Distribution System Topology Identification for DER Management Systems Using Deep Neural Networks

Abstract

For DER management systems (DERMS) to manage and coordinate the DER units, awareness of distribution system topology is necessary. Most of the approaches developed for the identification of distribution network topology rely on the accessibility of network model and load forecasts, which are logically not available to DERMS. In this paper, the application of deep neural networks in pattern recognition is availed for this purpose, relying only on the measurements available to DERMS. IEEE 123 node test feeder is used for simulation. Six switching configurations and operation of two protective devices are considered, resulting in 24 different topologies. Monte Carlo simulations are conducted to explore different DER production and load values. A two-hidden layer feed-forward deep neural network is used to classify different topologies. Results show the proposed approach can successfully predict the switching configurations and status of protective devices. Sensitivity analysis shows that the positive and negative sequence components of the voltage (from DER units and substation) have the most contribution to discrimination among different switching configurations.