Abstract: In this paper,a prediction model of PV output interval based on improved ensemble empirical mode decomposition（MEEMD）and quasi affine transformation（QUATRE）optimized BILSTM was proposed. Principal component analysis（PCA）was used to reduce dimension of time series,and K-means clustering algorithm was used to divide dimension reduction data into three types of meteorological data. Then MEEMD was used to decompose the output sequence of each type of PV and input it into the short-term PV output interval prediction model jointly constructed by QUATRE optimized BILSTM neural network and kernel density estimation（KDE）algorithm. Finally,the interval prediction performance of the model is evaluated based on the example of Ningxia photovoltaic power station. The experimental results show that the model can generate a high level of photovoltaic prediction interval and provide reliable decision-making guarantee for the economic and stable operation of the power system.