Inferential Sensors Developed Using Three-Dimensional Pareto-Front Genetic Programming

• US 20100049340A1
• Filed: 02/21/2008
• Published: 02/25/2010
• Est. Priority Date: 03/19/2007
• Status: Active Grant

• Current Assignee: Dow Global Technologies LLC (Dow, Inc.)
• Original Assignee: Dow Global Technologies, Inc. (Dow, Inc.)
• Inventors: Kordon, Arthur Karl, Smits, Guido Freddy
• Granted Patent: US 8,250,006 B2
• Time in Patent Office: Days
• US Class Current: 700/31
• CPC Class Codes: G05B 13/026 using a predictor
• CPC Class Codes: G05B 13/0265 the criterion being a learning criterion

Abstract

A predictive algorithm for predictive at least one output variable based on a plurality of input variables is developed using a genetic programming technique that evolves a population of candidate algorithms through multiple generations. Within each generation, the candidate algorithms are evaluated based on three fitness criteria: (i) an accuracy criterion that evaluates each candidate algorithm'"'"'s ability to predict historical measurements of the at least one output variable based on corresponding historical measurements of the input variables; (ii) a complexity criterion that evaluates each candidate algorithm'"'"'s complexity; and (iii) a smoothness criterion that evaluates each candidate algorithm'"'"'s nonlinearity. The predictive algorithm may be implemented in an inferential sensor that is used to monitor a physical, chemical, or biological process, such as an industrial process in an industrial plant.

Claims

1. A method of developing a predictive algorithm for predicting at least one output variable based on a plurality of input variables, said method comprising:

obtaining a first set of data representative of a physical, chemical, or biological process, said first set of data including first-set measurements of said at least one output variable and corresponding first-set measurements of said input variables;

evolving a plurality of candidate algorithms using a genetic programming technique that applies at least three fitness criteria, said at least three fitness criteria including an accuracy criterion that evaluates each candidate algorithm'"'"'s ability to predict said first-set measurements of said at least one output variable based on said corresponding first-set measurements of said input variables, a complexity criterion that evaluates each candidate algorithm'"'"'s complexity, and a smoothness criterion that evaluates each candidate algorithm'"'"'s nonlinearity; and

selecting one of said candidate algorithms as said predictive algorithm.

2. The method of claim 1, wherein said physical, chemical, or biological process is an industrial process.

3. The method of claim 1, wherein said at least one output variable relates to emissions monitoring.

4. The method of claim 1, wherein said at least one output variable is a physical property, chemical property, mechanical property, material property, or biological property.

5. The method of claim 1, wherein said predictive algorithm is optimal among said candidate algorithms with respect to said three fitness criteria.

6. The method of claim 1, wherein selecting one of said candidate algorithms as said predictive algorithm comprises:

determining which of said candidate algorithms lie on a three-dimensional Pareto-front with respect to said three fitness criteria to obtain a plurality of Pareto-front algorithms.

7. The method of claim 6, wherein selecting one of said candidate algorithms as said predictive algorithm further comprises:

applying at least one additional selection criterion to select said predictive algorithm from among said Pareto-front algorithms.

8. The method of claim 1, further comprising:

obtaining a second set of data representative of said physical, chemical, or biological process, said second set of data including second-set measurements of said at least one output variable and corresponding second-set measurements of said input variables; and

testing said predictive algorithm using said second set of data.

9. The method of claim 8, wherein testing said predictive algorithm using said second set of data comprises:

evaluating said predictive algorithm'"'"'s ability to predict said second-set measurements of said at least one output variable based on said corresponding second-set measurements of said input variables.

10. A method of monitoring a physical, chemical, or biological process, said method comprising:

determining current values of a plurality of measurable parameters of said physical, chemical, or biological process;

providing said current values to an inferential sensor that is configured to use at least one predictive algorithm that predicts an additional parameter of said physical, chemical, or biological process based on said measurable parameters, said at least one predictive algorithm having been optimized with respect to accuracy, complexity, and smoothness; and

said inferential sensor using said at least one predictive algorithm to determine an inferred value of said additional parameter based on said current values of said measurable parameters.

11. The method of claim 10, wherein determining current values of a plurality of measurable parameters of said physical, chemical, or biological process comprises:

measuring at least one of said available parameters using a hardware sensor.

12. The method of claim 10, wherein said process is an industrial process.

13. The method of claim 12, further comprising:

controlling said industrial process based on said inferred value of said additional parameter.

14. The method of claim 13, wherein controlling said industrial process based on said inferred value of said additional parameter comprises:

comparing said inferred value with a preset value.

15. The method of claim 10, wherein said additional parameter is a physical property, chemical property, mechanical property, material property, or biological property.

16. The method of claim 10, further comprising:

developing said at least one predictive algorithm using a genetic programming technique applied to historical values of said measurable parameters and historical values of said additional parameter, said historical values having been previously obtained for said physical, chemical, or biological process.

17. The method of claim 16, wherein said historical values of said measurable parameters define a first range, and wherein at least a subset of said current values of said measurable parameters are outside of said first range.

18. The method of claim 16, wherein said historical values of said additional parameter define a second range, and wherein said inferred value of said additional parameter is outside of said second range.

19. A system for monitoring a physical, chemical, or biological process, said system comprising:

at least one hardware sensor for obtaining measured values of a plurality of parameters of said physical, chemical, or biological process; and

an inferential sensor for determining an inferred value of an additional parameter of said physical, chemical, or biological process based on said measured values, said inferential sensor being configured to determine said inferred value using at least one predictive algorithm that has been optimized with respect to accuracy, complexity, and smoothness using a genetic programming technique.

20. The system of claim 19, further comprising:

a control system for controlling said physical, chemical, or biological process based on said inferred value.