh, K.C. A multimodal approach to chaotic renewable energy prediction using meteorological and historical information. Appl. Soft Comput. 2022, 118, 108487. [Google Scholar] [CrossRef]
  13. Rosas, M.A.T.; Pérez, M.R.; Pérez, E.R.M. Itineraries for charging and discharging a BESS using energy predictions based on a CNN-LSTM neural network model in BCS, Mexico. Renew. Energy 2022, 188, 1141–1165. [Google Scholar] [CrossRef]
  14. Sasaki, K.; Aki, H.; Ikegami, T. Application of model predictive control to grid flexibility provision by distributed energy resources in residential dwellings under uncertainty. Energy 2022, 239, 122183. [Google Scholar] [CrossRef]
  15. Zhang, S.; Chen, Y.; Xiao, J.; Zhang, W.; Feng, R. Hybrid wind speed forecasting model based on multivariate data secondary decomposition approach and deep learning algorithm with attention mechanism. Renew. Energy 2021, 174, 688–704. [Google Scholar] [CrossRef]
  16. Wang, L.; He, Y.; Li, L.; Liu, X.; Zhao, Y. A novel approach to ultra-short-term multi-step wind power predictions based on encoder–decoder architecture in natural language processing. J. Clean. Prod. 2022, 354, 131723. [Google Scholar] [CrossRef]
  17. Yan, J.; Möhrlen, C.; Göçmen, T.; Kelly, M.; Wessel, A.; Giebel, G. Uncovering wind power forecasting uncertainty sources and their propagation through the whole modelling chain. Renew. Sustain. Energy Rev. 2022, 165, 112519. [Google Scholar] [CrossRef]
  18. Chen, W.-H.; You, F. Sustainable building climate control with renewable energy sources using nonlinear model predictive control. Renew. Sustain. Energy Rev. 2022, 168, 112830. [Google Scholar] [CrossRef]
  19. Ser, J.D.; Casillas-Perez, D.; Cornejo-Bueno, L.; Prieto-Godino, L.; Sanz-Justo, J.; Casanova-Mateo, C.; Salcedo-Sanz, S. Randomization-based machine learning in Renewable Energy Prediction Problems: Critical Literature Review, New Results and Perspectives. Appl. Soft Comput. 2022, 118, 108526. [Google Scholar]
  20. Hu, G.; You, F. Renewable energy-powered semi-closed greenhouse for sustainable crop production using model predictive control and machine learning for energy management. Renew. Sustain. Energy Rev. 2022, 168, 112790. [Google Scholar] [CrossRef]
  21. Xue, J.; Ding, J.; Zhao, L.; Zhu, D.; Li, L. An option pricing model based on a renewable energy price index. Energy 2022, 239, 122117. [Google Scholar] [CrossRef]
  22. May, R.; Nygård, T.; Falkdalen, U.; Åstrom, J.; Hamre, Ø.; Stokke, B.G. Paint it black: Efficacy of increased wind turbine rotor blade visibility to reduce avian fatalities. Ecol. Evol. 2020, 10, 8927–8935. [Google Scholar] [CrossRef] [PubMed]
  23. Akdas, S.B.; Onur, M. Analytical solutions for predicting and optimizing geothermal energy extraction from an enhanced geothermal system with a multiple hydraulically fractured horizontal-well doublet. Renew. Energy 2022, 181, 567–580. [Google Scholar] [CrossRef]
  24. Gonz’alez-Longatt, F.; Wall, P.; Terzija, V. Wake effect in wind farm performance: Steady-state and dynamic behavior. Renew. Energy 2012, 39, 329–338. [Google Scholar] [CrossRef]
  25. Coro, G.; Trumpy, E. Predicting geographical suitability of geothermal power plants. J. Clean. Prod. 2020, 267, 121874. [Google Scholar] [CrossRef]
