Optimization and Economic Evaluation of Unconventional Reservoirs Hydraulic Fracturing Treatment Based on a Combined Simulation and Machine Learning Approach

This study introduces a novel and refined workflow for optimizing the fracturing operations based on simulation-assisted machine learning technique. Large datasets, ample efforts, and intensive time-consuming pose uncertainty and risk that human cognition is incapable of deciphering using direct simulation techniques only. The objective of this paper is to investigate the enhancement of a project's NPV through fracture parameters optimization. This includes fracture extensions, conductivity, distribution, fracture fluids, proppant types, and fracture job design. Specific UCR case study was used, and results were analyzed to verify the validity of the proposed workflow. Tight gas reservoir of 1 mD has been used for this case study. The first step of the proposed workflow started with data acquisition and data input in MFRAC^TM fracturing simulator. A set of 25 proppants was chosen according to the fracture closure stress, schedule, conductivity, and project NPV. Afterwards, fracture fluid was optimized based on fluid loss coefficient, treatment schedule, productivity, etc. Furthermore, the CMG commercial simulator was utilized to generate the required mathematical model using optimized fracture proppant and fracturing fluid. Finally, a designated machine learning-assisted random forest algorithm was used to select the effective fracturing parameters. Optimization efforts showed the best proppant and fracturing fluid selection. Among the 25 proppants tested, HSP proppant resulted in the highest Net Present Value (NPV). This proppant enhanced the well productivity to 2.46 times of the original productivity and boosted the NPV to more than $13.88 Million. This optimized proppant was used for the investigation of different fracturing fluids. Among which H006 resulted in the highest fracture half-length and NPV as opposed to the other fluids. CMG simulator was used for data generation and an assisted machine learning Algorithm was used for optimizing other fracture parameters, such as the number of fractures and anisotropy that were found influential for the fracturing treatment. Neither over-fitting nor under-fitting was observed during training and testing of the ML model with a coefficient of determination value of 0.9795. This study provides more insight into optimization of hydraulic fracturing design through a combined simulation and machine learning approach. The paper promotes further application of hydraulic fracturing jobs in unconventional reservoirs based on the proposed efficient workflow.