VSP pre-drilling formation velocity prediction and driving velocity modeling method

Abstract

The invention discloses a VSP pre-drilling formation velocity prediction and driving velocity modeling method. In order to solve the problems that in the prior art, an underground structure of a pre-drilling area is fuzzy, and the precision of a speed model is insufficient, the method comprises the steps that firstly, the acoustic logging speed is corrected through the high-precision longitudinal wave speed obtained through a zero well spacing VSP; secondly, according to the rock physical relationship between the speed and density of a target layer before drilling of an adjacent region and the first arrival travel time prediction before drilling, time-depth constraint is constructed in combination with VSP multi-wave information, and a pre-drilling region initial speed model is established; based on a VSP corridor and the corrected acoustic curve, obtaining an optimal seismic wavelet; and establishing a target function suitable for VSP pre-drilling speed inversion, and obtaining a pre-drilling speed inversion result by using VSP corridor data. And finally, under the constraint of a stratigraphic structure obtained by ground seismic data interpretation, finishing updating of a ground seismic velocity model by using the velocity obtained by VSP zero offset and pre-drilling inversion, so that the velocity model is more matched with underground real velocity structure characteristics.

Description

Technical Field

[0001] This invention belongs to the field of oil and gas seismic exploration, and specifically relates to a method for predicting formation velocity and driving velocity modeling before VSP drilling. Background Technology

[0002] In oil and gas exploration and development, formation velocity models are a crucial foundation for depth migration imaging, structural mapping, reservoir prediction, and drilling engineering deployment. Especially in the pre-drilling stage, their accuracy directly impacts the reliability of key engineering decisions such as wellbore trajectory optimization, casing program design, formation pressure prediction, and drilling risk control. As exploration targets continue to expand into deeper, ultra-deep, and fractured complex reservoirs, higher demands are placed on pre-drilling velocity models in terms of timeliness, accuracy, and consistency with geological structures. However, due to a lack of sufficient prior subsurface information in the pre-drilling stage, it is difficult to accurately grasp formation interface morphology, fault distribution, and the location of velocity abrupt changes. This leads to initial velocity models constructed based on empirical layered assumptions or simple trend extrapolation often deviating significantly from the actual subsurface medium, failing to meet the requirements for high-precision background models in inversion. Zero-source-spacing (VSP) data, due to its high signal-to-noise ratio, strong vertical resolution, and direct measurement of formation velocities near the well, plays an irreplaceable role in velocity modeling and calibration. Although zero-source-spacing VSP (velocity-split) effectively provides high-precision vertical velocity information, its utilization in existing studies remains limited, focusing primarily on velocity calibration or quality control, particularly failing to fully realize its potential in pre-drilling velocity prediction. Pre-drilling velocity prediction based on VSP corridors typically employs seismic inversion, but this method suffers from strong ill-conditioning due to the lack of prior pre-drilling velocity information. To mitigate the ill-conditioning of VSP pre-drilling velocity inversion, existing inversion techniques generally employ regularization to improve stability. While regularization constraints can suppress noise to some extent, they are not selective for structural boundaries and velocity abrupt changes, often weakening velocity gradients, obscuring fault locations, and smoothing out key stratigraphic interfaces, thus causing the final inversion results to deviate from the true geological structure. The combination of these problems makes traditional methods unsuitable for VSP pre-drilling velocity prediction, thereby affecting the accuracy of velocity models and reducing migration imaging quality and pre-drilling trajectory planning. To meet the needs of pre-drilling prediction under complex geological conditions, there is an urgent need to develop a new pre-drilling velocity inversion method that can integrate multi-source information such as VSP, surface seismic data, and well logging, reduce dependence on unreliable prior models, and preserve the structural boundary and velocity abrupt change characteristics.

[0003] To address these challenges, scholars both domestically and internationally have proposed various improvement strategies, such as using multivariate regression models to model formation velocity prediction, constructing alternative feature representations using deep convolutional neural networks, and automatically completing velocity estimation through recurrent neural networks. These strategies have significantly improved the accuracy of velocity models and greatly enhanced the intelligence and adaptability of velocity modeling.

