A New Fitness-Landscape-Driven Particle Swarm Optimization

Fitness landscape is an evolutionary mechanism and fitness landscape theory has developed considerably since it was proposed by Sewall Wright in the 1930s. In evolutionary algorithms, some characteristic information by analyzing the fitness landscape can be obtained to improve the optimization performance of algorithms. This paper introduces a new fitness-landscape-driven particle swarm optimization (FLDPSO). In the method, the correlation metric between fitness value and distance is obtained by charactering the fitness landscape of optimization problems. Then, two new proposed variants of particle swarm optimization (PSO) are developed to improve the optimization performance. Moreover, a selection mechanism based on this metric is introduced to select a fitter variant from these two variants. Finally, the experimental simulation is executed on 18 benchmark functions to assess the optimization performance of the proposed FLDPSO algorithm. The results show that FLDPSO can improve optimization accuracy and convergence very well.

Automated summary

Fitness landscape is an evolutionary mechanism that can improve optimization performance by analyzing the fitness landscape. This paper introduces a new fitness-landscape-driven particle swarm optimization algorithm, characterizing the fitness landscape to improve optimization performance and introducing a selection mechanism for choosing better variants. Experimental results show that the proposed algorithm significantly improves optimization accuracy and convergence.