Research on Identification Method of Gypsum Salt Layer and Subsalt Faults Based on Transformer Enhanced U-Net++ Network

The accurate identification of gypsum salt layers and sub salt faults is a key challenge in the exploration of complex oil and gas reservoirs. Due to the strong heterogeneity and plastic flow characteristics of the gypsum salt layer, it has a strong shielding and blocking effect on the underlying strata, resulting in weak seismic reflection energy and unclear displacement of faults events. Traditional manual interpretation methods and conventional deep learning techniques are difficult to identify directly and effectively. In response to this issue, this article proposes a Transformer-enhanced U-Net++ based method for identifying gypsum salt layers and intelligent prediction of salt faults, using a two-stage strategy of “identifying the cover first and then searching for faults”. Firstly, a multi-head self-attention mechanism is introduced in the U-Net++ network architecture to enable the network to establish long-range spatial position dependencies, and combined with the ReLU loss function, to achieve accurate identification of the boundary of the gypsum salt layer. Then, using the extracted bottom interface of the gypsum salt layer, the geometric continuity of the bottom interface is analyzed by calculating the gradient field and curvature field, and the development positions of the subsurface faults are indirectly determined based on the faulting and deformation caused by fault activity. The test results of the actual 3D seismic data in the work area show that the model training loss curve is stable and convergent, and the discontinuous 3D shape of the bottom interface of the gypsum salt layer clearly reflects the influence of faults on the overlying strata; the faults distribution obtained based on the continuity analysis of the bottom interface is in good agreement with the results of manual interpretation, while the effect of neural network fault recognition directly on seismic data is significantly poor, with serious omissions and fragmentation problems. The comparative experiment fully verified the effectiveness and superiority of the strategy proposed in this paper. This method integrates the identification of gypsum salt layers and fault prediction into a unified intelligent interpretation framework, providing a reliable technical approach for predicting complex subsurface faults in gypsum salt layer development areas.