Applied Reinforcement Learning Method for Pressure Control in Managed Pressure Drilling

During drilling operations, maintaining an optimal bottom hole pressure (BHP) in this narrow pressure margin between the pore and fracture pressure gradient is critical to avoid damage to the formation and the well. Back-pressure Managed pressure drilling (MPD) is widely used in these unconventional drilling prospects as it controls the BHP by adjusting the surface pressure using a back-pressure choke manifold at the outlet. The standard technique used for automated control of the MPD choke is the proportional-integral-derivative (PID) controller. However, the physical system in the well during drilling operations is highly non-linear with respect to pressure and flow rates. Alternative control methods are therefore being explored. In this research, we explore a novel automation method for MPD by training an intelligent tuning agent for a PID control system. The objective is to explore an automated and optimized control strategy assisted by reinforcement learning algorithms and to verify the feasibility and efficiency of this approach.