多变量混合概率网络分层建模方法及其在决策中应用研究

项目名称： 多变量混合概率网络分层建模方法及其在决策中应用研究

项目编号： No.70801024

项目类型： 青年科学基金项目

立项/批准年度： 2009

项目学科： 金属学与金属工艺

项目作者： 胡笑旋

作者单位： 合肥工业大学

项目金额： 19万元

中文摘要： 在科学研究方面，（1）研究了复杂决策问题的建模流程与方法。分别从复杂决策问题的特征分析、概率网络分层建模方法、决策建模中的知识获取和融合方法、多主体协作方法以及决策任务的调度方法等方面进行了研究；（2）研究了多变量混合的决策模型推理方法。分别研究了多变量混合的贝叶斯网络推理方法、层次影响图推理方法以及面向复杂决策问题的证据推理方法；（3）进行了应用研究。分别研究了贝叶斯网络在复杂系统故障诊断中的应用、模糊贝叶斯网络在电信客户流失预测中的应用，以及综合决策模型在服务供应商选择问题中的应用。 在论著方面，共在国内外学术期刊、学术会议上发表论文27 篇，其中被SCI 检索4篇，被EI 检索15 篇。 在人才培养方面，项目支持了3名博士研究生和3名硕士研究生。

中文关键词： 概率网络；决策建模；知识获取；影响图；贝叶斯网络

英文摘要： As to scientific research, firstly the decision modeling process and methodolgy for complex decision-making problems was considered, including characteristics of complex decision-making problems, hierarchical probability network modeling methods, knowledge acquisition and fusion methods, multi-agent collaborative approaches, scheduling algorithms of decision-making tasks. Secondly, the reasoning methods of decision models with multi-type variables are discussed. Specifically, we studied the Bayesian network inference algorithm for mult-type variables, hierarchical influence diagrams inference algorithm and the evidential reasoning algorithm, respectively. Finally, some applications for real-world decision-making problems are studied, such as the application of Bayesian network for fault diagnosis of complex systems, the application of fuzzy Bayesian network for the loss of customers forecast in telecommunication company and the application of integrated decision model for the vendor selection problem. As to article, 27 papers were published, among which 4 papers were indexed by SCI, 15 papers were indexed by EI. As to student cultivation, this project has supported 3 PhD candidates and 3 graduatestudents.

英文关键词： probabilistic network; decision modeling; knowledge acquisition; influence diagarm; bayesian network