Application of seismic modelling and deep neural network for 4D seismic inversion

In this study, a novel methodology for time-lapse seismic data inversion is proposed. Utilizing a deep learning (DL) model that has been trained on synthetically generated data, the methodology provides maps of saturations and pore pressure differences across the field. The synthetic data set generation is based on reservoir modelling results, well logs, geomechanical, and geophysical data from the target field. To test the efficacy of the proposed methodology, it was applied to a mature field within the Caspian basin, where a substantial level of production activities has already taken place. The obtained inversion results are qualitatively consistent with fluid production data and conventional 4D interpretation results. Additionally, the proposed methodology facilitated quantitative analysis: provided water changes values were validated using the water cut values in wells. The 4D inversion results provided can be used to separate the effect of saturations and pressure changes, improving the field development.