Waveform protected local scale traveltime inversion method independent of source wavelet

Abstract

The application discloses a waveform-protected non-source wavelet local scale traveltime inversion method, relates to a non-source wavelet full waveform inversion method established by using a convolution and deconvolution method, and is combined with local scale traveltime inversion, and simultaneously solves the period jump of full waveform inversion and the source wavelet dependence problem. First, the influence of the source wavelet in the observation data and the simulation data is eliminated by using the convolution and deconvolution method, and the original seismic data waveform characteristics are protected. Secondly, local scale traveltime information of the observation data and the simulation data is extracted, and a waveform-protected non-source wavelet local scale traveltime inversion objective function is constructed. Then, the partial derivative of the waveform-protected non-source wavelet local scale traveltime inversion objective function to the velocity parameter is deduced in detail, and the corresponding underground velocity parameter gradient operator of the adjoint state method is deduced. Finally, the stable update of the velocity model parameter is realized by using an L-BFGS optimization algorithm. Finally, the effectiveness of the method is verified by the seismic data test.

Description

Technical Field

[0001] This invention relates to a waveform-protected seismic wavelet-independent local-scale travel time inversion method. Specifically, it is a seismic inversion method that is independent of the seismic wavelet and is established by using convolution and deconvolution methods. This method is combined with a local-scale travel time inversion method to obtain high-precision inversion of seismic wave velocity parameters. Background Technology

[0002] Full waveform inversion (FWI) is an advanced technique that utilizes waveform information from seismic data to construct a seismic inversion objective function and obtain high-precision inversion results of subsurface physical parameters (Lailly, 1983; Tarantola, 1984; Plessix, 2006). Despite significant progress in FWI, as a highly nonlinear optimization problem, its application still faces numerous challenges. One core difficulty is the lack of low-frequency seismic data with high signal-to-noise ratios. This lack of low frequencies often causes the inversion process to get trapped in local minima, posing a significant challenge to practical applications (Pratt et al., 1998; Virieux and Operato, 2009). In particular, when seismic data lacks low-frequency components below 3 Hz, waveform matching problems are easily triggered, leading to severe period jumps. Furthermore, the accuracy of the source wavelet also plays a crucial role in the waveform matching accuracy of FWI; errors in the source wavelet will severely affect the accuracy of velocity modeling.

[0003] To address the period jump problem in FWI (Focus-In-Weight Seismic) analysis, previous researchers have conducted extensive studies and proposed various methods to mitigate it. These mainly include: frequency-division multi-scale inversion methods, weakly nonlinear objective functions, and low-frequency reconstruction of seismic data. Multi-scale FWI methods first utilize low-frequency seismic data and then gradually increase the frequency of the seismic data participating in FWI, achieving a scaled update of the velocity model from low wavenumber components to high wavenumber components (Bunks et al., 1995; Pratt et al., 1998). Therefore, frequency-division multi-scale FWI methods can effectively alleviate the wavenumber discontinuity problem in velocity model inversion caused by directly using full-band seismic data for FWI, thereby improving the inversion accuracy of the velocity model. Regarding the weakly nonlinear objective function, it can be achieved by decomposing complex seismic wavefields or increasing the similarity between observed and simulated data, thereby reducing the nonlinearity of the objective function and mitigating the risk of the FWI algorithm getting trapped in local minima. Examples include travel-time inversion, first-arrival inversion, early-arrival inversion, layer stripping inversion based on attenuation functions, and phase-corrected FWI (Sheng et al., 2006; Yang et al., 2018; Choi and Alkhalifah, 2015; Luo et al., 2018; Hu et al., 2018). In addition, there are low-frequency reconstruction FWI methods for seismic data, which can reconstruct missing low-frequency components from seismic data lacking low-frequency signals, such as envelope inversion methods (Wu et al., 2014; Hu et al., 2022).

