Well Construction Action Planning and Automation through Finite-Horizon Sequential Decision-Making

• Published in: Energies (Basel) 2022-08, Vol.15 (16), p.5776
• Main Authors: Saini, Gurtej Singh, Erge, Oney, Ashok, Pradeepkumar, van Oort, Eric
• Format: Article
• Language: English
• Subjects: Algorithms; Analysis; Automation; Cleaning; Construction; Construction accidents & safety; Decision making; Design and construction; digital twinning; Digital twins; Drilling; Drilling muds; Drillstrings; Feedback; Fluid flow; hole cleaning; Horizon; Hydraulics; Markov decision process; Markov processes; Mechanical properties; Methods; Planning; Reinforcement; reward shaping; Rheological properties; Rheology; Sequences; sequential decision-making; Simulation; well construction; Wells
• ISSN: 1996-1073; 1996-1073
• DOI: 10.3390/en15165776

Well construction operations require continuous complex decision-making and multi-step action planning. Action selection at every step demands a careful evaluation of the vast action space, while guided by long-term objectives and desired outcomes. Current human-centric decision-making introduces a degree of bias, which can result in reactive rather than proactive decisions. This can lead from minor operational inefficiencies all the way to catastrophic health and safety issues. This paper details the steps in structuring unbiased purpose-built sequential decision-making systems. Setting up such systems entails representing the operation as a Markov decision process (MDP). This requires explicitly defining states and action values, defining goal states, building a digital twin to model the process, and appropriately shaping reward functions to measure feedback. The digital twin, in conjunction with the reward function, is utilized for simulating and quantifying the different action sequences. A finite-horizon sequential decision-making system, with discrete state and action space, was set up to advise on hole cleaning during well construction. The state was quantified by the cuttings bed height and the equivalent circulation density values, and the action set was defined using a combination of controllable drilling parameters (including mud density and rheology, drillstring rotation speed, etc.). A non-sparse normalized reward structure was formulated as a function of the state and action values. Hydraulics, cuttings transport, and rig state detection models were integrated to build the hole cleaning digital twin. This system was then used for performance tracking and scenario simulations (with each scenario defined as a finite-horizon action sequence) on real-world oil wells. The different scenarios were compared by monitoring state–action transitions and the evolution of the reward with actions. This paper presents a novel method for setting up well construction operations as long-term finite-horizon sequential decision-making systems, and defines a way to quantify and compare different scenarios. The proper construction of such systems is a crucial step towards automating intelligent decision-making.