Forecasting green hydrogen production: An assessment of renewable energy systems using deep learning and statistical methods

•Analyzed nine years (2015–2023) of meteorological and photovoltaic data from Beni Mellal, Morocco.•Evaluated monocrystalline, polycrystalline, and amorphous PV panels for green hydrogen production.•Polycrystalline panels achieved the highest hydrogen output, peaking at 60.65 kg in 2018.•Seasonal patterns, especially in July, significantly impacted hydrogen production efficiency.•Advanced models (MLP, Random Forest, LSTM-CNN) outperformed traditional methods for hydrogen production forecasting. This study presents a comprehensive analysis of nine years (January 2015 to December 2023) of meteorological and photovoltaic data from the Faculty of Science and Technology in Beni Mellal, Morocco, focusing on solar irradiance, ambient temperature, and forecasting green hydrogen production. Three photovoltaic technologies—monocrystalline, polycrystalline, and amorphous panels—are compared to assess their hydrogen production efficiency. The findings indicate that polycrystalline panels generally achieved the highest hydrogen production, peaking at 60.65 kg in 2018, while amorphous panels experienced a marked decline in performance by 2023. These results emphasize the substantial role of seasonal patterns and environmental factors in influencing production efficiency across the different photovoltaic technologies. The study further reveals notable variations in the efficiency of PV-hydrogen systems among the panel types. Monocrystalline panels exhibited the widest efficiency range, peaking at 9.96% in 2018 and dropping to a low of 5.56% in 2015. In comparison, polycrystalline panels displayed a narrower range, while amorphous panels showed efficiency fluctuations, peaking at 7.71% in 2019. These disparities underscore the importance of adapting system design to environmental conditions to enhance the performance of PV-hydrogen systems. To forecast hydrogen production, the study applied various advanced machine learning (ML) models, including Support Vector Regression (SVR) and Random Forest. The Random Forest model demonstrated the highest accuracy for polycrystalline panels, with a Mean Absolute Error (MAE) of 0.0024 and a Mean Absolute Percentage Error (MAPE) of 1.6011%. In addition, deep learning (DL) methods, specifically the Multilayer Perceptron (MLP) and Long Short-Term Memory Convolutional Neural Network (LSTM-CNN), were utilized. The MLP model emerged as the best performer for monocrystalline panels, achieving an R2 of 0.9430. For amorphous panels, the MLP m