Natural fracture network model using Gaussian simulation and machine learning algorithms

Abstract

In this paper, a fracture network model is proposed to enhance the understanding of subsurface fracture characterization. The model combines geostatistical methods such as sequential indicators and Gaussian simulations. The model uses data from natural faults in Kazakhstan to predict the segment, azimuth, and length of fractures in unknown areas. The model is validated by comparing the simulated fracture networks with the original fracture data and by hiding some regions within the fracture network. The results show that the geostatistical methods perform better than the machine learning algorithm for azimuth prediction, while the machine learning algorithm performs better for length prediction. In addition, the validation of the fracture network model is conducted by comparing the production curve profiles in the tracer test setting. They are in good agreement.