Surrogate-Assisted and Population Characteristics’ Transfer for Expensive Optimization Problems

To save computational costs for expensive optimization problems, an approach of surrogate-assisted expensive optimization problem based on population characteristics’ transfer (SAEO-PCT) is proposed in this paper. The method is based on the framework of multi-task evolutionary optimization. Specifically, the expensive problem to be optimized is regarded as one task, and the other task is based on the cheap surrogate model. The population distribution characteristics and elite individuals in the evolution are regarded as the population characteristics’ knowledge. In the process of evolution, the cross transfer between the tasks is adopted to regularly transfer the population characteristics in the surrogate-assisted optimization task to the time-consuming optimization task. At the same time, the population characteristics are also transferred within the task, and the population renewal is achieved through population reconstruction, to speed up the time-consuming calculation and optimization task. Benchmark experiments show that SAEO-PCT can successfully accelerate the convergence of the time-consuming optimization problems and save computational costs.