Performance prediction from communication data

Patent metadata

Field	Value
Publication number	US-11604969-B2
Application number	US-201916553465-A
Country	US
Kind code	B2
Filing date	Aug 28, 2019
Priority date	Sep 18, 2018
Publication date	Mar 14, 2023
Grant date	Mar 14, 2023

Abstract

Systems and methods for predicting system device failure are provided. The method includes representing device failure related data associated with the devices from a predetermined domain by temporal graphs for each of the devices. The method also includes extracting vector representations based on temporal graph features from the temporal graphs that capture both temporal and structural correlation in the device failure related data. The method further includes predicting, based on the vector representations and device failure related metrics in the predetermined domain, one or more of the devices that is expected to fail within a predetermined time.

First claim

What is claimed is:

1. A method for predicting system device failure, comprising:
representing device failure related data associated with a plurality of devices from a predetermined domain by at least one temporal graph for each of the plurality of devices;
extracting, by a processor device, vector representations based on temporal graph features from the at least one temporal graph that capture both temporal and structural correlation in the device failure related data;
predicting, based on the vector representations and at least one device failure related metric in the predetermined domain, at least one of the plurality of devices that is expected to fail within a predetermined time;
outputting the at least one of the plurality of devices with a predicted device failure metric; and
replacing the at least one of the plurality of devices prior to an impending failure, based on the failure metric,
wherein representing the device failure related data further comprises:
performing, by the processor device, temporal graph construction using training data for at least one entity in a training phase to determine at least one first temporal graph; and performing temporal graph feature extraction from the at least one first temporal graph in the training phase to derive at least one first feature vector.

2. The method as recited in claim 1 , further comprising:
training, by the processor device, a plurality of models for system failure prediction via graph learning (SFPGL) for device failure prediction based on the at least one temporal graph.

3. The method as recited in claim 2 , wherein training the plurality of models further comprises:
preparing training data;
learning prediction models from the plurality of models based on the training data; and
performing model selection to find a best model of the plurality of models.

4. The method as recited in claim 1 , further comprising:
performing, by the processor device, temporal graph construction in a testing phase to determine at least one second temporal graph;
performing temporal graph feature extraction from the at least one second temporal graph in the testing phase to derive at least one second feature vector; and
performing device failure prediction based on the at least one second feature vector and outputting at least one prediction result.

5. The method as recited in claim 4 , wherein the device failure related data includes communication data and device profile data and performing temporal graph construction iii the testing phase further comprises:
encoding the communication data and the device profile data.

6. The method as recited in claim 4 , wherein performing temporal graph construction further comprises:
generating at least one multi-scale temporal graph at multiple time granularities.

7. The method as recited in claim 4 , wherein performing temporal graph feature extraction from the at least one second temporal graph further comprises
profiling a time series based on at least one of a raw value, a statistic measurement and a temporal differential measurement.

8. The method as recited in claim 4 , wherein performing temporal graph feature extraction from the at least one second temporal graph further comprises:
deriving structure features from a one-hop metric.

9. The method as r d in claim 4 , wherein performing temporal graph feature extraction from the at least one second temporal graph further comprises:
deriving structure features from a multi-hop metric.

10. The method as recited in claim 1 , wherein each at least one temporal graph is represented as a stream or graphs<G 1 to G x >, where G i is a graph that records communication data and profile data for devices at time t i .

11. The method as recited in claim 1 , wherein nodes and edges of each at least one temporal graph are associated with attributes, where node attributes include all information only relevant to a node at time t i , and edge attributes include information relevant to a corresponding communication.

12. A computer system for predicting system device failure, comprising:
a processor device operatively coupled to a memory device, the processor device being configured to:
represent device failure related data associated with a plurality of devices from a predetermined domain by at least one temporal graph for each of the plurality of devices;
extract vector representations based on temporal graph features from the at least one temporal graph that capture both temporal and structural correlation in the device failure related data;
predict, based on the vector representations and at least one device failure related metric in the predetermined domain, at least one of the plurality of devices that is expected to fail within a predetermined time; and
replace the at least one of the plurality of devices prior to an impending failure, based on the failure metric,
wherein, when representing the device failure related data, the processor device is further configured to:
perform temporal graph construction using training data for at least one entity in a training phase to determine at least one first temporal graph; and
perform temporal graph feature extraction from the at least one first temporal graph in the training phase to derive at least one first feature vector.

13. The system as recited in claim 12 , wherein the processor device is further configured to:
train a plurality of models for system failure prediction via graph learning (SFPGL) for device failure prediction based on the at least one temporal graph.

14. The system as recited in claim 12 , wherein, when raining the plurality of models, the processor device is further configured to:
prepare training data;
learn prediction models from the plurality of models based on the training data; and
perform model selection to find a best model of the plurality of models.

15. The system as recited in claim 14 , wherein the processor device is further configured to:
perform device failure prediction by feeding the vector representations into the best model.

16. The system as recited in claim 12 , wherein each at least one temporal graph is represented as a stream of graphs<G 1 to G x >, where G i is a graph that records communication data and profile data for devices at time t i .

17. The system as recited in claim 12 , wherein nodes and edges of each at least one temporal graph are associated with attributes, where node attributes include all information only relevant to a node at time t i , and edge attributes include information relevant to a corresponding communication.

18. A computer program product for predicting performance of a plurality of devices, the computer program product comprising a non-transitory computer readable storage medium having program instructions embodied therewith, the program instructions executable by a computing device to cause the computing device to perform the method comprising:
representing device failure related data associated with the plurality of devices from a predetermined domain by at least one temporal graph for each of the plurality of devices;
extracting vector representations based on temporal graph features from the at least one temporal graph that capture both temporal and structural correlation in the device failure related data;
predicting, based on the vector representations and at least one performance metric in the predetermined domain, at least one of the plurality of devices that is expected to fail within a predetermined time; and
replacing the at least one of the plurali

Assignees

• Nec Lab America Inc
• Nec Corp

Inventors

• Cheng Wei
• Tang Luan
• Song Dongjin
• Zong Bo
• Chen Haifeng
• Ni Jingchao
• Yu Wenchao

Classifications

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/084 Primary

  Backpropagation, e.g. using gradient descent · CPC title

• G06N5/01

  Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound · CPC title

• G06N20/10

  using kernel methods, e.g. support vector machines [SVM] · CPC title