Power anomaly detection

Nine defined use cases that cross-check billing, ELTEK measurements, equipment load and temperature to find the energy losses invisible to any single system — from phantom loads to dying HVAC.

9

energy-efficiency use cases defined in the catalog

4 sources

billing, power systems, equipment telemetry and weather — joined per site

Site-level

anomalies ranked by recoverable kWh, so the queue starts with the money

The idea

The production power suite already reconciles invoiced vs. measured energy. This program goes a layer deeper: it cross-checks measured vs. expected — what the site’s equipment, configuration and weather say consumption should be — and flags the gap. The catalog includes:

• Energy balance (ELTEK vs. equipment) — the rectifier reports more draw than the installed equipment can account for: phantom loads, unauthorized consumers, metering faults.

• Temperature-aware consumption — sites whose energy use doesn’t track outside temperature the way their cooling setup says it must.

• Cooling & HVAC inefficiency — air-conditioning that works ever harder for the same thermal result, caught months before it fails in a heatwave.

• Peer clustering — sites compared against their structural twins (same equipment class, same climate); the outliers are the audit list.

At a glance

Status: In Development

Deployed at: Catalog defined with CETIN SK

Domain: Energy

Built with: Billing data · ELTEK · Ericsson load · Temperature · BitSwan pipelines

Why it matters

Energy waste at sites is individually small and collectively enormous — and it hides precisely because each monitoring system sees only its own slice. The rectifier doesn’t know what equipment is installed; the bill doesn’t know the weather. Joined, the same data turns into a ranked queue of recoverable kilowatt-hours.

Where it stands

The catalog of nine use cases is defined and the required connectors (billing, ELTEK, Ericsson load) are already in production for the reconciliation suite — adding temperature and the detection pipelines is the build step. The pattern follows the proven anomaly-detection approach: per-site models, persistence filters, operator review.