Hybrid linear-neural network process control

• US 6,278,962 B1
• Filed: 10/02/1998
• Issued: 08/21/2001
• Est. Priority Date: 05/03/1996
• Status: Expired due to Fees

• Current Assignee: Aspentech Corporation
• Original Assignee: Aspen Technology (Telefonaktiebolaget LM Ericsson)
• Inventors: Guiver, John P., Klimasauskas, Casimir C.
• Primary Examiner(s): Teska, Kevin J.
• Assistant Examiner(s): JONES, HUGH M
• Application Number:
  US09/165,854
• Time in Patent Office: 1,054 Days
• Field of Search: 703/13, 703/2
• US Class Current: 703/13
• CPC Class Codes: G05B 13/027 using neural networks only
• CPC Class Codes: G05B 13/0275 using fuzzy logic only
• CPC Class Codes: G05B 19/042 using digital processors G05B19/05 takes precedence

Abstract

A hybrid analyzer having a data derived primary analyzer and an error correction analyzer connected in parallel is disclosed. The primary analyzer, preferably a data derived linear model such as a partial least squares model, is trained using training data to generate major predictions of defined output variables. The error correction analyzer, preferably a neural network model is trained to capture the residuals between the primary analyzer outputs and the target process variables. The residuals generated by the error correction analyzer is summed with the output of the primary analyzer to compensate for the error residuals of the primary analyzer to arrive at a more accurate overall model of the target process. Additionally, an adaptive filter can be applied to the output of the primary analyzer to further capture the process dynamics. The data derived hybrid analyzer provides a readily adaptable framework to build the process model without requiring up-front knowledge. Additionally, the primary analyzer, which incorporates the PLS model, is well accepted by process control engineers. Further, the hybrid analyzer also addresses the reliability of the process model output over the operating range since the primary analyzer can extrapolate data in a predictable way beyond the data used to train the model. Together, the primary and the error correction analyzers provide a more accurate hybrid process analyzer which mitigates the disadvantages, and enhances the advantages, of each modeling methodology when used alone.

Claims

1. A method for modeling a process having one or more disturbance variables as process input conditions, one or more corresponding manipulated variables as process control conditions, and one or more corresponding controlled variables as process output conditions, said method comprising:

picking one or more selected variables from said disturbance variables and said manipulated variables;

providing said selected variables to a data derived primary analyzer and an error correction analyzer;

generating a primary output from said selected variables using said data derived primary analyzer;

generating a predicted error output from said selected variables using said error correction analyzer;

summing the output of said primary and error correction analyzers;

presenting said summed output to a distributed control system;

selecting and time-shifting pre-determining variables from said distributed control system using a run-time delay and variable selector;

presenting the output of said run-time delay and variable selector to said data derived primary analyzer and said error correction analyzer;

picking one or more training variables from disturbance variables and manipulated variables stored in said data repository, said training variables having a corresponding training controlled variable;

determining, said delay and variable settings from said training variables;

providing said training variables to a training primary analyzer and a training error correction analyzer;

generating a training primary output from said training variables using said training primary analyzer;

subtracting said training primary output from said training controlled variable to generate a feedback variable;

generating a predicted training error output from said training variables and said feedback variable using said training error correction analyzer, wherein said generating a predicted training error output further includes training a neural network partial least squares error correction analyzer, wherein said neural network partial least squares error correction analyzer has a non-linear function f(t_h) and an error function, wherein said training input vector is defined as $X=\underset{h=1}{\sum <br><br>}<br><br>\text{\hspace{1em}<br><br>}<br><br>t_h<br><br>P_h^{\mathrm{\prime <br><br>}}+E={\mathrm{TP}}^{\mathrm{\prime <br><br>}}+E,$

wherein said training primary output is defined as $Y=\underset{h=1}{\sum <br><br>}<br><br>\text{\hspace{1em}<br><br>}<br><br>u_h<br><br>q_h^{\mathrm{\prime <br><br>}}+F={\mathrm{UQ}}^{\mathrm{\prime <br><br>}}+F,$

wherein Y further equals TBQ′

+F, further comprising the step of minimizing said error function ∥

u_h−

f(t_h)∥

² in said neural network partial least squares error correction analyzer;

summing said training primary output and said predicted training error output;

updating said delay and variable settings and said model parameters;

computing a difference between said summed output of said summed training primary output and said predicted training error output and said training controlled variable;

repeating said step of determining delay and variable through said step of computing a difference until said the performance of said analyzer on a test data set reaches an optimum point;

storing said delay and variable settings in said run-time delay and variable selector;

storing said model parameters in said data derived primary analyzer and said error correction analyzer.

2. The process of claim 1, wherein said generating a primary output from said selected variables further comprises applying a linear model in the data derived primary analyzer.

3. The process of claim 2, wherein said applying a linear model step further comprises the step of applying a Partial Least Squares (PLS) model to generate said primary output.

4. The process of claim 1, wherein said generating a predicted error output step further comprises the step of applying a non-linear model in the error correction analyzer.

5. The process of claim 1, wherein said generating a predicted error output from said selected variables using said error correction analyzer further comprises the step of applying a neural network to generate said predicted error output.

6. The process of claim 5, wherein said neural network applying step further comprises the steps of:

computing a derivative of said primary output;

integrating said derivative; and

correcting said primary output.

7. The process of claim 1, wherein said training input vector is defined as $X=\sum _{}$

h=1

 



th

Ph′

+E=TP′

+E,

wherein said training primary output is defined as $Y=\underset{h=1}{\sum <br><br>}<br><br>\text{\hspace{1em}<br><br>}<br><br>u_h<br><br>q_h^{\mathrm{\prime <br><br>}}+F={\mathrm{UQ}}^{\mathrm{\prime <br><br>}}+F,$

wherein Y further equals TBQ′

+F, said training primary analyzer generating a regression model between T and U.

8. The process of claim 7, wherein said generating a primary output step further includes:

generating {circumflex over (t)}_h=E<sub>h−

1</sub>w_h;

generating E_h=E<sub>h−

1</sub>−

{circumflex over (t)}_hP′

_h; and

generating the primary output Y=Σ

b_h{circumflex over (t)}_hq′

h.