Residual activation neural network

• US 6,363,289 B1
• Filed: 01/12/1999
• Issued: 03/26/2002
• Est. Priority Date: 09/23/1996
• Status: Expired due to Fees

• Current Assignee: Rockwell Automation Technologies Incorporated (Rockwell Automation, Inc.)
• Original Assignee: Pavilion Technologies Incorporated (Rockwell Automation, Inc.)
• Inventors: Liano, Kadir, Ferguson, Ralph Bruce, Hartman, Eric Jon, Keeler, James David
• Primary Examiner(s): Gordon, Paul P.
• Application Number:
  US09/228,962
• Time in Patent Office: 1,169 Days
• Field of Search: 700/28, 700/44, 700/45, 700/47, 700/48, 700/29, 700/30, 706/15, 706/16, 706/21, 706/23, 706/26, 706/31
• US Class Current: 700/48
• CPC Class Codes: G05B 13/027 using neural networks only

Abstract

A plant (72) is operable to receive control inputs c(t) and provide an output y(t). The plant (72) has associated therewith state variables s(t) that are not variable. A control network (74) is provided that accurately models the plant (72). The output of the control network (74) provides a predicted output which is combined with a desired output to generate an error. This error is back propagated through an inverse control network (76), which is the inverse of the control network (74) to generate a control error signal that is input to a distributed control system (73) to vary the control inputs to the plant (72) in order to change the output y(t) to meet the desired output. The control network (74) is comprised of a first network NET 1 that is operable to store a representation of the dependency of the control variables on the state variables. The predicted result is subtracted from the actual state variable input and stored as a residual in a residual layer (102). The output of the residual layer (102) is input to a hidden layer (108) which also receives the control inputs to generate a predicted output in an output layer (106). During back propagation of error, the residual values in the residual layer (102) are latched and only the control inputs allowed to vary.

Claims

1. A control network for controlling a system having system control inputs for receiving system control variables and desired system outputs, the system outputs being a function of the system control variables and external influences on the system, comprising:

a control network input for receiving as network inputs the current system control variables and desired system outputs;

a control network output for outputting predicted system control variables necessary to achieve the desired system outputs;

a processing system for processing the received system control variables through an inverse representation of the system that represents the dependencies of the system output on the system control variables parameterized by an estimation of the external influences to provide the predicted system control variables to achieve the desired system outputs; and

an interface device for inputting the predicted system control variables that are output by said control network output to the system as system control variables to achieve the desired system outputs.

2. The control network of claim 1, wherein the control variable inputs are variables that can be manipulated.

3. The control network of claim 1, wherein the inverse representation of said processing system is a general non-linear inverse representation.

4. The control network of claim 1 wherein said processing system further comprises:

an estimation network for estimating the external influences on the system and output estimated external influences; and

means for parameterizing the inverse representation of the system with the estimated influences.

5. The control network of claim 4, wherein said processing system comprises:

a first intermediate output for providing a predicted system output;

a first intermediate processing system for receiving the system control variables from said control network input and the estimated external influences from said estimation network for processing through a predictive model of the system to generate the predicted system outputs for output from said intermediate output;

an error generation device for comparing the predicted system outputs to the desired system outputs and generating an error representing the difference therebetween;

a second intermediate processing system for processing the error through the inverse representation of the system that represents the dependencies of the system output on the system control variables parameterized by the estimated external influences to output predicted control variable change values; and

a control system for inputting said predicted control variable change values to the input of said first intermediate processing system for summing with the control variable input to provide a summed control variable value, and processing the summed control variable through said first processing system to minimize said error and output the summed control variable value as the predicted control variables.

6. The control network of claim 5, wherein said second intermediate processing system comprises:

a neural network having an input layer for receiving said error;

an output layer for providing the predicted output of the system;

a hidden layer for mapping said input layer to said output layer through an inverse representation of the system that represents the dependencies of the system output on the system control variables parameterized by the external influences to provide as an output from the output layer the control variable change values.

7. The control network of claim 5 wherein said control system utilizes a gradient descent procedure to minimize said error.

