A dynamic selective neural network ensemble method for fault diagnosis of steam turbine

A new dynamic selective neural network ensemble method for fault diagnosis of steam turbine is proposed. Firstly, a great number of diverse BP neural network models are produced. Secondly, the error matrix is calculated and the K-nearest neighbor algorithm is used to predict the generalization errors of different neural networks on each testing sample. Thirdly, the individual networks whose generalization errors are in a threshold will be dynamically selected and a conditional generalized variance minimization method is used to choose the most suitable ensemble members again. Finally, the predictions of the selected neural networks with weak correlations are combined through majority voting. The practical applications in fault diagnosis of steam turbine show the proposed approach gives promising results on performance even with smaller learning samples, and it has higher accuracy and efficiency compared with other methods.