US-12626266-B1 - Detecting and reacting to unseen long term event in demand forecasting

Abstract

A system and method are disclosed for detecting and reacting to unseen events in demand forecasting, comprising preparing, by a server, data from a supply chain domain and entity to predict a demand, selecting features in the prepared data and training a machine learning model, generating a demand prediction with residual time series corrections, using the machine learning model, monitoring the demand prediction and data from the supply chain domain and entity to detect an occurrence of an unseen event; and in response to detecting the occurrence of the unseen event, revising the demand prediction by updating the machine learning model. The system and method further comprises detecting that a greater than threshold increase has occurred in a prediction of a target variable, and revising the demand prediction further comprises performing a second residual time series corrections that incorporates time information associated with the unseen event.

Inventors

• Felix Christopher Wick
• Sunny Kumar
• Trapti Singhal

Assignees

• Blue Yonder Group, Inc.

Dates

Publication Date

20260512

Application Date

20220608

Claims (17)

1. 1 . A system comprising a computer, the computer comprising a server and configured to: prepare data received from a supply chain domain and a supply chain entity to predict a demand associated with the supply chain domain and the supply chain entity; select features in the prepared data and train a machine learning model associated with the supply chain domain and the supply chain entity; generate a demand prediction with a first residual time series corrections, using the machine learning model associated with the supply chain domain and the supply chain entity, wherein the demand prediction further comprises a list of individual causal factors with a corresponding relative strength of the individual causal factors in the demand prediction; automatically monitor the demand prediction and data from the supply chain domain and the supply chain entity to detect an occurrence of an unseen event; in response to detecting the occurrence of the unseen event, automatically revise the demand prediction by updating the machine learning model for a duration of the event, wherein revising the demand prediction further comprises performing a second residual time series corrections using the machine learning model, wherein the second residual time series corrections incorporates time information associated with the unseen event; and initiate manufacturing of one or more components based, at least in part on, the revised demand prediction.
2. 2 . The system of claim 1 , wherein detecting the occurrence of the unseen event further comprises detecting that a greater than threshold increase has occurred in a prediction of a target variable of the machine learning model.
3. 3 . The system of claim 2 , wherein the revised demand prediction further comprises the first residual time series correction and the second residual time series correction, the second residual time series correction being applied only during a time span of the unseen event.
4. 4 . The system of claim 1 , wherein the unseen event comprises an unseen event for which the machine learning model has not previously been trained to predict demand.
5. 5 . The system of claim 1 , wherein the selected features are one or more selected from a group consisting of: product price, product family, product location, weather, price elasticity, a local event, a global event, a holiday and a week of a year.
6. 6 . The system of claim 1 , wherein the server is further configured to prepare the data received by performing at least one of: normalizing the data, dropping or deleting null values, dropping or deleting corrupted values, or dropping or deleting blank values.
7. 7 . A computer-implemented method, comprising preparing, by a server, data received from a supply chain domain and a supply chain entity to predict a demand associated with the supply chain domain and the supply chain entity; selecting, by the server, features in the prepared data and training a machine learning model associated with the supply chain domain and the supply chain entity; generating, by the server, a demand prediction with a first residual time series corrections, using the machine learning model associated with the supply chain domain and the supply chain entity, wherein the demand prediction further comprises a list of individual causal factors with a corresponding relative strength of the individual causal factors in the demand prediction; automatically monitoring, by the server, the demand prediction and data from the supply chain domain and the supply chain entity to detect an occurrence of an unseen event; in response to detecting the occurrence of the unseen event, automatically revising, by the server, the demand prediction by updating the machine learning model for a duration of the event, wherein revising the demand prediction further comprises performing, by the server, a second residual time series corrections using the machine learning model, wherein the second residual time series corrections incorporates time information associated with the unseen event; and initiating, by the server, manufacturing of one or more components based, at least in part on, the revised demand prediction.
8. 8 . The computer-implemented method of claim 7 , wherein detecting the occurrence of the unseen event further comprises detecting, by the server, that a greater than threshold increase has occurred in a prediction of a target variable of the machine learning model.
9. 9 . The computer-implemented method of claim 8 , wherein the revised demand prediction further comprises the first residual time series correction and the second residual time series correction, the second residual time series correction being applied only during a time span of the unseen event.
10. 10 . The computer-implemented method of claim 7 , wherein the unseen event comprises an unseen event for which the machine learning model has not previously been trained to predict demand.
11. 11 . The computer-implemented method of claim 7 , wherein the selected features are one or more selected from a group consisting of: product price, product family, product location, weather, price elasticity, a local event, a global event, a holiday and a week of a year.
12. 12 . The computer-implemented method of claim 7 , further comprising preparing, by the server, the data received by performing at least one of: normalizing the data, dropping or deleting null values, dropping or deleting corrupted values, or dropping or deleting blank values.
13. 13 . A non-transitory computer-readable storage medium embodied with software, the software when executed: prepares, by a server, data received from a supply chain domain and a supply chain entity to predict a demand associated with the supply chain domain and the supply chain entity; selects features in the prepared data and trains a machine learning model associated with the supply chain domain and the supply chain entity; generates a demand prediction with a first residual time series corrections, using the machine learning model associated with the supply chain domain and the supply chain entity, wherein the demand prediction further comprises a list of individual causal factors with a corresponding relative strength of the individual causal factors in the demand prediction; automatically monitors the demand prediction and data from the supply chain domain and the supply chain entity to detect an occurrence of an unseen event; in response to detecting the occurrence of the unseen event, automatically revises the demand prediction by updating the machine learning model for a duration of the event, wherein the software when executed revises the demand prediction further by performing a second residual time series corrections using the machine learning model, wherein the second residual time series corrections incorporates time information associated with the unseen event; and initiates manufacturing of one or more components based, at least in part on, the revised demand prediction.
14. 14 . The non-transitory computer-readable storage medium of claim 13 , wherein the software when executed detects the occurrence of the unseen event further by detecting that a greater than threshold increase has occurred in a prediction of a target variable of the machine learning model.
15. 15 . The non-transitory computer-readable storage medium of claim 14 , wherein the revised demand prediction further comprises the first residual time series correction and the second residual time series correction, the second residual time series correction being applied only during a time span of the unseen event.
16. 16 . The non-transitory computer-readable storage medium of claim 13 , wherein the unseen event comprises an unseen event for which the machine learning model has not previously been trained to predict demand.
17. 17 . The non-transitory computer-readable storage medium of claim 13 , wherein the selected features are one or more selected from a group consisting of: product price, product family, product location, weather, price elasticity, a local event, a global event, a holiday and a week of a year.

