Forecasting for Resource Allocation

Patent metadata

Field	Value
Publication number	US-2019197413-A1
Application number	US-201715855823-A
Country	US
Kind code	A1
Filing date	Dec 27, 2017
Priority date	Dec 27, 2017
Publication date	Jun 27, 2019
Grant date	—

Abstract

Forecasting resource allocation is disclosed. An example method includes receiving operating data from a resource; applying periodicity tests to the received operating data using a plurality of sketches of time series of prior operating data, the periodicity tests generating periodic components; applying regression models to the received operating data, the regression models collectively generating a trend component, each regression model being applied over a different time scale of a plurality of time scales; computing a trend model using the periodic components and a trend component; determining a random process describing the historical evolution of the trend model; and calculating and providing a mean prediction, an upper bound, and a lower bound for resource utilization at a future time using the trend model and a predicted distribution.

First claim

What is claimed is:

1 . A computer-implemented method for resource utilization forecasting comprising:
receiving operating data from a resource; applying periodicity tests to the received operating data using a plurality of sketches of time series of prior operating data, the periodicity tests generating periodic components; applying regression models to the received operating data, the regression models collectively generating a trend component, each regression model being estimated over a different time scale of a plurality of time scales; computing a trend model using the periodic components and the trend component; determining a random process describing the historical evolution of the trend model; and calculating and providing a mean prediction, an upper bound, and a lower bound for resource utilization at a future time using the trend model and a predicted distribution.

2 . The method of claim 1 , wherein each of the plurality of sketches holds a different version of the time series of prior operating data, each of the plurality of sketches has a different amount of compression of the time series of prior operating data, and the plurality of sketches are used by the periodicity tests to test for a range of periodicities using the different amounts of compression in the plurality of sketches.

3 . The method of claim 2 , wherein the periodicity tests include a hypothesis test for fitting multiple smooth periodic components, the periodic components including at least one of hourly, daily, weekly, weekend, and weekday.

4 . The method of claim 3 , wherein the periodicity tests further include a hypothesis test for periodic spikes in the received operating data and a first subset of sketches from the plurality of sketches of time series of prior operating data.

5 . The method of claim 4 , wherein the periodicity tests further include a test for periodicity in the variance in the received operating data and a second subset of sketches from the plurality of sketches of time series of prior operating data.

6 . The method of claim 5 , wherein the periodicity tests include a test for arbitrary periodic behavior in the received operating data and a third subset of sketches from the plurality of sketches of time series of prior operating data, the test for arbitrary period behavior including an analysis of cyclic autocorrelation.

7 . The method of claim 1 , wherein computing the trend model includes adding the periodic components and the trend component.

8 . The method of claim 1 , wherein the regression models of the trend component are combined using a weighted sum and weights of the weighted sum are adjusted over the forecasting interval.

9 . The method of claim 1 , wherein determining the predicted distribution of the trend model includes fitting a Wiener process to the historical evolution of the regression models.

10 . The method of claim 8 , wherein determining the random process describing the evolution of the trend model includes modeling statistical properties of at least one of time between step discontinuities and signed magnitudes of the step discontinuities.

11 . The method of claim 1 further comprising:
allocating resources using the mean prediction, the upper bound, and the lower bound.

12 . A system comprising:
a processor; and a memory coupled to the processor, the memory storing instructions executable by the processor to perform a method, the method comprising:
receiving operating data from a resource;
applying periodicity tests to the received operating data using a plurality of sketches of time series of prior operating data, the periodicity tests generating periodic components;
applying a plurality of regression models to the received operating data, each regression model being applied over a different time scale;
computing a trend model using the periodic components and the trend component;
determining a random process describing the historical evolution of the trend model; and
calculating and providing a mean prediction, an upper bound, and a lower bound for resource utilization at a future time using the trend model and a predicted distribution.

13 . The system of claim 11 , wherein each of the plurality of sketches holds a different version of the time series of the prior operating data, each of the plurality of sketches has a different amount of compression of the time series of the prior operating data, and the plurality of sketches are used by the periodicity tests to test for a range of periodicities using the different amounts of compression of the plurality of sketches.

14 . The system of claim 12 , wherein the periodicity tests include a hypothesis test for fitting multiple smooth periodic components, the periodic components including at least one of hourly, daily, weekly, weekend, and weekday.

15 . The system of claim 13 , wherein the periodicity tests further include a hypothesis test for periodic spikes in the received operating data and a first subset of sketches from the plurality of sketches of time series of prior operating data.

16 . The system of claim 14 , wherein the periodicity tests further include a test for variance in the received operating data and a subset of sketches from the plurality of sketches of time series of prior operating data.

17 . The system of claim 15 , wherein the periodicity tests include a test for arbitrary periodic behavior in the received operating data and a third subset of sketches from the plurality of sketches of time series of prior operating data, the test for arbitrary period behavior including an analysis of cyclic autocorrelation.

18 . The system of claim 11 , wherein computing the trend model includes adding the periodic components and the trend component.

19 . The system of claim 11 , wherein the regression models of the trend component are combined using a weighted sum and weights of the weighted sum are adjusted over the forecasting interval.

20 . The system of claim 11 , wherein determining the predicted distribution of the trend model includes fitting a Wiener process to the historical evolution of the regression models.

21 . The system of claim 18 , wherein determining the predicted distribution of the trend model includes modeling statistical properties of at least one of time between step discontinuities in the received operating data and signed magnitudes of the step discontinuities.

22 . A system comprising:
means for receiving operating data from a resource; means for applying periodicity tests to the received operating data using a plurality of sketches of time series of prior operating data, the periodicity tests generating periodic components; means for applying a plurality of regression models to the received operating data, each regression model being applied over a different time scale; means for computing a trend model using the periodic components and the trend component; means for determining a random process describing the historical evolution of the trend model; and means for calculating and providing a mean prediction, an upper bound, and a lower bound for resource utilization at a future time using the trend model and a predicted distribution.

Assignees

• Elasticsearch Bv

Inventors

• Veasey Thomas
• Dodson Stephen

Classifications

• G06N20/00

  Machine learning · CPC title

• G06Q10/06315

  Needs-based resource requirements planning or analysis · CPC title

• G06Q10/06314

  Calendaring for a resource · CPC title

• G06F16/252

  between a Database Management System and a front-end application · CPC title

• G06N5/02 Primary

  Knowledge representation; Symbolic representation · CPC title