Industries / Energy, water and waste / Electric transmission and distribution / Smart metering
Outages, leaks and theft
Meter data shows where the power has gone off, where water or gas is escaping, and where energy is used but not paid for. Northeast Group, a consultancy, put the world's losses to electricity theft and other unbilled use at US$96 billion a year in 2017.
7
Milestones achieved, from 1983 to 2024.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1983
Gas meters that shut themselves
Azbil Kimmon began full-scale production of "My Safe," a Japanese city gas meter that cuts off the gas on its own in an earthquake of intensity 5 or higher, and in 1986 added a pressure-sensitive quake sensor and self-checks.
Credit
Azbil Kimmon
2010
Smart meters hacked for cash
An FBI alert from May 2010, published by the journalist Brian Krebs in 2012, said people were reprogramming smart meters with an optical probe and a laptop or stopping them with magnets, and estimated a Puerto Rico utility's losses could reach US$400 million a year. The FBI believed former employees of the meter maker and of the utility were altering meters for cash.
Lessona smart meter does not stop theft if insiders and the meter's local service port are left unguarded.
Credit
FBI (alert); Brian Krebs (publication)
2012
Meters report the outage
During Superstorm Sandy, PECO in Pennsylvania used the "last gasp" message a smart meter sends when it loses power, and queried meters to confirm which homes already had power back. With 10% of meters fitted, it avoided 6,119 crew trips and restored customers two to three days sooner than otherwise; its older one-way meters had managed to send outage alarms from only about 1% of meters.
Credit
PECO; US Department of Energy (report)
2016
Tamper alarms find theft
The US Department of Energy reported that utilities in its smart grid grant programme confirmed an average of 951 tamper cases each, and that CenterPoint Energy recovered more than US$4.5 million from 2012 to 2014 by finding slow meters, unregistered meters and theft.
Lessonsome alarms came from faulty meters, not theft, and false alarms raised the cost of field checks.
Credit
US Department of Energy; CenterPoint Energy
2018
A public theft dataset
Zheng and colleagues released daily readings of 42,372 State Grid Corporation of China customers over 1,035 days, 3,615 of them confirmed thieves, with their paper on detecting theft with neural networks; it became the field's standard test set.
Lessonan independent re-analysis reports that many public results above 0.90 on the F1 score leak test data into training, and gets 0.49 when the leak is removed.
Credit
Zheng et al.; State Grid Corporation of China
2022
Court restores anti-theft meters
The Brazilian state of Amazonas banned centralised meter boxes, which put meters out of reach on theft-resistant lines with a display for each customer; Justice Luís Roberto Barroso of the Supreme Court suspended the ban in October 2022 as an invasion of federal powers. The distributors' association Abradee said blocking the meters would cost R$41.6 million in 2022 alone.
Lessonmeters moved out of the customer's sight lose trust unless customers can still read their own figures.
Credit
Supreme Federal Court of Brazil; Abradee
2024
Meters find household leaks
Thames Water said in November 2024 that smart water meters had helped find about 84,000 leaks on customers' own pipes over the previous decade, saving 120 million litres a day. South East Water in Melbourne reported about 3.1 billion litres saved from 2018 to June 2025 with 200,000 digital meters. Both figures are the companies' own.
Credit
Thames Water; South East Water
Watch for
Meters that spot line faults
Sense says software in newer smart meters detects arcing, downed lines and failing equipment, and that in a pilot with Georgia Power and Alabama Power it caught every vegetation-related problem "with zero false negatives to date."
Achieved whena utility or regulator publishes event counts checked against the utility's own records, with false-alarm rates and meter numbers.
Claimed by
Sense, February and June 2026
Open
01
An honest theft benchmark
Fixed, published train, validation and test splits of the State Grid of China theft data, split by customer, with a script that re-runs widely cited theft detectors under rules that keep test data out of training, and a public table of each paper's reported and reproduced scores.
The briefCredit · unclaimed02
A household leak benchmark
A public set of hourly household water meter readings with each leak confirmed or ruled out on site, and a scoring script, starting from the simplest detector, continuous flow overnight.
The briefCredit · unclaimed
Open 01 · the brief
An honest theft benchmark
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
Fixed, published train, validation and test splits of the State Grid of China theft data, split by customer, with a script that re-runs widely cited theft detectors under rules that keep test data out of training, and a public table of each paper's reported and reproduced scores.
Why it is still open
An independent repository shows scores falling from above 0.90 to 0.49 when the leaks are removed, but no shared benchmark with frozen splits or a public scoreboard was found.
Built with what exists
The public dataset; Python with scikit-learn, LightGBM and PyTorch; the authors' released code where it exists.
First step
Publish the customer-level split with checksums and the leak-free baseline.
Done looks like
Frozen splits with checksums, and reported against reproduced scores for at least 10 published papers.
Open 02 · the brief
A household leak benchmark
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of hourly household water meter readings with each leak confirmed or ruled out on site, and a scoring script, starting from the simplest detector, continuous flow overnight.
Why it is still open
The best public leak dataset describes itself as "the first fully labeled and publicly available dataset for leak detection and localization," but it comes from a laboratory rig with pipe sensors, not customer meters; no public household meter data with confirmed leak labels was found.
Built with what exists
Anonymised alert histories from a utility that already sends leak alerts; Python; open time-series models; privacy steps such as removing addresses and coarsening times.
First step
Ask Thames Water or South East Water for anonymised leak alerts with their confirmed outcomes.
Done looks like
At least 500 households with confirmed labels, published with a DOI, with precision and recall for the overnight-flow baseline and at least one model.