System and method for improving data coverage in modeling systems

• US 20090300052A1
• Filed: 05/30/2008
• Published: 12/03/2009
• Est. Priority Date: 05/30/2008
• Status: Active Grant

• Current Assignee: Caterpillar Incorporated
• Original Assignee: Caterpillar Incorporated
• Inventors: Seskin, Michael, Mason, James Robert, Grichnik, Anthony James
• Granted Patent: US 8,086,640 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G05B 17/02 electric

Abstract

A method for modifying data coverage in a modeling system is disclosed. The method may include obtaining data records relating to a plurality of input variables and one or more output parameters and selecting a plurality of input parameters from the plurality of input variables. The method may further include evaluating a coverage of the data records in a modeling space and modifying the coverage of the data records, if a data coverage condition is detected. The method may also include generating a computational model indicative of interrelationships between the plurality of input parameters and the one or more output parameters based on the data records.

Claims

1. A method for modifying data coverage in a modeling system, comprising:

obtaining data records relating to a plurality of input variables and one or more output parameters;

selecting a plurality of input parameters from the plurality of input variables;

evaluating a coverage of the data records in a modeling space;

modifying the coverage of the data records, if a data coverage condition is detected; and

generating a computational model indicative of interrelationships between the plurality of input parameters and the one or more output parameters based on the data records.

2. The method of claim 1, wherein evaluating the coverage of the data records further includes:

dividing the modeling space into a plurality of hyper-quadrants;

calculating a data density for each hyper-quadrant based on a number of data records in the respective hyper-quadrant;

generating a histogram based on the calculated data densities;

determining a statistical difference in a data density distribution based on the generated histogram; and

identifying the data coverage condition if the statistical difference exceeds a first threshold.

3. The method of claim 2, wherein the statistical difference is determined as a standard deviation of the data density distribution.

4. The method of claim 2, wherein evaluating the coverage of the data records further includes marking sparse hyper-quadrants with data densities below a second threshold, if the data coverage condition is identified.

5. The method of claim 1, wherein modifying the coverage of the data records further includes receiving additional data records from a user for sparse hyper-quadrants in the modeling space.

6. The method of claim 1, wherein modifying the coverage of the data records further includes removing data records from dense hyper-quadrants in the modeling space.

7. The method of claim 1, wherein modifying the coverage of the data records further includes creating additional data records for sparse hyper-quadrants in the modeling space via a symmetric random scatter process proposing input vectors of interest.

8. The method of claim 7, wherein creating the additional data records further includes:

determining starting ranges of the plurality of input parameters;

creating a centroid data record including midpoint values of the starting ranges of the plurality of input parameters;

randomly creating a plurality of sample data records of the plurality of input parameters within the starting ranges of the input parameters;

creating a plurality of SRS data records of the sample data records as symmetric scatters of the plurality of sample data records with respect to the centroid data record;

creating the additional data records by combining the centroid data record, the sample data records, and the SRS data records; and

removing the additional data records that fall out of the sparse hyper-quadrants.

9. The method of claim 1, further including statistically validating the computational model.

10. The method of claim 1, further including optimizing the computational model by maximizing a zeta statistic of the computational model.

11. A computer-based modeling system, comprising:

a database containing data records related to a plurality of input variables and one or more output parameters; and

a processor configured to;

obtain data records relating to the plurality of input variables and one or more output parameters;

select a plurality of input parameters from the plurality of input variables;

evaluate a coverage of the data records in a modeling space;

modify the coverage of the data records, if a data coverage condition is detected; and

generate a computational model indicative of interrelationships between the plurality of input parameters and the one or more output parameters based on the data records.

12. The computer-based modeling system of claim 11, wherein to evaluate the coverage of the data records, the processor is further configured to:

divide the modeling space into a plurality of hyper-quadrants;

calculate a data density for each hyper-quadrant based on a number of data records in the respective hyper-quadrant;

generate a histogram based on the calculated data densities;

determine a statistical difference in a data density distribution based on the generated histogram; and

identify the data coverage condition if the statistical difference exceeds a first threshold.

13. The computer-based modeling system of claim 12, wherein the statistical difference is determined as a standard deviation of the data density distribution.

14. The computer-based modeling system of claim 12, wherein to evaluate the coverage of the data records, the processor is further configured to mark sparse hyper-quadrants with data densities below a second threshold, if the data coverage condition is identified.

15. The computer-based modeling system of claim 11, wherein to modify the coverage of the data records, the processor is further configured to:

receive additional data records from a user for sparse hyper-quadrants in the modeling space;

remove data records from dense hyper-quadrants in the modeling space;

or create additional data records for sparse hyper-quadrants in the modeling space via a symmetric random scatter process.

16. The computer-based modeling system of claim 15, wherein to create the additional data records, the processor is further configured to:

determine starting ranges of the plurality of input parameters;

create a centroid data record including midpoint values of the starting ranges of the plurality of input parameters;

randomly create a plurality of sample data records of the input parameters within the starting ranges of the plurality of input parameters;

create a plurality of SRS data records of the sample data records as symmetric scatters of the plurality of sample data records with respect to the centroid data record;

create the additional data records by combining the centroid data record, the sample data records, and the SRS data records; and

remove the additional data records that fall out of the sparse hyper-quadrants.

17. The computer-based modeling system of claim 11, wherein the processor is further configured to:

statistically validate the computational model; and

optimize the computational model by maximizing a zeta statistic of the computational model.

18. A computer readable medium having stored thereon instructions for modifying data coverage in a modeling system, the instructions including:

instructions for obtaining data records relating to a plurality of input variables and one or more output parameters;

instructions for selecting a plurality of input parameters from the plurality of input variables;

instructions for evaluating a coverage of the data records in a modeling space;

instructions for modifying the coverage of the data records, if a data coverage condition is detected; and

instructions for generating a computational model indicative of interrelationships between a plurality of input parameters and the one or more output parameters based on the data records.

19. The computer readable medium of claim 18, wherein the instructions for evaluating coverage of the data records further include:

instructions for dividing the modeling space into a plurality of hyper-quadrants;

instructions for calculating a data density for each hyper-quadrant based on a number of data records in the respective hyper-quadrant;

instructions for generating a histogram based on the calculated data densities;

instructions for determining a statistical difference in a data density distribution based on the generated histogram; and

instructions for identifying the data coverage condition if the statistical difference exceeds a first threshold.

20. The computer readable medium of claim 18, wherein instructions for modifying coverage of the data records further include:

instructions for receiving additional data records from a user for sparse hyper-quadrants in the modeling space;

instructions for removing data records from dense hyper-quadrants in the modeling space;

or

instructions for creating additional data records for sparse hyper-quadrants in the modeling space via a symmetric random scatter process.