Enhanced solar power prediction using CNN and ladybug beetle optimization algorithm

Abstract

Integrating renewable energy sources like solar power into the grid necessitates accurate prediction methods to optimize their utilization. This paper proposes a novel approach that combines Convolutional Neural Networks (CNN) with the Ladybug Beetle Optimization (LBO) algorithm to forecast solar power generation efficiently. Many traditional models, for predicting power often struggle with accuracy and efficiency when it comes to computations. To overcome these challenges, we utilize the capabilities of CNN to extract features and recognize patterns from past irradiance data. The CNN structure is skilled at capturing relationships within the input data allowing it to detect patterns that are natural in solar irradiance changes. Additionally, we apply the LBO algorithm inspired by how ladybug beetles search for food to tune the parameters of the CNN model. LBO imitates how ladybug beetles explore to find solutions making it effective in adjusting the hyperparameters of the CNN. This research utilizes a dataset with solar irradiance readings to train and test the proposed CNN-LBO framework. The performance of this model is assessed using evaluation measures, like Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), MAPE, and $R^{2}$ value. The experimental outcomes indicate that our hybrid CNN-LBO method surpasses existing techniques in terms of efficiency.

Keywords

Convolutional neural networks LBO solar power prediction optimization MAE

1. Introduction

The rising interest, in energy sources concerns about the environment and goals for energy have sparked a significant focus on solar power as a clean and abundant alternative [1]. Photovoltaic (PV) systems have been widely adopted across commercial and industrial sectors contributing to a more diverse energy mix and reducing reliance on fossil fuels. Grid operators use forecasting to estimate and plan energy output efficiently [2]. Precise solar forecasting allows operators with assets to optimize their device control perception consistently. Solar forecasts remain reliable amidst differences in grid capabilities. These forecasts convert conditions into the plants electricity output helping the solar industry predict plant efficiency under environmental conditions [3, 4, 5].

To seamlessly integrate power into the grid accurate forecasting techniques are vital, for optimizing energy generation, consumption and distribution. Solar power predictions empower grid operators, energy traders and stakeholders to make decisions regarding energy scheduling, resource allocation and ensuring grid stability [6]. Traditional methods, for predicting power often rely on approaches or physical models which may struggle to capture the intricate temporal and spatial dynamics present in solar irradiance patterns. Additionally, these methods may lack the flexibility needed to adapt to changing conditions and face limitations in terms of scalability and computational efficiency [7, 8].

In response to these challenges researchers are increasingly turning to machine learning and optimization techniques to improve the accuracy and efficiency of power prediction models. Convolutional Neural Networks (CNN) a type of deep learning technique has demonstrated capabilities in extracting patterns and spatial relationships from high dimensional data [9, 10]. This quality makes CNN particularly well suited for tasks involving power prediction. Furthermore, nature inspired optimization algorithms have emerged as tools for optimizing machine learning model parameters providing robustness and scalability in optimization processes. An example of such an algorithm is the Ladybug Beetle Optimization (LBO) algorithm, inspired by the foraging behavior of ladybug beetles, in nature. LBO replicates the exploration exploitation balance observed in systems enabling parameter tuning and model optimization [11].

This study introduces a framework that combines CNN with the LBO algorithm for energy solar power prediction. By combining the power of learning to extract features and recognize patterns, with the optimization capabilities of LBO, for refining models our method seeks to improve the precision, validation and computational effectiveness of power prediction systems. This strategy enhances the accuracy of term power forecasts and strengthens the dependability and consistency of integrating solar power into the grid.

The rest of the paper is arranged as follows: Section 2 reviews related work; Section 3 covers materials, methodologies, and the CNN-LBO framework. Section 4 covers the experimental setup, findings, and performance analysis. Section 5 provides the conclusion and directions for future research.

2. Literature review

Solar power prediction has garnered significant attention from researchers and practitioners due to its critical importance in optimizing solar energy integration into the grid and improving the overall efficiency of renewable energy systems. This section reviews the existing literature on solar power prediction techniques, focusing on deep learning approaches and optimization algorithms [12, 13].

