System and method for improving model product property estimates

• US 5,402,333 A
• Filed: 06/15/1992
• Issued: 03/28/1995
• Est. Priority Date: 06/15/1992
• Status: Expired due to Fees

• Current Assignee: E. I. du Pont de Nemours and Company (Corteva, Inc.)
• Original Assignee: E. I. du Pont de Nemours and Company (Corteva, Inc.)
• Inventors: Cardner, David V.
• Primary Examiner(s): GORDON, PAUL P
• Application Number:
  US07/898,603
• Time in Patent Office: 1,016 Days
• Field of Search: 364/151, 364/153, 364/164, 364/554, 364/510
• US Class Current: 700/31
• CPC Class Codes: G05B 13/042 in which a parameter or coefficient is automatically adjusted to optimise the performance
• CPC Class Codes: Y02P 90/02 Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]

Abstract

The present invention is a system and method for improving stream composition and/or product property estimates from process models, which better reflect the true conditions of the process. One or more process simulation models are run on a computer in parallel with the actual process to provide estimates of stream composition and/or product properties to be used to control the process. Adjustments are made to the models to maintain them in alignment with continuously measured key process variables that are closely related to stream composition and/or product properties where such a relationship exists. This greatly improves the ability of the model to track the actual process. Additional adjustments are made to both the models and to the model estimates based on differences between measured and calculated stream composition/product properties. The combination of model calibration adjustments and direct correction of model estimates greatly improves the accuracy of the stream composition/product property estimates. Adjustments are made to model input variables and model parameters through use of controllers acting on the difference between measured and calculated properties. Differences obtained using relative error or the logarithm of the ratio of the measured and calculated values produce improved adjustments of component concentrations at low concentrations. Artificial measurements of stream composition and/or product properties can be synthesized by combining model composition estimates and flows at two or more locations in the process with an analytical measurement to produce and artificial analytical value at a second location in the process.

Claims

1. A computer-based method for processing stream composition and/or product property estimates generated from dynamic process models, said stream composition and/or product property estimates used to control a physical process, comprising the steps of:

(1) measuring, in a process comprising one or more units and two or more streams, pressure, temperature, and flow of said streams to produce raw process variable signals;

(2) comparing said current raw process variable signals with previously predicted process variable signals to determine error measures;

(3) using said error measures to adjust the dynamic process models to reduce future estimates of said error measures;

(4) predicting future process variable signals and future stream composition and/or product property signals from the dynamic process models;

(5) using said predicted future stream composition and/or product property values to directly or indirectly change the state of a control element(s) of said process; and

(6) operating said process using said control element(s) to produce a product.

2. The computer based method of claim 1, wherein said current raw process variable signals are first processed to eliminate gross errors and provide improved estimates of the true values of said current raw process variable signals before comparing said improved estimates of the true values with said previously predicted process variable signals.

3. The computer-based method of claim 1, wherein said error measures are determined by comparing said current raw process variable signals with said previously predicted process variable signals to determine arithmetic errors.

4. The computer-based method of claim 3, wherein said error measures are determined by comparing said current raw process variable signals with said previously predicted process variable signals to determine relative errors based on said arithmetic errors divided by the range of said current raw process variable signals.

5. The computer-based method of claim 3, wherein said error measures are determined by comparing said current raw process variable signals with said previously predicted process variable signals to determine relative errors based on said arithmetic errors divided by the measured or calculated values.

6. The computer-based method claim 1, wherein said error measures are determined by comparing said current raw process variable signals with said previously predicted process variable signals to determine relative errors based on the logarithm of the ratio of said current raw process variable signals to said previously predicted process variable signals.

7. The computer-based method of claim 1, further comprising the steps of:

using said error measures as input to a controller; and

using said controller to adjust independent model variables in said model to reduce future estimates of said error measures.

