Abstract: In view of the unstable output power of photovoltaic power generation and the difficulty of implementing the power generation prediction model at the present stage,the short-term photovoltaic power generation prediction method based on long and short memory neural network optimization is studied. By analyzing the distribution characteristics of neural network,the initial data is substituted into the data optimization model,the target weight is calculated iteratively,and the selfcyclic product method is introduced to obtain the optimal optimization function of the model. The separability is calculated through the class spacing between the data to be predicted,the data is divided into comparison sequence and reference sequence,the class cluster correlation degree of the data at each unit time in the reference sequence is analyzed,and the data weight at the next time is extracted according to the quantitative value of correlation degree so that the prediction of short-term photovoltaic power generation data is completed. Simulation results show that the proposed method has high prediction accuracy, intuitive model structure, easy implementation,strong data inclusiveness,and can effectively realize power generation data prediction.