CN115453623A - UNet-based receiving function and surface wave frequency dispersion joint inversion method - Google Patents
UNet-based receiving function and surface wave frequency dispersion joint inversion method Download PDFInfo
- Publication number
- CN115453623A CN115453623A CN202211145748.1A CN202211145748A CN115453623A CN 115453623 A CN115453623 A CN 115453623A CN 202211145748 A CN202211145748 A CN 202211145748A CN 115453623 A CN115453623 A CN 115453623A
- Authority
- CN
- China
- Prior art keywords
- unet
- surface wave
- data set
- function
- inversion
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 239000006185 dispersion Substances 0.000 title claims abstract description 37
- 238000000034 method Methods 0.000 title claims abstract description 30
- 230000006870 function Effects 0.000 claims abstract description 66
- 238000012549 training Methods 0.000 claims abstract description 31
- 238000004364 calculation method Methods 0.000 claims abstract description 12
- 238000013135 deep learning Methods 0.000 claims abstract description 8
- 230000004913 activation Effects 0.000 claims description 5
- 238000011176 pooling Methods 0.000 claims description 4
- 230000007547 defect Effects 0.000 abstract description 2
- 238000012360 testing method Methods 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000001514 detection method Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000008034 disappearance Effects 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 239000011435 rock Substances 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
- 230000017105 transposition Effects 0.000 description 1
- 238000009827 uniform distribution Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V1/00—Seismology; Seismic or acoustic prospecting or detecting
- G01V1/28—Processing seismic data, e.g. for interpretation or for event detection
- G01V1/282—Application of seismic models, synthetic seismograms
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V1/00—Seismology; Seismic or acoustic prospecting or detecting
- G01V1/28—Processing seismic data, e.g. for interpretation or for event detection
- G01V1/30—Analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Remote Sensing (AREA)
- General Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Environmental & Geological Engineering (AREA)
- Geology (AREA)
- General Life Sciences & Earth Sciences (AREA)
- Geophysics (AREA)
- Theoretical Computer Science (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Geophysics And Detection Of Objects (AREA)
Abstract
The invention discloses a UNet-based receiving function and surface wave frequency dispersion joint inversion method, which comprises the following steps of: s1: constructing a speed model, and performing forward calculation to obtain a training data set; s2: and (3) constructing a UNet, training by using a training data set, and taking a receiving function to be inverted and surface wave frequency dispersion data as the input of the UNet to obtain the velocity structures of the crust and the upper mantle. The receiving function and surface wave frequency dispersion joint inversion method based on UNet can reduce non-uniqueness in inversion, improve inversion efficiency and stability, explore applicability of a deep learning method in joint inversion, can be used as a substitute for conventional inversion, improve defects of conventional inversion, and lay a foundation for geophysical real-time inversion interpretation.
Description
Technical Field
The invention belongs to the technical field of earth detection, and particularly relates to a UNet-based receiving function and surface wave frequency dispersion joint inversion method.
Background
The receive function is a reflection seismology-like technique for detecting the structure of the crust and upper mantle below the station. Generally speaking, a teleseismic body wave comprises a seismic source time function and a comprehensive effect of medium attribute influence on a propagation path, the purpose of a receiving function is to remove the influence of the seismic source time function and instruments and obtain P-SV conversion seismic phases from the lower part of a station, and the seismic phases comprise velocity information of a main discontinuous interface in a rock ring. In contrast to other seismic methods, the receive function requires only a single seismic station to achieve depth sounding. Therefore, the receiving function is widely applied to rock ring structure detection and the study of the morphology of the Mohol surface.
The inversion of the receiving function is generally considered to have strong non-uniqueness, which results in that the receiving function has weak constraint on the underground absolute velocity and the inversion has strong dependence on an initial model. The surface wave dispersion can well reflect the overall seismic wave velocity value of the underground geologic body, but cannot well invert the interface position of the geologic body; the receive function is the opposite, it can reflect well the position of the interface under the station, but lacks the constraint ability to absolute velocity, the insensitivity comes from the trade-off property of velocity and depth, i.e. shallow low-speed structure and deep high-speed structure have almost the same receive function response, and the simultaneous inversion of receive functions with different slownesses is not enough to eliminate the velocity uncertainty in depth. Therefore, the joint inversion of the surface wave and the receiving function has natural rationality, the receiving function and the surface wave frequency dispersion of the same station are jointly inverted, the advantage complementation of the receiving function and the surface wave frequency dispersion is realized, and S-wave speed information below the station can be better restrained.
