Abstract: Aiming at the fluctuation of photovoltaic output will seriously affect the stable operation of power system when a high proportion of PV is connected to the power grid,a combined photovoltaic output short-term prediction model based on mean impact value and improved particle swarm optimization neural network was proposed. Firstly,the MIV-PSO-BPNN model was developed by selecting the total radiation,direct radiation,scattering radiation,relative humidity,temperature,wind speed and rainfall,as the main factors affecting on the PV power generation. Based on the Rapid Miner data mining,the mean impact value of rainfall on PV output is0.0099,the influence is small and is not considered as the model input variables. Then,the improved particle swarm optimization（PSO）was applied to optimize the weights and thresholds of BP neural networks. Finally,the data of photovoltaic power station of Shanghai T2-2 Pudong International Airport was used for verification. The comparison results show that the MIV-PSO-BP neural network model can used effectively for PV power generation prediction,and has application value in practice.