关键词: 贝叶斯网络, 结构学习, 遗传禁忌搜索, GATS贝叶斯网络结构学习, 航班延误模型

Abstract: Based on the study of Bayesian Networks structure learning by Genetic Algorithm（GA） and Tabu Search（TS）, GATS_BNSL, the algorithm of Bayesian Networks structure learning by hybrid GA and TS, is put forward. The method of TS is applied into evolution of populations to descendant populations of Bayesian Networks structure learning based on GA. Instead of crossover and mutation, the TS crossover and TS mutation are suggested. Contrastive experimental results show the learning advantage of GATS-BNSL. Meanwhile, this method is applied to build the flight delay model of a large hub airfield by using real data. Causality of multi-factors that lead to flight delays are displayed with less learning time and higher learning accuracy.

Key words: Bayesian Networks（BN）, structure learning, Genetic-Tabu Search（GATS）, GATS_BNSL, flight delay model