Neural networks for intelligent control

• US 6,882,992 B1
• Filed: 09/01/2000
• Issued: 04/19/2005
• Est. Priority Date: 09/02/1999
• Status: Expired due to Fees

• Current Assignee: IPU Power Management, LLC
• Original Assignee: Paul J. Werbos
• Inventors: Werbos, Paul J.
• Primary Examiner(s): Knight, Anthony
• Assistant Examiner(s): Holmes, Michael B
• Application Number:
  US09/654,396
• Time in Patent Office: 1,691 Days
• Field of Search: 706/48
• US Class Current: 706/48
• CPC Class Codes: G05B 13/027 using neural networks only
• CPC Class Codes: G05B 23/024 Quantitative history assessment, e.g. mathematical relationships between available data; Functions therefor; Principal component analysis [PCA]; Partial least square [PLS]; Statistical classifiers, e.g. Bayesian networks, linear regression or correlation analysis; Neural networks
• CPC Class Codes: G06N 3/105 Shells for specifying net layout

Abstract

A method and system for implementing a neuro-controller. One example of a neuro-controller is a brain-like stochastic search. Another example is a neuro-controller for controlling a hypersonic aircraft. Using a variety of learning techniques, the method and system provide adaptable control of external devices (e.g., airplanes, plants, factories, and financial systems).

Claims

1. A computer program product, comprising:

a computer storage medium and a computer program code mechanism embedded in the computer storage medium for causing a neural network to control an external device, the computer program code mechanism comprising;

a first computer code device configured to represent U(u,X) in computer-readable form, where U(u,X) is a family of max problems;

a second computer code device configured to implement a learning algorithm; and

a third computer code device configured to utilize the second computer code device on the computer-readable form of U(u,X) to learn to find u which maximizes U(u,X).

2. The computer program product as claimed in claim 1, wherein the first computer code device comprises a computer-readable form of U(u,X) that represents a family of all traveling salesman problems.

3. An artificial neural network to control an external device, comprising:

means for representing a family of max problems, U(u,X); and

learning means for learning to find u which maximizes U(u,X).

4. The artificial neural network of claim 3, wherein U(u,X) represents a family of all traveling salesman problems.