Abstract: In order to solve the uncertainty of fuzzier parameter, a fuzzy rough clustering method based on generalized granularity self-encoder is proposed. The segmentation threshold of each cluster was optimized based on shadow set, and all patterns were divided into different approximate regions. The multiple granularity approximation region was used to capture the uncertainties caused by fuzzy parameters, including the uncertainty caused by the fuzzy coefficient, the fuzziness generated by the boundary region and overlapping regions. Furthermore, a multiple level granularity self-encoder was established to evaluate the quality of the clustering model. The experimental results on multiple datasets show that this method can effectively mine uncertain information and improve the clustering performance.