Table 2 A DFNN was designed with three hidden layers, each comprising nine neurons. The model used the Tanh activation function and was optimized using the CG algorithm, with L2 regularization (0.06) applied to improve generalization and prevent overfitting. Input seismic attributes were normalized to the range [− 1, 1], and model validation was performed using a leave-one-well-out strategy to ensure robust prediction of the 3D acoustic impedance volume.

From: A physics-informed deep learning approach for 3D acoustic impedance estimation from seismic data: application to an offshore field in the Southwest Iran