基于双层潜隐结构的工业过渡过程监测方法研究

The smooth running of transient process plays a critical role in the quality reliability and safety of industrial systems. This project focuses on the complex characteristics of transient process, such as nonlinearity, non-stationary and time varying, etc. Existing transient process monitoring methods suffer from problems like insufficient consideration of quality-related information and incomplete extraction of higher order information. In order to alleviate these problems, this project divides the monitoring problem into three stages, i.e., process sub-mode division, modeling and monitoring, and model updating. More specifically, the research can be summarized as: 1) Considering the significant differences between data in different transient periods, a distributed random projection method will be proposed to divide the transient process into several sub-modes by iteratively updating the optimum match between process variables and quality variables; 2) A double-layer hidden latent structure will be developed based on the maximum higher order information expansion method to deal with both the low order and high order information of process data, based on which a modeling and monitoring approach will be developed for data from different sub-modes; 3) A new locally-weighting scheme will be proposed to update the process model by considering both the moving window confidence and quality-related local weights of each data sample. The research of this project will further enhance the process monitoring of complex industrial system; and is of significant theoretical and practical importance to the efficient and smooth operation of industrial processes.

工业过渡过程的正常运行与否直接影响产品质量的可靠性和系统运行安全。本项目重点围绕过渡过程中的非线性、非平稳与时变等复杂特性，针对过渡过程监测中存在的质量相关信息考虑不充分、高阶信息提取不完全等问题，将过渡过程监测研究分解为过程阶段划分、过程建模与故障检测以及模型自适应更新三部分，并分别开展以下内容研究：1)根据不同阶段内数据特性相异的特点，提出一种分布式随机投影方法，迭代求取各阶段内数据与相应质量相关信息之间的最优匹配，实现过渡过程阶段划分；2)设计最大高阶信息展开方法，并结合低阶信息提取，构建双层潜隐结构分别对阶段内不同阶次质量相关信息开展建模与故障检测方法研究；3)建立滑动窗内质量相关局部权重，并结合滑动窗置信度，创建新型局部加权方法，获取质量相关的样本全局权重，实现模型自适应更新。本项目旨在进一步丰富复杂工业过程监测的研究，对工业生产过程高效稳定运行有着重要的理论意义与应用价值。

本项目计划以工业过渡过程监测为目标，针对过渡过程监测中存在的质量相关信息考虑不充分、高阶信息提取不完全等问题，将工业过渡过程监测研究分解为过程阶段划分、过程建模与故障检测以及模型自适应更新三部分。研究面向数据的自相关性、非平稳等复杂特性的工业过渡过程阶段建模与故障检测方法。在过渡过程模型自适应更新方法研究中，针对动态过程中的动态关系以及变量间相互关系和非平稳状态的问题开展深入研究。针对过渡过程的时变特性，充分挖掘过程中全局以及局部特征，在捕捉全局信息的同时获取更多的局部信息，用于过程监测。针对传统故障分类器权重设置不准确的问题，研究权重细化工业过程故障分类决策融合系统，通过少量且高效的指标反映分类结果，提高决策系统的科学性和合理性。课题团队根据项目申请书中的技术路线顺利推进本项目研究内容的开展，已顺利完成本项目既定研究目标。