A new 3D earthquake travel-time simulation method based on PIUFNO

Abstract

The present invention is applicable to the field of seismic exploration technology, and provides a novel 3D seismic travel time simulation method based on PIUFNO, including the following specific steps: step 1, establishing a PIUFNO network; step 2, determining the input and output of PIUFNO; step 3, determining the loss function of PIUFNO, performing network training and outputting the results; step 4, inputting a new model after the network training and performing seismic travel time simulation. This method effectively solves the problems of difficulty and low efficiency in simulating the travel time of a 3D model caused by gridding limitations of traditional numerical methods. After one training, the 3D seismic travel time of multiple earthquake sources can be quickly obtained by inputting different velocity models and corresponding background travel times, which significantly reduces the calculation cost and realizes a more efficient and practical 3D seismic travel time simulation. This efficiency improvement is of great significance to enhancing the overall effect of 3D seismic exploration and its practical application in the fields of oil and gas resource development.

Description

Technical Field

[0001] The invention belongs to the technical field of seismic exploration, and in particular relates to a novel three-dimensional seismic travel time simulation method based on PIUFNO. Background Art

[0002] Oil and gas resources are strategic resources that are essential for national development, so oil and gas exploration is an important task that affects the national economy and people's livelihood. By using the properties of seismic wave propagation, geophysical methods such as seismic imaging can be used to detect the internal structure of the earth. Seismic travel time is an important information transmitted by seismic waves, and seismic travel time simulation has always been an important topic of cross-scale research in geophysics.

[0003] The basis of earthquake travel time simulation is to solve the Eikonal equation, which is a nonlinear first-order partial differential equation. There are two most commonly used numerical solution methods: the fast march method (FMM) and the fast scanning method (FSM). Both methods are based on finite difference approximation and require the underground model to be meshed. In addition, numerical algorithms have theoretical limitations and can usually only perform travel time simulation on a single model and a fixed source. With the increase of dimensions, especially when performing three-dimensional travel time simulation, the computational complexity and cost increase significantly, making it more difficult to achieve efficient and accurate three-dimensional simulation. Therefore, a new three-dimensional earthquake travel time simulation method based on the physical information Unet Fourier neural operator (PIUFNO) is proposed. Summary of the invention

[0004] The purpose of the embodiment of the present invention is to provide a novel 3D seismic travel time simulation method based on PIUFNO, aiming to solve the problems raised in the above background technology.

[0005] The embodiment of the present invention is implemented as follows: a novel 3D seismic travel time simulation method based on PIUFNO can obtain seismic travel time information by solving the Eikonal equation in an isotropic medium. The Eikonal equation can be expressed as:

[0006]

[0007] Where T represents the travel time of the earthquake first arrival wave, v represents the velocity model, represents the gradient operator, x = (x, y, z) and x s =(x s ,y s ,z s ) are the coordinates of the model and the source, T(x s ) = 0 means that the time at the source is zero. In order to avoid the singularity of the source, T is decomposed into the additive form T = T 0 +τ, τ is the factorization time factor, T 0is the background travel time corresponding to the velocity value at the earthquake source, and the expression is:

[0008]

[0009] Among them, |xx s | is the seismic wave propagation from position x to position x s The distance, v 0 is the velocity at the earthquake source. 0 +τ is substituted into formula 1, and the factorized eikonal equation can be expressed as:

[0010]

[0011] Among them, τ(x s )=0 means that the factorization factor at the earthquake source is zero.

[0012] The method comprises the following specific steps:

[0013] Step 1. Establish PIUFNO network;

[0014] Step 2: Determine the input and output of PIUFNO;

[0015] Step 3: Determine the loss function of PIUFNO and perform network training;

[0016] Step 4: After the network training is completed, input the new model and perform earthquake travel time simulation.

