ABSTRACT Accurate identification of transmission line fault causes is crucial for enhancing grid reliability and improving fault handling efficiency. Although deep learning techniques have advanced fault cause identification, existing methods still exhibit deficiencies in capturing global contextual information, suppressing noise interference and effectively integrating multi‐perspective information such as weather and temporal data. To address these limitations, this paper proposes a multi‐perspective fault identification framework integrating wavelet attention enhancement and adaptive correlation fusion. The method initially constructs an energy matrix through a multi‐scale continuous wavelet transform, utilizing prior knowledge of fault‐related frequency components to guide attention weight allocation, thereby enhancing critical signal features while suppressing noise. Subsequently, a learnable weight matrix is introduced to explicitly model the differential contributions of the static features—including weather, date and time—to the fused features, enabling adaptive multi‐perspective fusion. Furthermore, a cross‐attention mechanism is designed to capture deep interactions between dynamic fault recording features and the static contextual features, achieving cross‐perspective deep integration. Experimental results on a real‐world fault dataset demonstrate the superior performance of the proposed method, attaining an accuracy of 94.29% and a recall of 93.10%. These values exceed those of the second‐best configuration by approximately 0.86 and 3.1 percentage points, respectively. Multiple other evaluation metrics also significantly outperform existing baseline methods.
Zhang et al. (2026) studied this question.