Machine Learning for Solar Radiation and Photovoltaic Systems: Evidence, Validation, and Operational Applications

Danjuma Maiwada, Abubakar; Ahmad, Mardiana Idayu; Abba, Sani; Shakir, Mohammad Aliff

doi:10.65582/aifsc.2026.014

Artificial Intelligence for Sustainable Cities

Volume 1, 2026, Article 2026.014

Review

Machine Learning for Solar Radiation and Photovoltaic Systems: Evidence, Validation, and Operational Applications

, , ,

¹Environmental Technology Division, School of Industrial Technology, Universiti Sains Malaysia, 11800, Penang, Malaysia

²Department of Civil Engineering, Prince Mohammad Bin Fahd University, Al Khobar, 31952, Saudi Arabia

Published: 19 October 2026
Pages: 253-282
Received: 21 July 2026
Revised: 16 October 2026
Accepted: 16 October 2026
DOI: 10.65582/aifsc.2026.014
Access: Open access, CC BY 4.0

Published 19 October 2026Pages 253-282doi.org/10.65582/aifsc.2026.014

Received 21 July 2026Revised 16 Oct 2026Accepted 16 Oct 2026Open accessunder CC BY 4.0

Download PDF30 pages · 5.8 MB

Check for updates

Abstract

Machine learning is used in solar-energy forecasting and management, but reported gains depend on the task, data and evaluation design. This review combines descriptive Scopus bibliometric mapping with a critical narrative synthesis of applications in irradiance estimation, photovoltaic power prediction, fault diagnosis, energy management and grid integration. The photovoltaic search yielded 1,954 unique records published between 2010 and 2026; a separate solar-energy search returned 6,767 results. Their VOSviewer maps are interpreted independently. We examine technical studies by prediction target, forecast horizon, input data, climate, validation method, benchmark and operational purpose. Results from different sites, targets and test designs cannot establish a universal ranking of model families. Classical, ensemble and deep-learning methods each have useful applications, but their reported accuracy should be interpreted against appropriate baselines and leakage-free temporal tests. External-site evaluation and calibrated uncertainty are essential when forecasts inform operational decisions. Physics-guided approaches may improve consistency, whereas digital twins, graph learning, federated learning and autonomous control require more task-specific validation. The review identifies transparent search reporting, reproducible data processing and cross-climate testing as priorities. Because the complete search history and technical screening records are unavailable, the synthesis is descriptive rather than systematic, and bibliometric counts do not measure model quality.

Keywords. Machine learning; Solar radiation Prediction; Deep learning; Smart grid integration; Digital twin

1. Introduction

Solar energy (SE) is central to decarbonisation, but its operational value depends on accurate characterisation and prediction of the solar resource. Global horizontal irradiance (GHI), direct normal irradiance (DNI), and diffuse horizontal irradiance (DHI) determine the energy available to PV, concentrating solar power (CSP), concentrating PV, and building-integrated systems. Clouds, aerosols, atmospheric composition, and local meteorology make these quantities strongly variable in space and time, creating uncertainty for energy-yield assessment and system operation (Firoozi et al. 2025; Jia et al. 2021). As PV penetration expands across utility, distributed, and urban energy systems, resource forecasting must increasingly support storage scheduling, market participation, demand response, and grid-support functions (Morey et al. 2023). The problem extends beyond weather variability. Sensor noise, calibration errors, missing data, geographical differences, partial shading, soiling, temperature effects, ageing, and inverter behaviour introduce uncertainty from measurement to final AC power. Purely empirical models often lack transferability, whereas physics-only models may be computationally demanding or unable to capture site-specific nonlinearities. ML and deep learning (DL) can fuse ground measurements, satellite observations, numerical weather prediction (NWP), imagery, and operational data, but reported performance remains sensitive to data quality, validation design, and deployment context (Guo et al. 2025; Paletta et al. 2023). Research has consequently progressed from support vector machines (SVM), random forests (RF), and artificial neural networks (ANN) to ensembles, convolutional and recurrent networks, transformers, metaheuristic hybrids, and physics-informed learning. Recent reviews already address computer-vision solar forecasting (Paletta et al. 2023) and physics–ML hybrid PV prediction (Santos et al. 2024). The present synthesis connects those strands to prediction targets, forecast horizons, validation and operational decisions. This fragmentation obscures how model choice, radiation component, data source, forecast horizon, and operational objective interact across the solar-energy value chain (Hamza and Modu 2025). As illustrated in Figure 1, advanced ML connects solar-radiation data with forecasting, optimisation, and intelligent energy management. This review combines descriptive Scopus bibliometric maps with a selective critical technical synthesis of ML for solar-resource and solar-energy management. It: (a) maps publication growth, collaboration, and thematic evolution; (b) compares classical, ensemble, deep, hybrid, and emerging learning paradigms; (c) links GHI, DNI, and DHI modelling to PV forecasting, MPPT, fault diagnosis, farm optimisation, load management, and grid integration; and (d) identifies research priorities in physical consistency, transfer learning, uncertainty quantification, interpretability, and operational deployment. The aim is not to identify a universally superior algorithm, but to clarify which modelling strategies are most defensible under different data, climate, and decision contexts. Table 1 compares the scope of this review with two relevant reviews.

**Fig. 1.** Graphical overview of advanced machine learning in solar energy Full size

Table 1. Scope of this review relative to two relevant recent reviews.

Review	Emphasis	Distinction in this article
Paletta et al. (2023)	Computer vision for solar forecasting	Connects forecast evidence to PV and downstream decisions.
Santos et al. (2024)	Hybrid physics–ML PV prediction	Appraises target, horizon and validation across model families.
Present review	Bibliometric maps and selective technical synthesis	Uses a common data–target–horizon–validation–purpose framework.

2. Review Approach and Bibliographic Scope

The scope and organisation of the review are presented in Figure 2. The evidence base was organised around three linked questions: how ML methods for solar modelling have evolved, which model families are most appropriate for different solar-energy tasks, and which limitations still prevent robust operational deployment. Early studies established the value of nonlinear data-driven models for radiation estimation. SVM and ANN improved estimation where direct measurements were unavailable or meteorological inputs were incomplete (Assi et al. 2010; Chen et al. 2011; Gastón et al. 2013). Subsequent work expanded the problem from static estimation to short-term forecasting by combining weather classes, fuzzy logic, time-delay networks, and SVM-based predictors (Chen et al. 2013; Detyniecki et al. 2012; Wu and Chan 2012). The literature then shifted toward multi-source forecasting and comparative benchmarking. NWP post-processing, transmissivity-based features, sky cameras, ensemble trees, and hybrid optimisation improved prediction across different horizons, while comparative studies repeatedly showed that performance depends on climate, data resolution, and forecast lead time rather than on a single universally dominant model (Diagne et al. 2013; Perez et al. 2013; Prakash and Singh 2014; West et al. 2014; Zeng and Qiao 2013). Table 2 summarises selected 2015–2025 studies as examples; it is not the complete bibliometric corpus. The synthesis compares studies within their own target, data, horizon and validation settings rather than pooling noncomparable errors.

**Fig. 2.** Analytical framework used to distinguish data sources, prediction targets, model families, validation evidence and operational decisions. Full size

Table 2. Selected machine-learning studies for solar-energy applications (2015–2025); examples, not a systematically screened corpus

