Diversified Knowledge Transfer Strategy for Multitasking Particle Swarm Optimization.

Area of Science:

• Artificial Intelligence
• Optimization Algorithms
• Computational Intelligence

Background:

• Evolutionary multitasking optimization (EMTO) leverages shared knowledge across tasks for improved efficiency.
• Existing EMTO methods often prioritize convergence, potentially neglecting diversity and leading to local optima.
• The need for strategies that exploit both convergence and diversity knowledge in EMTO is critical.

Purpose of the Study:

• To propose a novel diversified knowledge transfer strategy for multitasking particle swarm optimization (DKT-MTPSO).
• To address the limitations of existing EMTO approaches in exploiting knowledge for diversity.
• To enhance the ability of EMTO to alleviate local optimization problems.

Main Methods:

• An adaptive task selection mechanism manages knowledge transfer based on population evolution.
• A diversified knowledge reasoning strategy captures both convergence and diversity information.
• A diversified knowledge transfer method expands solution regions using varied transfer patterns for comprehensive search space exploration.

Main Results:

• DKT-MTPSO demonstrated superior performance compared to state-of-the-art EMTO algorithms on benchmark test suites.
• Experimental results validated the algorithm's effectiveness in alleviating local optimization.
• The algorithm's practicality was confirmed through a real-world application study.

Conclusions:

• The proposed DKT-MTPSO effectively balances convergence and diversity knowledge for improved EMTO.
• The strategy successfully mitigates local optimization issues inherent in traditional EMTO methods.
• DKT-MTPSO offers a promising advancement for multitasking optimization problems.