Abstract: To meet the ever-increasing demands of massive data mining that traditional methods are not able to deal with, a new algorithm was proposed based on rough set attribute reduction method and improved classical K-meas clustering algorithm by applying big data mining technology, so as to realize parrallel computation on MapReduce framework of Hadoop platform. Taking the massive data of a 600 MW coal-fired power unit as the mining object, the new algorithm was applied to optimize the adjustable operation parameters that may affect the boiler efficiency under typical load conditions. Results show that the new method can be used to determine the optimal target values of massive operation data, with remarkable effects in energy saving and emission reduction; the optimal target values mined by the new algorithm represent historical optimal reachable values, which may serve as a reference for operation optimization of similar boilers.