浮体式太阳能发电系统的强化学习基准测试与原型开发：结合棕熊优化算法的实验研究与LSTM建模

Abstract This study conducts comprehensive comparative experimental investigation, performance assessment analysis, and enhanced artificial intelligence (AI) modeling of solar floating photovoltaic (SFPV) and ground-mounted solar PV (GSPV) systems. Both systems—SFPV and GSPV—are installed, tested, and compared under identical harsh environmental conditions in Bahrain’s Gulf, in Al-Khobar, Saudi Arabia, with a detailed assessment of electric power output, PV panel surface temperature, PV DC voltage, and current, as well as energy yield and efficiency. More so, a hybrid artificial intelligence framework integrating Light Gradient-Boosting Machine (LightGBM), Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM) models, fine-tuned through the utilization of an innovative Brown Bear Optimization Algorithm (BBOA) are also developed to forecast electrical power generation and PV panel surface temperature for both SFPV and GSPV systems. The experiments indicate that the SFPV system improved the average PV electrical power and accumulated net daily electrical energy by 59.25% and 69.70%, as well as reduced the PV panel surface temperature by 32.36% compared to that of the SGPV system, respectively. Moreover, statistical evaluations highlighted the LSTM-BBOA model achieved superior robustness over the investigated AI models (LGBM-BBOA, GRU, LSTM , LGBM) in performance prediction, evidenced by the maximal determination coefficient (R 2 ) of 0.9998 and 0.9999, and the minimal RMSE values of 0.5031 and 0.0007 for predicting the SFPV’s electric power and module surface panel temperature, respectively. Conclusively, the study provides valuable insights into the benchmarking of AI techniques for improving smart-grid integration and operational efficiency of floating solar installations, aligning with the innovation objectives of sustainable energy development.