Data Based Construction of A Multi-Step Predictor for Multivariable Nonlinear Stochastic Systems - With Applications to Soft-Sensing, Process Monitoring and Predictive Control

基于数据的多变量非线性随机系统多步预测器构建 - 应用于软测量、过程监控和预测控制

• 批准号: 9979873
• 负责人: Jay Lee
• 金额: $ 25.73万
• 依托单位: Purdue Research Foundation
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 1999
• 资助国家: 美国
• 起止时间: 1999-09-15 至 2000-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9979873&HistoricalAwards=false
• 关键词: Data; Based; Construction; Multi; Step

AbstractProposal Title: Data Based Construction of a Multi-Step Predictor for Multivariable Nonlinear Stochastic Systems-with Applications to Soft-Sensing, Process Monitoring and Predictive ControlProposal Number: CTS-9979873Principal Investigator: Jay H. LeeInstitution: Purdue UniversityThe main objective of this research is to develop a framework for using plant data for the development of a reduced-order nonlinear multistep prediction model, capable of predicting a long-term behavior of relevant variables on the basis of incoming measurements, in the presence of both deterministic and stochastic input changes. Such a model is an essential component of the envisioned computer-based system that performs the integrated function of process monitoring, soft sensing, and predictive control. The PI will draw upon the subspace identification paradigm that enables the engineer to build a state-space prediction model directly from input-output data. The proposed research will address the following topics: (1) extension of linear subspace identification to nonlinear stochastic systems, involving the development of a nonlinear projection method for the construction of state-space models, (2) a comparative evaluation of the nonlinear projection-based method against other competing options including the direct time-series based design and the internally recurrent neural network based design, (3) development of a robust model predictive control method for piecewise-linear state-space models, (5) development of a software package for the integrated control scheme, and (6) testing of the developed methods on distillation columns and reactors. The development of a model-based estimator and controller is of great industrial importance, particularly for a process in which control variables are complex functions of process variables such as temperature and composition.

AbstractProposal Title: Data Based Construction of a Multi-Step Predictor for Multivariable Nonlinear Stochastic Systems-with Applications to Soft-Sensing, Process Monitoring and Predictive ControlProposal Number: CTS-9979873Principal Investigator: Jay H. LeeInstitution: Purdue UniversityThe main objective of this research is to develop a framework for using plant data for the development of a reduced-order nonlinear multistep prediction model, capable在存在确定性和随机输入变化的情况下，根据传入的测量来预测相关变量的长期行为。 这样的模型是设想的基于计算机的系统的重要组成部分，该系统执行过程监视，软传感和预测控制的集成功能。 PI将利用子空间标识范例，使工程师可以直接从输入输出数据直接构建状态空间预测模型。 The proposed research will address the following topics: (1) extension of linear subspace identification to nonlinear stochastic systems, involving the development of a nonlinear projection method for the construction of state-space models, (2) a comparative evaluation of the nonlinear projection-based method against other competing options including the direct time-series based design and the internally recurrent neural network based design, (3) development of a robust model predictive control method for piecewise-linear州空间模型，（5）为集成控制方案开发软件包，以及（6）在蒸馏柱和反应堆上测试开发的方法。 基于模型的估计器和控制器的开发具有极大的工业重要性，特别是对于控制变量是过程变量（例如温度和组成）复杂函数的过程。