Solar Panel Degradation Prediction using Machine Learning: A Comprehensive Approach

Solar Panel Degradation Prediction using Machine Learning: A Comprehensive Approach | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Solar Panel Degradation Prediction using Machine Learning: A Comprehensive Approach Deepanshu, Kartik Garg, Harshit Mittal, Vivek Yadav, Omkar Singh Kushwaha This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6297947/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Solar photovoltaic (PV) systems are central to the world's movement toward renewable power, but their performance declines with time owing to a combination of environmental expo- sure and usage stress. In this research, we suggest a hybrid machine learning system that incorporates multi-source data such as device logs, weather history, customer endpoints, and network endpoints in order to make precise predictions about solar panel degradation. The data, which was obtained from the London Datastore and recorded by UK Power Networks for 480 days, is processed to obtain significant features capturing electrical performance as well as environmental conditions. High-level feature extraction methods were used to obtain stress measures like temperature stress, humidity stress, solar exposure, voltage drop stress, current drop stress, and total harmonic distortion (THD) stress. Fifteen regression models were trained and compared based on mean absolute error (MAE), mean squared error (MSE), root mean squared error (RMSE), and the coefficient of determination (R2) [ 1 , 2 ]. Our top-performing hybrid ensemble, which was built by stacking an artificial neural network (ANN), XGBoost, and Random Forest, recorded an R2 value above 0.96. These findings highlight the effectiveness of combining various data sources and advanced feature engineering for proactive maintenance and enhanced operational efficiency of PV systems. Artificial Intelligence and Machine Learning solar panel degradation photovoltaic systems machine learning hybrid models data integration feature engineering Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 1 Introduction The international energy scene is being revolutionized as countries move towards renewable sources of energy. Of these, solar photovoltaic (PV) systems have become a bedrock technology owing to their scalability, reducing cost of installation, and ability to mitigate greenhouse gas emissions [1, 2]. Yet, with time, solar panels undergo natural performance degradation as a result of intrinsic aging and extrinsic stress factors—high ambient temperatures, high humidity levels, and electrical load fluctuations—collectively reducing energy yield and raising maintenance costs [3, 4]. Conventional degradation models, usually empirical or simple linear regression in nature, are typically not sufficient to capture the intricate and nonlinear interactions between these stress factors [5]. Recent developments in machine learning (ML) have brought strong capabilities to analyze large-scale, multi-source data to represent these complexities with greater accuracy [6, 7]. In this research, we introduce a hybrid ML framework in its entirety that combines data from device logs, weather history, customer endpoints, and network endpoints to make solar panel degradation predictions with high accuracy. Our approach entails a careful data processing and normalization coupled with sophisticated feature engineering to identify the important stress metrics like temperature stress, humidity stress, sun exposure, voltage drop stress, current drop stress, and total harmonic distortion (THD) stress. The degradation of solar panels over time can be quantified using key equations such as the Current Drop Calculation (Eq. 8) and the Voltage Drop Calculation (Eq. 7). These provide a mathematical basis for analyzing efficiency loss.We compare the performance of fifteen regression models and hybrid ensemble approaches on the basis of metrics such as mean absolute error (MAE), mean squared error (MSE), root mean squared error (RMSE), and the coefficient of determination (R 2 ) [8]. The overall aim of this research is to enable informed insights for proactive maintenance and increased operational efficiency of PV systems, hence contributing towards the larger movement towards sustainable energy [9]. 2 Methodology This section presents the different machine learning models and methodologies employed in this study. We present the models used, the configurations, as well as the performance metrics applied to measure the performance. The models selected were regression-based, ensemble learning techniques, and neural networks, and each was subjected to testing in terms of determining their performance on the target variable. We further describe the preprocessing techniques, the hyperparameter tuning methods, as well as validation methods applied in order to promote stable model performance 2.1 Data Collection and Integration Data used in this research were extracted from the London Datastore's Photovoltaic (PV) Solar Panel Energy Generation Data [ 16 ]. The data consist of voltage, current, power, energy, and weather readings that UK Power Networks gathered over a 480-day period. Measurements were taken at 10-minute intervals (at 1-minute intervals during the summer season) and averaged out into hourly minimums and maximums. Device logs, customer end-points, and network endpoints' data were combined using shared identifiers like serial number and date-time, creating an aggregated dataset with more than 70 features that capture both operational behavior and environmental factors [ 8 , 9 ]. 2.2 Data Processing A multi-step data processing pipeline was utilized in order to maintain the quality and integrity of the dataset. Missing values and extreme outliers (e.g., cases of high solar irradiation with zero PV output or energy levels below 1 kWh under high irradiance) were detected and deleted according to domain-specific thresholds [ 10 ]. The date-time data was then normalized to enable the extraction of temporal features (e.g., day, month, and hour). All the features of interest were numericized and normalized to the 0–1 range using the following formula: Normalised Feature = (original value – minimum value) / (maximum value – minimum value)……………………………………………………………………………………………………(1) Lastly, information from the different sources were combined into one unified dataset that encompasses both the electrical performance and environmental conditions influencing solar panel degradation. 2.3 Feature Extraction A rigorous feature extraction procedure was conducted to obtain metrics that can capture the stresses on solar panels that cause degradation. Correlation analysis, as well as heatmaps (as shown in Fig. 1), was utilized for the verification of feature selection. The following essential features were obtained: Time-Based Feature The days since they were installed were calculated to reflect the natural aging of the panels TimeDegradation(t) = α · t………………………………………………………………...…………(2) where t is the number of days since installation. Environmental Stress Metrics Environmental stress is represented as the compounded effect of weather-related variables TempStress(t) = max {0, HiTemp(t) − 25}...……………………………………………………….(3) HumidityStress(t) = OutHum(t) × Rain(t)………………………………………………………...(4) SolarExposure(t) = SolarRad(t) × HiSolarRad(t)……………………………………………….(5) The total environmental stress is computed as: EnvironmentalStress(t) = TempStress(t) + HumidityStress(t) + SolarExposure(t).