Deep learning has grown among the prediction tools used within renewable energy options. Solar energy belongs to the options with the lowest atmosphere impact after considering their limitations. In the last five years, Brazil has seen the expansion of wind and solar options almost all over the country, and to preserve the Amazon rainforest, the use of solar energy has helped large and small cities towards a greener future. The novelty of this research covers the use of Deep Learning with data from twelve cities in the state of Amazonas to forecast solar irradiation (W.h/m 2 ) within 30 days. The data input came from ground stations, as much as possible, and NASA satellite models, with a daily time aggregation. The types of neural networks considered are Long Short-Term Memory (LSTM), a Multi-Layer Perceptron (MLP), and an LSTM Gated Recurrent Unit (GRU). Among the metrics used to check the algorithm’s performance, the Mean Absolute Percentage Error (MAPE) indicates that the values of this research are coherent with other scenarios to forecast solar energy; the boundary conditions were not the same, however. The lowest MAPE was observed in the city of Labrea with the LSTM GRU.
Forecasting models employing machine learning (ML) and deep learning (DL) have become fundamental for assessing the technical feasibility of renewable energy systems. Among these, solar energy stands out as a renewable energy option, particularly relevant for supporting the preservation of the Amazon rainforest. This study introduces a novel approach using ML and DL methods—integrated with Universal Kriging and Holt-Winters (time series) models — to forecast solar irradiance (kWh/m2) in cities across the state of Amazonas. The analysis is grounded in the Data Science cycle, with input data sourced from both ground stations and satellite products. Forecasting performance was evaluated for short-term horizons (one to three days ahead) across three representative cities. The hybrid SARIMAX-CNN-LSTM, SARIMAX-CNN-Transformer, and SARIMAX-TCN models achieved MAPE values ranging from 18.1% to 26.6% for the different forecast horizons and cities. These results are consistent with existing literature and reinforce the suitability of advanced ML/DL approaches for solar energy forecasting in highly variable and challenging environments such as the Amazon Basin.
Assessing whether data-driven photovoltaic (PV) models can generalize beyond the plants used for training remains a critical challenge when operational data are available for only a limited number of installations. This study proposes a methodological framework for cross-plant transfer assessment, focusing on spatial generalization rather than within-plant temporal fitting. Daily operational data from three utility-scale PV plants are combined with satellite-derived irradiance and meteorological variables to construct a normalized, dimensionless technical performance indicator, closely related to established PV normalization metrics and designed to reduce scale and site effects. Forecasting models—including decision-tree ensembles (Random Forest and gradient-boosted trees), multilayer perceptrons (MLP), and convolutional–recurrent neural networks (CNN–LSTM and CNN–GRU)—are evaluated by training on data pooled from two plants and testing exclusively on the third. Hyperparameters are optimized using Optuna under a unified tuning strategy. Performance is assessed using MAE, sMAPE, and RMSE, with emphasis on metrics appropriate for normalized targets. Results show that no single model dominates across all transfer scenarios and that forecasting performance degrades under spatial transfer in a plant-dependent manner, even after normalization. Diebold–Mariano tests against a Kriging-based spatial baseline provide strong statistical evidence that the learning-based models deliver predictive power beyond spatial smoothness alone across all evaluated plants. The contribution of this work lies in quantifying the limits of cross-plant generalization and in providing a statistically grounded framework for comparative PV performance assessment in data-scarce settings.
The renewable energies have confirmed their potential to reshape the energy matrix of several countries, in the last two decades, reinforcing the set of actions to better protect the environment. F...