基于动态不确定因果图的复杂系统故障诊断推理与决策方法研究

Aiming at solving the problems in fault diagnosis of large-scale complex systems, such as modeling complexity and computational overload, etc., this study systematically investigates the algorithm and mathematical foundation of Weighted Logic Inference theory based on Dynamic Uncertain Causality Graph. WLI provides mechanisms for compact representation of uncertain multi-valued causality, and accurate and efficient diagnostic inference under situations of incomplete information. Motived by the inhomogeneous structural characteristics which commonly emerge in real complex networks, this study intends to capture the fault spreading behaviors of complex system and identify the highly influential nodes in fault spreading process on the basis of diagnosis and prediction network. The dynamics of structural load distribution and node’s uncertain causal influence are measured under conditions of real-time failures, so as to formulate the dynamic adaptive decisions. The intuitive graphical causality representation makes the diagnostic reasoning and decision-making processes accessible and reliable. It is promising that the proposed method can suggest operators with prompt and correct decisions of on-line fault localization, fault prediction, protective measures and system recovery in a targeted manner.

本项目针对大规模复杂系统故障诊断建模及推理运算的复杂性等问题，为满足多赋值情况下不确定性因果知识的简洁表达以及不完备信息下的精确、高效的诊断推理要求，在动态不确定因果图的基础上进行权重逻辑推理算法其数理基础的系统化研究。依据复杂网络中常见的非同质结构特性，本研究以故障诊断与预测网络来建模复杂故障传播行为，并通过测度实时故障工况下的网络结构负载分布和节点因果影响的动态性来识别具有高传播影响力的节点，从而形成动态适应性决策。该研究将诊断推理和决策的过程以直观可解释的因果图形式呈现，可在复杂系统在线运行不中断的情况下快速、准确地定位故障根源、预测故障发展，有针对性地给出系统保护和修复策略。

本研究面向核电工业等大规模复杂系统的高可靠性运维决策需求，针对其中广泛存在的不确定性、时变性、数据不足等挑战，着力于克服当前智能故障诊断决策方法在工程应用常见的模型建造困难、运算复杂度高、可解释性差等问题。课题基于动态不确定因果图进行故障传播因果网络建模，开展了权重逻辑推理算法（WLI）的系统化研究。WLI引入绑定权重系数的逻辑事件推理机制，可确保变量状态概率的自动归一性和链式推理的自我依赖性，课题完成了WLI的规范化定义、推理算法补充、运算性质探析，并就理论相容性和自洽性开展了详细论证。WLI下不确定性因果关系的简洁、不完备表达及精确、高效的概率推理，可实现准确的故障定位与演化预测。在此基础上，为在故障工况下有效地介入实时故障传播控制和系统恢复策略，进一步结合诊断推理所得故障传播因果网络，开展了网络传播动态行为特征分析。通过评估网络结构负载分布和节点因果影响的动态性和非同质性，识别关键故障传播节点，从而形成适应时变性故障情境的最优化故障决策方法。.课题结合实际工业系统、核电站模拟机和仿真复杂网络等实验平台，对所建立的不确定性因果建模、诊断推理和动态适应性决策等方法进行了综合实证应用和测试分析。结果表明，所构建的理论方法在复杂系统故障诊断与决策方面表现出可解释性强、高效、准确等工程适用性特征，可在系统在线运行不中断的情况下，辅助系统操作人员快速定位故障根源、分析演化趋势、进行系统保护和修复，提高系统可靠性。