Three-Phase Voltage Fault Diagnosis Based on Parallel Dual-Stream Attention Network and Cross-Modal Fusion

In modern power systems, voltage quality is directly related to the reliability and stability of power supply. Abnormalities in amplitude, phase, frequency, etc. may occur during the transmission and distribution of three-phase AC voltage. If these faults are not detected and diagnosed in time, they may lead to equipment damage or system instability. Traditional methods rely heavily on manual feature extraction and signal processing algorithms, making it difficult to maintain high recognition accuracy under complex working conditions. To address the above issues, this paper proposes a deep learning model that integrates wavelet transform and dual-stream attention. This method differs from the simple series structure in previous studies. Instead, it adopts a parallel dual-stream architecture: the temporal stream performs denoising and feature enhancement on the original voltage signal through discrete wavelet transform, and then captures the dynamic temporal dependency of faults through a bidirectional long short-term memory network (BiLSTM) and an attention mechanism. The time-frequency current converts the signal into a time-frequency graph through continuous wavelet transform and inputs it into a multi-branch convolutional neural network (CNN) and a convolutional attention module (CBAM) to mine frequency-domain discriminative features at multiple scales. To further integrate the dual-stream information, this paper designs a cross-modal attention mechanism, which dynamically aggregates the key information in the time-series stream under the guidance of the time-frequency stream features, forming a robust comprehensive fault characterization. Finally, this paper conducts experiments using a dataset containing normal three-phase AC voltage data and fault data of three types of faults. The experimental results show that this method outperforms traditional methods and mainstream deep learning models in terms of fault classification accuracy and robustness.