Research Article
A Kriging Model-Based Expensive Multiobjective Optimization Algorithm Using R2 Indicator of Expectation Improvement
| Input: | | : Maximum number of evaluations by expensive function | | : Ratio of the number of initial samples to | | : Integer used to generate weight vector | | Output: | | : Nondominated solution set | | (1) Initialization: Obtain design variable limits and , number of objective , etc. according to MOP to be solved. Set and . | | (2) Generate sample points in design space using optimal Latin hypercube sampling. | | (3) for = 1 to do | | (4) Calculate the expensive objective function values for sample point | | (5) , i.e., add sample point to | | (6) end for | | (7) Generate weight vectors in objective space and save them in set . | | (8) Find non-dominated solutions in and put them into | | (9) fordo | | (10) fordo | | (11) Using PSO to find the best hyperparameter | | (12) Build Kriging model of the -th objective function based on | | (13) end for | | (14) According to EIR2 indicator, apply PSO to find the best infilling point | | (15) Calculate expensive objective function | | (16) Update | | (17) Find the non-dominated solutions in and put them into | | (18) end for return Output as approximated Pareto set |
|