基于贝叶斯学习的复杂系统研制风险演化分析  

Complex system development risk evolution analysis based on Bayesian learning

作　　者：徐一帆 吕建伟[1] 史跃东 狄鹏[1] XU Yifan;Lü Jianwei;SHI Yuedong;DI Peng(Department of Management Engineering and Equipment Economics,Naval University of Engineering,Wuhan 430033,China;College of Naval Architecture and Marine,Naval University of Engineering,Wuhan 430033,China)

机构地区：[1]海军工程大学管理工程与装备经济系,武汉430033 [2]海军工程大学舰船与海洋学院,武汉430033

出　　处：《系统工程理论与实践》2019年第6期1580-1590,共11页Systems Engineering-Theory & Practice

基　　金：国家自然科学基金(71401171);装备预研基金项目(6140002050101);军队院校“2110”三期建设项目(4142D4616)~~

摘　　要：大型复杂装备的系统结构和研制流程呈现网络化特征,研究风险演化机理有助于控制风险、降低复杂性.通过系统动态过程建模仿真获取数据样本,运用贝叶斯学习从仿真数据样本中提炼风险演化网络,识别不同风险等级的节点之间存在的关联关系,降低了仅凭经验构建风险网络的主观性.对贝叶斯学习获得的风险网络进行概率推理,在总体高风险等级下计算风险网络节点的风险后验概率分布,进而确定风险演化关键节点和传播链路.最后,通过与复杂网络特征指标评估下的静态特征进行对比分析,研究风险网络动态特征与静态特征的差异性,结果表明网络结构特征和风险传播的动态特征共同决定了风险演化关键节点和传播链路.System structures and developing process of large-scale complex weapon are characterized with networks.Research on risk evolution mechanism is useful to control risk and reduce complexity.Based on dynamic data samples presented by system process modeling and simulations,risk evolution networks was refined by Bayesian learning to recognize correlations among nodes with different risk levels.By this way,subjectivity is reduced than risk networks built only by experience.The risk networks from Bayesian learning was further used to implement Bayesian inference and calculate risk posterior probability distributions of risk network nodes under the conditions of system total risk at high level,and then the critical nodes and propogation chains of risk evolution were identified.Finally,comparing with static characteristics based on complex network characteristic indicators,the differences between dynamic and static characteristics of risk networks were under discussion.It is found that the critical nodes and chains of risk evolution are jointly determined by network structure characteristics and dynamic features of risk evolution.

关 键 词：风险演化 设计结构矩阵 贝叶斯网络 复杂系统 复杂网络 

分 类 号：C931.2[经济管理—管理学]