Prediction of turbine vibration trend based on smooth Support Vector Regression

Rotor-to-stator impact-rub of rotor is a kind of common fault of steam turbine. Traditional method of fault modeling, such as statistical theory and artificial neural network, usually gets a non-linear model because of the complexity of turbine system. Based on the need of analysis for impact-rub degree and fault trend, Support Vector Regression (SVR) arithmetic is imported and used for time series analysis and prediction. SVR suggests a best tradeoff between complexity of model and learning ability to establish a linear model in high-dimension feature space. The model built by SVR is able to reflect the implicit mechanism in the data set of time series. But SVR arithmetic has longer running time and large need of memory, so an improved arithmetic of SVR was imported, which uses smooth method and advances operation capability of standard SVR method, and is very fit for the time series of small sample size. Simulation indicates that smooth SVR method has higher speed, better precision and generalization ability than neutral network, and is effective to analyze the trend of rotor-to-stator impact-rub fault for steam turbine, which has guiding significance for forecasting fault and maintenance of steam turbine.