[0004] The travel time information of seismic waves reflects the low wavenumber velocity structure and is the basis for improving the initial velocity model of FWI (Luo and Schuster, 1991). However, the resolution of travel time inversion results is relatively low, and the detection depth is limited (Sheng et al., 2006; Leung et al., 2006; Luo et al., 2016; Jiang and Zhang, 2018). To address this issue, Luo and Schuster (1991) used cross-correlation to calculate travel time differences and then developed a travel time inversion method based on the wave equation to improve the inversion resolution. Zhang et al. (2018) proposed a time-delayed FWI method, using a window function to extract refracted waves to update the velocity model of strong contrast salt domes, effectively alleviating the cycle skipping problem in FWI. Hu et al. (2020) converted seismic data to a local scale and proposed a local-scale travel time inversion method based on the wave equation, effectively handling the interference of multiple seismic events in cross-correlation calculations and using full-wavefield travel time information to obtain low wavenumber components in deep regions. Wu et al. (2022) introduced an elastic time-delay inversion method, which effectively improved the inversion accuracy of salt dome boundary information. Hu et al. (2024) proposed a local-scale travel-time inversion method based on elastic waves, which can simultaneously obtain the macroscopic structure of P-wave and S-wave models. Hu et al. (2025) combined the simultaneous source strategy with the local-scale travel-time inversion method of elastic wave field, which significantly improved the computational efficiency.

[0005] Furthermore, the accuracy of the source wavelet directly affects the inversion accuracy of FWI. Choi et al. (2005) proposed a frequency-domain source wavelet-independent FWI method, which eliminates the influence of the wavelet on the inversion results by dividing the absolute value of the amplitude of the frequency-domain seismic data by the reference trace. Choi and Alkhalifah (2011) proposed using the reference trace and convolving it with the seismic data to make the observed and simulated data have the same source wavelet, thus realizing a source wavelet-independent FWI method. Choi and Min (2012) constructed a logarithmic objective function in the frequency domain and extended the source wavelet-independent inversion method to elastic media. Aorui et al. (2015) combined the source wavelet-independent method with envelope inversion, achieving low wavenumber component velocity modeling while avoiding the period jump problem caused by inaccurate source wavelets. Zhang et al. (2016) proposed that only the direct wave waveform needs to be used as reference trace information and convolved with elastic wave seismic data to obtain a source wavelet-independent elastic wave FWI method. Zhang et al. (2018) combined the source-independent wavelet convolution concept with the direct envelope inversion method, successfully eliminating the influence of wavelet inaccuracy on the direct envelope inversion method. Li Qingyang et al. (2017) and Fang et al. (2019) applied the source-independent wavelet convolution method to the elastic wave least squares reverse time migration method, effectively solving the problem of wavelet inaccuracy in migration imaging. Jiang et al. (2023) combined the source-independent wavelet convolution method with Wasserstein distance FWI, alleviating the dual problems of period jump and wavelet inaccuracy to some extent. Song et al. (2026) combined the source-independent wavelet theory method with the deep learning FWI method based on physical information constraints, eliminating the influence of the source wavelet on FWI within the theoretical framework of automatic differentiation.

[0006] Because it is difficult to extract accurate source wavelets from actual seismic data, and because the source wavelets of each shot are different during actual seismic exploration operations, the accuracy of FWI velocity modeling is severely affected. Furthermore, local-scale travel-time inversion requires the extraction of local-scale travel-time information through window functions. Directly using convolution methods independent of source wavelets would significantly alter the waveform characteristics of the seismic data, leading to travel-time calculation errors and affecting the final accuracy of velocity modeling. Therefore, how to achieve a local-scale travel-time inversion method independent of source wavelets while preserving the original waveform characteristics has become a pressing technical problem for those skilled in the art. Summary of the Invention

[0007] To address the above problems, this invention proposes a waveform-protected local-scale travel time inversion method that does not rely on the source wavelet. While preserving the original characteristics of the waveform, it realizes a local-scale travel time inversion method that does not rely on the source wavelet, obtains the low wavenumber component of the subsurface velocity model, and alleviates the periodic jump problem of FWI.

