Neural net based disturbance predictor for model predictive control

• US 5,408,406 A
• Filed: 10/07/1993
• Issued: 04/18/1995
• Est. Priority Date: 10/07/1993
• Status: Expired due to Term

• Current Assignee: Honeywell Incorporated (Honeywell International Inc.)
• Original Assignee: Honeywell Incorporated (Honeywell International Inc.)
• Inventors: Mathur, Anoop K., Gopinath, Ravi S.
• Primary Examiner(s): Envall, Jr., Roy N.
• Assistant Examiner(s): Garland, Steven R.
• Application Number:
  US08/133,542
• Time in Patent Office: 558 Days
• Field of Search: 364/148, 364/165, 364/176, 364/177, 364/180, 395/21-23, 395/903-910
• US Class Current: 700/45
• CPC Class Codes: G05B 13/027 using neural networks only
• CPC Class Codes: Y10S 706/903 Control

Abstract

A control loop for controlling a process or plant which controls the process or plant via an actuator. The control loop receives from the process or plant a signal representative of the process or plant output. The loop includes a nominal controller that generates a control signal for the actuator which is used only in the absence of a predicted disturbance to the process or plant signal from a disturbance mode controller unit having a neural network conditioned for predicting and indicating a disturbance.

Claims

1. A control loop for controlling a process having an output signal to control the process having an actuator, connected to the actuator for effecting control of the process, the control loop having an input signal representative of the process output, said control loop comprising:

a nominal controller, having a first output for controlling, and connected to, the actuator, having a first input for receiving a signal representative of a process output, second input and third inputs,

a disturbance mode controller unit (DMCU) having a first input for receiving a signal representative of the process output, having a second input connected to the first output of said nominal controller, having an output connected to the second input of said nominal controller, and having a third input; and

a periodicity indicator having an output connected to the third input of said disturbance mode controller unit; and

wherein;

said disturbance mode controller unit comprises a disturbance predictor unit having a neural network which contains likelihoods of expected disturbances with reference to said periodicity indicator, and in view of the signal representative of the process output at the first input of said disturbance mode controller unit and the first output of said nominal controller for controlling the actuator, at the second input of said disturbance mode controller unit, to predict a time and extent of an expected disturbance, and provides on the output of said disturbance mode controller unit a signal indicative of any predicted imminent disturbance, which enters the second input of said nominal controller; and

said nominal controller, having a setpoint signal at the third input of said nominal controller and the signal representative of the process output at the first input of said nominal controller to generate the output for controlling the actuator, in absence of the signal indicative of any imminent disturbance, and in presence of the signal indicative of any imminent disturbance, the said nominal controller generates its output from the signal indicative of any imminent disturbance from said disturbance mode controller unit, for controlling the actuator.

2. A control loop as set forth in claim 1 wherein said DMCU employs a signal from said disturbance predictor to said nominal controller output for a period of time selected by a user unless the process output varies from the setpoint by a user selected amount within said user selected time period.

3. A controller as set forth in claim 1 wherein the DMCU is a neural network trained in simulation on expected disturbance models.

4. A controller as set forth in claim 1 wherein a DMCU output is a signal indicating only whether or not a disturbance is predicted.

5. A controller as set forth in claim 1 wherein a DMCU output is a signal indicating the profile of y predicted by the DMCU, wherein Y is at least one output variable vector.