摘要
数据驱动方法是一种黑箱模型,具有自主挖掘和构建数据内在关系的优点。随着感知设备的发展和计算能力的提升,数据驱动方法在化工过程故障诊断的研究领域体现出了更大的优势。本文介绍了各类数据驱动方法的原理和作用,并分析了其各自的优缺点与实际应用方向,总结得出深度学习和集成学习是数据驱动方法未来研究重点。同时,本文回顾了近五年来国内外数据驱动方法在化工过程故障诊断中的研究与应用,综合分析了现阶段该领域的研究情况,表明将多种数据驱动方法进行组合来解决化工过程问题的思路具有一定的有效性。并进一步给出了关于数据异常、时间滞后等问题的研究方向。最后,本文建议更多地从方法的机理出发对方法进行研究和优化,在未来的研究思考中应更着重于实用性和时效性。
The data-driven method is a black-box model,which has the advantage of autonomously mining and constructing the internal relationship of data.The development of perception equipment and the improvement of computing power highlight the advantages of the data-driven method,which can better independently mine and build the internal relationship of data.This paper introduces the principles and functions of various data-driven methods,analyzes the advantages and disadvantages of the methods and the practical application direction,and draws the conclusion that deep learning and integrated learning are the key research points of data-driven methods in the future.This paper reviews the research and application of data-driven methods for fault diagnosis in chemical processes in the recent five years,and finally comprehensively analyzes the current research situation in this field.The analysis shows that it is effective to solve problems in chemical processes by combining multiple data-driven methods.Furthermore,the research direction of data anomaly and time lag is provided.In this paper,it suggested that the mechanism of the method should be studied and optimized.In the future,the development and research of data-driven methods for fault diagnosis in chemical processes should focus on two points,namely“practicality”and“timeliness”.
作者
姚羽曼
罗文嘉
戴一阳
YAO Yuman;LUO Wenjia;DAI Yiyang(College of Chemistry and Chemical Engineering,Southwest Petroleum University,Chengdu 610500,Sichuan,China;School of Chemical Engineering,Sichuan University,Chengdu 610065,Sichuan,China)
出处
《化工进展》
EI
CAS
CSCD
北大核心
2021年第4期1755-1764,共10页
Chemical Industry and Engineering Progress
关键词
故障诊断
数据驱动方法
化工过程
特征提取
时间序列
fault detection and diagnosis
data-driven method
chemical process
feature extraction
time series