………(6) These characteristics were chosen due to their high correlation with degradation, as validated by correlation heatmaps and other plots (see Figs. 2 ). Operational Stress Metrics Operational stress is extracted from electrical performance variations VoltageDrop(t) = (V GEN MAX(t) – V GEN MIN(t)) / V GEN MAX(t)……………………………..(7) CurrentDrop(t) = (I GEN MAX(t) – I GEN MIN(t)) / I GEN MAX(t)………………………………..(8) THDStress(t) = thdV MAX(t) + thdI GEN MAX(t)……………………………………………………..(9) The cumulative operational stress is then: OperationalStress(t) = VoltageDrop(t) + CurrentDrop(t) + THDStress(t)………………………….(10) Maintenance Impact Cleaning and maintenance (e.g., repairs)vcan reduce degradation. While no explicit maintenance data are available, the model framework can include them if available MaintenanceImpact(t) = \(\:{\sum\:}_{\left\{i=1\right\}}^{t}\:\left(CleaningEvents\left(i\right)+RepairEvents\left(i\right)\right)\) ………………(11) In this study, the maintenance impact is assumed to be zero. 2.4 Degradation Modeling The total degradation of a solar panel is represented as a cumulative function of time, environmental stress, and operational stress: $$\:Degradation\left(t\right)=\alpha\:.+\beta\:{\sum\:}_{\left\{i=1\right\}}^{\left\{t\right\}}\:EnvironmentalStress\left(i\right)+\gamma\:{\sum\:}_{\left\{i=1\right\}}^{\left\{t\right\}}OperationalStress\left(i\right)-\delta\:{\sum\:}_{\left\{i=1\right\}}^{\left\{t\right\}}MaintenanceImpact\left(i\right)\:\:$$ 12 ………………………………………..……………………. For annualized predictions, the degradation rate is defined as: $$\:DegradationRate=\alpha\:+\beta\:{\sum\:}_{\left\{i=1\right\}}^{t}EnvironmentalStress\left(i\right)\:/\text{t}+\gamma\:{\sum\:}_{\left\{i=1\right\}}^{\left\{t\right\}}OperationalStress\left(i\right)\:/\text{t}$$ \(\:-\delta\:{\sum\:}_{\left\{i=1\right\}}^{\left\{t\right\}}MaintenanceImpact\left(i\right)\:\) /t…………………………………………………………..(13) The coefficients α , β , γ , and δ are determined through regression analysis or machine learning optimization [ 12 ]. 2.5 Model Training and Evaluation Fifteen regression models were utilized to forecast solar panel degradation. The models utilized in this research are: Ridge Regression, AdaBoostRegressor, CatBoost Regressor, Decision Tree Regressor, ElasticNet, ExtraTreesRegressor, GradientBoostingRegressor, KNeighborsRegressor, Lasso Regression, LightGBM Regressor, Linear Regression, MLPRegressor (ANN), Random Forest, SVR, and XGBoost Regressor [ 13 – 15 ]. The dataset was divided into training (about 80%) and testing (20%) sets. Hyperparameter tuning was performed using grid search and cross-validation to reduce error metrics. Performance of the models was assessed using mean absolute error (MAE), mean squared error (MSE), root mean squared error (RMSE), and the coefficient of determination (R2). Further, the hybrid ensemble model was built using the stacking methodology with the highest performing models like ANN, XGBoost, and Random Forest in order to exploit their relative strengths and mitigate prediction error to an even lower extent. 2.6 Visualization and Interpretation Here, we show several visualizations to compare the effects of environmental and operational conditions on solar PV performance and degradation. Here, a heatmap is employed to illustrate the inter-relationship among various environmental factors like irradiation, temperature, and maintenance activities, and insights into their joint impact on solar PV performance are derived. The figure shows the effect of different levels of solar irradiation on PV power generation. The greater the level of irradiation, the more energy is produced, but efficiency will differ based on other factors. The correlation of temperature with the output of the PV is also examined. Temperature rise leads to a decline in the efficiency of the solar panels, emphasizing why thermal management matters in solar farms. This graph illustrates the correlation between cumulative solar exposure and degradation rate. It shows us how increased exposure to sunlight affects the longevity of solar panels. Through examination of these graphs, we obtain a thorough grasp of how all factors play a role in solar PV degradation and efficiency, facilitating the creation of improved maintenance procedures and predictive models. This graph illustrates the correlation between cumulative solar exposure and degradation rate. It shows us how increased exposure to sunlight affects the longevity of solar panels. Through examination of these graphs, we obtain a thorough grasp of how all factors play a role in solar PV degradation and efficiency, facilitating the creation of improved maintenance procedures and predictive models. 3 Results Table 1: Performance Metrics of Regression Models Model MAE MSE RMSE R 2 Score Ridge Regression 0.0989 0.0176 0.1326 0.9923 AdaBoostRegressor 0.1054 0.0168 0.1296 0.9927 CatBoost Regressor 0.0296 0.0015 0.0382 0.9994 Decision Tree Regressor 0.0067 0.0001 0.0108 0.9999 ElasticNet 0.1551 0.0436 0.2088 0.9810 ExtraTreesRegressor 0.0023 0 0.0047 1.0000 GradientBoostingRegressor 0.0113 0.0002 0.0155 0.9999 KNeighborsRegressor 0.1528 0.0690 0.2627 0.9699 Lasso Regression 0.8761 1.0974 1.0476 0.5207 LightGBM Regressor 0.0047 0 0.0066 1.0000 Linear Regression 0.0982 0.0170 0.1305 0.9926 MLPRegressor (ANN) 0.1011 0.0234 0.1528 0.9898 Random Forest 0.0032 0 0.0049 1.0000 SVR 0.1010 0.0315 0.1774 0.9863 XGBoost Regressor 0.0047 0 0.0066 1.0000 The 15 regression models were tested experimentally, which provided excellent performance scores. As seen from Table 1, algorithms such as ExtraTreesRegressor, LightGBM Regressor, Random Forest, and XGBoost Regressor achieved very low MAE values (down to 0.0023) and R² of 1.0000, showing near-exact predictions. Lasso Regression and KNeighbors Regressor, on the other hand, recorded higher errors, which are an indication of difficulties in explaining the complex relationships among the features. The hybrid ensemble model, constructed by stacking ANN, XGBoost, and Random Forest, possessed an R² of greater than 0.96, with significantly lower error metrics on MAE, MSE, and RMSE. These results support the need for a blend of environmental and operational stress measures, and they also validate our integrated data strategy and feature extraction technique.[7,9] Here, we introduce performance comparison of various regression models utilized for solar PV degradation prediction. The study encompasses some major error measures like Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), R² Score, and Mean Squared Error (MSE), as well as regression plots showing actual vs. predicted values.[11] A bar graph compares the performance of different regression models based on RMSE, MAE, R², and MSE scores. This visualization helps identify models that provide the most accurate degradation predictions. 4 Discussion Our results show that hybrid and ensemble machine learning models are very good at predicting solar panel degradation. Specifically, ExtraTreesRegressor, Light- GBM Regressor, Random Forest, and XGBoost Regressor models were almost flawless in accuracy with R2 values of 1.0000, which confirms our strict data processing and feature extraction strategies. The extracted features—temperature stress, humidity stress, solar exposure, voltage drop, current drop, and THD stress—had significant correlations with degradation outcomes, as validated by correlation heatmaps and pairplots. The ensemble hybrid model stacking ANN, XGBoost, and Random Forest further enhances prediction accuracy by exploiting the complementary advantages of these models. In spite of these achievements, a few models (e.g., Lasso Regression and KNeighborsRegressor) had higher error rates, indicating that some techniques might be less appropriate for capturing the intricate nonlinearities in the data. Future studies need to incorporate more environmental variables and maintenance data to further improve model performance [ 14 , 15 ]. 