8. The computer-based method of claim 1, further comprising the steps of:

comparing said error measures with its moving average error to determine a variance;

comparing said variance with the moving average variance to determine a deviation;

using said deviation and a normal frequency distribution to determine a weighing factor for said error measures;

using said weighing factor, said error measures, and said moving average error to determine a new moving average error;

using said new moving average error as input to a controller; and

using said controller to adjust alignment factors in said model to reduce future estimates of said error measures.

9. The computer-based method of claim 8, further comprising the steps of:

determining the number of consecutive times said deviation exceeds a threshold with the same sign for said error measures to determine a count;

using an artificial deviation and the normal frequency distribution to determine a weighting factor for said error measures;

using said weighting factor, said error measures, and said moving average error to determine a new moving average error;

using said new moving average error as input to a controller;

using said controller to adjust alignment factors in said model to reduce future estimates of said error measures.

10. The computer-based method of claim 9, wherein said artificial deviation is obtained by dividing said deviation by said count.

11. The computer-based method of claim 9, wherein said artificial deviation is obtained by multiplying said deviation by an amplification factor.

12. The computer-based method of claim 9, wherein said artificial deviation is obtained by multiplying said deviation by said amplification factor and dividing by said count.

13. The computer-based method of claim 8, further comprises the step of using said variance, said average variance, and a fixed weighting factor to determine a new moving average variance.

14. The computer-based method of claim 13, wherein said fixed weighting factor is adjusted by dividing by said deviation when said deviation exceeds a threshold.

15. The computer-based method of claim 1, further comprising the step of operating said process to create said current raw process variable signals.

18. The computer-based system of claim 3, further comprising a controller, for changing a state of a final control element of said process, connected directly or indirectly to use said predicted stream composition and/or product property signals as inputs, and connected to change said state in accordance with an operating objective.

16. A computer-based system for processing dynamic model generated stream composition and/or product property estimates, comprising:

(a) two or more sensors for producing raw pressure and temperature signals from two or more streams in a process comprising one or more units and two or more streams;

(b) one or more computer-based process dynamic models connected to receive limit checked pressure, temperature, and level signals, reconciled flows, feed composition signals, and model alignment factors, to produce predicted pressure and temperature signals and predicted stream composition and/or product property signals;

(c) a process variable comparison module connected to use said raw pressure and temperature signals and said predicted pressure and temperature signals to determine an error between said raw signals and said predicted signals;

(d) a statistical filter module connected to take said error and determine a new moving average error from the statistical history of said error;

(e) an alignment module connected to take said new moving average error and produce one or more model alignment factors and connected to adjust said one or more computer-based process dynamic models based on said one or more model alignment factors;

(f) a stream composition and/or product property controllers, connected to directly or indirectly use said predicted stream composition and/or product property signals as input, for producing a controller output signal in accordance with said input and a control objective; and

(g) an actuator connected to change one or more states of said process in accordance with said controller output signal.

17. A computer-based system for processing model generated stream composition and/or product property estimates, comprising:

two or more sensors for producing raw stream composition and/or product property signals from two or more streams in a process comprising one or more units and two or more streams;

one or more computer-based process models connected to receive reconciled flows and pressure, temperature, and level signals, said raw stream composition and/or product property signals, and model calibration factors, and produce predicted stream composition and/or product property signals;

a product property transform module connected to use said reconciled flows, said raw stream composition and/or product property signals, and said predicted stream composition and/or product property signals to produce artificial stream composition and/or product property signals for streams lacking raw stream composition and/or product property signals;

a product property comparison module connected to use said raw stream composition and/or product property signals, said artificial stream composition and/or product property signals, and said predicted stream composition and/or product property signals to determine an error between said raw signals and said predicted signals;

a statistical filter module connected to take the said error and determine a new moving average error from the statistical history of said error;

a calibration module connected to take the said new moving average error and produce model calibration factors for adjusting said process models; and

a product property corrector module connected to take said new moving average error and said predicted stream composition and/or product property signals and apply said new moving average error to said predicted stream composition and/or product property signals to produce a corrected stream composition and/or product property signal.

