SYSTEM FAILURE PREDICTION USING LONG SHORT-TERM MEMORY NEURAL NETWORKS

• US 20170293542A1
• Filed: 04/04/2017
• Published: 10/12/2017
• Est. Priority Date: 04/06/2016
• Status: Active Grant

• Current Assignee: NEC Corporation
• Original Assignee: NEC Laboratories America Inc (NEC Corporation)
• Inventors: Xu, Jianwu, Zhang, Ke, Zhang, Hui, Min, Renqiang, Jiang, Guofei
• Granted Patent: US 10,289,509 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 11/008 Reliability or availability analysis
• CPC Class Codes: G06F 11/2263 using neural networks
• CPC Class Codes: G06N 3/044 Recurrent networks, e.g. Hopfield networks
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G06N 3/084 Backpropagation, e.g. using gradient descent

Abstract

Methods for system failure prediction include clustering log files according to structural log patterns. Feature representations of the log files are determined based on the log clusters. A likelihood of a system failure is determined based on the feature representations using a neural network. An automatic system control action is performed if the likelihood of system failure exceeds a threshold.

Claims

1. A method for system failure prediction, comprising:

clustering a plurality of log files according to structural log patterns;

determining feature representations of the log files based on the log clusters;

determining a likelihood of a system failure based on the feature representations using a neural network; and

performing an automatic system control action if the likelihood of system failure exceeds a threshold.

2. The method of claim 1, wherein determining feature representations comprises determining two, and only two, feature representations.

3. The method of claim 2, wherein the two feature representations consist of pattern distribution among clustered logs and term frequency-inverse document frequency.

4. The method of claim 1, wherein clustering the plurality of log files comprises performing unsupervised, hierarchical clustering.

5. The method of claim 1, wherein determining the likelihood of system failure comprises processing the feature representations using a long short-term memory recurrent neural network.

6. The method of claim 5, further comprising training the long short-term memory recurrent neural network from training data and user labels on failure time periods.

7. The method of claim 1, wherein determining the likelihood of a system failure comprises determining a binomial distribution based on an output of the neural network and a binary target vector with two complementary classes.

8. The method of claim 1, wherein the plurality of log files comprise a plurality of different log formats.

9. The method of claim 8, wherein the plurality of different log formats comprise different delimiter formats.

10. A method for system failure prediction, comprising:

clustering a plurality of log files having heterogeneous log formats according to structural log patterns using unsupervised, hierarchical clustering;

determining feature representations of the log files based on the log clusters, wherein the feature representations consist of pattern distribution among clustered logs and term frequency-inverse document frequency;

determining a likelihood of a system failure based on the feature representations using a long short-term memory neural network a binomial distribution based on an output of the long short-term memory neural network and a binary target vector with two complementary classes; and

performing an automatic system control action if the likelihood of system failure exceeds a threshold.

11. A system for system failure prediction, comprising:

a pattern learning module comprising a processor configured to cluster a plurality of log files according to structural log patterns;

a feature extraction module configured to determine feature representations of the log files based on the log clusters;

a failure prediction module configured to determine a likelihood of a system failure based on the feature representations using a neural network; and

a system control module configured to perform an automatic system control action if the likelihood of system failure exceeds a threshold.

12. The system of claim 11, wherein the feature extraction module is further configured to determine two, and only two, feature representations.

13. The system of claim 12, wherein the two feature representations consist of pattern distribution among clustered logs and term frequency-inverse document frequency.

14. The system of claim 11, wherein the pattern learning module is further configured to perform unsupervised, hierarchical clustering.

15. The system of claim 11, wherein the failure prediction module is further configured to process the feature representations using a long short-term memory recurrent neural network.

16. The system of claim 11, wherein the failure prediction module is further configured to determine a binomial distribution based on an output of the neural network and a binary target vector with two complementary classes.

17. The system of claim 11, wherein the plurality of log files comprise a plurality of different log formats.

18. The system of claim 17, wherein the plurality of different log formats comprise different delimiter formats.