Neural network algorithms are one of the main contents of deep learning. In recent years, there have been a number of applications for inversion of geophysical data using neural network techniques. These applications have focused mainly on the field of seismic exploration, electromagnetic exploration and natural seismic research. Although machine learning has been widely applied to inversion of various single methods of geophysical, there is no systematic study on joint inversion of a receiving function and surface wave dispersion by using a deep learning technique at present. The supervised learning method can establish effective mapping between data and a model by utilizing a large number of data sets obtained by a forward algorithm so as to complete inversion, and the thought is widely applied to deep learning inversion and produces a large amount of results. More importantly, the same neural network framework is easily migrated to different datasets, even if the two data come from completely different domains. This attribute allows us to build the same neural network model for joint inversion of multiple data, such as the receive function and surface wave dispersion joint inversion.
Disclosure of Invention
In order to solve the problems, the invention provides a receiving function and surface wave frequency dispersion joint inversion method based on UNet.
The technical scheme of the invention is as follows: a receiving function and surface wave frequency dispersion joint inversion method based on UNet comprises the following steps:
s1: constructing a speed model, and performing forward calculation to obtain a training data set;
s2: and constructing the UNet, training by using a training data set, and taking the receiving function to be inverted and the surface wave dispersion data as the input of the UNet to obtain the velocity structures of the crust and the upper mantle.
Further, step S1 comprises the following sub-steps:
s11: constructing a four-layer speed model comprising an upper crust, a middle crust, a lower crust and an upper mantle;
s12: interpolating the four-layer velocity model to obtain a first data set and a second data set;
s13: and performing forward calculation of a receiving function and the surface wave frequency dispersion on the first data set and the second data set to obtain a training data set.
Further, in step S12, when the disturbance of the velocity model is 10%, the four-layer velocity model is interpolated to obtain a first data set.
Further, in step S12, when the disturbance of the velocity model is 20%, each layer of the velocity model is divided into two sublayers to obtain an eight-layer velocity model, and the eight-layer velocity model is interpolated to obtain a second data set.
Further, in step S2, the UNet includes 17 convolutional layers, 3 pooling layers, and 3 transposed convolutional layers.
Further, in step S2, the ELU function is an activation function of the convolution layer of UNet.
Further, in step S2, the objective function of UNet in the training process
The calculation formula of (c) is:
where N represents the total number of training samples, m i Represents the corresponding label of the ith sample, D (-) represents the inversion operator realized by deep learning, D i Representing the input data and theta representing the hyperparameter that needs to be updated in the UNet backpropagation.
The beneficial effects of the invention are: the UNet-based receiving function and surface wave frequency dispersion joint inversion method provided by the invention can reduce non-uniqueness in inversion, improve inversion efficiency and stability, explore the applicability of a deep learning method in joint inversion, can be used as a substitute for conventional inversion, improve the defects of the conventional inversion and lay a foundation for geophysical real-time inversion interpretation.
Drawings
FIG. 1 is a flow chart of a method for joint inversion of a receive function and a surface wave dispersion;
FIG. 2 is a schematic view of a synthetic one-dimensional laminar mantle model for UNet training;
FIG. 3 is a framework of a deep learning model for joint inversion;
FIG. 4 (a) is a diagram showing the evolution of the loss function during training;
FIG. 4 (b) is a histogram of the predicted results and label residuals of a portion of the test set during the training process;
fig. 5 is a diagram illustrating UNet prediction results and uncertainty estimation.
Detailed Description
The embodiments of the present invention will be further described with reference to the accompanying drawings.
Forward calculation: and calculating to obtain a receiving function and surface wave frequency dispersion data by utilizing a seismic wave propagation theory based on the earth velocity model.
As shown in fig. 1, the present invention provides a method for jointly inverting a receiving function and a surface wave dispersion based on a UNet network, comprising the following steps:
s1: constructing a speed model, and performing forward calculation to obtain a training data set;
s2: and (3) constructing a UNet, training by using a training data set, and taking a receiving function to be inverted and surface wave frequency dispersion data as the input of the UNet to obtain the velocity structures of the crust and the upper mantle.
