Process control using neural network
First Claim
1. A control system for a continuous physical process characterized by directly controlled process variables and indirectly controlled process variables, said variables having current values and having had historical values that define the state of the physical process wherein the current values of directly controlled process variables are measured and wherein the current value of at least one indirectly controlled process variable is controlled indirectly by control of the current values of a plurality of directly controlled process variables comprising:
- control means responsive to set point values for establishing the current values of the directly controlled process variables at said set point values applied to said control means,means for implementing a trainable neural network having a plurality of input neurons for having input values applied thereto and at least one output neuron for providing an output value,means for training the neural network to provide a predicted value for the at least one indirectly controlled process variable at an output neuron, said predicted value corresponding to the input values of the neural network,means for measuring the values of directly controlled process variables,means for establishing and continuously updating a computer database to store the current and historical values of measured process variables,computer means for establishing the input values at the input neurons of the neural network based upon the values of process variables stored in the computer database,computer means for establishing set point values to be applied to control means based upon values at an output neuron,said control system so constructed and arranged that said computer means for establishing set point values, after the neural network has been trained to predict the value of the at least one indirectly controlled process variable, changes at least one set point value to cause the predicted value of the at least one indirectly controlled process variable to approach a desired value.
1 Assignment
0 Petitions
Accused Products
Abstract
A control system and method for a continuous process in which a trained neural network predicts the value of an indirectly controlled process variable and the values of directly controlled process variables are changed to cause the predicted value to approach a desired value.
-
Citations
31 Claims
-
1. A control system for a continuous physical process characterized by directly controlled process variables and indirectly controlled process variables, said variables having current values and having had historical values that define the state of the physical process wherein the current values of directly controlled process variables are measured and wherein the current value of at least one indirectly controlled process variable is controlled indirectly by control of the current values of a plurality of directly controlled process variables comprising:
-
control means responsive to set point values for establishing the current values of the directly controlled process variables at said set point values applied to said control means, means for implementing a trainable neural network having a plurality of input neurons for having input values applied thereto and at least one output neuron for providing an output value, means for training the neural network to provide a predicted value for the at least one indirectly controlled process variable at an output neuron, said predicted value corresponding to the input values of the neural network, means for measuring the values of directly controlled process variables, means for establishing and continuously updating a computer database to store the current and historical values of measured process variables, computer means for establishing the input values at the input neurons of the neural network based upon the values of process variables stored in the computer database, computer means for establishing set point values to be applied to control means based upon values at an output neuron, said control system so constructed and arranged that said computer means for establishing set point values, after the neural network has been trained to predict the value of the at least one indirectly controlled process variable, changes at least one set point value to cause the predicted value of the at least one indirectly controlled process variable to approach a desired value. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8)
-
-
9. A control system for a process characterized by directly controlled process variables and indirectly controlled variables, all said process variables having values that define the state of the process wherein the values of a plurality of process variables are measured and wherein the value of at least one process variable is controlled indirectly by control of a plurality of directly controlled process variables comprising:
-
control means for establishing the value of the directly controlled variables at set point values applied to said control means, means for implementing a neural network having a plurality of input neurons for having input values applied thereto and at least one output neuron for providing an output value, computer means for storing an input value for input neurons of the neural network and at least one computer executable rule to establish the input value associated with an input neuron, computer means for storing an output value for output neurons of the neural network and at least one computer executable rule for an output neuron, computer means for establishing a process database defining the state of the process including the instantaneous values of the process variables, means for continuously updating the process database to reflect the values of the measured process variables, computer means for executing a rule associated with an input neuron to establish the value of the input neuron, computer means for executing a rule associated with an output neuron for establishing set point values to be applied to control means, means for applying the set point values to the control means, means for training the neural network to generate a predicted value corresponding to the value of said indirectly controlled variable at an output neuron from values at input neurons corresponding to measured process variables, said control system so constructed and arranged that after the neural network has been trained to generate the predicted value of said indirectly controlled variable a rule associated with said at least one output neuron changes at least one set point value to cause the predicted value of the indirectly controlled variable to approach a directed value. - View Dependent Claims (10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31)
-
Specification