Abstract: In this paper, a PV output portfolio forecasting model is constructed by integrating principal component analysis（PCA）, an improved K-means clustering method, dynamic time warping（DTW）, and a long-short term memory（LSTM） neural network. Based on the PCA method to extract the principal component factors of meteorological elements, the improved K-means clustering method and DTW algorithm are innovatively used to generate a set of historical day samples with a high degree of internal correlation and similar weather characteristics to the day to be predicted. Then, the LSTM neural network is combined to build a PV power prediction model based on the selection of similar days, which finally achieves the accurate prediction of power generation of a PV plant in Yunnan. The comparison results with other prediction models show that the combined prediction model constructed in this paper has better prediction performance and broad application prospects.