基于多元时间序列分析的复杂系统复合故障诊断研究

复杂系统工艺连续、系统结构关系庞杂，存在极强的时态和关联复杂性，其运行过程中影响可靠性的因素不易检测鉴别和量化，使得系统中存在复合故障成为常态。作为故障诊断和容错控制技术中的重难点问题，本申请关注数据驱动的复合故障诊断方法。针对复杂系统运行状态的安全控制和可靠性增长需求，拟对复杂系统运行参数在时域上的关联关系以及在复合故障诊断中的应用展开研究。本申请拟使用系统过程运行参数形成的多元时间序列建立复杂系统的运行工况模型；根据先验知识对独立故障在时间序列上的特征关键点进行筛选，并形式化表示、实现高维工况时序数据在低维特征集上的映射；通过多元时间序列分析刻画故障行为传播轨迹，建立针对故障行为的关键特征传播关联网络；基于上述模型研究故障辨识分离策略，以发现复杂系统中可能存在的多个故障模式，实现复合故障诊断。研究希望从根本上脱离对物理模型的依靠约束，是对数据驱动的系统安全控制理论方法发展和更新。

Complex systems, with a continuous process and complex structural relationships profile, have time and associated characteristics. And factors that affect reliability of running process are hard to be detected and quantified, making their precise mathematical model are difficult to establish. Multiple fault diagnosis Is a key issue in both fault diagnosis and fault-tolerance control technology fields. Focusing on the growth demand in safety control and reliability of running state of complex systems, this project carry out data driven based fault diagnosis, aiming at researching time domain based correlation analysis of operating parameters of complex systems and applied multiple fault diagnosis. This project build operational states models based on multivariate time series formed by operating parameters. Then we filtrate by using the relationship between priori knowledge and key Feature of parameters with time series, Building the mapping visually from low-dimensional space with data sets to high-dimensional working space. Finally, key feature spread based relevance network are established by multivariate time series correlation analysis and fault propagation path. The key scientific problems of multiple fault diagnosis are studied by finding potential multiple failure modes in complex systems with established model based fault recognition and failure prediction strategy. The goal is to shake off the yoke of Physical Model, develop Methodology of data driven based system safety control theory.