融合专家综合知识的贝叶斯网络参数学习方法  

Bayesian Network Parameter Learning Method Integrating Comprehensive Knowledge of Experts

作　　者：鹿浩 张鸿洲[1,2] 钟寒 宿亚杰 石晶莹 LU Hao;ZHANG Hong-zhou;ZHONG Han;SU Ya-jie;SHI Jing-ying(College of Information and Network Safety, People's Public Security University of China, Beijing 100038, China;Public Security Risk Prevention and Control Ministry of Education Engineering Research Center, People's Public Security University of China, Beijing 100038, China;Zhongshan Branch of Dalian Public Security Bureau, Dalian 116001, China)

机构地区：[1]中国人民公安大学信息网络安全学院,北京100038 [2]中国人民公安大学公共安全风险防控教育部工程研究中心,北京100038 [3]大连市公安局中山分局,大连116001

出　　处：《科学技术与工程》2021年第21期8944-8950,共7页Science Technology and Engineering

基　　金：国家重点研发计划(2016YFC0801003);公安部技术研究竞争性遴选项目(2019JZX005)。

摘　　要：贝叶斯网络模型是经典概率图模型,目前已经广泛应用到各个领域中。在贝叶斯网络模型的参数获取方面,以往的方法往往仅依靠数据集或者专家知识。实际情况中,数据集往往存在缺失或者存在噪声,而单一专家指定的参数存在较大的主观因素,两类方法所得参数与实际存在较大偏差。提出基于D-S证据理论的专家综合知识结合小数据集的贝叶斯网络参数学习方法,克服了依靠数据集小的情况下造成的参数不准确或依靠单一专家知识主观性较强的问题。通过实验验证,提出的方法在小数据集的情况下,所获得的贝叶斯网络参数更为准确。并将提出的方法用于公安机关刑事案件线索研判,综合专家知识与小数据集获取模型参数,研判结果能够较好地反映实际情况,证明了方法的有效性。Bayesian network is a classic probability graph model,which has been widely used in various fields.In terms of parameter acquisition of Bayesian network model,previous methods often rely solely on dataset or expert knowledge.In actual situations,the dataset is often missing or noisy,and the parameters specified by a single expert have large subjective factors,and the parameters obtained by the two types of methods have large deviations from the actual situation.A Bayesian network parameter learning method based on D-S evidence theory with multiple expert knowledge combined with small data sets was proposed,which overcomes the problem of inaccurate parameters caused by small data sets or strong subjectivity of relying on single expert knowledge.Through experimental verification,the method proposed can obtain more accurate Bayesian network parameters in the case of small data sets.The method was used in the research and judgment of criminal case clues in public security organs,and the model parameters were obtained by integrating expert knowledge and small data sets.The research and judgment results can better reflect the actual situation and prove the effectiveness of the method.

关 键 词：贝叶斯网络 D-S证据理论 参数学习 刑事侦查 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]