To obtain detailed, direct breakdowns of energy consumption, individual consumers’ shares of total consumption must be identified. The objective of the Bundesamt für Energie-funded (BFE, Federal Office of Energy) «SmartNIALMeter» project is to use non-intrusive appliance load monitoring (NIALM) methods for this purpose. This approach is based on the fact that every electricity consumer generates a specific signature in the power grid, thus every activation leaves behind an electric «finger print».
The project’s NIALM approach is based on technologies used in the field of machine learning. Deep learning algorithms such as «recurrent neural networks» and «convolutional neural networks» in particular have produced promising findings thanks to the astonishing capabilities which these artificial neuronal networks have to recognize and extract patterns.
The project is scheduled to run for two years and started in September 2017 in cooperation with partners Energie Thun, EKZ, Landis+Gyr, smart-me and Bits 2Energy-Lab. The BFE is providing most of funding. Test data which can be trained and validated using the algorithms is currently being gathered in several households. In addition to this, a cloud infrastructure is being established to facilitate field testing of a system for some 50 pilot households in a later project phase.