基于WD-MNPE-CVA的带钢热轧全流程动态过程监测方法
Dynamic process monitoring based on WD-MNPE-CVA for hot strip mill plant-wide process
董洁 1陈柔汝 1彭开香1
作者信息
- 1. 北京科技大学自动化学院,北京 100083;北京科技大学工业过程知识自动化教育部重点实验室,北京 100083
- 折叠
摘要
过程监测技术是保障复杂工业全流程安全、高质、高效运行的有效手段.考虑带钢热轧过程的"非线性、多模态、动态性"等特征,本文提出一种基于改进的加权差分邻域保持嵌入-规范变量分析(WD-MNPE-CVA)的带钢热轧全流程动态过程监测方法.首先,针对过程数据的非线性、多模态特性,采用加权差分方法进行数据预处理;其次,基于过程的机理知识进行流程划分,并开发了一种改进的邻域保持嵌入算法,基于样本点之间的欧氏距离和余弦距离,获得每个样本点更准确的邻域关系,进而基于规范变量分析建立每个子流程的局部动态监测模型;最后,采用贝叶斯推理建立全局的动态过程监测模型,通过带钢热轧实际过程故障数据验证了该方法的有效性.
Abstract
Process monitoring technology is an effective measure to ensure the safety and efficiency of the hot strip mill plant-wide process.Considering the"nonlinear,multimode,dynamic"characteristics of the hot strip mill process,a weighted difference-modified neighborhood preserving embedding-canonical variable analysis(WD-MNPE-CVA)method for dynamic process monitoring of plant-wide process is proposed in this paper.Firstly,in view of the nonlinear and mul-timode characteristics existing in the process data,the weighted difference method is used to preprocess the process data.Then the plant-wide process is divided based on the mechanism knowledge.An improved neighborhood preserving embed-ding algorithm is developed to obtain more accurate neighborhood relationship of each sample point based on Euclidean distance and cosine distance between sample points,and then establish local dynamic monitoring model of each subpro-cess based on canonical variable analysis.Finally,a global dynamic process monitoring model is established by Bayesian inference,and the effectiveness of the proposed method is verified by the actual fault data of hot strip mill process.
关键词
过程监测/邻域保持嵌入/规范变量分析/贝叶斯融合/带钢热轧全流程Key words
process monitoring/neighborhood preserving embedding/canonical variable analysis/Bayesian fusion/hot strip mill plant-wide process引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据