具有多操作特性的间歇工业过程监测技术研究

项目名称： 具有多操作特性的间歇工业过程监测技术研究

项目编号： No.61503169

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： 王亚君

作者单位： 辽宁工业大学

项目金额： 22万元

中文摘要： 该项目针对具有多操作特性的间歇工业过程监测的关键技术难题展开研究。由于操作频繁和随机等特点，导致多操作间歇工业过程具有：批次数据严重不等长、数据呈严重非高斯分布及建模数据不全等问题，这使得传统的基于多元统计监测方法无法适用。本项目针对具有多操作特性的间歇工业过程，提出以操作粗划分和细化分为基础的一系列建模与监测方法。主要研究内容包括：（1）针对批次数据不等长的过程，采用基于重要点（IP）的建模方法进行监测，避免轨迹同步问题同时可减少核主成分分析（KPCA）算法的运算复杂度；（2）针对非高斯分布的数据，研究用近邻子集标准化方法完成数据的高斯转化；（3）针对建模数据不全问题，研究邻近模型（前、后模型）同时监测的方法实现对新操作数据的在线监测以及模型在线更新。本项目为监测具有多操作特性的间歇工业过程提供更有效的解决方法与实现技术。

中文关键词： 间歇式建模；数据驱动建模；离线建模

英文摘要： The purpose of this project is to address the key technical problems in batch industrial processes with multi-operation features. Since the operations has the characteristics of frequence and randomness, multi-operation batch industrial process features seriously unequal batch data and non-Gaussian distribution of data as well as imcomplete modeling data. As a result, it is not suitable for current multivariate statistical monitoring method. This project focuses on batch industrial process with multi-operation features and a series of modeling and monitoring methods based on both coarse partition and refinement partition of operations are thus proposed. Main contents include: (1) For the process with unequal batch data, important point (IP) based modeling method is adopted so as to avoid trajectory synchronization and meanwhile to reduce the computation complexity of kernel principal component analysis (KPCA) method; (2) For data with non-Gaussian distribution feature, neighbor subset normalization method is explored to implement Gaussian distribution transformation; (3) For the process with incomplete modeling data type, the adjacent models (the front and back models) monitoring method is further studied to realize online monitoring of new coming data and online update of model. This research is to develop more efficient methods or technologies for monitoring complex batch industrial process with multi-operation features.

英文关键词： Batch Modeling;Data-driven Modeling;Offline Modeling