Fast Prediction of Hydraulic Fracturing Geometry Using Machine Learning

For shale reservoirs, hydraulic fracturing design plays an important role, and with the help of machine learning (ML) methods, a novel approach will be presented for fast prediction of hydraulic fracture geometry. By integrating data from several wells of different blocks and machine learning algorithms were developed for the fast prediction of fracture geometry. The methodology involves hundreds of simulations using Gohfer, incorporating engineering, geological, and pumping parameters. Fracturing design parameters were extracted using Python and then given as input to ML models to train and validate. Furthermore, various parameters were identified as influencing fracture geometry. In parallel, analysis was conducted by building and comparing different models, and the effect of complexity on ML models. The analysis provides great results for predicting results from the Random Forest, XGBoost, and Voting Regressor models, with reduced validation errors for proppant cutoff length, fracture width, and fracture height. The Voting Regressor showed excellent performance by achieving the lowest values for MSE of 0.1729 and R square of 0.9786. This work demonstrates the efficiency of ML-based models and compares their accuracy with three different models after the integration of multi-source data for fast predicting fracture geometry, offering advantages in terms of speed and efficiency.