A computationally fast multimodal optimization with push enabled genetic algorithm

Multimodal optimization problems have landscape with multiple global/local optima and the task of a multimodal-optimization algorithm is to locate these optimal points. Population based approaches like genetic algorithms have been successfully implemented to find these optimal points in single run. Techniques adopted by these solvers usually attempt at segregating the population into multiple clusters through a modified selection operator. Due to multimodality, relatively more computational effort in terms of number of function evaluations is required to converge at multiple peaks. In one of the recent studies, a non-uniform mapping approach for binary-coded variables was proposed to attain faster convergence in single-objective unimodal problems. This non-uniform mapping approach acts as a push-operator for genetic algorithms with real-coded variables. In this work, we implement push-operator along with a niching algorithm to solve multimodal optimization problems. Results show significant reduction in the number of function evaluations to reach mostly all optimal points of certain benchmark problems. The work also encourages implementation of push-operator in other evolutionary algorithms.