19. The computer-based system of claim 17, wherein said system further comprises:

a stream composition and/or product property controllers, connected to directly or indirectly use said predicted stream composition and/or product property signal as input, for producing a controller output signal in accordance with said input and a control objective; and

an actuator for changing the state of the process, connected to change said state in accordance with said controller output signal.

20. A computer-based method for processing dynamic model generated stream composition and/or product property estimates, comprising the steps of:

(1) measuring, in a process comprising one or more units and two or more streams, stream composition and/or product property signals of said streams to produce raw stream composition and/or product property signals;

(2) comparing said raw stream composition and/or product property signals with previously predicted stream composition and/or product property signals to determine an error measure;

(3) using said error measure to adjust dynamic process models to reduce future estimates of said error measure;

(4) predicting, using said dynamic process models, future process stream composition and/or product property signals for one or more of said streams;

(5) using said predicted future stream composition and/or product property values to directly or indirectly change the state of a control element(s) of said process; and

(6) operating said process using said control element(s) to produce a product.

21. The computer-based method of claim 20, further comprising the steps of:

comparing said error measure with its moving average error to determine the variance;

comparing said variation with the moving average variance to determine the deviation;

using said deviation and the normal frequency distribution to determine a weighing factor for said error measure;

using said weighing factor, said error measure, and said moving average error to determine a new moving average error;

using said new moving average error as input to a controller;

using said controller to adjust calibration factors in the said model to reduce future estimates of said error.

22. The computer-based method of claim 21, further comprising the steps of:

determining the number of consecutive times said deviation exceeds a threshold with the same sign for said error measure to determine a count;

using an artificial deviation and the normal frequency distribution to determine a weighting factor for said error measure;

using said weighting factor, said error measure, and said moving average error to determine a new moving average error;

using said new moving average error as input to a controller;

using said controller to adjust alignment factors in said model to reduce future estimates of said error measure.

23. The computer-based method of claim 22, wherein said artificial deviation is obtained by dividing said deviation by said count.

24. The computer-based method of claim 22, wherein said artificial deviation is obtained by multiplying said deviation by an amplification factor.

25. The computer-based method of claim 22, wherein said artificial deviation is obtained by multiplying said deviation by said amplification factor and dividing by said count.

26. The computer-based method of claim 21, further comprises the step of using said variance, said average variance, and a fixed weighting factor to determine a new moving average variance.

27. The computer-based method of claim 26, wherein said fixed weighting factor is adjusted by dividing by said deviation when said deviation exceeds a threshold.

28. The computer-based method of claim 20, wherein said error measure is determined by comparing said raw stream composition and/or product property signals with said predicted stream composition and/or product property signals to determine an arithmetic error.

29. The computer-based method of claim 20, wherein said error measure is determined by comparing said raw stream composition and/or product property signals with said predicted stream composition and/or product property signals to determine a relative error based on the arithmetic difference divided by the range of said raw stream composition and/or product property signals.

30. The computer-based method of claim 20, wherein said error measure is determined by comparing said raw stream composition and/or product property signals with said predicted stream composition and/or product property signals to determine a relative error based on the arithmetic difference divided by the measured or calculated composition and/or product property signals.

31. The computer-based method of claim 20, further comprises the step of determining a relative error based on the logarithm of the ratio of said raw stream composition and/or product property signals to said predicted stream composition and/or product property signals.

32. The computer-based method of claim 20, wherein said current raw stream composition and/or product property measurements may be obtained from continuous signals or may refer to measurements on samples taken more than one computer cycle in the past, wherein said predicted future stream composition and/or product property values refer to values computed immediately prior to taking samples, the frequency of sampling for said stream composition and/or product property measurements may be fixed or variable, and a delay between sampling and transmission of measured values may be fixed or variable.

33. The computer-based method of claim 20, further comprising the step of operating said process to create said raw stream composition and/or product property signals.