基于偏最小二乘法的非线性工业过程监测方法研究
发布时间：2018-09-18 13:48
【摘要】：过程监测技术的出现是工业过程趋向自动化、智能化的标志,作为保障系统安全稳定运行的关键要素,其不可或缺性日益凸显。以往朴素的过程监测技术着眼于工业过程机理模型。然而,针对复杂程度日益提高的现代工业过程系统,即便辅以先进的模型辨识手段,基于物理化学先验知识的精确过程模型也越来越难以构建,这已成为工业控制学术界的共识。为此学术界将目光转向由传感器技术、网络通信技术、计算机数据处理技术的发展而带来的海量工业过程数据。显然,这些工业过程历史运行数据中蕴含着变量间的相关关系,充分挖掘数据内部的信息将极大地助力过程监测方法的研究。现有的基于数据的过程监测机制研究大多面向线性的、静态的工业过程,虽然这些研究取得了一定的成果,但仍然难以满足实际工业系统中非线性的过程监测需求。针对这样的情况,本文以偏最小二乘法为理论根基,试图建立一套完整的工业过程监测体系,使其能够适应于线性、非线性以及动态过程的监测。本文首先介绍标准偏最小二乘算法原理,针对偏最小二乘算法的缺点,介绍一种以完全分解数据空间为核心思想的改进型偏最小二乘算法,并讨论该方法在故障检测中的应用,为后续算法的提出奠定理论基础。接下来本文探讨核偏最小二乘算法在非线性过程监测中的应用问题。在此基础上,提出一种基于核偏最小二乘算法的在线非线性过程故障检测方法。同时引入小波变换对数据情况复杂的非线性工业过程进行监测。以数值算例和污水处理系统为应用背景,验证所述方法的可行性,为监测非线性、数据情况复杂的工业过程提供一种行之有效的解决方案。为了寻求动态过程监测问题的解决方法,本文基于解构过程数据以及过程模型的思想,将多子阶段模型方法与核偏最小二乘法相结合,其核心思想是在对动态过程进行监测之前,先对考察的过程进行精确建模。这一部分内容给出了非线性动态工业过程监测方法的新实践。
[Abstract]:The appearance of process monitoring technology is the symbol of the trend of industrial process towards automation and intelligence. As a key element to ensure the safe and stable operation of the system, its indispensable is becoming more and more prominent. Previous simple process monitoring technology focused on the industrial process mechanism model. However, for modern industrial process systems with increasing complexity, even with advanced model identification methods, precise process models based on prior knowledge of physical chemistry are becoming more and more difficult to construct. This has become the consensus of industry control academia. For this reason, the academic circles turn their attention to the massive industrial process data brought by the development of sensor technology, network communication technology and computer data processing technology. Obviously, these industrial process historical running data contain the correlation relation between variables, fully mining the information inside the data will greatly help the research of the method of process monitoring. Most of the existing data-based process monitoring mechanisms are oriented to linear and static industrial processes. Although some achievements have been made in these studies, it is still difficult to meet the needs of nonlinear process monitoring in practical industrial systems. In this paper, based on the partial least square method, we try to establish a complete industrial process monitoring system, which can adapt to linear, nonlinear and dynamic process monitoring. In this paper, the principle of standard partial least squares algorithm is introduced, and an improved partial least squares algorithm based on the idea of completely decomposing data space is introduced, and its application in fault detection is discussed. It lays a theoretical foundation for the following algorithm. Then this paper discusses the application of kernel partial least squares algorithm in nonlinear process monitoring. On this basis, an on-line nonlinear process fault detection method based on kernel partial least squares algorithm is proposed. At the same time, wavelet transform is introduced to monitor nonlinear industrial processes with complex data. Taking numerical examples and sewage treatment system as the application background, the feasibility of the method is verified and an effective solution is provided for the monitoring of nonlinear and complex industrial processes. In order to find a solution to the problem of dynamic process monitoring, based on the idea of deconstruction process data and process model, this paper combines the multi-sub-stage model method with the kernel partial least square method. Its core idea is to model the process accurately before monitoring the dynamic process. In this part, the new practice of nonlinear dynamic industrial process monitoring method is given.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP274
本文编号：2248112
[Abstract]:The appearance of process monitoring technology is the symbol of the trend of industrial process towards automation and intelligence. As a key element to ensure the safe and stable operation of the system, its indispensable is becoming more and more prominent. Previous simple process monitoring technology focused on the industrial process mechanism model. However, for modern industrial process systems with increasing complexity, even with advanced model identification methods, precise process models based on prior knowledge of physical chemistry are becoming more and more difficult to construct. This has become the consensus of industry control academia. For this reason, the academic circles turn their attention to the massive industrial process data brought by the development of sensor technology, network communication technology and computer data processing technology. Obviously, these industrial process historical running data contain the correlation relation between variables, fully mining the information inside the data will greatly help the research of the method of process monitoring. Most of the existing data-based process monitoring mechanisms are oriented to linear and static industrial processes. Although some achievements have been made in these studies, it is still difficult to meet the needs of nonlinear process monitoring in practical industrial systems. In this paper, based on the partial least square method, we try to establish a complete industrial process monitoring system, which can adapt to linear, nonlinear and dynamic process monitoring. In this paper, the principle of standard partial least squares algorithm is introduced, and an improved partial least squares algorithm based on the idea of completely decomposing data space is introduced, and its application in fault detection is discussed. It lays a theoretical foundation for the following algorithm. Then this paper discusses the application of kernel partial least squares algorithm in nonlinear process monitoring. On this basis, an on-line nonlinear process fault detection method based on kernel partial least squares algorithm is proposed. At the same time, wavelet transform is introduced to monitor nonlinear industrial processes with complex data. Taking numerical examples and sewage treatment system as the application background, the feasibility of the method is verified and an effective solution is provided for the monitoring of nonlinear and complex industrial processes. In order to find a solution to the problem of dynamic process monitoring, based on the idea of deconstruction process data and process model, this paper combines the multi-sub-stage model method with the kernel partial least square method. Its core idea is to model the process accurately before monitoring the dynamic process. In this part, the new practice of nonlinear dynamic industrial process monitoring method is given.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP274
【参考文献】
相关期刊论文 前1条
1 柴天佑;;生产制造全流程优化控制对控制与优化理论方法的挑战[J];自动化学报;2009年06期
相关博士学位论文 前2条
1 叶洪涛;基于免疫算法的污水处理系统预测及优化控制研究[D];华南理工大学;2011年
2 葛志强;复杂工况过程统计监测方法研究[D];浙江大学;2009年
相关硕士学位论文 前3条
1 张景欣;基于复杂可测变量的非线性过程智能监控方法研究[D];哈尔滨工业大学;2016年
2 谢晓晨;大数据意义下的非线性工业过程预测问题研究[D];哈尔滨工业大学;2014年
3 杨婷;污水处理过程的模糊建模与控制[D];哈尔滨工业大学;2011年
，本文编号：2248112
本文链接：https://www.wllwen.com/kejilunwen/zidonghuakongzhilunwen/2248112.html
