Machine Intelligence at the NOC

The network operations center (centralized monitoring and control station for telecommunications networks) is primarily concerned with fault management and performance management to maintain network efficiency and customer satisfaction. Ericsson's already highly automated NOC automation has been enhanced with machine intelligence technologies such as advanced analytics, deep learning, and machine inference to drive smart network operations that combine the high performance and ease of use of new 5G systems.

What Are The Specific Challenges Of Machine Intelligence @ NOC?

The key challenges of NOC management today, outlined in the following figure, are:

• ·         Troubleshoot billions of service alarms
• ·         Process approximately 20 million notifications for workflow management by NOC experts.
• ·         Handle millions of emails from the service desk
• ·         Higher costs due to low utilization of workflow management.

Incident management is an area that already uses an expert system-based framework. However, due to the ever-changing nature of the network, both in terms of technology and implementation, it is very difficult to maintain human-written rules in such expert systems. By automating incidents in a data and domain independent manner, automation of network operations center can be vastly improved without the need for expert rules. As an example, failure of one node can lead to cascading failure of the other node, resulting in a large number of alarms. Machine learning techniques can be used to discover concurrent patterns in this set of alarms and other events, allowing you to quickly identify the root cause in most failure scenarios. This frees the NOC operations team to focus on more complex challenges.

How Complex Is This?

Common NOC alarm handling involves mapping incoming alarms to incidents using enrichment, aggregation, deduplication, and correlation techniques. This is difficult due to the unevenness of the alarm information caused by the multi-technology and multi-vendor solutions used in today's communication networks. This heterogeneity makes it difficult to create a harmonious vision of the network and greatly increases the complexity associated with fault detection and resolution.

 Can You Afford To Encode Your Domain Knowledge Long-Term?

Current noc team include rule-based processing of alarms from various sources, such as nodes, service management systems, element / network management systems. The rules are written to transform domain-specific information into an overview of the network in the NOC, and also include coded rules for handling / correlating alarms for proper grouping. I am

                         

The development of such rules is time consuming and manual. Continuous network change with new types of network nodes and the resulting new types of alarms also complicate rule development and maintenance. Also, rule generation / update must be done frequently. Otherwise, the rules database will be incomplete or inaccurate.

Does That Mean Stopping The Development Of Domain-Oriented Rules?

This does not mean that traditional rule development has disappeared, but rather that it is enhanced by a domain-independent, data-driven approach. Additionally, automatic detection of possible correlations between alarms improves the rule-based approach when rules are incomplete or when knowledge of a particular domain has not yet been acquired.

The data-driven approach enables the identification of correlation between domains and the generation of data-based information. Over time, the system evolves into a fully automated solution.

NOC Data Driven Automation

Here is a case study on automatic incident formation, root cause, and self-healing scenarios that we have been working on as part of our investigation.

We apply artificial intelligence principles (data mining and data science) to discover patterns of behavior from large historical data sets. These behaviors or patterns essentially signify the correlation between alarms and co-occurrence patterns. One of the interesting aspects of our approach is that it not only evaluates it as time series data, but it also processes most of the symbolic or categorical information collected from the network from which the potential behavior also considered how to identify it.

This approach helps domain professionals learn unknown evolving behavior patterns when the environment is multi-technology and multi-vendor. These correlated and grouped patterns allow automatic grouping of alarms, configuring automatic detection of network incidents, root cause, and self-healing stages.

With this approach, you can achieve intelligent grouping of alarms and tickets with minimal manual intervention. By automatically identifying critical and missing groups, you can reduce or eliminate manual rule development and reduce the total number of trouble tickets.

Automatic Incident Detection

The grouping of fault conditions and alarms is

Embeds network information such as alarms and events into the phone company's knowledge graph, and includes underlying raw and insightful derived information to enable intelligent and automated NOC behavior.

Automatically capture network data behavior for alarms and events in a digitized, data-driven version. This is called the machine learning generation (ML) rule.

This approach allows you to use automatic identification of merged and hardened states instead of checking individual alarms one by one. The data-based function automatically creates compound conditions from historical information. That is, pattern mining techniques are used to intelligently group alarms across domains. These compound conditions are replaced as ML generation rules that help detect a group of alarms called incidents.

Frequent patterns are sets of elements, subsequences, or substructures that occur in a dataset with a certain frequency. Frequent pattern mining algorithms range from frequent item set mining, sequential pattern mining, structured pattern mining, correlation mining, associative classification, and frequent pattern-based grouping. Finding patterns in your data can help you extract associations, correlations, and other interesting relationships between your input data. Telecommunication network data conforms to an extended variant of the pattern mining algorithm to generate generated machine learning rules.