Supply chain demand forecasting and planning

• US 20020169657A1
• Filed: 10/29/2001
• Published: 11/14/2002
• Est. Priority Date: 10/27/2000
• Status: Active Grant

• Current Assignee: JDA Software Group Incorporated (RedPrairie Holding Incorporated)
• Original Assignee: Manugistics Group Incorporated (RedPrairie Holding Incorporated)
• Inventors: Singh, Namita, Olasky, S. Jason, Welch, William F. Jr., Cluff, Kellie S.
• Granted Patent: US 7,080,026 B2
• Time in Patent Office: Days
• US Class Current: 705/10
• CPC Class Codes: G06Q 10/06 Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
• CPC Class Codes: G06Q 30/0202 Market predictions or forecasting for commercial activities

Abstract

Disclosed herein are systems and methods for demand forecasting that enable multiple-scenario comparisons and analyses by letting users create forecasts from multiple history streams (for example, shipments data, point-of-sale data, customer order data, return data, etc.) with various alternative forecast algorithm theories. The multiple model framework of the present invention enables users to compare statistical algorithms paired with various history streams (collectively referred to as “models”) so as to run various simulations and evaluate which model will provide the best forecast for a particular product in a given market. Once the user has decided upon which model it will use, it can publish forecast information provided by that model for use by its organization (such as by a downstream supply planning program). Embodiments of the present invention provide a system and method whereby appropriate demand responses can be dynamically forecasted whenever given events occur, such as when a competitor lowers the price on a particular product (such as for a promotion), or when the user'"'"'s company is launching new sales and marketing campaigns. Preferred embodiments of the present invention use an automatic tuning feature to assist users in determining optimal parameter settings for a given forecasting algorithm to produce the best possible forecasting model.

Claims

1. A method for forecasting future demand, said method comprising:

providing a plurality of demand forecasting algorithms and a plurality of demand data history streams, said history streams pertaining to a particular product at a location;

creating more than one model for the product at the location, each said model comprising a pairing of one of said demand forecasting algorithms and one of said demand data history streams;

calculating a forecast for each model by statistical regression of each model'"'"'s history stream using its paired model'"'"'s forecasting algorithm;

comparing said forecasts for said more than one models and identifying certain forecasts as finalized; and

publishing said finalized forecasts.

2. The method according to claim 1, wherein said step of calculating a forecast for each model comprises creating an initial forecast for each model such that said initial forecast has a minimum error when compared to its demand data history stream and tuning said initial forecast to make it feasible.

3. The method according to claim 2, wherein said step of tuning said initial forecast comprises balancing a plurality of feasibility factors selected from the group consisting of smoothness of a dynamic mean, stability of a seasonal profile, distribution of positive and negative errors, forecast explosion, and continuity of error of the initial forecast during recent history.

4. The method according to claim 1, wherein each forecast is identified according to a demand forecasting unit, and wherein said demand forecasting unit comprises a demand unit, a demand group, a location, and a model.

5. The method according to claim 1, further comprising adjusting said published finalized forecasts according to a ratio of actual demand data over demand predicted by said forecasts.

6. The method according to claim 5, wherein said ratio is multiplied by a remainder of the predicted demand of the published finalized forecast to produce an adjusted published forecast.

7. The method according to claim 6, wherein more than one models created allows a user to compare model variant forecasts, location variant forecasts, and forecasts for similar products.

8. The method according to claim 6, wherein calculating a forecast comprising identifying a significant number of forecast terms and preparing forecast using said significant number of forecast terms.

9. A system for creating forecasts of future demand, said system comprising:

means for providing a plurality of demand forecasting algorithms and a plurality of demand data history streams, said history streams pertaining to a particular product at a location;

means for creating more than one model for the product at the location, each said model comprising a pairing of one of said demand forecasting algorithms and one of said demand data history streams;

means for calculating a forecast for each model by statistical regression of each model'"'"'s history stream using its paired model'"'"'s forecasting algorithm;

means for comparing said forecasts for said more than one models and identifying certain forecasts as finalized; and

means for publishing said finalized forecasts.

10. A system according to claim 9, wherein said means for calculating a forecast is adapted to select history data related to said history stream, and perform a least-squares regression on said history data.

11. A method for forecasting future demand, said method comprising the steps of:

setting up a database and defining Demand Forecasting Units (DFUs), each of said DFUs pertaining to a particular product at a particular location;

preparing a forecast for each of the DFUs;

reviewing said prepared forecasts; and

for one or more of said DFUs, publishing said reviewed forecasts associated with said DFUs.

12. The method of claim 11 further comprising the steps of:

adjusting one or more of said published forecasts; and

republishing said adjusted forecasts.

13. The method of claim 11, wherein said forecast preparing step further comprises:

identifying active DFUs, and for each of said active DFUs;

preparing a history stream;

associating a demand forecasting algorithm to said history stream; and

using said history stream and said demand forecasting algorithm to generate an active DFU forecast.

14. The method of claim 13, wherein said forecast preparing step further comprises fine-tuning said history stream and said demand forecasting algorithm.

15. The method of claim 11, wherein said forecast preparing step further comprises adjusting parameters used to form said DFU forecast.

16. The method of claim 13, wherein said forecast preparing step further comprises finalizing said DFU forecasts for each of said active DFUs.

17. The method of claim 16, wherein said forecast preparing step further comprises closing a forecast period for each of said DFU forecasts.

18. The method of claim 13, wherein said step of using said history stream and said demand forecasting algorithm to generating an active DFU forecast further comprises:

selecting history data related to said history stream, and

performing a least-squares regression on said history data.

19. The method of claim 18, wherein said step of selecting history data related to said history stream comprises identifying a maximum number of terms in said history stream, and wherein the step of performing a least-squares regression comprises using only the maximum number of terms.

20. The method of claim 19, wherein said step of using said history stream and said demand forecasting algorithm to generating an active DFU forecast further comprises testing of said terms for significant amplitude.

21. The method of claim 20, wherein said step of using said history stream and said demand forecasting algorithm to generating an active DFU forecast further comprises using only said terms having significant amplitude greater than a preset minimum.