In an embodiment of the present invention, step S1 includes the following sub-steps:
s11: constructing a four-layer speed model comprising an upper crust, a middle crust, a lower crust and an upper mantle;
s12: interpolating the four-layer velocity model to obtain a first data set and a second data set;
s13: and performing forward calculation of a receiving function and surface wave frequency dispersion on the first data set and the second data set to obtain a training data set.
Step S11 to step S13 are methods of constructing the velocity model.
In the embodiment of the present invention, in step S12, when the disturbance of the velocity model is 10%, the four-layer velocity model is interpolated to obtain the first data set.
In the embodiment of the present invention, in step S12, when the disturbance of the velocity model is 20%, each layer of the velocity model is divided into two sublayers, so as to obtain an eight-layer velocity model, and the eight-layer velocity model is interpolated, so as to obtain a second data set.
In the embodiment of the invention, for any neural network model, the training data set occupies a core position in the optimization of the model performance, contains the characteristics of the model needing to be learned, and determines the prediction effect to a great extent. When the data is good enough, the parameter space of the model has a large elasticity. Within a certain model hyper-parameter space, the precision difference of the training results is small although different model parameters are set.
For geophysical inversion, the core problem of training data is how to set a set of geological models that is reasonable and as complete as possible. In the invention, the training data set is a receiving function waveform obtained by generating a large number of models on the basis of a standard one-dimensional layered earth model and performing forward modeling. As shown in FIG. 2, a basic four-layer velocity model is first established comprising upper, middle and lower hulls, and an upper mantle, the S-wave velocity V of the upper mantle s The velocity of the rest layers changes randomly within a given disturbance R (10%) on the basis of a standard earth reference model. Corresponding P wave velocity V p According to a fixed ratio V p /V s Density calculated as 1.73 from empirical relationship ρ =0.77+0.32v s Obtaining; and then, interpolating each four-layer model to obtain the final 80-layer model with the layer thickness of 0.75km, and generating 10000 models as a first data set, wherein the models enable the network to learn the most basic structural characteristics of the shell mantle. In order to improve the capability of the model for predicting the complex speed structure, the value of R is expanded to 20%, then the basic four-layer model is expanded into eight layers, namely each layer is divided into two sublayers, the speed and the depth of each layer randomly change within the range of R, interpolation is carried out on each eight-layer model to obtain the final 80-layer model, and 40000 models are generated as a second data set.
The present invention uses a seismology computer program package to forward calculate the receive function and the surface wave dispersion for these models (i.e., knowing the velocity model to solve for the observed response). The effective time window of the receiving function is 5 seconds before the arrival of the seismic phase of the P wave and 25 seconds after the arrival, each receiving function has 300 sampling points in total, and the sampling frequency is 10Hz. In the forward calculation of the model, the ray parameters are set to random values from 0.04 to 0.075 to improve the generalization capability of the model. The periodic range of the surface wave dispersion is 10-40 s; the number of the frequency points is 16 in total and is set to be equal interval. Finally, the 300 receive functions and the 16 dispersion data are interpolated to 80 samples as two inputs to UNet.
In the embodiment of the present invention, in step S2, the UNet includes 17 convolutional layers, 3 pooling layers, and 3 transposed convolutional layers.
In the embodiment of the present invention, in step S2, the ELU function is an activation function of the convolution layer of UNet.
In the embodiment of the invention, in step S2, the objective function of UNet in the training process
The calculation formula of (2) is as follows:
where N represents the total number of training samples, m i Represents the corresponding label of the ith sample, D (-) represents the inversion operator realized by deep learning, D i Representing the input data and theta representing the hyperparameter that needs to be updated in the UNet backpropagation.
The framework of the depth model for joint inversion and the framework of the separately inverted receive function are shown in fig. 3. The network comprises 4 layers on the left and the right, and totally comprises 17 convolutional layers, 3 pooling layers and 3 transposition convolutional layers. Because a large number of negative numbers exist in the received function data, an ELU activation function is selected to act on the output of the convolution layer, and the activation function can avoid information loss and gradient disappearance of a negative value part. The convolution operation is followed by a Batch _ normalization to normalize the data to further prevent gradient disappearance or gradient explosion phenomena, while possibly increasing the regularization effect.
