Issue link: https://fhpublishing.uberflip.com/i/1218318
M ore consum- ers and more connections. More solar panels, more heat pumps - and lots more electric vehicles (EVs). Keeping the lights on has never been more chal - lenging. And with customers increasingly reliant on supply continuity, the threats of inter- ruptions and consequent pen- alties are major preoccupations for every distribution network operator (DNO). Low voltage (LV) fault loca - tion and management has been a largely reactive process for the best part of a century; and a‚er all that time it's still a far from perfect science. The first indica- tor of a problem in that tricky last mile can o‚en be when a customer phones in to say their lights are flickering or they've lost power altogether. But even then, the underly - ing cause - or causes - that have NETWORK / 23 / MARCH 2020 number of customers being directly affected. UK wide, our devices have re- corded in excess of one million real customer fault incidents and over 15 million hours of load monitoring data over the last few years. As the strain on DNOs' ageing infrastructure increases, the number of faults is rising faster than ever. A proactive approach to LV fault monitoring With upwards of half a million LV substations in the UK, the cost of putting a discrete moni- toring device on every circuit is clearly prohibitive. As the distribution landscape becomes ever more challenging, what's therefore needed is a more intel - ligent mechanism for DNOs to anticipate and remediate faults before customers are aware that there's a problem. At Kelvatek, we're harnessing our deep understanding of LV network behaviours to create innovative new fault detec - tion and management strate- gies. Drawing on the talents of so‚ware engineers, physicists, mathematicians and data science experts, we are using data from UK energy assets to formulate predictive models that help network operators do their day jobs more efficiently. Powerful artificial intel - ligence and machine learning algorithms can si‚ through voltage patterns, sensor signals and historical datasets from LV cables and other assets, uncov- ering the real story hidden in gigabytes of network data. The main object of their interest is a huge historic data triggered an inbound customer report may be unclear. There could be a genuine fault event that's caused a blown fuse on the substation. Equally, a circuit could be a temporarily over - loaded when lots of customers happen to be drawing power at that moment. While these LV ca- bles could easily be 80 years old or more, digging up the streets to check their condition clearly isn't an option. Traditionally LV distribution networks have been character - ised by a single supply point. While a blown fuse means lost power for customers connected directly to that point, nobody else on the network is affected. More recently, we've seen the rise of meshed networks fed from multiple points. If a fuse blows, other circuits on the net- work can help take the strain. However, there's a downside to this approach. Faults on one circuit can propagate rapidly, leading to a significantly higher set of real-life fault behaviours, collected by tens of thousands of connected monitoring devices sitting on operator's networks. This extensive library of fault information is augmented with dynamic load patterns and a wide range of data points from other sources. Sharing informa - tion with DNO partners gives us a detailed picture of cable routes and connectivity, health and history plus the number and location of smart meters and other third-party devices on the network - and of course data from our devices. This granular, continually updated picture about the network infrastructure itself is further augmented by live and historic insights into other fac - tors that subtly shape customer demand patterns. This can be everything from prevailing weather conditions to economic and social/demographic data about a neighbourhood and its inhabitants. As an example, an area that's dominated by afflu - ent households with solar pan- els installed, and a nearby Tesla dealership, gives some strong hints about expected customer demand patterns. Actionable intelligence: a forensic picture of network health Aggregating and analysing this uniquely valuable dataset allows the team to create a so- phisticated virtual model of an operator's network. In turn, this yields a forensic picture of that network's current health and demand trends over time. Armed with this intelligence, we can help operators antici - pate future faults or problem- atic changes in load profiles and predict when these issues are likely to affect customer service. Our DNO partners can thus enjoy a significant operational advantage with the ability to allocate resources in good time - whether it's fitting a fuse, deploying network flexibility measures or adopting tradi - tional network reinforcement techniques. KEY POINTS l Strain on DNOs' ageing infrastructure is ever increasing l Understanding of low voltage network behaviours allows for better fault detection and management l Enhance customer service by anticipating faults or changes in load