The data from the operating thermal power units are very large and contains complex noise

which is hard to eliminate the noise thoroughly. To solve this problem

an improved empirical mode decomposition filtering algorithm based on the state recognition(SREMD) was proposed. On the basis of empirical mode decomposition(EMD)

this SREMD algorithm firstly eliminates the significantly abnormal data by using a filtering method which is based on the Pauta criterion. Then

it recognizes the running states of the unit according to continuous change rate of the data. Finally

it performs corresponding EMD denoising according to the steady state and transition process of unit operation

to adapt to the state switching characteristics. Moreover

this algorithm was applied for actual unit operation data. The results show that

the proposed SREMD algorithm effectively completed the pre-processing of the monitoring data of thermal power units and achieved a better denoising effect on the basis of maintaining the overall trend of the signal.