Applying Deep Learning-based concepts for the detection of device misconfigurations in power systems

Abstract

The electrical energy system is undergoing major changes due to the necessity for more sustainable energy generation and the following increased integration of novel grid-connected devices, such as inverters or electric vehicle supply equipment. To operate reliably in novel circumstances, as created by the decentralization of generation, power systems usually need grid supportive functions provided by these devices. These functions include control mechanisms such as reactive power dispatch used for voltage control or active power reduction depending on the voltage. As the main contribution of this work, an approach for the development of the detection of misconfigured (e.g., wrongly parameterized control curve) grid devices using solely operational data is proposed. By generating and analyzing operational data of power distribution grids, a Deep Learning-based approach is applied to the detection problem given. An end-to-end framework is used to synthesize and process the data as well as to apply machine learning techniques to it. The results offer insights into the applicability and possible ways to improve the proposed solution and how it could be employed by grid operators. The findings show that DL methods, in contrast to traditional machine learning, can be used for the problem at hand and that the framework developed offers the necessary tools to fine-tune and scale the solution for broader usage.