Intelligent node failure prediction and ticket triage solution

Patent metadata

Field	Value
Publication number	US-11252052-B1
Application number	US-202117180706-A
Country	US
Kind code	B1
Filing date	Feb 19, 2021
Priority date	Nov 13, 2020
Publication date	Feb 15, 2022
Grant date	Feb 15, 2022

Abstract

Embodiments provide for prediction and mitigation of network faults. Information associated network nodes may be compiled and used to generate network analytical records (NARs). A first model may be executed against the NARs to predict faults associated with one or more nodes of the network. Tickets are generated for predicted faults and stored in a ticket database. The tickets may be analyzed to predict executable actions to mitigate the faults associated with each ticket. To analyze the tickets, ticket data may be compiled and used to generate ticket analytical records (TARs). A second model may be executed against the TARs predict actions to resolve the predicted faults. The predicted actions may be executed to mitigate the impact that the faults have on the network, which may include preventing the faults entirely (e.g., via preventative maintenance) or minimizing the impact of the faults via use of the predicted actions.

First claim

What is claimed is:

1. A method for predicting and mitigating faults of network nodes, the method comprising:
receiving, by one or more processors, node data associated with nodes of a network, wherein the node data comprises at least performance data for each of the nodes of the network and alarm data associated with nodes of the network;
generating, by the one or more processors, network analytical record (NARs) based on the node data, wherein the NARs comprise node metrics associated with one or more nodes of the network and information associated with a first fault;
evaluating, by the one or more processors, the NARs against a model configured to:
compare the node metrics of the NARs to a pattern of node metrics corresponding to the first fault; and
determine a node of the one or more nodes of the network predicted to experience the first fault based on whether a portion of the node metrics corresponding to the node exhibit the pattern of node metrics corresponding to the first fault;

generating, by the one or more processors, a ticket associated with the node predicted to experience the first fault;
storing the ticket associated with the node predicted to experience the first fault at a ticket management system;
generating ticket analytical records (TARs) based at least in part on ticket data, the ticket data including tickets associated with faults predicted by the model, wherein each of the TARs includes a set of ticket metrics associated with the tickets;
evaluating, by the one or more processors, the TARs against a second model configured to:
identify self-healing tickets within tickets and non-self-healable tickets within the tickets; and
determine, for each of the non-self-healable tickets, actions predicted to resolve faults corresponding to each of the non-self-healable tickets; and

executing, by the one or more processors, at least one action predicted to resolve at least one of the faults corresponding to at least one the non-self-healable tickets.

2. The method of claim 1 , wherein generating the NARs comprises:
generating backward rolling data based on the node data;
generating forward rolling data based on the node data; and
merging the backward rolling data and forward rolling data to produce a first NAR, wherein the node metrics of the NARs are derived from the backward rolling data and the first fault is determined based on a forward rolling metrics.

3. The method of claim 2 , wherein the alarm data comprises information associated with status alarms and information associated with failure alarms, and wherein the backward rolling data is generated based on the failure alarms.

4. The method of claim 3 , wherein the node data further comprises troubled ticket data and device disconnection data, and wherein the backward rolling data is generated based on the performance data, the status alarms, the troubled ticket data, and the device disconnection data.

5. The method of claim 1 , further comprising:
generating historic NARs based on historical data;
fitting a base model to the historic NARs;
tuning characteristics of the base model based on the fitting to produce the model; and
providing the model to a fault prediction engine configured to predict faults within the network.

6. The method of claim 1 , wherein the second model is a multistage classification model that includes a first classifier and a second classifier, wherein identifying the self-healing tickets and the non-self-healing tickets is based on evaluation of a first TAR against the first classifier, and determining the actions predicted to resolve the faults corresponding to each of the non-self-healable tickets is based on an evaluation of the non-self-healable tickets associated with the first TAR against the second classifier.

7. The method of claim 1 , further comprising:
determining confidence scores for the actions predicted to resolve faults corresponding to each of the non-self-healable tickets, wherein each of the confidence scores indicates a likelihood that a predicted action will resolve the fault of the corresponding non-self-healable ticket; and
determining to assign the at least one ticket for execution or investigation based on the confidence score.

8. The method of claim 1 , further comprising:
constructing an action dictionary based at least in part on historical ticket data;
tagging historical tickets included in the historical ticket data with one or more tags derived from the action dictionary; and
training the second model based on the tagged historical ticket data.

9. The method of claim 8 , wherein the ticket data comprises metadata, ticket update data, and troubled tickets data, and wherein generating a first TAR comprises:
extracting log data from the ticket update data and the troubled tickets data;
extracting action data from the log data based on the action dictionary;
generating bucketed data based on the metadata, the ticket update data, and the troubled tickets data, wherein the bucketed data comprises: bucketed metadata derived from the metadata; bucketed log data derived from the ticket update data and the troubled tickets data; and bucketed failure data derived from the troubled tickets data, the ticket update data, and the metadata; and
aggregating the bucketed data and a portion of the action data to produce the first TAR, wherein a second TAR comprises the bucketed data of the first TAR and a different portion of the action data.

10. A non-transitory computer-readable medium storing instructions that, when executed by one or more processors, cause the one or more processors to perform operations for predicting and mitigating faults of network nodes, the operations comprising:
receiving node data associated with nodes of a network, wherein the node data comprises at least performance data for each of the nodes of the network and alarm data associated with nodes of the network;
generating network analytical records (NARs) based on the node data, wherein a first NAR of the NARs comprises a first set of node metrics associated with one or more nodes of the network and information associated with a first fault;
evaluating the first NAR against a model configured to:
compare the first set of node metrics of the first NAR to a pattern of node metrics corresponding to the first fault; and
determine a node of the one or more nodes of the network predicted to experience the first fault based on whether a portion of the first set of node metrics corresponding to the node exhibit the pattern of node metrics corresponding to the first fault;

generating a ticket associated with the node predicted to experience the first fault;
storing the ticket associated with the node predicted to experience the first fault at a ticket management system;
generating ticket analytical records (TARs) based at least in part on ticket data, the ticket data including tickets associated with faults predicted by the model, wherein each of the TARs includes include ticket metrics derived, at least in part, on the tickets;
evaluating, by the one or more processors, the TARs against a second model configured to:
identify self-healing tickets within the tickets and non-self-healable tickets within the tickets; and
determine, for each of the non-self-healable tickets, actions predicted to resolve faults corresponding to each of the non-self-healable tickets; and

executing, by the one or more processors, at least one action predicted to resolve at least one of the faults corresponding to at least one the non-self-healable tickets.

11. The non-transitory computer-readable medium of claim 10 , wherein the alarm data comprises information associated with status alarms

Assignees

• Accenture Global Solutions Ltd

Inventors

• Babu Balasubramani Shoban
• Navinchandra Jivani Meet
• Chaudhary Ruchi
• Sanyal Kaushik

Classifications

• H04L41/149

  for prediction of maintenance · CPC title

• H04L41/5074 Primary

  Handling of user complaints or trouble tickets · CPC title

• H04L41/147 Primary

  for predicting network behaviour · CPC title

• G06N20/00

  Machine learning · CPC title

• H04L41/0654

  using network fault recovery (ring fault isolation or reconfiguration in loop networks without recovery actions by a network management system H04L12/437) · CPC title