Abstract: Accurate prediction of photovoltaic power generation is conducive to improvement of dispatching management for the PVconnected power grid,but the present PV prediction method is low in accuracy and weak in adaptability to different weather types. This paper explores and presents a prediction method which combines the similar day and immune genetic neural network（IGA-BP）.Under the frameworks of the method,based on the weather type,temperature and wind speed, a weather similarity day discrimination model combining grey correlation analysis and cosine similarity index is constructed;and with the meteorological characteristic vector of similar days as the input,the IGA-BP power prediction model is established. The prediction accuracy of the IGABP model proposed in the paper is compared with that of the GA-BP and BP models by using the measured data. The results show that the IGA-BP model has high accuracy under different weather types,with an average RMSE of 14.142% and an average TIC of 0.01758 respectively,which are better than other models,indicating that the IGA-BP model can improve the power prediction accuracy and also has high adaptability.