Method for process system identification using neural network

• US 5,740,324 A
• Filed: 07/11/1996
• Issued: 04/14/1998
• Est. Priority Date: 10/10/1990
• Status: Expired due to Fees

• Current Assignee: Honeywell (Honeywell International Inc.)
• Original Assignee: Honeywell (Honeywell International Inc.)
• Inventors: Mathur, Anoop, Samad, Tariq
• Primary Examiner(s): Davis, George B.
• Application Number:
  US08/678,130
• Time in Patent Office: 642 Days
• Field of Search: 395/23, 395/902, 395/903, 395/906, 395/22
• US Class Current: 706/16
• CPC Class Codes: G05B 17/02 electric
• CPC Class Codes: Y10S 706/902 Application using ai with detail of the ai system
• CPC Class Codes: Y10S 706/903 Control
• CPC Class Codes: Y10S 706/906 Process plant

Abstract

The method of making the tool, for process system identification that is based on the general purpose learning capabilities of neural networks. The method can be used for a wide variety of system identification problems with little or no analytic effort. A neural network is trained using a process model to approximate a function which relates process input and output data to process parameter values. Once trained, the network can be used as a system identification tool. In principle, this approach can be used for linear or nonlinear processes, for open or closed loop identification, and for identifying any or all process parameters.

Claims

1. A method for making a neural network tool for identifying parameters of a system which is modeled by an equation:

##EQU6## wherein x(t) is a response of the system, t_p is a time constant parameter of the system, K_p is a gain parameter of the system and θ

is a delay parameter of the system, said method comprising the steps of;

providing a neural network having an arrangement of processing elements, each of said elements having an input and an output, and adjustable weights connecting the outputs of some of said elements to the inputs of other of said elements, said network having input and output terminal means and target setting terminal means;

providing learning algorithm operational means for said network for adjusting said weights wherein output values on said output terminal means are biased to converge respectively to target values applied to said target setting terminal means;

making a data process system model of said equation and utilizing said model to generate sets of training data for said neural network with each of said sets having selected values of said parameters within respective predetermined ranges and a resulting response which is said x(t); and

sequentially applying said sets of training data to said neural network with each of said sets having said resulting response thereof applied to said input terminal means and said values of said parameters being applied to said target setting terminal means.

2. The method according to claim 1 wherein said system is accurately described by a linear first order equation.

3. The method according to claim 1 wherein each of said sets of training data applied to said network has only one of said parameters applied to said target setting terminal means.

4. The method according to claim 3 wherein said one of said parameters is said delay parameter of the system θ

.

5. The method according to claim 1 wherein each of said sets of training data includes a stimulus value for said model which is applied to said input terminal means of said network.

6. The method according to claim 5 wherein said stimulus value is a step input.

7. A neural network tool developed by a method for making the neural network tool for identifying parameters of a system which may be modeled by an equation:

##EQU7## wherein x(t) is a response of the system, t_p is a time constant parameter of the system, K_p is a gain parameter of the system and θ

is a delay parameter of the system, said method comprising the steps of;

providing a neural network having an arrangement of processing elements, each of said elements having an input and an output, and adjustable weights connecting the outputs of some of said elements to the inputs of other of said elements, said network having input and output terminal means and target setting terminal means;

providing learning algorithm operational means for said network for adjusting said weights wherein output values on said output terminal means are biased to converge respectively to target values applied to said target setting terminal means;

making a data process system model of said equation and utilizing said model to generate sets of training data for said neural network with each of said sets having selected values of said parameters within respective predetermined ranges and a resulting response which is said x(t); and

sequentially applying said sets of training data to said neural network with each of said sets having said resulting response thereof applied to said input terminal means and said values of said parameters being applied to said target setting terminal means.

8. The neural network tool developed by the method according to claim 7 wherein said system is accurately described by a linear first order equation.

9. The neural network tool developed by the method according to claim 7 wherein each of said sets of training data applied to said network has only one of said parameters applied to said target setting terminal means.

10. The neural network tool developed by the method according to claim 9 wherein said one of said parameters is said delay parameter of the system θ

.

11. The neural network tool developed by the method according to claim 7 wherein each of said sets of training data includes a stimulus value for said model which is applied to said input terminal means of said network.

12. The method according to claim 11 wherein said stimulus value is a step input.