Comparison of Machine Learning-Based Breakdown Pressure Prediction and Initiation Criteria in Hydraulic Fracturing Testing

The breakdown pressure ([Formula: see text]) is a critical index for assessing the effectiveness of reservoir stimulation. Achieving precise and efficient evaluation of [Formula: see text] under multiple influencing factors is essential for deepening understanding of the underlying mechanisms in reservoir stimulation. Based on the current yet incomplete knowledge, the current analytical models and initiation criteria cannot provide a reliable prediction on [Formula: see text]. Thus, this work uses the predictive power of machine learning to tackle these challenges head-on. Four variants of the support vector machine (SVM) algorithms were employed, including SVM-Linear, SVM-Poly, SVM-radial basis function (RBF), and SVM-Sigmoid. These algorithms were trained and tested using a dataset of 168 entries from hydraulic fracturing studies. Among these, the SVM-RBF model demonstrated superior performance, achieving the lowest mean absolute error (MAE) and root mean square error (RMSE), and attaining the highest coefficient of determination (R). The analysis also underscored the importance ranking of tensile strength, elastic modulus, vertical stress ([Formula: see text]), and horizontal stress ([Formula: see text]) in predicting the [Formula: see text], which were identified as key parameters for optimizing the machine learning model’s training process. By comparing the machine learning predictions to initiation criteria using data from glutenite and shale hydraulic fracturing tests, the machine learning approaches were found to be more accurate. Consequently, the SVM-based machine learning algorithm can accurately predict the [Formula: see text] in hydraulic fracturing of rock considering key parameters. These findings not only enhance our understanding of reservoir stimulation mechanisms but also offer a novel perspective for future research and applications in the field.