Method and apparatus utilizing neural networks to predict a specified signal value within a multi-element system

• US 5,349,541 A
• Filed: 01/23/1992
• Issued: 09/20/1994
• Est. Priority Date: 01/23/1992
• Status: Expired due to Fees

• Current Assignee: Electric Power Research Institute
• Original Assignee: Electric Power Research Institute Incorporated
• Inventors: Khadem, Mostafa, Colley, Robert W., Alexandro, Frank J. Jr., Ipakchi, Ali
• Primary Examiner(s): Teska, Kevin J.
• Assistant Examiner(s): CHOI, KYLE JAEHUN
• Application Number:
  US07/824,794
• Time in Patent Office: 971 Days
• Field of Search: 364/148, 364/150, 364/495, 364/550, 364/551.01, 364/578, 395/20-24, 395/906, 395/907, 395/914, 395/915, 376/214-216
• US Class Current: 703/18
• CPC Class Codes: G06N 3/105 Shells for specifying net layout
• CPC Class Codes: Y10S 706/907 Power plant

Abstract

A method and apparatus for predicting a signal value for a target element within a multi-element system is disclosed. The method includes modeling the multi-element system by defining fundamental physical relationships between the target element and other elements within the system. The resultant system model is in the form of a set of coupled non-linear differential equations. These differential equations are then approximated into linearized models about an operating point or series of operating points corresponding to the system behavior. The linearized differential equations are then subjected to a coupling analysis. The coupling analysis is employed to determine dynamic coupling between instruments. The coupling analysis assesses the degree of observability of the system and associated elements. The coupling analysis may be based upon observability tests, gramian analyses, or modal analyses. Based upon the coupling analysis, coupled elements are selected. The coupled elements correspond to system elements which are strongly coupled to the target element. A neural network is then trained using previous process values corresponding to the coupled elements. Thereafter, present operating system values corresponding to the coupled elements are fed to the trained neural network. The trained neural network processes the present operating system values to render a predicted value for the target element. This predicted value is then compared to the present system value to determine whether the target element is operating correctly.

Claims

1. An apparatus for predicting a signal value for a target element within a multi-element system, said apparatus comprising:

a storage device containing a system model defining fundamental physical relationships between said target element and said multi-element system, said fundamental physical relationships of said system model being defined in accordance with the laws of physics;

a processor interpreting said system model to identify a reduced data set of coupled elements within said system model, said reduced data set of coupled elements corresponding to elements within said multi-element system which are coupled to said target element;

a neural network trained with process values corresponding to said reduced data set of coupled elements, said neural network processing measured system values corresponding to said reduced data set of coupled elements to render a predicted value for said target element;

means for comparing said predicted value from said neural network to a measured value for said target element to determine whether said target element is malfunctioning; and

an indicator responsive to said comparing means to identify when said target element is malfunctioning.

2. The apparatus of claim 1 wherein said processor includes

means for linearizing said system model by defining a plurality of differential equations characterizing said system model;

and

means for executing a coupling analysis on said differential equations to form a matrix.

3. The apparatus of claim 2 wherein said executing means performs a modal analysis.

4. A method for predicting a signal value for a target element within a multi-element system, said method comprising the steps of:

storing in a memory a system model defining fundamental physical relationships between said target element and said multi-element system, said fundamental physical relationships of said system model being defined in accordance with the laws of physics;

interpreting said system model with a processor to identify a reduced data set of coupled elements within said system mode, said reduced data set of coupled elements corresponding to elements within said multi-element system which are coupled to said target element;

training a neural network with process values corresponding to said reduced data set of coupled elements;

utilizing said neural network to process measured system values corresponding to said reduced data set of coupled elements to render a predicted value for said target element;

comparing said predicted value from said neural network to a measured value for said target element to determine whether said target element is malfunctioning; and

actuating an indicator responsive to said comparing step to identify when said target element is malfunctioning.

5. The method of claim 4 wherein said interpreting step includes the steps of:

linearizing said system model by defining a plurality of differential equations characterizing said system model; and

executing a coupling analysis on said differential equations to form a matrix.

6. The method of claim 5 wherein said executing step includes a modal analysis.