Robust Process Identification with Dynamic Feature Analysis

过程工厂中的每个人，从工厂经理到工程师到技术人员，都依赖于大量的数据，这些数据在日常分析和决策中起着重要作用。常见的过程控制实践是借助过程知识开发基于数据的模型。但是，随着现代过程数据在维度、多样性和复杂性方面的增加，传统的分析工具已经无法跟上这种复杂数据的冲击。数据的高维性和数据采集过程中的不规则性给基于数据的建模带来了许多挑战，从而使传统建模技术的有效性受到严重质疑。因此，过程控制研究团体在不断增加的压力下交付分析工具，以应对过程工业现代实践的挑战。******为了应对这种压力，并受到过程工业面临的现实挑战的激励，本提案的短期目标是为在现代数据集中存在高维和不规则性的过程识别中遇到的基本问题提供解决方案。该研究项目将开发新的过程建模技术，通过这些技术，可以有效地利用大量数据，实现安全和智能的过程操作。从长远来看，目标是通过使用复杂的过程数据，为过程系统的识别和控制开发一个综合框架。建模和控制器设计是密不可分的。在复杂数据存在的情况下，建模和控制设计问题的纠缠是一个重大的挑战，也是一个相对未触及的领域。本研究计划将有助于建立一种新的基于数据的控制设计理论和方法。******我们的方法同时处理两个关键问题：数据维度和数据不规则性。首先，我们建立了一种新的动态特征分析方法，然后使该方法在存在数据不规则性的情况下具有鲁棒性。我们的解决方案将适用于广泛的行业采用或将采用自动化系统。我们的研究项目将培养在数据分析和基于数据的建模方面高素质的年轻人。他们将成为下一代技术领导者，将这些技术整合到加工工厂中，以提高加拿大工业的竞争力，并带头推动在全球范围内销售解决方案。

Everyone in a process plant, from plant managers to engineers to technicians, relies on a massive amount of data, which plays a significant role in daily analysis and decision making. Common process control practice is to develop models based on data with the aid of process knowledge. But as modern process data has increased in dimensionality, diversity and complexity, traditional analytical tools have been unable to keep up with this onslaught of complex data. High dimensionality of data and irregularities during data collection pose many challenges in data-based modeling, thereby casting serious doubt on the validity of traditional modeling techniques. As a result, the process control research community is under ever increasing pressure to deliver analytic tools to cope with the challenges of the modern day practices of the process industries.******Responding to this pressure and motivated by the real-life challenges faced by process industries, the shorter term objective of this proposal is to provide a solution to fundamental problems encountered in process identification in the presence of high dimensionality and irregularities in modern datasets. This research program will develop new process modeling techniques by which this enormous amount of data can be fruitfully utilized, to achieve safe and intelligent process operations. In the long term, the objective is to develop an integrated framework for identification and control of process systems by employing complex process data. Modeling and controller design are inseparable. The entangling of modeling and control design problems in the presence of complex data poses a significant challenge and is a relatively untouched field. This research program will contribute to the establishment of a new data-based control design theory and methodology. ******Our methodology deals with two critical problems simultaneously: data dimensionality and data irregularities. First, we establish a new dynamic feature analysis methodology, and then we make the methodology robust in the presence of data irregularities. Our solutions will be applicable to a wide range of industries that employ or will employ automation systems. Our research program will train young people who are highly qualified in data analytics and data-based modeling. They will be the next generation of technical leaders who will integrate these technologies into process plants to boost the competitiveness of Canadian industry and spearhead the drive to sell solutions worldwide.