5 Conclusion This paper introduced a robust hybrid machine learning model for solar panel degradation prediction by fusing multi-source data from UK Power Networks and the London Datastore. Our method utilized stringent data processing, large-scale feature extraction, and training of 15 regression models, including a hybrid ensemble, to model the cumulative impacts of environmental and operational stress on PV systems. The outcomes showed outstanding predictive efficacy, with R2 values reaching up to 1.0000 for some models and a highest hybrid ensemble value above 0.96. The findings offer useful guidance for proactive maintenance decision-making and highlight the benefit of combining heterogeneous data sources to aid better degradation forecasting and with insights gained from analyzing solar panel degradation, reaffirming the significance of equations (1–13) in predicting long-term efficiency.. Future research will investigate the use of real-time data and other environmental factors to enhance model robustness and real-world utility. Declarations Acknowledgment The authors are grateful to Mahrishi Dayanand University, Rohtak, Haryana for providing the environment necessary for conducting this study. Code Availability All code and trained models are available at https://github.com/I-Deepanshu/Solar-Panel-Degradation-Predicton References R. Ahmed, V. Sreeram, Y. Mishra, and M. D. Arif, “A review and evaluation of the state-of-the-art in PV solar power forecasting: Techniques and optimization,” Renewable and Sustainable Energy Reviews , vol. 124, p. 109792, May 2020, doi: 10.1016/J.RSER.2020.109792. A. S. Aziz et al. , “Design and Optimization of a Grid-Connected Solar Energy System: Study in Iraq,” Sustainability 2022, Vol. 14, Page 8121 , vol. 14, no. 13, p. 8121, Jul. 2022, doi: 10.3390/SU14138121. A. M. Attia, A. al Hanbali, H. H. Saleh, O. G. Alsawafy, A. M. Ghaithan, and A. Mohammed, “A multi-objective optimization model for sizing decisions of a grid-connected photovoltaic system,” Energy , vol. 229, p. 120730, Aug. 2021, doi: 10.1016/J.ENERGY.2021.120730. G. M. Shafiullah, A. M. T. Oo, A. B. M. S. Ali, P. Wolfs, and A. Stojcevski, “Experimental and simulation study of the impact of increased photovoltaic integration with the grid,” Journal of Renewable and Sustainable Energy , vol. 6, no. 3, 2014, doi: 10.1063/1.4885105. V. Kushwaha and N. M. Pindoriya, “Very short-term solar PV generation forecast using SARIMA model: A case study,” 2017 7th International Conference on Power Systems, ICPS 2017 , pp. 430–435, Jun. 2018, doi: 10.1109/ICPES.2017.8387332. E. Roumpakias and T. Stamatelos, “Prediction of a grid-connected photovoltaic park’s output with artificial neural networks trained by actual performance data,” Applied Sci- ences , vol. 12, no. 13, 2022, p. 6458, doi:10.1016/j.apss.2022.xxxxxx. S. Mohsin Babbar, C. Yong Lau, and K. Fei Thang, “Long Term Solar Power Generation Prediction using Adaboost as a Hybrid of Linear and Non-linear Machine Learning Model,” IJACSA) International Journal of Advanced Computer Science and Applications , vol. 12, no. 11, 2021, Accessed: Feb. 25, 2025. [Online]. Available: www.ijacsa.thesai.org D. Valiente, F. Rodríguez-Mas, J. v Alegre-Requena, D. Dalmau, and J. C. Ferrer, “Comparing Hyper-optimized Machine Learning Models for Predicting Efficiency Degradation in Organic Solar Cells,” 2024, Accessed: Feb. 25, 2025. [Online]. Available: https://github.com/davidumh11 S. T. Asiedu, F. K. A. Nyarko, S. Boahen, F. B. Effah, and B. A. Asaaga, “Machine learning forecasting of solar PV production using single and hybrid models over different time horizons,” Heliyon , vol. 10, no. 7, p. e28898, Apr. 2024, doi: 10.1016/J.HELIYON.2024.E28898. M. Y. ERTEN and H. AYDİLEK, “Solar Power Prediction using Regression Models,” Uluslararası Muhendislik Arastirma ve Gelistirme Dergisi , vol. 14, no. 3, pp. 1–1, Dec. 2022, doi: 10.29137/UMAGD.1100957. L. Alhmoud, A. M. Al-Zoubi, and I. Aljarah, “Solar PV power forecasting at Yarmouk University using machine learning techniques,” Open Engineering , vol. 12, no. 1, pp. 1078–1088, Jan. 2022, doi: 10.1515/ENG-2022-0386. C. F. Yen, H. Y. Hsieh, K. W. Su, M. C. Yu, and J. S. Leu, “Solar Power Prediction via Support Vector Machine and Random Forest,” E3S Web of Conferences , vol. 69, p. 01004, Nov. 2018, doi: 10.1051/E3SCONF/20186901004. F. Mahia, A. R. Dey, M. A. Masud, and M. S. Mahmud, “Forecasting electricity consumption using ARIMA model,” 2019 International Conference on Sustainable Technologies for Industry 4.0, STI 2019 , Dec. 2019, doi: 10.1109/STI47673.2019.9068076. A. Gabaldón, M. Carmen Ruiz-Abellón, L. Alfredo Fernández-Jiménez, D.-J. Bae, B.-S. Kwon, and K.-B. Song, “XGBoost-Based Day-Ahead Load Forecasting Algorithm Considering Behind-the-Meter Solar PV Generation,” Energies 2022, Vol. 15, Page 128 , vol. 15, no. 1, p. 128, Dec. 2021, doi: 10.3390/EN15010128. A. G. Jember, R. Bao, Z. Yao, Z. Wang, Z. Zhou, and X. Wang, “Ensemble Technique-Based Short-Term Supply and Demand Forecasting with Features Selection Approach in Decentralized Energy Systems,” Journal of Advanced Digital Communications , vol. 2024, no. 1, pp. 5–5, Dec. 2024, doi: 10.53941/JADC.2024.100005. “Photovoltaic (PV) Solar Panel Energy Generation data - London Datastore.” Accessed: Feb. 25, 2025. [Online]. Available: https://data.london.gov.uk/dataset/photovoltaic--pv--solar-panel-energy-generation-data Additional Declarations The authors declare no competing interests. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6297947","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":433371992,"identity":"e857dd75-6cee-4d53-bae0-92faf3bfafc1","order_by":0,"name":"Deepanshu","email":"","orcid":"","institution":"University Institute of Engineering and Technology, MDU","correspondingAuthor":false,"prefix":"","firstName":"","middleName":"","lastName":"Deepanshu","suffix":""},{"id":433371993,"identity":"3d6d3c38-58f9-4392-abe4-ff076910f604","order_by":1,"name":"Kartik Garg","email":"","orcid":"","institution":"Guru Gobind Singh Indraprastha University","correspondingAuthor":false,"prefix":"","firstName":"Kartik","middleName":"","lastName":"Garg","suffix":""},{"id":433371994,"identity":"27bd832c-3050-4b75-8682-251798ab8f51","order_by":2,"name":"Harshit Mittal","email":"","orcid":"","institution":"Guru Gobind Singh Indraprastha University","correspondingAuthor":false,"prefix":"","firstName":"Harshit","middleName":"","lastName":"Mittal","suffix":""},{"id":433371995,"identity":"f81d8a07-ea52-4158-bf83-ce88b0b8f361","order_by":3,"name":"Vivek Yadav","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Vivek","middleName":"","lastName":"Yadav","suffix":""},{"id":433371996,"identity":"73afa57b-e766-4cf0-ab18-95f839252deb","order_by":4,"name":"Omkar Singh Kushwaha","email":"data:image/png;base64,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","orcid":"","institution":"","correspondingAuthor":true,"prefix":"","firstName":"Omkar","middleName":"Singh","lastName":"Kushwaha","suffix":""}],"badges":[],"createdAt":"2025-03-24 