文献类型：期刊文献

英文题名：Multifactorial Evolutionary Optimization Algorithm Based on Online Knowledge Transfer

作者：Yuan, Jin[1,2]; Feng, Xiang[1,2]; Yu, Huiqun[1,2]

机构：[1] Department of Computer Science and Engineering, East China University of Science and Technology, Shanghai, 200237, China; [2] Shanghai Engineering Research Center of Smart Energy, Shanghai, 200237, China

年份：2023

外文期刊名：SSRN

收录：EI(收录号：20230113786)

语种：英文

外文关键词：Benchmarking - E-learning - Evolutionary algorithms - Knowledge management - Learning algorithms - Learning systems - Optimization - Statistics

摘要：The concept of evolutionary multitasking has gained significant interest in the evolutionary computing community, aiming at solving multiple optimization tasks at once through knowledge transfer in evolutionary algorithms. These algorithms derive common knowledge from different (possibly similar) optimization tasks. However, in many practical situations and scenarios, not all problems are correlated. Mindless knowledge transfer between tasks with low similarity can lead to a degradation of performance, known as negative knowledge transfer. To prevent this, this paper proposes a multitasking optimization framework, called the online transfer multifactorial evolutionary algorithm (OTMFEA), that is equipped with both an online knowledge transfer strategy and an outlier individuals removal strategy. The proposed online knowledge transfer strategy aligns the estimated distribution of offspring with that of the parents in real-time during the optimization process, thus promoting positive inter-task knowledge transfer and suppressing negative knowledge transfer. Additionally, an outlier individuals removal strategy is introduced, where an outlier removal model is built for each task. During the optimization process, the outlier removal threshold is updated through online learning of the offspring's distribution, in order to filter out individuals that contribute to the positive knowledge transfer of the current task and avoid the optimization from converging to a local optimum. To assess the performance of OT-MFEA, experiments were carried out using a set of synthetic benchmark test problems. The results indicate that OT-MFEA accelerates the convergence of the optimization and effectively prevents negative knowledge transfer, achieving a significant performance improvement compared to the other advanced MFO methods. ? 2023, The Authors. All rights reserved.