KMCGO: Kriging-Assisted Multi-objective Constrained Global Optimization
The Kriging method based single-objective optimization have been preventing the application of some engineering design problems. The main challenge is how to explore a method that can improve convergence accuracy and reduce time cost under the conditions of parallel simulation estimation. For this purpose, a Kriging-assisted multi-objective constrained global optimization (KMCGO) algorithm is proposed. In KMCGO, Kriging models of expensive objective and constraint functions are firstly constructed or updated with the sampled data. And then, the objective, root mean square error and feasibility probability, which will be predicted by Kriging models, are used to construct three optimization objectives. After optimizing the three objectives by the NSGA-II solver, the new sampling points produced by the Pareto optimal solutions will be further screened to obtain better design points. Finally, four numerical tests and a design problem are checked to illustrate the feasibility, stability and effectiveness of the proposed method
KeywordsConstrained global optimization Surrogate models Kriging Infill search criterion Multi-objective optimization
This work is supported by the National Natural Science Foundation of China (No. 51775472, No. 51675197, No. 51575205).
- 6.Jeong, S., Yamamoto, K., Obayashi, S.: Kriging-based probabilistic method for constrained multi-objective optimization problem. In: AIAA 1st Intelligent Systems Technical Conference, pp. 1–12 (2004)Google Scholar
- 12.Tang, B.: Latin Hypercube Designs. Encyclopedia of Statistics in Quality and Reliability. Wiley, New Jersey (2008)Google Scholar