Characterization of Horizontal Fractures in Shale Gas Reservoirs Using a Rock-Physics-Based Method Integrated With SA-PSO Algorithm and CNN

Natural fractures play a crucial role in shale reservoir characterization. While vertical fractures can be estimated using amplitude variation with azimuth (AVAz) inversion methods, predicting horizontal fractures remains limited due to their intricate seismic responses. This article addresses this gap by introducing a rock-physics-based method for predicting horizontal fracture parameters using well-log and seismic data. A shale model is developed using rock physics methods to quantify elastic properties associated with horizontal fractures. The sensitivity of the elastic properties to horizontal fracture parameters is analyzed to validate their potential for fracture prediction. Subsequently, a model-based inversion approach is proposed to extract horizontal fracture parameters from logging data. This method integrates a simulated annealing particle swarm optimization (SA-PSO) algorithm to ensure robustness and convergence in calculations. The results confirm the efficacy of the proposed method in estimating horizontal fracture parameters in boreholes. Based on the results obtained using the proposed method and well-log data, a prediction model is constructed to capture complex correlations between horizontal fracture density and elastic properties by employing a convolutional neural network (CNN) architecture. Following successful training and validation, the established model predicts horizontal fracture density in shales using elastic properties obtained from seismic inversion. The results align closely with estimates derived from logging data and are consistent with the geological characteristics of the studied area. This study presents a valuable method for predicting horizontal fractures using geophysical logging and seismic data, offering valuable insights into natural fracture characterization in shales.