Application of hybrid deep learning models for seasonal solar irradiance forecasting in the city of Pala.

Accurate prediction of solar irradiance is crucial for designing, controlling, and monitoring solar energy systems. However, its intermittent nature poses a significant challenge to achieving satisfactory forecasting results. This study aims to design new hybrid models and evaluate the performance of different forecasting models. The models combine Attentive Meteo-Embedding with long-term memory (AME-LSTM), Attentive Meteo-Embedding and convolutional neural networks (AME-CNN), Attentive Meteo-Embedding and gated recurrent units (AME-GRU) using the Adam optimization algorithm. The models take temperature, relative humidity, and wind direction as input variables for the years 2015–2023. The results show that the AME-GRU model performs exceptionally well in summer and spring, while the AME-CNN model excels in summer and autumn. However, the AME-LSTM model has notable limitations, with higher errors in winter. This study can contribute to the development of a hybrid power plant for the city of Pala.