Glasgow researchers use machine learning to build network digital twin

Researchers at the University of Glasgow have developed a new way to test networks, which they claim is 25,000 times faster than traditional approaches.

Shenjia Ding, a research student at the university’s School of Computing Science, used automatically generated digital twins – built with machine learning – to test two complex American and European computer networks, with 12 and 37 nodes, respectively.

The test included six different types of traffic, including web browsing, video streaming and file downloads, alongside continuous congestion and background noise to simulate real conditions.

The team’s digital twin took 4.78 seconds to test the speed of the networks, compared with the 33 hours needed to run the same tests using a traditional simulator.

As internet traffic and data volumes grow exponentially, the scientists say their approach could become a “practical, scalable and cost-effective” approach to testing and managing networks.

Traditional network testing involves using simulators to mimic real-world scenarios and data traffic to test the performance, security and reliability of a computer network. The researchers used automated machine learning (AutoML) to build the digital twin, which they said not only speeds up the process of building machine learning tools, but can also be used by non-experts with limited machine learning expertise.

“Our results show that testing computer networks with automatically generated digital twins can achieve high accuracy and significantly faster speeds than traditional simulator-based testing,” said Ding. “We’re demonstrating a very promising alternative to manual and time-consuming testing that also relies heavily on professional expertise.”

Testing computer networks with automatically generated digital twins can achieve high accuracy and significantly faster speeds than traditional simulator-based testing

Paul Harvey, a co-author of the research and a senior lecturer at Glasgow University’s School of Computing Science, is also a co-investigator for TransiT, a collaboration between Heriot-Watt University in Edinburgh, the University of Glasgow and 70 industry partners, funded by the UK Research and Innovation Engineering and Physical Sciences Research Council.

TransiT is looking to identify the fastest, least-risky and lowest-cost pathways to transport decarbonisation in the UK. Harvey believes the research shows how using machine learning to build digital twins could be applied in other network settings, such as transport. 
 
“Transport, like computing, is seeing enormous growth in data volumes, and in both instances, the pressure on the communications networks carrying all this data is immense,” Harvey explained.

“By proving that we can use machine learning to build digital twins – which is another time-consuming and laborious task – we are highlighting the huge potential of this research to also test and optimise transport and other networks that we rely on daily.”

He said Ding’s research could potentially support TransiT, particularly in its goal of creating a “digital twin factory” that can automate the production of digital twins for transport settings.

The researchers plan to focus on validating the digital twin’s update mechanisms and cost, assessing performance in real-time network environments, and conducting a comparative study across diverse network scenarios.

Ding will present a paper, Automated digital twin generation for network testing: A multi-topology validation, which looks at the use of automated digital twins in network management, at the 2026 IEEE International Conference on Communications (ICC) in Glasgow later this month.

The paper is co-authored by Paul Harvey and David Flynn from the University of Glasgow.