Research areas.

The connectivity of an LV network — which customers share conductors, where branches join, where the boundary lies — is rarely documented well. Utility GIS records are often incomplete, out-of-date, or simply absent for older sections.

We develop methods to infer the network tree from smart-meter data alone. The meters already record voltage and power at customer premises; the question is how to read the structure out of those signals without site visits or extra sensors.

Topology gets worked on the most because almost every other method here depends on it — load flow, loss accounting, fault location, and downstream phase or impedance estimation all assume a network you can trust.

OpenVerifying topology estimates at scale without ground-truth GIS to compare against — and at the same time, building better methods that estimate better trees and work on larger networks.

LV networks are three-phase, but each customer connects to just one. Which phase a customer is on affects load balance, voltage quality, and how their meter readings should be interpreted. Phase records in the field are frequently incomplete or wrong — particularly for older networks where connections have been changed and not redocumented.

We infer phase membership from the correlation structure of voltage time-series across the smart-meter fleet. Customers on the same phase exhibit a distinct correlation signature compared to customers on different phases, even without measurements at the substation. The "grouping" step clusters all customers simultaneously into three phase classes — no labels, no priors.

OpenWe're implementing a variety of phase-grouping methods so they can be tested against each other — working out which approaches are most effective, and under what conditions.

Distinct from topology, impedance estimation focuses on the electrical properties of the conductors themselves — specifically the resistance R and reactance X of each cable. These values determine how voltage drops across the network, how losses accumulate, and how power flows under varying load conditions.

As with topology, utility records of cable impedances are often unreliable or missing. We're developing methods to estimate impedance from smart-meter observations, either jointly with topology or as a follow-on step. Accurate impedances also act as a cross-check: they should be consistent with the topology that produced them.

OpenHow much of impedance is identifiable from meter data alone, vs. how much needs a prior over standard cable types — and whether joint topology + impedance estimation outperforms the two-step approach.

Distribution transformers sit between the medium-voltage grid and LV customers. Knowing how a transformer is loaded, what voltage it is presenting, and whether it's under thermal or operational stress is valuable for network management — but instrumenting transformers directly is expensive and impractical at scale.

Virtual sensing estimates transformer-level quantities from the smart-meter readings already collected at customers downstream. No hardware change at the transformer itself. The aim is to make transformer monitoring tractable for utilities where dedicated sensor deployment isn't economically viable.

OpenDeveloping different methods to estimate different transformer parameters — total loading, voltage, thermal state — and characterising how each method holds up under typical LV measurement conditions.

Knowing the current state of a network is useful; knowing how it responds to change is essential for planning. Situation analysis is a tool for scenario exploration — taking the network model produced by the other methods and stress-testing it against hypothetical futures, so problems can be found and reasoned about before they arrive in the field.

Scenarios we want to support out-of-the-box:

OpenHow operators want to interact with scenario results — single answer? envelope of outcomes? a sensitivity ranking of which assumptions move the needle most? The right output format is itself a research question.