Authors	Country	Models Used / Methods	Performance Criteria / Indicators	Results / Remarks
Piri et al. (2015)	Iran	SVR models with Polynomial and RBF kernels for daily SR estimation at two Iranian stations.	Comparison of modeled vs measured daily radiation.	Polynomial SVR performed best for Bojnurd, while RBF SVR was best for Zahedan; SVR successfully captured nonlinear climate-dependent radiation patterns.
Ramli et al. (2015)	Saudi Arabia	SVM vs ANN models to predict SR on tilted PV surfaces in Jeddah and Qassim.	RMSE, Correlation Coefficient (CC), Magnitude of Relative Error (MRE), and computation time.	SVM outperformed ANN with higher CC (0.918-0.976), lower MRE, and shorter training/testing time, showing better robustness for tilted-surface radiation prediction.
Prema and Rao (2015)	India	Time-series models: Moving Average (MA), Simple Exponential Smoothing (SES), Double Exponential Smoothing (DES), and Decomposition for 1-day-ahead SI	Forecast error (%), similar to MAPE.	Decomposition model using 2 months of data achieved the lowest error (9.28%), outperforming MA, SES, and DES for short-term irradiance forecasting.
Almeida et al. (2015)	Spain	Nonparametric Quantile Regression Forests (QRF) for day-ahead PV power forecasting using WRF-NWP predictors.	Cv MBE, prediction (quantile) intervals, validation on multiple PV plants.	QRF achieved cv MBE < 1.3% and provided accurate point forecasts plus well-calibrated uncertainty bands, making it suitable for operational PV forecasting.
Wu and Wang (2016)	China	Hybrid SOM-OPELM neural model with multi-step strategies (Recursive, Dir Rec, MISMO) for SR prediction; compared to BP and ARIMA at three sites.	RMSE, MAE, MAPE, MBE for multiple horizons and locations.	SOM-OPELM with Dir Rec or MISMO strategies consistently produced lower errors than BP, Recursive SOM-OPELM, and ARIMA, particularly for multi-step forecasts, demonstrating strong accuracy and stability.
Belaid and Mellit (2016)	Algeria	SVM regression with RBF kernel for daily and monthly GSR in an arid climate using simple inputs (Tmax, Tmin, Tmean, Tdiff, S0, H0).	NRMSE, RMSE, MAPE, MBE, correlation coefficient (R); comparison with ANN and empirical models.	Best SVM configurations achieved NRMSE = 13%, MAPE = 10-11%, R = 0.89-0.90 and outperformed ANN and several empirical models, confirming SVM’s effectiveness in arid-region solar estimation.
Alzahrani et al. (2017)	USA	Deep Recurrent Neural Network (DRNN-LSTM) for short-term SI forecasting; compared with Feedforward NN (FNN) and SVR.	RMSE, MBE for training and testing.	DRNN-LSTM achieved the lowest testing RMSE (0.086 vs SVR 0.11 and FNN 0.16), showing superior ability to capture temporal dependencies and outperform shallow NN and SVM models.
Sivaneasan et al. (2017)	Singapore	ANN with fuzzy-logic pre-processing and adaptive error-correction for 5-minute-ahead solar forecasting.	MAPE for pure ANN, ANN+fuzzy, and ANN+fuzzy+error-correction.	ANN with fuzzy logic and error-correction reduced MAPE to 29.6%, improving accuracy by 16.7% over pure ANN and 13.5% over ANN+fuzzy, and giving more consistent performance under rapidly changing weather.
Moosa et al. (2018)	India	ANN, RF, and GB models for hourly, +24 h, and +48 h GHI forecasting using atmospheric, NWP, and astronomical features.	Comparative error plots / rankings for GHI prediction.	RF and GB provided the most accurate and stable GHI forecasts; ANN showed higher variability and dependence on initialization.
Yang et al. (2018)	USA	Benchmark of 68 ML/statistical models (RF, GB, SVR variants, MLP, GP, LASSO, kNN, etc.) for 1-hour-ahead GHI forecasting across 7 locations and 5 climate zones.	nRMSE, nMBE; daily and two-year overall accuracy; climate-specific performance; training time.	Tree-based ensemble models (RF, GB, XGB) consistently achieved strong long-term performance; no single universal best model was found, with preferred models depending on climate and site; trade-offs between accuracy and training cost were highlighted.
Yousif et al. (2019)	Oman	MLR forecasting models using multi-boundary scoring (2-4 boundaries) for long-term air temperature prediction (1991-2015).	R², Prob > F, significance tests, model diagnostics.	Models attained high R² (0.9335-0.9977); forecasts closely matched observed temperature trends, suggesting the approach is reliable for temperature-based renewable-energy and planning studies.
Kisi et al. (2020)	Turkey	Bayesian Model Averaging (BMA) vs ANN, ELM, RBF, WANN, WELM, WRBF for SR estimation in a Mediterranean climate.	RMSE, NSE, R² across models.	BMA achieved the highest accuracy, reducing RMSE by 16-57% relative to other models; wavelet-hybrid ANN/ELM/RBF variants improved performance compared to their standalone counterparts.
Tharani et al. (2020)	India	Linear Regression, Neural Network Regression, RF, Extra Trees for renewable (solar) energy forecasting.	RMSE, MAE (implied), comparison of predicted vs actual curves.	Random Forest and Extra Trees achieved the best prediction accuracy; Linear Regression and SVR underperformed, while NN regression showed competitive results but slightly below tree ensembles.
Jebli et al. (2021)	Morocco	Pearson correlation-based feature selection combined with LR, RF, SVR, and ANN for real-time and short-term SE prediction.	MAE, MSE, RMSE, R² for various horizons and models.	RF and ANN delivered the highest accuracy for both real-time and short-term forecasting; LR and SVR produced noticeably larger errors. Model generalization was confirmed using an additional site in Brazil (Pirapora).
Ağbulut et al. (2021)	Turkey	ANN, SVM, k-NN, and DL for daily GSR prediction in four provinces representing different solar regimes.	R², RMSE, rRMSE, MBE, MABE, MAPE, t-stat.	All ML models performed well (R² 0.855-0.936), but ANN was the best-fitting algorithm, followed by DL and SVM; k-NN had the weakest performance. ANN exceeded t-critic thresholds and provided the most reliable predictions.
Gao et al. (2022)	Japan	Deep generative LSTM model for multi-step (≥24 h) SI prediction using historical measurements and temperature weather forecasts in Tokyo.	Comparison of multi-step prediction errors vs conventional regression LSTM under different horizons and forecast lead times.	Generative LSTM mitigated error accumulation and improved accuracy by 7.7% over regression LSTM; incorporating 1-day-ahead temperature forecasts improved accuracy substantially, while older or daily-averaged temperature forecasts gave limited benefit.
Sansine et al. (2022)	French	Hybrid PSO-optimized models (PSO-XGBoost, PSO-LSTM, PSO-GBRT) for day-ahead hourly SI forecasting using AROME NWP plus historical data; probabilistic forecasts via residual-based quantile mapping.	Deterministic metrics (e.g., RMSE, MAE) and probabilistic metrics using 38%, 68%, 95%, 99% prediction intervals.	PSO-LSTM achieved the best overall deterministic and probabilistic performance among all tested models (ANN, CNN, RF, LSTM, GBRT, XGBoost and their PSO variants), delivering accurate day-ahead forecasts and well-calibrated prediction intervals suitable for microgrid operation.
Ledmaoui et al. (2023)	Morocco	Comparison of ML models (SVR, ANN, DT, RF, GAM, XGBoost) for daily SE production forecasting using one year of PV plant data in Benguerir.	RMSE, MAE, MASE, R² (train/test).	ANN emerged the most effective model, with the lowest RMSE and MASE and also with the highest R². In this study, optimal ANN architecture identified for deployment in PV operational forecasting and maintenance planning.
Chakraborty et al. (2023)	India	Ensemble learning (Bagging, Boosting, Stacking, Voting) plus classical regressors for PV power prediction using SRRA meteorological data and a 10 kWp rooftop plant; integrated feature-selection/data-mining pipeline.	R², RMSE and related regression error measures over 7 months of field data.	Stacking and Voting ensembles achieved =96% prediction accuracy, outperforming individual base learners; the proposed framework generalized well and was recommended for larger PV plants in similar climates.
Chaaban and Alfadl (2024)	Saudi Arabia	Comparative ML study (ANN, RF, LSTM, k-NN) for PV energy yield prediction using historical PV output, meteorological, and SI data from a large industrial PV plant.	MAE, RMSE under different input configurations (meteorological-only vs meteorological + irradiation).	Random Forest consistently outperformed ANN, LSTM, and k-NN; with meteo-only inputs it achieved MAE = 0.06 and RMSE = 0.15, and using meteo+irradiation improved performance further (MAE = 0.03, RMSE = 0.09), offering robust, stable PV forecasts.
Tanoli et al. (2024)	Pakistan / Saudi Arabia	ML models (LR, DT, R, FFNN, LSTM) for surface SR prediction over 45 glaciers in KPK and Gilgit using CERES data (2001-2021).	MSE, MAE, nRMSE, R² for each model and region.	Feed-forward neural network (FFNN) gave the best performance (nRMSE = 0.07-0.08, R² = 0.89-0.92), outperforming LR, DT, RF, and LSTM; results are useful for renewable planning and glacier/climate impact assessment.
Reddy et al. (2025)	India	Global-scale solar potential assessment using multiple ML models with XGBoost selected for final deployment; inputs include geospatial, meteorological, infrastructure layers and CMIP6 climate projections (1980-2099).	RMSE, MAE for SE potential; validation in case-study regions (Mindanao, Gobi-Altai, Peloponnese).	XGBoost outperformed alternative models, achieving RMSE of 0.97 kWh/m² and MAE of 0.76 kWh/m². It also provides long-term, climate-informed solar potential and techno-economic indicators for rapid site screening and planning.
Pourmir and Miri (2025)	Iran	Weather-based SE forecasting using statistical tests (Kolmogorov-Smirnov, correlation), PCA for dimensionality reduction, and weighted linear regression, boosted trees, CNN, LSTM, CNN-LSTM, and Temporal Convolutional Network (TCN).	Training/testing errors (%), parameter count, computation time.	TCN delivered the best balance of accuracy and complexity, with lowest errors and fewer parameters than CNN-LSTM and other deep models; CNN-LSTM also performed strongly but at higher computational cost. TCN was recommended as an efficient, adaptable forecasting model across locations.

2.1. Database Selection and Evidence Sets

This review combines bibliometric mapping with a critical narrative discussion of machine learning in solar-energy systems. Scopus provided two related, separately analysed bibliographic sets. The photovoltaic search used “Machine Learning AND Photovoltaics” in the article title, abstract and keywords fields. Its export contains 1,954 distinct Scopus EIDs from 2010–2026: 1,143 articles, 606 conference papers, 146 reviews and 59 other records. Two records are marked “Retracted” and are retained in that descriptive total but not used as technical evidence. A broader search for “Machine Learning AND Solar Energy” returned 6,767 results. Because the two searches cover different, potentially overlapping populations, we interpret their VOSviewer maps separately and do not combine their counts. The 2026 entries represent a partial publication year and are not treated as a complete annual trend. The technical synthesis draws on relevant published studies of irradiance and photovoltaic forecasting, fault detection, energy management and grid applications; it is not a screened subset of either bibliometric set. We interpret reported results by target variable, forecast horizon, input source, climate, sample size, chronological validation, prevention of preprocessing leakage, benchmark choice, uncertainty assessment and testing at another site or period. These criteria guide the discussion rather than constitute a numerical quality score or a formal risk-of-bias assessment. The retained materials do not establish the exact search dates, full executed queries, filters, duplicate-removal log or screening decisions. VOSviewer version 1.6.20 is visible in the analysis records, but the original counting and normalisation settings, thesaurus and native network files are unavailable for verification. We therefore present descriptive maps and a selective technical synthesis, without a PRISMA flow or claims of systematic-review completeness. These constraints limit independent reproduction and cross-study comparisons.

