Input/output modeling for process control

Conference · Sat Dec 31 04:00:00 EST 1994

OSTI ID:103013

Holcomb, T R; Rhodes, C A; Morari, M ^[1]

---

1. California Institute of Technology, Pasadena, CA (United States)

Traditionally, for ease of operation and analysis, technological systems (manufacturing systems, communication systems, etc.) have been designed to behave in as linear a manner as possible. Increasingly tight specifications, environmental considerations and economic pressures are pushing the operational windows into regions where assumptions of linearity tend to break down. Despite a flurry of activity in the area of nonlinear dynamical systems (both analysis and control systems design) we see very few applications of these new results. One important reason is the lack of suitable nonlinear dynamic models. The objective of this project is the development of tools for building nonlinear models based on input/output data. The developed techniques build on results from nonlinear time series analysis, classical statistics and chemometrics. We show how to use the False Nearest Neighbor method to determine the order of a nonlinear ARMAX model, and explore the local linear structure of the input/output map via PLS (Partial Least Squares or Projection to Local Structures). By putting PLS in the general context of {open_quotes}significance regression{close_quotes} we are able to deal effectively with collinear data sets involving errors on the inputs ({open_quotes}Measurement Error Model{close_quotes}) and outliers ({open_quotes}Robust Regression{close_quotes}).

• Research Organization: Argonne National Lab., IL (United States)
• OSTI ID: 103013
• Report Number(s): CONF-9404137--; ON: DE94017694
• Country of Publication: United States
• Language: English

Related Subjects

99 GENERAL AND MISCELLANEOUS
CHEMICAL REACTORS
CONTROL SYSTEMS
FUNCTIONAL MODELS
NONLINEAR PROBLEMS
PROCESS CONTROL