Neural Network Phase Control Property Modeling Based on Complex Structural Constraints

The low-frequency attribute model is an important part of seismic inversion, and its accuracy directly affects the precision of seismic inversion. With the continuous deepening of exploration, the exploration targets have gradually shifted from shallow to deep, from conventional reservoirs to concealed reservoirs, and from simple structures to complex structures. The transformation of exploration targets also simultaneously places higher demands on the accuracy of inversion methods. For areas with rapid lateral changes in strata deposition, complex structures, and few wells, traditional methods of constructing low-frequency attribute models based on theoretical drives are difficult to obtain accurate seismic low-frequency attribute models, which restricts the accuracy of seismic inversion. This paper proposes an intelligent phase-controlled attribute modeling method based on complex structure constraints. Under the constraints of complex structure models, using well logging curves as labels and seismic attributes as training samples, local neural network training is conducted within the sequence structure model to establish a seismic low-frequency model. This method comprehensively utilizes information such as strata sequence, well logging curves, and seismic data to establish a high-precision low-frequency attribute model, overcomes the serious problems of mathematization and modelization in conventional methods, significantly improves the seismic inversion effect, and provides important technical support for reservoir prediction.