The invention discloses a modeling method for a distributed photovoltaic output power short-term prediction model. The method comprises the steps of obtaining history photovoltaic output data from a power mechanism, obtaining meteorological factor data which correspond to the sampling moment of photovoltaic output from a meteorological monitoring point, and conducting normalization processing; by utilizing the Pearson correlation coefficient analysis method, obtaining the correlation coefficient between the photovoltaic output and illumination intensity, temperature and wind speed data by analyzing; analyzing the meteorological factor data with the correlation coefficient not less than 0.5 through the gray relational analysis grade method, and determining a similar day of the day to be predicted; regarding the photovoltaic output data of the similar day and the meteorological factor data with the correlation coefficient not less than 0.5 as input conditions of the model, and building an epsilon-SVR support vector regression prediction model; determining model parameters through two stages, and obtaining the complete distributed photovoltaic output power short-term prediction model. By means of the method, the problems that in the prior art, a traditional statistical method is adopted for predicting the short-term photovoltaic output, the applicability is low, and the prediction accuracy is low are solved.