A General Easy and Similar Task-aided Evolutionary Optimization Framework for Expensive Problems

Surrogate-Assisted Evolution Algorithms (SAEAs) use surrogate models to approximate true objective functions and they have been successfully used to solve expensive problems. However, when solving expensive problems, the complicated search space makes solvers hard to locate the promising areas. To effectively solve expensive problems, this paper attempts to improve the solving ability and find more promising solutions by learning experience from performing easier tasks. And thus, an Easy And Similar Task-aided Evolutionary Optimization (EASTEO) framework is proposed. Specifically, an easy and similar task is specially designed to obtain more useful knowledge and experience. And then, the solution experience of this created task can be transferred to the original expensive task, which can help evolve the expensive task. This work uses commonly used benchmark test problems to verify the performance of the proposed framework.