[0008] The technical solution of this invention is as follows: First, the influence of the source wavelet in the observation data and simulation data is eliminated by using convolution and deconvolution methods to protect the original seismic waveform characteristics; second, the local scale travel time information of the observation data and simulation data is extracted to construct the waveform protection objective function that is independent of the local scale travel time of the source wavelet; then, the partial derivative of the waveform protection objective function that is independent of the local scale travel time of the source wavelet with respect to the velocity parameters and the gradient operator of the underground velocity parameters corresponding to the adjoint state method are derived in detail; finally, the L-BFGS optimization algorithm is used to achieve stable updating of the velocity model parameters.

[0009] Specifically, the following steps are included: Step 1: Use Geoeast seismic data processing software to remove noise from seismic data, and use the processed seismic data as the data input for the waveform-protected local scale travel time inversion method without source wavelet dependence; Step 2: Based on the first arrival travel time information of the seismic data, construct an initial velocity model as the model input for the forward modeling simulation of the source-protected wavelet local scale travel time inversion method. Step 3: Set the source wavelet as the wavelet input for the forward modeling simulation of the waveform protection method that does not rely on the local scale travel time inversion of the source wavelet; Step 4: Set up the seismic exploration and observation system, and perform forward modeling using the source wavelet and initial velocity model, storing the forward propagation wave field. The equation for the isotropic medium wave with constant density can be expressed as: , Where v is the velocity of sound; u is the sound wave field; x and z represent the horizontal and vertical directions, respectively; and t represents time. The finite difference algorithm is used to solve the above partial differential equations to achieve forward modeling of seismic wave propagation underground. Step 5: Source-independent wavelet processing of seismic data: Extract the direct wave waveforms of the near-source traces from both the observed and simulated data as reference traces, and perform convolution and deconvolution processing on the seismic data. Specifically, the extracted direct wave waveforms of the near-source traces from the observed and simulated data can be represented as follows: , , in and These represent the simulated data and the observed data closest to the epicenter, respectively. For window functions, specifically, it can be expressed as: , in The cutoff time is determined by the arrival time of the direct seismic wave and can be read from the seismic record. 'n' represents the order of the window function; we choose n=10. A smaller n value results in a smoother window function, while a larger n value leads to more pronounced changes. Then, the reference trace is convolved and deconvolved with the observed and simulated data, respectively, to obtain observational and simulated data independent of the source wavelet. , , Where * denotes convolution; Indicates deconvolution; and Let represent the observed data and simulated data after convolution and deconvolution, respectively. Since seismic data can be viewed as the convolution of the Green's function and the source wavelet, the above equation can be expressed as: , , in Represent the Green's function; This represents the source wavelet. The source wavelets of both observed and simulated data after convolution and deconvolution are... This effectively eliminates the influence of the difference between the source wavelets of the observed data and the simulated data on the inversion results; Step 6: Local Scale Decomposition of Seismic Data: Using local scale transformation, extract the local scale waveform features of the observed and simulated data after source wavelet processing. The forward and inverse local scale transformations of the observed and simulated data are as follows: , , , , in This represents local-scale simulation data; This