AI future-proofs power network
Image credit: IET
An award-winning project is having a positive impact on the Scottish electricity network.
A good test of the worthiness of an award winner comes from revisiting that project further into its working life. Will the water that has passed under the bridge wash away the gloss? That is certainly not the case with the Networks Constraints Early Warning System (NCEWS), which won the 2019 E&T Innovation of the Year at the IET’s annual Innovation Awards.
This collaboration between Heriot-Watt University and SP Energy Networks (SPEN) used large-scale smart meter / asset data to develop advanced machine-learning algorithms that can extract information about missing cable assets and voltage excursions. This enables high prediction accuracy, even if, as is often the case, some data is missing.
It was a project initially supported by both the Knowledge Transfer Partnership and the Network Innovation Allowance (aimed at funding network operators’ innovation projects), but what started as an interesting project has evolved into an important tool for SPEN, as the company’s smart systems manager Fiona Fulton explains.
“We started to use that data network model in our business. We’ve used the data science piece around the backfilling cables to provide complete models for design", she says. "You can imagine when we are designing a reinforcement network we are taking into consideration quite a large number of cables and assets as part of that design and we are able to demonstrate now how that can automatically export a nearly complete network model into the design tools. We are able to run many more scenarios as part of those designs.”
It has become the platform to identify what assets (notably cables) are missing from the network, and what will be needed for development of the network.
It comes at a critical point for electricity suppliers. In bygone years companies would count the properties, estimate where properties were likely to be built and multiply the combined number by an estimated demand. “To be honest that has worked, the service has been pretty stable, but we can see that fundamentally changing,” claims Fulton. “It’s not going to take too many unmanaged electric car superchargers down the street to exceed the rating on your local substation.”
Another potentially disruptive element is input to the grid from independent renewable supplies. The options to counter this are either massive reinforcement, which requires planning, or being “much cleverer about how we manage the LV network”. Fulton continues: “A lot of this is being driven by just being sensible and how we prepare for what could be a gradual or could be a step change in the way people use or deploy low-carbon tech. It has been a key driver, but also we have all this data coming available from smart meters and we have to make sure we use it as effectively as we can.”
Interestingly, the project kicked off based on the premise that there would be an influx of data provided by home smart meters, but the model has been opened up to include inputs from a variety of assets, such as smart meters, low-voltage monitors, higher-voltage monitors, and electric-vehicle charging points. The more information the more accurate the model and the better the analysis it can perform.
Dr Valentin Robu, associate professor and co-director of Smart Systems Group at Heriot-Watt University, describes the project as far more than just building a model: “The initial stage of the project involved developing a data platform for visualising the network and for pulling information from the network, using the geographical information system. The input of the Heriot-Watt University team in NCEWS was to develop the machine learning and AI algorithms that can extract useful information from the data available in the SPEN platform.”
He also stressed that AI is often seen as a silver bullet, but the reality is that the success of this project revolves around rigorous testing of the model as new data becomes available and features are added. Close cooperation with the SPEN team is key, as practically useful AI techniques cannot be developed in isolation.
As is often the case for award-winning work, Robu points out it resulted from a team effort: “The key person was Dr Maizura Mokhtar, but supporting her were myself and Professor David Flynn from Heriot-Watt University, as well as Caroline Loughran, Jim Whyte and Fiona Fulton – all from SP Energy Networks – and Ciaran Higgins from Derryherk Ltd.”
E&T Innovation Awards
For last year’s winner it was more than just picking up a trophy.
“We have a clear plan for moving to more data science and large-scale analytics, as set out in our recently published Digitalisation Strategy, which we’re pushing forward with.
"However, we were delighted to have our efforts recognised externally as it makes it clear our focus is shared across the industry, and acknowledges the work of our team to get us to this point,” says Fiona Fulton, smart systems manager, SP Energy Networks, E&T Innovation of the Year award winner in 2019.
The new-look E&T Innovation Awards (previously the IET Innovation Awards), feature a range of new categories to reflect engineering excellence in these unprecedented times in which we find ourselves living:
■ Digital Health and Social Care
■ Future Power & Energy
■ Communications & IT
■ Manufacturing 4.0
■ Sustainable Planet
■ Protecting Society & Saving Lives
■ Cyber Security
■ Smarter World
■ Model-Based Engineering
■ Intelligent Systems
■ Future Unicorn
■ Diversity & Inclusion Impact
■ Small Idea, Big Impact: Global Challenge
■ Leader of the Year
The deadline for entries is 3 July 2020, so don’t delay.
Winners will be announced at a gala dinner on 19 November 2020 at The Pavilion at the Tower of London.
For more details on the categories and how to enter, visit us at eandt.theiet.org/innovation
Sign up to the E&T News e-mail to get great stories like this delivered to your inbox every day.