Ozbek et al. [14] presented a supervised neural method with an LSTM deep net – the method aimed to predict the electricity generation of a 1.15MW solar PV power plant one hour ahead. Two distinct data-driven methodologies, adaptive ANFIS and fuzzy c-means with grid partition, were integrated with the recommended deep neural network. Empirical data is used to confirm the accuracy of the data generated by the models. The comparison results showed that the approach described yielded the most favorable results. A multi-stage model is proposed for co-planning wind stations, energy storage systems, and power line construction, including severe weather events [15]. A deep learning model using TLSM systems is outlined for predicting annual peak loads. The proposed model’s MILP method is solved using the Wielders Transform. The effectiveness of the proposed model is evaluated using a customized IEEE RTS experimental configuration.

Weng et al. [16] proposed a wind power time-series architecture with BiLSTM to tackle energy consumption issues. Renewable energy will play a crucial role in the future energy system, particularly by introducing urban energy networks. Renewable power sources such as wind have not significantly contributed to the energy supply owing to their unreliability and instability. The secure running of the power system relies on precisely forecasting wind power generation. The findings are drawn using wind data gathered by the urban energy networks. A technique based on Deep Learning for maximizing and controlling power was suggested by Ashok et al. [17]. These controllers use many control strategies, including monitoring voltages, phases, and frequencies on both sides of the fixed switch to ensure system stability. Establishing a sustainable energy economy via the widespread use of renewable energy sources requires regulatory structures that are effective and efficient. Predictive analysis emphasizes the need for control techniques to establish a reliable and efficient micropower system.

Shekher et al. [18] proposed integrating deep learning models, such as CNN and RNN algorithms, to overcome the gap in power usage predictions. The CNN stack exhibits the best accuracy level among the evaluated models when predicting energy consumption and solar power production. It has an average absolute percentage deviation of 1.71%, a MAE of 0.015%, and a RMSE of 0.23%, respectively. However, in the context of energy generation from wind turbines, the recurrent neural network stands out as the most precise model. The score is 0.070, with a median absolute percentage error of 2.65. Moreover, the MAE and ABE both have a root value of 0.38. The research employed training and validation data on solar power production and wind turbine power generation, which the International Renewable Energy Organization gave.

Wang et al. [19] proposed an advanced hybrid prediction system that incorporates meteorological factors throughout its development. The constraints of single-data dimension prediction are overcome by introducing a feature selection module. This module allows for selecting features and assigning weights to external meteorological parameters during the forecast process. Second, to develop deterministic combinatorial prediction models, both shallow and deep learning models are used flexibly, and multi-objective intelligent optimization procedures are proposed. To successfully satisfy the requirements of a wide variety of consumer information, the module can effectively enhance the diversity of prediction models while simultaneously completely weighing the accuracy and stability of prediction. In conclusion, an interval prediction model is developed in order to further enrich the PV power forecast system from the point of view of uncertainty analysis [20, 21].

This review of literature showcases methods employed in predicting power encompassing traditional techniques, deep learning strategies and optimization algorithms. While conventional approaches offer forecasts, machine learning and deep learning methods enhance precision and scalability. Incorporating optimization algorithms boosts the effectiveness and productivity of models used for forecasting power. The suggested model that merges CNN, with the Ladybug Beetle Optimization algorithm seeks to harness the advantages of learning and nature inspired optimization, for prediction of solar energy.

3. Materials and methods

In the methodology section the process is detailed for creating the data preprocessing steps introducing a power prediction framework that focuses on energy efficiency using a Convolutional Neural Network (CNN) and the Ladybug Beetle Optimization (LBO) algorithm. This section covers aspects such, as data preprocessing, designing the model architecture optimizing parameters conducting training sessions and evaluating the results.

3.1 Data preprocessing

In the power field getting the data ready, for analysis or modeling involves a series of steps like cleaning, transforming and organizing it. This includes filtering out any data points and selecting the important features for estimating loads later on. In 2019 Nova Scotia Community College gathered 23,400 recordings in Halifax during daylight hours with a 10-minute interval between readings. Any data that looks unusual should be removed away to avoid causing issues during model building and potentially leading to overfitting. Outdated assets are left out during the selection process. A good feature selector helps improve prediction accuracy and speeds up the process by focusing on features. It’s important to divide the dataset into training, validation and test sets, for training models and evaluating their performance.

