The invention provides a self-correction inspection photovoltaic power short-term prediction method. The method comprises the steps of near ground shortwave radiation prediction, data compiling, radiation-power relation calibration, model retraining inspection judgment and short-term power prediction output. According to the method, the problem of photovoltaic short-term power prediction under thecondition of poor photovoltaic power station data quality, such as lack of the actually measured meteorological monitoring data, abrupt change in photovoltaic power station meteorological conditionsand photovoltaic component dust attachment and other situations, can be solved. According to the technical scheme, the high-precision near ground shortwave radiation produced by the numerical weatherprediction WRF pattern and the actually measured power data of the photovoltaic power station act as the input of the prediction model and then progressive optimization is performed through a series of correction and prediction evaluation so that the precision of the photovoltaic short-term prediction model can be enhanced. The experiment result shows that the algorithm has great universality andcan be greatly used for photovoltaic short-term power prediction of different regions of photovoltaic power stations.