A Novel GAN-Based Technique Towards Smarter Seismic Inversion

Abstract Seismic surveying is a crucial and effective method for exploring hydrocarbon reserves, continually attracting the attention of the upstream oil and gas sectors. Seismic inversion, a pivotal technique for investigating the properties of underlying geological strata, presents a significant problem. This work diverges from the traditional strategy of solely improving or amalgaming existing seismic inversion techniques. We have utilized a novel generative-adversarial network (GAN) method, a deep learning technique, specifically trained for the seismic inversion process. This unique methodology seeks to eliminate certain critical obstacles, such as the calculation of the inversion matrix, the initialization of the wavelet, and the management of limitations associated with the restricted frequency band of seismic amplitudes in seismic inversion. This research has produced significant outcomes. By employing the generative-adversarial deep learning method, we have successfully addressed the previously described obstacles and attained remarkable quality and accuracy in our outcomes. We performed seismic inversion utilizing actual data from an oil field, attaining a remarkable accuracy rate of 97.5%. The accuracy percentage is corroborated by the validation data and the mean squared error (MSE), so strengthening the reliability of the suggested method. Moreover, the acoustic impedance of the five test wells consistently registered below 0.125 units, underscoring the superiority of our results. The correlation coefficient among these test wells varied from a minimum of 96% to a maximum of 99%. The acoustic impedances derived from the band-limited technique and the model-based method exhibited correlation coefficients of 71% and 83%, respectively. The application of the generative-adversarial algorithm in the inversion process highlights its smart modern effectiveness. It possesses the capacity to completely update and enhance the traditional seismic inversion method, heralding a new epoch in seismic exploration and inversion.