Demand classification based pipeline system for time-series data forecasting

Patent metadata

Field	Value
Publication number	US-10685283-B2
Application number	US-201916726616-A
Country	US
Kind code	B2
Filing date	Dec 24, 2019
Priority date	Jun 26, 2018
Publication date	Jun 16, 2020
Grant date	Jun 16, 2020

Abstract

A pipeline system for time-series data forecasting using a distributed computing environment is disclosed herein. In one example, a pipeline for forecasting time series is generated. The pipeline represents a sequence of operations for processing the time series to produce modeling results such as forecasts of the time series. The pipeline includes a segmentation operation for categorizing the time series into multiple demand classes based on demand characteristics of the time series. The pipeline also includes multiple sub-pipelines corresponding to the multiple demand classes. Each of the sub-pipelines applies a model strategy to the time series in the corresponding demand class. The model strategy is selected from multiple candidate model strategies based on predetermined relationships between the demand classes and the candidate model strategies. The pipeline is executed to determine the modeling results for the time series.

First claim

The invention claimed is:

1. A system comprising:
a processing device; and
a memory device comprising instructions that are executable by the processing device for causing the processing device to:
access a pipeline for forecasting a plurality of time series, the pipeline representing a sequence of operations for processing the plurality of time series and generating forecasts for the plurality of time series, wherein the pipeline includes:
(i) a segmentation operation for categorizing the plurality of time series into a plurality of demand classes, each time series in the plurality of time series being categorized into a particular demand class among the plurality of demand classes based on demand characteristics of the time series;
(ii) a plurality of sub-pipelines corresponding to the plurality of demand classes, each sub-pipeline among the plurality of sub-pipelines defining respective sub-operations to be performed on the time series in a respective demand class among the plurality of demand classes, the respective sub-operations comprising:
a model strategy operation configured for applying a model strategy to the time series in the respective demand class to determine modeling results for the respective demand class, wherein the model strategy is selected from among a plurality of candidate model strategies based on predetermined relationships between the plurality of demand classes and the plurality of candidate model strategies, wherein the modeling results comprise forecasts generated by applying a plurality of forecasting models specified in the model strategy to the time series in the respective demand class, and wherein a first forecasting model applied to a first time series in the demand class is different from a second forecasting model applied to a second time series in the demand class; and

(iii) a merge operation for combining the modeling results from the plurality of sub-pipelines; and

execute the pipeline to determine the modeling results for the plurality of time series.

2. The system of claim 1 , wherein the memory device further includes instructions that are executable by the processing device for causing the processing device to:
include a graphical visualization of the pipeline in a graphical user interface, the graphical visualization depicting the sequence of operations spatially organized in an order in which sequence of operations is executed in the pipeline, wherein the graphical visualization of the pipeline comprises a visual representation of a modeling operation representing the plurality of sub-pipelines;
detect an interaction with the visual representation of the modeling operation; and
in response to detecting the interaction, cause a second graphical user interface to be displayed for showing information about the plurality of demand classes.

3. The system of claim 2 , wherein the second graphical user interface is configured to display a table view showing the information about the plurality of demand classes, and wherein the memory device further includes instructions that are executable by the processing device for causing the processing device to:
receive a query comprising one or more filter criteria; and
in response to receiving the query:
filter the plurality of demand classes based on the one or more filter criteria to identify a filtered set of demand classes; and
display information about the filtered set of demand classes in the table view.

4. The system of claim 3 , wherein the table view showing the information about the plurality of demand classes comprises an indication of a number of time series in each demand class among the plurality of demand classes.

5. The system of claim 3 , wherein the information about the plurality of demand classes comprises an indication of a respective model strategy used for each demand class among the plurality of demand classes and an aggregate error value associated with the respective model strategy.

6. The system of claim 2 , wherein the second graphical user interface comprises user interface controls associated with the plurality of demand classes, and wherein the memory device further includes instructions that are executable by the processing device for causing the processing device to:
detect a selection of a user interface control; and
in response to detecting the selection of the user interface control, display a graphical user interface showing a graphical visualization of a sub-pipeline among the plurality of sub-pipelines corresponding to the user interface control, the graphical visualization of the sub-pipeline including visual representations of the respective sub-operations in the sub-pipeline that are spatially organized in an order in which the respective sub-operations are executed in the sub-pipeline.

7. The system of claim 2 , wherein the graphical visualization of the pipeline comprises a visual representation of the segmentation operation, and wherein the memory device further includes instructions that are executable by the processing device for causing the processing device to:
detect an interaction with the visual representation of the segmentation operation; and
in response to detecting the interaction with the visual representation of the segmentation operation, cause a third graphical user interface to be displayed for showing a profile of the plurality of demand classes.

8. The system of claim 1 , wherein the plurality of demand classes include a retired class, a short class, a low-volume class, a year-round-seasonal class, a year-round-non-seasonal class, a year-round intermittent class, a year-round-seasonal-intermittent class, an in-season-intermittent class, or an in-season-non-intermittent class.

9. The system of claim 1 , wherein applying the model strategy to the time series in the respective demand class comprises:
determining a default model strategy corresponding to the respective demand class based on predefined relationships between default model strategies and demand classes, the default model strategy specifying one or more forecasting models;
applying the one or more forecasting models in the default model strategy to the time series in the respective demand class; and
comparing results of the one or more forecasting models to determine one or more champion forecasts for the time series in the respective demand class.

10. The system of claim 1 , wherein the plurality of candidate model strategies comprise a non-seasonal model, retired series, a seasonal model, a time series regression model, a temporal aggregation model, a panel series neural network, a multi-stage model, or a stacked model.

11. The system of claim 10 , wherein generating the panel series neural network involves training a neural network model based on default attributes assigned to a set of time series, and wherein the stacked model includes a neural network model configured to generate and feed a first forecast into a time series model, the time series model configured to generate a second forecast based on the first forecast.

12. A method comprising:
access a pipeline for forecasting a plurality of time series, the pipeline representing a sequence of operations for processing the plurality of time series and generating forecasts for the plurality of time series, wherein the pipeline includes:
(i) a segmentation operation for categorizing the plurality of time series into a plurality of demand classes, each time series in the plurality of time series being categorized into a particular demand class among the plurality of demand classes based on demand characteristics of the time series;
(ii) a plurality of sub-pipelines corresponding to the plurality

Assignees

• Sas Inst Inc

Inventors

• Li Yue
• Trovero Michele Angelo
• Helmkamp Phillip Mark
• Brzezicki Jerzy Michal
• Frazier Macklin Carter
• Haley Timothy Patrick
• Solomonson Randy Thomas
• Kim Sangmin
• Mills Steven Christopher
• Chien Yung-Hsin
• Hodgin Ron Travis
• Xie Jingrui

Classifications

• G06N3/045

  Combinations of networks · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/0499

  Feedforward networks · CPC title

• G06N3/08 Primary

  Learning methods · CPC title

• H04L43/10

  Active monitoring, e.g. heartbeat, ping or trace-route · CPC title