Abstract: Taking centrifugal pump of mine drainage system as research object, the paper introduced basic principles, related cases, advantages and disadvantages, unsolved problems of four kinds of life prediction methods based on data driven, namely machine learning method, multivariate statistical analysis method, characteristic extracting method and information fusion method, and expounded their applications in centrifugal pump life prediction. Meanwhile it pointed out development trend of the centrifugal pump life prediction: life recession index should be diversified, and all kinds of normal centrifugal pump indexes show that the centrifugal pump works normally, the more comprehensive consideration, the higher the prediction reliability; decision layer information should be highly integrated, and factors such as vibration signal, dynamic friction torque and pump head will change with decay of life, so results will be better to fuse these information into life prediction; feature layer information should be integrated, and integrating various prediction models or establishing a hybrid model with many advantages can better meet industrial requirements.