Parallel Genetic Algorithm Based Crowding Scheme Using Neighbouring Net Topology
Pradipta Kumar Nanda, Priyadarshi Kanungo, Durga Prasad Muni · 2003
In this article, the notion of crowding is em- ployed to maintain stable subpopulations at niches of a multi modal nonlinear function. In this work, we have attempted to parallelize the crowding scheme and hence, propose new concepts of net topology while devising the parallel scheme based on coarse grained parallelization. Besides, we have also proposed a new interconnection model which takes care of the intra deme migration. The use of Generalized Crossover (GC) operator is found to be superior to that of the scheme using two point crossover operators. The effect of different net topology, based on the neighbourhood structure, upon the solution is investigated. It was found that the net topology with second order neighbourhood structure is good enough to yield satisfactory results. faster than that of the net using only the inter deme migration. Generalized Crossover operator (GC) as proposed by Nanda et al(7) is introduced to explore the diversity and the quality of the solution. The effect of the net structure with different neighbourhood is investigated and compared with that of the fully connected network. It is observed that the network with the second order neighbourhoodproduces satisfactory results. Since, the migration is allowed among the demes in the selected neighbourhood structure, the computational burden is substantially reduced as compared to a fully connected net. Although, effect of migration policies, rate of migration, number of demes and size of the demes on the quality of the solution has been investigated, for the sake of illustration simulation results are presented only for the migration policy where the Good migrants of a demes replaces the bad migrants of other demes.