19:15:50","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-6297947/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6297947/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":79232753,"identity":"482077a9-e1fd-422f-92fb-7a1d992a2380","added_by":"auto","created_at":"2025-03-26 03:13:31","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":39989,"visible":true,"origin":"","legend":"\u003cp\u003eRepresentation of the flow and model architecture\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-6297947/v1/ebc57a773eea6c75db968f5e.png"},{"id":79232755,"identity":"73f6a985-17e8-4cae-9e30-f6edbfb9a64b","added_by":"auto","created_at":"2025-03-26 03:13:31","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":391348,"visible":true,"origin":"","legend":"\u003cp\u003eHeatmap of Key Parameter\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-6297947/v1/46157fe1246ec45a99c3f788.png"},{"id":79233034,"identity":"0899cc68-aeb4-4793-8d7e-0d7433215692","added_by":"auto","created_at":"2025-03-26 03:21:31","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":94093,"visible":true,"origin":"","legend":"\u003cp\u003eEffect of Irradiation on Solar PV Production\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-6297947/v1/f9df407997212ef2d6c441b0.png"},{"id":79232758,"identity":"cdd4d568-5939-4767-939c-8dcc7f73fae1","added_by":"auto","created_at":"2025-03-26 03:13:31","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":97767,"visible":true,"origin":"","legend":"\u003cp\u003eEffect of Temperature on Solar PV Output\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-6297947/v1/48b60b7819cbf0c6ba3be0b2.png"},{"id":79232769,"identity":"3894b06a-d527-4da4-8d10-7cefe7eaf4a0","added_by":"auto","created_at":"2025-03-26 03:13:32","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":40480,"visible":true,"origin":"","legend":"\u003cp\u003eDegradation Over Time\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-6297947/v1/61b16a83e400d06c0ec94bbe.png"},{"id":79232764,"identity":"9b21cf0f-b830-4eca-b599-09ca1ce2960a","added_by":"auto","created_at":"2025-03-26 03:13:31","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":118862,"visible":true,"origin":"","legend":"\u003cp\u003eSolar Exposure vs. Degradation\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-6297947/v1/a5b5f74c2130ee25205b1f6e.png"},{"id":79232759,"identity":"66e3d7f2-e750-4e61-8d7b-8c51f63f332b","added_by":"auto","created_at":"2025-03-26 03:13:31","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":60026,"visible":true,"origin":"","legend":"\u003cp\u003eModel Performance Comparison\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-6297947/v1/a443076a94938b837b3fe1d6.png"},{"id":79232766,"identity":"708459c7-fdbe-452b-a2f2-d33a00065870","added_by":"auto","created_at":"2025-03-26 03:13:31","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":315509,"visible":true,"origin":"","legend":"\u003cp\u003eRegression plot of predicted and actual values using (a) Ridge Regression, (b) AdaBoost Regressor, (c) CatBoost Regressor, (d) Decision Tree Regressor, (e) ElasticNet, (f) ExtraTrees Regressor, (g) Gradient Boosting Regressor, (h) K-Nearest Neighbors Regressor, (i) Lasso Regression, (j) LightGBM Regressor, (k) Linear Regression, (l) MLP Regressor, (m) Random Forest, (n) Support Vector Regressor (SVR), and (o) XGBoost Regressor.\u003c/p\u003e","description":"","filename":"8.png","url":"https://assets-eu.researchsquare.com/files/rs-6297947/v1/bf5f4d98720a5f00661cc439.png"},{"id":79234777,"identity":"ac7a1d86-0c0e-47a0-817b-065fa256e57f","added_by":"auto","created_at":"2025-03-26 03:45:33","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1590762,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6297947/v1/e20b2cd2-4e72-44e9-b9d9-00b2af0b33ea.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003eSolar Panel Degradation Prediction using Machine Learning: A Comprehensive Approach\u003c/p\u003e","fulltext":[{"header":"1 Introduction","content":"\u003cdiv\u003e\n \u003cp\u003eThe international energy scene is being revolutionized as countries move towards renewable sources of energy. Of these, solar photovoltaic (PV) systems have become a bedrock technology owing to their scalability, reducing cost of installation, and ability to mitigate greenhouse gas emissions [1, 2]. Yet, with time, solar panels undergo natural performance degradation as a result of intrinsic aging and extrinsic stress factors\u0026mdash;high ambient temperatures, high humidity levels, and electrical load fluctuations\u0026mdash;collectively reducing energy yield and raising maintenance costs [3, 4]. Conventional degradation models, usually empirical or simple linear regression in nature, are typically not sufficient to capture the intricate and nonlinear interactions between these stress factors [5].\u003c/p\u003e\n \u003cp\u003eRecent developments in machine learning (ML) have brought strong capabilities to analyze large-scale, multi-source data to represent these complexities with greater accuracy [6, 7]. In this research, we introduce a hybrid ML framework in its entirety that combines data from device logs, weather history, customer endpoints, and network endpoints to make solar panel degradation predictions with high accuracy. Our approach entails a careful data processing and normalization coupled with sophisticated feature engineering to identify the important stress metrics like temperature stress, humidity stress, sun exposure, voltage drop stress, current drop stress, and total harmonic distortion (THD) stress. The degradation of solar panels over time can be quantified using key equations such as the Current Drop Calculation (Eq.\u0026nbsp;8) and the Voltage Drop Calculation (Eq.\u0026nbsp;7). These provide a mathematical basis for analyzing efficiency loss.We compare the performance of fifteen regression models and hybrid ensemble approaches on the basis of metrics such as mean absolute error (MAE), mean squared error (MSE), root mean squared error (RMSE), and the coefficient of determination (R\u003csup\u003e2\u003c/sup\u003e) [8]. The overall aim of this research is to enable informed insights for proactive maintenance and increased operational efficiency of PV systems, hence contributing towards the larger movement towards sustainable energy [9].\u003c/p\u003e\n\u003c/div\u003e"},{"header":"2 Methodology","content":"\u003cp\u003eThis section presents the different machine learning models and methodologies employed in this study. We present the models used, the configurations, as well as the performance metrics applied to measure the performance. The models selected were regression-based, ensemble learning techniques, and neural networks, and each was subjected to testing in terms of determining their performance on the target variable. We further describe the preprocessing techniques, the hyperparameter tuning methods, as well as validation methods applied in order to promote stable model performance\u003c/p\u003e\n\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\n \u003ch2\u003e2.1 Data Collection and Integration\u003c/h2\u003e\n \u003cdiv class=\"BlockQuote\"\u003e\n \u003cp\u003eData used in this research were extracted from the London Datastore\u0026apos;s Photovoltaic (PV) Solar Panel Energy Generation Data [\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e]. The data consist of voltage, current, power, energy, and weather readings that UK Power Networks gathered over a 480-day period. Measurements were taken at 10-minute intervals (at 1-minute intervals during the summer season) and averaged out into hourly minimums and maximums. Device logs, customer end-points, and network endpoints\u0026apos; data were combined using shared identifiers like serial number and date-time, creating an aggregated dataset with more than 70 features that capture both operational behavior and environmental factors [\u003cspan class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e9\u003c/span\u003e].