A Pilot Study on Data-Driven Misconfiguration Detection in Existent Power Distribution Systems

Distributed generation (DG) and electronic-based load units in power distribution grids introduce new challenges for distribution grid operators (DSOs). To maintain stable grid operations these devices often provide grid-supporting functionalities, whose configurations are difficult to monitor for
DSOs. Advanced smart grid automation approaches are essential to manage the numerous units that need monitoring with limited sensory equipment. In this work, misconfigurations of control curves at the installation of devices are to be monitored. The grid specifications and operational data of two existing grid segments were used. These data are used to simulate cases in which a device is configured correctly or misconfigured. The simulated grid operational data is then used as training data for Machine Learning (ML) algorithms, which are used to make
statements about whether the measured operational data stem from operation under a misconfigured control curve or not. The DSO confirmed the results for one case under scrutiny, for the other no clear statement could be made. The results show that the presented approach can serve the needs of DSOs and offer a viable solution for automating misconfiguration detection in distribution grids, even though further field testing is needed.