A multi-model fitting algorithm for extracting a fracture network from microseismic data

Fractures are increasingly employed in tectonic movement and earthquake risk analyses. Because fracture connectivity influences fluid flow pathways and flow rates, fractures are studied to evaluate sites for CO2 sequestration, radioactive waste storage and disposal, petroleum production, and geothermal energy applications. Discrete fracture networks are an effective method for imaging fractures in three-dimensional geometric models and for analyzing the fluid behavior that cause movements in fracture zones. Microseismic event monitoring data can be used to analyze the event source mechanisms and the geometry, distribution, and orientation of the fractures generated during the event. This study proposes a method for simultaneously imaging multi-fracture networks using microseismic monitoring data. The random sample consensus and propose, expand, and re-estimate labels algorithms commonly used in multi-model fitting were integrated to produce an upgraded method that accommodates geophysical data for faster and more accurate simultaneous multi-fracture model imaging within a point cloud. The accuracy of the method was improved using circular calculation and density-based spatial clustering of applications with noise, such that the estimated fracture orientations correspond well to those at the actual locations. The proposed algorithm was applied to synthetic data to assess the impact of considering orientation and outlier data on the model results. The errors in the results when considering orientation were 1.32% and 0.83% for the strike and dip angles, respectively, and those without considering were 20.23% and 24.63% respectively. In addition, the errors in the results obtained from data containing many outliers were 1.89% and 1.64% for the strike and dip angles, respectively. Field microseismic data were also used to depict fractures representing the dominant orientation, and the errors of the strike and dip angle estimates were 2.89% and 2.83%, respectively. These results demonstrate the suitability of the algorithm for fast and accurate field data modeling.