In order to research the efficiency degradation of condensing steam turbines

the calculation and prediction of exhaust enthalpy in conventional research methods is avoided

and two concepts

namely the efficiency reference period and the efficiency degradation index

are put forward

by analyzing the working principle of steam turbines. Based on historical data

a model of the turbine in efficiency reference period is established by deep belief network. The initial data are filtered steadily by limiting the range of data in continuous sliding intervals

and the model is trained by the initial data combined with the steady data. By comparing the real-time output power of the turbine with the ideal output power of the model

the degradation degree of the real-time turbine efficiency relative to the reference period is obtained. The application results show that

the spiral decline trend of turbine efficiency degradation index before overhaul is consistent with the slow decline of turbine performance over time

and the rapid rise of turbine efficiency degradation index after overhaul is consistent with the increase of turbine current capability after overhaul. This method can effectively characterize the efficiency trend of the turbine

and provide an effective reference for users to master the performance of the turbine in real time.