Global Search of Genetic Algorithm Enhanced by Multi-basin Dynamic Neighbor Sampling

We propose a pioneering enhanced genetic algorithm to find a global optimal solution without derivatives information. A new neighbor sampling method driven by a multi-basin dynamics framework is used to efficiently divert from one existing local optimum to another. The method investigates the rectangular-box regions constructed by dividing the interval of each axis in the search domain based on information of the constructed multi-basins, and then finds a better local optimum. This neighbor sampling and the local search are repeated alternately throughout the entire search domain until no better neighboring local optima could be found. We improve the quality of solutions by applying genetic algorithm with the resulting point as an initial population generator. We fulfill two kinds of simulations, benchmark problems and a financial application, to verify the effectiveness of our proposed approach, and compare the performance of our proposed method with that of direct search, genetic algorithm, particle swarm optimization, and multi-starts.