The receive function and the surface wave dispersion can be considered as two parallel input channels because the two data reflect the true regularity of the model from different angles. As if the input of a picture were composed of three channels representing different pixels, each channel containing a partial feature of the picture.
Fig. 4 (a) shows the evolution of the loss function in the training process, and it can be seen that similar to the individual inversion of the receive function, the errors of the training set and the test set of the joint inversion gradually decrease with the increase of the iteration number, and finally are both lower than 0.001, which indicates that the model can well learn the features in the training set. The residuals of the prediction results and the labels of part of the test sets are shown in fig. 4 (b), the histogram is normally distributed, and the maximum is located at almost 0, which indicates the least square assumption of the residual error and the good performance of the algorithm, and also indicates that the joint inversion can well constrain the Vs absolute velocity value.
After the UNet model is trained, a new receiving function and surface wave dispersion data are used as input, and the network can quickly give an output earth crust velocity model, namely a prediction result, through forward propagation.
To illustrate the effectiveness and applicability of the proposed inversion scheme, the present invention, as shown in FIG. 5, presents multiple synthetic model inversion results from a test data set, depth is Depth, V s Is the S-wave velocity, the bold line represents the true model, the bold line with small dot marks represents the prediction result, i.e. the average of 100 models (indicated by thin lines) that apply 10% random noise estimates to the observed data. The uncertainty of the inversion is further estimated by applying 10% random noise to the "observed" receive function. The disturbance δ r obeys a uniform distribution δ r U (- σ, + σ), where σ is the uncertainty of each point correspondence. Each model corresponds to a receiving function waveform and a surface wave frequency dispersion curve of 100 disturbances, and the receiving function waveform and the surface wave frequency dispersion curve are used as input of U-Net to predict the 1D shear wave speed structure. The predicted model forms a narrow band, which shows that U-Net has certain anti-noise capability and can better predict underground V s And (4) distribution.
It will be appreciated by those of ordinary skill in the art that the embodiments described herein are intended to assist the reader in understanding the principles of the invention and are to be construed as being without limitation to such specifically recited embodiments and examples. Those skilled in the art can make various other specific changes and combinations based on the teachings of the present invention without departing from the spirit of the invention, and these changes and combinations are within the scope of the invention.
Claims (7)
1. A receiving function and surface wave dispersion joint inversion method based on UNet is characterized by comprising the following steps:
s1: constructing a speed model, and performing forward calculation to obtain a training data set;
s2: and constructing the UNet, training by using a training data set, and taking the receiving function to be inverted and the surface wave dispersion data as the input of the UNet to obtain the velocity structures of the crust and the upper mantle.
2. The UNet-based receive function and surface wave dispersion joint inversion method of claim 1, wherein the step S1 comprises the following sub-steps:
s11: constructing a four-layer speed model comprising an upper crust, a middle crust, a lower crust and an upper mantle;
s12: interpolating the four-layer velocity model to obtain a first data set and a second data set;
s13: and performing forward calculation of a receiving function and the surface wave frequency dispersion on the first data set and the second data set to obtain a training data set.
3. The UNet-based receive function and surface wave dispersion joint inversion method of claim 2, wherein in step S12, when the disturbance of the velocity model is 10%, the four-layer velocity model is interpolated to obtain the first data set.
4. The UNet-based receive function and surface wave dispersion joint inversion method of claim 2, wherein in step S12, when the disturbance of the velocity model is 20%, each layer of velocity model is divided into two sub-layers to obtain an eight-layer velocity model, and the eight-layer velocity model is interpolated to obtain the second data set.
5. The UNet-based receive function and surface wave dispersion joint inversion method of claim 1, wherein in step S2, UNet includes 17 convolutional layers, 3 pooling layers, and 3 transpose convolutional layers.
6. The UNet-based reception function and surface wave dispersion joint inversion method according to claim 5, wherein in the step S2, the ELU function is an activation function of a convolution layer of UNet.