  26. Ikram, R.M.A.; Dai, H.-L.; Ewees, A.A.; Shiri, J.; Kisi, O.; Zounemat-Kermani, M. Application of improved version of multi verse optimizer algorithm for modeling solar radiation. Energy Rep. 2022, 8, 12063–12080. [Google Scholar] [CrossRef]
  27. Abdelghany, R.Y.; Kamel, S.; Sultan, H.M.; Khorasy, A.; Elsayed, S.K.; Ahmed, M. Development of an improved bonobo optimizer and its application for solar cell parameter estimation. Sustainability 2021, 13, 3863. [Google Scholar] [CrossRef]
  28. Liu, Z.; Fan, G.; Sun, D.; Wu, D.; Guo, J.; Zhang, S.; Yang, X.; Lin, X.; Ai, L. A novel distributed energy system combining hybrid energy storage and a multi-objective optimization method for nearly zero-energy communities and buildings. Energy 2022, 239, 122577. [Google Scholar] [CrossRef]
  29. Tercan, S.M.; Demirci, A.; Gokalp, E.; Cali, U. Maximizing self-consumption rates and power quality towards two-stage evaluation for solar energy and shared energy storage empowered microgrids. J. Energy Storage 2022, 51, 104561. [Google Scholar] [CrossRef]
  30. Mostafa, N.; Ramadan, H.S.M.; Elfarouk, O. Renewable energy management in smart grids by using big data analytics and machine learning. Mach. Learn. Appl. 2022, 9, 100363. [Google Scholar] [CrossRef]
  31. Mehrpooya, M.; Raeesi, M.; Pourfayaz, F.; Delpisheh, M. Investigation of a hybrid solar thermochemical water-splitting hydrogen production cycle and coal-fueled molten carbonate fuel cell power plant. Sustain. Energy Technol. Assess. 2021, 47, 101458. [Google Scholar] [CrossRef]
  32. Almeshaiei, E.; Al-Habaibeh, A.; Shakmak, B. Rapid evaluation of micro-scale photovoltaic solar energy systems using empirical methods combined with deep learning neural networks to support systems’ manufacturers. J. Clean. Prod. 2020, 244, 118788. [Google Scholar] [CrossRef]
  33. Koo, C.; Li, W.; Cha, S.H.; Zhang, S. A novel estimation approach for the solar radiation potential with its complex spatial pattern via machine-learning techniques. Renew. Energy 2019, 133, 575–592. [Google Scholar] [CrossRef]
  34. Munawar, U.; Wang, Z. A framework of using machine learning approaches for short-term solar power forecasting. J. Electr. Eng. Technol. 2020, 15, 561–569. [Google Scholar] [CrossRef]
  35. AlKandari, M.; Ahmad, I. Solar power generation forecasting using ensemble approach based on deep learning and statistical methods. Appl. Comput. Inf. 2020; ahead-of-print. [Google Scholar]
  36. Nejati, M.; Amjady, N. A New Solar Power Prediction Method Based on Feature Clustering and Hybrid-Classification-Regression Forecasting. IEEE Trans. Sustain. Energy 2022, 13, 1188–1198. [Google Scholar] [CrossRef]
  37. Zhang, Y.; Chen, Y. Application of hybrid model based on CEEMDAN, SVD, PSO to wind energy prediction. Environ. Sci. Pollut. Res. 2022, 29, 22661–22674. [Google Scholar] [CrossRef]
  38. Zhao, N.; You, F. Sustainable power systems operations under renewable energy induced disjunctive uncertainties via machine learning-based robust optimization. Renew. Sustain. Energy Rev. 2022, 161, 112428. [Google Scholar] [CrossRef]
  39. Fathy, A.; Alharbi, A.G.; Alshammari, S.; Hasanien, H.M. Archimedes optimization algorithm based maximum power point tracker for wind energy generation system. Ain Shams Eng. J. 2022, 13, 101548. [Google Scholar] [CrossRef]
