Multiobjective Multitasking Optimization With Decomposition-Based Transfer Selection.

Area of Science:

• Computational intelligence
• Evolutionary computation
• Optimization algorithms

Background:

• Multiobjective multitasking optimization (MTO) aims to solve multiple MOPs concurrently, leveraging experience transfer for efficiency.
• Inadequate selection of transfer solutions can negatively impact MTO performance, potentially leading to suboptimal results.
• Existing MTO algorithms face challenges in effectively identifying and transferring beneficial search experiences.

Purpose of the Study:

• To propose a novel multiobjective multitasking evolutionary algorithm (MMTEA-DTS) that enhances transfer learning in MTO.
• To introduce a decomposition-based transfer selection mechanism to quantify and select high-potential solutions for knowledge transfer.
• To develop a hybrid transfer evolution method for diversifying search experiences across tasks.

Main Methods:

• The MMTEA-DTS algorithm decomposes MTO tasks into subproblems.
• Transfer potential is quantified using the performance improvement ratio of associated subproblems.
• A hybrid transfer evolution strategy is employed to diversify transferred search experiences.

Main Results:

• MMTEA-DTS demonstrated superior performance on benchmark and real-world MTO problems.
• The decomposition-based transfer selection effectively identified high-potential solutions for knowledge transfer.
• The hybrid transfer evolution method facilitated diversified knowledge sharing across tasks, accelerating convergence.

Conclusions:

• MMTEA-DTS offers an effective approach to improve knowledge transfer in multiobjective multitasking optimization.
• The proposed transfer selection and evolution strategies enhance convergence speed and solution quality.
• The algorithm shows significant advantages over recently proposed MTO methods.