AI for Smart Grid Operations
Smart meters produced the richest network data utilities have ever had and, in most operators, the least used. The value is in what it reveals about the low-voltage network nobody could previously see.
A smart meter programme is usually justified on billing and delivers something more valuable: the first real visibility of what happens on the low-voltage network. Most operators are still using it for billing.
AI for smart grid operations covers advanced metering infrastructure analytics, distributed energy resource detection and forecasting, voltage and power quality analysis, flexibility and demand response dispatch, non-technical loss detection, and low-voltage network state estimation.
AMI data, and the resolution question
What you can do with smart meter data is determined almost entirely by its resolution and completeness, and both were usually specified for settlement rather than for network analysis.
| Resolution | What it supports | Typical situation |
|---|---|---|
| Monthly consumption | Billing only | Not a network dataset |
| Half-hourly consumption | Load profiling, basic forecasting | Common; useful but limited |
| Half-hourly with voltage | Voltage compliance, LV visibility | Where the real value begins |
| Minute-level or better | Power quality, DER detection | Uncommon; transforms what is possible |
| Event and alarm data | Outage detection, tamper | Frequently discarded at collection |
| Incomplete coverage | Limits everything | Estimation across gaps becomes the problem |
Voltage data is the underused half of the meter
Most operators collect consumption and discard or ignore voltage, which is the measurement that makes low-voltage network analysis possible: compliance monitoring, phase imbalance, identifying which customers sit on which transformer, detecting distributed generation. If your meters record voltage and your systems drop it, recovering that is one of the highest-return data engineering projects available in a distribution business.
Seeing the low-voltage network for the first time
Topology is frequently wrong and now checkable
Which customer is on which transformer, and which phase, is recorded in systems that have drifted from reality over decades. Correlating voltage patterns across meters identifies actual connectivity, and it commonly finds errors in the low single-digit percentages of connections — enough to matter.
Distributed generation is detectable from the meter
Unregistered solar generation has a distinctive signature in consumption and voltage data. Detecting it matters for planning, for voltage management and increasingly for network safety during outages.
Voltage compliance becomes measurable rather than sampled
Statutory voltage limits were historically checked by spot measurement. AMI voltage data makes compliance continuously measurable across the whole population, which is both an obligation and an opportunity to target reinforcement precisely.
Phase imbalance is expensive and invisible without this
Imbalance increases losses and stresses equipment, and identifying it at scale requires exactly the data smart meters produce.
Flexibility, losses and where advisory ends
- Non-technical loss detection is a well-posed problem. Consumption patterns inconsistent with premises characteristics, tamper events and network-level energy balance together identify candidates for investigation.
- Investigation capacity sets the threshold. As with inspection, more candidates than you can visit is a backlog rather than a result.
- Flexibility dispatch is an optimisation with hard constraints. Network limits are physical, contractual obligations are firm, and both are inputs rather than variables.
- Customer fairness matters in loss detection. A false accusation of theft is a serious harm; the output is an investigation lead and never a conclusion.
- State estimation support is close to the line. Improving an operator's view is advisory; automated action on that view is a safety component.
Where advisory ends in a distribution network
Estimating network state, flagging voltage excursions and recommending actions to a control engineer sits outside the AI Act's critical infrastructure high-risk category on the reading that matters: it informs a person who decides. A system that switches, curtails or reconfigures automatically is a safety component whose failure could cause physical damage — with conformity assessment, high-risk obligations from December 2027 and functional safety reconciliation attached. We build the first and treat the second as a separate programme requiring your protection engineers.
How an engagement runs
Start with what the meter data can actually support, which is usually more than is being used.
AMI data assessment
Resolution, completeness, whether voltage is retained, and what the data can genuinely support.
Topology and data foundation
Meter data joined to network model, with actual connectivity verified against records.
Build
LV visibility and voltage compliance, DER detection, or non-technical loss candidates ranked to investigation capacity.
