摘要
针对贝叶斯网络构建过程中先验知识的获取问题,将AHP/D-S证据理论引入到贝叶斯网络参数学习中。设计了应用AHP/D-S证据理论整合专家先验知识,综合单调性约束和近等式约束进行参数学习的算法,并进行了仿真案例研究。结果表明:该算法从原理上能够进一步提高贝叶斯网络参数学习的精度,仿真结果也明显优于极大似然估计和无先验信息的最大后验估计的结果,为贝叶斯网络参数学习过程中先验知识的获取提供了一种新的方法。
Aiming at the problem of prior knowledge acquisition in the process of Bayesian network construction,AHP/D-S evidence theory was introduced into Bayesian network parameter learning.An algorithm was proposed for parameter learning by using AHP/D-S evidence theory to integrate expert prior knowledge and integrating monotonic constraints and near-equal constraints.And simulation cases were studied which indicated that applying AHP/D-S evidence theory to integrate expert prior knowledge can improve the accuracy of Bayesian network parameter learning in principle,and the si-mulation results are obviously better than maximum likelihood estimate(MLE)and maximum a prosterior(MAP)without prior information.This offers a new method for acquiring prior knowledge in the Bayesian network parameter learning process.
作者
魏曙寰
曾强
陈砚桥
WEI Shu-huan;ZENG qiang;CHEN Yan-qiao(College of Power Engineering, Naval Univ. of Engineering, Wuhan 430033, China;Unit No. 91697, Qingdao 266400, China)
出处
《海军工程大学学报》
CAS
北大核心
2021年第6期19-24,共6页
Journal of Naval University of Engineering
基金
国家自然科学基金资助项目(51779262)
海军工程大学自然科学基金资助项目(425517K156)。
关键词
D-S证据理论
贝叶斯网络
参数学习
单调性约束
近等式约束
D-S evidence theory
Bayesian network
parametric learning
monotonic constraint
approximate equality constraint