Description

CROSS REFERENCE TO RELATED APPLICATIONS The present disclosure is related to that disclosed in the U.S. Provisional Application No. 63/212,911, filed Jun. 21, 2021, entitled “Detecting and Reacting to Unseen Long Term Event in Demand Forecasting,” and U.S. Provisional Application No. 63/208,635, filed Jun. 9, 2021, entitled “Automated Supply Chain Demand Forecasting Pipeline.” U.S. Provisional Application Nos. 63/212,911 and 63/208,635 are assigned to the assignee of the present application. The present invention hereby claims priority under 35 U.S.C. § 119 (e) to U.S. Provisional Application Nos. 63/212,911 and 63/208,635. TECHNICAL FIELD The present disclosure relates generally to detecting and reacting to unseen, long-term events and, in particular, detecting and reacting to unseen, long-term events in supply chain demand forecasting utilizing machine learning models. BACKGROUND Events may comprise important features in supply chain demand forecasting and play important roles in supply chain model predictions and the training of machine learning models. For a particular product, events may be positive, such as sales increasing due to an exterior factor, or negative, such as sales decreasing due to, for example, extreme weather. Long-term events may last and affect the sales of one or more products for weeks or months at a time, whereas short-term events may only last for a day or even less. Events may also be classified as seen or unseen, wherein seen events have been seen earlier in history, but unseen events have not been seen and are detected only in model prediction scenarios, in which one or more machine learning models are asked to make demand predictions without having previously been trained to model and predict demand associated with an unseen event. Long-term, unseen events, such as, for example, the sudden arrival of the COVID-19 pandemic, may be harder to detect due to unknown characteristics and are therefore difficult to respond to, which is undesirable. BRIEF DESCRIPTION OF THE DRAWINGS A more complete understanding of the present invention may be derived by referring to the detailed description when considered in connection with the following illustrative figures. In the figures, like reference numbers refer to like elements or acts throughout the figures. FIG. 1 illustrates a supply chain network, according to a first embodiment; FIG. 2 illustrates the forecasting system, the archiving system, and the planning and execution system of FIG. 1 in greater detail, according to an embodiment; FIG. 3 illustrates demand prediction method, according to an embodiment; and FIG. 4 illustrates an unseen event method, according to an embodiment. DETAILED DESCRIPTION Aspects and applications of the invention presented herein are described below in the drawings and detailed description of the invention. Unless specifically noted, it is intended that the words and phrases in the specification and the claims be given their plain, ordinary, and accustomed meaning to those of ordinary skill in the applicable arts. In the following description, and for the purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the various aspects of the invention. It will be understood, however, by those skilled in the relevant arts, that the present invention may be practiced without these specific details. In other instances, known structures and devices are shown or discussed more generally in order to avoid obscuring the invention. In many cases, a description of the operation is sufficient to enable one to implement the various forms of the invention, particularly when the operation is to be implemented in software. It should be noted that there are many different and alternative configurations, devices and technologies to which the disclosed inventions may be applied. The full scope of the inventions is not limited to the examples that are described below. As described in more detail below, embodiments of the following disclosure provide a demand forecasting system and method that detects starting and ending conditions for long-term unseen events and modifies demand predictions to incorporate the effects of the long-term unseen events into the accuracy of the demand predictions for the duration of the unseen event. Embodiments provide one or more machine learning models which may utilize one or more causal factors X to predict a volume Y (target or label). Having predicted a volume, the one or more machine learning models may then apply one or more residual time series corrections that incorporate differences between previous target and causal predictions time series data to correct, update, or modify the predicted volume to improve forecasting accuracy. Embodiments contemplate automatically detecting the starting and ending conditions for the unseen event and performing residual correction actions during the duration of the unseen event to increase the accuracy