Gaussian process regression based inertia emulation and reserve estimation for grid interfaced photovoltaic system

Accurate power reserve estimation for a Photovoltaic Generator (PVG) is of paramount importance to combat frequency changes in a smart grid. Standalone PVG lacks inertia, or an internal power reserve due to power electronic converter grid-interface. Operating a PVG at deloaded percentage of its maximum power capacity mimics an internal power reserve, simulating the Automatic Generation Control (AGC) feature of synchronous machines. Thus, a deloaded PVG releases or absorbs the reserve according to the frequency variations for the grid stability. Moreover, an efficient switching between various reserves during grid operation is required. The common reserve estimation technique is to apply PVG manufacturer's specification based deterministic approach. In this work, we compare the deterministic modeling results with a statistical learning model of Gaussian Process Regression (GPR). The GPR model is trained by dataset of PVG maximum power values evaluated by load line analysis in a simulation, according to the irradiance and historical temperature of Abbottabad, Pakistan. The trained model performance is compared with the deterministic model in a simulation, where the PVG is saturated to turn on a synchronous generator. Time difference of turning on the backup generator between GPR model and deterministic modeling validates the importance of accurate reserve estimation.