Fracture Identification and Porosity Prediction of Carbonate Reservoirs Based on Neural Network Simulation

Abstract

Carbonate reservoirs have characteristics such as diverse reservoir types, complex structures, and strong heterogeneity, resulting in complex logging responses. Therefore, it is necessary to study logging characterization methods suitable for complex carbonate reservoirs. The dissolution of deep carbonate rocks in karst depressions in central Sichuan is relatively weak, but the development of fractured reservoirs makes it difficult to effectively apply conventional interpretation methods. A multi-layer perceptual neural network model based on artificial intelligence was used to establish a fracture identification and porosity prediction model based on logging data, combined with the measured physical properties of G1 well. The results indicate that fractures are generally developed in the study area, accounting for over 30%; The research area is mainly composed of dense reservoirs, with an average porosity of 2.7% in the normal matrix section and 2.6% in the fracture developed section. The verification with actual physical properties also indicates that the model has a prediction accuracy of 78%, which has high application value.