A comparative study of novel object-based geostatistical algorithm and direct sampling method on fracture network modeling

• Published in: Stochastic environmental research and risk assessment Vol. 37; no. 2; pp. 777 - 793
• Main Authors: Shakiba, Sima, Doulati Ardejani, Faramarz
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.02.2023; Springer Nature B.V
• Subjects: Algorithms; Aquatic Pollution; Chemistry and Earth Sciences; Comparative studies; Computational Intelligence; Computer Science; Configurations; Earth and Environmental Science; Earth Sciences; Energy resources; Energy sources; Environment; Flow simulation; Fluid dynamics; Fluid flow; Geostatistics; Geothermal energy; Geothermal resources; Groundwater; Image processing; Iterative methods; Math. Appl. in Environmental Science; Mathematical models; Modelling; Oil reservoirs; Original Paper; Particle swarm optimization; Physics; Probability Theory and Stochastic Processes; Sampling; Sampling methods; Statistical analysis; Statistics for Engineering; Waste Water Technology; Water Management; Water Pollution Control; Water resources; Weights & measures; Discrete fracture network; Spatial characteristics; Direct sampling (DS) simulation; Multi-objective optimization problem; Particle swarm optimization
• ISSN: 1436-3240; 1436-3259
• DOI: 10.1007/s00477-022-02320-0

Accurate discrete fracture network modeling is a significant requirement for fluid flow simulation in various applications such as managing groundwater resources, simulating oil and gas reservoirs, and modeling geothermal energy resources. The existing fracture network modeling approaches are often unsuccessful in regenerating spatial variability and can only characterize the fracture geometries by statistical probability distributions. In addition, the alternative geostatistical methods to address these limitations suffer from a smoothing effect and reproducing fracture patterns due to the use of the two-point statistics technique. In this paper, a comparative study between the new object-based iterative fracture network modeling algorithm and the geostatistical direct sampling (DS) method is performed. The presented algorithm starts with an initial configuration to directly model the statistical geometry of the fracture network and uses particle swarm optimization algorithm to impose four different constraints and include its spatial variability. Each constraint is defined in the form of the difference between spatial properties of the reference configuration and of the generated model using L2-norm criterion characterized by common specific filtering functions in the image processing. Both employed methods are applied on a real 2-Dimentional fracture network image from an exposed wall and their performance is assessed by four different criteria including classification correctness rate (CCR), indicator variogram ( γ ), degree of consistency ( r ), and average relative error (ARE). Results show the superiority of the presented algorithm over the DS method in regenerating the real fracture network configuration with CCR = 0.99, r = 0.994, ARE = 2.63, and the same indicator variogram function.