Three-quarters of network issues are identified by end users, and over a third of problems are caused by manual network changes, according to data from ZK Research. Many operations and maintenance (O & M) staff spend as much as 90% of their time trying to locate faults. China Telecom is tackling these related challenges through TM Forum’s proof-of-concept Catalyst program.“Our multi-vendor network is too complex to locate all the problems,” said Qian Bing, Technical Director of Network AI Center, China Telecom. “This is an important problem for us to solve.”
China Telecom is collaborating with Huawei, ZTE, AsiaInfo and BOCO Inter-Telecom to explore a solution. The team’s project name,
Glaucus Precision on Telco Data, gives a clue as to the focus of the work. In Greek mythology, Glaucus was a mortal who turned into a prophetic sea-god through eating a magical herb.
The team is exploring the use of artificial intelligence (AI), unique data quality control technology, standardized data cleansing and quality analysis to prepare multi-vendor network data for AI model training. The aim is to locate, solve and predict problems automatically – with the goal of saving staff time and the business money, as well as dramatically improving customer experience.
The project is part of China Telecom’s major ¥10 million ($1.42 million) drive over the next two years to capitalize on AI technology.
Preparing dataEach Catalyst participant plays a specific role. Huawei was responsible for preparing the data using AI-driven pre-processing for metadata definition, multi-data-source correlation, data labeling and real-time monitoring of data quality
ZTE and AsiaInfo are responsible for data analysis to find abnormal cells and provide early warnings. Huawei and BOCO Inter-Telecom provide traffic predictions to further evaluate abnormal cells, predict changes to trends in the short term and forecast demand patterns for the long term.
In the pilot the team used the solution developed through the Catalyst in 15,000 cells in Guangdong and Guangxi, which resulted in a reduction in network complaints at China Telecom by 12%. The mean time to resolve problems was also reduced dramatically by 80%, while network operations center (NOC) efficiency was up 20%. Further, the team said power efficiency could be improved by 5% due to traffic pattern predictions.
The team reported achieving 89% accuracy in the evaluation of existing anomaly issues and 97% accuracy for both short- and long-term predictions.
“97% accuracy for traffic prediction is the highest we have seen anywhere in the world,” said Giles Vincent, Senior Consultant, Huawei.
TakeawaysFrom this work, the team has gleaned some key takeaways that apply to all telcos and their partners:
- Data quality is key: The mantra ‘garbage in, garbage out’ applies.
- Operations management center (OMC) consistency is crucial, particularly in multi-vendor networks.
- Labeling automation is important: When it comes to machine learning, everything should be labelled using automation for speed and accuracy.
China Telecom plans to use the Catalyst’s solution to optimize its recently launched 5G network.
At the recent TM Forum Digital Transformation Asia event in Kuala Lumpur, the
Glaucus Precision on Telco Data team won the
Outstanding Catalyst – business impact award.
The team used a number of TM Forum assets in their solution and will contribute their findings back for the benefit of the whole community. These include the
Open Digital Architecture (a more agile replacement for traditional operational and business support system architecture) and
Open APIs.
“TM Forum has so many wonderful assets – use them,” Vincent encouraged.
The team will contribute a new Open API related to data acquisition and managing data flows.
Bing said: “Through projects such as this, we hope to optimize each of our algorithms with our partners.”
