A Study Of Adaptive Surrogate Model Selection Strategy And Filling Criterion In Expensive Single-objective Optimization Algorithm

There are a large number of expensive single-objective optimization problems in real production and life.The expensive single-objective optimization problems require a lot of time and resources for real computation during the optimization process.Therefore,in many cases,satisfactory optimization results cannot be obtained with limited resources and limited time.Through limited sample training,the surrogate model can replace the real computation in a certain environment to solve the large consumption of resources and time in the optimization process.In the study of surrogate models,how to manage the models and how to effectively select individuals for real computation is still a problem that needs to be studied in depth.Therefore,this paper makes the following research for the single-objective optimization problem assisted by surrogate models,and the main work is as follows:1 An adaptive model selection assisted evolutionary algorithm with multiple populations(AMSEA-MP).AMSEA-MP adopts a multiple populations search strategy,which better balances the diversity and convergence of the population.At the same time,an adaptive method is adopted in the aspect of model management.Based on the distance of the objective function value between the individual and the training sample for the adaptive selection model is used to estimate the individual objective function value,thereby improving the accuracy of the estimation.A number of experiments are conducted on the CEC 2005 benchmark problems and on the Polly coding optimization design problem for spread-spectrum radar.The results are compared to other algorithms for solving expensive optimization problems.It shows that the proposed algorithm can obtain better solutions in a limited computational budget of objective evaluation.2 Modified an adaptive model selection assisted evolutionary algorithm with multiple populations(MAMSEA-MP).In order to further improve the accuracy of the model simulating function,both the distance in the objective space and the distance in the decision space are taken into account when estimating the population individuals in the MAMSEA-MP algorithm.The decision space distance parameter is introduced in the algorithm,and when selecting points for real computation,some points with large uncertainty are used for real computation to increase the diversity of samples in the database,which improves the accuracy of model prediction and allows the model to simulate the problem better.The algorithm is also experimented on the CEC2005 test function as well as the Polly coding optimization design problem for spread-spectrum radar and compared with algorithms that have solved expensive optimization problems in recent years.The experimental results show that the algorithm proposed in this paper can acquire better solutions than other algorithms.