Hybrid Stacking with Targeted Residual Learning for PV Forecasting

Deep Learning for Solar Power Forecasting by Integrating Historical and Meteorological Data

With the global shift toward green energy, solar photovoltaic (PV) power has expanded rapidly. However, the unpredictable nature of PV generation, caused by changing weather conditions like irradiance and temperature, challenges grid stability and power scheduling. Therefore, developing smart forecasting models for high-precision PV power prediction is essential for modern grid management. This paper introduces an optimized forecasting framework using Long Short-Term Memory (LSTM) networks. By integrating historical power generation data with localized meteorological factors, a multivariate predictive model was developed and validated using empirical data from a 100 MW PV plant. Based on Pearson correlation analysis, four key features—temperature, direct normal irradiance (DNI), relative humidity, and cloud cover—were chosen as the main drivers of PV output. A multivariate LSTM model was then trained and carefully tested using time series cross-validation. Results show the multi-feature architecture consistently outperforms and surpasses single-feature benchmarks. Specifically, the model achieved a peak R2 of 0.9864 and minimum MAE of 1.4057 kW. Across four validation sets, R2 remained stable (0.9755–0.9836), with most errors tightly bounded within ±5 kW. These findings confirm the model’s excellent accuracy and its value for improving power system dispatch and resource planning.

A Novel Two-Stage Residual-Corrected Stacking Framework for Photovoltaic Power Output Prediction

Proper prediction of photovoltaic (PV) power output is essential in ensuring the successful incorporation of solar energy in the smart grid systems and energy management systems. Current single-algorithm models are often ineffective to represent the compound non-linear interactions between meteorological variables, time variations and irradiance dynamics that cause solar generation variability. This paper presents the Solar-Adaptive Hybrid Ensemble (SAHE) which is a new two-stage stacking model that uses a new Random Forest (RF) base learner with an XG Boost residual-correction meta-learner, supplemented by solar-domain feature engineering such as clearness index, irradiance polynomial transforms, lagged target variables and cyclic temporal encodings. The SAHE framework has a Root Mean Squared Error (RMSE) equal to 23.8850 kWh, Mean Absolute Error (MAE) equal to 18.6632 kWh, coefficient of determination (R 2) equal to 0.8079, Mean Absolute Percentage Error (MAPE) equal to 8.7608% and Pearson Correlation Coefficient (PCC) equal to 0.9113 on the test set, compared to all comparison models, in all reported measures.

Stacked Ensemble–Neural Network Hybrid Framework for Autonomous Control and Optimization of Solar Photovoltaic Energy Systems

The growing demand for renewable energy around the world, many countries are adding solar power to their energy programs. Solar photovoltaic (PV) systems can affect the stability and quality of the electricity grid since solar radiation can come and go, especially in big installations. Solar fluctuations can lead to either excessive or insufficient power generation, therefore accurate forecasting is essential for effective energy management and system integration. A major area of study is autonomous control and optimisation of solar photovoltaic energy systems. This study presents data preparation and transformation methodologies aimed at enhancing data quality and model efficacy in forecasting. Kernel Density Estimation (KDE) and Pearson Correlation Coefficient (PCC) feature selection help find important characteristics and cut down on prediction mistakes. LSTM and XGBoost are the basic models that DES-XG, a frequently used stacked ensemble method, employs. Extreme gradient boosting combines the outputs of basic learners to create final predictions. Tests reveal that the proposed DES-XG model works better than both the LSTM and XGBoost models on their own, with an accuracy of 95.42% and better stability and consistency across case studies.

Truncated-Regularized Kernel Ridge Regression Hybrid Model for Solar Irradiance Prediction and Renewable Energy Forecasting

The significance of Solar Irradiance Prediction & Renewable Energy Forecasting for efficient energy utilisation is highlighted by the fast adoption of renewable energy systems, which is being propelled by falling PV costs and rising fossil fuel prices. Solar energy sources are beneficial to the environment, but they are difficult to predict because of their inherent variability and non-stationary behaviour. In response, the research suggests a solid method for predicting solar irradiance, which is integral to PV power generation. Dimensionality reduction using PCA follows extensive data pretreatment, which includes outlier removal and normalisation. The next step is to create a hybrid predictive model that forecasts both the short and long term using a TRKRR fast-trainable statistical learning approach using a variety of historical meteorological information. In testing, the suggested TRKRR-based model achieved a minimum MSE of 0.00142, demonstrating that it outperforms previous algorithm-trained models, especially in extremely dynamic weather situations. With the presentation of a scalable, accurate, and dependable model that can handle variability in renewable energy data, this research concludes that Solar Irradiance Prediction & Renewable Energy Forecasting has made substantial advancements, paving the way for better integration of solar power into modern energy systems.

Machine learning-based prediction of solar PV power using experimental data

A Hybrid Ensemble Learning Strategy for Enhancing Solar Photovoltaic Forecasting in Renewable Energy Systems

The successful forecast of the solar photovoltaic (PV) power processing is essential in increasing grid stability, and in the ultimate inclusion of renewable energy. This paper is a research project aiming to introduce a hybrid classifier of Random Forest (RF), K-Nearest Neighbors (KNN), and Elastic Net (EN) to enhance multi-regional solar PV power forecasting using time-series. Preprocessing of Hourly PV generation data was done by taking out the temporal feature like hour, day and month to improve forecasting. To compare the proposed hybrid model with individual algorithms, the coefficient of determination (R2), mean absolute error (MAE) were used to evaluate the model. The results of the experiment indicate that the hybrid model had an R 2 of 0.91, which was higher than RF (0.88), KNN (0.82), and EN (0.79), and nearly halved the MAE, relative to single models. The results prove that the combination of multiple regression methods would lead to better prediction robustness, as well as lower variance and increase forecasting accuracy, which makes the suggested method an appropriate tool to study smart grids and sustainable energy planning.