DEEP LEARNING-BASED BINARY CLASSIFICATION OF ISLANDING CONDITIONS IN A HYDROGEN ENERGY-BASED DISTRIBUTED GENERATION SYSTEM

Abstract

This paper presents a deep long short-term memory (DLSTM) with a binary-label classifier method proposed for binary classification of islanding and non-islanding events in a hydrogen energy-based distributed generator (DG) system. Deep learning (DL)-based method eliminated the lack of performance of conventional intelligent islanding detection methods that uses feature extraction, feature selection, and event classification. Besides, the proposed method has provided the need for a processing-intensive filtering process to reduce noise from the signal. The proposed islanding detection method has a 98.33% accuracy rate under no-noise, and 97.66% high-level noise conditions. In the proposed method, the non-detection zone (NDZ) is almost zero, and the detection time is under the defined IEEE 929-2000 standards. Experimental and simulative data results show that the LSTM-based islanding detection method outperforms the algorithms in recent studies in terms of noise immunity and accuracy. © 2022 Proceedings of WHEC 2022 - 23rd World Hydrogen Energy Conference: Bridging Continents by H2. All rights reserved.