报告主题：Dynamic Latent Variable Regression for Inferential Sensor Modeling and Supervised Monitoring
This seminar will be on system modeling and process monitoring with multivariate statistical methods. Specifically, the presentation will (1) review some existing multivariate statistical approaches for process modeling and monitoring; and (2) introduce a novel latent variable regression algorithm and its dynamic counterpart.
With the advent of Industry 4.0, which contributes to fostering a "smart factory" nowadays, the volume of collected data has increased explosively with the aid of advanced data sensory technologies. Mining for meaningful information from voluminous data and transforming it to valuable knowledge for smart decision making have become an emerging field. Multivariate statistical analysis has demonstrated its capability and superiority in extracting and interpreting important patterns from process and quality data in many areas, including computer science, biomedical engineering, materials science, chemical engineering and management science. In this presentation, some popular multivariate statistical approaches will be reviewed, such as projection to latent structures (PLS) and canonical correlation analysis (CCA), and their pros and cons will be analyzed as well. Afterwards, the new regularized latent variable regression (rLVR) algorithm will be covered, which employs consistent inner and outer modeling objectives, and it has shown to retain better predition performance over PLS and CCA. The dynamic version, dynamic rLVR (DrLVR), and its monitoring framework, will also be designed for dynamic data modeling and monitoring.
Qinqin Zhu博士是滑铁卢大学化学工程系的助理教授，滑铁卢人工智能研究所（Waterloo.AI）和滑铁卢可持续能源研究所（WISE）教师。在南加州大学计算机科学系和化学工程系分别获得硕士学位和博士学位。在滑铁卢大学任教之前，她曾在美国Facebook Inc.担任高级研究科学家。她的研究主要致力于在大数据时代开发先进的统计机器学习方法，过程数据分析技术和优化算法，并将其应用于统计过程监控和故障诊断，解决了过程系统工程领域中的理论挑战和具有实际重要性的问题。她的团队致力于通过利用数学建模和优化的力量，开发高级的多元统计分析算法，以增强复杂工程系统中的决策能力。