Method and apparatus for adaptive control

• US 5,335,164 A
• Filed: 07/31/1991
• Issued: 08/02/1994
• Est. Priority Date: 07/31/1991
• Status: Expired due to Term

• Current Assignee: Universal Dynamics, Inc. (Pentafin SpA)
• Original Assignee: Universal Dynamics, Inc. (Pentafin SpA)
• Inventors: Gough, William A. G. Jr., Lyon, David L.
• Primary Examiner(s): Smith, Jerry
• Assistant Examiner(s): GORDON, PAUL P
• Application Number:
  US07/739,013
• Time in Patent Office: 1,098 Days
• Field of Search: 364/149, 364/151, 364/163, 364/153, 364/148
• US Class Current: 700/29
• CPC Class Codes: G05B 13/024 in which a parameter or coefficient is automatically adjusted to optimise the performance

Abstract

The present invention provides a method and apparatus for adaptive control which does not require a predetermined model of the process to be controlled. The method comprises the steps of 1) setting an initial state vector; 2) setting an initial parameter vector; 3) setting an initial prediction parameter gain; 4) set the initial covariance matrix; 5) performing a state update 6) estimating the model error; 7) updating the parameter vector, 8) updating the co-variance matrix, and 9) updating the controller output.

Claims

1. A method of generating a control signal for controlling a process, during each of a plurality of sampling instants, said process having at least one input variable and one output variable, the application of said control signal to said process constituting a control action causing a variation of said input variable, said method comprising:

a) providing an adaptive controller adapted to store a model of said process comprising a state vector and a model parameter vector and to generate said control action;

b) providing means for measuring said output variable of said process, generating a signal representing said measurement and communicating said measurement to said adaptive controller;

c) selecting a number n of orthonormal series functions to model said process;

d) providing a signal to define an initial state vector l(t)_t=O = ##EQU12## where d_i is a constant, and storing said initial state vector in said adaptive controller;

e) providing a signal to define an initial predicted model parameter vector= ##EQU13## where h_i is a constant, and storing said initial predicted model parameter vector in said adaptive controller;

f) performing an update of said state vector in said controller using said orthonormal series functions to provide a history of said control actions and any feedforward input variables up to that sampling instant;

g) measuring said process output variable and calculating a model error in the predicted model parameter vector by comparing said predicted model parameter vector to said measured process output variable;

h) updating the model parameter vector with said model error by a Kalman filter recursive least squares method;

i) calculating the change in controller output required to bring said process output variables to a desired value based on a predicted process output variable obtained from said updated model parameter vector and said updated state vector;

j) generating a control action based on said calculated required change in controller output;

k) repeating steps f) through j) at each sampling instant;

l) storing a reference model parameter vector in said controller during a period of satisfactory controller operation; and

m) periodically replacing said model parameter vector in said controller with said reference vector if a steadily decreasing prediction parameter gain is detected.

2. The method of claim 1 wherein said recursive least squares method is an EFRA parameter estimation.

3. The method of claim 1 wherein said orthonormal series functions are Laguerre functions.

4. The method of claim 1 wherein the initial state vector 1(t)= ##EQU14##

5. The method of claim 1 wherein the initial predicted model parameter vector= ##EQU15##

6. The method of claim 1 wherein an initial prediction parameter gain is set at 100.

7. The method of claim 1 wherein an initial covariance matrix for use in updating said model parameter vector is set at (10)*I, where I is the identity matrix.