Machine learning for computation of droop controller coefficients to improve the frequency nadir and ROCOF of a stand-alone microgrid

Load Frequency Control (LFC) is vital to keep up the nominal frequency in power systems. LFC initiates control actions to bring back the grid frequency to the desired level when it varies from the nominal value. Addition or removal of load causes deviation in frequency of the power system. This is taken care by inertia present in conventional generators. Inverter based renewable power systems accommodated with Distributed Generators (DGs) have less inertia when compared to conventional synchronous generator based power system. Less inertia in DGs have to be handled by inverter controllers to minimize the frequency deviation. Designing of droop controller for DGs necessitates the determination of proper coefficients for the controller. This research discusses a machine learning methodology that use the Stochastic Gradient Descent (SGD) algorithm to compute the droop coefficients. The controller designed using these coefficients is incorporated to ameliorate the system voltage and frequency profile. The droop coefficients computed using SGD are implemented in the controller and tested in a stand-alone microgrid system powered by solar photovoltaic panel. The droop coefficients assumed as + 0.05 in the conventional controller are replaced with the coefficients determined using the proposed machine learning techniques. The results shows that the Rate of Change of Frequency (RoCoF) and frequency nadir of the system has been improved when compared with conventional droop controller.