[0017] In a further technical solution, in step 1, PIUFNO converts the velocity model v and the background travel time T 0 As input, the corresponding perturbation travel time τ = TT 0 As output, the mapping between them is established through the Unet Fourier Neural Operator (UFNO);

[0018] In local transformation, p j p j and p j+1 is the input and output of the current module, input p j After forward Fourier transform, linear high-frequency filtering, inverse Fourier transform, U-net feature transform and local linear transform, the output p is finally obtained. j+1 .

[0019] In a further technical solution, in step 2, the input includes the velocity model v and the background travel time T 0 , the velocity model v provides velocity information, and the background travel time T 0 The source location information is provided, and after completing the PIUFNO training, the corresponding model and the disturbance travel time τ of the corresponding source are directly output.

[0020] In a further technical solution, in step 3, the physical information loss function is as follows:

[0021]

[0022] Among them, N v is the number of velocity models used, iv is the velocity model index value;

[0023] A further technical solution is that in step 4, after the network training is completed, different velocity models v and corresponding background travel times T are input. 0 , output the disturbance travel time τ of the corresponding model and the corresponding earthquake source, and finally after the addition operation T = T 0 +τ to obtain the simulated travel time results.

[0024] The embodiment of the present invention provides a novel 3D seismic travel time simulation method based on PIUFNO, which effectively solves the problem of low efficiency of traditional numerical methods that they can only simulate one model at a time. The present invention can realize 3D seismic travel time simulation of multiple earthquake sources through one training, and is applicable to different velocity models. This method significantly reduces the computational cost and improves the efficiency and practicality of 3D seismic travel time simulation. When testing typical geological velocity models, the method shows excellent performance and can accurately simulate 3D seismic travel time at any earthquake source position. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a 3D travel time simulation framework based on PIUFNO;

[0026] Figure 2 It is the Unet Fourier neural operator module;

[0027] Figure 3 The velocity model for network input (a and b are two different velocity models);

[0028] Figure 4 is the background travel time input to the network (a and b are the background travel times corresponding to the two velocity models respectively);

[0029] Figure 5 The velocity model and background travel time input after training and the travel time predicted by PIUFNO simulation (a is the velocity model, b is the corresponding background travel time T 0 , c is the travel time simulated by PIUFNO). DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0031] The specific implementation of the present invention is described in detail below in conjunction with specific embodiments.

[0032] An embodiment of the present invention provides a novel 3D seismic travel time simulation method based on PIUFNO. In an isotropic medium, seismic travel time information can be obtained by solving the Eikonal equation. The Eikonal equation can be expressed as:

[0033]

[0034] Where T represents the travel time of the earthquake first arrival wave, v represents the velocity model, represents the gradient operator, x = (x, y, z) and x s =(x s ,y s ,z s ) are the coordinates of the model and the source, T(x s ) = 0 means that the time at the source is zero.

[0035] In order to avoid the singularity of the earthquake source, T is decomposed into the multiplication form T = T 0 +τ, τ is the factorization time factor, T 0 is the background travel time corresponding to the velocity value at the earthquake source, and the expression is:

[0036]

[0037] Among them, |xx s | is the seismic wave propagation from position x to position x s The distance, v 0 is the velocity at the earthquake source. 0 +τ is substituted into formula 1, and the factorized eikonal equation can be expressed as:

[0038]

[0039] Among them, τ(x s )=0 means that the factorization factor at the earthquake source is zero.

[0040] The method comprises the following specific steps:

[0041] Step 1. Establish PIUFNO network;

[0042] The established PIUFNO network is shown in the figure below: Figure 1As shown, PIUFNO combines the velocity model v and the background travel time T 0 As input, the corresponding perturbation travel time τ = TT 0 As output, the mapping between them is established through the Unet Fourier Neural Operator (UFNO).

[0043] The core of PIUFNO is the UFNO block, such as Figure 2 As shown, where F represents the forward fast Fourier transform; F -1 represents inverse Fourier transform; R represents linear transform to filter out high-frequency components; U represents U-net feature transform; W represents local linear transform; σ represents Relu activation function; + represents input p j After three transformations, the results are added together. j and p j+1 is the input and output of the current module, input p j After forward Fourier transform, linear high-frequency filtering, inverse Fourier transform, U-net feature transform and local linear transform, the output p is finally obtained. j+1 .

