Forecasting Solar Power Generation Using a Novel Hybrid Deep Learning Model with Attention Mechanisms

The transition from fossil fuels to renewable energy sources has gained great importance due to environmental concerns and the nature of unsustainable energy production methods. Among the alternative renewable energy sources, solar power has a prominent place due to its abundance, accessibility and ecological benefits, but the variability and unpredictability of solar irradiance creates considerable challenges in predicting solar power generation. This study proposes a novel attention-based hybrid deep learning method for forecasting of solar power generation using real-world data. The used dataset combines power production records at the inverter level and weather data at the plant level. In the proposed model, deep learning approaches such as Convolutional Neural Networks and Long Short-Term Memory are employed in a hybrid architecture that includes the attention mechanism and a comparison is made with widely used machine learning methods. The hybrid attention model exhibited substantially higher predictive accuracy on the test set compared to conventional methods, and the results confirm the ability of attention-driven hybrid architectures to capture the complex, nonlinear dynamics of solar power prediction tasks.