System and method for forecasting high-sellers using multivariate bayesian time series

Patent metadata

Field	Value
Publication number	US-10453026-B2
Application number	US-201514641075-A
Country	US
Kind code	B2
Filing date	Mar 6, 2015
Priority date	Mar 6, 2015
Publication date	Oct 22, 2019
Grant date	Oct 22, 2019

Abstract

A system and method for grouping units for forecasting purposes is presented. A sales forecast for a set of stock keeping units (SKUs) is desired. The SKUs are separated into clusters based on the similarity of the SKUs. Then a set of Bayesian multivariate dynamic linear models is chosen to be used to calculate a sales forecast for each of the clusters of SKUs. The accuracy of each dynamic linear model is determined in a training procedure and a set of weights for each dynamic linear model is calculated. Thereafter, the weights can be used with the dynamic linear models to create a weighted average forecast model. The training procedure can be run periodically to maintain the accuracy of the weights. Each procedure can operate on a sliding window of data. Other embodiments are also disclosed herein.

First claim

What is claimed is:

1. A method comprising:
receiving, from one or more non-transitory storage modules, a set of stock keeping units (SKUs);
creating, at one or more processing modules, a set of one or more clusters of SKUs from the set of SKUs;
storing, at the one or more non-transitory storage modules, the set of one or more clusters of SKUs;
choosing, at the one or more processing modules, a set of Bayesian multivariate dynamic linear models (DLMs) to be used to calculate a sales forecast for each SKU of at least a portion of the set of SKUs using the set of one or more clusters of SKUs, wherein the set of one or more clusters of SKUs comprises the at least the portion of the set of SKUs;
determining, at the one or more processing modules, a set of training weights of a conditional predictive distribution for each Bayesian multivariate DLM in the set of Bayesian multivariate DLMs with different values of variance loading parameters that determine relative magnitudes of local levels and seasonal coefficients in the Bayesian multivariate DLMs by:
obtaining, from the one or more non-transitory storage modules, first historical sales from a first historical time segment for the set of one or more clusters of SKUs;
forecasting, at the one or more processing modules, first forecasted sales for the first historical time segment for the set of one or more clusters of SKUs:
(1) using, in the set of Bayesian multivariate DLMs, first preceding sales of the set of one or more clusters of SKUs that preceded the first historical sales of the set of one or more clusters of SKUs; and
(2) without using, in the set of Bayesian multivariate DLMs, the first historical sales from the first historical time segment;

storing, at the one or more non-transitory storage modules, the first forecasted sales;
obtaining, from the one or more non-transitory storage modules, second historical sales from a second historical time segment for the set of one or more clusters of SKUs;
forecasting, using the one or more processing modules, second forecasted sales for the second historical time segment for the set of one or more clusters of SKUs:
(1) using, in the set of Bayesian multivariate DLMs, second preceding sales of the set of one or more clusters of SKUs that preceded the second historical sales of the set of one or more clusters of SKUs; and
(2) without using, in the set of Bayesian multivariate DLMs, the second historical sales from the second historical time segment;

storing, at the one or more non-transitory storage modules, the second forecasted sales; and
retrospectively regressing, at the one or more processing modules, the first historical sales and the second historical sales with the first forecasted sales and the second forecasted sales from each Bayesian multivariate DLM of the set of Bayesian multivariate DLMs by a time-series cross-validation;

storing, at the one or more non-transitory storage modules, the set of training weights;
using the set of training weights to calculate, at the one or more processing modules, the sales forecast for each SKU of at least the portion of the set of SKUs;
purchasing inventory for each SKU of the at least the portion of the set of SKUs based on the sales forecast for each SKU of the at least the portion of the set of SKUs;
receiving the inventory for each SKU of the at least the portion of the set of SKUs at a warehouse or a brick and mortar store;
recording sales data for the inventory for each SKU of the at least the portion of the set of SKUs; and
repeating a step of determining, at the one or more processing modules, the set of training weights of the conditional predictive distribution for each Bayesian multivariate DLM in the set of Bayesian multivariate DLMs with the different values of the variance loading parameters that determine the relative magnitudes of the local levels and the seasonal coefficients in the Bayesian multivariate DLMs using the sales data for the inventory for each SKU of the portion of the set of SKUs as the first historical sales or the second historical sales, wherein:
the set of Bayesian multivariate DLMs are specified as:

Y
t

=

μ
t

+

ψ
⁡

(
t
)

⁢

α
t

+

ϵ
t

⁢

⁢
where
⁢

⁢

ϵ
t

~

N
⁡

(

0
,

σ
2

⁢

I
n

)

⁢

(
observational
)

⁢

[

μ
t

α
t

]

=

[

μ

t
-
1

α

t
-
1

]

+

Assignees

• Wal Mart Stores Inc
• Walmart Apollo Llc

Inventors

• Ray Shubhankar
• Jha Abhay

Classifications

• G06Q10/087 Primary

  Inventory or stock management, e.g. order filling, procurement or balancing against orders · CPC title

• G06Q30/0202 Primary

  Market predictions or forecasting for commercial activities · CPC title

• G06Q10/0872

  using inventory planning · CPC title