一种用于BN结构学习的自适应遗传退火算法

模拟退火算法（SAA）和遗传算法（GA）作为智能算法是结构学习的重要方法．针对两种典型算法存在收敛速度慢或过早陷入局部最优的问题，利用GA进行选择，通过SAA进行搜索并利用独立性测试信息自适应引导算法的进化，提出一种自适应遗传模拟退火算法（AGSAA），应用于贝叶斯网络（BN）结构学习．仿真结果表明AGSAA在学习的准确性和运行效率上均要优于SAA．

Adaptive genetic annealing algorithm for BN structure learning

Simulated annealing algorithm （SAA） and genetic algorithm （GA） as the intelligence algorithm are the key method of structure learning. Aimed at the problem that the rate of convergence is slow and/or get involved into the local optimization untimely in those two typical algorithms, this paper puts forward an adaptive genetic simulated annealing algorithm （AGSAA）, available for Bayesian networks （BN） structure learning, by using GA to select, using SAA to search and utilizing the independent test information to guide adaptively the evolution of algorithm. Simulation results show that AGSAA is superior to SAA in accuracy of learning and operational efficiency.