文献类型：期刊文献

中文题名：化工过程动态软仪表技术发展现状及展望

英文题名：Development Status and Prospect of Dynamic Soft Instrument Technology in Chemical Process

作者：田洲[1];杨逸俊[1];王振雷[1]

机构：[1]华东理工大学化工过程先进控制和优化技术教育部重点实验室,上海200237

年份：2020

卷号：41

期号：8

起止页码：1

中文期刊名：自动化仪表

外文期刊名：Process Automation Instrumentation

收录：CSTPCD

基金：国家重点研发计划基金资助项目(2018YFB1701103)。

语种：中文

中文关键词：动态软仪表;长短时记忆;循环网络;动态建模;在线校正

外文关键词：Dynamic soft sensor;Long short-term memory(LSTM);Recurrent neural network(RNN);Dynamic modeling;On-line correction

摘要：软仪表是解决化工过程中质量变量难以实时测量的有效手段。化工生产是动态生产过程,即产品指标不仅受到操作变量当前值的影响,而且受到变量历史信息的影响。辅助变量时序长度的选择通常依赖工程经验。近年来,以动态神经网络为载体的动态软仪表技术凭借其可以反映过程变量增量间存在的动态关系的能力,成为软仪表建模的重要研究方向。详细介绍了动态软仪表的主要结构与工作原理,重点针对数据高维度、非线性和动态性等复杂过程的动态软仪表发展现状和建模方法进行了综述与分析,并讨论了动态软仪表的在线实现方法,最后对动态软仪表技术发展进行了展望。
Soft sensor is an effective method to solve the difficulty of real-time measurement of quality variables in the process.Chemical production is a dynamic production process;that is,the product specification is not only affected by the current value of the operating variable,but also by the historical information of the variable.The selection of the lag time of the auxiliary variable usually depends on the engineering experience.In recent years,dynamic soft sensor technology based on dynamic neural networks has become an important research area for soft sensor modeling by its ability to reflect the dynamic relationship between process variable increments.The main structure and working principle of a dynamic soft sensor are introduced in detail,and the development process and modeling methods of a dynamic soft sensor with high-dimensional,non-linear and dynamic data and other sophisticated features are selective analysisummarized and analyzed.The on-line implementation of dynamic soft meter is discussed,and finally looks forward to the development of dynamic soft sensor technology.