Real-Time AI Geosteering for Horizontal Well Trajectory Optimization

Abstract

Abstract Active petrophysical geosteering in reservoir formations has become a very promising practice and commonplace to improve express formation evaluation and maximize hydrocarbon production from reservoirs. The present study showcases the effectiveness of utilizing artificial intelligence (AI) framework to process resistivity logging while drilling (LWD) data for the purpose of real-time optimizing well trajectories and maximizing hydrocarbon production in pay zones during drilling horizontal section. We introduce a novel framework that automatically adjusts planned well trajectories during horizontal drilling. The framework takes the planned wellbore trajectory, reservoir model porosity, and ultra-deep resistivity LWD data as input. Hydrocarbon saturation volume is then calculated using the Archie equation. Subsequently, optimization algorithms correct the planned trajectory to maximize wellbore production using hydrocarbon saturation volume. The framework delivers an optimal wellbore trajectory performing real-time formation evaluation, guiding the drill bit through highly saturated pay zones. The proposed framework was tested on a 2D synthetic dataset using various optimization algorithms, including reinforcement learning algorithms for continuous action spaces (PPO, DDPG, TwinDDPG) and Q-learning and evolutionary algorithms. The evolutionary group of optimization algorithms achieved the highest efficiency with baseline hyperparameter settings, improving cumulative oil saturation per drilled meter by up to 32.5%. Reinforcement learning algorithms needs to be further explored because they have promising results but still high computational complexity. The evolutionary algorithm was then verified on a 3D Groningen field dataset, determining optimal hyperparameters such as differential evolution strategy, population size, mutational constant, and maximum number of iterations. The framework improved cumulative oil saturation per drilled meter by up to 6%, significantly increasing the average number of penetrated hydrocarbon-saturated pay zones along the drilling path. The developed AI-based framework presents an innovative approach for real-time automatic correction of well drilling trajectories, maximizing well productivity. This method can significantly aid in the interpretation and optimization of decision-making related to geosteering.