When a smart meter loses supply, it has enough residual charge to send a final message — a last-gasp event. Across a meter base of a few hundred thousand, these events form a live picture of where supply exists. Most utilities receive them, log them, and never use them for anything but a post-hoc audit.
The reason is not indifference. It is that a last-gasp event on its own is close to useless. It tells you a meter lost power. It does not tell you whether the street lost power, whether a device tripped, whether the meter itself failed, or which asset to send a crew to. Making the signal actionable requires the network topology — and the topology is exactly what most utilities do not have in a machine-readable, connected form.
From events to a location
With a connected model, the inference is tractable. Cluster last-gasp events by time window. Map each affected meter to its service point and trace upstream. The smallest network element common to all affected meters, and not common to unaffected meters still reporting, is the probable fault location.
Two refinements matter in practice. First, absence of evidence is evidence: meters that did not report a last-gasp but are downstream of a candidate device argue against that device, provided their communication path is independent. Second, protection operation from SCADA constrains the search space sharply when it is available, and should be treated as a strong prior rather than an equal input.
The inference is only as good as the topology. Utilities that skip the network model and buy the analytics get neither.
What changes operationally
The obvious gain is speed — crews dispatched to a device rather than a reported address. The less obvious gains are often larger.
- Restoration is verified rather than assumed, because power-restored events confirm supply at each consumer before the job closes.
- Reliability indices are computed from measured consumer-level durations instead of estimated affected counts.
- Nested outages become visible: a section restored while a single service drop remains dead no longer disappears from the picture.
- Call centre load falls, because affected consumers can be notified before they call.
The honest prerequisites
This is not a drop-in analytics purchase. It needs communication coverage good enough that last-gasp messages actually arrive, a network model that is connected and reasonably current, and a consumer index that ties meters to the right transformer. Utilities that have those three things get a large step change. Utilities that do not should build them first, because every other application of meter data depends on the same foundations.
