High-Accuracy Solar Energy Forecasting Using Hybrid AI Models: Optimizing Performance in Thermal and Mechanical Conversion Systems

Abstract

Accurate forecasting of solar photovoltaic (PV) energy generation remains a challenging task due to the strong variability and nonlinear influence of meteorological conditions, which directly affects the reliability of energy management systems. This paper proposed the comparative framework of models for highly accurate forecasting of solar photovoltaic (PV) energy generation, utilizing state-of-the-art machine learning (ML) and deep learning (DL) approaches to substantially boost forecasting reliability under varying environmental conditions. All five models, Multi-Layer Perceptron (MLP), Extreme Gradient Boosting (XGBoost), Long Short-Term Memory (LSTM) and families of models, gated recurrent unit (GRU) and bidirectional LSTM (BiLSTM) have been implemented and tested using realistic meteorological data and irradiance data directly retrieved from the GRIDouble project, supported in I-NERGY from the HORIZON 2020 program of the European Union. More specifically, the dataset consisted of hourly measurements of such variables as air temperature, cloud opacity, diffuse and direct irradiance, and global horizontal irradiance, coupled with solar energy production reading. To maximize the learning power of time-patterned data and to minimize its variance, a comprehensive pre-processing pipeline was implemented that included temporal features extraction, cyclical encoding, as well as min–max normalization. Experiments confirmed that XGBoost yielded the highest forecasting performance with R² = 0.9478, MSE = 3191.15, RMSE = 56.49, MSRE = 27.59. The remaining models, bidirectional LSTM and MLP, demonstrated R² = 0.8662 and R² = 0.8581, respectively, highlighting their substantial ability to model the temporal and nonlinear environment. In summary, XGBoost can be praised as the top model with the achieved the best performance in this experimental setting. It also presents incredible potential for smart-grid optimization and renewable energy forecasting.