skip to main content
OSTI.GOV title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: Adaptive model predictive process control using neural networks

Abstract

A control system for controlling the output of at least one plant process output parameter is implemented by adaptive model predictive control using a neural network. An improved method and apparatus provides for sampling plant output and control input at a first sampling rate to provide control inputs at the fast rate. The MPC system is, however, provided with a network state vector that is constructed at a second, slower rate so that the input control values used by the MPC system are averaged over a gapped time period. Another improvement is a provision for on-line training that may include difference training, curvature training, and basis center adjustment to maintain the weights and basis centers of the neural in an updated state that can follow changes in the plant operation apart from initial off-line training data. 46 figs.

Inventors:
; ;
Publication Date:
Research Org.:
University of California
Sponsoring Org.:
USDOE, Washington, DC (United States)
OSTI Identifier:
527745
Patent Number(s):
US 5,659,667/A/
Application Number:
PAN: 8-373,736
Assignee:
Univ. of California Office of Technology Transfer, Alemeda, CA (United States) PTO; SCA: 320303; PA: EDB-97:125344; SN: 97001843784
DOE Contract Number:
W-7405-ENG-36
Resource Type:
Patent
Resource Relation:
Other Information: PBD: 19 Aug 1997
Country of Publication:
United States
Language:
English
Subject:
32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATION; PROCESS CONTROL; NEURAL NETWORKS; CONTROL SYSTEMS; INDUSTRIAL PLANTS; TRAINING; ON-LINE SYSTEMS

Citation Formats

Buescher, K.L., Baum, C.C., and Jones, R.D. Adaptive model predictive process control using neural networks. United States: N. p., 1997. Web.
Buescher, K.L., Baum, C.C., & Jones, R.D. Adaptive model predictive process control using neural networks. United States.
Buescher, K.L., Baum, C.C., and Jones, R.D. Tue . "Adaptive model predictive process control using neural networks". United States. doi:.
@article{osti_527745,
title = {Adaptive model predictive process control using neural networks},
author = {Buescher, K.L. and Baum, C.C. and Jones, R.D.},
abstractNote = {A control system for controlling the output of at least one plant process output parameter is implemented by adaptive model predictive control using a neural network. An improved method and apparatus provides for sampling plant output and control input at a first sampling rate to provide control inputs at the fast rate. The MPC system is, however, provided with a network state vector that is constructed at a second, slower rate so that the input control values used by the MPC system are averaged over a gapped time period. Another improvement is a provision for on-line training that may include difference training, curvature training, and basis center adjustment to maintain the weights and basis centers of the neural in an updated state that can follow changes in the plant operation apart from initial off-line training data. 46 figs.},
doi = {},
journal = {},
number = ,
volume = ,
place = {United States},
year = {Tue Aug 19 00:00:00 EDT 1997},
month = {Tue Aug 19 00:00:00 EDT 1997}
}
  • A control system for controlling the output of at least one plant process output parameter is implemented by adaptive model predictive control using a neural network. An improved method and apparatus provides for sampling plant output and control input at a first sampling rate to provide control inputs at the fast rate. The MPC system is, however, provided with a network state vector that is constructed at a second, slower rate so that the input control values used by the MPC system are averaged over a gapped time period. Another improvement is a provision for on-line training that may includemore » difference training, curvature training, and basis center adjustment to maintain the weights and basis centers of the neural in an updated state that can follow changes in the plant operation apart from initial off-line training data.« less
  • The authors work in controlling chemical processes included the following: (1) to develop neural networks and training procedures that are well suited to: small amounts of off-line training data for on-line control of systems with substantial time lags; (2) to develop generic Model Predictive Control (MPC) software; and (3) to control the following simulated systems using MPC: continuously stirred tank reactor with jacket dynamics; plasma etching model for semiconductor manufacture; and distillation columns. Details descriptions are given of the three points.
  • Characteristics of the plasma in a plasma-based manufacturing process step are monitored directly and in real time by observing the spectrum which it produces. An artificial neural network analyzes the plasma spectrum and generates control signals to control one or more of the process input parameters in response to any deviation of the spectrum beyond a narrow range. In an embodiment, a plasma reaction chamber forms a plasma in response to input parameters such as gas flow, pressure and power. The chamber includes a window through which the electromagnetic spectrum produced by a plasma in the chamber, just above themore » subject surface, may be viewed. The spectrum is conducted to an optical spectrometer which measures the intensity of the incoming optical spectrum at different wavelengths. The output of optical spectrometer is provided to an analyzer which produces a plurality of error signals, each indicating whether a respective one of the input parameters to the chamber is to be increased or decreased. The microcontroller provides signals to control respective controls, but these lines are intercepted and first added to the error signals, before being provided to the controls for the chamber. The analyzer can include a neural network and an optional spectrum preprocessor to reduce background noise, as well as a comparator which compares the parameter values predicted by the neural network with a set of desired values provided by the microcontroller. 7 figs.« less
  • Characteristics of the plasma in a plasma-based manufacturing process step are monitored directly and in real time by observing the spectrum which it produces. An artificial neural network analyzes the plasma spectrum and generates control signals to control one or more of the process input parameters in response to any deviation of the spectrum beyond a narrow range. In an embodiment, a plasma reaction chamber forms a plasma in response to input parameters such as gas flow, pressure and power. The chamber includes a window through which the electromagnetic spectrum produced by a plasma in the chamber, just above themore » subject surface, may be viewed. The spectrum is conducted to an optical spectrometer which measures the intensity of the incoming optical spectrum at different wavelengths. The output of optical spectrometer is provided to an analyzer which produces a plurality of error signals, each indicating whether a respective one of the input parameters to the chamber is to be increased or decreased. The microcontroller provides signals to control respective controls, but these lines are intercepted and first added to the error signals, before being provided to the controls for the chamber. The analyzer can include a neural network and an optional spectrum preprocessor to reduce background noise, as well as a comparator which compares the parameter values predicted by the neural network with a set of desired values provided by the microcontroller.« less
  • A control system for optimizing a chemical looping ("CL") plant includes a reduced order mathematical model ("ROM") that is designed by eliminating mathematical terms that have minimal effect on the outcome. A non-linear optimizer provides various inputs to the ROM and monitors the outputs to determine the optimum inputs that are then provided to the CL plant. An estimator estimates the values of various internal state variables of the CL plant. The system has one structure adapted to control a CL plant that only provides pressure measurements in the CL loops A and B, a second structure adapted to amore » CL plant that provides pressure measurements and solid levels in both loops A, and B, and a third structure adapted to control a CL plant that provides full information on internal state variables. A final structure provides a neural network NMPC controller to control operation of loops A and B.« less