摘要
Evolutionary multitasking (EMT) with the ability to tackle multiple different tasks has attracted more and more attention. The transferred knowledge among tasks can simultaneously improve the solving efficiency of all optimization problems in the evolutionary process. However, if the way of knowledge transfer is inappropriate, the negative knowledge transfer will make a bad effect on the performance of EMT algorithms. It is worth studying that how to promote the positive knowledge transfer across tasks. In this paper, a multiobjective multitasking algorithm named EMT-DAVT is introduced. The proposed algorithm contains two components, namely subspace distribution alignment (DA) strategy and decision variable transfer (VT) mechanism. In DA strategy, a learned mapping matrix is utilized to align the distributions in the subspaces and reduces the divergence between subpopulations belonging to different tasks. Then, VT mechanism is implemented which can further promote the positive information transfer. The two proposed strategies interact with each other to improve the knowledge transfer. Finally, the search engine is designed to balance exploration and exploitation. The experimental results on the multiobjective multitasking test suites demonstrate that EMT-DAVT outperforms other classical multiobjective multitasking evolutionary algorithms.
学科与关键词
- Multi-objective evolutionary multitasking algorithm based on cross-task transfer solution matching strategy?Sun, Hao*; Chen, Pengfei; Hu, Ziyu; Wei, LixinISA TRANSACTIONS, 2023, 138: 504-520.被引 3
- An Effective Knowledge Transfer Approach for Multiobjective Multitasking OptimizationLin, Jiabin; Liu, Hai-Lin*; Tan, Kay Chen; Gu, FangqingIEEE Transactions on Cybernetics, 2021, 51(6): 3238-3248.被引 93
- Cultural transmission based multi-objective evolution strategy for evolutionary multitaskingXu, Zhiwei; Liu, Xiaoming; Zhang, Kai*; He, JuanjuanInformation Sciences, 2022, 582: 215-242.被引 14
