2026· IEEE Transactions on Instrumentation and Measurement· Vol 75, pp. 9007315-9007315· 0 citations· 42 references
Abstract
Voltage sags can lead to substantial economic losses, making accurate source location the critical first step toward effective mitigation. Given that the absence of detailed network parameters constrains the use of impedance-based approaches, this study explores deep-learning-based alternatives for source location. To address the challenge of limited training samples—arising from sparse measurements in distribution networks—this work proposes a knowledge transfer framework employing a collaborative training strategy for voltage sag source location (VSSL). An interpretability-enhanced graph convolutional neural network (IIGCN) is developed, using the voltage magnitude depth difference (VMDD) derived from recorded waveforms as input features. A cross-region knowledge transfer framework is then introduced to capture general mapping relationships between VMDD and corresponding source positions. IIGCN models for different subregions are trained collaboratively, with parameter exchange and compensation mechanisms ensuring robust performance despite limited training data. The proposed approach is evaluated on both synthetic and real-world systems, achieving about 85 % accuracy under stringent training conditions. Comparative analyses demonstrate that the method consistently outperforms existing techniques, delivering superior VSSL performance across diverse application scenarios.
Results show that the proposed LSTM-DND framework is effective for single-fault location estimation under simulated operating conditions, although further validation with mixed-noise and field-recorded data is still required.
A. M. Saidina Omar, M. K. Osman, Siti Sarah Mat Isa et al.· Journal of Renewable Energy...· 0 citations
Comparative benchmarks against mainstream state-of-the-art machine learning approaches verify the superior efficacy and prediction precision of the presented spatiotemporal feature fusion neural network.
Yi Ran, Yong-Guang Ding, Yan-Ming Mai et al.· Frontiers in Energy Research· 0 citations
Deep Transfer Learning for State Estimation (DeepTL4SE), which combines Physics-Informed Data Generation (PIDG) with supervised transfer learning, is presented, which produces a point estimate of the current electrical state.
Zhen-De Zhang, Xianglong Li, Shengxin Kong et al.· Energies· 0 citations
An uncertainty-aware encoding and decoding framework that fused stochastic mixed-integer quadratic programming approaches, with deep neural network (DNN) classification was developed, showing the suggested approach was able to provide voltage regulation actions that were close to optimal while achieving significant com...
A fault-location method that integrates distributed-parameter physics-based features with a convolutional block attention module-based convolutional neural network (CBAM-CNN) to demonstrate high fault-location accuracy within the scope tested and provide preliminary evidence of applicability to field recordings.
Faguang Chen, Meng-Zhou Li, Shibin Fan et al.· Electronics· 0 citations
A knowledge-guided multimodal learning framework that combines raw electrical sequences with temporal, statistical, frequency-domain, and contextual descriptors and integrates their complementary representations through attention-based multimodal fusion is developed.