[0044] Step 2: Determine the input and output of PIUFNO;

[0045] The purpose is to use PIUFNO to simulate the travel time of multiple earthquake sources and arbitrary velocity models at the same time, and to establish the input [v; T 0 ] and the output τ. The input v provides speed information, T 0 Provided with the earthquake source location information, after completing the PIUFNO training, the corresponding model and the disturbance travel time τ of the earthquake source can be directly output. Finally, after a simple addition operation T = T 0 +τ can get the final travel time result.

[0046] Step 3: Determine the loss function of PIUFNO and perform network training;

[0047] The deep learning framework based on physical information does not need to generate a large amount of training data and manual labels, which reduces the cost of data generation and enhances the interpretability of the deep learning framework. The physical information loss function is as follows:

[0048]

[0049] Among them, N v is the number of velocity models used, and iv is the velocity model index value.

[0050] Step 4: After the network training is completed, input the new model and perform earthquake travel time simulation.

[0051] After the training is completed, by inputting different velocity models v and the corresponding background travel time T 0 , output the disturbance travel time τ of the corresponding model and the corresponding earthquake source, and finally after the addition operation T = T 0 +τ to obtain the simulated travel time results.

[0052] To verify the effectiveness of this method, Figure 3 a and Figure 3 b shows the two velocity models and the background travel time T corresponding to the two velocity models 0 (Respectively as Figure 4 a and 4b) are input into the network for training. This model contains 60×60×60 grid points, the spatial sampling interval is 25m, the number of training times is 500, and the Adam optimizer is used for network optimization training. The learning rate is 0.0001, and after 250 training times, the learning rate becomes half of the original.

[0053] By training PIUFNO, we can predict the travel time of earthquakes with different source locations and different velocity models. Figure 5 The velocity model shown in a and Figure 5 b shows the corresponding background travel time T 0 Input into the network, set the source location at the center (0.75, 0.75, 0.75), and the travel time simulated by PIUFNO is as follows Figure 5 As shown in c.

[0054] According to experiments, the time required for this method to simulate the travel time using the displayed model is 0.054 seconds, while the traditional numerical method takes 0.095 seconds to simulate the travel time of the same model, which greatly improves the overall calculation efficiency. This efficiency improvement is of great significance to enhancing the overall effect of 3D seismic exploration and its practical application in fields such as oil and gas resource development.

[0055] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A new 3D seismic travel time simulation method based on PIUFNO, characterized in that: The specific steps include: Step 1: Establish the PIUFNO network, namely the physical information U-net Fourier neural operator network; Step 2: Determine the input and output of PIUFNO; Step 3: Determine the loss function of PIUFNO and perform network training; Step 4: After the network training is completed, the new model is input and earthquake travel time simulation is performed; In step 1, PIUFNO converts the velocity model v And background time T 0 as input, the corresponding disturbance travel time τ = T - T 0 as output, and the mapping between them is established through UFNO, where T It represents the travel time of the earthquake first arrival wave, and UFNO is the U-net Fourier neural operator; In local transformation, and are the input and output of the current module, input After forward Fourier transform, linear high-frequency filtering, inverse Fourier transform, U-net feature transform and local linear transform, the output is finally obtained. ; In step 2, the input includes the velocity model v And background time T 0, speed model v Provides speed information, background travel time T 0 provides the source location information. After completing PIUFNO training, the corresponding model and the disturbance travel time of the corresponding source are directly output. τ ; In step 3, the physical information loss function is as follows: ； in, N v is the number of velocity models used, iv is the velocity model index value, That is, it represents the loss function, represents the gradient operator, Represents the sum of the values ​​output by each model; In step 4, after the training is completed, the new velocity model v and the background travel time are input. T 0, output the disturbance travel time of the corresponding model and the corresponding source τ , and finally after addition operation T = T 0+ τ Get the final simulated travel time result.

Images