基于参数和结构优化的置信规则库推理方法研究

Belief rule-base inference methodology using the evidential reasoning (RIMER) approach need to be based on belief rule-base in the inference procedure. Therefore, there is an important means to improve robustness for the RIMER as the core of intelligent decision-making by optimizing the parameters and structure of belief rule-base, which requires urgently quantitative theoretical and methodological guidance. The project aims to make optimization for parameter and structure of belief rule-base as the research background, and proposes parameter optimization approach, structure optimization approach and both parameters and structure of the parallel approach for belief rule-base. The main research contents include: (1) reconstructing belief rule-base parameter optimization model using hierarchy analysis; (2) proposing a new approach for parameter optimization based on the optimization method and swarm intelligence algorithm; (3) creating continuously structure optimization approach based on the multiple criteria decision-making TOPSIS methodology; (4) dividing the state interval of belief rule-base based on belief rule analysis and proposing a new approach for structure optimization based on rough set theory and cluster analysis method; (5) reconstructing or combining parameter optimization approach and structure optimization approach by creating the state transition table for belief rule-base; (6) extending serial computing of optimization approach to parallel computing using space partitioning strategy. The research outcomes have important theoretical and practical significance to the development and perfection of RIMER approach, especially for the optimization theory and methods of belief rule-base.

置信规则库推理（RIMER）方法的推理过程需建立在置信规则库（BRB）的基础上，因此，优化BRB中参数与结构是提高RIMER方法鲁棒性的重要手段，迫切需要科学的定量化的理论和方法指导。本项目拟以BRB参数和结构优化为研究背景，提出BRB的参数优化方法、结构优化方法和兼顾参数与结构的并行优化方法。主要研究内容有：（1）利用层次结构分析重构BRB参数优化模型；（2）结合最优化方法与群智能算法提出BRB参数优化的新方法；（3）创建用于持续优化BRB结构的多准则决策TOPSIS方法；（4）凭借置信规则分析划分BRB的状态区间及联合粗糙集理论和聚类分析方法构建结构优化新方法；（5）创建BRB的状态转移表并依此重构或合并参数优化方法和结构优化方法；（6）利用空间划分策略将BRB串行优化方法扩展成并行优化算法。以上研究成果对发展和完善RIMER方法，尤其是BRB的优化理论与方法具有重要的理论和实际意义。

基于证据推理的置信规则库推理方法（RIMER）在置信规则库的知识表达框架基础上完成决策推理。因此，优化置信规则库的参数和结构对提升RIMER方法具有重要的作用。本项目从参数优化和结构优化两个方面对置信规则库推理方法进行研究，提出置信规则库参数优化的理论和方法、置信规则库结构优化的理论和方法、置信规则库参数和结构联合优化的理论和方法、RIMER方法优化理论和方法，从参数和结构两个方面对置信规则库进行优化从而达到优化RIMER方法的目的。在此基础上，本项目还新增了置信规则库的应用研究，考虑不同应用背景下置信规则库推理方法的应用。本项目共完成学术论文27篇，其中SCIE收录论文7篇，累计培养毕业硕士研究生7名。项目的研究成果在输油管道泄漏、桥梁风险评估等方面取得较高的推理准确性，对相关领域的应用具有一定的理论指导意义和经济价值。