Phase Identification

Adaptricity offers a Phase Identification service that can identify the phases of loads based on SmartMeter power measurements linked to the loads and 3-phase transformer power measurements. Currently, the phases can only be identified for single-phase loads.

The phase identification algorithm is based on a machine learning model. For each transformer circuit, a machine learning model is trained based on the grid topology and a set of SmartMeter and transformer measurements. This machine learning model is then used in subsequent grid data updates to infer phase information for the loads contained in the grid.

Prerequisites

• There are some requirements for your data:

  • SmartMeter power measurements
  • 3-phase transformer

• Access to our Phase Identification service

Testing with Example Data

• In the "Examples" project of Adaptricity, open the project Settings and go to the "Data Integration" tab.
• Click "Edit API endpoints" and fill in the "Phase identification" URL and credentials provided to you.
• Select the "Uster Unknown Phases" grid as the grid to update.

• Further down in the settings, there are more Phase Identification-specific options. In particular, you can choose if the data for training the machine learning models should be generated using a power flow computation or just aggregate the measured powers. The latter will neglect grid losses and therefore the trained phase identification model will forfeit some of its accuracy, but it will also significantly reduce the time required to generate the training data.

• Save the settings and start a data update.

It will take a while to generate the training data for all the transformer circuits and train the corresponding models. Once the models are ready, they will be used in the subsequent data updates, adding phase information to any loads in the grids that don't come with any phase information assigned.

In the Data Update Report you will see that the phase identification has been performed successfully, and you will get warnings if the models failed to train to acceptable accuracy (which typically happens for grids for which insufficient data is available, or if the measurements across phases are too similar):

In the example data, the meters get assigned deterministic phases so that it is easy to assess the accuracy of the phase identification models. Loads have meter IDs of the form load_meter_n for integer numbers n.

• If $n\equiv 0(\mathrm{mod}3)$, the designated phase is A;
• if $n\equiv 1(\mathrm{mod}3)$, the phase is B;
• if $n\equiv 2(\mathrm{mod}3)$, the phase is C.

Please note that the phase identification models are probabilistic models, so they might not predict all phases with a 100% accuracy.