3.2 CNN method

This segment delves into the forecasting of power using Convolutional Neural Networks (CNN) which involves employing learning methods to enhance the precision and dependability of solar power prediction. Predicting power is essential, for energy management grid stability and integrating renewable energy sources into the grid. CNN utilizes convolution and pooling as techniques for extracting features within the model. Pooling layers reduce the size of feature maps generated by layers. This reduction aids in simplifying the networks calculations while preserving information. By encapsulating features in each area pooling layers improve the model’s consistency concerning translation and resilience to adjustments. Connected layers amalgamate extracted features from layers and facilitate abstract thinking at a higher level. In power prediction connected layers merge spatial and temporal data to capture intricate temporal dependencies and correlations. The quantity of neurons in layers alongside activation functions determines the networks’ ability to represent nonlinear connections within data sets. The output layer of CNN estimates forthcoming values for power generation over a specified forecast period. Depending on the forecasting objective the output layer could include one neuron, for predicting a value or multiple neurons for predicting time series simultaneously.

Creating a CNN model to predict power involves selecting the network structure adjusting key parameters and using methods, like regularization and optimization to make sure forecasts of solar energy generation are precise and dependable. It’s important to go through a trial-and-error process considering real world scenarios and expert knowledge to guarantee the model works effectively in situations. Depending on the input, the convolutional layers provide output, and a unit centered at $A$ and $B$ also produces output, as seen in the following.

$\displaystyle V_{i\,A,B}^{j}=U_{i\,A,B}^{j}+\mathop{\sum}\limits_{w=1}^{K_{1}^% {w-1}}\mathop{\sum}\limits_{q=-K_{1}^{h}}^{A_{1}^{h}}\mathop{\sum}\limits_{t=-% K_{2}^{h}}^{K_{2}^{h}}({E_{i,w}^{h}})*({V_{w\,L+q,B+t}^{h-1}})$ (1)

In Eq. (1), * represents the convolutional operator, $V_{i\,A,B}^{j}$ indicates the fixed feature map from $V^{\text{th}}$ convolutional layer centered at $A, B$ . The weights of the layers are denoted as $E_{i,w}^{h}$ , and $U_{i\,A,B}^{j}$ represents the bias, which is finely tuned by XGBoost. The parameters $w$ , $q$ , and $t$ correspond, to the feature maps that generate the output from each filter. RELU stands for Rectified Linear Activation Unit. According to Eq. (2), RELU functions, as a piecewise linear function that outputs the input if its positive; otherwise, it returns zero. It’s practice to apply RELU following the convolution layer.

$\displaystyle V_{i}^{j}=\textit{fun}({V_{w}^{j-1}})$ (2)

Fully connected layers: Prediction mode and a fully connected layer are used to safely identify nodes using the abstractive characteristics extracted from convolutional layers. The outcome of a connected layer is represented by the expression

$\displaystyle O_{n}^{j}=U_{i\,A,B}^{j}+\mathop{\sum}\limits_{w=1}^{K_{1}^{w-1}% }\mathop{\sum}\limits_{q=-K_{1}^{h}}^{A_{1}^{h}}\mathop{\sum}\limits_{t=-K_{2}% ^{h}}^{K_{2}^{h}}({E_{i,w}^{h}})*({V_{w\,A+q,B+t}^{h-1}})$ (3)

In this context $E_{i,w}^{h}$ denotes the weight connecting a unit at $({A,B})$ in the $w^{\text{th}}$ feature map of layer $({i-1})$ with the $j^{\text{th}}$ unit, in layer $i$ . $H_{n}^{j}$ is described as the anticipated result. Figure 1 illustrates the flow of data through the different layers, including the dimensions of the feature maps at each stage and labeling the key components of the architecture.

Figure 1.

Classification model of CNN.

The developed approach integrates new versions of XGBoost and attention processes significantly to improve the CNN model’s performance in classifying solar power datasets. For the classification job described in this technique, the final pre-processed dataset is used to train the CNN model by the architecture shown in Fig. 1.

