In order to diagnose the economic performance of unit online， this paper presents a new algorithm to forecast the exhaust enthalpy in the steam turbine online based on support vector machine（SVM）. This online forecasting method establishes the complicated relation model between the steam turbine exhaust enthalpy and the relative operating parameters， considering the coupling performance of every parameter， and has advantages of high forecasting accuracy， global optima property and more generalized performance. Applying this SVM method to a 200 MW unit and a 300 MW one， the SVM model has been trained to forecast the steam turbine exhaust enthalpy in the test samples set. For the 200 MW unit， the mean square root error and the mean relative error is 0.110% and 0.101%， being 38.87 and 38.11 percent of those from the BP network model， 52.38 and 49.75 percent of those from the RBF network model， respectively; Simultaneously， for the 300 MW unit， the mean square root error and the mean relative error is 0.057% and 0.069%， being 29.61 and 25.45 percent of those from the BP network model， 41.57 and 34.97 percent of those from the RBF network model， respectively. These results show that the SVM method could achieve higher accuracy than the BP and RBF neural networks and prove satisfactory in the forecasting demand very well.