A Pre-Drilling Prediction Method of Formation Tri-Pressures Based on Machine Learning

In response to the issues of low prediction reliability and insufficient computational efficiency associated with conventional models in calculating formation tri-pressures for offshore deep and ultra-deep oil and gas exploration, a machine learning-based method was used. In the method, a deep neural network (DNN) model was used as the main structure, and rock mechanics physical models were integrated as physical constraints. By jointly optimizing data-fitting errors and physical residual terms in the loss function, the mathematical embedding of the physical constraints was achieved. Additionally, it l incorporated multi-source data, including geological, logging, mud logging , and field operational data, to establish a physics − data-driven formation tri-pressure prediction model. Model performance was evaluated using root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²). The model was applied in the Wushi Block of the South Weixi Depression in the Beibuwan Basin. The results demonstrate minimal discrepancies between predicted and actual formation tri-pressure values, indicating high prediction accuracy, with results consistent with actual application. This machine learning-based pre-drilling prediction method of formation tri-pressures can provide technical support for safe and efficient drilling in deep and ultra-deep formations.