2.2. Descriptive Bibliometric Results

Figure 3 displays the original country and keyword visualisations for the two separate searches. The keyword maps show solar forecasting and machine-learning terms alongside PV, grid-management and meteorological terms. These are qualitative readings of the displayed maps, not evidence of model superiority, thematic evolution or operational adoption. The earlier analysis reported 16,018 distinct keywords, 1,626 with at least five occurrences and selection of 1,000 by total link strength. The underlying term list and VOSviewer processing log were unavailable, so these intermediate counts and threshold choices could not be independently verified from the photovoltaic csv. The photovoltaic country-selection screenshot lists the five countries with greatest displayed total link strength (TLS): China (475 documents; TLS 245), Saudi Arabia (123; 227), United States (289; 194), United Kingdom (104; 130) and India (278; 121). Their TLS values total 917, giving the percentages in the lower pie of Figure 3 after rounding: 26.7%, 24.8%, 21.2%, 14.2% and 13.2%, respectively. These percentages describe the combined TLS of only those five selected countries, not shares of publications. The upper panel represents the separate solar-energy search and is not pooled with the photovoltaic data. A co-authorship link reflects collaboration within a retrieved set; it does not establish research quality or model performance.

**Fig. 3.** Original VOSviewer country and keyword maps for the separate machine-learning-and-solar-energy (upper) and machine-learning-and-photovoltaics (lower) searches. The pie percentages are each of five selected countries’ combined total link strength, not shares of all publications; the underlying VOSviewer artwork is unchanged. Full size

3. Solar Energy Resources and Management: An Overview

3.1. Solar Radiation Types

Solar-resource information supports decisions from long-term site selection and capacity planning to short-term dispatch, storage scheduling, and grid balancing. Figure 4 presents a visual overview of the solar energy resources and management framework adopted in this study. ML is particularly useful because it can integrate ground observations, satellite products, NWP outputs, remote-sensing indices, and historical operating data while representing nonlinear atmospheric variability (Sulaiman and Mustaffa 2024; Sun et al. 2022; Liu et al. 2022). For practical solar-energy modelling, the principal surface components are GHI, DNI, and DHI, whose physical meaning and measurement constraints determine the appropriate target, input data, and downstream application (Liu et al. 2022).

**Fig. 4.** Visualized overview of SERM used in this study Full size

At the surface, solar radiation is modified by absorption, scattering, reflection, and solar geometry. The three principal components satisfy the horizontal-plane relationship $G H I = D H I + D N I cos (θ_{z})$ , where theta z is the solar zenith angle. This identity provides a useful physical-consistency check for measurements and component-wise ML predictions (Liu et al. 2022).

3.1.1. Global Horizontal Irradiance (GHI)

GHI is the total shortwave radiation received on a horizontal surface and is the most widely measured target for PV resource assessment. It combines the horizontal projection of the direct beam with diffuse radiation and is commonly measured by pyranometers or obtained from satellite and reanalysis products. Because GHI responds strongly to cloud and aerosol variability, it is central to PV yield estimation and short-term power forecasting (Liu et al. 2022; Lauret et al. 2024).

3.1.2. Direct Normal Irradiance (DNI)

DNI is the direct beam received by a surface normal to the Sun and is the key resource for CSP and concentrating PV. It is measured with a sun-tracking pyrheliometer and is highly sensitive to clouds, aerosols, and solar geometry. Long-term DNI observations are less widely available than GHI, so ML, satellite retrievals, and decomposition models are often used to estimate it for dispatch and thermal-storage planning (Paletta et al. 2023; Lee 2025).

3.1.3. Diffuse Horizontal Irradiance (DHI)

DHI is the sky-scattered component received on a horizontal surface after excluding the direct beam. It can be measured with a shaded pyranometer or derived from GHI and DNI. DHI becomes especially important under cloudy, polluted, shaded, and urban conditions, and the diffuse fraction is frequently used to classify sky states. Its stronger dependence on scattering processes makes it difficult to estimate, motivating ML models that use cloud, aerosol, and image-derived information while preserving the GHI-DNI-DHI relationship (Rajagukguk et al. 2021; Rumbayan et al. 2012).

3.2. Key Challenges in Solar Energy Resources and Management

Solar-energy management is constrained by a coupled atmosphere-measurement-technology problem. Rapid weather transitions and cloud motion create short-term ramps that historical time series or coarse NWP fields may miss (Juncklaus Martins et al. 2022). Ground sensors introduce calibration drift, soiling, shading, and data gaps, with DNI and DHI particularly affected by sparse measurement coverage (Daniel-Durandt and Rix 2023; Esposito et al. 2024). Models also face geographical shift because irradiance statistics differ across maritime, arid, monsoon, mountain, and urban environments (Kumar and Suzuki 2019). Finally, the irradiance-to-power relationship is nonlinear because of cell temperature, spectral response, partial shading, soiling, ageing, mismatch, and inverter controls (Rahman et al. 2023). These interacting sources of uncertainty limit model transferability and motivate quality control, domain adaptation, uncertainty-aware learning, and physically consistent prediction (Liu et al. 2022; Dai et al. 2025; Shata et al. 2025).

3.3. Solar Energy Management Tasks

SE management spans forecasting, control, diagnostics, optimisation, demand coordination, and grid support (Singh et al. 2024). These tasks share data but differ in time scale and decision objective. Forecasting estimates irradiance or PV power from seconds to days ahead; MPPT seeks the instantaneous operating point that maximizes module output; fault diagnosis detects soiling, hotspots, degradation, inverter failures, and wiring anomalies; farm optimisation coordinates trackers, inverters, maintenance, curtailment, storage, and dispatch; demand response aligns flexible loads with variable PV supply; and grid-integration tools support ramp-rate control, voltage regulation, reserve planning, and stability assessment. ML contributes by learning nonlinear relationships, extracting patterns from images and SCADA streams, and converting forecasts into adaptive control or maintenance decisions. The main value therefore lies not in prediction alone, but in linking prediction to operational action.

3.3.1. PV Power Forecasting

PV power forecasting combines irradiance, meteorological, NWP, image, and historical power data to support balancing, reserve allocation, storage scheduling, market participation, and plant operation. LSTM, CNN, and ensemble models are widely used because they can represent nonlinear temporal and spatial variability (Ansong et al. 2025; Das et al. 2025).

3.3.2. Maximum Power Point Tracking (MPPT)

ML-based MPPT learns the relationship between environmental and electrical variables and the optimal operating point, improving adaptation under rapid irradiance changes and partial shading compared with fixed-rule methods. Its practical role is to reduce oscillation and power loss while adapting to temperature, ageing, and changing module conditions (Guanghua et al. 2025; Douli et al. 2025).

3.3.3. PV Fault Diagnosis

For fault diagnosis, ML analyses electrical measurements, I-V curves, thermal or electroluminescence images, and SCADA streams to detect and classify anomalies and to support predictive maintenance. This approach is more scalable than manual inspection for large PV fleets, although performance depends on representative fault data and robust validation (Kull et al. 2025; Taghezouit et al. 2024).

3.3.4. Solar Farm Control and Optimization

At solar-farm scale, ML supports tracker and inverter optimisation, soiling and maintenance decisions, curtailment, storage dispatch, and economic operation. These applications require models that balance energy yield with equipment health and grid or market constraints rather than maximizing prediction accuracy alone (Adekanbi et al. 2024; Polleux et al. 2022).

3.3.5. Load Management and Demand Response

Load forecasting and demand response use ML to shift flexible demand, including HVAC, industrial loads, and electric-vehicle charging, toward periods of high PV availability. Better coordination reduces peaks, curtailment, and storage stress, but requires models that account for user behaviour as well as generation uncertainty (Firoozi et al. 2025; Banad et al. 2025).

3.3.6. Grid Stability Integration

For grid stability, ML supports early warning, reserve scheduling, voltage and frequency control, ramp-rate management, and coordinated PV-battery operation. These uses are particularly important in weak or isolated grids, where rapid PV fluctuations can have system-level consequences (Polleux et al. 2022).

4. Machine Learning Techniques used in Solar Energy

ML methods in solar energy range from interpretable classical learners to data-intensive deep architectures and hybrid physics-data frameworks. Model selection should reflect data volume, input structure, forecast horizon, uncertainty requirements, computational resources, and deployment constraints rather than algorithm popularity (Santos et al. 2024). Figure 5 distils the principal standalone, ensemble, deep, hybrid, and emerging approaches reviewed here.

