A Drilling-Parameter-Driven Well-Seismic Integration Method for Dynamic Velocity Updating

Summary Accurate formation velocity prediction is essential in drilling engineering, supporting pore pressure estimation and guiding drilling fluid design in strongly heterogeneous strata. Conventional horizon-constrained interpolation, relying on static log velocity, performs poorly with sparse wells or absent real-time logs. To address these limitations, this study proposes a drilling-parameter-driven well-seismic integration method for dynamic velocity updating. A hybrid temporal convolutional network–long short-term memory–feedforward neural network (TCN-LSTM-FNN) model captures nonlinear relationships between drilling parameters and velocity, enabling real-time predictions. By integrating offset well data, real-time drilling parameters, and seismic horizon constraints, a continuously updated low-frequency velocity field is constructed. The interpretability of the model is improved by mapping key drilling parameters to velocity trends. Furthermore, seismic waveform and amplitude details are added to the inversion to enhance resolution and accuracy. The method was validated on the Marmousi2 model, whose strong heterogeneity is ideal for testing velocity prediction. Numerical experiments demonstrate accurate velocity predictions ahead of the drill bit in complex geological settings. Applied to field data from an offshore oil field in eastern China, the method dynamically updated the velocity during drilling, providing reliable predictions around the borehole and ahead of the bit in log-lacking intervals and improving adaptability to complex geological conditions.