Less Is More: A Small-Scale Learning Particle Swarm Optimization for Large-Scale Optimization.

Area of Science:

• Evolutionary Computation
• Optimization Algorithms

Background:

• Large-scale optimization problems (LSOPs) are crucial in evolutionary computation.
• Existing algorithms often use large populations, leading to excessive fitness evaluations (FEs) and hindering population evolution.
• This limits the refinement of solution accuracy in LSOPs.

Purpose of the Study:

• To propose a novel algorithm, small-scale learning particle swarm optimization (SSLPSO), for effectively solving LSOPs.
• To reduce the number of FEs and prolong evolutionary generations for improved solution accuracy.
• To adaptively adjust evolutionary behavior based on the population's state.

Main Methods:

• Developed a small-scale learning mechanism updating only up to two representative individuals per generation.
• Introduced a representative individual selection (RIS) strategy to identify convergence and diversity representatives.
• Implemented a representative individual learning (RIL) strategy with specialized learning methods for selected individuals.
• Proposed an adaptive strategy adjustment (ASA) method for dynamic control of evolutionary behavior.

Main Results:

• SSLPSO demonstrated significantly superior or comparable performance against state-of-the-art large-scale optimization algorithms on IEEE CEC2010 and IEEE CEC2013 test suites.
• The algorithm effectively conserves FEs and extends evolutionary generations.
• Validated its applicability to real-world problems, such as large-scale constrained water distribution network optimization.

Conclusions:

• SSLPSO offers an efficient and effective approach to solving LSOPs.
• The small-scale learning mechanism and adaptive strategies contribute to enhanced solution accuracy and computational efficiency.
• SSLPSO shows strong potential for practical applications in complex optimization scenarios.