There are many kinds of energy loss indicators in power plant, and there are some relevance among the various indicators. So extraction of the key indicators plays an important role between in energy loss analysis of power plants and optimal management of thermal power plants. Based on the characteristics of these indicators, the idea of rough sets was applied to the energy loss analysis of thermal power plants, then we proposed a new algorithm -- use fuzzy C means algorithm (FCM) to discrete cluster the energy loss indicators of thermal power plant, and then analysis simplified the results with algorithm Johnson. Real experiments (Chaozhou 1,2 and Ningde 3,4 assembling units which of the same type in the SIS system under the THA working condition)’ results had proved high accuracy and valuable of the algorithm.