Abstract: Improving the prediction accuracy of solar power generation is of great significance for ensuring the safe dispatch and stable operation of power systems. This paper presents a photovoltaic power prediction model combining empirical mode decomposition（EMD）,principal component analysis（PCA）,and long-term and short-term memory neural networks（LSTM）. Fully consider the five environmental factors that restrict the photovoltaic output power,firstly,use EMD to decompose the environmental factor sequence to obtain the change of the data signal on different time scales,reduce the non-stationarity of the environmental factor sequence;secondly,use PCA to extract the characteristic sequence. The key impact factor is to eliminate the correlation and redundancy of the original sequence and reduce the dimension of the model input. Finally,the LSTM network is used to perform dynamic time modeling of the multivariate feature sequence to realize the prediction of photovoltaic power generation. The eight-month measured data from a power station in Shanxi Province is used for verification. The experimental results show that the prediction model has higher accuracy than the traditional photovoltaic power prediction method.