Yuki Trisnoaji, Singgih Dwi Prasetyo, Zainal Arifin, Mochamad Subchan Mauludin
The increase in global carbon dioxide (CO2) emissions highlights the urgent need to shift to a low-carbon energy system, where green hydrogen offers a strategic, clean, and sustainable solution. Indonesia has high potential for solar and wind energy, which can be combined through a hybrid PV–wind turbine on-grid system to support green hydrogen production. However, tropical weather variability presents challenges to power stability and electrolyzer efficiency. This study combines a systematic meta-analysis approach with deep learning-based predictive modeling to assess the technical and economic performance of such systems in tropical Indonesia. A meta-analysis of 52 global studies shows that electrolyzer capacity factor is the most critical factor affecting hydrogen production cost (LCOH), with a strong negative correlation (R2 = 0.92). Meteorological data from Tardamu Station, Kupang (2010–2025), were used to train three recurrent neural network models—namely RNN, LSTM, and GRU—to forecast solar radiation (SS) and wind speed (FF_AVG). The evaluation indicates that the RNN achieves the highest accuracy, with MSE = 0.0099, RMSE = 0.0994, and R2 = 0.9988, outperforming both the LSTM and the GRU. Seasonal forecasts for 2025–2045 confirm that RNN remains stable against long-term climate variations. Energy simulations estimate a total annual output of 23,937.7 kWh, with an average surplus of 78.9%, supporting the production of 0.4054 kg H2/day at a LCOH of USD 2.84/kg. This study demonstrates that integrating meta-analysis with deep learning improves the prediction accuracy and economic feasibility of a hybrid PV–wind–hydrogen system in tropical climates. The RNN-based approach is recommended as an adaptive prediction tool for real-time optimization and long-term planning toward Net Zero Emissions by 2060 in Indonesia. © 2026 Elsevier Ltd.
Power Plant Engineering Technology, State University of Malang, Malang, 65145, Indonesia; Post-doctoral Researcher, Department of Mechanical Engineering, Universitas Sebelas Maret, Surakarta, 57126, Indonesia; Department of Mechanical Engineering, Universitas Sebelas Maret, Surakarta, 57126, Indonesia; Department of Informatics Engineering, Universitas Wahid Hasyim, Semarang, 50236, Indonesia