**Fig. 5.** Standalone models, deep learning, ensemble, hybrid and emerging techniques used in SERM application Full size

4.1. Classical Machine Learning Methods

Classical methods remain strong baselines for structured meteorological and operational datasets. ANN provide flexible nonlinear approximation; SVM/SVR are effective for small-to-medium datasets with carefully tuned kernels; decision trees are interpretable but unstable as single learners; RF improve robustness through bagging and feature randomization; GPR provides predictive distributions and explicit uncertainty; and fuzzy systems or ANFIS incorporate rule-based reasoning. Their lower data and computational requirements often make them attractive for site-specific applications and embedded systems (Mohammed et al. 2025).

4.1.1. Artificial Neural Networks (ANN)

ANN map meteorological, temporal, and historical inputs to irradiance or PV power through nonlinear hidden layers. They are versatile and widely used for forecasting, missing-data reconstruction, and component estimation, but require careful regularisation and hyperparameter control to avoid overfitting and unstable results (André et al. 2024).

4.1.2. Support Vector Machines (SVM)

SVM/SVR use kernel functions to represent nonlinear relationships while controlling model complexity. They have performed well for radiation estimation and PV power prediction when datasets are not very large, although accuracy depends on kernel and hyperparameter selection and standard formulations do not directly provide probabilistic uncertainty (Zhang 2025).

4.1.3. Decision Trees and Random Forests (DT and RF)

Decision trees represent predictions as interpretable split rules but can overfit and vary substantially with the training sample. RF reduce this variance by averaging many trees trained on bootstrapped data and random feature subsets. RF handle nonlinear interactions, mixed features, and variable-importance analysis effectively, which explains their strong performance in many irradiance and PV forecasting benchmarks (Sun et al. 2024).

4.1.4. Gaussian Process Regression (GPR)

GPR is a nonparametric probabilistic method that returns both a mean prediction and predictive uncertainty through a chosen covariance kernel. This makes it useful for risk-aware forecasting and scheduling, especially when data are moderate in size and uncertainty estimates are operationally important (Najibi et al. 2021).

4.1.5. Fuzzy Logic (FL) and ANFIS

Fuzzy logic represents uncertain inputs through linguistic membership functions and transparent if-then rules, while ANFIS learns fuzzy-system parameters from data. These methods are useful for MPPT, control, and moderate-sized forecasting problems where interpretability matters, but model complexity can grow rapidly as the number of inputs and rules increases (Júnior et al. 2023; Sharkawy and Bekhiti 2026).

4.2. Ensemble Learning (EL)

Ensemble learning combines multiple predictors to reduce bias, variance, or model-specific error. Bagging and RF stabilize high-variance learners; boosting methods such as XGBoost, LightGBM, and CatBoost sequentially correct residual error; and stacking uses a meta-learner to combine heterogeneous models. For structured solar datasets, ensembles often provide an effective accuracy-computation compromise and can outperform single ANN or SVR models when feature quality is high (Ramírez-Rivera and Guerrero-Rodríguez 2024; Anonto et al. 2025).

4.3. Deep Learning (DL) Architectures

DL is most advantageous when large, high-resolution, sequential, spatial, or image-based datasets are available. LSTM networks model temporal dependence in irradiance and power series; CNN extract spatial features from sky or satellite imagery; CNN-LSTM hybrids integrate spatial and temporal information; and attention-based transformers can model long sequences and multi-step dependencies. Their benefits, however, must be weighed against training cost, data requirements, and reduced interpretability (Barancsuk et al. 2025; Seman et al. 2026).

4.4. Hybrid and Optimization-Enhanced Models

Hybrid models improve learning by combining optimisation, decomposition, or physical knowledge with ML. Metaheuristics such as GA and PSO tune network weights, kernels, hyperparameters, or feature subsets; wavelet, STL, and EMD decompositions separate multiscale or seasonal structure before prediction; and statistical-ML hybrids combine complementary assumptions. Physics-informed ML is especially promising because radiative, electrical, or energy-balance relations can be embedded as constraints or regularisation, improving consistency and potentially reducing extrapolation error (Santos et al. 2024). Figure 6 delineates the structural frameworks of the PSO and GA algorithms.

**Fig. 6.** Structure of PSO and GA algorithms Full size

4.5. Emerging Methods

Emerging methods shift the field from isolated forecasting toward distributed and adaptive energy intelligence. RL learns control policies for PV-battery dispatch, demand response, and MPPT; GNN represent spatial and electrical relationships among PV strings, inverters, feeders, or microgrids; federated learning may enable collaborative training across sites under stated privacy and communication assumptions; and digital twins may couple live measurements with virtual system models for monitoring, prediction, and what-if analysis. The reviewed evidence offers application-specific examples for reinforcement learning (Fang et al. 2025) and urban graph modelling (Li et al. 2025); broad deployment benefits of federated or agentic systems remain unestablished here. Safety, communication, cyber-security, validation and computation should be assessed separately. The emerging trends in SE applications using GNN and DT technologies are presented in Table 3.

Table 3. Emerging Trends for SE applications using GNN and DT Technology

Authors	Country / Case Study	Models Used	Performance Indicators	Results / Remarks
Li et al. (2025)	China / Hong Kong / USA (New York City)	Shadow-Attention Graph Neural Network (SAGNN), GNN	RMSE, R², spatial resolution (1 m), PV yield estimation	Proposed a novel SAGNN that explicitly models inter-building shading effects. Achieved high-resolution urban PV irradiation prediction over 1.08 million buildings.
Tripathi and Pandey (2025)	India	CatBoost + GA, DE, Bayesian Optimization + DT	RMSE, MAE, R²	Introduced an integrated forecasting framework combining ML, evolutionary optimization, and Digital Twin (DT) correction.
Olayiwola et al. (2025)	United Kingdom	DT frameworks integrated with AI/ML, IoT, autonomous systems	Qualitative benchmarking, system reliability, operational scalability	Provided a comprehensive review of Digital Twin concepts for PV systems. Highlighted DTs’ role in predictive maintenance, fault diagnosis, and real-time monitoring.

No single model is universally best. Classical learners remain competitive for moderate structured datasets, ensemble, deep and physics-guided methods have different data and validation requirements. Their relative performance cannot be ranked across heterogeneous studies. Model comparison should therefore consider accuracy, uncertainty, robustness, transferability, interpretability, and deployment cost together. Figure 7 illustrates key emerging trends in SERM through Agentic AI in panel (a) and Digital Twin technology in panel (b).

**Fig. 7.** Emerging Trend in SERM (a) Agentic AI (b) Digital Twin in SERM Full size

5. Machine Learning Applications in Solar Energy

ML now supports the full solar-energy value chain, from resource estimation to grid operation. The most important distinction across applications is the decision being supported: a model suitable for minute-ahead nowcasting may be inappropriate for long-term site assessment or real-time control. Figure 8 integrates these applications.

**Fig. 8.** ML applications in SERM Full size

5.1. Solar Radiation (SR) Forecasting

SR forecasting predicts GHI, DNI, or DHI across horizons from minutes to days. Very short-term methods benefit from sky cameras, satellite imagery, cloud motion, and recent measurements; intraday forecasts often combine temporal ML with meteorological inputs; and day-ahead forecasts increasingly post-process NWP outputs. LSTM, CNN, ensembles, decomposition hybrids, and metaheuristic-tuned models can improve accuracy, but fair comparison requires climate-specific baselines, persistence benchmarks, and consistent forecast horizons. The strongest studies treat forecasting as an uncertainty problem rather than reporting point-error metrics alone.

5.2. PV Power Output Prediction

PV power prediction adds module temperature, system design, degradation, inverter behaviour, shading, and soiling to irradiance uncertainty. Direct power models can learn the full weather-to-power mapping, while staged models first forecast irradiance and then convert it to power. RF and ensembles are often effective for structured plant data, whereas LSTM and hybrid models are useful when temporal dependence is strong. Operationally useful models should remain reliable during ramps, rare weather, curtailment, and equipment changes rather than only under average conditions.

5.3. Solar Farm Management and Optimization

Solar-farm management extends prediction into action. ML can identify abnormal energy loss, prioritize cleaning and maintenance, forecast component failure, optimize trackers and inverter loading, coordinate storage, and support curtailment or dispatch decisions. RL and digital-twin approaches can further integrate forecasting and control, but deployment requires explicit constraints, safe fallback behaviour, and evaluation against economic and reliability outcomes rather than prediction error alone.

5.4. Solar Resource Assessment (SRA)

Solar resource assessment increasingly combines ground observations with satellite, reanalysis, GIS, and climate-projection data. ML can correct bias, fill spatial gaps, downscale coarse products, and rank candidate sites. For long-term planning, transferability and climate non-stationarity are more important than small gains on a single historical test set; validation should therefore span multiple locations, seasons, and where possible future climate conditions.

5.5. Load Forecasting (LF) and Smart Grid Integration

In smart grids and urban energy systems, ML-based load forecasts can be coordinated with PV forecasts, batteries, flexible buildings, and electric-vehicle charging. The resulting problem is inherently multivariate and decision-oriented: forecast errors affect reserve needs, curtailment, comfort, electricity cost, and network constraints. Future systems therefore require joint probabilistic forecasting and control rather than independently optimising solar generation and demand.

6. Comparative Analysis of Machine Learning Models

Comparing solar ML models requires more than ranking a single error statistic. Performance depends on target variable, forecast horizon, temporal resolution, climate, sample size, feature set, validation strategy, and benchmark choice. Table 4 compares model families and the evidence needed to interpret them for a specified target, horizon and operational purpose.

