A new robust kernel ridge regression classifier for islanding and power quality disturbances in a multi distributed generation based microgrid

This paper presents a new robust and dimensionally reduced kernel ridge regression classifier (RKRR) for classification of islanding and power quality (PQ) disturbances in a distributed generation based microgrid. A grid connected active distribution network has been considered in this paper where inconsistency in PV generation is emphasized. To design an effective PQ monitoring solution the multi PV based microgrid is conditioned to operate in grid synchronous mode (i.e. according to IEEE 1547). Further to identify the detection threshold and extract relevant features for classification from the noisy voltage signals in a microgrid for different islanding and power quality disturbances, two advanced signal processing techniques like variational mode decomposition (VMD) and empirical wavelet transform (EWT) have been applied. To improve the reliability of the KRR and to make it robust under noisy conditions, a weight loss matrix has been derived and incorporated in the new formulation. To reduce the dimensionality of the KRR during training a lower set of randomly chosen data samples from the training set known as support vectors are used. This approach reduces the size of the kernel matrix used for inversion and produces reasonably accurate classification and results in a substantial reduction in execution time.