Particle swarm optimization with a modified learning strategy and blending crossover
Proceedings of the IEEE Symposium Series on Computational Intelligence, SSCI 2017
Particle Swarm Optimization (PSO) is a simple yet elegant derivative-free algorithm for solving continuous, multi-modal, non-convex and multi-dimensional optimization problems of widely different nature. However, one concern about the conventional PSO is that, it suffers from premature convergence, due to quick loss of diversity. In this paper, we propose an improvised version of the standard PSO (PSO-ML), which integrates a novel learning strategy with a genetic crossover scheme to circumvent this limitation. The suggested novel learning strategy generates a single robust exemplar vector by dynamically learning from individual dimensions of three guiding solutions: personal best, global best, and local best for each particle. A genetic crossover scheme, the blending crossover is also integrated with the PSO model, to enhance exploration through rapid search of the function space between and around each pair of particles in the swarm. PSO-ML is tested on 25 standard benchmark functions of the IEEE CEC (Congress on Evolutionary Computation) 2013. The results are then compared against other state-of-the-art algorithms, thus illustrating the advantages of PSO-ML in terms of accuracy and computational cost.
Panda, Aditya; Mallipeddi, Rammohan; and Das, Swagatam, "Particle swarm optimization with a modified learning strategy and blending crossover" (2018). Conference Articles. 103.