As the depletion of traditional fossil energy becomes more serious，the use of solar energy for photovoltaic power generation has been an important direction for global countries to adjust their energy structure.There is an urgent need to improve the prediction accuracy of photovoltaic power generation capacity.A short-term prediction model for photovoltaics （PV） coupled with a weather research and forecasting model for solar energy （WRF-Solar ） and irradiance correction is proposed to enhance the accuracy and reliability of short-term prediction of photovoltaic power.Firstly，WRF-Solar is used for numerical prediction of dynamic downscaling weather to obtain future meteorological factors，including irradiance.Then，a random forest （RF） is used to correct the predicted irradiance.On this basis，long-term and short-term neural networks，backpropagation neural networks，and stepwise cluster analysis are employed to establish a short-term prediction model for photovoltaic power.Finally，the actual operation data of a 40 MW photovoltaic power station is used to compare the models.The results show that the irradiance corrected by the RF model is closer to the real value，and the average absolute error rate is reduced by 56.06 percentage points.Compared with the prediction results of the other two models，the model of long-term and short-term neural networks demonstrates the best prediction effect，and the meaan absolute percentage error is increased by 4.13 percentage points，indicating that the combined model can further improve the accuracy of power prediction.