  40. Wang, Y.; Li, X.; Yu, X.; Zhu, J.; Shen, P.; Wang, Z.L.; Cheng, T. Driving-torque self-adjusted triboelectric nanogenerator for effective harvesting of random wind energy. Nano Energy 2022, 99, 107389. [Google Scholar] [CrossRef]
  41. Angadi, S.; Yargatti, U.R.; Suresh, Y.; Raju, A.B. Speed sensorless maximum power point tracking technique for SEIG-based wind energy conversion system feeding induction motor pump. Electr. Eng. 2022, 104, 2935–2948. [Google Scholar] [CrossRef]
  42. Anisa, E.; Berrada, A.; Bakhouya, M. Optimal sizing and deployment of gravity energy storage system in hybrid PV-Wind power plant. Renew. Energy 2022, 183, 12–27. [Google Scholar]
  43. Yang, J.J.; Yang, M.; Wang, M.X.; Du, P.J.; Yu, Y.X. A deep reinforcement learning method for managing wind farm uncertainties through energy storage system control and external reserve purchasing. Int. J. Electr. Power Energy Syst. 2020, 119, 105928. [Google Scholar] [CrossRef]
  44. Rushdi, M.A.; Rushdi, A.A.; Dief, T.N.; Halawa, A.M.; Yoshida, S.; Schmehl, R. Power prediction of airborne wind energy systems using multivariate machine learning. Energies 2020, 13, 2367. [Google Scholar] [CrossRef]
  45. Aksoy, B.; Selbaş, R. Estimation of wind turbine energy production value by using machine learning algorithms and development of implementation program. Energy Sources Part A 2021, 43, 692–704. [Google Scholar] [CrossRef]
  46. Wang, L.; Tao, R.; Hu, H.; Zeng, Y.-R. Effective wind power prediction using novel deep learning network: Stacked independently recurrent autoencoder. Renew. Energy 2021, 164, 642–655. [Google Scholar] [CrossRef]
  47. Meka, R.; Alaeddini, A.; Bhaganagar, K. A robust deep learning framework for short-term wind power forecast of a full-scale wind farm using atmospheric variables. Energy 2021, 221, 119759. [Google Scholar] [CrossRef]
  48. El-Aziz, R.M.A. Renewable power source energy consumption by hybrid machine learning model. Alex. Eng. J. 2022, 61, 9447–9455. [Google Scholar] [CrossRef]
  49. Zhao, Y.; Foong, L.K. Predicting electrical power output of combined cycle power plants using a novel artificial neural network optimized by electrostatic discharge algorithm. Measurement 2022, 198, 111405. [Google Scholar] [CrossRef]
  50. Naik, K.R.; Rajpathak, B.; Mitra, A.; Sadanala, C.; Kolhe, M.L. Power management scheme of DC micro-grid integrated with Photovoltaic-Battery-Micro hydro power plant. J. Power Sources 2022, 525, 230988. [Google Scholar] [CrossRef]
  51. Fonseca, J.D.; Commenge, J.-M.; Camargo, M.; Falk, L.; Gil, I.D. Multi-criteria optimization for the design and operation of distributed energy systems considering sustainability dimensions. Energy 2021, 214, 118989. [Google Scholar] [CrossRef]
  52. Ma, Z.; Li, M.-J.; Zhang, K.M.; Yuan, F. Novel designs of hybrid thermal energy storage system and operation strategies for concentrated solar power plant. Energy 2021, 216, 119281. [Google Scholar] [CrossRef]
  53. Sahu, P.C.; Baliarsingh, R.; Prusty, R.C.; Panda, S. Novel DQN optimised tilt fuzzy cascade controller for frequency stability of a tidal energy-based AC microgrid. Int. J. Ambient Energy 2022, 43, 3587–3599. [Google Scholar] [CrossRef]
  54. Chen, C.; Hu, Y.; Karuppiah, M.; Kumar, P.M. Artificial intelligence on economic evaluation of energy efficiency and renewable energy technologies. Sustain. Energy Technol. Assess. 2021, 47, 101358. [Google Scholar] [CrossRef]
