Industry Reference | Energy and Utilities
Part of the industry reference. Restructured from the energy deep dive for quick lookup. Differential privacy is a natural fit for energy use cases (grid planning, demand forecasting) — see the privacy models reference for DP parameter guidance.
Core challenge: NILM inference
Non-Intrusive Load Monitoring (NILM) can infer household activities from electricity consumption patterns at 15-minute granularity. Even basic smart meter data reveals behavioral information.
What NILM reveals
| Pattern | What it indicates | Resolution needed |
|---|---|---|
| Morning spike timing | Wake-up time, work schedule | 15-min |
| Midday sustained load | Work-from-home vs. away | Hourly aggregation reduces this (~70% accuracy vs >90% at 15-min) but does not eliminate it |
| Evening cooking pattern | Meal times, household size | 15-min |
| Continuous baseline loads | Medical equipment, always-on devices | Any granularity |
| EV charging signature | Vehicle ownership, charging schedule, battery size | 15-min |
| Extended zero consumption | Vacancy (burglary risk) | Daily |
Key finding: 5 consecutive half-hourly readings re-identify 90%+ households. Even rounding to 100W, 7 readings still re-identify 40%+. Re-pseudonymization doesn't help — deep learning can re-link across pseudonym rotations.
Data type treatment
| Data type | Granularity | Re-ID risk | Treatment |
|---|---|---|---|
| Smart meter electricity | 15-min intervals (96/day) | Very high | Temporal aggregation (15min → hourly → daily); DP for aggregates |
| EV charging sessions | Per-session (timestamp, location, energy, duration, charging curve) | High — location sequences + charging curve reveal vehicle model + movement | Location generalization, temporal bucketing, session-level pseudonyms, DP |
| District heating/cooling | Hourly or 15-min | High — thermal consumption arguably more privacy-sensitive than electricity in cold climates | Same as electricity; smaller datasets mean weaker k-anonymity |
| Green energy / carbon tracking | Per-facility, per-hour | Medium–high for small facilities | Aggregate to organizational level or temporal windows; DP for public stats |
Smart meter regulatory landscape
| Regulation / standard | Jurisdiction | Key requirement |
|---|---|---|
| EU Electricity Directive 2019/944 Art. 23 | EU | Smart metering data under GDPR; data management delegated to member states |
| EU Implementing Regulation 2023/1162 | EU | Interoperability requirements; non-discriminatory access with consent |
| BSI TR-03109 (Smart Meter Gateway) | Germany | Hardware-enforced data minimization: different granularities for different recipients |
| Netherlands metering law | NL | Consumer can refuse smart meter; 15-min data only with explicit consent |
| EU Energy Efficiency Directive 2023/1791 | EU | All district heating/cooling meters remotely readable by 2027 |
| OCPI / OICP | EU (EV roaming) | Persistent pseudonyms enable cross-operator tracking — privacy risk |
| ISO 15118 Plug&Charge | International | Cryptographic identity chains for EV authentication |
Common client starting position
Smart meter rollout creating new privacy obligations. DSOs (Distribution System Operators) are the primary data holder. Analytical uses (grid planning, demand forecasting) all work on aggregates — DP is the natural fit.
Aggregate to hourly, generalize to substation/feeder level, suppress groups below threshold. This covers most grid planning and demand forecasting use cases without the full four-zone architecture.