基于实时学习的带确定扰动过程监测方法研究

发布时间：2018-07-06 19:44

本文选题：过程监测 + 实时学习　；参考：《哈尔滨工业大学》2017年硕士论文

【摘要】：过程监测在保证系统的安全性和可靠性方面起着至关重要的作用,对于现代工业过程的发展来说是一个关键的研究课题。随着社会的发展,现代工业过程变得日益智能化,系统内部结构错综复杂,很难通过机理分析或定性分析达到建立数学模型的目的,因而在过程监测方面,传统的基于机理模型的方法和基于知识的方法具有非常大的局限性,在现代工业系统中,难以被广泛的应用和推广。值得注意的是,现代工业过程在系统运行时往往会产生大量隐含过程特性信息的历史数据和实时数据,如何充分挖掘这些数据中的有效信息为过程监测提供新的思路成为国内外学者普遍关注的重点,这也是基于数据的过程监测方法的研究内容。目前,已有的过程监测算法大多是针对线性过程而提出的,但实际上,现代工业过程常常是复杂的非线性动态系统。针对此现实,本文将在线性静态系统过程监测思路的基础上,提出一种用于非线性动态系统的过程监测方法,为解决上述难题提供一种新的研究方法和思路。首先,本文采用一种实时学习方法用于解决复杂非线性动态系统难以精确建模的问题,结合多变量统计过程监测中最经典的主元统计法,实现对现代工业系统的过程监测,同时为后续算法的提出提供解决思路。然后,本文在基于实时学习的主元统计分析法实现思路上,针对现代工业过程常常受确定扰动的问题,提出一种适用于线性静态系统的过程监测算法,该算法通过残差评估进行故障决策,并将其与实时学习建模算法相结合,用于复杂非线性动态系统的过程监测。同时将提出的算法与其他非线性算法数值仿真对比,以说明提出的算法在故障误检率、漏检率以及鲁棒性等性能方面具有优势。最后,本文介绍污水处理过程采用基于数据方法的必要性,并将提出算法与其他非线性算法均应用到污水处理过程中,实验分析证明本文所提方法的先进性。
[Abstract]:Process monitoring plays an important role in ensuring the safety and reliability of the system and is a key research topic for the development of modern industrial processes. With the development of society, the modern industrial process becomes more and more intelligent, the internal structure of the system is complicated, it is difficult to establish the mathematical model through mechanism analysis or qualitative analysis, so in the aspect of process monitoring, The traditional methods based on mechanism model and knowledge have great limitations and are difficult to be widely applied and popularized in modern industrial systems. It is worth noting that modern industrial processes often produce a large number of historical and real-time data with implicit process characteristics information when the system is running. How to fully mine the effective information from these data to provide new ideas for process monitoring has become the focus of attention of scholars at home and abroad, which is also the research content of data-based process monitoring methods. At present, most of the existing process monitoring algorithms are proposed for linear processes, but in fact, modern industrial processes are often complex nonlinear dynamic systems. In view of this reality, based on the idea of process monitoring in linear static systems, a process monitoring method for nonlinear dynamic systems is proposed in this paper, which provides a new research method and train of thought for solving the above problems. First of all, a real-time learning method is used to solve the problem of complex nonlinear dynamic system which is difficult to model accurately, and the most classical principal component statistics method in multivariable statistical process monitoring is used to realize the process monitoring of modern industrial system. At the same time, it provides the solution for the following algorithm. Then, based on the realization of principle component statistical analysis method based on real-time learning, a process monitoring algorithm for linear static systems is proposed to solve the problem that modern industrial processes are often subject to deterministic disturbances. The algorithm is applied to process monitoring of complex nonlinear dynamic systems by residual evaluation, which is combined with real-time learning modeling algorithm. At the same time, the proposed algorithm is compared with other nonlinear algorithms to show that the proposed algorithm has advantages in fault error detection rate, miss detection rate and robustness. Finally, this paper introduces the necessity of adopting data-based method in the process of sewage treatment, and applies the proposed algorithm and other nonlinear algorithms to the process of sewage treatment. The experimental analysis proves the advanced nature of the proposed method.
【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】