Welcome: for thesis, the well-known premature convergence problem of particle swarm optimization (PSO) was addressed. The primary goals were: (i) efficiently detect the occurrence of premature convergence, (ii) infer the uncertainty per dimension from the swarm state upon detection, and (iii) regroup the swarm within a plausible subset of the original search space defined in proportion to each dimension’s uncertainty. This allows the swarm to be liberated from the state of premature convergence and continue searching for better solutions instead of starting a new search or stagnating in place.
Simultaneously, thousands of combinations of the
(i) inertia weight, (ii) social acceleration coefficient, and
(iii) cognitive acceleration coefficient were tested to determine whether proper parameter selection alone could prevent premature convergence. Results were consistent with Kennedy’s observation that favoring the social coefficient can improve performance , and parameters were discovered that significantly outperformed the traditional Clerc’s equivalents (i.e. c1 = c2 = 1.49618, w = 0.729840); however, swarm regrouping was less problem-dependent and more generally applicable across the benchmark suite than parameter selection at combating premature convergence.
Regrouping PSO (RegPSO) has been published in thesis and presented at the 2009 IEEE International Conference on Systems, Man, and Cybernetics.
Legal disclaimer: the content of this website is protected by copyright laws and should not be misunderstood to be in the public domain. If you make use of any information on this website, please cite thesis or the website itself.
 J. Kennedy, "The particle swarm: social adaptation
of knowledge," in Proceedings of IEEE
International Conference on Evolutionary
Computation, Indianapolis, IN, 1997, pp. 303-308.