Clustered Parent Centric Normal Cross-Over for Multimodal Optimization

Genetic algorithms are mainly modeled on basic four steps of parent selection, crossover, offspring evaluation and replacement. It is not possible to model this for direct application to multimodal landscapes. In this paper we propose a novel algorithm in which GA named Parent Centric Normal Crossover is modified and works on a clustered population to tackle multimodal problems. We suggest a dynamic clustering scheme to maintain stable yet variable number of clusters of variable size which can tackle multimodal landscapes, and a Crossover Rate operator in GA for controlled convergence to tackle complex multimodal functions. The algorithm has been tested over widely used benchmarks from single dimension to complex composite functions and compared with other State of the art EAs. The results clearly prove C-SPC-PNX to be a robust multimodal optimization technique.