3.3 Ladybug Beetle Optimization algorithm

In this section, Ladybug Beetle Optimization (LBO) [22] was applied after classification using the CNN approach to optimize further the results obtained from the CNN classifier. The LBO is applied to optimize the parameters or hyperparameters of the CNN model to improve its performance further. LBO can optimize parameters such as learning rates, filter sizes, dropout rates, and other architectural parameters of the CNN model. In this case, a uniform random update is made to the ladybug population in the search space. The fitness function is determined by verifying the ladybug population. Also, everyone is escorted in an organized manner to the hot zone. The ladybugs’ ability to migrate from one spot to another is a result of their coordinated nature. Ladybugs create a distinctive signal that they use to follow one another. At this point, the member seen at the beginning may feel the hottest part more than the others. This model incorporates a mutation process to achieve a balance between the exploration and exploitation sectors.

At each stage, the ladybug’s location is refreshed and confirmed. Afterward, the goal function values are utilized to choose the excellent group after fusing the current and existing ladybug positions. The new position is used here to validate and update the next iteration. A whole member of the population is selected using the created model to update the population in all iterations.

The mathematical representation, for the location of the $p^{\text{th}}$ ladybug can be found in Eq. (4).

$\displaystyle L_{p}({j+1})=L_{p}(j)+\textit{rand}\times({L_{q}(j)-L_{c}(j)})+% \textit{rand}\times|{S_{P}}|^{-\frac{j}{D(j)}}\times L_{p}(j)$ (4)

In this case $S_{p}$ represents the proportion of the cost of the $p^{\text{th}}$ ladybug compared to the cost of all ladybugs, in the $j^{\text{th}}$ round. Its validity is confirmed through Eq. (5).

$\displaystyle S_{p}=\frac{g({L_{p}(j)})}{\mathop{\sum}\nolimits_{s=1}^{D_{j}}g% ({L_{s}(j)})}$ (5)

In this method the roulette wheel selection process is used to choose $q^{\text{th}}$ ladybug replaces the position of $p^{\text{th}}$ ladybug. The distances, between the ladybugs $D(j)$ are obtained from a range between 0 and 1 divided into sections. Initially a random number is assigned to determine the location of each ladybug. The ladybug in the position is then used to select the function. The vector representing the population is denoted as $Z$ along, with $D(j)$ as shown in Eq. (6).

$\displaystyle Z=[{Z_{1},Z_{2},\ldots,Z_{D^{j}}}],Z_{p}=e^{-\beta\frac{g({L_{p}% ,(j)})}{g_{WS}}}$ (6)

In this case the favourable goal, in the phase is referred to as $g_{WS}$ while the roulette wheel pressure factor is denoted as $\beta$ . When $\beta$ has a value the roulette wheel selection opts for the ladybugs, within the group. Figure 2 illustrates the workflow process of the proposed model.

Figure 2.

Workflow process of the proposed model.

Exploring the unexplored places in the search area of the ladybug population throughout the mutation process might be a challenging task in the updation phase. These programs efficiently accelerate the search operation and also randomly conduct mutation. The number of decision variables, in the $p^{\text{th}}$ ladybug is confirmed using Eq. (7).

$\displaystyle X_{m}=\textit{round}({X\ast\mu_{r}})$ (7)

In this scenario the mutation rate is denoted as $\mu_{r}$ . The decision variable length is labeled as $X$ . The $X_{m}$ variables featured in the path of the $p^{\text{th}}$ ladybug are chosen at random. Subsequently the random variables, within the region are substituted to determine the ladybugs position. Insects frequently get. Vanish, in locations because of chilly conditions and they migrate to different regions because of the cold. The mathematical depiction of the elimination process in the ladybug is. The tally of ladybugs, at each stage is shown in Eq. (8).

$\displaystyle D({j+1})=\textit{round}\left({D(j)-\textit{rand}\times D(j)\left% ({\frac{f}{f_{mx}}}\right)}\right)$ (8)

In this scenario the term $f$ represents the count of validation while $f_{mx}$ denotes the count of functional validation. When the termination condition includes the number of iterations, Eq. (9) is used to calculate and determine the count of ladybugs, across all iterations.