  55. Mostafa, M.H.; Aleem, S.H.E.A.; Ali, S.G.; Ali, Z.M.; Abdelaziz, A.Y. Techno-economic assessment of energy storage systems using annualized life cycle cost of storage (LCCOS) and levelized cost of energy (LCOE) metrics. J. Energy Storage 2020, 29, 101345. [Google Scholar] [CrossRef]
  56. Sayed, E.T.; Olabi, A.G.; Alami, A.H.; Radwan, A.; Mdallal, A.; Rezk, A.; Abdelkareem, M.A. Renewable Energy and Energy Storage Systems. Energies 2023, 16, 1415. [Google Scholar] [CrossRef]
  57. Mujtaba, A.; Jena, P.K.; Bekun, F.V.; Sahu, P.K. Symmetric and asymmetric impact of economic growth, capital formation, renewable and non-renewable energy consumption on environment in OECD countries. Renew. Sustain. Energy Rev. 2022, 160, 112300. [Google Scholar] [CrossRef]
  58. Roga, S.; Bardhan, S.; Kumar, Y.; Dubey, S.K. Recent technology and challenges of wind energy generation: A review. Sustain. Energy Technol. Assess. 2022, 52, 102239. [Google Scholar] [CrossRef]
  59. Rajasingam, N.; Rasi, D.; Deepa, S.N. Optimized deep learning neural network model for doubly fed induction generator in wind energy conversion systems. Soft Comput. 2019, 23, 8453–8470. [Google Scholar] [CrossRef]
  60. Huang, Y.; Li, J.; Hou, W.; Zhang, B.; Zhang, Y.; Li, Y.; Sun, L. Improved clustering and deep learning based short-term wind energy forecasting in large-scale wind farms. J. Renew. Sustain. Energy 2020, 12, 066101. [Google Scholar] [CrossRef]
  61. Shirzadi, N.; Nasiri, F.; El-Bayeh, C.; Eicker, U. Optimal dispatching of renewable energy-based urban microgrids using a deep learning approach for electrical load and wind power forecasting. Int. J. Energy Res. 2022, 46, 3173–3188. [Google Scholar] [CrossRef]
  62. Zhang, S.; Li, X. Future projections of offshore wind energy resources in China using CMIP6 simulations and a deep learning-based downscaling method. Energy 2021, 217, 119321. [Google Scholar] [CrossRef]
  63. Zhang, G.; Hu, W.; Cao, D.; Huang, Q.; Chen, Z.; Blaabjerg, F. A novel deep reinforcement learning enabled sparsity promoting adaptive control method to improve the stability of power systems with wind energy penetration. Renew. Energy 2021, 178, 363–376. [Google Scholar] [CrossRef]
  64. He, B.; Ye, L.; Pei, M.; Lu, P.; Dai, B.; Li, Z.; Wang, K. A combined model for short-term wind power forecasting based on the analysis of numerical weather prediction data. Energy Rep. 2022, 8, 929–939. [Google Scholar] [CrossRef]
  65. Li, Y.; Wang, R.; Li, Y.; Zhang, M.; Long, C. Wind power forecasting considering data privacy protection: A federated deep reinforcement learning approach. Appl. Energy 2023, 329, 120291. [Google Scholar] [CrossRef]
  66. Rajesh, P.; Muthubalaji, S.; Srinivasan, S.; Shajin, F.H. Leveraging a dynamic differential annealed optimization and recalling enhanced recurrent neural network for maximum power point tracking in wind energy conversion system. Technol. Econ. Smart Grids Sustain. Energy 2022, 7, 19. [Google Scholar] [CrossRef]
  67. Lin, G.-Q.; Li, L.-L.; Tseng, M.-L.; Liu, H.-M.; Yuan, D.-D.; Tan, R.R. An improved moth-flame optimization algorithm for support vector machine prediction of photovoltaic power generation. J. Clean. Prod. 2020, 253, 119966. [Google Scholar] [CrossRef]
