Computational Technology Resources - CCC

J. Song, W. Zhang, "A stepwise Bayesian updating approach by enhancing an active learning Gaussian process regression model", in P. Iványi, J. Logo, B.H.V. Topping, (Editors), "Proceedings of the Sixth International Conference on Soft Computing, Machine Learning and Optimisation in Civil, Structural and Environmental Engineering", Civil-Comp Press, Edinburgh, UK, Online volume: CCC 5, Paper 2.5, 2023, doi:10.4203/ccc.5.2.5

Keywords: Bayesian updating, active learning, Gaussian process regression, metamodeling.

Abstract

Bayesian updating framework is regarded as a promising approach for probabilistic calibration and uncertainty quantification, and the main obstacle for practical engineering problems is the high computational cost, especially for time-consuming models. The attractive point of traditional Bayesian updating with structural reliability methods (BUS) is to reformulate Bayesian updating into a structural reliability problem by constructing the limit state function with the likelihood and an auxiliary random variable. This paper proposes a step-wise Bayesian updating approach, by developing a varying observation domain-based strategy to reduce the dimensionality and nonlinearity of the constructed limit state function. In our work, the reliability problem is decomposed into a series of sub-problems by attributing random samples of the auxiliary random variable to each sub-problem. Thereafter, the exponential relation in each limit state function degenerates into squared linear additive type, so as to comprehensively reduce the nonlinearity. To overcome the inefficiency caused by rare event, this paper further develops an active learning procedure based on Gaussian process regression (GPR) to approximate a series of induced limit states as well as the acceptance rates. The main advantage of this procedure is of sharing the common performance function evaluations which can largely reduce the computational cost.

download the full-text of this paper (PDF, 9 pages, 501 Kb)

go to the previous paper
go to the next paper
return to the table of contents
return to the volume description

	Computational & Technology Resources an online resource for computational, engineering & technology publications
	not logged in - login
Front Page Browse CCC CCP CSETS CTR IJRT Other Authors Search Purchase Guide FAQ Contact us	Civil-Comp Conferences ISSN 2753-3239 CCC: 5 PROCEEDINGS OF THE SIXTH INTERNATIONAL CONFERENCE ON SOFT COMPUTING, MACHINE LEARNING AND OPTIMISATION IN CIVIL, STRUCTURAL AND ENVIRONMENTAL ENGINEERING Edited by: P. Iványi, J. Logo and B.H.V. Topping Paper 2.5 A stepwise Bayesian updating approach by enhancing an active learning Gaussian process regression model J. Song^1,2 and W. Zhang^1,2 ¹School of Mechanical Engineering, Northwestern Polytechnical University, China ²State IJR Center of Aerospace Design and Additive Manufacturing, China doi:10.4203/ccc.5.2.5 Full Bibliographic Reference for this paper J. Song, W. Zhang, "A stepwise Bayesian updating approach by enhancing an active learning Gaussian process regression model", in P. Iványi, J. Logo, B.H.V. Topping, (Editors), "Proceedings of the Sixth International Conference on Soft Computing, Machine Learning and Optimisation in Civil, Structural and Environmental Engineering", Civil-Comp Press, Edinburgh, UK, Online volume: CCC 5, Paper 2.5, 2023, doi:10.4203/ccc.5.2.5 Keywords: Bayesian updating, active learning, Gaussian process regression, metamodeling. Abstract Bayesian updating framework is regarded as a promising approach for probabilistic calibration and uncertainty quantification, and the main obstacle for practical engineering problems is the high computational cost, especially for time-consuming models. The attractive point of traditional Bayesian updating with structural reliability methods (BUS) is to reformulate Bayesian updating into a structural reliability problem by constructing the limit state function with the likelihood and an auxiliary random variable. This paper proposes a step-wise Bayesian updating approach, by developing a varying observation domain-based strategy to reduce the dimensionality and nonlinearity of the constructed limit state function. In our work, the reliability problem is decomposed into a series of sub-problems by attributing random samples of the auxiliary random variable to each sub-problem. Thereafter, the exponential relation in each limit state function degenerates into squared linear additive type, so as to comprehensively reduce the nonlinearity. To overcome the inefficiency caused by rare event, this paper further develops an active learning procedure based on Gaussian process regression (GPR) to approximate a series of induced limit states as well as the acceptance rates. The main advantage of this procedure is of sharing the common performance function evaluations which can largely reduce the computational cost. download the full-text of this paper (PDF, 9 pages, 501 Kb) go to the previous paper go to the next paper return to the table of contents return to the volume description
Back to top	©Civil-Comp Limited 2023 - terms & conditions