基于差分进化策略的贝叶斯网络结构学习方法  

Bayesian Network Structure Learning Method Based on Differential Evolution Strategy

作　　者：牛艳飞 马洁[1] NIU Yan-fei;MA Jie(School of Automation,Beijing Information Science&Technology University,Beijing 100192,China)

机构地区：[1]北京信息科技大学自动化学院,北京100192

出　　处：《计算机仿真》2021年第1期242-246,255,共6页Computer Simulation

基　　金：国家自然科学基金资助项目(61273173)。

摘　　要：针对目前主流的利用启发式搜索算法进行贝叶斯网络结构学习时,初始种群难以确定且容易陷入局部最优的问题,提出了基于部分互信息和改进差分进化算法相结合的混合算法。算法首先利用节点之间的部分互信息为依据构建初始种群,再将动态因子引入差分进化算法平衡了算法的全局寻优和局部搜索能力,最后对贝叶斯网络结构进行寻优。在两个标准网络Asia和Car网络中进行仿真,并与遗传算法和爬山算法进行对比,仿真结果表明算法在冗余边、缺失边、反向边以及算法的学习性能方面均有不同程度的提升,算法能够得到较好的贝叶斯网络结构,并有更高的数据拟合度。A hybrid algorithm based on partial mutual information and improved differential evolution algorithm is proposed to solve the problem that the initial population is difficult to be determined and easy to be trapped into the local optimum when using heuristic search algorithm to learn the structure of Bayesian network.Firstly,the initial population was constructed based on the mutual information among nodes,and then the dynamic factors were introduced into the differential evolution algorithm to balance the global and local search capabilities.In two standard networks,the Asia and the Car network simulation experiments were implemented and compared with the genetic algorithm and Hill-Climbing algorithm.The results show that the algorithm has different degrees of improvement in redundancy,missing,reversed side,and learning performance,and the algorithm can get better Bayesian network structure and has a higher data fitting.

关 键 词：贝叶斯网络 结构学习 部分互信息 差分进化算法 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]