Enhancing Breakdown Pressure Predictions in Ultra-Tight Formations through Robust Machine Learning Techniques

Accurately estimating rock breakdown pressure is crucial for designing effective hydraulic fracturing operations, particularly in unconventional ultra-tight reservoirs where hydrocarbon extraction is challenging. However, conducting experimental studies on hydraulic fracturing is both time-consuming and costly. To address this, we employed robust machine learning (ML) tools to estimate the breakdown pressure. The research comprised two stages: an extensive experimental phase followed by the development of ML prediction models using the obtained data. The ML models were trained using experimental factors such as injection rate, confining stress, fluid viscosity, and rock characteristics, including unconfined compressive strength, Poisson's ratio, tensile strength, porosity, permeability, and bulk density. Six machine learning techniques-K-Nearest Neighbor (KNN), Random Forest (RF), Decision Trees (DT), artificial neural networks (ANN), gradient boosting (GB), and adaptive gradient boosting (Adaboost)-were employed to construct the prediction models. With the optimal settings for the ML models, the breakdown pressure of the tight formations was accurately predicted with a 99% accuracy. The proposed ML approaches not only offer significant cost savings but also serve as a quick evaluation tool to assess the development prospects of tight rocks.