Prior Distribution Selection Criterion in Accelerated Degradation Testing Bayesian Optimization Design Based on Bayes Factors

Accelerated degradation testing (ADT) Bayesian optimization design can determine prior distribution of model parameters by prior information, which avoiding the uncertainty to the result of design brought by assuming parameter selection arbitrarily. However, different people may have different tendencies when choosing prior distributions through the same prior information. It may result in different schemes of optimization design. Hence, the challenge how to obtain prior distribution through prior information effectively is urgent to be solved. The use of Bayes factors is a method of Bayesian model comparison in classical method of hypothesis testing. It quantifies the support for a model over another on the basis of observed data. This kind of technical definition of "support" can be reference to compare prior distributions in ADT Bayesian optimization design. This article proposes a criterion of prior distribution selection in ADT Bayesian optimization design based on Bayes factors, which provides a new path for choosing prior distribution.