Table 4. Model families and evidence needed for a valid solar-energy comparison

Model family	Typical input and use	Report in a fair comparison	Interpretive limit
SVR and Gaussian processes Zeng and Qiao (2013); Najibi et al. (2021)	Moderate structured meteorological records; irradiance or PV power	Same-site temporal test, tuned baseline, data volume and uncertainty if claimed	Kernel choice and scalability affect results.
Random forests and boosting Chakraborty et al. (2023)	Tabular weather and plant data; short-term power or resource estimation	Persistence or physical baseline, site and climate, horizon, missing-data protocol	Trees may extrapolate poorly beyond observed conditions.
Recurrent and convolutional networks Ansong et al. (2025); Barancsuk et al. (2025)	Sequential measurements, sky images or satellite fields	Training size, compute, chronological split, image availability, baseline	Larger error reductions at one site do not establish transfer.
Hybrid and physics-guided models Liu et al. (2022); Santos et al. (2024)	Joint physical variables and observational data	Physical-closure check plus independent forecast skill and ablation	A physically consistent prediction may still be inaccurate.
Graph and control methods Fang et al. (2025); Li et al. (2025)	Spatial shading, networks, batteries and dispatch states	External-site or operational test, safety constraints, decision cost	Application examples do not prove general deployment benefit.

6.1. Performance Metrics

Common metrics include MAE and RMSE for absolute error, MAPE or normalized RMSE for scale-independent comparison, R² or correlation for association, and bias measures for systematic over- or under-prediction. No metric is sufficient alone: RMSE emphasises large errors, MAPE can be unstable near zero, and high R² does not guarantee low operational error. For real deployment, studies should also report uncertainty calibration, inference time, model size, robustness to missing or shifted data, and performance during high-impact ramps. Evaluation should use temporally separated testing and, where relevant, external-site validation rather than random splits that leak seasonal information.

6.2. Model Strengths and Weaknesses

Model strengths are application-specific. SVM and GPR are attractive for smaller datasets, with GPR adding uncertainty estimates but scaling poorly to very large samples. RF and boosting methods are robust on structured data and offer useful feature importance, although tree ensembles may struggle to extrapolate beyond the training range. ANN and LSTM capture nonlinear temporal behaviour but require careful tuning and adequate data. CNN and transformers are powerful for imagery and long sequences but are computationally intensive. Hybrid and physics-informed models can improve accuracy and consistency, yet additional complexity is justified only when gains persist under external validation. Interpretability, maintainability, and computational cost should therefore be treated as model-selection criteria, not afterthoughts.

6.3. Benchmark Datasets

Reliable benchmarking depends on accessible, quality-controlled datasets with clear metadata. NREL and other public solar archives support reproducibility, but many published studies still use private single-site datasets, inconsistent preprocessing, and different train-test periods. Progress would be accelerated by benchmark suites spanning climates, forecast horizons, sensor types, PV technologies, and operational disturbances, together with standard persistence, physical, and ML baselines. Such benchmarks should include missing data, sensor drift, and distribution shift rather than only clean historical records.

7. Research Trends in Machine Learning Applications for Solar Energy

Across the selected technical studies, solar ML evolved through four broad phases: early ANN and SVM estimation; wider use of RF, boosting, and systematic model comparison; rapid adoption of LSTM, CNN, imagery, and hybrid decomposition after the mid-2010s; and, more recently, physics-informed learning, transformers, digital twins, federated learning, GNN, and RL. The central trend is a shift from isolated point prediction toward integrated, multi-source, uncertainty-aware, and decision-coupled systems. At the same time, the literature increasingly recognizes that gains reported at one site do not guarantee cross-climate or operational robustness.

8. Challenges and Research Gaps in Machine Learning-Based Solar Energy Systems

Five research gaps remain decisive. First, data quality and representativeness are inconsistent, particularly for DNI, DHI, faults, and extreme weather. Second, geographical and temporal distribution shift limits transfer across climates, plants, and changing operating conditions. Third, many studies optimize average prediction error while neglecting uncertainty, ramp events, rare failures, and economic consequences. Fourth, complex DL and hybrid models often lack interpretability, reproducibility, and fair comparison with strong baselines. Fifth, the transition from retrospective modelling to real-time deployment remains weak because latency, cyber-security, communication failures, maintenance, and safe control are rarely evaluated. Addressing these gaps requires open benchmarks, external validation, uncertainty-aware metrics, interpretable modelling, and closer integration between atmospheric physics, PV engineering, and operational decision-making.

9. Future Direction

Future research should prioritize models that remain reliable under changing climates, locations, sensors, and system configurations. Physics-informed and hybrid learning can constrain predictions with radiative and electrical knowledge; transfer and domain adaptation can reduce the need for complete retraining; probabilistic and conformal methods can communicate forecast uncertainty; and federated learning can support multi-site learning while preserving data ownership. Digital twins and RL can connect forecasting with maintenance and control, while GNN can represent networked PV and urban energy infrastructures. The critical methodological shift is from maximizing benchmark accuracy to demonstrating generalisation, physical consistency, calibrated uncertainty, computational feasibility, and measurable operational value.

10. Policy Implications and Practical Recommendations

Policy should support the data and governance infrastructure required for trustworthy AI-enabled solar systems. Public agencies and grid operators should promote interoperable solar, weather, and operational datasets; standardize quality control and metadata; and require transparent reporting of validation protocols and uncertainty. Investment in open benchmarks and shared testbeds would reduce duplicated effort and improve comparison across regions. For cities, planning frameworks should link rooftop and building-integrated PV forecasts with storage, flexible demand, electric mobility, and distribution-network constraints so that AI supports system-level decarbonisation rather than isolated asset optimisation. Regulation should also address model accountability. Grid codes and procurement rules for AI-enabled inverters, forecasting services, digital twins, and autonomous energy-management systems should define minimum requirements for robustness, cyber-security, human oversight, fallback operation, and post-deployment monitoring. Capacity building is equally important: regions with high solar potential but limited data infrastructure need investment in sensing, data stewardship, and technical skills to avoid widening the digital divide in renewable-energy deployment.

Limitations of the review. No auditable PRISMA flow, independent dual screening or complete extraction file was supplied. Consequently, the 1,954 PV export records and the 6,767-result solar-energy screenshot describe bibliometric searches, not the number of studies critically reviewed. Their retrieval counts cannot establish an eligible-study count or a duplicate-removal total. Performance conclusions remain conditional on the study designs and reported benchmarks.

11. Conclusions

This review combines two descriptive bibliometric maps with a selective technical discussion of ML in solar radiation and PV operations. No single algorithm is supported as universally superior: irradiance and PV-power studies differ in target, climate, data source, forecast horizon and validation. Directly comparable studies support task-specific insights, including the value of imagery in short-range irradiance forecasting (Ansong et al. 2025), physics-informed features and interpretation (Liu et al. 2022), and probabilistic benchmarking (Lauret et al. 2024). Broad claims of ensemble or deep-learning superiority do not follow from heterogeneous reported errors. The main evidential limitation is that the technical studies were not drawn from a fully documented systematic selection and were not formally scored for leakage, sample size, external-site testing, calibration or code availability. The exact bibliometric retrieval dates and full native files are unavailable for independent reconstruction, and 2026 is incomplete. The review therefore treats cross-climate transfer, physical closure, uncertainty and operational value as evaluation priorities. Digital twins, reinforcement learning, graph methods, federated learning and agentic systems remain emerging, application-specific directions; deployment claims require field validation, safety assessment and decision-level outcomes (Fang et al. 2025; Li et al. 2025).

Declarations

Funding. No Funding

Conflicts of interest. The authors declare no competing interests

Data availability. The photovoltaic bibliometric analysis used a Scopus CSV containing 1,954 records and unique EIDs, together with Scopus export and VOSviewer screenshots. The separate solar-energy export, full executed queries, retrieval dates, deduplication log, thesaurus and native VOSviewer map files were unavailable for this revision. A shareable identifier and derived-metadata file should be deposited with the final article where permitted; the search history and processing files are required to reproduce the original maps fully.

Ethics. This review and bibliometric study analysed published literature and bibliographic records and involved no human participants or animal subjects. Ethical approval and informed consent were not applicable.