$\displaystyle D({j+1})=\textit{round}\left({D(j)-\textit{rand}\times D(j)\left% ({\frac{j}{j_{mx}}}\right)}\right)$ (9)

In this case the phrase $j_{mx}$ represents the iteration while $j$ stands for the iteration itself. Algorithm 1 provides the outline for the CNN-LBO process.

Algorithm 1: Pseudocode of CNN-LBO
Input: Initialize the population $O_{i}=O_{1},O_{2},O_{3},\ldots,O_{n}$
Output: $\textit{Best}_{\textit{sol}}$
While ( $t\leqslant\textit{itr}_{\max}$ ) do
If ( $\textit{round}>r_{n}$ )
$d=$ 0
While ( $d=$ 2) do
Adjust the $q^{\text{th}}$ ladybug using a selection method to a roulette wheel.
End While
Check the validity of the following action;
Check the updated cost function, for the population.
Else
Determine the quantity of $p^{\text{th}}$ ladybugs using Eq. (8).
Create a number at random within a range.
Renew the random variables
End if
End while
Collect both the original group of individuals.
Arrange the group based on a cost function.
Find the best solution

4. Experimental setup and result analysis

The experimental results are obtained using a MATLAB-developed model and a 2.0 GHz Intel Core i3 CPU with 8 GB of RAM. During 2019 at Nova Scotia Community College (NSCC), 23400 recordings were collected in Halifax during daylight hours with a 10-minute solar interval. The data records are used to apply the suggested model. With a total of 5240 values (3668 for training and 1572 for testing), we analyzed the solar power generation predictions hourly for both the winter and summer seasons. Various matrices, including MPE, RMSE, MAE, and $R^{2}$ , are used to assess the performance of the suggested model. There are other alternative prediction measuring approaches. However, this research only focuses on four factors to determine forecast accuracy. The prediction matrices facilitate the comparative analysis of several prediction models. Table 1 displays the values of several optimized parameters achieved using the proposed algorithm.

Table 1
Simulation parameters of CNN-LBO

Parameters	Values
Filter	256 Cov1/2
Kernel size	3
Learning rate	0.05
Max pooling	2
Activation function	ReLU
Epochs	40
Population size	30
Coefficient vector	[0, 2]
$\theta$	[ $-$ 1, 1]
$p$	0.1

4.1 Evaluation performance

The CNN-LBO model is proposed to analysis using evaluation measures including MPE, RMSE, MAE, and $R^{2}$ by Eqs (10)–(13). In order to provide a more accurate assessment of the CNN-LBO, this research compares it with other techniques such as LSTM, XGB, and RBFN for predicting solar power. The data set comprises 5240 samples gathered at 10-minute intervals over both winter and summer days.

$\displaystyle\textit{MAPE}=\left(\frac{1}{N}\mathop{\sum}\limits_{i=1}^{N}% \frac{|{T_{t}({x_{i}})-O_{p}({x_{i}})}|}{T_{t}({x_{i}})}\right)\ast 100$ (10) $\displaystyle\textit{RMSE}=\sqrt{\frac{1}{N}\mathop{\sum}\limits_{i=1}^{N}|T_{% t}({x_{i}})-O_{p}({x_{i}})|^{2}}$ (11) $\displaystyle\textit{MAE}=\frac{1}{N}\mathop{\sum}\limits_{i=1}^{N}|{T_{t}({x_% {i}})-O_{p}({x_{i}})}|$ (12) $\displaystyle R^{2}=1-\frac{\mathop{\sum}\nolimits_{i=1}^{N}({T_{t}({x_{i}})-O% _{p}({x_{i}})})^{2}}{\mathop{\sum}\nolimits_{i=1}^{N}(T_{t}({x_{i}})-\tilde{T}% _{t})^{2}}$ (13)

Solar power prediction has been based on two distinct seasons: winter and summer. The results were analyzed using these seasons. For the purpose of determining the model’s accuracy, only data from a single month is taken into consideration out of the full season. The month of December is regarded as the winter season, whereas May is considered the summer season. Figure 3 presents a comparative examination of the suggested approach on a cold day compared to existing methods. Figure 4 presents a comparative study of several approaches used to determine the proposed method’s performance accuracy throughout the summer.

