An Improved Method for Comprehensive Learning Particle Swarm Optimization

Particle Swarm Optimization (PSO) is a population-based stochastic search technique for solving optimization problems, which has been proven to be effective in wide applications in scientific and engineering domains. However, it is inefficient when searching in complex problems spaces. Lots of improved PSO variants with different features have been proposed, such as Comprehensive Learning PSO (CLPSO). CLPSO is an enhanced PSO variant by adopting a better learning strategy that has some chance to choose other particles' historical best information to update velocity. Comparing with the standard PSO, CLPSO has successfully improved the diversity of population and hence avoids the deficiency of premature convergence and local optima. However, this algorithm causes slow convergence speed, especially during the late state of searching process. In this paper, an improved CLPSO algorithm is proposed, termed as ICLPSO, to accelerate convergence speed and keep diversity of population at the same time. We set the learning probability based on particles' own fitness and adaptively construct different learning exemplars for different particles according to particles' own features and properties, which is a more appropriate learning strategy for particles' optimization. Experimental results show that the performance of ICLPSO is better than standard CLPSO and some other peer algorithms, using the functions both on unimodal and multimodal.