Method and apparatus for modeling dynamic and steady-state processes for prediction, control and optimization

• US 6,738,677 B2
• Filed: 11/22/2002
• Issued: 05/18/2004
• Est. Priority Date: 05/06/1996
• Status: Expired due to Term

• Current Assignee: Rockwell Automation Technologies Incorporated (Allen-Bradley Co., Inc.)
• Original Assignee: Pavilion Technologies Incorporated (Rockwell Automation, Inc.)
• Inventors: Timmer, Douglas, Piche, Stephen, Martin, Gregory D., Boe, Eugene, Havener, John P., Gerules, Mark, Keeler, James David
• Primary Examiner(s): Cuchlinski, Jr., William A.
• Assistant Examiner(s): Marc, McDieunel, Marc, McDieunel
• Application Number:
  US10/302,923
• Publication Number:
  US 20030078684A1
• Time in Patent Office: 543 Days
• Field of Search: 700/27, 700/37, 700/29, 700/44, 700/28, 700/42, 700/32, 700/45, 700/72, 703/2, 703/16, 703/18, 703/23
• US Class Current: 700/44
• CPC Class Codes: G05B 13/048 using a predictor
• CPC Class Codes: G05B 17/02 electric

Abstract

A method for providing independent static and dynamic models in a prediction, control and optimization environment utilizes an independent static model (20) and an independent dynamic model (22). The static model (20) is a rigorous predictive model that is trained over a wide range of data, whereas the dynamic model (22) is trained over a narrow range of data. The gain K of the static model (20) is utilized to scale the gain k of the dynamic model (22). The forced dynamic portion of the model (22) referred to as the b_i variables are scaled by the ratio of the gains K and k. The b_i have a direct effect on the gain of a dynamic model (22). This is facilitated by a coefficient modification block (40). Thereafter, the difference between the new value input to the static model (20) and the prior steady-state value is utilized as an input to the dynamic model (22). The predicted dynamic output is then summed with the previous steady-state value to provide a predicted value Y. Additionally, the path that is traversed between steady-state value changes.

Claims

1. A dynamic controller for controlling operation of a system by predicting a change in the dynamic parameter values to the system to effect a change in the output of the system from a current output value at a first time to a desired output value at a second and future time, comprising:

a dynamic predictive model for receiving the current input value and the desired output value and predicting the dynamic response of the system at different time positions between the first time and the second time to define a dynamic operation path of the system between the current output value and the desired output value at the second time; and

an optimizer for optimizing the operation of the dynamic controller at each of the different time positions from the first time to the second time in accordance with a predetermined optimization method that optimizes the objectives of the dynamic controller to achieve a desired dynamic operation path, such that the objectives of the dynamic predictive model vary as a function of time.

2. The dynamic controller of claim 1, wherein said dynamic predictive model comprises:

a dynamic forward model operable to receive input values at each of said different time positions and map said received input values through a stored representation of the system to provide a predicted dynamic output value;

an error generator for comparing the predicted dynamic output value to the desired output value and generating a primary error value as the difference therebetween for each of said time positions;

an error minimization device for determining a change in the input value to minimize the primary error value output by said error generator;

a summation device for summing said determined input change value with the original input value for each time position to provide a future input value; and

a controller for controlling the operation of said error minimization device to operate under control of said optimizer to minimize said primary error value in accordance with said predetermined optimization method.

3. The dynamic controller of claim 2, wherein said controller controls the operation of said summation device to iteratively minimize said primary error value by storing the summed output from said summation device in a latch in a first pass through said error minimization device and input the latch contents to said dynamic forward model in a subsequent pass and for a plurality of subsequent passes, with the output of said error minimization device summed with the previous contents of said latch with said summation device, said latch containing the current value of the input on the first pass through said dynamic forward model and said error minimization device, said controller outputting the contents of said latch as the input to the system after said primary error value has been determined to meet the objectives in accordance with said predetermined optimization method.

4. The dynamic controller of claim 2, wherein said dynamic forward model is a dynamic linear model with a fixed gain.

5. The dynamic controller of claim 4 and further comprising a gain adjustment device for adjusting the gain of said linear model for substantially all of said time positions.

6. The dynamic controller of claim 5, wherein said gain adjustment device comprises:

a non-linear model for receiving an input value and mapping the received input value through a stored representation of the system to provide on the output thereof a predicted output value, and having a non-linear gain associated therewith;

said linear model having parameters associated therewith that define the dynamic gain thereof; and

a parameter adjustment device for adjusting the parameters of said linear model as a function of the gain of said non-linear model for at least one of said time positions.

