Evolutionary multitasking with evolutionary trend alignment in subdomains

作     者：Du, Wenhao Ren, Zhigang Cole, Jack Zou, Xiaotian Wang, Chaowen 

作者机构：Xi An Jiao Tong Univ Sch Automat Sci & Engn Xian Peoples R China Univ Exeter Dept Comp Sci Exeter England 

出 版 物：《EXPERT SYSTEMS WITH APPLICATIONS》 (Expert Sys Appl)

年 卷 期：2025年第269卷

核心收录：

学科分类：1201[管理学-管理科学与工程(可授管理学、工学学位)] 0808[工学-电气工程] 08[工学] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：National Natural Science Foundation of China 

主　　题：Evolutionary multitasking Domain adaptation Domain decomposition Evolutionary trend alignment 

摘      要：Evolutionary multitasking (EMT) seeks to address multiple tasks by leveraging inter-task knowledge transfer for an enhanced comprehensive performance. However, its effectiveness may be severely diminished when the tasks involved share little similarity. Some domain adaptation (DA) methods are developed to augment inter-task similarity through learning and applying transformation mappings between tasks. Nevertheless, they generally treat each task as an indivisible domain, which prevents them from fully and precisely leveraging task-specific characteristics. To overcome this limitation, this study proposes a multifactorial evolutionary algorithm based on subdomain evolutionary trend alignment (SETA-MFEA). SETA-MFEA first adaptively decomposes each task into several subdomains, each of which exhibits a relatively simple fitness landscape. A novel DA technique named SETA is then designed and utilized to establish precise inter-subdomain mappings by determining and aligning the evolutionary trends of the corresponding subpopulations. The derived mappings enable various subpopulations, whether within the same task or different tasks, to share a consistent evolutionary trend, so that intra- and inter-task complementary evolutionary information can be transferred through SETA-based intersubdomain crossovers. Extensive testing on several commonly used multitasking/many-tasking benchmark suites, as well as in a real-world application, demonstrates that SETA-MFEA shows competitive performance compared to two single-task evolutionary algorithms and six classic or state-of-the-art EMT algorithms.