Growth Parameter Conditioning of Genetic Discrete Fracture Networks to Trace Length Distribution Data

Abstract

ABSTRACT: Genetic discrete fracture network (DFN) modeling is a dynamic modeling approach that simulates fracture growth processes in DFN generation. The modeling procedure incorporates growth-related input parameters, such as the growth time, step size, and coefficients of nucleation and growth, which are difficult to deduce directly from survey data. This study introduces a method for conditioning these growth parameters to trace length distribution using an optimization procedure, where the growth parameters are fine-tuned to produce genetic DFNs that align with the observed distribution. The optimization objective function is formulated to minimize the difference in trace length distributions and the number of traces between the observed trace map and that derived from a genetic DFN. In case studies involving synthetic and real trace maps, the conditioned genetic DFNs reproduced trace maps that exhibit a high level of fitness to the observed data. Considering the representation of complex growth processes and fracture interactions, these models can provide a robust foundation for DFN extrapolation beyond the surveyed areas. 1. INTRODUCTION Natural fracture networks can be modeled in the form of discrete fracture networks (DFNs) that capture the spatial characteristics of the observed network. In DFNs, fractures are simplified as discrete objects, commonly as disks, which replicate the geometric properties of fractures. The geometric properties of these fracture disks, including size and orientation, can be determined by analyzing the distributions of surveyed data and directly sampling values from these distributions to match the observed data. This modeling approach is termed as "stochastic" or "Poisson" DFN modeling, where fractures are randomly generated by corresponding statistical models (Lei et al., 2017). While stochastic DFNs follow the statistical models of geometric properties, such as size and orientation, they exhibit clear limitations in reproducing the organization or topology of fracture networks, especially the specific fracture terminations in the form of T-type intersections (Selroos et al., 2022). The stochastic DFN approach fundamentally lacks physical integrity as it is purely statistical and does not consider the physical process involved in fracture network generation.