Method for forecasting using a genetic algorithm

• US 7,213,007 B2
• Filed: 12/24/2002
• Issued: 05/01/2007
• Est. Priority Date: 12/24/2002
• Status: Expired due to Fees

• Current Assignee: Caterpillar Incorporated, DSM B.V.
• Original Assignee: Caterpillar Incorporated
• Inventors: Grichnik, Anthony J.
• Primary Examiner(s): Knight; Anthony
• Assistant Examiner(s): Holmes; Michael B.
• Application Number:
  US10/327,676
• Publication Number:
  US 20040139041A1
• Time in Patent Office: 1,589 Days
• Field of Search: 706/21
• US Class Current: 706/21
• CPC Class Codes: G06N 3/126 Evolutionary algorithms, e.g. genetic algorithms or genetic programming

Abstract

A method and system is provided to perform a process for forecasting the characteristics of a target item based on historical characteristic data. The process includes determining one or more equations that when solved present a representation of forecasted characteristics of the target item, wherein each of the one or more equations includes one or more variables. Further, the process involves determining a data value for each of the one or more variables that best represents the characteristics of the target item and solving the one or more equations using the determined data values to forecast the characteristics of the target item.

Claims

1. A computer-implemented method of forecasting characteristics of a target item, comprising:

determining one or more equations associated with forecasted characteristics of the target item, wherein each of the one or more equations includes one or more variables;

determining a data value for each of the one or more variables that represents the characteristics of the target item more so than other data values;

forecasting the characteristics of the target item by solving the one or more equations using the determined data values and a time period data value corresponding to a future time interval; and

providing the forecasted data,

wherein determining the data value includes performing a process that implements a genetic algorithm that generates a plurality of chromosomes corresponding to the data values for the one or more variables.

2. The method of claim 1, wherein the genetic algorithm determines a fitness value for each of the plurality of chromosomes in a population table, wherein the fitness value is derived from a goal function that is based on an average error and maximum error associated with a relationship between historical data and calculated data corresponding to the characteristics of the target item over a period of time.

3. The method of claim 2, wherein determining a data value for each of the one or more variables includes:

identifying a chromosome with a fitness value that is converging toward a target value.

4. The method of claim 1, wherein determining one or more equations includes:

determining boundary values for each of the one or more variables.

5. The method of claim 4, wherein the boundary values are based on at least one of a minimum and maximum value of monitored characteristics of the target item over a previous period of time, a time interval when the minimum value was monitored, and a time interval when the maximum value was monitored.

6. The method of claim 4, wherein the one or more equations includes a sinusoidal equation including a gain, rate, and phase variable.

7. The method of claim 6, wherein the boundary values for the gain variable is (−

Δ

P*10) and (Δ

P*10), wherein Δ

P is the difference between a minimum and maximum value of monitored characteristics of the target item over a previous period of time.

8. The method of claim 6, wherein the boundary values for the rate variable is −

Δ

t and Δ

t, wherein Δ

t is the difference between a time interval corresponding to a maximum value of monitored characteristics of the target item and a time interval corresponding to a minimum value of monitored characteristics of the target item took place.

9. The method of claim 6, wherein the boundary values for the phase variable is 0 and (10*Δ

t*Π

/2), wherein Δ

t is the difference between a time interval corresponding to a maximum value of monitored characteristics of the target item and a time interval corresponding to a minimum value of monitored characteristics of the target item.

10. The method of claim 4, wherein the one or more equations includes a linear equation including a slope and y-intercept variable.

11. The method of claim 10, wherein the boundary values for the slope variable is (Δ

P/−

Δ

t/2) and (Δ

P/Δ

t/2), wherein At is the difference between a time interval corresponding to a maximum value of monitored characteristics of the target item and a time interval corresponding to a minimum value of monitored characteristics of the target item and Δ

P is the difference between a minimum and maximum value of monitored characteristics of the target item over a previous period of time.

12. The method of claim 10, wherein the boundary values for the y-intercept variable is (−

10*min) and (10*max), wherein mm and max are the minimum and maximum values, respectively, of monitored characteristics of the target item over a previous period of time.

13. The method of claim 1, wherein the target item is a tangible product and the characteristics include at least one of a number of tangible products manufactured, delivered, shipped, bought, sold, and ordered.

