Accurate solar irradiance forecasting is essential for renewable energy in tropical regions like Java-Bali, where weather variability poses major challenges. This study compared statistical (SARIMA) and neural network-based models (LSTM, GRU, NARNET, WNN), along with hybrid approaches, to identify the most effective prediction method. SARIMA was selected for its ability to capture consistent seasonal and linear trends, while NNVs model nonlinear relationships, especially in unstable weather. The proposed models were benchmarked against persistence and ARIMA baselines. The hybrid SARIMA-NARNET model achieved superior accuracy, with an MAE of 1.9287 W/m², RMSE of 2.5197 W/m², and a remarkably low MAPE of 0.3084%. Additionally, the DCL strategy demonstrated adaptive energy management, yielding daily energy savings of 16%–17% compared to static methods, with even greater efficiency at extreme confidence levels. These findings highlight the potential of hybrid modeling and adaptive control for optimizing solar energy use in tropical climates.