Accurate meteorological information is the foundation for achieving precise prediction of photovoltaic power generation; However, distributed photovoltaic power plants are usually scattered in layout, have small installed capacity, and generally lack dedicated meteorological stations; Based on this, a distributed photovoltaic power short-term prediction method using JSD Informer model was proposed; By analyzing the numerical weather forecast of meteorological stations around the photovoltaic power station, the mapping relationship between the power generation of the photovoltaic power station and its geographic information; Accurately quantify the distribution differences of multi-source meteorological data, integrate numerical weather forecasts from multi-source meteorological stations, and calculate meteorological information data for the location of distributed photovoltaic power stations; Combining the fused meteorological information and utilizing the Johnson Shannon divergence to optimize the sparse self attention mechanism of the Inframer network, a distributed photovoltaic power generation short-term prediction model is constructed; Comparative experimental analysis shows that the proposed method has high prediction accuracy and good robustness. The root mean square error and average absolute error of distributed photovoltaic power prediction calculation are as low as 0.10 and 0.06, respectively.