An Error-Pursuing Adaptive Uncertainty Analysis Method Based on Bayesian Support Vector Regression

The Bayesian support vector regression (BSVR) metamodel is widely used in various engineering fields to analyze the uncertainty arising from uncertain parameters. However, accuracy of BSVR based on traditional one-shot sampling method fails meet requirements analysis complex systems. To this end, an error-pursing adaptive presented by combining a new scheme. This scheme was improved error-pursuing active learning function that named, herein, adjusted mean square error (AMSE), which guides metamodel’s design experiments (DoE). During process, AMSE combines and leave-one-out cross-validation estimate prediction entire space. Stepwise refinement achieved placing sampled regions at locations with large errors. Six benchmark analytical functions featuring different dimensions were validate proposed method. effectiveness then further illustrated more realistic application overhung rotor system.