Vulnerability Assessment of Equipment Excited by Disturbances Based on Support Vector Machine and Gaussian Process Regression

摘要

In this article, a nonparametric vulnerability assessment method is proposed for the equipment excited by transient electromagnetic disturbances. By carrying out the proposed method, the probability distributions of multistate failure thresholds could be obtained for the cases of lacking prior knowledge, such as the effect mechanism, related distribution characteristics and training data. In fact, the proposed method combines support vector machine (SVM) and Gaussian process regression (GPR). SVM may extract the classification hyperplane from a small number of test samples, and obtain the continuous classification indicators subsequently. Furthermore, the regression results of classification indicators based on GPR can provide both mean values and their confidence intervals. To deal with the multilevel effect assessment, the "one against the rest" strategy is applied and probabilities of each level can be assessed simultaneously. Finally, a case study of an electronic system is carried out to illustrate the applicability and effectiveness of the proposed method.