IntroductionThe pore pressure of formations is a critical factor in assessing reservoir stability, designing drilling programs, and predicting production dynamics. Traditional methods often rely on limited well-logging data and empirical formulas to derive one-dimensional formation pressure models, which are inadequate for accurately reflecting the three-dimensional distribution of pore pressure in complex geological structures.MethodsTo address this challenge, this study leverages the temporal characteristics of well-logging and seismic data, employing the Mamba technique in conjunction with high-precision seismic inversion results, to construct a pore pressure prediction model. The model is a structured state-space model designed to process complex time-series data, and improve efficiency through parallel scan algorithm, making it suitable for large-scale three-dimensional data prediction. Initially, the deep learning model is trained and optimized by collecting and analyzing well-logging data, including key parameters such as acoustic time difference and density. Advanced seismic inversion techniques are then employed to obtain three-dimensional elastic properties like subsurface velocity and density, which serve as input features for the trained deep learning model.ResultsThrough complex nonlinear mappings, the model effectively captures the intrinsic relationship between input attributes and formation pressure, enabling accurate spatial distribution prediction of formation pore pressure. Research findings indicate that this method not only achieves high-precision formation pressure predictions but also reveals lateral variations in pore pressure that are challenging to detect using traditional methods.DiscussionThis provides robust technical support for the precise management and efficient development of oil and gas fields. With this method, oilfield engineers can more accurately assess formation pressure, optimize drilling programs, reduce accident risks, and enhance production efficiency.