US 12,736,699 B1 · Fresnel zone picking model and training method therefor, and diffraction wave imaging method based on machine learning

Abstract

Provided are a Fresnel zone picking model and training method therefor, and a diffraction wave imaging method based on machine learning. The training method includes acquiring seismic wave training data; labeling a Fresnel zone label for the set of horizontal slice images of the initial 3D dip-angle gathers to obtain a set of true labels; inputting the set of horizontal slice images and the set of true labels of the initial 3D dip-angle gathers into a pre-constructed Fresnel zone picking model for iterative training, so as to obtain a prediction result of the Fresnel zone; and updating parameters of the Fresnel zone picking model for each training based on the true label of the Fresnel zone and a loss value of the prediction result until a preset iteration condition is reached. The diffraction wave imaging method performs the diffraction wave imaging by using the trained Fresnel zone picking model.

A diffraction wave imaging method based on machine learning, wherein the method comprises:

acquiring seismic wave detection data of a target work zone, wherein the seismic wave detection data comprises a set of horizontal slice images of 3D dip-angle gathers corresponding to all imaging points on all imaging lines in a seismic wave imaging space;
inputting the set of horizontal slice images of the 3D dip-angle gathers into a pre-trained Fresnel zone picking model to pick up a Fresnel zone of a horizontal slice corresponding to each of the imaging points;
muting the picked Fresnel zone in the set of horizontal slice images of the 3D dip-angle gathers;
and
stacking the set of horizontal slice images of the 3D dip-angle gathers after muting the Fresnel zone, so as to obtain a diffraction wave imaging profile for identifying small-scale variations of subsurface microstructures;
wherein the Fresnel zone picking model is trained by a training method comprising:
acquiring seismic wave training data of a target work zone, wherein the seismic wave training data comprises a set of horizontal slice images of initial 3D dip-angle gathers corresponding to all imaging points on a target imaging line in a seismic wave imaging space;
labeling a Fresnel zone label for the set of horizontal slice images of the initial 3D dip-angle gathers to obtain a set of true labels;
inputting the set of horizontal slice images and the set of true labels of the initial 3D dip-angle gathers into a pre-constructed Fresnel zone picking model for iterative training, so as to obtain a prediction result of the Fresnel zone;
and
updating parameters of the Fresnel zone picking model for each training based on the true labels of the Fresnel zone and a loss value of the prediction result until a preset iteration condition is reached;
wherein the step of labeling a Fresnel zone label for the set of horizontal slice images of the initial 3D dip-angle gathers to obtain a set of true labels comprises:
picking up a Fresnel zone in each horizontal slice in the set of horizontal slice images of the initial 3D dip-angle gathers;
muting the picked Fresnel zone;
and
producing the set of true labels according to the set of horizontal slice images after muting the Fresnel zone;
and
wherein the Fresnel zone is in an elliptical shape on the horizontal slices, and the step of picking up a Fresnel zone in each horizontal slice in the set of horizontal slice images of the initial 3D dip-angle gathers comprises:
acquiring 3D dip-angle gather data corresponding to the horizontal slice for each horizontal slice in the set of horizontal slice images of the initial 3D dip-angle gathers, wherein the 3D dip-angle gather data comprises a seismic wave two-way traveltime, a dip-angle in x-direction and a dip-angle in y-direction corresponding to a seismic trace, and a main frequency of the seismic wave;
calculating radii of the Fresnel zone in two directions based on the 3D dip-angle gather data and a picking formula of the Fresnel zone, and determining a range of the Fresnel zone based on the radii in the two directions and a stationary phase point, wherein the picking formula of the Fresnel zone is:
γ
x
=
arccos
⁡
(
T
0
/
cos
⁡
(
φ
x
⁢
0
)
-
f
d
/
4
T
0
/
cos
⁡
(
φ
x
⁢
0
)
)
,
γ
y
=
arccos
⁡
(
T
0
/
cos
⁡
(
φ
y
⁢
0
)
-
f
d
/
4
T
0
/
cos
⁡
(
φ
y
⁢
0
)
)
,
where γx is a radius of the Fresnel zone in x-direction;
γy is a radius of the Fresnel zone in y-direction;
T0 is the seismic wave two-way traveltime;
fd is the main frequency of the seismic wave;
(φx0, φy0) is the stationary phase point of the 3D dip-angle gathers at an imaging point on a corresponding horizontal slice;
and φx0, φy0 respectively correspond to stratum dip-angles in x-direction and y-direction of the stratum where the imaging point is located.

93 more elements

Fresnel zone picking model and training method therefor, and diffraction wave imaging method based on machine learning

Life of this patent

Fresnel zone picking model and training method therefor, and diffraction wave imaging method based on machine learning

Life of this patent

Abstract

Claims8