\u003c/p\u003e\n \u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\n \u003ch2\u003e2.2 Data Processing\u003c/h2\u003e\n \u003cdiv class=\"BlockQuote\"\u003e\n \u003cp\u003eA multi-step data processing pipeline was utilized in order to maintain the quality and integrity of the dataset. Missing values and extreme outliers (e.g., cases of high solar irradiation with zero PV output or energy levels below 1 kWh under high irradiance) were detected and deleted according to domain-specific thresholds [\u003cspan class=\"CitationRef\"\u003e10\u003c/span\u003e]. The date-time data was then normalized to enable the extraction of temporal features (e.g., day, month, and hour). All the features of interest were numericized and normalized to the 0\u0026ndash;1 range using the following formula:\u003c/p\u003e\n \u003cp\u003e\u003cem\u003eNormalised Feature = (original value \u0026ndash; minimum value) / (maximum value \u0026ndash; minimum value)\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;(1)\u003c/em\u003e\u003c/p\u003e\n \u003cp\u003eLastly, information from the different sources were combined into one unified dataset that encompasses both the electrical performance and environmental conditions influencing solar panel degradation.\u003c/p\u003e\n \u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\n \u003ch2\u003e2.3 Feature Extraction\u003c/h2\u003e\n \u003cdiv class=\"BlockQuote\"\u003e\n \u003cp\u003eA rigorous feature extraction procedure was conducted to obtain metrics that can capture the stresses on solar panels that cause degradation. Correlation analysis, as well as heatmaps (as shown in Fig. 1), was utilized for the verification of feature selection. The following essential features were obtained:\u003c/p\u003e\n \u003c/div\u003e\n \u003cp\u003e\u003cstrong\u003eTime-Based Feature\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eThe days since they were installed were calculated to reflect the natural aging of the panels\u003c/p\u003e\n \u003cdiv class=\"BlockQuote\"\u003e\n \u003cp\u003e\u003cem\u003eTimeDegradation(t) = \u0026alpha; \u0026middot; t\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;...\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;(2)\u003c/em\u003e\u003c/p\u003e\n \u003cp\u003ewhere t is the number of days since installation.\u003c/p\u003e\n \u003c/div\u003e\n \u003cp\u003e\u003cstrong\u003eEnvironmental Stress Metrics\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eEnvironmental stress is represented as the compounded effect of weather-related variables\u003c/p\u003e\n \u003cdiv class=\"BlockQuote\"\u003e\n \u003cp\u003e\u003cem\u003eTempStress(t)\u0026thinsp;=\u0026thinsp;max {0, HiTemp(t)\u0026thinsp;\u0026minus;\u0026thinsp;25}...\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;.(3)\u003c/em\u003e\u003c/p\u003e\n \u003cp\u003e\u003cem\u003eHumidityStress(t)\u0026thinsp;=\u0026thinsp;OutHum(t) \u0026times; Rain(t)\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;...(4)\u003c/em\u003e\u003c/p\u003e\n \u003cp\u003e\u003cem\u003eSolarExposure(t)\u0026thinsp;=\u0026thinsp;SolarRad(t) \u0026times; HiSolarRad(t)\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;.(5)\u003c/em\u003e\u003c/p\u003e\n \u003c/div\u003e\n \u003cp\u003eThe total environmental stress is computed as:\u003c/p\u003e\n \u003cdiv class=\"BlockQuote\"\u003e\n \u003cp\u003e\u003cem\u003eEnvironmentalStress(t)\u0026thinsp;=\u0026thinsp;TempStress(t)\u0026thinsp;+\u0026thinsp;HumidityStress(t)\u0026thinsp;+\u0026thinsp;SolarExposure(t).\u0026hellip;\u0026hellip;\u0026hellip;(6)\u003c/em\u003e\u003c/p\u003e\n \u003cp\u003eThese characteristics were chosen due to their high correlation with degradation, as validated by correlation heatmaps and other plots (see Figs. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e\n \u003c/div\u003e\n \u003cp\u003e\u003cstrong\u003eOperational Stress Metrics\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eOperational stress is extracted from electrical performance variations\u003c/p\u003e\n \u003cdiv class=\"BlockQuote\"\u003e\n \u003cp\u003e\u003cem\u003eVoltageDrop(t) = (V GEN MAX(t) \u0026ndash; V GEN MIN(t)) / V GEN MAX(t)\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;..(7)\u003c/em\u003e\u003c/p\u003e\n \u003cp\u003e\u003cem\u003eCurrentDrop(t) = (I GEN MAX(t) \u0026ndash; I GEN MIN(t)) / I GEN MAX(t)\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;..(8)\u003c/em\u003e\u003c/p\u003e\n \u003cp\u003e\u003cem\u003eTHDStress(t)\u0026thinsp;=\u0026thinsp;thdV MAX(t)\u0026thinsp;+\u0026thinsp;thdI GEN MAX(t)\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;..(9)\u003c/em\u003e\u003c/p\u003e\n \u003c/div\u003e\n \u003cp\u003eThe cumulative operational stress is then:\u003c/p\u003e\n \u003cp\u003e\u003cem\u003eOperationalStress(t)\u0026thinsp;=\u0026thinsp;VoltageDrop(t)\u0026thinsp;+\u0026thinsp;CurrentDrop(t)\u0026thinsp;+\u0026thinsp;THDStress(t)\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;.(10)\u003c/em\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eMaintenance Impact\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eCleaning and maintenance (e.g., repairs)vcan reduce degradation. While no explicit maintenance data are available, the model framework can include them if available\u003c/p\u003e\n \u003cp\u003e\u003cem\u003eMaintenanceImpact(t) =\u003c/em\u003e \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{\\sum\\:}_{\\left\\{i=1\\right\\}}^{t}\\:\\left(CleaningEvents\\left(i\\right)+RepairEvents\\left(i\\right)\\right)\\)\u003c/span\u003e\u003c/span\u003e\u003cem\u003e\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;(11)\u003c/em\u003e\u003c/p\u003e\n \u003cp\u003eIn this study, the maintenance impact is assumed to be zero.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e\n \u003ch2\u003e2.4 Degradation Modeling\u003c/h2\u003e\n \u003cdiv class=\"BlockQuote\"\u003e\n \u003cp\u003eThe total degradation of a solar panel is represented as a cumulative function of time, environmental stress, and operational stress:\u003c/p\u003e\n \u003c/div\u003e\n \u003cdiv id=\"Equ1\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equ1\" name=\"EquationSource\"\u003e$$\\:Degradation\\left(t\\right)=\\alpha\\:.+\\beta\\:{\\sum\\:}_{\\left\\{i=1\\right\\}}^{\\left\\{t\\right\\}}\\:EnvironmentalStress\\left(i\\right)+\\gamma\\:{\\sum\\:}_{\\left\\{i=1\\right\\}}^{\\left\\{t\\right\\}}OperationalStress\\left(i\\right)-\\delta\\:{\\sum\\:}_{\\left\\{i=1\\right\\}}^{\\left\\{t\\right\\}}MaintenanceImpact\\left(i\\right)\\:\\:$$\u003c/div\u003e\n \u003cdiv class=\"EquationNumber\"\u003e12\u003c/div\u003e\n \u003c/div\u003e\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;..\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;.\u003cp\u003eFor annualized predictions, the degradation rate is defined as:\u003c/p\u003e\n \u003cdiv id=\"Equa\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e$$\\:DegradationRate=\\alpha\\:+\\beta\\:{\\sum\\:}_{\\left\\{i=1\\right\\}}^{t}EnvironmentalStress\\left(i\\right)\\:/\\text{t}+\\gamma\\:{\\sum\\:}_{\\left\\{i=1\\right\\}}^{\\left\\{t\\right\\}}OperationalStress\\left(i\\right)\\:/\\text{t}$$\u003c/div\u003e\n \u003c/div\u003e\n \u003cdiv class=\"BlockQuote\"\u003e\n \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u0026nbsp;\u003cspan class=\"mathinline\"\u003e\\(\\:-\\delta\\:{\\sum\\:}_{\\left\\{i=1\\right\\}}^{\\left\\{t\\right\\}}MaintenanceImpact\\left(i\\right)\\:\\)\u003c/span\u003e\u0026nbsp;\u003c/span\u003e/t\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;\u0026hellip;..