Tables 2 show the two models’ performance requirements for the winter and summer days, respectively. The data across all performance metrics on both days clearly show that the CNN-LBO model performs the best.

Table 2

Performance evaluation of both winter and summer day

Season	Method	MAPE	MAE	RMSE	$R^{2}$
Winter	CNN-LBO	3.784	289.4	381.6	0.914
	LSTM	3.987	312.5	422.1	0.927
	XGB	4.235	456.6	552.6	0.951
	RBNF	4.897	711.3	798.3	0.998
Summer	CNN-LBO	0.311	758.7	943.5	0.699
	LSTM	0.463	833.3	1117.1	0.726
	XGB	0.689	1167.3	1656.3	0.933
	RBNF	0.983	1522.7	1893.6	0.992

Figure 3.

Comparison analysis of winter day.

Figure 4.

Comparison analysis of summer day.

The challenging problem lies in attaining reduced temporal complexity rather than only focusing on enhancing accuracy in classification tasks and reducing error rates for solar forecasting data. Consequently, we compared the time complexity between the proposed model and the baselines – the evaluation of the models’ training and testing durations, which are shown in Table 3. The training and testing times of the Proposed CNN-LBO method were compared with those of the LSTM, XGB, and RBNF methods.

Table 3

Various methods of training and testing time

Method	Training time (second)	Testing time (second)
CNN-LBO	376.238	2.506
LSTM	487.612	4.117
XGB	645.441	6.528
RBNF	892.204	8.482

The study individual undergo training, followed by a series of experiments done using the dataset. Figures 5 and 6 show the test accuracies of several approaches, together with the training loss and validation loss.

Figure 5.

Comparison analysis of training loss.

Figure 6.

Comparison analysis of validation loss.

5. Conclusion

In this paper, the integration of renewable energy sources like solar power into the energy grid necessitates accurate forecasting techniques to optimize energy utilization and grid stability. In this study, we proposed a novel framework for energy-efficient solar power prediction using Convolutional Neural Networks (CNN) and the Ladybug Beetle Optimization (LBO) algorithm. Through the integration of CNN, our framework leverages the representational power of deep learning to extract spatial features and capture temporal dependencies in solar irradiance data effectively. CNN have shown remarkable capabilities in pattern recognition tasks, making them well-suited for analyzing complex solar irradiance dynamics. Furthermore, incorporating the LBO algorithm enables efficient parameter tuning and optimization of the CNN model’s hyperparameters. The proposed model converges to optimal or near-optimal solutions, enhancing the performance and scalability of the solar power prediction. Experimental results demonstrated that the proposed CNN-LBO framework outperforms existing methods in terms of prediction accuracy, MPE, RMSE, MAE, and $R^{2}$ . Harnessing the synergies between deep learning and nature-inspired optimization techniques offers a promising solution for enhancing the efficiency of solar power forecasting systems. Future research directions may explore the application of hybrid models, ensemble methods, and real-time optimization techniques to improve the performance and applicability of solar power forecasting systems in diverse operational environments.

References

Abou Houran

Bukhari

Zafar

Mansoor

Chen

. COA-CNN-LSTM: Coati optimization algorithm-based hybrid deep learning model for PV/wind power forecasting in smart grid applications. Applied Energy. 2023 Nov 1; 349: 121638.

Zouli

. Design of solar power-based hybrid desalination predictive method using optimized neural network. Desalination. 2023 Nov 15; 566: 116854.

Ahmed

Sreeram

Mishra

Arif

. A review and evaluation of the state-of-the-art in PV solar power forecasting: Techniques and optimization. Renewable and Sustainable Energy Reviews. 2020 May 1; 124: 109792.

Sharadga

Hajimirza

Balog

. Time series forecasting of solar power generation for large-scale photovoltaic plants. Renewable Energy. 2020 May 1; 150: 797-807.

Irshad

Islam

Gari

Algarni

Alqahtani

Imteyaz

. Arithmetic optimization with hybrid deep learning algorithm based solar radiation prediction model. Sustainable Energy Technologies and Assessments. 2023 Jun 1; 57: 103165.

