Optimizing energy production predictions in hybrid PV-wind systems: Integrating deep learning techniques and particle swarm optimization for improved performance in tropical climates

The energy transition towards a sustainable system requires a predictive approach and accurate technical-economic planning, especially in tropical regions with complex climate characteristics. This study aims to examine the potential of renewable energy based on hybrid solar panels (PV) and wind turbines through long-term climate data analysis and artificial intelligence modeling. Climate data from 2010 to 2025 were analyzed using descriptive statistics to identify the distribution of sunshine duration (SS) and average wind speed (FF_AVG). Predictions were performed using deep recurrent neural networks (DRNNs) and long short-term memory (LSTM) architectures and validated using cross-validation metrics. The prediction results were then applied in energy production simulations and optimized using particle swarm optimization (PSO) to determine the optimal system capacity and Levelized Cost of Energy (LCOE). The optimal hybrid configuration consists of 30 kWp PV capacity and a 500 W wind turbine to supply a daily electrical load of 13.5 kWh. The LSTM model outperformed DRNN, achieving an MSE of 4.8251 and R² of 0.3245 for SS, and an MSE of 0.3124 and R² of 0.2368 for FF_AVG. Simulation results indicate that solar energy dominates production during the dry season, while wind energy contributes as a complementary source in the rainy season. PSO optimization successfully reduced the LCOE to Rp505/kWh, with an average daily energy surplus of 95.9 kWh and a total accumulation of 8479 kWh over 90 days. Environmentally, the proposed hybrid system is estimated to reduce carbon dioxide emissions by approximately 10.4 tons CO₂ per year compared to conventional grid electricity. This study concludes that integrating LSTM with PSO improves prediction reliability, system efficiency, and sustainability of hybrid PV–wind systems. Despite remaining challenges in solar prediction accuracy, this research provides a valuable contribution to tropical renewable energy development through artificial intelligence and economic optimization.