Filling the gaps of microresistivity imaging data by combining structure dual generative adversarial networks and diffusion models

Abstract Gaps in microresistivity imaging data pose challenges for reservoir characterization and formation analysis. To address this, we propose a novel approach that employs two cascaded deep-learning algorithms to fill these gaps. First, a conditional texture and structure dual generative (CTSDG) adversarial network is used to reconstruct the missing portions of the microresistivity imaging data. Next, a diffusion model with enhanced loss functions is employed to improve the resolution of gap-filled imaging data. The new loss function stabilizes the training process by combining traditional reconstruction loss with a weighted contrastive loss that adapts over time. Given the persistent gaps in microresistivity imaging data, we utilize acoustic borehole imaging and core data as training datasets for the CTSDG network. The trained models are then applied to microresistivity imaging data to produce microresistivity imaging data without gaps. This method was tested on data from a gas field, successfully restoring key geological features such as lithology, formation boundaries, fractures, and borehole collapse within the previously incomplete areas. The results demonstrate that the proposed method effectively recovers missing geological features, providing enhanced data for reservoir characterization and formation analysis.