项目名称: 基于动态数据融合的多模型软测量方法及其工业应用
项目编号: No.61203213
项目类型: 青年科学基金项目
立项/批准年度: 2013
项目学科: 自动化学科
项目作者: 刘瑞兰
作者单位: 南京邮电大学
项目金额: 26万元
中文摘要: 软测量技术能有效估计工业过程中难测的质量变量,针对复杂的工业过程,单一的模型往往不能满足精度要求,另外由于种种原因会发生过程数据失效的情况,本项目提出了基于动态数据融合的多模型软测量方法,开展过程变量数据失效的处理与恢复、子模型的建立和多个子模型的动态融合及误差分析工作。采用选择多个具有一定冗余的输入变量子集建立不同的软测量模型的方法解决数据失效问题,从时间冗余的角度使用状态估计方法恢复多个变量的少量失效数据问题。提出交互多模型算法的动态线性融合方法和基于粒子滤波的动态非线性融合方法。并从误差传递的角度分析恢复后的数据误差对子模型及综合模型输出的影响。以PX氧化过程中4-CBA浓度及晶体粒径的软测量为研究对象,通过实测数据来检验方法的有效性。本项目提出的失效数据的处理和交互多模型动态线性融合、粒子滤波动态非线性融合思路以及误差分析方法,在软测量领域具有较强的借鉴意义和较好的应用前景。
中文关键词: 软测量;动态数据融合;交互多模型;粒子群算法;深度学习
英文摘要: Soft sensor technology can be used to estimating quantities that are too difficult or expensive to be measured in process industry. Single soft sensor model can not meet the estimation precision in complex industry process, and missing data problems are usully happend under some circumstances such as malfunction of sensors, interruption of data transmission channels and so on. In this project, dynamic data fusion methods based on multiple soft sensors are proposed, which include missing or irregular data treatment and recovery, sub model building, dynamic data fusion and error analysis. Input sets with redundant variables are selected to different sub models to deal with missing data problem. State estimation methods are applied to recover the missing or irrrgular data in multiple varaibles. Interacting multiple models methods based dynamic data fusion are proposed for linear intergrating of sub models. Particle filtering methods are proposed for non-linear intergration of sub models. The precision of intergrating models with recovering data are defined by error transfer formula. The efficiency of the proposed methods will be demonstrated through 4-CBA content estimation and crystal particle size estimation in PX purification process. The proposed methods, including missing or irregular data recovery, dynamic li
英文关键词: soft sensor;dynamic data fusion method;interacting multiple model;particle swarm optimum algorithm;deep learning