APPLICATION OF REPLICA EXCHANGE MCMC METHOD TO BAYESIAN STRUCTURAL MODEL UPDATING BY DUAL SAMPLING METHOD

Abstract

In Bayesian structure model update, it is not possible to assume the independence in the posterior distribution of each parameter in the uncertainty estimation due to the correlation between the parameters. To overcome this, dual sampling methodology that estimates the simultaneous posterior distribution tail space separately from the expected value estimation has been proposed, however the existing second stage tail space estimation using GA has a problem in term of computational efficiency. In this study, authors introduced the replica exchange MCMC method, which is a typical wide-area spatial estimation method, as the second-stage tail space estimation method, and verified its effectiveness using the same actual bridge data as in previous study. As a result, it was clarified that the proposed method can estimate the tail space of the posterior distribution as well as the existing GA method with about 1/4 of the calculation time.