Summary Real-time monitoring and accurate control of bottomhole pressure (BHP) play a decisive role in maintaining the mechanical balance of the formation-wellbore system during drilling operations. Due to the difficulty in real-time transmission of downhole parameters, surface measurements often fail to reflect the true downhole conditions, limiting the adaptability of existing intelligent models to dynamically changing complex drilling environments and constraining their field applications. In this paper, we propose a surface-parameter-driven BHP prediction approach with real-time calibration. By analyzing the multiphase flow mechanics in the wellbore and the coupling correlation between surface and downhole data, key surface parameters are selected as auxiliary prediction tasks to construct a multitask temporal neural network model that outputs both surface features and BHP jointly. During model inference, an online error feedback mechanism dynamically updates the shared hidden layer weights using real-time surface data and prediction errors, significantly enhancing the model’s adaptability to current complex downhole conditions. Results demonstrate that compared with traditional single-task models, the proposed method reduces prediction error by 6%, achieving an accuracy of 93% in BHP monitoring. This study provides an innovative framework for real-time and precise monitoring of BHP under complex downhole environments.
Xia et al. (Thu,) studied this question.