8. The control network of claim 5 wherein said control system utilizes a Monte Carlo technique to minimize said error.

9. The control network of claim 8, wherein said representation stored in said residual activation network is a non-linear representation of the dependency of the state variables on the control variables and the representation in said hidden layer of said main neural network comprises a non-linear representation of the system output as a function of the input control variables and said residual.

10. The control network of claim 5, wherein said second intermediate processing system and said estimation network comprise:

a residual activation neural network having;

a residual neural network for receiving as inputs in an input layer the system control variables and non-manipulatable system state variables dependant on the system control variables, and mapping the received system control variables through a hidden layer to an output layer, the hidden layer having a representation of the dependencies of the system state variables on the system control variables to provide as an output from said output layer predicted state variables,

a residual layer for determining as a residual the difference between the system state variables and the predicted state variables as an estimation of the external influences on the system, and

a latch for latching said residual determined in said residual layer after determination thereof; and

a main neural network having;

an input layer for receiving the system control variables and said latched residual,

an output layer for outputting a predicted system output, and

a hidden layer for mapping said input layer to said output layer through a representation of the system as a function of the system control variable inputs and said latched residual, said main neural network operating in an inverse mode to receive at the output layer said error and back propagate said error through said hidden layer to said input layer with said residual latched in said latch to output from said input layer said predicted control variable change values.

11. A network for predicting system outputs and for receiving control variables and state variables, with the state variables having dependencies on the control variables, the control network for projecting out the dependencies of the state variables on the control variables, comprising:

a residual activation neural network for generating an estimation of external influences on the system and having;

an input layer for receiving the control variables,

an output layer for outputting predicted state variables,

a hidden layer for mapping said input layer to said output layer through a representation of the dependencies of the state variables on the control variables to generate said predicted state variables, and

a residual layer for determining as a residual the difference between said predicted state variables and the input state variables, said residual comprising an estimation of external influences on the system; and

a main neural network having;

an input layer for receiving as inputs the control variables and said residual,

an output layer for outputting a predicted system output, and

a hidden layer for mapping said input layer to said output layer through a representation of the system as a function of the control variables and said residual.

12. The network of claim 11, and further comprising:

means for generating an error between the predicted system output and a desired system output;

a latch for latching said residual in said input layer of said main neural network; and

means for operating said main neural network to provide the inverse of said associated representation and back propagate said error through said main neural network from said output layer to the control variable inputs of said input layer of said main neural network to generate predicted control variable change values necessary to achieve said desired system output.

13. The control network of claim 12, wherein said means for generating said error comprises:

a predictive model neural network for providing a representation of the system and for receiving the control variables and the state variables as inputs and predicting the output of the system as a predicted system output; and

a difference device for receiving said desired system output and said predicted system output and generating said error.

14. The control network of claim 12, wherein said means for back propagating error through said main neural network comprises means for back propagating error through said main neural network to define said predicted control variable change values, and iteratively summing said change values with the control variables to minimize said error in accordance with a back propagation-to-activation technique.

15. A predictive network for predicting the operation of a system in response to receiving manipulatable control variables and non-manipulatable state variables, comprising:

a residual network for projecting out the dependencies of the state variables on the control variables to generate an estimation of external influences on the system and having;

an input for receiving input control variables,

an output for outputting predicted state variables as a function of the input control variables,

a residual processing system for processing the input control variables through a representation of the dependencies of the state variables on the control variables to provide predicted state variables for output by said output, and

a residual layer for determining the difference between the input state variables and the predicted state variables, the difference comprising a residual, said residual comprising the estimation of external influences on the system; and

a main network having;

an input for receiving as inputs the input control variables and said residual,

an output for outputting a predicted output representing the predicted output of the system, and

a main processing system for processing the input control variables and said residual through a representation of the system as a function of the control variables and said residual.

16. The predictive network of claim 15 wherein said input, said output and said processing system of said residual network comprise a residual neural network having:

an input layer for receiving the input control variables;

an output layer for outputting said predicted state variables; and

a hidden layer for mapping said input layer to said output layer through a non-linear representation of the dependencies of the state variables on the control variables.

