The transient output prediction of solar power source is of great significance for power grid stability analysis, power quality control and fault diagnosis. To this goal, the discrete model and linear prediction model of ideal solar power source were established. Then, a regularized least square prediction scheme was proposed to estimate the unchanged model parameters. When the power source model parameters vary, the prediction model parameters are continuously updated in real-time by the Sliding Rectangle Window (SRW) Recursive Least Square (RLS) method. Unlike the standard RLS, SRW-RLS adopts a data update strategy based on sliding rectangle window, which improves the tracking performance and prediction accuracy. The experimental results show that the proposed prediction schemes achieve good prediction accuracy and SRW-RLS is able to adapt well to the changes in the parameters of power source model.