Abstract: Fault-controlled reservoirs represent important targets for the exploration and exploitation of unconventional hydrocarbons. Accurately identifying multi-scale faults directly determines the reservoir productivity and economic benefits. Although seismic attributes are commonly used for fault prediction, a single attribute struggles to characterize multi-scale faults simultaneously. To address this challenge, this study proposes a novel method for fault identification based on varying-scale decomposition and hierarchical fusion of seismic attributes. First, multiple optimized seismic attributes for fault identification are extracted, and multi-scale fault responses are separated from noise components using a multi-scale decomposition algorithm based on the rolling guidance filter (RGF). Then, distinct fusion models tailored to small-, medium-, and large-scale faults are established. Specifically, a Gaussian-weighted absolute value maximization (GAVM) rule is adopted to achieve adaptive weighted fusion of small-scale attributes, thus enhancing weak fault responses. The weighted local energy is combined with a weighted sum of eight-neighborhood-based modified Laplacian (WLE-WSEML) for medium-scale attribute fusion, thereby highlighting the spatial distribution of medium-scale faults. A stacked sparse autoencoder (SSAE) is used to conduct deep non-linear fusion of large-scale attributes, achieving fine-scale characterization of large-scale fault structures. Finally, multi-scale fusion results are integrated to form comprehensive attributes for fault characterization. Tests on synthetic and field seismic data demonstrate that the proposed method in this study can effectively overcome the issues of feature redundancy and information loss inherent in traditional attribute fusion algorithms. The proposed method significantly improves the resolution of minor- and small-scale faults and yields clearer boundaries and enhanced noise-resistant performance for large-scale faults. Therefore, this study provides reliable technical support for accurate prediction of complex fault-controlled reservoirs.