本项目针对大规模复杂动态系统的特点，利用多元时间序列分析方法进行因果关系建模，构建拓扑搜索和因果推断相结合的溯源诊断双层策略，在下层通过多元时间序列分析的预测提升类分析方法实现因果拓扑获取，在上层结合事件和报警信息以及实验设计和知识引入完成因果推断。预期可形成融合时间序列分析和过程连接性知识的变量间因果拓扑获取的系统化方法体系；提出适用于动态系统分析的因果推断方法，特别是异常数据小样本或无样本条件下的因果推断方法；建立拓扑搜索和因果推断相结合的溯源诊断双层策略，用于大规模复杂动态系统的故障定位。将所研究的方法在核电厂、炼油厂和网络服务的信息系统上进行应用验证和案例分析。所建立的模型和推理框架可在异常状况发生时实现自动的溯源推理，锁定故障源，以提高系统实时监控的智能水平，保证系统的安全稳定运行。

Aiming at the characteristics of large-scale complex and dynamic systems, we build causal models based on multivariate time series analysis, to construct a two-level strategy for root cause analysis by combining topology search and causal inference: in the lower level, causal topology is captured by a class of prediction improvement analysis methods based on multivariate time series analysis, and in the upper level, event and alarm information as well as experiment design and knowledge is introduced for causal inference. We expect to establish a systematic methodology framework for capturing causal topology between variables by combining time series analysis and process connectivity knowledge, propose causal inference methods suitable for dynamic system analysis, in particular causal inference methods under the circumstance of small sample or no sample of abnormal data, and construct a two-level strategy for root cause analysis by combining topology search and causal inference, which can be used for fault localization of large-scale complex and dynamic systems. The proposed methods will be tested with case studies in nulear power plants, refineries and information systems for network services. The constructed framework of models and inference can realize automatic root cause inference and root cause localization under abnormal situations, to improve the intelligence level of real-time monitoring of the sytem to assure its safe and steady operation.