Machine discovery of aberrant operating states

• US 10,438,124 B2
• Filed: 12/11/2017
• Issued: 10/08/2019
• Est. Priority Date: 01/25/2017
• Status: Active Grant

• Current Assignee: CenturyLink Intellectual Property LLC (Lumen Technologies, Inc.)
• Original Assignee: CenturyLink Intellectual Property LLC (Lumen Technologies, Inc.)
• Inventors: Kirk, Ryan
• Primary Examiner(s): Zhen, Li B.
• Assistant Examiner(s): Haedi, Selene A.
• Application Number:
  US15/837,858
• Publication Number:
  US 20180211171A1
• Time in Patent Office: 666 Days
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 5/045 Explanation of inference; Explainable artificial intelligence [XAI]; Interpretable artificial intelligence
• CPC Class Codes: H04L 67/10 in which an application is distributed across nodes in the network software deployment G06F8/60; multiprogramming arrangements G06F9/46

Abstract

Novel tools and techniques for the machine discovery of aberrant states are provided. A system includes a plurality of network devices, and a decision system in communication with the plurality of network devices. Each of the plurality of network devices may be configured to generate a respective data stream. The decision system may include a processor and a non-transitory computer readable medium including instructions executable by the processor to obtain, via the plurality of network devices, one or more data streams. The decision system may build a historic model of a data stream, and determine a predicted value of the data stream at a future time, based on the historic model. The decision system may be configured to determine whether an anomaly has occurred based on a variation between a current value of the data stream and the predicted value of the data stream.

Claims

1. A system comprising:

a plurality of network devices associated with a cloud platform, each network device of the plurality of network devices configured to generate a respective data stream that includes a current value of a performance metric in real-time;

a decision system in communication with the plurality of network devices, the decision system comprising;

a processor;

a non-transitory computer readable medium comprising instructions executable by the processor to;

obtain, via the plurality of network devices, one or more data streams, each of the one or more data streams comprising real-time time-series data indicative of a network activity generated by a respective network device;

build a historic model of historic data for a data stream of the one or more data streams;

generate a windowed data stream, wherein the data stream includes a series of data points in time, wherein the windowed data stream is a list of fixed length data comprising historic values of the data stream;

add a new data point indicative of the current value of the data stream to the windowed data stream, wherein each new data point causes an ejection of a corresponding historic data point of the window;

determine one or more phase offsets for the windowed data stream, wherein the series of data points of the windowed data stream is shifted by the phase offset;

determine a phase weight for each respective phase offset of the one or more phase offsets;

determine, in real-time, a predicted value of the data stream at a future time, based on the historic model and based, at least in part, on the one or more phase offsets as weighted by a respective phase weight;

determine a variation between the predicted value and the current value of the data stream at the future time;

determine whether an anomaly has occurred based on whether the variation exceeds a threshold variation, wherein the threshold variation is determined as a function of the historic model; and

update the historic model based on the determination of whether the anomaly has occurred.

2. The system of claim 1, wherein the instructions are further executable by the processor to:

poll one or more of the plurality of network devices for respective data streams.

3. The system of claim 1, wherein the decision system further comprises a data stream buffer, wherein one or more network devices of the plurality of network devices are configured to publish periodically data points to the data stream buffer, wherein the data stream buffer is configured to associate the data points to their respective data streams.

4. The system of claim 1, wherein the instructions are further executable by the processor to:

determine a second predicted value of the data stream at a second future time, based at least in part on the predicted value.

5. The system of claim 1, wherein the instructions are further executable by the processor to:

retrieve metadata associated with the data stream, the metadata indicative of a range of a minimum value and maximum value of the data stream based on a type of the performance metric associated with the data stream.

6. The system of claim 1, wherein the instructions are further executable by the processor to:

determine a pattern of anomalies in the historic model of the data stream; and

trigger an alert responsive to a determination of the pattern of anomalies in the data stream.

7. The system of claim 1, wherein the instructions are further executable by the processor to:

predict, based on the presence of anomalies in the data stream, whether an incident will occur at respective network device of the plurality of network devices associated with the data.

8. The system of claim 1, wherein the instructions are further executable by the processor to:

perform a remedial action responsive to determining the presence of the anomaly on the data stream, wherein the remedial action includes determining one or more network devices of the plurality of network devices associated with the anomaly.

9. The system of claim 1, wherein the instructions are further executable by the processor to:

receive analyst feedback indicative of whether to flag the variation as an anomaly; and

update the historic model based on the analyst feedback.

10. An apparatus comprising:

a processor;

a non-transitory computer readable medium comprising instructions executable by the processor to;

obtain, via a plurality of network devices, one or more data streams, each of the one or more data streams comprising real-time time-series data indicative of a network activity generated by a respective network device;

build a historic model of historic data for a data stream of the one or more data streams;

generate a windowed data stream, wherein the data stream includes a series of data points in time, wherein the windowed data stream is a list of fixed length data comprising historic values of the data stream;

add a new data point indicative of the current value of the data stream to the windowed data stream, wherein each new data point causes an ejection of a corresponding historic data point of the window;

determine one or more phase offsets for the windowed data stream, wherein the series of data points of the windowed data stream is shifted by the phase offset;

determine a phase weight for each respective phase offset of the one or more phase offsets;

determine, in real-time, a predicted value of the data stream at a future time, based on the historic model and based, at least in part, on the one or more phase offsets as weighted by a respective phase weight;

determine a variation between the predicted value and the current value of the data stream at the future time;

determine whether an anomaly has occurred based on whether the variation exceeds a threshold variation, wherein the threshold variation is determined as a function of the historic model; and

update the historic model based on the determination of whether the anomaly has occurred.

11. The apparatus of claim 10, wherein the instructions are further executable by the processor to:

poll one or more of the plurality of network devices for respective data streams.

12. The apparatus of claim 10, wherein the instructions are further executable by the processor to:

determine the threshold variation for a data stream based, at least in part, on a range of a minimum value and a maximum value of the data stream, wherein the range of the minimum value and the maximum value is updated in real-time, and is based, at least in part, on metadata associated with a type of the performance metric associated with the data stream.

13. The apparatus of claim 10, wherein the instructions are further executable by the processor to:

trigger an alert responsive to a determination that an anomaly has occurred in the data stream for two consecutive polling cycles, wherein a new data point of the data stream is generated periodically every polling cycle.

14. The apparatus of claim 10, wherein the instructions are further executable by the processor to:

perform a remedial action responsive to determining the presence of the anomaly on the data stream, wherein the remedial action includes determining one or more network devices of the plurality of network devices associated with the anomaly.