IN Brief:
- UK Power Networks is analysing millions of half-hourly commercial and industrial electricity readings using AI.
- The work is intended to improve visibility of demand patterns across the low voltage distribution network.
- More detailed network information can support investment and connection decisions as commercial electricity use changes.
UK Power Networks is applying artificial intelligence to millions of half-hourly electricity readings from commercial and industrial customers, extending its use of data analytics to improve visibility of demand on the low voltage network.
The work is intended to give planners a more detailed view of how business electricity consumption changes through the day than would be available through manual analysis of large numbers of individual readings. The resulting demand profiles can help inform investment decisions and connection assessments across the distribution system.
Commercial and industrial loads are particularly difficult to generalise. Offices, workshops, warehouses, shops and manufacturing facilities operate to different schedules, while individual sites can change substantially when production equipment, heating systems, charging infrastructure or other electrical loads are added.
A half-hourly profile therefore provides different information from an annual consumption figure or a single assumed peak. A local transformer may appear lightly loaded for much of the year but approach its limit during a repeatable period each weekday, while another part of the network can retain significant headroom because customers there follow different operating patterns.
Turning millions of readings into useful engineering information is a scale problem well suited to automated analysis. Models can group recurring patterns, identify unusual behaviour and estimate utilisation across large numbers of network assets far more quickly than engineers could inspect the raw data manually.
UK Power Networks has already been using smart meter information, monitoring equipment and machine learning to improve visibility of its low voltage system. Its Smart Grid Index says a predictive data science model uses monitored substations to estimate utilisation across the wider network, reducing the need for dedicated monitoring equipment at every location.
The company says physical monitoring covers 14% of secondary substations in that model, representing around 30% of households, while the resulting analysis is used to forecast utilisation across the entire low voltage network. Commercial and industrial interval data adds another source of evidence to the same wider move towards more detailed planning.
The distinction matters as connection demand becomes more varied. A business seeking a larger electricity supply may be adding machinery, converting heat from gas to electricity, installing vehicle chargers or changing operating hours. Those changes can create sharp increases in local peak demand even where annual consumption does not rise at the same rate.
More accurate information about the existing load can therefore change the reinforcement question. If a transformer or cable has genuine spare capacity during the periods when a new customer intends to operate, some work may be avoidable. If detailed data shows a constraint that broad assumptions had missed, investment may need to be brought forward.
Neither outcome can be decided by AI alone. Electrical connection assessments still have to consider transformer and cable ratings, voltage limits, fault levels, protection arrangements and the behaviour of the network under abnormal conditions. Data analysis can improve the starting assumptions, but engineering limits still determine what can be connected safely.
The same caution applies to forecasts. Historical readings describe how a customer used electricity at the time they were recorded, not how the site is guaranteed to operate in future. A factory can add a production line, a warehouse can become highly automated or a fleet operator can install a large charging system within a relatively short period.
Models therefore need to be updated as demand changes rather than treated as permanent descriptions of a network. Their value lies in processing volumes of data frequently enough to identify changing patterns and direct engineering attention towards areas where a constraint or unexpected opportunity is emerging.
Low voltage networks present a particularly large data problem because of the number of assets involved. UK Power Networks operates a very large population of secondary substations and associated feeders, making comprehensive physical monitoring expensive and difficult to deploy quickly.
Using monitored sites and smart meter information to estimate utilisation elsewhere provides an alternative. It does not create the same evidence as a sensor installed directly on each transformer, but it can narrow uncertainty and identify the locations where additional monitoring or conventional engineering studies are most valuable.
The company’s digitalisation programme also makes aggregated smart meter information available through its open data activities, while its wider network planning processes increasingly use monitoring and analytics to decide where traditional reinforcement or other interventions are required.
That becomes more important as electricity flows become less predictable. Rooftop solar can reduce daytime imports or create export, batteries can switch rapidly between demand and generation, and electric vehicle charging can add substantial short-duration loads. Commercial sites can combine several of those technologies behind one connection.
As those patterns become more common, historic assumptions about a typical business customer become less reliable. The network increasingly has to be understood from measured behaviour and regularly updated forecasts rather than broad demand categories alone.
The current AI work is therefore better understood as a visibility and planning tool than an autonomous network controller. UK Power Networks is using computing power to turn a very large body of interval consumption data into information that engineers and connection teams can use when deciding where capacity exists and where intervention may be required.
The practical test will be whether those estimates remain accurate as individual businesses change how they consume electricity. If they do, the approach gives the distribution network another way to direct reinforcement towards genuine constraints while making better use of assets that still have capacity available.