Photovoltaic(PV) generation prediction has great significance for the stability and security of power grid after the PV grid-connected. This paper puts forward a model of very short-term photovoltaic power forecasting method based on similar days and wavelet neural networks. By using historical weather information from the PV power generation system, meteorological feature vectors are established,and similar days are found based on Computation Grey Correlation Degree. Autocorrelation analysis was used to discover historical output power which has great relation with predicted output power.The input vectors of WNN predictive model were historical output power, wind speeds,irradiance,temperature,To predict the output power of the forecasting time directly.The simulation result show that this model has high accuracy,and can provide an effective and feasible way to forecast the PV system very short-term power output.