Condition prediction of hydroturbine generating units using least squares support vector regression with genetic algorithms

Abstract

The least squares support vector regression (LSSVR), a least squares version of standard support vector regression, is applied in condition forecast of hydroturbine generating units (HGUs) by its vibration signal time series in this paper. An effective LSSVR model can only be built under suitable parameters. A novel approach, named as GA-LSSVR, is proposed in this paper, which searches for the optimal parameters of LSSVR model using real-value genetic algorithms and adopts the optimal parameters to construct the LSSVR model. The peak-peak value (ppv) time series data of the stator vibration signals in HGUs were used as the data set. The experimental results are shown that the GA-LSSVR model outperforms the existing BP neural network approaches and the simple LSSVR based on the mean absolute percent error criterion.