17. The predictive network of claim 15 wherein said main network comprises a main neural network having:

an input layer for receiving the input control variables and said residual output by said residual layer;

an output layer for outputting said predicted output representing the predicted output of the system; and

a hidden layer for mapping said input layer to said output layer through a non-linear representation of the system as a function of the control variables and said residual.

18. The predictive network of claim 17 wherein said main network has the hidden layer thereof trained through back propagation as a function of known input control variables and residuals from said residual layer, said residuals generated by said residual network, and said output layer of said main network having input thereto known target predicted outputs.

19. A method for controlling a system having system outputs and system control inputs for receiving system control variables and desired system outputs, the system outputs being a function of the system control variables and external influences on the system, comprising the steps of:

receiving the current system control variables and desired system outputs;

processing the received system control variables through an inverse representation of the system that represents the dependencies of the system output on the system control variables parameterized by an estimation of the external influences to provide the predicted system control variables necessary to achieve the desired system outputs;

outputting on an output the predicted system control variables necessary to achieve the desired system outputs; and

controlling the system with the predicted system control variables.

20. The method of claim 19 wherein the inverse representation of the processing system is a general non-linear inverse representation.

21. The method of claim 19 wherein the control variables are variables that can be manipulated.

22. The method of claim 19 and further comprising:

estimating the external influences on the system as estimated external influences; and

parameterizing the inverse representation of the system with the estimated external influences.

23. The method of claim 22 wherein the step of processing comprises:

processing in a first intermediate processing step the system control variables and the estimated external influences through a predictive model of the system to generate the predicted system outputs for output from an intermediate output;

comparing the predicted system outputs to the desired system outputs and generating an error representing the difference therebetween; and

processing in a second intermediate processing step the error through the inverse representation of the system that represents the dependencies of the system output on the system control variables parameterized by the estimated external influences to output predicted control variable change values; and

changing the input control variables to the first intermediate step by the control variable change values to provide the predicted system control variables.

24. The method of claim 23 wherein the second intermediate processing step comprises:

receiving on input layer of a neural network the error;

mapping the neural network input layer to a neural network output layer through a neural network hidden layer having stored therein a local representation of the system parameterized by the external influences; and

operating the neural network in an inverse relationship wherein the error is received as an input in the output layer and propagated through the hidden layer having a local inverse representation of the system that represents the dependencies of the system output on the system control variables parameterized by the estimate external influences to provide as an output from the neural network input layer the predicted system control variable change values, wherein the error is back propagated through the neural network hidden layer to the neural network input layer.

25. The method of claim 23 wherein the first intermediate processing step includes the step of estimating and comprises:

receiving the system control variables on an input layer to a residual neural network and mapping the received system control variables to a residual neural network output layer through a hidden layer, the hidden layer having a representation of the dependencies of non-manipulatable system state variables on the system control variables to provide from the output layer predicted state variables as a function of the system control variables, the residual comprising the estimation of the external influences;

determining as a residual the difference between the system state variables and the predicted state variables;

latching the determined residual after determination thereof;

receiving the system control variables and the latched residual on an input layer of a main neural network; and

mapping the input layer of the main neural network to an output layer of the main neural network through a main neural network hidden layer having stored therein a representation of the system as a function of the system control variable inputs and the residual, to output from the output layer the predicted system outputs.

26. The method of claim 25 wherein the step of changing the input control variables comprises iteratively changing the input control variables by summing with the predicted control variable change values to minimize the error in accordance with a gradient descent technique.

27. A method for predicting system outputs from system inputs, comprising:

inputting the system inputs to a first input layer of a first neural network;

mapping the system inputs from the first input layer of the first neural network through a non-linear representation of the system in a first hidden layer to a first output layer, the non-linear representation having a residue associated therewith;

inputting the system inputs to a second input layer of a second neural network;

mapping the inputs from the second input layer through a second hidden layer in the second neural network to a second output layer in the second neural network, the second hidden layer having stored therein a non-linear representation of the residue of the hidden layer in the first neural network; and

linearly mapping the output of the first and second output layers to a primary output layer to provide the sum of the outputs of the first and second output layers.