Forecast model calibration

• US 10,657,547 B2
• Filed: 10/07/2016
• Issued: 05/19/2020
• Est. Priority Date: 03/28/2016
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Singhee, Amith
• Primary Examiner(s): Jeanty, Romain
• Application Number:
  US15/287,873
• Publication Number:
  US 20170278116A1
• Time in Patent Office: 1,320 Days
• Field of Search: 705 723
• CPC Class Codes: G06Q 10/06315 Needs-based resource requirements planning or analysis
• CPC Class Codes: G06Q 30/0202 Market predictions or forecasting for commercial activities

Abstract

A system and method perform calibration of a forecast model for resource allocation. The method includes receiving inputs to the forecast model derived from historical data for a period of time, and executing the forecast model to obtain one or more forecast levels for each interval within the period of time, the forecast level corresponding with a quantified forecast of a forecast parameter that is forecast by the forecast model for the interval. Obtaining an actual level for each interval within the period of time according to the historical data is followed by comparing the one or more forecast levels with the actual level for the period of time according to a metric to adjust a mapping within the forecast model between values of the quantified forecast and the forecast levels based on the comparing to obtain a calibrated forecast model. The calibrated forecast model is used for resource allocation.

Claims

1. A computer-implemented method of performing calibration of a forecast model for resource allocation, the method comprising:

receiving inputs to the forecast model derived from historical data for a period of time;

executing, using a processor, the forecast model to obtain one or more forecast levels for each interval within the period of time, wherein the forecast level corresponds with a quantified forecast of a forecast parameter that is forecast by the forecast model for the interval;

obtaining, using the processor, an actual level for each interval within the period of time according to the historical data;

comparing, using the processor, the one or more forecast levels with the actual level for the period of time according to a metric;

adjusting, using the processor, a mapping within the forecast model between values of the quantified forecast and the forecast levels based on the comparing to obtain a calibrated forecast model; and

deploying resources, using the processor, based on calibrated forecast levels output by the calibrated forecast model, wherein the deploying resources includes deploying equipment and personnel, scheduling maintenance, ordering retail inventory, allocating compute resources, or prioritizing activities.

2. The method according to claim 1, wherein obtaining the one or more forecast levels includes obtaining a probability associated with each forecast level.

3. The method according to claim 2, wherein the comparing the one or more forecast levels with the actual level includes computing a quantity based on the probability associated with each forecast level.

4. The method according to claim 3, further comprising determining the metric based on the quantity.

5. The method according to claim 1, wherein the obtaining the actual level includes determining the actual value that corresponds with an actual value of the forecast parameter.

6. The method according to claim 1, wherein the adjusting the mapping includes determining a lower bound of a range of values of the quantified forecast that correspond with each forecast level.

7. The method according to claim 6, wherein the determining the range of values is based on an exhaustive sampling, a random sampling, or a deterministic sampling.

8. The method according to claim 6, wherein the determining the range of values is based on a binary search, a gradient descent search, quadratic programming, heuristic optimization, or sequential quadratic programming.

9. A system to calibrate a forecast model for resource allocation, the system comprising:

a memory device configured to store historical data; and

a processor configured to receive inputs to the forecast model derived from the historical data for a period of time, execute the forecast model to obtain a forecast level for each interval with in the period of time, obtain an actual level for each interval within the period of time according to the historical data, compare the one or more forecast levels with the actual level according to a metric, and adjust a mapping within the forecast model between the quantified forecast and the forecast levels to improve a result of the comparing and obtain a calibrated forecast model, wherein the calibrated forecast model is used to deploy equipment and personnel, schedule maintenance, order retail inventory, allocate compute resource, or prioritize activities.

10. The system according to claim 9, wherein the processor obtains the one or more forecast levels based on obtaining a probability associated with each forecast level.

11. The system according to claim 10, wherein the processor compares the one or more forecast levels with the actual level based on computing a quantity based on the probability associated with each forecast level and determining the metric based on the quantity.

12. The system according to claim 9, wherein the processor adjusts the mapping based on determining a lower bound of a range of values of the quantified forecast that correspond with each forecast level.

13. The system according to claim 12, wherein the processor determines the range of values based on an exhaustive sampling, a random sampling, a deterministic sampling, a binary search, a gradient descent search, quadratic programming, heuristic optimization, or sequential quadratic programming.

14. A computer program product for calibrating a forecast model to perform resource allocation, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor to perform a method comprising:

receiving inputs to the forecast model derived from historical data for a period of time;

executing the forecast model to obtain one or more forecast levels for each interval within the period of time, wherein the forecast level corresponds with a quantified forecast of a forecast parameter that is forecast by the forecast model for the interval;

obtaining an actual level for each interval within the period of time according to the historical data;

comparing the one or more forecast levels with the actual level for the period of time according to a metric;

adjusting a mapping within the forecast model between values of the quantified forecast and the forecast levels based on the comparing to obtain a calibrated forecast model; and

deploying resources based on calibrated forecast levels output by the calibrated forecast model, wherein the deploying resources includes deploying equipment and personnel, scheduling maintenance, ordering retail inventory, allocating compute resources, or prioritizing activities.

15. The computer program product according to claim 14, wherein obtaining the one or more forecast levels includes obtaining a probability associated with each forecast level.

16. The computer program product according to claim 15, wherein the comparing the one or more forecast levels with the actual level includes computing a quantity based on the probability associated with each forecast level.

17. The computer program product according to claim 16, further comprising determining the metric based on the quantity.

18. The computer program product according to claim 14, wherein the obtaining the actual level includes determining the actual value that corresponds with an actual value of the forecast parameter.

19. The computer program product according to claim 14, wherein the adjusting the mapping includes determining a lower bound of a range of values of the quantified forecast that correspond with each forecast level.

20. The computer program product according to claim 19, wherein the determining the range of values is based on an exhaustive sampling, a random sampling, a deterministic sampling, a binary search, a gradient descent search, quadratic programming, heuristic optimization, or sequential quadratic programming.