Lateral changes in velocity about faults can give rise to fault shadow (FS) zones on time-migrated data volumes, which can result in structural interpretation artifacts in the fault trap reservoir. To address this issue, we have adopted a new reconstruction method of FS distortion structures based on a deep learning fully connected network (FCN). We use the 3D stratigraphic dip attributes to quantitatively delineate the extent of the FS zone. Then, we train a model to construct a nonlinear trend surface based on the structures of the stratigraphic reflectors that fall outside of the shadow zone. Finally, we use this nonlinear trend surface to compensate for the distorted structure within the FS zone. We calibrate our method using synthetic data and find that the method can accurately recover structural data within the FS distortion zone. We then test the effectiveness of our workflow by applying it to recover real FS distortion structures in the Pearl River Mouth Basin of the South China Sea. The results confirm that our method significantly reduces drilling depth errors in the FS zone. Compared to the traditional polynomial fitting method, the multilayer, multiparameter, and flexible nonlinear activation function of FCN is more capable of reconstructing nonlinear geologic structures in the FS zone. We find the FCN-based geologic reconstruction method to be efficient and effective for exploring potential structures in the FS zone and thereby in avoiding the risks of structural failure.