References

Adekanbi, Michael L.; Alaba, Ezekiel S.; John, Toluwalope J.; Tundealao, Tomi D.; Banji, Titilope I. (2024) 'Soiling loss in solar systems: A review of its effect on solar energy efficiency and mitigation techniques', Cleaner Energy Systems, 7, pp. 100094. Available at: https://doi.org/10.1016/j.cles.2023.100094
Ağbulut, Ümit; Gürel, Ali Etem; Biçen, Yunus (2021) 'Prediction of daily global solar radiation using different machine learning algorithms: Evaluation and comparison', Renewable and Sustainable Energy Reviews, 135, pp. 110114. Available at: https://doi.org/10.1016/j.rser.2020.110114
Almeida, Marcelo Pinho; Perpiñán, Oscar; Narvarte, Luis (2015) 'PV power forecast using a nonparametric PV model', Solar Energy, 115, pp. 354-368. Available at: https://doi.org/10.1016/j.solener.2015.03.006
Alzahrani, Ahmad; Shamsi, Pourya; Dagli, Cihan; Ferdowsi, Mehdi (2017) 'Solar Irradiance Forecasting Using Deep Neural Networks', Procedia Computer Science, 114, pp. 304-313. Available at: https://doi.org/10.1016/j.procs.2017.09.045
André, P.S.; Dias, L.M.S.; Correia, S.F.H.; Carneiro Neto, A.N.; Ferreira, R.A.S. (2024) 'Artificial neural networks for predicting optical conversion efficiency in luminescent solar concentrators', Solar Energy, 268, pp. 112290. Available at: https://doi.org/10.1016/j.solener.2023.112290
Anonto, Hasanur Zaman; Hossain, Md Ismail; Shufian, Abu; Sheikh, Protik Parvez; Alam, Sadman Shahriar; Sayem, Md. Shaoran; Shovon, S M Tanvir Hassan (2025) 'Analyzing energy consumption trends and environmental influences: A time-series study on temperature, renewables, and demand correlations', Cleaner Energy Systems, 12, pp. 100209. Available at: https://doi.org/10.1016/j.cles.2025.100209
Ansong, Martin; Huang, Gan; Nyang’onda, Thomas N.; Musembi, Robinson J.; Richards, Bryce S. (2025) 'Very short-term solar irradiance forecasting based on open-source low-cost sky imager and hybrid deep-learning techniques', Solar Energy, 294, pp. 113516. Available at: https://doi.org/10.1016/j.solener.2025.113516
Assi, A.; Al-Shamisi, M.; Jama, M. (2010) 'Prediction of monthly average daily global solar radiation in Al Ain City, UAE, using artificial neural networks', Proceedings of the 25th European Photovoltaic Solar Energy Conference, Valencia, Spain.
Banad, Yaser M.; Sharif, Sarah S.; Rezaei, Zahra (2025) 'Artificial intelligence and machine learning for smart grids: from foundational paradigms to emerging technologies with digital twin and large language model-driven intelligence', Energy Conversion and Management: X, 28, pp. 101329. Available at: https://doi.org/10.1016/j.ecmx.2025.101329
Barancsuk, Lilla; Groma, Veronika; Kocziha, Barnabás (2025) 'Hybrid ultra-short term solar irradiation forecasting using resource-efficient multi-step long-short term memory', Renewable Energy, 247, pp. 122962. Available at: https://doi.org/10.1016/j.renene.2025.122962
Belaid, S.; Mellit, A. (2016) 'Prediction of daily and mean monthly global solar radiation using support vector machine in an arid climate', Energy Conversion and Management, 118, pp. 105-118. Available at: https://doi.org/10.1016/j.enconman.2016.03.082
Chaaban, Alain K.; Alfadl, Najd (2024) 'A comparative study of machine learning approaches for an accurate predictive modeling of solar energy generation', Energy Reports, 12, pp. 1293-1302. Available at: https://doi.org/10.1016/j.egyr.2024.07.010
Chakraborty, Debojyoti; Mondal, Jayeeta; Barua, Hrishav Bakul; Bhattacharjee, Ankur (2023) 'Computational solar energy – Ensemble learning methods for prediction of solar power generation based on meteorological parameters in Eastern India', Renewable Energy Focus, 44, pp. 277-294. Available at: https://doi.org/10.1016/j.ref.2023.01.006
Chen, Ji-Long; Liu, Hong-Bin; Wu, Wei; Xie, De-Ti (2011) 'Estimation of monthly solar radiation from measured temperatures using support vector machines – A case study', Renewable Energy, 36(1), pp. 413-420. Available at: https://doi.org/10.1016/j.renene.2010.06.024
Chen, S.X.; Gooi, H.B.; Wang, M.Q. (2013) 'Solar radiation forecast based on fuzzy logic and neural networks', Renewable Energy, 60, pp. 195-201. Available at: https://doi.org/10.1016/j.renene.2013.05.011
Dai, Haonan; Zhang, Yumo; Wang, Fei (2025) 'A Day-Ahead PV Power Forecasting Method Based on Irradiance Correction and Weather Mode Reliability Decision', Energies, 18(11), pp. 2809. Available at: https://doi.org/10.3390/en18112809
Daniel-Durandt, Francisca Muriel; Rix, Arnold Johan (2023) 'Automating Quality Control of Irradiance Data with a Comprehensive Analysis for Southern Africa', Solar, 3(4), pp. 596-617. Available at: https://doi.org/10.3390/solar3040032
Das, Opy; Dahlioui, Dounia; Zafar, Muhammad Hamza; Akhtar, Naureen; Moosavi, Syed Kumayl Raza; Sanfilippo, Filippo (2025) 'Ultra-short term PV power forecasting under diverse environmental conditions: A case study of Norway', Energy Conversion and Management: X, 27, pp. 101072. Available at: https://doi.org/10.1016/j.ecmx.2025.101072
Detyniecki, Marcin; Marsala, Christophe; Krishnan, Ashwati; Siegel, Mel (2012) 'Weather-based solar energy prediction', 2012 IEEE International Conference on Fuzzy Systems, pp. 1-7. Available at: https://doi.org/10.1109/fuzz-ieee.2012.6251145
Diagne, Maimouna; David, Mathieu; Lauret, Philippe; Boland, John; Schmutz, Nicolas (2013) 'Review of solar irradiance forecasting methods and a proposition for small-scale insular grids', Renewable and Sustainable Energy Reviews, 27, pp. 65-76. Available at: https://doi.org/10.1016/j.rser.2013.06.042
Douli, Abdelhak Samir; Badoud, Abd Essalm; Ould Bouamama, Belkacem; Boukerdja, Mahdi (2025) 'Development and validation of a novel model-agnostic machine learning framework for photovoltaic systems performance assessment under complex weather conditions', Results in Engineering, 28, pp. 108237. Available at: https://doi.org/10.1016/j.rineng.2025.108237
Esposito, Elena; Leanza, Gianni; Di Francia, Girolamo (2024) 'Comparative Analysis of Ground-Based Solar Irradiance Measurements and Copernicus Satellite Observations', Energies, 17(7), pp. 1579. Available at: https://doi.org/10.3390/en17071579
Fang, Huasong; Zhang, Huayue; Wen, Shuli; Li, Zhong; Zeng, Zhilin; Zhu, Miao; Lin, Pengfeng (2025) 'A reinforcement learning based real-time energy management method for mobile microgrid considering photovoltaic uncertainty', International Journal of Electrical Power & Energy Systems, 170, pp. 110844. Available at: https://doi.org/10.1016/j.ijepes.2025.110844
Firoozi, Ali Akbar; Firoozi, Ali Asghar; Maghami, Mohammad Reza (2025) 'Harnessing photovoltaic innovation: Advancements, challenges, and strategic pathways for sustainable global development', Energy Conversion and Management: X, 27, pp. 101058. Available at: https://doi.org/10.1016/j.ecmx.2025.101058
Gao, Yuan; Miyata, Shohei; Akashi, Yasunori (2022) 'Multi-step solar irradiation prediction based on weather forecast and generative deep learning model', Renewable Energy, 188, pp. 637-650. Available at: https://doi.org/10.1016/j.renene.2022.02.051
Gastón, Martín; Pagola, Íñigo; Fernández-Peruchena, Carlos M.; Ramírez, Lourdes; Mallor, Fermin (2013) 'A new adaptive methodology of global-to-direct irradiance based on clustering and kernel machines techniques', 15th SolarPACES Conference, Berlin. Available at: https://hal.science/hal-00919064
Guanghua, Lyu; Soomar, Arsalan Muhammad; Shah, Syed Hadi Hussain; Shaikh, Shoaib; Musznicki, Piotr (2025) 'Maximum power point tracking strategies for solar PV systems: A review of current methods and future innovations', Results in Engineering, 28, pp. 107227. Available at: https://doi.org/10.1016/j.rineng.2025.107227
Guo, Yuxiang; Han, Qiang; Li, Tan; Fu, Huichu; Liang, Meng; Zhang, Siwei (2025) 'Robust Photovoltaic Power Forecasting Model Under Complex Meteorological Conditions', Mathematics, 13(11), pp. 1783. Available at: https://doi.org/10.3390/math13111783
Hamza, Mukhtar Fatihu; Modu, Babangida (2025) 'A bibliometric analysis on renewable energy microgrids integrating hydrogen storage: Strategies for optimal sizing and energy management', Scientific African, 27, pp. e02609. Available at: https://doi.org/10.1016/j.sciaf.2025.e02609
Jebli, Imane; Belouadha, Fatima-Zahra; Kabbaj, Mohammed Issam; Tilioua, Amine (2021) 'Prediction of solar energy guided by pearson correlation using machine learning', Energy, 224, pp. 120109. Available at: https://doi.org/10.1016/j.energy.2021.120109
Jia, Dongyu; Hua, Jiajia; Wang, Liping; Guo, Yitao; Guo, Hong; Wu, Pingping; Liu, Min; Yang, Liwei (2021) 'Estimations of Global Horizontal Irradiance and Direct Normal Irradiance by Using Fengyun-4A Satellite Data in Northern China', Remote Sensing, 13(4), pp. 790. Available at: https://doi.org/10.3390/rs13040790
Juncklaus Martins, Bruno; Cerentini, Allan; Mantelli, Sylvio Luiz; Loureiro Chaves, Thiago Zimmermann; Moreira Branco, Nicolas; von Wangenheim, Aldo; Rüther, Ricardo; Marian Arrais, Juliana (2022) 'Systematic review of nowcasting approaches for solar energy production based upon ground-based cloud imaging', Solar Energy Advances, 2, pp. 100019. Available at: https://doi.org/10.1016/j.seja.2022.100019
Júnior, Jorge S. S.; Mendes, Jérôme; Souza, Francisco; Premebida, Cristiano (2023) 'Survey on Deep Fuzzy Systems in Regression Applications: A View on Interpretability', International Journal of Fuzzy Systems, 25(7), pp. 2568-2589. Available at: https://doi.org/10.1007/s40815-023-01544-8
Kisi, Ozgur; Alizamir, Meysam; Trajkovic, Slavisa; Shiri, Jalal; Kim, Sungwon (2020) 'Solar Radiation Estimation in Mediterranean Climate by Weather Variables Using a Novel Bayesian Model Averaging and Machine Learning Methods', Neural Processing Letters, 52(3), pp. 2297-2318. Available at: https://doi.org/10.1007/s11063-020-10350-4
Kull, Karl; Asad, Bilal; Khan, Muhammad Amir; Naseer, Muhammad Usman; Kallaste, Ants; Vaimann, Toomas (2025) 'Faults, Failures, Reliability, and Predictive Maintenance of Grid-Connected Solar Systems: A Comprehensive Review', Applied Sciences, 15(21), pp. 11461. Available at: https://doi.org/10.3390/app152111461
Kumar, Kondapalli Niranjan; Suzuki, Kentaroh (2019) 'Assessment of seasonal cloud properties in the United Arab Emirates and adjoining regions from geostationary satellite data', Remote Sensing of Environment, 228, pp. 90-104. Available at: https://doi.org/10.1016/j.rse.2019.04.024
Lauret, Philippe; Alonso-Suárez, Rodrigo; Amaro e Silva, Rodrigo; Boland, John; David, Mathieu; Herzberg, Wiebke; Le Gall La Salle, Josselin; Lorenz, Elke; Visser, Lennard; van Sark, Wilfried; Zech, Tobias (2024) 'The added value of combining solar irradiance data and forecasts: A probabilistic benchmarking exercise', Renewable Energy, 237, pp. 121574. Available at: https://doi.org/10.1016/j.renene.2024.121574
Ledmaoui, Younes; El Maghraoui, Adila; El Aroussi, Mohamed; Saadane, Rachid; Chebak, Ahmed; Chehri, Abdellah (2023) 'Forecasting solar energy production: A comparative study of machine learning algorithms', Energy Reports, 10, pp. 1004-1012. Available at: https://doi.org/10.1016/j.egyr.2023.07.042
Lee, Dong-Seok (2025) 'A Multi-Directional Pyranometer (CUBE-i) for Real-Time Direct and Diffuse Solar Irradiance Decomposition', Remote Sensing, 17(8), pp. 1336. Available at: https://doi.org/10.3390/rs17081336
Li, Zheng; Ma, Jun; Wang, Qian; Wang, Mingzhu; Jiang, Feifeng (2025) 'Enhancing urban solar irradiation prediction with shadow-attention graph neural networks: Implications for net-zero energy buildings in New York City', Sustainable Cities and Society, 120, pp. 106133. Available at: https://doi.org/10.1016/j.scs.2025.106133