Shivashankaraiah

Shashibhushan

Prema Kirubakaran

Anjanappa

Sharma

Mayiladuthurai Vaidyanathan

Taqui

Obaid

Alfarraj

. Developing a deep learning and reliable optimization techniques for solar photovoltaic power prediction. Electric Power Components and Systems. 2024 Feb 13: 1-5.

Moradzadeh

Moayyed

Mohammadi-Ivatloo

Vale

Ramos

Ghorbani

. A novel cyber-Resilient solar power forecasting model based on secure federated deep learning and data visualization. Renewable Energy. 2023 Jul 1; 211: 697-705.

Raman

Mewada

Meenakshi

Jayaseelan

Sharmila

Taqui

Al-Ammar

Wabaidur

Iqbal

. Forecasting the PV power utilizing a combined convolutional neural network and long short-term memory model. Electric Power Components and Systems. 2024 Jan 20; 52(2): 233-49.

Yadav

Chand

Sahoo

Kumar

. Comparative analysis of detection and classification of diabetic retinopathy by using transfer learning of CNN based models. Journal of Intelligent & Fuzzy Systems. 2022 Jan 1; 43(1): 985-99.

10.

Ibrahim

Gharghory

Kamal

. A hybrid model of CNN and LSTM autoencoder-based short-term PV power generation forecasting. Electrical Engineering. 2024 Jan 24: 1-7.

11.

Sahoo

Pandey

Amgoth

. A genetic algorithm inspired optimized cluster head selection method in wireless sensor networks. Swarm and Evolutionary Computation. 2022 Dec 1; 75: 101151.

12.

Amoura

Torres

Lima

Pereira

. Hybrid optimisation and machine learning models for wind and solar data prediction. International Journal of Hybrid Intelligent Systems. 2023(Preprint): 1-6.

13.

Jang

Bae

Park

Sung

. Solar power prediction based on satellite images and support vector machine. IEEE Transactions on Sustainable Energy. 2016 Mar 21; 7(3): 1255-63.

14.

Ozbek

Yildirim

Bilgili

. Deep learning approach for one-hour ahead forecasting of energy production in a solar-PV plant. Energy Sources, Part A: Recovery, Utilization, and Environmental Effects. 2022 Dec 21; 44(4): 10465-80.

15.

Moradi-Sepahvand

Amraee

Gougheri

. Deep learning based hurricane resilient coplanning of transmission lines, battery energy storages, and wind farms. IEEE Transactions on Industrial Informatics. 2021 Apr 20; 18(3): 2120-31.

16.

Weng

Pei

Ren

Jiang

Zheng

Liu

Gao

. Modeling and forecasting of wind power output of urban regional energy internet based on deep learning. InJournal of Physics: Conference Series. 2021; 1732(1): 012190. IOP Publishing.

17.

Ashok Babu

Mazher Iqbal

Siva Priyanka

Jithender Reddy

Sunil Kumar

Ayyasamy

. Power control and optimization for power loss reduction using deep learning in microgrid systems. Electric Power Components and Systems. 2024 Jan 20; 52(2): 219-32.

18.

Shekhar

Bhushan Mahato

Suman

Singh

Bhagyalakshmi

Prasad Sharma

Laxmi Kantha

Agraharam

Rajaram

. Demand side control for energy saving in renewable energy resources using deep learning optimization. Electric Power Components and Systems. 2023 Nov 26; 51(19): 2397-413.

19.

Wang

Zeng

. Hybrid ultra-short-term PV power forecasting system for deterministic forecasting and uncertainty analysis. Energy. 2024 Feb 1; 288: 129898.

20.

Khan

Javaid

Gamage

Taylor

Baig

. Heuristic algorithm based optimal power flow model incorporating stochastic renewable energy sources. IEEE Access. 2020 Aug 14; 8: 148622-43.

21.

Dhara

Rather

. Non-synchronous inertia estimation in a renewable energy integrated power system with reduced number of monitoring nodes. IEEE Transactions on Sustainable Energy. 2022 Dec 8; 14(2): 864-75.

22.

Safiri

Nikoofard

. Ladybug Beetle Optimization algorithm: Application for real-world problems. The Journal of Supercomputing. 2023 Feb; 79(3): 3511-60.