Domain: Energy & Environment
Automatic detection of anomalies in energy consumptions thanks to the analysis of smart-meters data from final customers.
The challenge will facilitate the detection of anomalies in the energy consumptions of final customers thanks to the analysis of two types of data:
- Recorded events in smart meters. All issues happening to an electricity meter like strong DC field, opening/closing of a meter, over/under voltage, on-off, missing voltage, etc. It is sometimes difficult to determine if an event is justified or not. It is the point for this challenge to detect when anomalies occur. A justified event is the one which is a consequence of the work performed onsite (example: “meter cover opened” if there is a work order for the smart meter that day).
- Work orders for maintenance. The work plan for the company’s workers visiting on-site facilities. This data is recorded in a different system of the company (MOW).
We look to automatically identify patterns for anomalies combining both sources of data. Anomalies can be categorised as fraud patterns, errors on meters, errors on the distribution network, etc.
At this point the discovery of anomalies is done only a month later than it happened, and ELEK-LJ would like to switch this process to a real-time one.
The challenge has the following sample datasets available for download
The overall goal of the company is testing solutions which allow budget savings thanks to a reduction of fraud and of technical issues. Concrete outcomes pursued should be:
- Detect anomalies remotely from the measuring centre: We already detect remotely some anomalies (errors on meters and network) but we do not have application for automated discovering anomalies. For some anomalies (stealing of electricity) we cannot locate precisely where the anomalies are happening remotely from measuring centre.
- Avoid human intervention when detecting anomalies: we need automatisation.
- Faster detection of fraud patterns and errors in meters and the distribution network. We currently discover anomalies after a month. The goal is to discover all anomalies one day after they happened and or real time.