Real-Time Data-Driven Updates for Look-Ahead Lithology Parameter Modeling in Geosteering

Summary Geosteering plays a vital role in directional drilling, particularly for horizontal and deviated wells, but challenges persist in accurately guiding well trajectories through complex and unpredictable geological environments. With this study, we tackle three critical challenges in geosteering: (1) reducing uncertainties in lithology parameter distribution in undrilled formations, which affects trajectory accuracy; (2) developing a reliable “look-ahead” lithology parameter prediction model that utilizes multiscale geophysical data to enhance predictive precision; and (3) efficiently integrating real-time logging-while-drilling (LWD) data to continuously refine lithology parameter predictions, minimizing uncertainty during drilling operations. Our research introduces an innovative approach by using a random forest (RF)-based framework to link LWD and multifrequency seismic data. This method enables dynamic updates to lithology parameter predictions by adjusting the weights of seismic data from various frequencies based on real-time LWD inputs. Blind well tests validate the framework, demonstrating a high correlation between predicted and actual acoustic impedance values along the full well section—initially exceeding 0.85, with improvements above 0.9 as drilling depth increases. The RF algorithm effectively utilizes impedance data from multifrequency seismic sources, incorporating geological prior information to enhance model robustness. This method enables reliable look-ahead lithology parameter prediction, anticipating lithological changes up to 400 m ahead of the drill bit. Our findings demonstrate that this approach effectively reconciles the resolution differences between logging and seismic data. By integrating geological priors from predrilling models with real-time, data-driven LWD insights, the methodology enhances both the efficiency and robustness of model updates in geosteering applications.