Data-Driven Fault Prediction for Electrical Machinery Based on Novel Sample Preprocessing

Focusing on the data-driven fault prediction for electrical machinery, a hyperplane model of one-class SVM using novel sample preprocessing is studied in this paper. By analyzing the distribution characteristics of the support vectors from the one-class SVM algorithm, a hyperplane modeling method with the samples owning some geometry characters is proposed. On this basis, a large number of novel samples, namely the non-support-vectors can be eliminated. The experimental results for electrical machinery show the effectiveness of the presented fault prediction method, and the novel sample preprocessing method can remarkably reduce the training samples and bring higher modeling efficiency.