Network Monitoring (Smart Grids)
A distribution network operator today manages between 10 and 35 million smart meters. Each meter transmits data every 10 to 30 minutes: consumption readings, load curves, technical alerts, connection status. This represents several billion measurement points per day. Operations teams must detect in real time voltage drifts, potential fraud, insulation faults, phase imbalances.
Regulatory obligations (TURPE, European network codes) impose strict response times and complete traceability. Meanwhile, legacy systems: designed for half-yearly readings: become saturated. Conventional relational databases struggle beyond a few million records per table. Data teams spend more time maintaining fragile pipelines than analysing.
And when a network incident occurs, reconstructing the chronology of events takes hours instead of minutes. Fine control of smart grids is not a luxury: it is an operational and regulatory obligation.
This solution belongs to the Data, Master Records & Reporting family.
Challenges holding back your performance
Current approaches demonstrate their limitations when confronted with data volume and velocity. Traditional relational databases (your database, your database in standard mode) collapse when exceeding several hundred million temporal rows. Teams cobble together overnight extractions to flat files, followed by Excel macros for analysis.Result: detection delays of 24 to 48 hours for anomalies that should be identified within minutes. Generic BI tools (your management tool, your visualisation tool) are not designed for queries on sliding temporal windows of several terabytes. Generic service providers propose oversized architectures, calibrated for web giants, with infrastructure and maintenance costs incompatible with regional utilities’ budgets.
Our Technical Approach
Our approach rests upon a clear separation between ingestion, storage and analysis. Ingestion utilises distributed message brokers capable of absorbing peaks of several hundred thousand events per second: typically during synchronised readings at midnight. Storage employs databases specialised in time series, designed to efficiently compress repetitive data and enable queries across time ranges without scanning entire tables.
Anomaly detection combines explicit business rules (voltage thresholds, consumption variances) with statistical models that learn the normal profiles of each delivery point. Alerting is hierarchical: operators only see what requires action, not a continuous stream of false positives. Governance incorporates encryption of personal data, GDPR-compliant pseudonymisation, and audit trails for CRE controls.
Network Supervision (Smart Grids)nnTime-Series Fundamentals
Optimised storage for timestamped data, with retention policies (Downsampling).
Massive Ingestion (Kafka)
Queuing of messages to handle peaks in readings without data loss.
Anomaly Detection
Algorithms identifying sudden voltage drops or overconsumption (leakage).
Operations Dashboards
Cartographic visualisation of network nodes (Grafana, bespoke tools).
Governance and Security
Encryption and anonymisation of domestic consumption readings.
Proactive Alerting
Creation of rules triggering notifications (SMS/Webhook) to on-call teams.
Technical Architecture
Several sources, one master record to arbitrate between them, then analysis: Core HR & Unified Repository and Observatories & Open Data stem from this same chain.
- Network Supervision (Smart Grids)nnTime-Series FundamentalsOptimised storage for timestamped data, with retention policies (Downsampling).
- Massive Ingestion (Kafka)Queuing of messages to handle peaks in readings without data loss.
- Anomaly DetectionAlgorithms identifying sudden voltage drops or overconsumption (leakage).
- Operations DashboardsCartographic visualisation of network nodes (Grafana, bespoke tools).
- Governance and SecurityEncryption and anonymisation of domestic consumption readings.
- Proactive AlertingCreation of rules triggering notifications (SMS/Webhook) to on-call teams.
Smooth and frictionless integration
We work with your existing data collection systems (concentrators, Itron head-ends, Landis+Gyr, Sagemcom). The ingestion connects in parallel to your current data flows, without modifying the collection chains validated by your telecommunications teams. The dashboards integrate with existing portals or operational tools (GE PowerOn, Schneider ATS). Customer and asset repositories remain in your information systems: we consume them in read-only mode.
No forced migration, no replacement of your dispatch tools.
Network Traffic Audit and Volume Analysis
Mapping of data sources, analysis of load peaks, identification of current bottlenecks. Deliverable: realistic sizing.
Network Supervision (Smart Grids) Architecture and Targeted POC
Deployment on a limited perimeter (one region, one type of meter). Validation of ingestion and querying performance on real data.
Industrialisation and integration
Extension across the entire asset base, connection to existing business tools, implementation of alerting rules and detection models.
Network Supervision and Autonomy
Training of operations and data teams, operational documentation, transition to support mode with progressive skills development.
Measurable results for your organisation
- Network anomaly detection time reduced from 24-48 hours to less than 15 minutes
- 60 to 80% reduction in storage volume through native time series compression
- Ingestion capacity validated at 500,000 events per second on standard infrastructure
- False positive rate on fraud alerts reduced by 70% following 6 months of learning
- Network incident reconstruction time reduced from several hours to a few minutes
- GDPR compliance and CRE requirements documented and auditable
Clarifying your decision-making
What is the infrastructure cost compared to our current systems?
On the monitoring projects Clarendis has delivered, infrastructure cost runs 30 to 50% below a conventional relational database approach, mainly through compression and the absence of oversized servers. Return on investment is measured above all in engineering time saved on pipeline maintenance.
How do you manage the transition without interrupting our operations?
The transition to the new monitoring platform runs in parallel with the existing system for two to three months in our deployments. Both systems coexist, which allows data consistency to be validated and teams to be trained. The switchover is progressive, by geographical area or by meter type.