Trial
Findings verified in the field, with topology corrections and loss investigations confirmed.
Operation
Continuous compliance monitoring, DER detection as installations grow, topology maintained rather than allowed to drift.
What you receive
A low-voltage network you can see, and a network model that matches reality.
AMI data assessment
What your meter data supports, and what is being discarded that should not be.
Topology verification
Actual customer-to-transformer and phase connectivity from voltage correlation.
Voltage compliance monitoring
Continuous across the population rather than sampled by spot measurement.
DER detection
Unregistered generation identified from meter signatures, for planning and safety.
Non-technical loss candidates
Ranked to investigation capacity, presented as leads rather than conclusions.
Flexibility dispatch support
Advisory, within network and contractual constraints treated as hard.
Is this the right starting point?
Worth being direct. There are situations in smart grid where custom AI work is the wrong spend, and those are listed rather than buried.
Worth doing if
- You have AMI data at half-hourly resolution or better and use it for billing only.
- Meters record voltage and your systems discard it.
- Low-voltage network topology records have drifted from reality.
- Unregistered distributed generation is affecting planning and voltage management.
- Non-technical losses are material and detection is manual.
Do something else if
- Meter data is monthly consumption only. That is a billing dataset, not a network one.
- AMI coverage is too sparse for population-level inference.
- You want automated switching or curtailment. That is a safety component programme.
- There is no field capacity to verify findings, so nothing can be confirmed or acted on.
Frequently asked questions
Marked up with FAQPage schema so these answers can surface directly in search results and inside AI assistant responses.
What can we actually do with smart meter data?
Considerably more than billing, and how much depends on resolution. Half-hourly consumption supports load profiling and forecasting; half-hourly consumption plus voltage is where low-voltage network analysis becomes possible — compliance monitoring, phase imbalance, connectivity verification, distributed generation detection. If your meters record voltage and your systems discard it, recovering that is one of the highest-return data projects in a distribution business.
Can meter data tell us our actual network topology?
Yes, and it commonly finds errors. Correlating voltage patterns across meters identifies which customers actually sit on which transformer and which phase, against records that have drifted over decades. Error rates in the low single-digit percentages are typical and they are enough to matter — for loss allocation, for planning, and for anything that depends on the network model being right.
How do we detect unregistered solar?
From the meter signature. Behind-the-meter generation produces distinctive patterns in consumption and voltage data, particularly around midday and on clear days. Detecting it matters for planning accuracy, for voltage management as penetration rises, and increasingly for network safety — knowing where generation exists changes what is safe to assume during an outage.
Is non-technical loss detection safe to deploy?
As an investigation lead, yes; as a conclusion, no. A false accusation of theft is a serious harm to a customer, so the output must route to a human investigation with the evidence attached rather than triggering an action. And set the candidate threshold to your actual investigation capacity — more candidates than you can visit is a backlog rather than a result, exactly as with inspection findings.
Where is the AI Act line in a distribution network?
At the point where a system acts rather than informs. Estimating state, flagging voltage excursions and recommending actions to a control engineer is advisory. A system that switches, curtails or reconfigures automatically is a safety component whose failure could cause physical damage, with high-risk obligations from 2 December 2027 and functional safety work attached. That decision belongs in week one, with your protection engineers in the room.
Related verticals
Organisations in smart grid usually share data, buyers or regulators with these. All fourteen are listed on the Energy & Utilities page.
AI for Power Generation and Utilities
Demand forecasting, network asset health, inspection analytics and outage response — advisory, not actuating.
Read more →AI for EV and Charging Infrastructure
Siting and utilisation, charger reliability, and smart charging under grid constraints.
Read more →AI for Renewable Energy
Generation forecasting, turbine and inverter health, and curtailment — where measurement beats modelling.
Read more →Tell us what the problem looks like.
Thirty minutes, no charge, no deck. We will tell you whether this is an AI problem, a data problem, or a process problem — and we will say when the honest answer is to buy something rather than build it.