Abstract: This paper proposes a cloud image exposure optimization method based on the all-sky imager,the ground-based cloud map is collected by the self-developed ground-based cloud map collection instrument,combined with cloud images with different exposures continuously shot at the same time,the cloud images are processed using dynamic range optimization algorithms.Perform feature extraction on the optimized cloud image,use image features as the input data of the prediction model,and establish a prediction model based on BP neural network.The verification results show that on the 5 min prediction scale,compared with the persistent model,the root mean square error of the established model is reduced by 14.31%.Compared with the existing research,the model established in this paper has a lower E_RMSE.