Fault prediction of power electronic circuits based on multi-scale relevance vector machine

Aiming at the issue that fault prediction of power electronic circuits is not accurate enough, a method based on multi-scale relevance vector machine (MSRVM) is proposed. Then the kernel parameters are optimized by Genetic Algorithm (GA) to avoid the negative affect on the performance of MSRVM by the non-ideal ones. The feasibility and advantages of MSRVM are proved by the fault prediction of a Buck converter circuit.