Performance analysis of advanced deep learning technics: Application to solar energy forecasting and management in several cities in Chad

• The design of a hybrid LLM_LSTM model based on the Adam optimization algorithm. • The use of deep learning algorithms to optimize energy management, offering a novel approach to solar radiation forecasting. • The collection and processing of local weather data to enable robust model training, ensuring the relevance of the results. Accurate solar radiation prediction is essential as it provides detailed solar resource information to investors and governments, promoting solar energy development. The main objective of this study is to analyze advanced forecasting techniques and apply them to solar radiation prediction to enhance energy management in various cities in Chad. To achieve this, eight deep learning algorithms were used to optimize energy management. This study proposes a hybrid artificial intelligence model that combines LLM and LSTM methods, utilizing the Adam optimization algorithm to make hourly solar radiation predictions over a seven-day period. The data, sourced from the Chad weather station for the period 2013–2023, was used to train the models. Additionally, the predicted values were used to calculate the power output of a photovoltaic (PV) system, as well as its efficiency and the energy that could be stored. The results indicate that the statistical indicators for each algorithm range from 43.197 to 123.393 W/m² for RMSE, from 9.356 to 90.762 W/m for MAE, from 0.155 to 20.194 W/m² for MedAE, and from 0.822 to 0.998 for R². The analysis demonstrates that the proposed hybrid LLM-LSTM model is more efficient than other models.