represents observational data at a local scale; Represents the Gaussian window function; The shift interval for the Gaussian window function; Step 7: Calculate the local travel time difference between the observed data and the simulated data: Use the cross-correlation algorithm to calculate the travel time difference between the two waveforms. , The travel time difference between observed data and simulated data ( It can be obtained using a cross-correlation algorithm: , in Indicates the maximum time difference. This represents the travel time difference between local-scale observed data and simulated data after processing independent of wavelet convolution and deconvolution. This indicates that there is no time difference, meaning the waveform matching is successful. If This indicates a travel time difference between the observed data and the simulated data, at which point the underground velocity field differs from the actual velocity field. Step 8: Constructing the Local-Scale Travel Time Inversion Objective Function: Since travel time information can better invert the low-wavenumber components of the subsurface velocity model, it can provide a better initial velocity model for FWI and alleviate the period jump problem during the inversion process. Therefore, the velocity inversion objective function can be constructed by calculating local-scale travel time information using seismic data processed by convolution and deconvolution, and can be expressed as follows: , in A direct mapping between the local-scale travel time difference and the seismic wavefield cannot be established. However, by using the implicit function differentiation rule and the cross-correlation function as the correlation equation, the mapping between the local-scale travel time difference and the seismic wavefield can be obtained. When the first derivative of the cross-correlation function with respect to time is 0, the following relationship holds: , Using the above equation as the correlation equation, the mapping relationship between local scale travel time difference and velocity parameters can be established: , in , , According to the chain rule, the partial derivatives of the local-scale seismic data after convolution and deconvolution processing with respect to the velocity parameter can be expressed as follows: , Input the results of convolution and deconvolution on the simulated data: , Further differentiation using the chain rule yields: , in Represents the cross-correlation algorithm; Step 10: Define the accompanying source for the waveform-protected wavelet-independent local-scale travel-time inversion method, and backpropagate the accompanying source to the model space to obtain the backpropagated wavefield, which is used to update the subsurface velocity model. Therefore, the corresponding accompanying source can be expressed as: ; Step 11: Calculate the gradient of the source-independent wavelet local-scale travel-time inversion objective function for waveform protection using the adjoint state method: , in This represents the propagating wave field generated by the earthquake source. This represents the backpropagating wavefield obtained through back propagation from the accompanying earthquake source. Finally, the gradient of the velocity parameters is obtained through zero-delay cross-correlation, and the velocity modulus is updated using an optimization algorithm. Step 12: Using the gradient information calculated in the previous step as input information for the L-BFGS optimization algorithm, update the velocity model. The iterative formula is as follows: , in Let L-BFGS be the velocity parameter at the k-th step. This is the step size for velocity updates. In the L-BFGS optimization algorithm, It is the inverse of an approximate Hessian matrix: , , , , , in To update the gradient, Through vector pairs and The inverse matrix of the approximate Hessian matrix is ​​calculated. It can be represented as: ; Step 13: Determine if the termination condition is met. If it is, output the result of the source-independent wavelet local scale travel time inversion based on waveform protection. If the termination condition is not met, continue to use the current inversion result as the initial velocity parameter model for the next cycle until the termination condition is met.