[0004] Existing research results indicate that the pre-drilling velocity inversion method still faces the following problems in practical applications: 1. During the VSP pre-drilling velocity inversion stage, the seismic wavelet morphology of the target formation is difficult to predict due to the lack of actual well calibration. Traditional methods often rely on statistical estimation or reference from adjacent wells, resulting in a low degree of matching between the wavelet phase and frequency components and the actual formation, which directly weakens the reliability of the VSP pre-drilling velocity inversion results. 2. The corridor overlay data is only a single seismic data point, which can be considered as seismic data with zero incident angle. It can only be used for acoustic impedance inversion, while the core requirement of engineering practice is formation velocity. The existing "impedance-velocity" secondary conversion process not only relies on uncertain empirical formulas for density, but also leads to the accumulation of prediction errors in the calculation chain. 3. In the pre-drilling environment lacking well-side constraints, the establishment of the initial velocity model often faces severe spatial uncertainties. Traditional pre-drilling inversion relies heavily on the accuracy of the initial model. If the initial model deviates from the actual formation trend, the inversion process is prone to getting trapped in local extrema, seriously affecting the accuracy of pre-drilling velocity prediction. Summary of the Invention

[0005] To address the challenges of accurate pre-drilling velocity inversion and driving velocity modeling using traditional methods, this invention proposes a pre-drilling formation velocity prediction method that integrates zero-source-distance VSP data (hereinafter referred to as zero-biased VSP) with surface seismic velocity information and well logging data, along with a linearized inversion-driven apparent velocity model update method. This invention fully leverages the advantages of zero-biased VSP in signal-to-noise ratio, resolution, and well-side realism, performing layer-by-layer fine correction on the initial velocity model established based on surface seismic data. This fundamentally overcomes the problems of insufficient VSP information utilization and low vertical resolution of the velocity model in traditional methods, significantly improving the geological rationality of the velocity model. Simultaneously, by introducing constraints such as the intersection points of first arrival waves and multi-wave reflections, this invention achieves reliable pre-drilling formation velocity prediction, resulting in higher consistency and credibility of the initial model in the off-well region. In the velocity inversion stage, addressing the problems of rapid error accumulation and reliance on unreliable priors leading to convergence difficulties or even local optima in traditional nonlinear inversion, this invention performs equivalent linearization of the nonlinear mapping between seismic records and the velocity model, transforming the nonlinear inversion into a controllable and stable linear inversion problem. This strategy effectively suppresses the amplification effect of errors during the iteration process and reduces the sensitivity to prior models, thereby significantly improving the stability, accuracy, and convergence efficiency of velocity updates. Finally, by correcting the ground seismic velocity model along the stratigraphic structure, the accuracy of the constructed velocity model is improved.

[0006] The technical solution of the present invention includes the following steps: S1. Using the VSP multi-wave intersection point and adjacent channel information, constrain the pre-drilling velocity, predict the pre-drilling velocity by fitting the first arrival curve, and construct a reasonable pre-drilling initial velocity model. S2. Combining known velocity density data from neighboring areas, the relationship between the two is fitted based on a rock physics model. Subsequently, seismic wavelets for inversion of the target segment are extracted using VSP corridor and well logging information, laying the foundation for pre-drilling velocity inversion. S3. Using the corridor waveform as the target, perform the inversion of the drilling speed, and finally update the offset speed.

[0007] Step S1 includes the following sub-steps: S11. Extract the first arrival travel time of the zero-biased VSP seismic profile and use curve fitting to fit the first arrival time before drilling. S12, extract the multi-wave intersection point from the zero-biased VSP seismic profile, combine the velocity information of adjacent seismic traces, constrain the fitted first arrival curve, calculate the pre-drilling velocity based on the pre-drilling VSP first arrival, and obtain a reliable pre-drilling initial velocity model.

