Design of a Novel Adaptive Weighted Multiview Deep Neural Network for Enhanced Islanding Detection in Smart Distribution Systems

The extensive adoption of renewable energy resources has introduced a challenge of unintentional islanding in distributed generation (DG) systems. If the islanded DG is not promptly isolated from the system, it may lead to potential damage to the DGs. This paper develops a novel adaptive weighted multiview deep neural network (AWM-DNN) model for enhanced islanding detection in smart distribution networks. The proposed approach extracts distinct views from three-phase(3- $\phi $ ) voltage signal—using time-domain (TD), frequency-domain (FD), and time-frequency-domain analysis (TFD). The proposed method employs synchrosqueezed wavelet transform to extract frequency-domain and time-frequency domain features from the modal voltage signal. These views are then adaptively weighted and provided to a combination of gated recurrent unit and fully connected layer to classify the islanding events in the microgrids. The AWM-DNN based islanding detection scheme (IDS) uses the multiview learning framework to accurately classify islanding and grid-tied scenarios. Exhaustive simulations conducted on the modified IEEE 34-bus system and International Electrotechnical Commission (IEC) microgrid show that the proposed IDS surpasses traditional methods in terms of detection accuracy, non-detection zone (NDZ), and robustness to noisy data. Comparative analysis further highlights the superiority of model superior generalization capabilities and resilience to uncertainties in measurement.