[0010] This invention effectively solves the problems of seismic waveform distortion caused by existing source-independent wavelet techniques, which lead to inaccurate calculation of local-scale travel time information and severely affect the accuracy of velocity modeling in travel time inversion. Therefore, based on previous research, this invention introduces convolution and deconvolution processing techniques into local-scale travel time inversion to protect waveform characteristics and establish a waveform-protected, source-independent wavelet-based local-scale travel time inversion objective function to recover accurate low-wavenumber components of the velocity model, providing a better initial velocity model for FWI.

[0011] The beneficial effects of this invention are as follows: This invention fully utilizes convolution and deconvolution processing techniques to protect waveform features, while also fully utilizing local-scale travel time information from the entire wavefield to invert the low wavenumber components of the subsurface velocity model. It also effectively solves the problem of severe waveform distortion caused by conventional methods that do not rely on the source wavelet. First, convolution and deconvolution processing is performed on the seismic data to ensure that the observed and simulated data have the same source wavelet, eliminating the influence of the source wavelet on local-scale travel time calculations. Second, local-scale transformation is performed on the seismic data after eliminating the source wavelet influence to extract local-scale travel time information from the observed and simulated data, constructing a local-scale travel time inversion objective function. Then, the partial derivatives of the waveform-protected, source-wavelet-independent local-scale travel time inversion objective function with respect to velocity parameters, as well as the velocity parameter gradient operator corresponding to the adjoint state method, are derived in detail. Finally, the L-BFGS optimization algorithm is used to synchronously update the velocity model, realizing a waveform-protected, source-wavelet-independent local-scale travel time inversion method, avoiding the influence of inaccurate source wavelets, and providing a better initial velocity model for conventional FWI methods. The Marmousi model was tested, which verified that the source-independent wavelet local scale travel time inversion method for waveform protection can obtain accurate low wavenumber components of the subsurface velocity model. Attached image description: Figure 1 This is a flowchart of a waveform protection method for local-scale travel time inversion of wavelet without relying on the seismic source. Figure 2 It is the Marmousi velocity model, (a) the actual velocity model; (b) the initial velocity model; Figure 3 These are waveform diagrams of the correct source wavelet and the incorrect source wavelet; Figure 4 These are the wavelet inversion results from the erroneous source: (a) FWI results; (b) local scale travel time inversion results. Figure 5 These are traditional travel-time inversion results that do not depend on the local scale of the source wavelet. The inversion results (4-6Hz) are corresponding to different Gaussian window lengths: (a) 0.4s; (b) 0.7s; (c) 1.1s; (d) 1.5s. Figure 6 The waveform protection is a local scale travel time inversion of the source wavelet, and the inversion results (4-6Hz) corresponding to different Gaussian window lengths are: (a) 0.4s; (b) 0.7s; (c) 1.1s; (d) 1.5s. Figure 7 It is waveform-protected local-scale travel time inversion of the source wavelet independent of the seismic source + multi-scale FWI results (4-15Hz), with Figure 6 As the FWI results of the initial velocity model, (a) with Figure 6 a represents the initial velocity model; (b) with... Figure 6b represents the initial velocity model; (c) with Figure 6 c represents the initial velocity model; (d) with Figure 6 d represents the initial velocity model. Detailed implementation method: To clearly illustrate the technical features of this patent, the following detailed description is provided through specific embodiments and in conjunction with the accompanying drawings.

[0012] FWI utilizes optimization algorithms for velocity updates, but its objective function exhibits strong nonlinear characteristics, making it prone to getting trapped in local minima during the inversion process. This leads to severe period jump problems in FWI, affecting the accuracy of the final velocity modeling. Furthermore, seismic data often struggles to acquire high signal-to-noise ratio (SNR) low-frequency components below 3Hz, causing significant waveform mismatches during the matching of observed and simulated data, resulting in erroneous velocity inversion results. In addition, the accuracy of the source wavelet directly impacts the accuracy of velocity modeling; incorrect source wavelets also cause mismatches between FWI observations and simulated data, leading to period jump problems. To address the period jump issues caused by missing low frequencies and inaccurate source wavelets in FWI, this invention proposes a waveform-protected, local-scale travel-time inversion method independent of the source wavelet. This method uses convolution and deconvolution to ensure that the observed and simulated data share the same source wavelet, eliminating the influence of the source wavelet on local-scale travel-time calculations. Then, using the local-scale travel-time information from the observed and simulated data, the low-wavenumber components of the velocity model are obtained, mitigating the period jump problem in the FWI process. The specific process is as follows: Figure 1 As shown.

[0013] To test the effectiveness of the method, this invention uses model data for testing. A real Marmousi velocity model is used, such as... Figure 2 As shown in Figure a, 40 seismic sources are uniformly distributed on the Earth's surface, with each source corresponding to 360 geophones. The seismic data recorded for the test was 4 seconds long, with a time interval of 2 milliseconds. The source wavelet was selected... Figure 3 The correct and incorrect wavelets are shown, and the velocity model is updated and iterated using the L-BFGS optimization algorithm to obtain the final high-precision inversion result.