  68. Bouzgou, H.; Gueymard, C.A. Fast short-term global solar irradiance forecasting with wrapper mutual information. Renew. Energy 2019, 133, 1055–1065. [Google Scholar] [CrossRef]
  69. Lee, M.-H. Identifying correlation between the open-circuit voltage and the frontier orbital energies of non-fullerene organic solar cells based on interpretable machine-learning approaches. Sol. Energy 2022, 234, 360–367. [Google Scholar] [CrossRef]
  70. Zambrano, A.F.; Giraldo, L.F. Solar irradiance forecasting models without on-site training measurements. Renew. Energy 2020, 152, 557–566. [Google Scholar] [CrossRef]
  71. Khan, W.; Walker, S.; Zeiler, W. Improved solar photovoltaic energy generation forecast using deep learning-based ensemble stacking approach. Energy 2022, 240, 122812. [Google Scholar] [CrossRef]
  72. Lim, S.C.; Huh, J.H.; Hong, S.H.; Park, C.Y.; Kim, J.C. Solar Power Forecasting Using CNN-LSTM Hybrid Model. Energies 2022, 15, 8233. [Google Scholar] [CrossRef]
  73. Stoean, C.; Zivkovic, M.; Bozovic, A.; Bacanin, N.; Strulak-Wójcikiewicz, R.; Antonijevic, M.; Stoean, R. Metaheuristic-Based Hyperparameter Tuning for Recurrent Deep Learning: Application to the Prediction of Solar Energy Generation. Axioms 2023, 12, 266. [Google Scholar] [CrossRef]
  74. Liu, Z.; Zhang, W.G.; Helfrich, K.R. Vertical Structure of Barotropic-to-Baroclinic Tidal Energy Conversion on a Continental Slope. J. Geophys. Res. Ocean. 2022, 127, e2022JC019130. [Google Scholar] [CrossRef]
  75. Omidvar, S.; Fagundes, M.; Woodson, C.B. Modification of Internal Wave Generation and Energy Conversion in the Nearshore Due to Tide-Tide and Tide-Wind Interactions. J. Geophys. Res. Ocean. 2022, 127, e2021JC017986. [Google Scholar] [CrossRef]
  76. Shen, Z.; Yao, Y.; Wang, Q.; Lu, L.; Yang, H. A novel micro power generation system to efficiently harvest hydroelectric energy for power supply to water intelligent networks of urban water pipelines. Energy 2023, 268, 126694. [Google Scholar] [CrossRef]
  77. Sari, B.; Sid, M.A. Deception attack monitoring in vulnerable hydroelectric generator system. Asian J. Control 2022, 24, 517–525. [Google Scholar] [CrossRef]
  78. Monahan, T.; Tang, T.; Adcock, T.A. A hybrid model for online short-term tidal energy forecasting. Appl. Ocean Res. 2023, 137, 103596. [Google Scholar] [CrossRef]
  79. Fujiwara, R.; Fukuhara, R.; Ebiko, T.; Miyatake, M. Forecasting design values of tidal/ocean power generator in the strait with unidirectional flow by deep learning. Intell. Syst. Appl. 2022, 14, 200067. [Google Scholar] [CrossRef]
  80. Chen, Y.C.; Yeh, H.C.; Kao, S.P.; Wei, C.; Su, P.Y. Water level forecasting in tidal rivers during typhoon periods through ensemble empirical mode decomposition. Hydrology 2023, 10, 47. [Google Scholar] [CrossRef]
Figure 1. Taxonomy on uses of ML techniques in sustainable energy sources.
Figure 1. Taxonomy on uses of ML techniques in sustainable energy sources.
Energies 16 06236 g001
Figure 2. VAWT and HAWT [58].
Figure 2. VAWT and HAWT [58].
Energies 16 06236 g002
Table 1. Research on basic approaches for power prediction, energy conversion, and wind energy forecasting.
Table 1. Research on basic approaches for power prediction, energy conversion, and wind energy forecasting.
