Title: An efficient method for Bayesian system identification based on Markov chain Monte Carlo simulation
Authors: Jia-Hua Yang
Addresses: Department of Disaster Mitigation for Structures, College of Civil Engineering, Tongji University, 1239 Siping Road, Shanghai, China
Abstract: This paper proposes an efficient method for identifying a dynamic system using measured accelerations. A practical mathematical model of a dynamic system is developed based on modal superposition for response prediction. To explicitly address uncertainties, system identification is treated as a Bayesian inference problem where the objective is to identify the posterior PDF conditional measured data. Unless a very simple system is considered, the posterior PDF is usually complicated in the sense that its significant region is concentrated in the neighbourhood of an extended and extremely complex manifold. An effective Markov chain Monte Carlo algorithm is developed to sample from the posterior PDF. Given the generated samples, a framework is proposed to systematically consider multiple models whose relative plausibility is quantified by the weightings depending on the PDF values of the samples. It is illustrated that the proposed method can handle both globally identifiable and unidentifiable problems.
Keywords: system identification; Bayesian updating; Markov chain Monte Carlo; MCMC; robust prediction.
International Journal of Lifecycle Performance Engineering, 2019 Vol.3 No.1, pp.20 - 34
Received: 11 Jul 2018
Accepted: 08 Jan 2019
Published online: 26 May 2019 *