Granted Patent B2

US 12,675,337 · App. 18/263,282 · Granted Jul 7, 2026

Inventors: Jieqiu Chen (Redmond, WA); Harsh Mahendra Mehta (Sammamish, WA); Boon Pin Yeap (Redmond, WA); Garet Marshall Anderson (Mount Vernon, WA); Dale Kenji Miyakusu (Campbell, CA); Vishnupriya Appiah Suresh Bapu (Sammamish, WA)

Assignee: Microsoft Technology Licensing, LLC

G06F9/5077G06F9/5072G06F2209/5019G06F2209/5022G06F2209/508

Abstract

Claims (55)

1 . A computer-implemented method for configuring a cloud service center comprising:

accessing usage data of resources of the cloud service center;

standardizing the usage data by applying a pre-processing operation to the usage data;

generating a plurality of forecast models based on the standardized usage data, the plurality of forecast models being configured to predict a demand of the resources of the cloud service center;

selecting a demand forecast model from the plurality of forecast models based on a ranking of the plurality of forecast models;

applying a post-processing operation to a demand forecast that is generated based on the selected demand forecast model, by, when the demand forecast is unusable, creating a new forecast by perturbing a last good forecast so that it is smoothly aligned with a latest actual of a time series of the usage data; and

configuring the cloud service center based on the post-processed demand forecast.

2 . The computer-implemented method of claim 1 , further comprising:

detecting that a current demand forecast deviates from a previous demand forecast;

detecting noise in recent usage data; and

in response to detecting that the current demand forecast deviates from the previous demand forecast model, and detecting the noise, shifting the previous demand forecast to align with the recent usage data.

3 . The computer-implemented method of claim 2 , further comprising:

determining a time-to-live metric based on the shifted previous forecast model and a capacity of the cloud service center, the time-to-live metric indicating a predicted time at which demands for the resources of the cloud service center exceed the resources of the cloud service center; and

detecting that the time-to-live metric transgresses a capacity threshold of the cloud service center,

wherein configuring the cloud service center is in response to the time-to-live metric transgressing the capacity threshold.

4 . The computer-implemented method of claim 1 , further comprising:

generating a graph in a graphical user interface that illustrates the demand forecast model and a time-to-live metric; and

generating a user-interactive graphical user interface element that enables an administrator of the cloud service center to allocate additional resources to the cloud service center relative to the time-to-live metric.

5 . The computer-implemented method of claim 4 , wherein the graphical user interface indicates a date based on the time-to-live metric.

6 . The computer-implemented method of claim 1 , wherein the pre-processing operation comprises at least one of a time-series standardization operation, a missing data imputation operation, an outlier detection and imputation operation, a seasonal decomposition operation, or a mathematical transformation operation.

7 . The computer-implemented method of claim 1 , wherein generating the plurality of forecast models is based on time series of the usage data, wherein the plurality of forecast models are based on one of a statistical method, and a machine learning method.

8 . The computer-implemented method of claim 1 , wherein selecting the demand forecast model is based on at least one of a performance metrics calculation, an ensemble machine learning technique, a backtesting, a model selection, or a forecast generation.

9 . The computer-implemented method of claim 1 , further comprising:

determining a set of base forecast models;

applying a backtesting procedure for each forecast model;

ensembling two base forecast models to generate a new combined forecast; and

ranking the base forecast models and the new combined forecast.

10 . The computer-implemented method of claim 1 , wherein the post-processing operation comprises at least one of: transforming the demand forecast to an original scale, restoring a decomposed forecast, forecasting for new services, overwriting forecasts that are unusable, overlaying an inorganic forecast on top of an organic forecast.

11 . A computing apparatus comprising:

a processor; and

a memory storing instructions that, when executed by the processor, configure the apparatus to:

access usage data of resources of a cloud service center;

standardize the usage data by applying a pre-processing operation to the usage data;

generate a plurality of forecast models based on the standardized usage data, the plurality of forecast models being configured to predict a demand of the resources of the cloud service center;

select a demand forecast model from the plurality of forecast models based on a ranking of the plurality of forecast models;

apply a post-processing operation to a demand forecast that is generated based on the selected demand forecast model, by, when the demand forecast is unusable, creating a new forecast by perturbing a last good forecast so that it is smoothly aligned with a latest actual of a time series of the usage data; and

configure the cloud service center based on the post-processed demand forecast.

12 . The computing apparatus of claim 11 , wherein the instructions further configure the apparatus to:

detect that a current demand forecast deviates from a previous demand forecast;

detect noise in recent usage data; and

in response to detect that the current demand forecast deviates from the previous demand forecast, and detecting the noise, shifting the previous demand forecast to align with the recent usage data.

13 . The computing apparatus of claim 12 , wherein the instructions further configure the apparatus to:

determine a time-to-live metric based on the shifted previous demand forecast and a capacity of the cloud service center, the time-to-live metric indicating a predicted time at which demands for the resources of the cloud service center exceed the resources of the cloud service center; and

detect that the time-to-live metric transgresses a capacity threshold of the cloud service center,

wherein configuring the cloud service center is in response to the time-to-live metric transgress the capacity threshold.

14 . The computing apparatus of claim 11 , wherein the instructions further configure the apparatus to:

generate a graph in a graphical user interface that illustrates the demand forecast model and a time-to-live metric; and

generate a user-interactive graphical user interface element that enables an administrator of the cloud service center to allocate additional resources to the cloud service center relative to the time-to-live metric.

15 . A non-transitory computer-readable storage medium, the computer-readable storage medium including instructions that when executed by a computer, cause the computer to:

access usage data of resources of a cloud service center;

standardize the usage data by applying a pre-processing operation to the usage data;

generate a plurality of forecast models based on the standardized usage data, the plurality of forecast models being configured to predict a demand of the resources of the cloud service center;

select a demand forecast model from the plurality of forecast models based on a ranking of the plurality of forecast models;

apply a post-processing operation to a demand forecast that is generated based on the selected demand forecast model, by, when the demand forecast is unusable, creating a new forecast by perturbing a last good forecast so that it is smoothly aligned with a latest actual of a time series of the usage data; and

configure the cloud service center based on the post-processed demand forecast.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (11)