Survey of Evolutionary Transfer Optimization Algorithms

Evolutionary Algorithm(EA) are heuristic algorithms that simulate natural biological evolution and exhibit good searchability and flexibility. They have been successfully applied to solve complex optimization problems. However, while solving the problem, EA default the prior knowledge to zero. However, as target problems seldom exist in isolation, experience gained from a task can be transferred to other related tasks. Evolutionary Transfer Optimization (ETO) algorithms utilize knowledge learning and transferring in related fields to achieve improved optimization efficiency and performance. This study introduces a basic classification of ETO algorithms. The core strategies, advantages, and disadvantages of the mainstream ETO algorithms are sorted out and analyzed from five perspectives: Source task selection, knowledge transfer, narrowing the search space difference, evolutionary algorithm search, and evolutionary resource allocation. Relevant papers on ETO published from 2014 to 2021 are retrieved through the China National Knowledge Infrastructure(CNKI) and Web of Science(WOS).The knowledge graph is used for data mining, information processing, knowledge measurements, and graph drawing. Based on the development trend of ETO and experience, the main challenges and future research directions are summarized.