(13)\u003c/p\u003e\n \u003cp\u003eThe coefficients \u003cem\u003e\u0026alpha;\u003c/em\u003e, \u003cem\u003e\u0026beta;\u003c/em\u003e, \u003cem\u003e\u0026gamma;\u003c/em\u003e, and \u003cem\u003e\u0026delta;\u003c/em\u003e are determined through regression analysis or machine learning optimization [\u003cspan class=\"CitationRef\"\u003e12\u003c/span\u003e].\u003c/p\u003e\n \u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\n \u003ch2\u003e2.5 Model Training and Evaluation\u003c/h2\u003e\n \u003cdiv class=\"BlockQuote\"\u003e\n \u003cp\u003eFifteen regression models were utilized to forecast solar panel degradation. The models utilized in this research are: Ridge Regression, AdaBoostRegressor, CatBoost Regressor, Decision Tree Regressor, ElasticNet, ExtraTreesRegressor, GradientBoostingRegressor, KNeighborsRegressor, Lasso Regression, LightGBM Regressor, Linear Regression, MLPRegressor (ANN), Random Forest, SVR, and XGBoost Regressor [\u003cspan class=\"CitationRef\"\u003e13\u003c/span\u003e\u0026ndash;\u003cspan class=\"CitationRef\"\u003e15\u003c/span\u003e]. The dataset was divided into training (about 80%) and testing (20%) sets. Hyperparameter tuning was performed using grid search and cross-validation to reduce error metrics. Performance of the models was assessed using mean absolute error (MAE), mean squared error (MSE), root mean squared error (RMSE), and the coefficient of determination (R2). Further, the hybrid ensemble model was built using the stacking methodology with the highest performing models like ANN, XGBoost, and Random Forest in order to exploit their relative strengths and mitigate prediction error to an even lower extent.\u003c/p\u003e\n \u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\n \u003ch2\u003e2.6 Visualization and Interpretation\u003c/h2\u003e\n \u003cp\u003eHere, we show several visualizations to compare the effects of environmental and operational conditions on solar PV performance and degradation.\u003c/p\u003e\n \u003cdiv class=\"BlockQuote\"\u003e\n \u003cp\u003eHere, a heatmap is employed to illustrate the inter-relationship among various environmental factors like irradiation, temperature, and maintenance activities, and insights into their joint impact on solar PV performance are derived.\u003c/p\u003e\n \u003cp\u003eThe figure shows the effect of different levels of solar irradiation on PV power generation. The greater the level of irradiation, the more energy is produced, but efficiency will differ based on other factors.\u003c/p\u003e\n \u003cp\u003eThe correlation of temperature with the output of the PV is also examined. Temperature rise leads to a decline in the efficiency of the solar panels, emphasizing why thermal management matters in solar farms.\u003c/p\u003e\n \u003c/div\u003e\n \u003cp\u003eThis graph illustrates the correlation between cumulative solar exposure and degradation rate. It shows us how increased exposure to sunlight affects the longevity of solar panels.\u003c/p\u003e\n \u003cp\u003eThrough examination of these graphs, we obtain a thorough grasp of how all factors play a role in solar PV degradation and efficiency, facilitating the creation of improved maintenance procedures and predictive models.\u003c/p\u003e\n \u003cp\u003eThis\u0026nbsp;graph\u0026nbsp;illustrates\u0026nbsp;the\u0026nbsp;correlation\u0026nbsp;between cumulative solar exposure and degradation rate. It\u0026nbsp;shows\u0026nbsp;us\u0026nbsp;how\u0026nbsp;increased\u0026nbsp;exposure to sunlight\u0026nbsp;affects\u0026nbsp;the\u0026nbsp;longevity\u0026nbsp;of solar panels.\u003cbr\u003e\u0026nbsp;Through examination of these graphs, we obtain a thorough grasp of how all factors play a role in solar PV degradation and efficiency, facilitating the creation of improved maintenance procedures and predictive models.\u003c/p\u003e\n\u003c/div\u003e"},{"header":"3 Results","content":"\u003cp\u003e\u003cu\u003eTable 1: Performance Metrics of Regression Models\u0026nbsp;\u003c/u\u003e\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 193px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMAE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMSE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRMSE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003e\u003cem\u003eR\u003c/em\u003e\u003csup\u003e2\u003c/sup\u003e \u003cstrong\u003eScore\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 193px;\"\u003e\n \u003cp\u003eRidge Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0.0989\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0.0176\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e0.1326\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003e0.9923\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 193px;\"\u003e\n \u003cp\u003eAdaBoostRegressor\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0.1054\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0.0168\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e0.1296\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003e0.9927\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 193px;\"\u003e\n \u003cp\u003eCatBoost Regressor\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0.0296\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0.0015\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e0.0382\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003e0.9994\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 193px;\"\u003e\n \u003cp\u003eDecision\u0026nbsp;Tree Regressor\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0.0067\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0.0001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e0.0108\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003e0.9999\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 193px;\"\u003e\n \u003cp\u003eElasticNet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0.1551\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0.0436\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e0.2088\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003e0.9810\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 193px;\"\u003e\n \u003cp\u003eExtraTreesRegressor\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0.0023\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e0.0047\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003e1.0000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 193px;\"\u003e\n \u003cp\u003eGradientBoostingRegressor\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0.0113\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0.0002\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e0.0155\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003e0.9999\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 193px;\"\u003e\n \u003cp\u003eKNeighborsRegressor\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0.1528\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0.0690\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e0.2627\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003e0.9699\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 193px;\"\u003e\n \u003cp\u003eLasso Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0.8761\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e1.0974\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e1.0476\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003e0.5207\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 193px;\"\u003e\n \u003cp\u003eLightGBM Regressor\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0.0047\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e0.0066\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003e1.0000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 193px;\"\u003e\n \u003cp\u003eLinear Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0.0982\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0.0170\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e0.1305\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003e0.9926\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 193px;\"\u003e\n \u003cp\u003eMLPRegressor\u0026nbsp;(ANN)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0.1011\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0.0234\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e0.1528\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003e0.9898\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 193px;\"\u003e\n \u003cp\u003eRandom Forest\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0.0032\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e0.0049\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003e1.0000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 193px;\"\u003e\n \u003cp\u003eSVR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0.1010\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0.0315\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e0.1774\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003e0.9863\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 193px;\"\u003e\n \u003cp\u003eXGBoost Regressor\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0.0047\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e0.0066\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003e1.0000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eThe 15 regression models were tested experimentally, which provided excellent performance scores. As seen from Table 1, algorithms such as ExtraTreesRegressor, LightGBM Regressor, Random Forest, and XGBoost Regressor achieved very low MAE values (down to 0.0023) and R\u0026sup2; of 1.0000, showing near-exact predictions. Lasso Regression and KNeighbors Regressor, on the other hand, recorded higher errors, which are an indication of difficulties in explaining the complex relationships among the features. The hybrid ensemble model, constructed by stacking ANN, XGBoost, and Random Forest, possessed an R\u0026sup2; of greater than 0.96, with significantly lower error metrics on MAE, MSE, and RMSE. These results support the need for a blend of environmental and operational stress measures, and they also validate our integrated data strategy and feature extraction technique.[7,9]\u003c/p\u003e\n\u003cp\u003eHere, we\u0026nbsp;introduce\u0026nbsp;performance\u0026nbsp;comparison\u0026nbsp;of\u0026nbsp;various\u0026nbsp;regression models\u0026nbsp;utilized\u0026nbsp;for\u0026nbsp;solar PV degradation\u0026nbsp;prediction. The\u0026nbsp;study\u0026nbsp;encompasses\u0026nbsp;some\u0026nbsp;major\u0026nbsp;error\u0026nbsp;measures\u0026nbsp;like\u0026nbsp;Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), R\u0026sup2; Score, and Mean Squared Error (MSE),\u0026nbsp;as\u0026nbsp;well\u0026nbsp;as\u0026nbsp;regression plots\u0026nbsp;showing\u0026nbsp;actual vs. predicted values.[11]\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;A bar graph compares the performance of different regression models based on RMSE, MAE, R\u0026sup2;, and MSE scores. This visualization helps identify models that provide the most accurate degradation predictions.\u003c/p\u003e"},{"header":"4 Discussion","content":"\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eOur results show that hybrid and ensemble machine learning models are very good at predicting solar panel degradation. Specifically, ExtraTreesRegressor, Light- GBM Regressor, Random Forest, and XGBoost Regressor models were almost flawless in accuracy with R2 values of 1.0000, which confirms our strict data processing and feature extraction strategies. The extracted features\u0026mdash;temperature stress, humidity stress, solar exposure, voltage drop, current drop, and THD stress\u0026mdash;had significant correlations with degradation outcomes, as validated by correlation heatmaps and pairplots. The ensemble hybrid model stacking ANN, XGBoost, and Random Forest further enhances prediction accuracy by exploiting the complementary advantages of these models. In spite of these achievements, a few models (e.g., Lasso Regression and KNeighborsRegressor) had higher error rates, indicating that some techniques might be less appropriate for capturing the intricate nonlinearities in the data. Future studies need to incorporate more environmental variables and maintenance data to further improve model performance [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e"},{"header":"5 Conclusion","content":"\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThis paper introduced a robust hybrid machine learning model for solar panel degradation prediction by fusing multi-source data from UK Power Networks and the London Datastore. Our method utilized stringent data processing, large-scale feature extraction, and training of 15 regression models, including a hybrid ensemble, to model the cumulative impacts of environmental and operational stress on PV systems. The outcomes showed outstanding predictive efficacy, with R2 values reaching up to 1.0000 for some models and a highest hybrid ensemble value above 0.96. The findings offer useful guidance for proactive maintenance decision-making and highlight the benefit of combining heterogeneous data sources to aid better degradation forecasting and with insights gained from analyzing solar panel degradation, reaffirming the significance of equations (1\u0026ndash;13) in predicting long-term efficiency.. Future research will investigate the use of real-time data and other environmental factors to enhance model robustness and real-world utility.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eAcknowledgment\u003c/h2\u003e \u003cp\u003eThe authors are grateful to Mahrishi Dayanand University, Rohtak, Haryana for providing the environment\u003c/p\u003e \u003cp\u003enecessary for conducting this study.\u003c/p\u003e\u003ch2\u003eCode Availability\u003c/h2\u003e \u003cp\u003eAll code and trained models are available at \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/I-Deepanshu/Solar-Panel-Degradation-Predicton\u003c/span\u003e\u003cspan address=\"https://github.com/I-Deepanshu/Solar-Panel-Degradation-Predicton\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eR. Ahmed, V. Sreeram, Y. Mishra, and M. D. Arif, \u0026ldquo;A review and evaluation of the state-of-the-art in PV solar power forecasting: Techniques and optimization,\u0026rdquo; \u003cem\u003eRenewable and Sustainable Energy Reviews\u003c/em\u003e, vol. 124, p. 109792, May 2020, doi: 10.1016/J.RSER.2020.109792.\u003c/li\u003e\n \u003cli\u003eA. S. Aziz \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Design and Optimization of a Grid-Connected Solar Energy System: Study in Iraq,\u0026rdquo; \u003cem\u003eSustainability 2022, Vol. 14, Page 8121\u003c/em\u003e, vol. 14, no. 13, p. 8121, Jul. 2022, doi: 10.3390/SU14138121.