Liu, Weijia; Liu, Yangang; Zhang, Tao; Han, Yongxiang; Zhou, Xin; Xie, Yu; Yoo, Shinjae (2022) 'Use of physics to improve solar forecast: Part II, machine learning and model interpretability', Solar Energy, 244, pp. 362-378. Available at: https://doi.org/10.1016/j.solener.2022.08.040
Mohammed, Fareeduddin; Boumaiza, Ameni; Sanfilippo, Antonio; Perez-Astudillo, Daniel; Bachour, Dunia (2025) 'A robust hybrid machine learning framework for short-term load forecasting: integrating multi-linear regression, long short-term memory, and feed-forward neural networks for enhanced accuracy and efficiency', Energy and AI, 22, pp. 100625. Available at: https://doi.org/10.1016/j.egyai.2025.100625
Moosa, Aaftaab; Shabir, Hamza; Ali, Huzefa; Darwade, Rishikesh; Gite, Balasaheb (2018) 'Predicting Solar Radiation Using Machine Learning Techniques', 2018 Second International Conference on Intelligent Computing and Control Systems (ICICCS), pp. 1693-1699. Available at: https://doi.org/10.1109/iccons.2018.8663110
Morey, Meghraj; Gupta, Nitin; Garg, Man Mohan; Kumar, Ajay (2023) 'A comprehensive review of grid-connected solar photovoltaic system: Architecture, control, and ancillary services', Renewable Energy Focus, 45, pp. 307-330. Available at: https://doi.org/10.1016/j.ref.2023.04.009
Najibi, Fatemeh; Apostolopoulou, Dimitra; Alonso, Eduardo (2021) 'Enhanced performance Gaussian process regression for probabilistic short-term solar output forecast', International Journal of Electrical Power & Energy Systems, 130, pp. 106916. Available at: https://doi.org/10.1016/j.ijepes.2021.106916
Olayiwola, Olufemi; Cali, Umit; Elsden, Miles; Yadav, Poonam (2025) 'Enhanced Solar Photovoltaic System Management and Integration: The Digital Twin Concept', Solar, 5(1), pp. 7. Available at: https://doi.org/10.3390/solar5010007
Paletta, Quentin; Terrén-Serrano, Guillermo; Nie, Yuhao; Li, Binghui; Bieker, Jacob; Zhang, Wenqi; Dubus, Laurent; Dev, Soumyabrata; Feng, Cong (2023) 'Advances in solar forecasting: Computer vision with deep learning', Advances in Applied Energy, 11, pp. 100150. Available at: https://doi.org/10.1016/j.adapen.2023.100150
Perez, Richard; Lorenz, Elke; Pelland, Sophie; Beauharnois, Mark; Van Knowe, Glenn; Hemker, Karl; Heinemann, Detlev; Remund, Jan; Müller, Stefan C.; Traunmüller, Wolfgang; Steinmauer, Gerald; Pozo, David; Ruiz-Arias, Jose A.; Lara-Fanego, Vicente; Ramirez-Santigosa, Lourdes; Gaston-Romero, Martin; Pomares, Luis M. (2013) 'Comparison of numerical weather prediction solar irradiance forecasts in the US, Canada and Europe', Solar Energy, 94, pp. 305-326. Available at: https://doi.org/10.1016/j.solener.2013.05.005
Piri, Jamshid; Shamshirband, Shahaboddin; Petković, Dalibor; Tong, Chong Wen; Rehman, Muhammad Habib ur (2015) 'Prediction of the solar radiation on the Earth using support vector regression technique', Infrared Physics & Technology, 68, pp. 179-185. Available at: https://doi.org/10.1016/j.infrared.2014.12.006
Polleux, Louis; Guerassimoff, Gilles; Marmorat, Jean-Paul; Sandoval-Moreno, John; Schuhler, Thierry (2022) 'An overview of the challenges of solar power integration in isolated industrial microgrids with reliability constraints', Renewable and Sustainable Energy Reviews, 155, pp. 111955. Available at: https://doi.org/10.1016/j.rser.2021.111955
Pourmir, Mohammadreza; Miri, Seyedeh Mohadeseh (2025) 'Machine learning methods comparison by using statistical tests in solar energy forecasting based on weather features', Engineering Applications of Artificial Intelligence, 162, pp. 112239. Available at: https://doi.org/10.1016/j.engappai.2025.112239
Prakash, Anuj; Singh, Sandeep Kumar (2014) 'Towards an efficient regression model for solar energy prediction', 2014 Innovative Applications of Computational Intelligence on Power, Energy and Controls with their impact on Humanity (CIPECH), pp. 18-23. Available at: https://doi.org/10.1109/cipech.2014.7019040
Prema, V.; Rao, K. Uma (2015) 'Development of statistical time series models for solar power prediction', Renewable Energy, 83, pp. 100-109. Available at: https://doi.org/10.1016/j.renene.2015.03.038
R. Singh, Arvind; Kumar, R. Seshu; Bajaj, Mohit; Khadse, Chetan B.; Zaitsev, Ievgen (2024) 'Machine learning-based energy management and power forecasting in grid-connected microgrids with multiple distributed energy sources', Scientific Reports, 14(1). Available at: https://doi.org/10.1038/s41598-024-70336-3
Rahman, Tuhibur; Mansur, Ahmed; Hossain Lipu, Molla; Rahman, Md.; Ashique, Ratil; Houran, Mohamad; Elavarasan, Rajvikram; Hossain, Eklas (2023) 'Investigation of Degradation of Solar Photovoltaics: A Review of Aging Factors, Impacts, and Future Directions toward Sustainable Energy Management', Energies, 16(9), pp. 3706. Available at: https://doi.org/10.3390/en16093706
Rajagukguk, Rial A.; Kamil, Raihan; Lee, Hyun-Jin (2021) 'A Deep Learning Model to Forecast Solar Irradiance Using a Sky Camera', Applied Sciences, 11(11), pp. 5049. Available at: https://doi.org/10.3390/app11115049
Ramírez-Rivera, Francisco A.; Guerrero-Rodríguez, Néstor F. (2024) 'Ensemble Learning Algorithms for Solar Radiation Prediction in Santo Domingo: Measurements and Evaluation', Sustainability, 16(18), pp. 8015. Available at: https://doi.org/10.3390/su16188015
Ramli, Makbul A.M.; Twaha, Ssennoga; Al-Turki, Yusuf A. (2015) 'Investigating the performance of support vector machine and artificial neural networks in predicting solar radiation on a tilted surface: Saudi Arabia case study', Energy Conversion and Management, 105, pp. 442-452. Available at: https://doi.org/10.1016/j.enconman.2015.07.083
Reddy, B. Nishant Sree; Gautam, Kumar; Pachauri, Nikhil (2025) 'Solar potential assessment using machine learning and climate change projections for long-term energy planning', Scientific Reports, 15(1). Available at: https://doi.org/10.1038/s41598-025-23661-0
Rumbayan, Meita; Abudureyimu, Asifujiang; Nagasaka, Ken (2012) 'Mapping of solar energy potential in Indonesia using artificial neural network and geographical information system', Renewable and Sustainable Energy Reviews, 16(3), pp. 1437-1449. Available at: https://doi.org/10.1016/j.rser.2011.11.024
Sansine, Vateanui; Ortega, Pascal; Hissel, Daniel; Hopuare, Marania (2022) 'Solar Irradiance Probabilistic Forecasting Using Machine Learning, Metaheuristic Models and Numerical Weather Predictions', Sustainability, 14(22), pp. 15260. Available at: https://doi.org/10.3390/su142215260
Santos, Leticia de Oliveira; AlSkaif, Tarek; Barroso, Giovanni Cordeiro; Carvalho, Paulo Cesar Marques de (2024) 'Photovoltaic power estimation and forecast models integrating physics and machine learning: A review on hybrid techniques', Solar Energy, 284, pp. 113044. Available at: https://doi.org/10.1016/j.solener.2024.113044
Seman, Laio Oriel; Stefenon, Stefano Frizzo; Yow, Kin-Choong; Coelho, Leandro dos Santos; Mariani, Viviana Cocco (2026) 'Multi-step short-term solar energy forecasting using Fourier-enhanced BiLSTM and neural additive models', Renewable Energy, 257, pp. 124738. Available at: https://doi.org/10.1016/j.renene.2025.124738
Sharkawy, Abdel-Nasser; Bekhiti, Belkacem (2026) 'Enhanced estimation and short-term forecasting of output power of a solar photovoltaic plant using adaptive neuro-fuzzy inference system (ANFIS)', Journal of Electrical Systems and Information Technology, 13(1). Available at: https://doi.org/10.1186/s43067-025-00302-0
Shata, Dina; Omrani, Sara; Drogemuller, Robin; Denman, Simon; Wagdy, Ayman (2025) 'Unlocking solar potential through machine learning techniques for roof geometry prediction- A review', Solar Energy, 302, pp. 113994. Available at: https://doi.org/10.1016/j.solener.2025.113994
Sivaneasan, B.; Yu, C.Y.; Goh, K.P. (2017) 'Solar Forecasting using ANN with Fuzzy Logic Pre-processing', Energy Procedia, 143, pp. 727-732. Available at: https://doi.org/10.1016/j.egypro.2017.12.753
Sulaiman, Mohd Herwan; Mustaffa, Zuriani (2024) 'Forecasting solar power generation using evolutionary mating algorithm-deep neural networks', Energy and AI, 16, pp. 100371. Available at: https://doi.org/10.1016/j.egyai.2024.100371
Sun, Yanwei; Li, Ying; Wang, Run; Ma, Renfeng (2022) 'Measuring dynamics of solar energy resource quality: Methodology and policy implications for reducing regional energy inequality', Renewable Energy, 197, pp. 138-150. Available at: https://doi.org/10.1016/j.renene.2022.07.122
Sun, Zhigang; Wang, Guotao; Li, Pengfei; Wang, Hui; Zhang, Min; Liang, Xiaowen (2024) 'An improved random forest based on the classification accuracy and correlation measurement of decision trees', Expert Systems with Applications, 237, pp. 121549. Available at: https://doi.org/10.1016/j.eswa.2023.121549
Taghezouit, Bilal; Harrou, Fouzi; Sun, Ying; Merrouche, Walid (2024) 'Model-based fault detection in photovoltaic systems: A comprehensive review and avenues for enhancement', Results in Engineering, 21, pp. 101835. Available at: https://doi.org/10.1016/j.rineng.2024.101835
Tanoli, Irfan Khan; Mehdi, Asqar; Algarni, Abeer D.; Fazal, Azra; Khan, Talha Ahmed; Ahmad, Sadique; Ateya, Abdelhamied A. (2024) 'Machine learning for high-performance solar radiation prediction', Energy Reports, 12, pp. 4794-4804. Available at: https://doi.org/10.1016/j.egyr.2024.10.033
Tharani, Kusum; Kumar, Neeraj; Srivastava, Vishal; Mishra, Sakshi; Pratyush Jayachandran, M. (2020) 'Machine learning models for renewable energy forecasting', Journal of Statistics and Management Systems, 23(1), pp. 171-180. Available at: https://doi.org/10.1080/09720510.2020.1721636
Tripathi, Sumit; Pandey, Dinesh Chandra (2025) 'A data-driven approach to renewable energy forecasting: integrating digital twins and evolutionary computing', Environment, Development and Sustainability. Available at: https://doi.org/10.1007/s10668-025-06457-0
West, Samuel R.; Rowe, Daniel; Sayeef, Saad; Berry, Adam (2014) 'Short-term irradiance forecasting using skycams: Motivation and development', Solar Energy, 110, pp. 188-207. Available at: https://doi.org/10.1016/j.solener.2014.08.038
Wu, Ji; Chan, C.K. (2012) 'The Prediction of Monthly Average Solar Radiation with TDNN and ARIMA', 2012 11th International Conference on Machine Learning and Applications, pp. 469-474. Available at: https://doi.org/10.1109/icmla.2012.225
Wu, Yujie; Wang, Jianzhou (2016) 'A novel hybrid model based on artificial neural networks for solar radiation prediction', Renewable Energy, 89, pp. 268-284. Available at: https://doi.org/10.1016/j.renene.2015.11.070
Yang, Dazhi; Yagli, Gokhan Mert; Quan, Hao (2018) 'Quality Control for Solar Irradiance Data', 2018 IEEE Innovative Smart Grid Technologies - Asia (ISGT Asia), pp. 208-213. Available at: https://doi.org/10.1109/isgt-asia.2018.8467892
Yousif, Jabar H.; Al-Balushi, Haitham A.; Kazem, Hussein A.; Chaichan, Miqdam T. (2019) 'Analysis and forecasting of weather conditions in Oman for renewable energy applications', Case Studies in Thermal Engineering, 13, pp. 100355. Available at: https://doi.org/10.1016/j.csite.2018.11.006
Zeng, Jianwu; Qiao, Wei (2013) 'Short-term solar power prediction using a support vector machine', Renewable Energy, 52, pp. 118-127. Available at: https://doi.org/10.1016/j.renene.2012.10.009
Zhang, Yandi (2025) 'Enhancing solar irradiance prediction for sustainable energy solutions employing a hybrid machine learning model; improving hydrogen production through Photoelectrochemical device', Applied Energy, 382, pp. 125280. Available at: https://doi.org/10.1016/j.apenergy.2025.125280

