Multitask Level-Based Learning Swarm Optimizer.

Area of Science:

• Artificial Intelligence
• Computational Intelligence
• Optimization Algorithms

Background:

• Evolutionary multitasking optimization (EMTO) leverages correlations between tasks for simultaneous optimization.
• Existing EMTO algorithms primarily use Differential Evolution (DE) and Genetic Algorithms (GA).
• Particle Swarm Optimization (PSO) offers faster convergence, particularly in later evolutionary stages, yet is less explored in EMTO.

Purpose of the Study:

• To propose a novel EMTO algorithm based on Particle Swarm Optimization (PSO).
• To introduce the Multitask Level-based Learning Swarm Optimizer (MTLLSO).
• To enhance knowledge transfer effectiveness in evolutionary multitasking.

Main Methods:

• MTLLSO maintains multiple populations, each optimizing a single task using Level-based Learning Swarm Optimizer (LLSO).
• High-level individuals (better fitness) guide low-level individuals (worse fitness) within and across populations.
• Information transfer uses high-level individuals from a source task to guide low-level individuals in a target task.

Main Results:

• MTLLSO demonstrated effectiveness on the CEC2017 benchmark problems.
• The proposed MTLLSO significantly outperformed other compared algorithms.
• The algorithm achieved a satisfying balance between self-evolution and knowledge transfer.

Conclusions:

• MTLLSO represents a promising new approach to evolutionary multitasking optimization using PSO.
• The level-based learning and targeted knowledge transfer mechanisms are effective.
• MTLLSO shows superior performance compared to existing methods in benchmark tests.