Method for model-free control of general discrete-time systems

• US 5,513,098 A
• Filed: 12/27/1994
• Issued: 04/30/1996
• Est. Priority Date: 06/04/1993
• Status: Expired due to Term

• Current Assignee: Johns Hopkins University
• Original Assignee: Johns Hopkins University
• Inventors: Cristion, John A., Spall, James C.
• Primary Examiner(s): RUGGIERO, JOSEPH F
• Application Number:
  US08/364,069
• Time in Patent Office: 490 Days
• Field of Search: 364/148, 364/149-151, 364/157, 364/158, 364/164, 364/165, 395/21-23, 395/904, 395/905, 395/912, 395/913, 395/919, 395/906
• US Class Current: 700/38
• CPC Class Codes: G05B 13/027 using neural networks only

Abstract

A method of developing a controller for general (nonlinear) discrete-time systems, where the equations governing the system are unknown and where a controller is estimated without building or assuming a model for the system. The controller is constructed through the use of a function approximator (FA) such as a neural network or polynomial. This involves the estimation of the unknown parameters within the FA through the use of a stochastic approximation that is based on a simultaneous perturbation gradient approximation, which requires only system measurements (not a system model).

Claims

1. A method for approximating a controller for a nonlinear, discrete-time, closed-loop, stochastic system wherein the nonlinear functions governing the system dynamics and measurement process are unknown, the method comprising the steps of:

obtaining information about the system;

inputting the information about the system into a data processing means;

approximating the controller using the data processing means and the information about the system comprising the steps of;

selecting a single function approximator to approximate directly the controller;

estimating the unknown parameters of the single function approximator in the controller using a stochastic approximation algorithm; and

using the single function approximator to approximate the controller, wherein an output of the single function approximator is the controller; and

using the controller to control the system.

2. The method as recited in claim 1, the selecting a single function approximator step comprising the step of selecting a single continuous function approximator to approximate the controller.

3. The method as recited in claim 1, the estimating the unknown parameters step comprising the step of estimating the unknown parameters of the single function approximator in the controller using a simultaneous perturbation stochastic approximation algorithm.

4. The method as recited in claim 3, the selecting a single function approximator step comprising the step of selecting a neural network to approximate the controller.

5. The method as recited in claim 4, the selecting a neural network step comprising the step of selecting a multilayered, feed-forward neural network to approximate the controller.

6. The method as recited in claim 4, the selecting a neural network step comprising the step of selecting a recurrent neural network to approximate the controller.

7. The method as recited in claim 3, the selecting a single function approximator step comprising the step of selecting a polynomial to approximate the controller.

8. The method as recited in claim 3, the selecting a single function approximator step comprising the step of selecting a spline to approximate the controller.

9. The method as recited in claim 3, the selecting a single function approximator step comprising the step of selecting a trigonometric series to approximate the controller.

10. The method as recited in claim 3, the selecting a single function approximator step comprising the step of selecting a radial basis function to approximate the controller.