AN ARTIFICIAL INTELLIGENCE APPROACH FOR THE IDENTIFICATION OF THE NORMATIVE BEHAVIOR OF DECENTRALIZED GENERATORS IN AN ISOLATED NETWORK

A sustainable, intentional, isolated grid operation could be achieved by allowing the in-feed from the available decentralized generators (DGs), like rooftop PV systems, together with the conventional ones. A concept for the emergency islanded operation was proposed based on the normative power reduction governed for higher frequencies. However, the exact amount of power reduction from the DG side is unknown since its normative requirements, intended for parallel connection, have changed within the past years in Germany, strongly differing on the frequency characteristic. One method estimates the aggregated power versus frequency behavior based on the commissioning date, as it corresponds to a valid grid code. Still, some discrepancies are observed when compared to field measurement data. This work proposes a neural network algorithm to identify the penetration level of each frequency characteristic inside the islanded grid and predict its power profile in the over-frequency region. Results show a promising capability to predict profiles closer to the recorded measurement data.