Density abrupt interface inversion method and system based on machine learning constraints

A density abrupt interface inversion method based on machine learning constraints, comprising following steps:

• S1, constructing an initial basin interface based on measured gravity data, and randomly generating a disturbed basin interface data set based on the initial basin interface;
• wherein a method for constructing the initial basin interface comprises determining an initial depth of the initial basin interface through a Bouguer plate model according to the measured gravity data of the basin in a target area, a calculation formula is as follows:
• Z
• i
• ⁢
• n
• ⁢
• i
• =
• g
• o
• ⁢
• b
• ⁢
• s
• ⁢
• Δ
• ⁢
• ρ
• 0
• 4
• ⁢
• 1
• .
• 8
• ⁢
• 9
• ⁢
• Δ
• ⁢
• ρ
• 0
• 2
• +
• β
• ⁢
• g
• o
• ⁢
• b
• ⁢
• s
• Zini is an initial depth of a basement of the basin model, gobs is the measured gravity data, Δρ0 is a density contrast between a ground surface and surrounding rock, and β is a gradient of the density contrast varying with depth;
• and
• a method for generating the disturbed basin interfaces comprises generating a plurality of sets of disturbed basin interfaces by a plurality of random transformation methods based on probability distribution;
• S2, obtaining a basin interface data set through Hadamard product operation based on the initial basin interface and the disturbed basin interface data set;
• S3, obtaining a high-resolution density interface model data set through filling density of overlying strata with advanced functions based on the basin interface data set;
• S4, performing forward calculation on the high-resolution density interface model data set to obtain a simulated gravity data set;
• S5, carrying out mathematical transformation on the simulated gravity data set and applying depth weighting to obtain a low-resolution migration density interface model data set;
• S6, optimizing a migration model deep learning network based on the high-resolution density interface model data set, the simulated gravity data set and the low-resolution migration density interface model data set, and mapping the measured gravity data and a primary migration density interface model corresponding to the measured gravity data based on the migration model deep learning network to obtain a high-resolution constrained density interface prior model;
• and
• S7, constructing a stable nonlinear loss function with multivariate density contraction constraints and performing regularization inversion based on the high-resolution constrained density interface prior model to obtain a high-resolution density interface model.