Enhancing reservoir characterization using deep learning aided model-based inversion: a case study of data from the coastal swamp zone in the Niger Delta

Abstract

Seismic Reservoir Characterization is a pivotal element in seismic data interpretation. This study describes a successful utilization of model-based seismic inversion methodology coupled with Probabilistic Neural Network for the identification of hydrocarbon reservoir zones within post-stack seismic data. The paper unfolds in two segments. Initially, Acoustic Impedance (AI) volume is extrapolated from seismic datasets via the application of the model-based inversion algorithm in the time domain. The strong correlation coefficient of 0.988 between synthetic and seismic data underscores the effectiveness of model-based inversion. Subsequently, a Probabilistic Neural Network (PNN) undergoes training, validation, and testing utilizing estimated porosity data at well locations, serving as internal attributes, and the outcomes from model-based inversion as external attributes. The trained Probabilistic Neural Network is then deployed across the seismic volume to delineate a three-dimensional map in total porosity. A notable high value of total porosity, ranging from 16 to 30%, observed within horizons A and B, indicates a substantial volume of void spaces within the rock formation. Therefore, the presence of such elevated porosity values in horizons A and B signifies the existence of a promising reservoir with ample capacity for fluid storage and migration, which is scientifically significant for hydrocarbon exploration and production. The findings from this study underscore the potential of merging model-based inversion and PNN for effective estimation of reservoir properties, particularly in scenarios where the relationship between porosity and acoustic impedance exhibits non-linear characteristics