AuthorProposed MethodAdvantageLimitations
Rajasingam et al. [59]A doubly fed induction generator that uses deep learning neural network and a Density-based Grey Artificial Bee Colony (D-GABC) algorithmThe controller with the doubly fed induction generator minimizes the reactive power in a lesser settling timeThe increased number of iterations minimizes the efficiency of the induction generator, due to its step response characteristics
Huang et al. [60]Power prediction approach that uses Density Peak Clustering (DPC) and a deep learning approachThe suggested power prediction approach has the capability to evaluate the prediction time by considering the relation among the wind turbinesIt is difficult to correlate the type of wind turbines in the power prediction approach that uses DPC
Shirzadi et al. [61]Mixed-Integer Linear Programming (MILP) method to optimize the system’s power reliability and the cost for operationsThe forecasting results from MILP breaks down the cost required for operation and enhances battery lifeThe results obtained from the MILP approach varies due to unexpected climatic changes and the unexpected climatic change probably diminishes the power generation capability
Zhang and Li [62]A downscaling approach on the basis of a Bidirectional Gated Recurrent Unit (BiGRU) to predict the offshore wind energy sourcesThe downscaling approach using BiGRU effectively detects the spatial patterns of downscaled wind energy and minimizes biasing state of suggested approachHowever, the downscaling approach using BiGRU developed uncertainties when it was evaluated in single models
Zhang et al. [63]Sparsity promoting adaptive control method to self-tune power system stabilizers along with a Deep Deterministic Policy Gradient (DDPG algorithm)The suggested DDPG approach has the ability to adjust multi- power system stabilizing parameters and helps in an optimistic parameter settingSince the suggested approach was a model-based technique, the presence of an accurate model is necessary to train the agent or else the self-tuning capability of the model will be reduced
Boyu He et al. [64]A Short-term Wind Power Forecasting (WPF) model referred as IOWA CNN-LSTM to predict the power forecasts using the Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) combined with Induced Ordered Weighted Average (IOWA) The IOWA operator can effectively combine the prediction results to CNN and LSTM which helps to enhance the power prediction accuracyHowever, the suggested approach considered the mapping relation among the speed series of the winds and other relevant factors
Yang Li et al. [65] A forecasting scheme based on federated learning and deep reinforcement learning (DRL) for short-term wind power forecasting referred as the Federated Deep Reinforcement Learning (FedDRL) with Deep Deterministic Policy Gradient (DDPG)The FedDRL has the capability to obtain accurate predictability in decentralized manner and ensure better forecasting accuracyHowever, the training efficiency of the model does not meet the requirement related to real-time dispatch power systems
P. Rajesh et al. [66] A hybrid approach to tract the maximal power of the Wind Energy Conversion System (WECS) which is a combination of the Dynamic Differential Annealed Optimization (DDAO) and Recalling Enhanced Recurrent Neural Network (RERNN) known as D 2 AORERN 2 The suggested approach has better search ability and helps to maintain system ability with desired performic uncertaintiesHowever, imbalance occurred when the huge load was prompted for a longer time duration
Table 2. Research based on machine learning and deep learning approaches for power prediction, energy conversion, and forecasting for solar energy.
Table 2. Research based on machine learning and deep learning approaches for power prediction, energy conversion, and forecasting for solar energy.
AuthorProposed MethodAdvantageLimitations
Lin et al. [67]A strategy based on inertia weighting and a Cauchy mutation operator was used to enhance performance of moth flame optimization and the prediction using Support Vector Machine (SVM)The suggested approach used moth flame optimization aided in improvising the prediction capability of photovoltaic energy and minimized the impact caused due to penetration of PV cells into the gridHowever, when excessive input was fed into model, training time of model increased, and the prediction accuracy decreased
Bouzgou and Gueymard [68]Wrapper Mutual Information Methodology (WMIM) was devised for solar irradiance forecasting which combined the mutual information and the Extreme Learning Machine (ELM) to forecast the horizonsThe suggested WMIM had advanced generalization ability which effectively selected the variables for time series forecastingHowever, the WMIM approach was not robust enough for time horizons and cloudy weather conditions
Lee [69]The tree-based machine learning model predicted the open circuit voltage of the Non-Fullerene Acceptors on the basis of Organic Solar Cells (NFA-OSC) by applying the electronic featuresThe suggested approach effectively extracted the non-linear maps among offset and Voc. Moreover, it predicted the linear interactions among the ground-reflected radiusHowever, the suggested approach lacked in reliable prediction as it did not consider factors such as illumination intensity, temperature, and the interfacial area
Zambrano and Giraldo [70]A methodological approach based on exogenous variables which correlated the solar irradiance with a multidimensional spaceThe suggested methodological approach effectively implemented the various stages of the solar PV power systemHowever, the prediction capability of the model was minimized due to the improper irradiances at unknown measurement values
Waqas Khan et al. [71] An improved generally applicable stacked ensemble algorithm known as DSE-XGB was introduced with the help of Artificial Neural Network (ANN) and Long Short Term Memeory (LSTM). The Extreme Gradient Boosting (XGB) algorithm was used to enhance the accuracy of PV generation The suggested DSE-XGB approach does not rely on the input features and helps to handle the uncertainties that occur during forecastHowever, the suggested approach does not select the trade-off among prediction gain and the evaluation time
Su-Chang Lim [72] A hybrid model which comprised a Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) for stable power generation forecastingThe CNN categorize the weather condition and the LSTM was used to stabilize the power forecast generation in an effective manner. However, the suggested approach exhibited errors when it was evaluated in a peaky point with inappropriate patterns
Catalin Stoean et al. [73] The Long Short-Term Memory and a Bi-directional LSTM were introduced to perform data collection and solar energy generation. Moreover, the Reptile Search Algorithm (RSA) was used in the process of fine tuning the hyperparametersThe RSA algorithm models the time series data of the PV output which is accompanied with the exogenous weather However, the suggested approach does not suit multi-objective time series forecasting
Table 3. Research on the basis of ML and DL for power prediction, energy conversion, and forecasting for hydro and tidal energy.
