Particle Swarm Optimization or Differential Evolution-A comparison

In the mid 1990s two landmark metaheuristics have been proposed: Particle Swarm Optimization and Differential Evolution. Their initial versions were very simple, but rapidly attracted wide attention. During the last quarter century hundreds of variants of both optimization algorithms have been proposed and applied in almost any field of science or engineering. However, no broader comparison of performance between both families of methods has been presented so far. In the present paper ten Particle Swarm Optimization and ten Differential Evolution variants, from historical ones from the 1990s up to the most recent ones from 2022, are compared on numerous single-objective numerical benchmarks and 22 real-world problems. On average Differential Evolution algorithms clearly outperform Particle Swarm Optimization ones. Such advantage of Differential Evolution over Particle Swarm Optimization is in contradiction with popularity: In the literature Particle Swarm Optimization algorithms are two-three times more frequently used than Differential Evolution ones. Problems for which Particle Swarm Optimization performs better than Differential Evolution do exist but are relatively few. Although this result may be an effect of the choice of specific variants, experimental settings or problems used for comparison, some re-consideration of algorithmic philosophy may be needed for Particle Swarm Optimization variants to make them more competitive.

Automated summary

This paper compares Particle Swarm Optimization and Differential Evolution, two landmark metaheuristics, and finds that the performance of Differential Evolution algorithms is clearly better than Particle Swarm Optimization ones. Despite being more commonly used in the literature, Particle Swarm Optimization algorithms are outperformed by Differential Evolution on single-objective numerical benchmarks and real-world problems. Therefore, there is a need to reconsider the algorithmic philosophy of Particle Swarm Optimization variants to enhance their competitiveness.