[0014] According to the procedure described in the book, the seismic data is first preprocessed to establish an initial velocity model. Figure 2 b), set the source wavelet ( Figure 3The invention first defines an observation system. Then, convolution and deconvolution are used to ensure that the observed and simulated data share the same source wavelet, eliminating the influence of the source wavelet on local-scale travel time calculations. Next, within the framework of local-scale decomposition, local-scale observed and simulated data are obtained, a local-scale travel time inversion objective function is established, and the partial derivatives of the waveform-protected, source-wavelet-independent local-scale travel time inversion objective function with respect to velocity parameters are calculated, thereby obtaining the gradient of the velocity parameter model. Finally, the L-BFGS optimization algorithm is used to continuously update the initial velocity parameter model to obtain the final high-precision inversion result. This invention compares with conventional multi-scale FWI methods (…). Figure 4 Traditional local-scale travel-time inversion methods that do not rely on source wavelets ( Figure 5 Waveform protection method based on local scale travel time inversion of wavelet without source dependence ( Figure 6 ), and waveform protection based on source-independent wavelet local-scale travel time inversion + multi-scale FWI method ( Figure 7 ).

[0015] The model parameters are as follows: Table 1: Test parameters of source-independent wavelet local-scale travel-time inversion method for waveform protection Model size Grid spacing lateral distance Vertical depth Speed ​​range 100×360 25m 9km 2.5km 1.5-4.0 km / s from Figure 3 As can be seen, the source wavelets used to generate the observational and simulation data differ significantly, with almost no overlap. This greatly increases the difficulty of FWI waveform matching. Directly using this wavelet for FWI will not yield any information about the subsurface velocity model. Figure 4 The inversion results of the erroneous source wavelet are shown. It can be seen that the erroneous source wavelet did not obtain any subsurface tectonic information, mainly because both FWI and local scale travel time inversion methods produced obvious periodic jump problems. Figure 5 The results are traditional source-independent wavelet local-scale travel time inversions. To test the impact of different Gaussian window lengths on the inversion results, four window function lengths of 0.4s, 0.7s, 1.1s, and 1.5s were selected for testing. It can be seen that macroscopic test information of the Marmousi velocity model can be obtained, but some outliers still exist on the left side of the model. This is mainly because the waveform information of the source-independent wavelet seismic data established using the convolution method is severely altered, leading to inaccurate local-scale travel time calculations, which in turn affects the low-wavenumber component modeling of the Marmousi velocity model. Figure 6 This demonstrates source-independent wavelet local-scale travel-time inversion with waveform protection. To test the impact of different Gaussian window lengths on the inversion results, four window function lengths of 0.4s, 0.7s, 1.1s, and 1.5s were selected for testing. Figure 5 and Figure 6Comparative analysis revealed that the velocity model established by waveform-protected local-scale travel time inversion of the wavelet without source protection is closer to the actual velocity model, and the velocity modeling results in the left region have been significantly improved. Figure 7 This demonstrates source-independent wavelet local-scale travel-time inversion + multi-scale FWI results (4-15Hz) for waveform protection. Figure 6 The FWI results were used as the initial velocity model. Figure 7 b and Figure 7 Comparison with the actual velocity model reveals that the final inversion result is very close to the actual velocity model, proving the effectiveness of the new method and avoiding the period jump problem caused by missing low frequencies and source wavelet errors in FWI. Numerical experimental results show that the waveform-protected, source wavelet-independent local-scale travel-time inversion + multi-scale FWI method has certain advantages in modeling low wavenumber components of velocity parameters and in independence from source wavelets.

[0016] There are many specific ways to implement this invention. The above description is only a preferred embodiment of this invention. It should be noted that for those skilled in the art, several improvements can be made without departing from the principle of this invention, and these improvements should also be considered within the scope of protection of this invention.

