An Improvement Proposal of Genetic Algorithms Based on Information Entropy and Game Theory

Abstract

In this paper, a hybrid genetic algorithm based on information entropy and game theory was proposed in order to cope with the defect that traditional genetic algorithm easily falls into local optimum. First of all, initial population was generated through the way that population diversity was calculated by information entropy. It, in combination with parallel genetic algorithm, performed evolutionary operation by adopting standard genetic algorithm (GA), partheno-genetic algorithm (PGA), and hybrid genetic algorithm integrating standard genetic algorithm and partheno-genetic algorithm (GA-PGA). Secondly, it operated game of complete information to population at parallel nodes in accordance with information entropy and fitness value of each sub-population, hoping to optimize the whole population. Finally, two program verification functions were introduced, i.e. Rosenbrock and Rastrigin, to analyze performance superiority of algorithm. It shows from the results that, when compared with the traditional genetic algorithm, this algorithm boasts of good optimization ability and solving precision and is characterized by high convergence rate and stability.