Multiscale Relevance Vector Machine Fault Prediction Based on Genetic Algorithm Optimization

Abstract

As the present fault prediction methods are not accurate enough, a multiscale relevance vector machine (MSRVM) based on genetic algorithm (GA) optimization is proposed. The kernel scales' number and kernel parameters are optimized by GA to improve the performance of the MSRVM. Its feasibility and advantages are proved by the fault prediction of a Buck converter circuit IntroductionThe fault prediction is a more advanced maintenance type than the fault prognosis [1], and it is the core content of prognostics and health management (PHM) of the equipment [2]. It can estimate the remaining useful life of the units or system and predict their future health conditions with the comprehensive utilization of all varieties of data, such as the parameters monitored, use state, current environment and working condition, early experiment data, and historical experience while assisted by all varieties of reasoning technologies, such as mathematics-physics models and artificial intelligence [3]. The fault prediction methods can be classified as the model-driven based method, the data-driven based method, and the reliability and statistics based method [4], from among which the data-driven based method has a broad application perspective because it can predict fault with the test data or the data from the sensors and does not need to build the model of the units or system.