基于数据特征选择与匹配的工业过程监测方法研究

项目名称： 基于数据特征选择与匹配的工业过程监测方法研究

项目编号： No.61503204

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 自动化技术、计算机技术

项目作者： 童楚东

作者单位： 宁波大学

项目金额： 20万元

中文摘要： 以故障检测与诊断为核心的过程监测是保证生产安全和提高产品质量的必要手段。由于现代复杂工业系统的大量涌现，不同任务需求和应用环境不断地给数据驱动的过程监测研究提出新的考验。已有的数据驱动方法在建立动态、非线性和其它复杂过程的监测模型时，欠缺对数据复杂性与方法可用性的考虑，衍生出很多问题亟待解决。为此，本项目拟以数据特征选择与匹配为主线，研究如何根据不同过程分析目的来设计相应的数据相关性、样本及变量的选择方法，并辅以相关的数据特征匹配策略，以简易化复杂大型过程数据分析与建模过程。另外，针对可参考故障样本不充分的问题，拟在故障特征选择的基础上，进一步开展如何利用数据特征匹配方法进行故障分类诊断的研究。项目将阐明特征选择与匹配在改善过程监测性能上的重要作用，旨在建立一套可靠又实用的故障检测与诊断方法，其成果将为丰富和完善数据驱动的过程监测方法体系提供理论与实验依据，具有重要的理论意义与应用价值。

中文关键词： 数据驱动；统计过程监测；特征选择与匹配；故障检测；故障诊断

英文摘要： Process monitoring with the aim of fault detection and diagnosis is the essential tool to ensure plant safety and improve production quality. With the emergence of diverse complex large-scale industrial systems, different mission requirements and also practical application conditions keep bringing in new challenges to process monitoring. Without enough respect to the complexity of process data and the feasibility of the monitoring methods, the existing data-driven techniques thus encounter many issues that needed to be addressed urgently in developing monitoring models for dynamic, nonlinear and other complex processes. To this end, the formulated project follows the feature selection and matching of process data, focuses on designing corresponding selection approaches of data correlation, samples and variables according to the different purposes of process analysis, so as to simplify the process of data analysis and modeling. Additionally, the project intends to attempt the problem caused by insufficient reference fault datasets by conducting research on classification-based fault diagnosis, which is implemented on the basis of faulty data feature selection. This project would demonstrate the importance of feature selection and matching in improving the monitoring performance, and a set of reliable and practical fault detection and diagnosis methods is then expected to be constructed finally. The results of this project would provide theoretical and experimental basis for enriching and perfecting data-driven process monitoring methodology, it is of great importance in both theory and application.

英文关键词： Data Driven;Statistical Process Monitoring;Feature Selection and Matching;Fault Detection;Fault Diagnosis