Data-driven topology detector for self-healing strategies in Active Distribution Networks

Abstract

The integration of distributed energy resources requires the implementation of control and automation functionalities in distribution networks, which allow them to operate in a more flexible, efficient, and reliable way. The operation of these functionalities causes topological changes on the network that must be identified since these affect protection, volt/var control, and state estimation, among others. This paper presents a data-driven topology detector for self-healing strategies in Active Distribution Networks (ADN). This approach uses machine learning (ML) techniques to obtain a trained model as a topology detector. The ML models are integrated into the Intelligent Electronic Device (IED) of ADN so that using the voltage and current signals measured locally determine the network’s topology. A features selection and tuning technique based on a multiverse optimizer are proposed to improve the ML model accuracy. This approach allows it to be implemented in decentralized architectures since each IED detects the system’s topology from local measurements and does not depend on the availability of the communication system. The proposed topology detector was validated on a modified IEEE 123 nodes test feeder considering six topology changes, five DER outages, and five load variations. The results obtained show accuracy values above 96%, which evidences a highlighted potential for real-life applications.