Method, system, and device for full waveform inversion on elastic wave seismic data
View Patent ↗Disclosed are a method, system, and device for full waveform inversion on elastic wave seismic data. The method includes: constructing a convolutional neural network full waveform inversion module; determining a multi-scale inversion frequency based on the observation data; performing low-pass filtering on the observation data; determining an initial velocity and an RTM image corresponding to the current inversion frequency; determining a current inversion result by using the convolutional neural network full waveform inversion module based on the initial velocity and the RTM image corresponding to the current inversion frequency; determining whether inversion performed at the multi-scale inversion frequency is completed; if the inversion is not completed, performing inversion on a next frequency, performing smoothing processing on the inversion result, and replacing the initial velocity corresponding to the current inversion frequency with an inversion result obtained after smoothing processing until the inversion performed at the multi-scale inversion frequency is completed.
1 . A method for full waveform inversion on elastic wave seismic data, comprising:
constructing a convolutional neural network full waveform inversion module, wherein the convolutional neural network full waveform inversion module is configured to determine an inversion result based on an initial velocity and a reverse time migration (RTM) image;
obtaining observation data, and determining a multi-scale inversion frequency based on the observation data, wherein the multi-scale inversion frequency comprises a plurality of inversion frequencies sorted in order from low to high, and inversion starts from a low frequency;
performing low-pass filtering on the observation data by using a current inversion frequency;
determining an initial velocity and an RTM image corresponding to the current inversion frequency based on the observation data obtained after low-pass filtering;
determining a current inversion result by using the convolutional neural network full waveform inversion module based on the initial velocity and the RTM image corresponding to the current inversion frequency;
determining whether inversion performed at the multi-scale inversion frequency is completed;
if the inversion is not completed, performing inversion on a next frequency based on the order of the inversion frequencies of the multi-scale inversion frequency, returning to the step of performing low-pass filtering on the observation data by using a current inversion frequency, performing smoothing processing on the current inversion result, and replacing the initial velocity corresponding to the current inversion frequency with an inversion result obtained after smoothing processing until the inversion performed at the multi-scale inversion frequency is completed; and
if the inversion is completed, using a last inversion result as a target inversion result.
2 . The method for full waveform inversion on elastic wave seismic data according to claim 1 , wherein the convolutional neural network full waveform inversion module comprises a P wave inversion module, an S wave inversion module, and a density inversion module.
3 . The method for full waveform inversion on elastic wave seismic data according to claim 1 , wherein the determining an initial velocity and an RTM image corresponding to the current inversion frequency based on the observation data obtained after low-pass filtering specifically comprises:
determining the initial velocity and the RTM image corresponding to the current inversion frequency by tomography or migration velocity analysis based on the observation data obtained after low-pass filtering.
4 . The method for full waveform inversion on elastic wave seismic data according to claim 1 , wherein the observation data is initially collected shot data of seismic data.
5 . A device for full waveform inversion on elastic wave seismic data, comprising: at least one processor, at least one memory, and a non-transitory computer program instructions stored in the non-transitory memory, wherein the processor executes the computer non-transitory program instructions to implement the method according to claim 1 .
6 . The device for full waveform inversion on elastic wave seismic data according to claim 5 , wherein the convolutional neural network full waveform inversion module comprises a P wave inversion module, an S wave inversion module, and a density inversion module.
7 . The device for full waveform inversion on elastic wave seismic data according to claim 6 , wherein the memory is a non-transitory computer-readable storage medium.
8 . The device for full waveform inversion on elastic wave seismic data according to claim 5 , wherein the determining an initial velocity and an RTM image corresponding to the current inversion frequency based on the observation data obtained after low-pass filtering specifically comprises:
determining the initial velocity and the RTM image corresponding to the current inversion frequency by tomography or migration velocity analysis based on the observation data obtained after low-pass filtering.
9 . The device for full waveform inversion on elastic wave seismic data according to claim 8 , wherein the memory is a non-transitory computer-readable storage medium.
10 . The device for full waveform inversion on elastic wave seismic data according to claim 5 , wherein the observation data is initially collected shot data of seismic data.
11 . The device for full waveform inversion on elastic wave seismic data according to claim 10 , wherein the memory is a non-transitory computer-readable storage medium.
12 . The device for full waveform inversion on elastic wave seismic data according to claim 5 , wherein the memory is a non-transitory computer-readable storage medium.
13 . A system for full waveform inversion on elastic wave seismic data, comprising:
a module construction unit, configured to construct a convolutional neural network full waveform inversion module, wherein the convolutional neural network full waveform inversion module is configured to determine an inversion result based on an initial velocity and a reverse time migration (RTM) image;
a multi-scale inversion frequency determining unit, configured to: obtain observation data, and determine a multi-scale inversion frequency based on the observation data, wherein the multi-scale inversion frequency comprises a plurality of inversion frequencies sorted in order from low to high, and inversion starts from a low frequency;
a low-pass filtering unit, configured to perform low-pass filtering on the observation data by using a current inversion frequency;
an input data determining unit, configured to determine an initial velocity and an RTM image corresponding to the current inversion frequency based on the observation data obtained after low-pass filtering;
an inversion result determining unit, configured to determine a current inversion result by using the convolutional neural network full waveform inversion module based on the initial velocity and the RTM image corresponding to the current inversion frequency;
a determining unit, configured to determine whether inversion performed at the multi-scale inversion frequency is completed;
an iterative unit, configured to: if the inversion is not completed, perform inversion on a next frequency based on the order of the inversion frequencies of the multi-scale inversion frequency, return to the low-pass filtering unit, perform smoothing processing on the current inversion result, and replace the initial velocity corresponding to the current inversion frequency with an inversion result obtained after smoothing processing until the inversion performed at the multi-scale inversion frequency is completed; and
a target inversion result determining unit, configured to: if the inversion is completed, use a last inversion result as a target inversion result.