In recent years,agent model assisted optimization algorithms for solving expensive constraint problems have attracted more and more attention.In this kind of algorithm,the strategy of selecting individuals for real con-straints and objective function calculation will directly affect the solution results of the algorithm.However,the cur-rent algorithm is not rigorous enough to update the model.In order to obtain a better feasible solution when the number of expensive evaluations is limited.In this algorithm,the feasible rule method is used for environment selec-tion,and an adaptive filling criterion is proposed according to whether the feasible solution is found so far.The fill-ing criterion of this method is:when the sample database does not find a feasible solution,use the feasibility proba-bility to select the individual with the greatest probability for real evaluation.Otherwise,if at least one feasible solu-tion is found,the individual with the largest constraint expectation improvement is selected to evaluate the real ex-pensive objective function and constraint function.This method improves the accuracy of Gaussian process model and the convergence ability of evolutionary population.In the experimental comparison,the operation results of 10 test functions and I-beam design optimization problems show that the algorithm has better optimization ability than the existing algorithms for expensive constrained optimization problems.
关键词
代理模型/约束优化/可行性概率/约束期望提高
Key words
surrogate model/constrained optimization/probability of feasibility/constrained expect improvement
维普
万方数据