Decomposition of Bayesian Networks Based on Hybrid Genetic Algorithm
Hongliang Yao · Jisuanji fangzhen · 2008
Because optimal decomposition of Bayesian Networks is a NP-complete problem,the paper proposes an algorithm of decomposition of Bayesian Network based on hybrid genetic algorithm by phases—PHGA.The PHGA algorithm divides the evolving process into some different stages.With larger population size,crossover probability and smaller selection pressure at early and middle stages the algorithm remarkably improves its ability of global searching and avoids the phenomenon of population prematurity.Moreover,with small population size,crossover probability and larger selection pressure at latter stage,the algorithm also introduces locally optimal operator—climbing hill algorithm which improves the ability of local searching and the speed of convergence.The experiment results show that the algorithm is superior to genetic algorithm and simulated annealing algorithm on three BN.