14. The method of claim 2, wherein the goal function is e_avg* (e_max)ⁿ wherein e_avg is an average error associated with a calculated characteristic data value and an historical characteristic data value for a corresponding chromosome, e_max is a maximum error value associated with the actual characteristic value and the calculated characteristic value at some interval within the period of time, and n is a positive integer value.

15. The method of claim 1, wherein determining a data value includes:

collecting a set of historical data values corresponding to characteristics of the target item for each of a plurality of intervals within a period of time;

determining, for each interval, a calculated data value corresponding to calculated characteristics of the target item;

determining, for each interval, an error value corresponding to a difference between the calculated and actual data values; and

determining an average error value based on the determined error values.

16. The method of claim 15, wherein determining a calculated data value includes:

solving, for each interval, the one or more equations using randomly selected data values for each of the corresponding one or more variables.

17. The method of claim 16, wherein the average error is used to determine a fitness value for each of a plurality of chromosomes in a population table.

18. The method of claim 1, further including:

determining a confidence factor reflecting a level of confidence in the forecasted value.

19. The method of claim 18, further including:

determining an error factor that reflects a range of characteristic data values associated with the forecasted characteristics.

20. A system for forecasting characteristics of a target item, comprising:

a database including a collection of historical data corresponding to monitored characteristics of the target item over a previous period of time; and

a forecasting computer system for executing instructions stored on a memory device that performs a forecasting process that determines predicted characteristics of the target item for a future point in time, wherein the forecasting process includes, when executed by the forecasting computer system;

determining one or more equations associated with forecasted characteristics of the target item, wherein each of the one or more equations includes one or more variables;

determining a data value for each of the one or more variables that represents the characteristics of the target item more so than other data values;

forecasting the characteristics of the target item by solving the one or more equations using the determined data values and a time period data value corresponding to a future time interval; and

providing the forecasted data.

21. The system of claim 20, wherein determining a data value includes:

retrieving, from the database, a set of historical data values corresponding to characteristics of the target item for each of a plurality of intervals within a selected period of time;

determining, for each interval, a calculated data value corresponding to calculated characteristics of the target item;

determining, for each interval, an error value corresponding to the difference between the calculated and historical data values; and

determining an average error value based on the determined error values.

22. The system of claim 21, wherein determining a calculated data value includes:

solving, for each interval, the one or more equations using randomly selected data values for each of the corresponding one or more variables.

23. The system of claim 22, wherein the average error is used to determine a fitness value for each of a plurality of chromosomes in a population table.

24. A computer-readable medium including instructions for performing a method, when executed by a processor, for forecasting characteristics of a target item, the method comprising:

determining one or more equations associated with forecasted characteristics of the target item, wherein each of the one or more equations includes one or more variables;

determining a data value for each of the one or more variables that represents the characteristics of the target item more so than other data values;

forecasting the characteristics of the target item by solving the one or more equations using the determined data values and a time period data value corresponding to a future time interval; and

providing the forecasted data.

25. The computer-readable medium of claim 24, wherein determining a data value for each of the one or more variables includes:

performing a process that implements a genetic algorithm that generates a plurality of chromosome that includes the best data values for the one or more variables.

26. The computer-readable medium of claim 25, wherein the genetic algorithm determines a fitness value for each of a plurality of chromosomes in a population table, wherein the fitness value is derived from a goal function that is based on an average error and maximum error associated with a relationship between historical data and calculated data corresponding to the characteristics of the target item over a period of time.

27. The computer-readable medium of claim 26, wherein determining a data value for each of the one or more variables includes:

identifying a chromosome with a fitness value that is converging toward an optimum value.

28. The computer-readable medium of claim 24, wherein determining one or more equations includes:

determining boundary values for each of the one or more variables.

29. The computer-readable medium of claim 28, wherein the boundary values are based on at least one of a minimum and maximum value of monitored characteristics of the target item over a previous period of time, a time interval when the minimum value was monitored, and a time interval when the maximum value was monitored.

30. The computer-readable medium of claim 24, wherein the method further includes:

determining a confidence factor reflecting a level of confidence in the forecasted value.

31. The computer-readable medium of claim 30, wherein determining a confidence factor includes:

determining an error factor that reflects a range of characteristic data values associated with the forecasted characteristics.