7. The method of joint inversion of UNet-based receive function and surface wave dispersion as claimed in claim 1, wherein in step S2, the objective function of UNet is trained during training
The calculation formula of (2) is as follows:
where N represents the total number of training samples, m i Represents the corresponding label of the ith sample, D (-) represents the inversion operator realized by deep learning, D i Representing the input data and theta representing the hyperparameter that needs to be updated in the UNet backpropagation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211145748.1A CN115453623A (en) | 2022-09-20 | 2022-09-20 | UNet-based receiving function and surface wave frequency dispersion joint inversion method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211145748.1A CN115453623A (en) | 2022-09-20 | 2022-09-20 | UNet-based receiving function and surface wave frequency dispersion joint inversion method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115453623A true CN115453623A (en) | 2022-12-09 |
Family
ID=84305259
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211145748.1A Pending CN115453623A (en) | 2022-09-20 | 2022-09-20 | UNet-based receiving function and surface wave frequency dispersion joint inversion method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115453623A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116520402A (en) * | 2023-06-02 | 2023-08-01 | 大连理工大学 | Multi-vibration phase wave field inversion method considering bulk wave and surface wave under rock half-space field |
| CN117631029A (en) * | 2024-01-26 | 2024-03-01 | 中国铁路设计集团有限公司 | Rayleigh surface wave dispersion curve inversion method based on joint algorithm |
-
2022
- 2022-09-20 CN CN202211145748.1A patent/CN115453623A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116520402A (en) * | 2023-06-02 | 2023-08-01 | 大连理工大学 | Multi-vibration phase wave field inversion method considering bulk wave and surface wave under rock half-space field |
| CN116520402B (en) * | 2023-06-02 | 2023-11-28 | 大连理工大学 | Multi-vibration phase wave field inversion method considering bulk wave and surface wave under rock half-space field |
| CN117631029A (en) * | 2024-01-26 | 2024-03-01 | 中国铁路设计集团有限公司 | Rayleigh surface wave dispersion curve inversion method based on joint algorithm |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115453623A (en) | UNet-based receiving function and surface wave frequency dispersion joint inversion method | |
| Brazell et al. | A machine-learning-based approach to assistive well-log correlation | |
| Wang et al. | Data-driven S-wave velocity prediction method via a deep-learning-based deep convolutional gated recurrent unit fusion network | |
| CN111948712B (en) | Pre-stack linear inversion method based on depth domain seismic record | |
| US20220413172A1 (en) | Method and system for seismic imaging using s-wave velocity models and machine learning | |
| CN108369287B (en) | Determining nodal depth and water column transit velocity | |
| Vamaraju et al. | Unsupervised physics-based neural networks for seismic migration | |
| CN110895348A (en) | Method, system and storage medium for extracting low-frequency information of seismic elastic impedance | |
| CN116047583A (en) | Adaptive wave impedance inversion method and system based on depth convolution neural network | |
| WO2022187685A1 (en) | Method and system for faster seismic imaging using machine learning | |
| Liu et al. | Physics-driven self-supervised learning system for seismic velocity inversion | |
| Wang et al. | Unsupervised learning for seismic internal multiple suppression based on adaptive virtual events | |
| Bazulin et al. | Determination of the elastic parameters of a VTI medium from sonic logging data using deep learning | |
| CN113625336A (en) | Seismic wave impedance thin layer inversion method based on full convolution neural network | |
| Shiroodi et al. | Shear wave prediction using committee fuzzy model constrained by lithofacies, Zagros basin, SW Iran | |
| CN111045084A (en) | Multi-wave self-adaptive subtraction method based on prediction feature extraction | |
| Li et al. | Pertinent multigate mixture-of-experts-based prestack three-parameter seismic inversion | |
| Meng et al. | Seismic impedance inversion using a multi‐input neural network with a two‐step training strategy | |
| Malvić et al. | Neural networks in petroleum geology as interpretation tools | |
| Jun Park et al. | Realistic synthetic data generation using neural style transfer: Application to automatic fault interpretation | |
| Pérez et al. | 3D initial salt body building for full-waveform inversion assisted by deep learning | |
| CN113296150B (en) | High-dimensional closed-loop network seismic inversion method under logging constraint | |
| CN113985480B (en) | AVO inversion method and device based on angle correction | |
| CN115061200B (en) | Method for suppressing interlayer multiple based on virtual homophase axis method and unsupervised neural network | |
| US11846740B2 (en) | First break picking for full wavefield travel-time inversion |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |