To enhance the forecasting capability of surface solar radiation under complex weather conditions, a prediction method based on time-specific forecasting and model fusion is proposed. First, astronomical radiation features are supplemented to reflect the periodicity of solar radiation intensity, and a radiation attenuation coefficient is introduced through weather classification methods to correct astronomical radiation, thereby strengthening its correlation. Secondly, a correlation analysis of the data reveals that surface irradiance is primarily closely related to astronomical radiation height, while other meteorological factors exhibit lower correlations. Consequently, the dataset is divided into several subsets based on temporal nodes to alter data distribution and improve the correlation among other key features.Given that using a single model for predictions across different datasets may lead to discrepancies in results, we employ Stacking algorithms to enhance model generalization capabilities. Additionally, by incorporating cross-validation and augmenting data with added Gaussian noise techniques, we optimize the Stacking model.Experimental results indicate that the optimized Stacking model effectively improves generalization ability while reducing overfitting risks; this proposed method can accurately identify the randomness of surface irradiance under complex weather conditions with an accuracy rate of 95.8% and a qualification rate of 96.7%, representing an improvement of 4% to 19% compared to traditional forecasting methods.