Time series data analysis

Patent metadata

Field	Value
Publication number	US-11122145-B2
Application number	US-201916670630-A
Country	US
Kind code	B2
Filing date	Oct 31, 2019
Priority date	Oct 31, 2019
Publication date	Sep 14, 2021
Grant date	Sep 14, 2021

Abstract

The present approach relates to the use of time series analyses to estimate times or time intervals when a user of IT resources is likely to schedule or request that an operation is run on those services. In certain implementations, the present approach performs forecasting using time series data and supervised machine learning techniques. These techniques may be used to help predict future times when an operation or operations may be requested for execution. Based on these predicted future time, automations (e.g., the automated execution of operations) may be scheduled so as to effectively utilize available resources and efficiently perform the operations.

First claim

The invention claimed is:

1. A processor-implemented method for predicting occurrence of an information technology (IT) operation, comprising:
receiving as an input a set of time series data comprising a series of time values corresponding to one or more types of operations performed on an information technology (IT) platform;
determining a plurality of time difference values for the set of time series data, wherein each time difference value comprises the time between different sequential operations performed on the IT platform;
pre-processing the set of time series data to remove time observations associated with outlier time difference values to derive a pre-processed dataset, wherein pre-processing the set of time series data comprises acts of:
in an iterative loop:
characterizing a set of clusters;
sorting the time difference values into respective clusters of the set of clusters; and
processing the respective clusters to remove clusters containing outlier time difference values; and

exiting the iterative loop when no further clusters are determined to be removable or a number of total time difference values fall below a threshold value, wherein an output from the iterative loop upon exiting comprises the pre-processed dataset;

configuring one or more time series analytic routines based on the pre-processed dataset to specify a final dataset;
processing the final data set using at least two time series analytic routines each generating a respective trained model;
calculating one or more measures of a fit of each respective trained model; and
selecting a final trained model for predicting future requests of the one or more types of operation based on the one or more measures of the fit for each respective trained model.

2. The method of claim 1 , wherein characterizing the set of clusters comprises specifying a number of clusters and a range of each cluster.

3. The method of claim 1 , wherein processing the respective clusters to remove the clusters containing the outlier time difference values comprises evaluating each cluster of the respective clusters based upon a discard threshold value.

4. The method of claim 1 , wherein the at least two time series analytic routines comprise machine learning routines.

5. The method of claim 1 , wherein the at least two time series analytic routines comprise one or both of a linear regression analytic routine or a random forest analytic routine.

6. A processor-based system, comprising:
one or more processing components;
one or more memory or storage components encoding routines executable by the one or more processing components, wherein the routines, when executed, causes actions to be performed comprising:
receiving as an input a set of time series data comprising a series of time values corresponding to one or more types of operations performed on an information technology (IT) platform;
determining a plurality of time difference values for the set of time series data, wherein each time difference value comprises the time between different sequential operations performed on the IT platform;
in an iterative loop:
characterizing a set of clusters;
sorting the time difference values into respective clusters of the set of clusters; and
processing the respective clusters to remove clusters containing outlier time difference values;

exiting the iterative loop when no further clusters are determined to be removable or a number of total time difference values fall below a threshold value, wherein an output from the iterative loop upon exiting comprises a pre-processed dataset;
configuring one or more time series analytic routines based on the pre-processed dataset to specify a final dataset;
processing the final data set using at least two time series analytic routines each generating a respective trained model;
calculating one or more measures of a fit of each respective trained model; and
selecting a final trained model for predicting future requests of the one or more types of operation based on the one or more measures of the fit for each respective trained model.

7. The processor-based system of claim 6 , wherein the IT platform comprises a client instance of a cloud-based platform and the one or more types of operations were scheduled or requested by a user of the client instance.

8. The processor-based system of claim 6 , wherein the series of time values comprise timestamps for the one or more types of operations.

9. The processor-based system of claim 6 , wherein characterizing the set of clusters comprises specifying a number of clusters and a range of each cluster.

10. The processor-based system of claim 6 , wherein processing the respective clusters to remove the clusters containing the outlier time difference values comprises evaluating each cluster of the respective clusters based upon a discard threshold value.

11. The processor-based system of claim 6 , wherein the at least two time series analytic routines comprise machine learning routines.

12. The processor-based system of claim 6 , wherein the at least two time series analytic routines comprise one or both of a linear regression analytic routine or a random forest analytic routine.

13. The processor-based system of claim 6 , wherein the one or more measures of the fit of each respective trained model comprise one or both of root mean squared error or mean absolute error.

14. The processor-based system of claim 6 , wherein the routines, when executed, cause further actions to be performed comprising:
using the final trained model to predict a next occurrence of the one or more types of operations; and
managing scheduling of the one or more type of operations or other operations based on the predicted next occurrence.

15. A method for selecting a model for predicting information technology (IT) operations, comprising:
pre-processing a set of time series data to derive a pre-processed dataset by performing acts of:
in an iterative loop:
characterizing a set of clusters;
sorting a plurality of time difference values into respective clusters of the set of clusters; and
processing the respective clusters to remove clusters containing outlier time difference values; and

exiting the iterative loop when no further clusters are determined to be removable or a number of total time difference values fall below a threshold value, wherein an output from the iterative loop upon exiting comprises the pre-processed dataset

processing the pre-processed data set comprising time difference values using at least two time series analytic routines, each generating a respective trained model, wherein each time difference value of the time difference values comprises the time between different sequential operations performed on an IT platform;
for each respective trained model, calculating one or both of root mean squared error (RMSE) or mean absolute error (MAE);
comparing the RMSE or MAE for each respective trained model; and
based on the comparison of the RMSE or MAE, selecting a final trained model for predicting future requests of one or more types of operations to be performed on the IT platform.

16. The method of claim 15 , wherein the IT platform comprises a client instance of a cloud-based platform and the one or more types of operations were scheduled or requested by a user of the client instance.

17. The method of claim 15 , wherein the at least two time series analytic routines comprise one or both of a linear regression analytic routine or a random forest analytic routine.

18. The method of claim 15 , further

Assignees

• Servicenow Inc

Inventors

• Goudarzi Khashayar
• Kuraku Vinayak Raju
• Vaddempudi Sharath

Classifications

• H04L41/149

  for prediction of maintenance · CPC title

• H04L41/40

  using virtualisation of network functions or resources, e.g. SDN or NFV entities · CPC title

• H04L67/60

  Scheduling or organising the servicing of application requests, e.g. requests for application data transmissions using the analysis and optimisation of the required network resources (admission control or resource allocation H04L47/70) · CPC title

• H04L41/147 Primary

  for predicting network behaviour · CPC title

• G06F11/3447

  Performance evaluation by modeling · CPC title