Guided hybrid genetic algorithm for solving global optimization problems

Abstract

The paper develops and implements a new algorithm for solving global optimization problems by combining genetic algorithm and quasi-Newton methods, which reproduces guided local search, and combines two successful modifications of the hybrid approach, the first of which BOHGA establishes a qualitative balance between local and global search, the second – HGDN – prevents re-exploration of previously explored areas of search space. In addition, a modified bump function and an adaptive scheme for determining its parameter – the radius of the "deflated" region of the objective function in the vicinity of the already found local minimum - were proposed to speed up the algorithm.