GNGTS 2024 - Atti del 42° Convegno Nazionale

Session 3.3 GNGTS 2024 Fig. 1 – (a) Scheme of the network architecture for a 3-shots acquisiton array. The input data and output velocity models are in DCT-domain. (b) Training and validaton are represented by the blue and orange curves, respectvely. The analysis is confned to the inital 1300 epochs, as indicated by the black-dashed curve, untl which convergence is achieved. Results To illustrate the capability of the trained network to propose Vs-models, it is presented below an example for synthetc and feld data. The feld data was taken from the InterPACIFIC project which also includes Vs measurements in boreholes located at 10 m of the acquisiton array. The borehole data was utlized to validate the obtained results. The synthetc data was constructed using the same array of the feld data. The acquisiton setng consists of 48 receivers separated 1 meter, and 3 sources (two of-end and one middle shots). Synthetc data example: Figure 2 shows the network’s predicton on the synthetc example. The data was created using the same Ricker wavelet as the training dataset. In Figure 2a it is shown the true and predicted velocity model. Note that the trained network accurately predicts both the main features and magnitudes of the true velocity model. In Figure 2b, a comparison between the observed and predicted data is presented. It is noteworthy that the data exhibit a perfect match and do not show any signs of cycle-skipping. Fig. 2 – Synthetc example: (a) S-velocity model predicted using the NN at epoch 1300. (b) Observed and predicted data comparison .