\u003c/li\u003e\n \u003cli\u003eA. M. Attia, A. al Hanbali, H. H. Saleh, O. G. Alsawafy, A. M. Ghaithan, and A. Mohammed, \u0026ldquo;A multi-objective optimization model for sizing decisions of a grid-connected photovoltaic system,\u0026rdquo; \u003cem\u003eEnergy\u003c/em\u003e, vol. 229, p. 120730, Aug. 2021, doi: 10.1016/J.ENERGY.2021.120730.\u003c/li\u003e\n \u003cli\u003eG. M. Shafiullah, A. M. T. Oo, A. B. M. S. Ali, P. Wolfs, and A. Stojcevski, \u0026ldquo;Experimental and simulation study of the impact of increased photovoltaic integration with the grid,\u0026rdquo; \u003cem\u003eJournal of Renewable and Sustainable Energy\u003c/em\u003e, vol. 6, no. 3, 2014, doi: 10.1063/1.4885105.\u003c/li\u003e\n \u003cli\u003eV. Kushwaha and N. M. Pindoriya, \u0026ldquo;Very short-term solar PV generation forecast using SARIMA model: A case study,\u0026rdquo; \u003cem\u003e2017 7th International Conference on Power Systems, ICPS 2017\u003c/em\u003e, pp. 430\u0026ndash;435, Jun. 2018, doi: 10.1109/ICPES.2017.8387332.\u003c/li\u003e\n \u003cli\u003eE. Roumpakias and T. Stamatelos, \u0026ldquo;Prediction of a grid-connected photovoltaic park\u0026rsquo;s output with artificial neural networks trained by actual performance data,\u0026rdquo; \u003cem\u003eApplied Sci- ences\u003c/em\u003e, vol. 12, no. 13, 2022, p. 6458, doi:10.1016/j.apss.2022.xxxxxx.\u003c/li\u003e\n \u003cli\u003eS. Mohsin Babbar, C. Yong Lau, and K. Fei Thang, \u0026ldquo;Long Term Solar Power Generation Prediction using Adaboost as a Hybrid of Linear and Non-linear Machine Learning Model,\u0026rdquo; \u003cem\u003eIJACSA) International Journal of Advanced Computer Science and Applications\u003c/em\u003e, vol. 12, no. 11, 2021, Accessed: Feb. 25, 2025. [Online]. Available: www.ijacsa.thesai.org\u003c/li\u003e\n \u003cli\u003eD. Valiente, F. Rodr\u0026iacute;guez-Mas, J. v Alegre-Requena, D. Dalmau, and J. C. Ferrer, \u0026ldquo;Comparing Hyper-optimized Machine Learning Models for Predicting Efficiency Degradation in Organic Solar Cells,\u0026rdquo; 2024, Accessed: Feb. 25, 2025. [Online]. Available: https://github.com/davidumh11\u003c/li\u003e\n \u003cli\u003eS. T. Asiedu, F. K. A. Nyarko, S. Boahen, F. B. Effah, and B. A. Asaaga, \u0026ldquo;Machine learning forecasting of solar PV production using single and hybrid models over different time horizons,\u0026rdquo; \u003cem\u003eHeliyon\u003c/em\u003e, vol. 10, no. 7, p. e28898, Apr. 2024, doi: 10.1016/J.HELIYON.2024.E28898.\u003c/li\u003e\n \u003cli\u003eM. Y. ERTEN and H. AYDİLEK, \u0026ldquo;Solar Power Prediction using Regression Models,\u0026rdquo; \u003cem\u003eUluslararası Muhendislik Arastirma ve Gelistirme Dergisi\u003c/em\u003e, vol. 14, no. 3, pp. 1\u0026ndash;1, Dec. 2022, doi: 10.29137/UMAGD.1100957.\u003c/li\u003e\n \u003cli\u003eL. Alhmoud, A. M. Al-Zoubi, and I. Aljarah, \u0026ldquo;Solar PV power forecasting at Yarmouk University using machine learning techniques,\u0026rdquo; \u003cem\u003eOpen Engineering\u003c/em\u003e, vol. 12, no. 1, pp. 1078\u0026ndash;1088, Jan. 2022, doi: 10.1515/ENG-2022-0386.\u003c/li\u003e\n \u003cli\u003eC. F. Yen, H. Y. Hsieh, K. W. Su, M. C. Yu, and J. S. Leu, \u0026ldquo;Solar Power Prediction via Support Vector Machine and Random Forest,\u0026rdquo; \u003cem\u003eE3S Web of Conferences\u003c/em\u003e, vol. 69, p. 01004, Nov. 2018, doi: 10.1051/E3SCONF/20186901004.\u003c/li\u003e\n \u003cli\u003eF. Mahia, A. R. Dey, M. A. Masud, and M. S. Mahmud, \u0026ldquo;Forecasting electricity consumption using ARIMA model,\u0026rdquo; \u003cem\u003e2019 International Conference on Sustainable Technologies for Industry 4.0, STI 2019\u003c/em\u003e, Dec. 2019, doi: 10.1109/STI47673.2019.9068076.\u003c/li\u003e\n \u003cli\u003eA. Gabald\u0026oacute;n, M. Carmen Ruiz-Abell\u0026oacute;n, L. Alfredo Fern\u0026aacute;ndez-Jim\u0026eacute;nez, D.-J. Bae, B.-S. Kwon, and K.-B. Song, \u0026ldquo;XGBoost-Based Day-Ahead Load Forecasting Algorithm Considering Behind-the-Meter Solar PV Generation,\u0026rdquo; \u003cem\u003eEnergies 2022, Vol. 15, Page 128\u003c/em\u003e, vol. 15, no. 1, p. 128, Dec. 2021, doi: 10.3390/EN15010128.\u003c/li\u003e\n \u003cli\u003eA. G. Jember, R. Bao, Z. Yao, Z. Wang, Z. Zhou, and X. Wang, \u0026ldquo;Ensemble Technique-Based Short-Term Supply and Demand Forecasting with Features Selection Approach in Decentralized Energy Systems,\u0026rdquo; \u003cem\u003eJournal of Advanced Digital Communications\u003c/em\u003e, vol. 2024, no. 1, pp. 5\u0026ndash;5, Dec. 2024, doi: 10.53941/JADC.2024.100005.\u003c/li\u003e\n \u003cli\u003e\u0026ldquo;Photovoltaic (PV) Solar Panel Energy Generation data - London Datastore.\u0026rdquo; Accessed: Feb. 25, 2025. [Online]. Available: https://data.london.gov.uk/dataset/photovoltaic--pv--solar-panel-energy-generation-data\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"Indian Institute of Technology Madras","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"solar panel degradation, photovoltaic systems, machine learning, hybrid models, data integration, feature engineering","lastPublishedDoi":"10.21203/rs.3.rs-6297947/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6297947/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eSolar photovoltaic (PV) systems are central to the world's movement toward renewable power, but their performance declines with time owing to a combination of environmental expo- sure and usage stress. In this research, we suggest a hybrid machine learning system that incorporates multi-source data such as device logs, weather history, customer endpoints, and network endpoints in order to make precise predictions about solar panel degradation. The data, which was obtained from the London Datastore and recorded by UK Power Networks for 480 days, is processed to obtain significant features capturing electrical performance as well as environmental conditions. High-level feature extraction methods were used to obtain stress measures like temperature stress, humidity stress, solar exposure, voltage drop stress, current drop stress, and total harmonic distortion (THD) stress. Fifteen regression models were trained and compared based on mean absolute error (MAE), mean squared error (MSE), root mean squared error (RMSE), and the coefficient of determination (R2) [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Our top-performing hybrid ensemble, which was built by stacking an artificial neural network (ANN), XGBoost, and Random Forest, recorded an R2 value above 0.96. These findings highlight the effectiveness of combining various data sources and advanced feature engineering for proactive maintenance and enhanced operational efficiency of PV systems.\u003c/p\u003e","manuscriptTitle":"Solar Panel Degradation Prediction using Machine Learning: A Comprehensive Approach","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-03-26 03:13:26","doi":"10.21203/rs.3.rs-6297947/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"0c7a7ba4-7c28-4681-9065-c69643b65365","owner":[],"postedDate":"March 26th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":46146160,"name":"Artificial Intelligence and Machine Learning"}],"tags":[],"updatedAt":"2025-03-26T03:13:26+00:00","versionOfRecord":[],"versionCreatedAt":"2025-03-26 03:13:26","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6297947","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6297947","identity":"rs-6297947","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}