Deep Learning-based Real-time Switching of Reconfigurable Microgrids

This paper proposes a new approach for finding the optimal switchings of reconfigurable microgrids (MGs) based on a deep learning technique. As the load and power generation of units vary with time, the network reconfiguration depends on both the current and previous status of load demand and generation units. To this end, the gated recurrent unit (GRU) algorithm, a time series model, is employed to solve the network reconfiguration. The GRU architecture is designed to learn the network topology characteristics, e.g., power injection and line impedance. Finally, the proposed technique is examined with the IEEE 33 microgrid test system. Results show that the deep learning-based technique is able to make accurate reconfiguration decisions in real time.