A sequential niching memetic algorithm for continuous multimodal function optimization

摘要

This work discusses a sequential niching algorithm for multiple optimal determination. The procedure consists of a sequence of genetic algorithms (GA) runs, which incorporate a gradient-based hill-climbing algorithm, and make use of a derating function and of niching and clearing techniques to promote the occupation of different niches in the function to be optimized. Thus, the algorithm searches the solution space eliminating from the fitness landscape previously located peaks forcing the individuals to converge into unoccupied niches. An effective search of the solution space is stimulated incorporating in the algorithm stages dedicated to find new promising domains in the variable space and stages that exploits the located promising regions. Unlike other algorithms the efficiency of the sequential niching memetic algorithm (SNMA) proposed in this work is not highly sensitive to the niche radius. Performance measurements with 37 standard test functions of dimensions ranging from 1 to 100 show that the SNMA has very good scalability and outperforms other algorithms in accurately locating multiple optima, both global and local.