The key problem for UK grid operators is that they have inherited a system which was designed for an era of centralised generation and relatively predictable consumption, running on assets which are getting old. Some underground low voltage (LV) cables, for example, were installed more than 100 years ago and have been patched and mended many times. Many transformers, which control voltages, were built well over 50 years ago and have yet to be upgraded to modern standards.

Rebuilding the grid infrastructure wholesale is not an option. It is far too expensive and disruptive. Nor is it entirely necessary, apart from targeted reinforcement and replacement. The real challenge is to make maximum use of the assets we already have by making them smarter – and using smarter technology to better manage them. So how can AI and deep ML help?

Old tech costs 
Take the thousands of miles of buried LV cables which are the final links between the grid and consumers. Deteriorating joints and disintegrating insulation cause many thousands of failures of supply each year – representing 40% of customer complaints and 75% of customer minutes lost being the average amount of time that a customer is without power in a year. The first thing a network operator knows about an incident is usually when customers report an outage, because a fuse on the network has ruptured and tripped out the power.

Finding and fixing LV cable faults has until now been expensive and difficult – as much as 80% of ongoing network costs. The first step is usually to replace the fuse and hope the problem goes away – but without knowing the exact cause and whether it might re-occur.

If that does not work, the conventional solution is to send a signal down the cable, which reflects from the failure site and gives an approximate location. The final step is to use more precise location technologies at the site and dig a hole. All of this involves time, money, unhappy customers and possible sanctions for unacceptable service from the regulator, Ofgem.

AI prevention is better than cure
The application of AI and deep ML to the LV cable problem has the potential to change operators’ approach radically – from reactive to predictive. Instead of waiting for a cable to fail before acting, new mathematics-based AI technology has been shown to have the ability to detect, classify and locate pre-fault events, long before they lead to actual failures.

Otherwise known as proto-faults or pecking faults, these are typically very small incidents of arcing, often transient, caused by water ingress. As the joint or insulation continues to deteriorate, they may worsen and lead to failure.

Under Ofgem’s Network Innovation Allowance scheme to support innovative technologies, we completed two projects to demonstrate AI and deep ML’s potential to tackle the LV cable fault issue. Developed and trialled in association with Scottish and Southern Electricity Networks (SSEN) and UK Power Networks, the result is the Synchronous Analysis and Protection System (SYNAPS).

SYNAPS works by taking voltage waveform data from sensors attached to LV cables and using hybrid ML to search for proto-faults, which can be as brief as a single microsecond. When they are detected, they are analysed and classified by specially developed algorithms to reveal the nature of the event and its precise location. For example, a cable identified by SYNAPS during trials as being in the process of failing was excavated and showed no external sign of problems – but subsequent forensic analysis confirmed damage in the joint. It was indeed on the road to failure.

Learning to be smarter
The nature of AI and deep ML is that the system becomes more accurate and faster, the more it learns. Early trials of SYNAPS were based on relatively little data about proto-faults in LV cables and the progression towards failure. This has changed rapidly with the input of real-world fault-failure scenarios in a wide range of conditions, and realistic simulations – and will continue to evolve as SYNAPS learns even more. But we still need to collect and input more data.

Much work has been successfully completed on calibrating SYNAPS for different types of LV network and classifying events in terms of their probabilities of leading to failures. Of the 300,000 proto-fault events recorded by the system, it identified only around 500 (1.5%) as significant enough to warrant further investigation.

In other words, it can distinguish between problem events and false alarms. And it can provide predictions of time-to-failure with ever-increasing accuracy to enable pro-active planned maintenance, delivering significant cost savings that are up to 75–80% less than the cost of unplanned maintenance.

From dumb to smart
The UK has more than a million LV substations (the boxes on poles and at roadsides near the end of networks). The great majority have no instrumentation, monitoring or communication capabilities whatsoever. Many larger substations contain assets which are decades old and are similarly dumb. With no data in or out, there is no opportunity to be part of a digitised, AI-enabled grid.

Where communications systems do exist across the grid, they are often from different eras, using varying protocols that are not compatible with each other. The good news is that a new generation of communications technology is emerging to achieve two things – retro-fit dumb assets to become data-enabled, and allow disparate communications systems to talk with each other.

Instead of waiting for a cable to fail before acting, new mathematics-based AI technology has been shown to have the ability to detect, classify and locate pre-fault events, long before they lead to actual failures.

AI for EVs
A fully digitised and communications-enabled grid is still a work in progress. But significant developments are moving forward in several areas, including the use of AI, deep ML and satellite navigation to help network operators handle rapidly expanding demand from EV charging, coupled with fluctuating power inputs from renewable generation. Because unpredictable load spikes could overwhelm LV networks, leading to faults and outages.

In the UK alone, the National Grid predicts 11mn EVs and plug-in hybrids by 2030 (up from 1.27mn in July 2022) and by 2050 up to 80% of households with an EV will be smart charging their cars. This presents problems – coping with the volume, location and timing of charging. Network operators need detailed, real-time and localised visibility of rapid load changes and their effects on grid performance and reliability, to balance generation and demand at local level and prevent issues.

A UK consortium of industry participants is addressing the problem in a project called ENERSYN, using a combination of satellite navigation, data gathering and ML technologies. Its aim is to give grid operators full visibility of potentially unexpected LV network loads due to EV charging.

Space technology
With backing from the European Space Agency (ESA), the ENERSYN partners are developing a new, standardised monitoring platform for subsystems and powerlines. The platform is based on gathering detailed, real-time information on the time and volume of load changes across LV grids, together with hyper-localised data on where the load changes are taking place.

Changes in load are monitored using shoebox-sized sensors to measure electrical current and voltage some hundred times per second, accurately identifying when EV charging is connected to, or disconnected from, the LV network. Trials have shown the platform can detect when EV charging is taking place in multiple specific locations, with an accuracy of 98.9% and timings precise to billionths of a second.

ML algorithms analyse waveform and location data, getting smarter as more data is gathered and analysed. The result promises to provide distribution companies with the fullest possible situational awareness of grid load changes and problems. A fine-grained picture that can then be analysed in various ways using deep ML. This could include intelligent predictions of future load patterns over time.

Future grid applications
SYNAPS and ENERSYN demonstrate that AI and deep ML have enormous potential for smarter management of our transmission and distribution system. Their application to LV cable fault prediction and EV charging are only a start. Far from being simply problem-prevention technologies, AI-based solutions have profound implications for smarter load balancing, better asset management, and 24/7 monitoring of the health of whole networks, leading to smarter maintenance and investment strategies.

But – and it is a big but – a smarter grid will only become reality if the assets of which it comprises are brought up to speed with the digital age. Because AI is driven by data. And data needs to be collected and communicated for it to be processed into useful management information and effective control systems.