Due to the intermittent and variable nature of renewable energy generation, the large-scale integration of renewable energy into the power grid poses challenges to system stability and reliability. Forecasting renewable energy generation is a vital technical method for maintaining the stability and security of the power system and for promoting the efficient utilization of renewable energy sources. This paper analyzes the domestic and international research status and challenges of power forecasting methods, and discusses renewable energy power prediction methods from three dimensions: time scale, prediction object, and prediction principle. It examines the causes of prediction errors from the perspectives of the data input link and the prediction model link. Then it proposes a comprehensive optimization method for renewable energy power forecasting based on multidimensional data fusion and multi-algorithm stacking. Optimization efforts are carried out in three aspects: meteorological source selection (via relevant indicator screening), power prediction algorithm selection (using stacking ensemble for fusion), and establishment of a multi-indicator evaluation system (covering indicators such as correlation and accuracy). The effectiveness of this method has been verified. Moreover, its prediction performance outperforms that of single models across different regions, seasons, and under special weather conditions, highlighting the potential of the hybrid modeling approach in improving the reliability of renewable energy forecasting.