ML Assisted T‐Model Feeder Reduction and Convex Fast Reconfiguration Method for Active Distribution Networks With DERs

ABSTRACT Deep electrification of energy systems and massive integration of distributed energy resources (DERs) push the growth of radial distribution networks (RDNs), requiring innovative methods to deal with the increased complexity and enhance current operational tasks. Network reconfiguration (NR) is one of the tools researched and developed during the last 50+ years for large distribution sizes for more robust operations. NR is vital for reliable operation and demands swift, accurate decision‐making, prompting distribution utilities to adopt faster solution methods. In this sense, a method is proposed in this paper with the objective of minimising the execution time of reconfigured large RDNs. A T‐Model feeder reduction method, a mapping artificial neural network (ANN) method to map feeder demand and power injections of DERs to T‐Model parameters, a reduced RDNs representation using T‐Models of feeders, and a network reconfiguration method leveraging this reduced network are proposed. The proposed T‐Model reduces the number of nodes in RDNs considering connected loads and DERs by around 55%, which leads to significantly reduced network representation and thereby execution time when reconfiguring the RDNs. The proposed method is tested on several systems, including the IEEE 123‐Bus network. The execution time is reduced by up to 74.53% while providing the accuracy of at least 97.17%. This method scales well and performs better for active larger RDNs.