Machine Learning for Solar Radiation and Photovoltaic Systems: Evidence, Validation, and Operational Applications

Abstract

1. Introduction

2. Review Approach and Bibliographic Scope

2.1. Database Selection and Evidence Sets

2.2. Descriptive Bibliometric Results

3. Solar Energy Resources and Management: An Overview

3.1. Solar Radiation Types

3.1.1. Global Horizontal Irradiance (GHI)

3.1.2. Direct Normal Irradiance (DNI)

3.1.3. Diffuse Horizontal Irradiance (DHI)

3.2. Key Challenges in Solar Energy Resources and Management

3.3. Solar Energy Management Tasks

3.3.1. PV Power Forecasting

3.3.2. Maximum Power Point Tracking (MPPT)

3.3.3. PV Fault Diagnosis

3.3.4. Solar Farm Control and Optimization

3.3.5. Load Management and Demand Response

3.3.6. Grid Stability Integration

4. Machine Learning Techniques used in Solar Energy

4.1. Classical Machine Learning Methods

4.1.1. Artificial Neural Networks (ANN)

4.1.2. Support Vector Machines (SVM)

4.1.3. Decision Trees and Random Forests (DT and RF)

4.1.4. Gaussian Process Regression (GPR)

4.1.5. Fuzzy Logic (FL) and ANFIS

4.2. Ensemble Learning (EL)

4.3. Deep Learning (DL) Architectures

4.4. Hybrid and Optimization-Enhanced Models

4.5. Emerging Methods

5. Machine Learning Applications in Solar Energy

5.1. Solar Radiation (SR) Forecasting

5.2. PV Power Output Prediction

5.3. Solar Farm Management and Optimization

5.4. Solar Resource Assessment (SRA)

5.5. Load Forecasting (LF) and Smart Grid Integration

6. Comparative Analysis of Machine Learning Models

6.1. Performance Metrics

6.2. Model Strengths and Weaknesses

6.3. Benchmark Datasets

7. Research Trends in Machine Learning Applications for Solar Energy

8. Challenges and Research Gaps in Machine Learning-Based Solar Energy Systems

9. Future Direction

10. Policy Implications and Practical Recommendations

11. Conclusions

Declarations

References

How to cite this article