Photovoltaic grid-connected applications pose challenges to the stable and safe operation of power systems. Solar power predicted by random forest based on numerical weather prediction model is proposed to enable grid-connected planning to balance fluctuations. Using downward solar short-wave radiation flux, temperature, wind speed, and other meteorological parameters obtained from the numerical weather forecast model, PV power forecast models with different initial time and durations under various weather conditions by random forest method are established. The correlations between observed and predicted power in daily, monthly and seasonal change are analyzed and both two powers are compared under three different detailed weather types. The results show that the proposed method can accurately simulate the change of solar power under different weather conditions in the middle-lower plain of the Yangtze River. This model has an acceptable amount of data computation and high prediction accuracy, so that it can apply to engineering.