Context
Low‑voltage (LV) networks have traditionally been maintained reactively, with interventions triggered after customer impact.
The large‑scale rollout of communicating smart meters creates the conditions to shift toward predictive maintenance by providing high‑frequency, real‑time power‑quality data at scale.
Challenge
Detecting early signs of LV network issues is crucial to reduce customer impact and improve operational readiness.
The practical challenge is to convert large volumes of smart‑meter power‑quality events (e.g., outages and voltage deviations) into clear, prioritised intervention recommendations that operations teams can act on in time.
AI Solution
CartoLine analyses smart‑meter power‑quality data (including power cuts and over‑/under‑voltage events) to identify suspected anomalies on the LV network and prioritise recommended interventions for operations managers.
The approach supports a shift from reactive to predictive maintenance workflows, enabled by large‑scale real‑time data processing and field feedback loops.
Impact & evidence
CartoLine reports high field validation and actionable prioritisation outcomes:
- 3,000+ cases were addressed with CartoLine support over two years of use.
- >95% of anomaly suspicions were confirmed as real anomalies in the field
- >50% of priority recommendations were followed by an actual outage within 15 days
Implementation / Adoption
Delivering predictive maintenance on LV networks depends on data scale, integration, and operational adoption.
Key enablers include:
- Accompanied organisational change management to embed predictive maintenance workflows and ensure adoption by field teams
- Large‑scale deployment of communicating smart meters on the LV network (Linky in France — 35M+ devices)
- Real‑time data collection infrastructure capable of processing billions of events, with significant client‑side data preparation (aggregation and processing)
- Internal data‑science capability to develop and improve the anomaly‑detection logic as data volumes grow
- Automatic report prioritisation so operations teams focus on the most urgent cases, supported by an active field feedback loop to improve recommendation relevance over time
Data
Execution
CartoLine is fully developed in‑house, with a European patent granted. Deployment is supported through standard integration (APIs/SDKs/connectors) and requires significant client‑side data preparation to aggregate and process smart‑meter event data at scale.
Data
CartoLine relies on smart‑meter power‑quality event data (including voltage events and power cuts), combined with LV network topology and incident history, and strengthened through a field feedback loop to validate anomalies and improve recommendations over time.
Data quality, availability and scale
The required data is fully available, but it requires cleansing via aggregation and processing.
Enedis reports that, over the last two years, the application processed around 2 billion voltage spikes, 150M high‑voltage swings, 800M low‑voltage swings, and 300M power cuts, highlighting that very large smart‑meter data volumes are a key replication barrier without a national rollout.
Future enhancements
Extension to CartoLine HTA for pre-localisation of MV (20 000 V) aerial network faults (using fault detectors on the MV network); development of more detailed diagnostics with precise recommended actions; potential extension to other failure types beyond general voltage variations.
Additional information
Intelligence artificielle : Enedis remporte un prix international à l’ISGAN Award 2023 pour son outil CartoLine BT | Enedis
L’IA Générative au service de l’énergie : des premiers pas vers une révolution ? | Wavestone