Multi-objective robust optimization using adaptive Kriging


Loading...

Date

2022-04-14

Publication Type

Other Conference Item

ETH Bibliography

yes

Citations

Altmetric

Data

Abstract

Accounting for uncertainties is crucial in the design of engineering systems. Various techniques have been developed for design optimization within a probabilistic framework. In this work, we consider simultaneously robust and multi-objective design optimization. While the former allows one to deal with uncertainties affecting the objective function, the latter allows for handling multiple conflicting objectives. Conservative quantiles are used as a single measure of robustness, trading-off the optimality and degree of robustness of the solution. The quantiles are computed using crude Monte Carlo simulation and are embedded within a classical multi-objective optimization algorithm, namely the non-dominated sorting genetic algorithm (NSGA-II). Such an approach is obviously computationally intensive. To alleviate this burden, we consider the use of surrogate models, and more specifically adaptive Gaussian process models. This approach is eventually adapted to problems with mixed continuous-categorical variables. After a validation on analytical examples, the proposed method is applied to building renovation where the goal is to find the optimal renovation strategy that minimizes both the environmental impact and the life cycle cost of a building. This is carried out in the context of life cycle analysis where there are numerous uncertainties that need to be accounted for.

Publication status

published

External links

Editor

Book title

Journal / series

Volume

Pages / Article No.

Publisher

ETH Zurich, Chair of Risk, Safety and Uncertainty Quantification

Event

SIAM Conference on Uncertainty Quantification (UQ22)

Edition / version

Methods

Software

Geographic location

Date collected

Date created

Subject

Robust optimization; Multi-objective optimization; Optimization under uncertainty; Kriging

Organisational unit

03962 - Sudret, Bruno / Sudret, Bruno check_circle

Notes

Conference lecture held on April 14, 2022.

Funding

Related publications and datasets