Claims

1. A waveform-protected method for local-scale travel time inversion of a wavelet independent of the seismic source, characterized in that, The influence of the source wavelet in both observed and simulated data is eliminated by using convolution and deconvolution methods to preserve the original seismic waveform characteristics. Secondly, local-scale travel time information is extracted from both observed and simulated data to construct a waveform protection objective function that is independent of the source wavelet's local-scale travel time inversion. Then, the partial derivatives of the waveform protection objective function with respect to velocity parameters, as well as the gradient operator for the underground velocity parameters corresponding to the adjoint state method, are derived in detail. Finally, the L-BFGS optimization algorithm is used to achieve stable updates of the velocity model parameters.

2. The waveform protection method for local-scale travel time inversion of wavelet without source dependence as described in claim 1, characterized in that, Specifically, the following steps are included: Step 1: Use Geoeast seismic data processing software to remove noise from seismic data, and use the processed seismic data as the data input for the waveform-protected local scale travel time inversion method without source wavelet dependence; Step 2: Based on the first arrival travel time information of the seismic data, construct an initial velocity model as the model input for the forward modeling simulation of the source-protected wavelet local scale travel time inversion method. Step 3: Set the source wavelet as the wavelet input for the forward modeling simulation of the waveform protection method that does not rely on the local scale travel time inversion of the source wavelet; Step 4: Set up the seismic exploration and observation system, and use the source wavelet and initial velocity model to perform forward modeling and store the forward propagation wave field; Step 5: Seismic data processing independent of source wavelet: Extract the direct wave waveforms of the near-source channel from the observation data and the simulation data respectively as reference channels, and perform convolution and deconvolution processing on the seismic data; Step 6: Local scale decomposition of seismic data: Using local scale transformation, extract the local scale waveform features corresponding to the observation data and simulation data after processing without relying on the source wavelet; Step 7: Calculate the local travel time difference between the observed data and the simulated data: Use the cross-correlation algorithm to calculate the travel time difference between the two waveforms; Step 8: Construct the local-scale travel-time inversion objective function; Step 9: Calculate the partial derivatives of the local-scale travel time difference objective function with respect to the velocity model parameters; Step 10: Define the accompanying source of the waveform protection non-source wavelet local scale travel time inversion method, and backpropagate the accompanying source to the model space to obtain the backpropagated wave field, which is used to update the underground velocity model. Step 11: Calculate the gradient of the source-independent wavelet local scale travel-time inversion objective function for waveform protection using the adjoint state method; Step 12: Use the gradient information calculated in the previous step as the input information for the L-BFGS optimization algorithm to update the velocity model. Step 13: Determine if the termination condition is met. If it is, output the result of the source-independent wavelet local scale travel time inversion based on waveform protection. If the termination condition is not met, continue to use the current inversion result as the initial velocity parameter model for the next cycle until the termination condition is met.

3. The waveform protection method for local-scale travel time inversion of the wavelet independent of the seismic source, as described in claim 2, is characterized in that... The constant density isotropic medium wave equation in the forward modeling of step 4 can be expressed as: ， Where v is the velocity of sound; u is the sound wave field; x and z represent the horizontal and vertical directions, respectively; and t represents time. The finite difference algorithm is used to solve the above partial differential equations to achieve forward modeling of seismic wave propagation underground.

4. The waveform protection method for local-scale travel time inversion of wavelet without source dependence as described in claim 2, characterized in that, The direct wave waveform of the near-source path of the observed data and simulated data extracted in step 5 can be specifically represented as follows: 、 ， in and These represent the simulated data and the observed data closest to the epicenter, respectively. For window functions, specifically, it can be expressed as: ， in The cutoff time is determined by the arrival time of the direct seismic wave and can be read from the seismic record. 'n' represents the order of the window function; we choose n=10. A smaller n value results in a smoother window function, while a larger n value leads to more pronounced changes. Then, the reference trace is convolved and deconvolved with the observed and simulated data, respectively, to obtain observational and simulated data independent of the source wavelet. 、 ， Where * denotes convolution; Indicates deconvolution; and Let represent the observed data and simulated data after convolution and deconvolution, respectively. Since seismic data can be viewed as the convolution of the Green's function and the source wavelet, the above equation can be expressed as: 、 ， in Represent the Green's function; This represents the source wavelet. The source wavelets of both observed and simulated data after convolution and deconvolution are... This effectively eliminates the influence of the difference between the source wavelet in the observed data and the simulated data on the inversion results.

