Particle swarm algorithms to solve engineering problems: A comparison of performance

Abstract

In many disciplines, the use of evolutionary algorithms to perform optimizations is limited because of the extensive number of objective evaluations required. In fact, in real-world problems, each objective evaluation is frequently obtained by time-expensive numerical calculations. On the other hand, gradient-based algorithms are able to identify optima with a reduced number of objective evaluations, but they have limited exploration capabilities of the search domain and some restrictions when dealing with noncontinuous functions. In this paper, two PSO-based algorithms are compared to evaluate their pros and cons with respect to the effort required to find acceptable solutions. The algorithms implement two different methodologies to solve widely used engineering benchmark problems. Comparison is made both in terms of fixed iterations tests to judge the solution quality reached and fixed threshold to evaluate how quickly each algorithm reaches near-optimal solutions. The results indicate that one PSO algorithm achieves better solutions than the other one in fixed iterations tests, and the latter achieves acceptable results in less-function evaluations with respect to the first PSO in the case of fixed threshold tests.