Table 3. Research on the basis of ML and DL for power prediction, energy conversion, and forecasting for hydro and tidal energy.
AuthorProposed MethodAdvantageLimitations
Liu et al. [74]The vertical profile approach for energy conversion in an ocean model based on slope shelf context due to barotropic tidal flowThe energy converted at thermocline plays a significant role in production of onshore energy radiation which is viable than the offshore radiationHowever, the vertical approach varies with the bottom slopes and leads to fluctuation at the time of energy conversion
Omidvar et al. [75]An approach based on wave generation and conversion of energy near the shore which is caused as a result of tide to tide and wind to wind interactionThe interaction that occurred among wind and tides resulted in high conversion rates and internal wave frequenciesHowever, the energy conversion system requires extensive calibration and testing to provide stability to the model
Shen et al. [76]A micro power generation system which tends to generate hydroelectric energy for supplying water through intelligent networksThe integrated permanent magnet generator used in the micro power generation system minimalized the total operating hours and improved the efficiencyHowever, the storage capacity of the battery was limited which meant only a minimal amount of energy could be stored and the overload to the battery produced heat and severely affected the efficiency of the power generation system
Sari and Sid [77]A vulnerable hydroelectric generator system to monitor the deception attack using a modified Kalman filterThe suggested system effectively addresses the region which leads to packet loss and the Kalman filter leads to conserve the structural properties of the hydroelectric conversion systemThe suggested hydroelectric generator system faces oscillations when the output of the power system stabilizer (PSS) is impacted by the attack
Thomas Monahan et al. [78]A hybrid model for short-term prediction of tidal currents based on Harmonic Residual Analysis (HRA). The HRA is combined with Linear Recurrent Forecasting (LRF) and High Order Fuzzy Time Series (HOFTS) The automated LRF was used for automated selection of components with acute forecasting results However, energy forecasting for more than the specified period using the suggestion results does not provide expected results
Ryo Fujiwara et al. [79]The tidal power generating system using Flaring Flanged Diffuser (FFD) to create power at an optimistic condition with increased velocity flow The suggested approach performs based on the correlation among fluid velocity and enhances the outlet diameter; this helps to generate more power in a minimal time period However, the installation cost of FFD was higher and needs to be periodically analyzed with human intervention
Yen-Chang Chen et al. [80] Ensemble Empirical Mode Decomposition (EEMD) and a stepwise regression model was introduced to forecast the water level of tidal river sourcesThe EEMD utilized in this research decomposes the signals of water levels from tidal river into several Intrinsic Mode Functions (IMFs)However, the suggested approach cannot determine the factors that affects the IMF
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Bhansali, A.; Narasimhulu, N.; Pérez de Prado, R.; Divakarachari, P.B.; Narayan, D.L. A Review on Sustainable Energy Sources Using Machine Learning and Deep Learning Models. Energies 2023, 16, 6236.

AMA Style

Bhansali A, Narasimhulu N, Pérez de Prado R, Divakarachari PB, Narayan DL. A Review on Sustainable Energy Sources Using Machine Learning and Deep Learning Models. Energies. 2023; 16(17):6236.

Chicago/Turabian Style

Bhansali, Ashok, Namala Narasimhulu, Rocío Pérez de Prado, Parameshachari Bidare Divakarachari, and Dayanand Lal Narayan. 2023. "A Review on Sustainable Energy Sources Using Machine Learning and Deep Learning Models" Energies 16, no. 17: 6236.

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop