×

Neural network predictive control cost function designer

  • US 20050049728A1
  • Filed: 08/28/2003
  • Published: 03/03/2005
  • Est. Priority Date: 08/28/2003
  • Status: Active Grant
First Claim
Patent Images

1. A method for tuning a cost function used by a neural network control system configured to control an operational plant having a known plant phase response to each of a range of known signals, the neural network control system including a neural network model and a cost function, the method comprising:

  • selecting parameters used in a cost function;

    selecting an input weight to be applied to a control output by the cost function;

    selectively incorporating predicted future states generated by a neural network model;

    iteratively applying a control input from a range of known signals;

    calculating a control output in response to the control input;

    determining a control system phase and a control system amplitude of the control output in response to the control input; and

    combining a known plant phase with regards to a known signal equivalent to the control input and the control system phase such that effectiveness of the cost function parameters, the input weight, and the selectively incorporated predicted future states is determinable.

View all claims
  • 1 Assignment
Timeline View
Assignment View
    ×
    ×