China Unicom uses AI knowledge graphs in move towards more autonomous networks

Who: China Unicom

What: Increasing the efficiency of fault handling across its mobile networks

How: China Unicom collaborated with Nokia and Baidu to develop and deploy an intelligent fault management tool based on AI knowledge graph technology to better identify and resolve network faults.

Results:

• Fault handling efficiency up by 67%
• Annual costs reduced by 11.3 million euros
• More accurate root cause analysis

Traditionally, network operation and maintenance has been a largely manual process. Not only is this time-consuming, but the results can be patchy, in part because employees bring different levels of experience, knowledge and expertise to the task of identifying and solving problems.

China Unicom, for example, found that 40% of its network fault management analysis was manual, and consumed 158,400 person-working-days per year at an annual cost equivalent to 18 million euros.

To top it all, the operator wasn’t even reaching the desired levels of efficiency. It took an experienced engineer between 30 minutes and 1 hour to deal with a 5G network fault, from identification, and root cause analysis, through to decision-making. This meant one person handled on average up to 12 problems a day.

China Unicom needed to go faster. It an extensive 5G network and at the end of October this year reported 251.943 million 5G package subscribers, as well as 474.387 million internet-of-things terminal connections, and 7,441 customers served by virtual 5G industry private networks.

It wasn’t only a question of resolving issues more quickly. China Unicom also wanted to greatly speed up the dissemination and use of knowledge. It points out that in the case of intelligent fault management, for example, it can take at least two months to acquire insight from the summaries provided by experts. Automation was the answer.

Automating network operations

In 2021 China Unicom set out to develop and deploy a network O&M knowledge center based on AI knowledge graph technology, which it could use to manage faults across multiple mobile domains, while addressing different scenarios, including network planning, construction, maintenance, optimization and operations. All while supporting rapid automation of its network operations.

“Based on knowledge graph technology, the solution can automatically collect, integrate and conclude multi-domain network operation and maintenance knowledge, realize the transformation from artificial experience to knowledge intelligence, and finally form a knowledge closed-loop to serve various scenarios of the Autonomous Networks,” according to China Unicom.

Capabilities include:

• A Knowledge management framework: This supports the flexible collection and conversion of multiple data sources, such as alarm data, inventory data and work order data. Through the combination of traditional AI algorithm and in-graph algorithm, it realizes the acquisition of intelligent fault handling knowledge and forms an applicable knowledge center. All knowledge is stored in the graph database, and inference capabilities are open to all third-party applications through pre-defined TM Forum Open APIs.
• Network domains: Mobile network cross-domain and cloud core cross-layer fault management are two common pain points in network operations, which usually require human expertise to resolve. However, a knowledge graph can quickly deduce the root cause and suggest processes with a high accuracy rate, using a generated graph network.
• Three principal knowledge graph applications: Alarm correlation, root cause recommendation, and decision tree self-discovery help improve the accuracy of fault root cause analysis by 27% and improve alarm processing time by 67%.
• Open architecture: The solution is decoupled from the CSP’s existing OSS system and draws on big data & AI capabilities. In addition, it is designed to be scalable, using a cloud-native architecture and a Kubernetes platform.
• Other capabilities include the use of RESTful Open APIs to all third-party systems and closed-loop knowledge collection. In addition, online AI graph algorithms and 13 offline AI graph algorithms improve the accuracy of intelligent analysis and decision making. The system also provides real time recommendation and 100% interpretable.

China Unicom has recorded numerous benefits from deploying the new system:

• Increased knowledge

Whereas it can take at least two months to manually summarize experts’ mobile network cross-domain fault handling insight, with a knowledge graph technology the time needed falls to between two and four weeks. In addition, there is a more comprehensive gathering of content, resulting in a 38% increase in knowledge.

• Cost reduction

Prior to deploying the knowledge graph, on average 40% of the network fault analysis was manual, which cost around 18 million euros per year and demanded approximately 158,400 person-working-days. Now only 15% of network fault analysis is manual, saving 11.3 million euros annually.

• More efficient network fault handling

With the knowledge graph most problems can be identified and analyzed automatically, reducing the average process time from 45 minutes previously to 15 minutes today.

• Greater root cause analysis accuracy

The operator’s Root Cause Analysis accuracy rate used to be around 70%; since deploying the AI knowledge graph this figure has increased to 89%.

• A fully interpretable inference process

Today 100% of the inference processes can be explained, compared to 75% in the past, which helps front-line engineers understand recommended actions.

TM Forum standards provide a framework for automation

China Unicom worked with Baidu and Nokia to design and implement the knowledge graph solution using TM Forum’s Autonomous Networks Framework (IG1218 Autonomous Networks – Business requirements & architecture Version 2.0.0, IG1253 Intent in Autonomous Networks Version 1.0.0) and Open APIs (TMF921 Intent Management API).

China Unicom won the 2023 Excellence Award for Excellence in Autonomous Operations. Find out more about their entry and all the 2023 winners. Submit your entry for the 2024 Excellence Awards here.