Data-Driven Misconfiguration Detection in Power Systems With Transformer Profile Disaggregation

Abstract

Rapid and necessary changes in the energy sector are leading to the rise of new, decentralized devices for generation and consumption in the electrical distribution grid. Such devices are inverter-connected photovoltaic (PV) generators, heat pumps (HP), or electric vehicle supply equipment (EVSE). These new components make the power grid operation more difficult as they display volatile behavior and therefore also need to provide grid-supporting functionalities. Distribution System Operators (DSOs) need to make sure these grid-supporting functionalities are performed correctly, in order to guarantee a safe and reliable operation of the grid. However, especially the low voltage distribution grid is still ill-equipped with sensors and therefore difficult to monitor. This contribution, therefore, presents a data-driven application for detection of misconfigurations using the data available at metering points of substations and selected voltage measurement points in combination with a transformer load profile disaggregation approach. The assembled application outlined is both functional, scalable, and easy to integrate into current monitoring schemes. Such a monitoring application has not been designed yet and is therefore novel. The data used were collected in a life-like laboratory setup and recreated using simulations in order to be able to test and validate both the detection as well as the disaggregation method. Two monitoring use cases of control functions are considered; the first one is a reactive power control of PV inverters, and the other one is a Demand Side Management (DSM) control of loads. The results presented offer insights into both the quality and performance of the application assembled. The best achieved performance is a F-score of 0.83, which also serves as a future benchmark as there are no comparable results to be found in literature. Furthermore, the influences of the individual methods of the approach are explored as well. The conclusions drawn show that a functional monitoring solution of reasonable reliability can be implemented using the methods presented and tested here. The application can serve as a decision support tool for DSOs requiring only minimal adjustments to the sensing infrastructure.