Deep Learning-based Islanding Detection Method for Droop-controlled Grid-Forming Inverters

This paper proposes a deep learning-based intelligent technique to overcome islanding detection challenges in droop controlled-based grid-forming inverters (GFM). Because of the effect of droop parameters of the power control on the non-detection zone (NDZ), the conventional islanding detection methods (IDM) are ineffective in GFM inverters. In addition, the active IDMs have opposite functionality to the working operation of the GFM inverters, which may contradict each other. Thus, there is a need to develop efficient IDMs capable to operate effectively even with unconventional droop gains and help maintain GFM functions. The proposed IDM is a two-stage process: (i) The first stage is to extract some distinguishable features from the root-mean-square (RMS) voltage and current signals. These signals are analyzed to determine the total harmonic distortion (THD) using the fast Fourier transform (FFT). (ii) In the second stage, a deep learning classifier based on a long-short-term memory (LSTM) recurrent neural network (RNN) is implemented to identify the islanding condition. The efficacy of the proposed IDM is tested and validated in the real-time RSCAD test system and IEEE-13 node feeder, respectively. the proposed IDM performance is evaluated even during weak grid conditions for various grid impedances. The results verify that, compared to other intelligent classifiers and previously reported techniques, the proposed IDM has remarkably high accuracy (100%) and reduced NDZ (2%) within the 35ms detection time.