Neural network predictive control method and system

Claim 3

The method of claim 1 wherein the step of modifying the input control signal comprises the steps of: modifying the trial control inputs based on the performance index; predicting another set of future output states n time steps into the future with the neural network, said future output states of the other set of future output states being representative of the output signal for the next n consecutive future states of the plant over n time steps into the future in response to the modified trial control inputs, said neural network predicting the other set of future output states based on the past input and output signals and the modified future trial control inputs without using a recurrent network; determining the performance index as a function of the other set of future output states, said performance index being indicative of plant performance over time in response to the modified trial control inputs; and repeating the steps of modifying the trial control inputs, generating the other set of future output states and determining the performance index until the desired performance index is reached.

Claim 4

The method of claim 1 wherein each of the trial control inputs has a signal level representative of the input control signal corresponding to the next n consecutive future states of the plant and further comprising the step of setting the signal levels of the trial control inputs equal to each other for each iteration.

Claim 5

The method of claim 1 further comprising the steps of generating an error signal as a function of the difference between the predicted future output states and the output signal and feeding the error back into the neural network.

Claim 6

The method of claim 1 wherein the performance index is a cost function.

Claim 7

The method of claim 1 wherein the step of determining the performance index includes calculating the following cost function: C = ∑ i = 1 n   ( G p * Y i 2 + G v * Y . i 2 + G I * u 2 ) * W i where: C=cost of a selected trial control input (u), i=horizon index, n=horizon window, G p =position gain, Y i =predicted future output state of the plant at horizon i, G v =velocity gain, G I =input gain, W i =future state weighting factor.

Claim 8

The method of claim 1 wherein the plant includes an actuator for controlling the plant as a function of the input control signal.

Claim 9

The method of claim 7 wherein the plant comprises an aircraft wing and the actuator controls the wing for flutter suppression.

Claim 10

The method of claim 1 wherein n defines a horizon window representing a plurality of time steps into the future.

Claim 13

The system of claim 10 wherein the neural network predicts another set of future output states n time steps into the future, said future output states of the other set of future output states being representative of the output signal for the next n consecutive future states of the plant over n time steps into the future, in response to the modified trial control inputs, said neural network predicting the other set of future output states based on the past input and output signals and the modified future trial control inputs, and wherein the processor determines the performance index as a function of the other set of future output states, said performance index being indicative of plant performance over time in response to the modified trial control inputs.

Claim 14

The system of claim 10 wherein each of the trial control inputs has a signal level representative of the input control signal corresponding to the next n consecutive future states of the plant and wherein the signal levels of the trial control inputs are equal to each other for each iteration.

Claim 15

The system of claim 10 further comprising a summing function for generating an error signal as a function of the difference between the predicted future output states and the output signal and a feedback loop for feeding the error back into the neural network.

Claim 16

The system of claim 10 wherein the performance index is a cost function.

Claim 17

The system of claim 10 wherein the performance index is calculated according to the following cost function: C = ∑ i = 1 n   ( G p * Y i 2 + G v * Y . i 2 + G I * u 2 ) * W i where: C=cost of a selected trial control input (u), i=horizon index, n=horizon window, G p =position gain, Y i =predicted future output state of the plant at horizon i, G v =velocity gain, G I =input gain, W i =future state weighting factor.

Claim 18

The system of claim 10 wherein the plant includes an actuator for controlling the plant as a function of the input control signal.

Claim 19

The method of claim 16 wherein the plant comprises an aircraft wing and the actuator controls the wing for flutter suppression.

Claim 20

The system of claim 10 wherein n defines a horizon window representing a plurality of time steps into the future.

Claim 21

A neural network predictive control system for suppressing structural vibration in a plant, said plant receiving an input control signal for controlling the plant and generating an output signal representative of a state of the plant in response to the received input control signal, said system comprising: a memory storing the input and output signals corresponding to m consecutive past states of the plant over m time steps into the past; a non-recurrent computer neural network having a parallel processing arrangement, said neural network being responsive to a set of trial control inputs for predicting a set of future output states n time steps into the future for each trial control input, said future output states being representative of the output signal corresponding to the next n consecutive future states of the plant over n time steps into the future in response to the trial control inputs, said trial control inputs being representative of the input control signal corresponding to the next n consecutive future states of the plant over n time steps into the future, said neural network predicting the future output states based on the past input and output signals and the future trial control inputs; and a processor for generating the input control signal for controlling the plant, said processor: generating the trial control inputs; determining a performance index as a function of the future output states, said performance index being indicative of plant performance over time in response to the trial control inputs; and modifying the input control signal as a function of the trial control inputs for controlling the plant so that the performance index reaches a desired value.

Claim 22

A neural network predictive control system for suppressing structural vibration in a plant, said plant receiving an input control signal for controlling the plant and generating an output signal representative of a state of the plant in response to the received input control signal, said system comprising: a memory storing the input and output signals corresponding to m consecutive past states of the plant over m time steps into the past; a non-recurrent computer neural network having a parallel processing arrangement, said neural network being responsive to a set of trial control inputs for predicting a set of future output states n time steps into the future for each trial control input, said future output states being representative of the output signal corresponding to the next n consecutive future states of the plant over n time steps into the future in response to the trial control inputs, said trial control inputs being representative of the input control signal corresponding to the next n consecutive future states of the plant over n time steps into the future, said neural network predicting the future output states based on the past input and output signals and the future trial control inputs; and a processor for generating the input control signal for controlling the plant, said processor: generating the trial control inputs; determining a performance index as a function of the future output states, said performance index being indicative of plant performance over time in response to the trial control inputs; and modifying the input control signal as a function of the trial control inputs for controlling the plant so that the performance index reaches a desired value.

Claim 23

The system of claim 19 wherein the plant comprises an aircraft wing and wherein the plant includes an actuator for suppressing vibration in the wing as a finction of the input control signal.