Modelling and Simulation of Managed Pressure Drilling (MPD) Using Simulink and Physics-Informed Neural Networks

Abstract

Managed Pressure Drilling (MPD) utilizes a sophisticated network of sensors, control systems, and specialized drilling equipment to continuously monitor pressure at various depths and adjust parameters to maintain a desired pressure profile. The primary objective of MPD is to achieve an optimal pressure balance within the wellbore, enabling safe and effective drilling operations without compromising the integrity of the formation. In this study, a MATLAB Simulink-based MPD model was developed and compared to the VICTUS software for validation. Mathematical connections between system components in the Simulink Editor were made by connecting blocks with signal lines. Subsystems were created in the model by grouping several blocks into single components. The model was based on data from a high-pressure-high-temperature onshore well located in River State, Nigeria, with a pressure of 15 000 psi at the top of the formation and a formation temperature of 150°C. Early kick detection was accomplished via Kick Influx Volume analysis in the MPD model. The critical gas level indicators from the Simulink MPD model indicate effective well control during the MPD operations. The results showed that the average surface backpressure (SBP) recorded by the VICTUS software and the MATLAB Simulink MPD model were in agreement. The Simulink model SBP predictions for coefficient of determination (R2), correlation coefficient (R), and Root Mean Square Error (RMSE) were 0.9928, 0.9964, and 0.4705, respectively. Additionally, a Physics Informed Neural Network (PINN) was used to predict SBP, with a testing R2of 0.9815. These findings indicate a significant and positive correlation between the VICTUS software, MATLAB Simulink MPD, and PINN models, developed as a cost-effective tool.