7. The dynamic controller of claim 6, wherein said gain adjustment device further comprises an approximation device for approximating the dynamic gain for a plurality of said time positions between the value of the dynamic gain at said first time and the determined dynamic gain at the one of said time positions having the dynamic gain thereof determined by said parameter adjustment device.

8. The dynamic controller of claim 7, wherein the one of said time positions at which said parameter adjustment device adjusts said parameters as a function of the gain of said non-linear model corresponds to the maximum at the second time.

9. The dynamic controller of claim 6, wherein said non-linear model is a steady-state model.

10. The dynamic controller of claim 2, wherein said error minimization device includes a primary error modification device for modifying said primary error to provide a modified error value, said error minimization device optimizing the operation of the dynamic controller to minimize said modified error value in accordance with said predetermined optimization method.

11. The dynamic controller of claim 10, wherein said primary error is weighted as a function of time from the first time to the second time.

12. The dynamic controller of claim 11, wherein said weighting function decreases as a function of time such that said primary error value is attenuated at a relatively high value proximate to the first time and attenuated at a relatively low level proximate to the second time.

13. The dynamic controller of claim 2, wherein said error minimization device receives said predicted output from said dynamic forward model and determines a change in the input value maintaining a constraint on the predicted output value such that minimization of the primary error value through a determined input change would not cause said predicted output from said dynamic forward model to exceed said constraint.

14. The dynamic controller of claim 2, and further comprising a filter determining the operation of said error minimization device when the difference between the predicted manipulated variable and the desired output value is insignificant.

15. The dynamic controller of claim 14, wherein said filter determines when the difference between the predicted manipulated variable and the desired output value is not significant by determining the accuracy of the model upon which the dynamic forward model is based.

16. The dynamic controller of claim 15, wherein the accuracy is determined as a function of the standard deviation of the error and a predetermined confidence level, wherein said confidence level is based upon the accuracy of the training over the mapped space.

17. A method for predicting an output value from a received input value, comprising the steps of:

modeling a set of static data received from a system in a predictive static model over a first range, the static model having a static gain of K and modeling the static operation of the system;

modeling a set of dynamic data received from the system in a predictive dynamic model over a second range smaller than the first range to model the dynamic operation of the system, the dynamic model having a dynamic gain k, and the operation of the dynamic model being independent of the operation of the static model;

adjusting the gain of the dynamic model as a predetermined function of the gain of the static model to vary the model parameters of the dynamic model;

predicting the dynamic operation of the system for a change in the input value between a first input value at a first time and a second input value at a second time;

subtracting the input value from a steady-state input value previously determined and inputting the difference to the dynamic model and processing the input through the dynamic model to provide a dynamic output value; and

adding the dynamic output value from the dynamic model to a steady-state output value previously determined to provide a predicted value.

18. The method of claim 17, wherein the predetermined function is an equality function wherein the static gain K is equal to the dynamic gain k.

19. The method of claim 17, wherein the static model is a non-linear model.

20. The method of claim 19, wherein the dynamic model for a given dynamic gain is linear.

21. The method of claim 20, wherein the step of adjusting the gain of the dynamic model as a function of the predetermined function of the gain of the static model is a non-linear operation.

22. The method of claim 17, wherein the static model and the dynamic model are utilized in a control function to receive as inputs the manipulated inputs of the system, the actual output from the system in addition to a desired output value at which the system is to operate, and then perform the step of predicting future manipulated inputs required to achieve the desired output.

23. The method of claim 22, and further comprising the step of optimizing the operation of the dynamic model in accordance with a predetermined optimization method between an initial steady-state value and a predicted final steady-state value that constitutes an input control variable to the system during the control operation.

24. The method of claim 23, wherein the step of optimizing comprises determining the dynamic gain k for multiple positions between the input steady-state input value and the final predicted steady-state input value that comprises the input control values to the system.

25. The method of claim 24, wherein the step of determining utilizes an algorithm that estimates the dynamic gain k independent of the operation of the static model.

26. The method of claim 25, wherein the algorithm is a quadratic equation.

27. The method of claim 23, wherein the step of predicting with the dynamic model utilized as a dynamic controller comprises the steps of:

predicting the dynamic operation of the system from the initial steady-state input value to the predicted steady-state input value to provide a predicted dynamic operation;

comparing the predicted dynamic operation to the desired steady-state value at the final desired output value and generating an error as the difference therebetween;

determining a change in the input value for input to step of predicting the dynamic operation which will vary the input value thereto; and

varying the change in the input value to minimize the error.

28. The method of claim 27, wherein the step of detennining the error comprises multiplying the determined error value by a predetermined weighting value that is a function of time from the first time to the second time.

29. The method of claim 27, wherein the predetermined weighting function attenuates the error for values proximate in time to the initial steady-state value at the first time and decreases the attenuation value as time increases to the final steady-state value at the second time.