5. The waveform protection method for local-scale travel time inversion of wavelet without source dependence as described in claim 2, characterized in that, The local scale transformations of the observed and simulated data in step 6 are as follows: 、 、 、 ， in This represents local-scale simulation data; This represents observational data at a local scale; Represents the Gaussian window function; The shift interval is the Gaussian window function.

6. The waveform protection method for local-scale travel time inversion of wavelet without source dependence as described in claim 2, characterized in that, In step 7, the cross-correlation algorithm is used to calculate the travel time difference between the two waveforms: ， The travel time difference between observed data and simulated data ( It can be obtained using a cross-correlation algorithm: ， in Indicates the maximum time difference. This represents the travel time difference between local-scale observed data and simulated data after processing independent of wavelet convolution and deconvolution. This indicates that there is no time difference, meaning the waveform matching is successful. If This indicates that there is a travel time difference between the observed data and the simulated data, at which point the underground velocity field differs from the actual velocity field.

7. The waveform protection method for local-scale travel time inversion of wavelet without source dependence as described in claim 2, characterized in that, In step 8, the velocity inversion objective function constructed by calculating local-scale travel time information using the seismic data processed by convolution and deconvolution can be expressed as: ， in Represents the velocity parameters of seismic waves; This represents a weighting factor, which improves the stability of the algorithm; Indicates the number of earthquake focal points. Indicates the number of detectors.

8. The waveform protection method for local-scale travel time inversion of wavelet without source dependence as described in claim 2, characterized in that, In step 9, the partial derivatives of the local-scale travel time difference objective function with respect to the velocity model parameters are calculated: ， in A direct mapping between the local-scale travel time difference and the seismic wavefield cannot be established. However, by using the implicit function differentiation rule and the cross-correlation function as the correlation equation, the mapping between the local-scale travel time difference and the seismic wavefield can be obtained. When the first derivative of the cross-correlation function with respect to time is 0, the following relationship holds: ， Using the above equation as the correlation equation, the mapping relationship between local scale travel time difference and velocity parameters can be established: ， in , , According to the chain rule, the partial derivatives of the local-scale seismic data after convolution and deconvolution processing with respect to the velocity parameter can be expressed as follows: ， Input the results of convolution and deconvolution on the simulated data: ， Further differentiation using the chain rule yields: ， in This represents the cross-correlation algorithm.

9. The waveform protection method for local-scale travel time inversion of the wavelet independent of the seismic source, as described in claim 2, is characterized in that... The accompanying earthquake source in step 10 can be represented as:

10. A waveform protection method for local-scale travel time inversion of a wavelet independent of the seismic source, as described in claim 2, is characterized in that... Step 11: Calculate the gradient of the source-independent wavelet local-scale travel-time inversion objective function for waveform protection using the adjoint state method: ， in This represents the propagating wave field generated by the earthquake source. This represents the backpropagating wavefield obtained through back propagation from the accompanying earthquake source. Finally, the gradient of the velocity parameters is obtained through zero-delay cross-correlation, and the velocity modulus is updated using an optimization algorithm.

11. The waveform protection method for local-scale travel time inversion of wavelet without source dependence as described in claim 2, characterized in that, In step 12, the gradient information calculated in the previous step is used as input information for the L-BFGS optimization algorithm to update the velocity model. The iterative formula is as follows: ， in Let L-BFGS be the velocity parameter at the k-th step. This is the step size for velocity updates. In the L-BFGS optimization algorithm, It is the inverse of an approximate Hessian matrix: 、 、 、 、 ， in To update the gradient, Through vector pairs and The inverse matrix of the approximate Hessian matrix is ​​calculated. It can be represented as: 。

Images