[0008] Step S2 includes the following steps: S21. Based on the known velocity and density data of the well section, establish a velocity-density empirical formula to realize the relationship between velocity and density. This invention selects the classic Gardner relation to establish a functional mapping relationship between velocity and density, and transforms the relationship between reflection coefficient and wave impedance into a direct relationship between reflection coefficient and velocity. S22. Since VSP processing employs deterministic wavelet deconvolution, it can be assumed that the wavelets contained in the VSP corridor seismic data are in the form of Ricker wavelets. By appropriately setting the Ricker wavelet frequency and phase range, different frequency and phase combinations of Ricker wavelets within the traversal range are convolved with the reflection coefficients of known well sections to generate seismic records. Subsequently, a similarity comparison analysis is performed between each synthetic record and the VSP corridor waveform. Based on the principle of maximizing the correlation coefficient, the wavelet with the highest similarity is selected as the wavelet for subsequent velocity inversion. Finally, the wavelet energy coefficient is determined according to the energy ratio between the synthetic record and the corridor waveform to ensure energy consistency between the synthetic record and the corridor waveform.

[0009] Step S3 includes the following steps: The difference between S31, VSP pre-drilling velocity inversion and surface seismic velocity inversion lies in the fact that the accurate formation velocity in the section near the bottom of the well can serve as a hard constraint for pre-drilling velocity inversion. Therefore, this invention constructs an initial velocity model for participating in pre-drilling velocity inversion, with the upper part representing the actual formation velocity above the bottom of the well and the lower part representing the initial velocity constructed using step S12. In subsequent velocity inversion, only the pre-drilling portion is iterated. Then, based on the velocity-density relationship extracted in S21, the reflection coefficient is directly expressed as a function of velocity. S32. The inversion loss function is minimized using VSP corridor records and synthetic seismic records. The initial velocity model is used as the starting point for inversion, and a loss function with TV regularization constraints is constructed. The sliding window technique is used to perform piecewise inversion iterations of formation velocity, and the inversion is iteratively performed by minimizing the loss function, thereby achieving high-precision prediction of pre-drilling velocity. S33. Based on the known stratigraphic structure obtained by interpreting ground seismic data, and combined with the zero-biased VSP velocity and the inverted pre-drilling velocity, the velocity model is updated along the stratigraphic structure to obtain an accurate velocity model that includes the pre-drilling area.

[0010] This method achieves precise matching between synthetic and measured seismic records by searching for combinations of wavelet frequency and phase parameters and combining them with energy matching. It effectively solves the prediction error problem caused by inaccurate wavelet parameters and provides effective support for improving inversion resolution.

[0011] This method establishes the relationship between velocity and density by using velocity density data and fitting Gardner formula coefficients. It also predicts pre-drilling velocity by combining the zero-biased VSP initial arrival fitting results. This method constrains and extrapolates the velocity in the unknown pre-drilling area, significantly improving the rationality and continuity of the pre-drilling initial velocity model and reducing the accumulation of deviations caused by improper velocity extrapolation.

[0012] This method constructs an initial velocity model before drilling by integrating the fitting first arrival time, multi-wave convergence information, and spatial consistency of adjacent seismic traces as constraints. This achieves the effect of establishing a high-precision geological background framework in deep strata where measured data is lacking, and significantly enhances the initial model's ability to characterize complex geological structures and the accuracy of predicting velocities in deep strata.

[0013] Compared with traditional methods, this invention significantly improves the accuracy, stability and resolution of pre-drilling velocity prediction, providing accurate and reliable technical support for velocity modeling under complex geological conditions. Attached Figure Description

[0014] Figure 1 This is a flowchart of the present invention; Figure 2 (1) Comparison of first arrival predictions with and without multi-wave convergence points; (2) Schematic diagram of time-depth relationship constraints of multi-wave convergence in Marmousi forward modeling seismic records; Figure 3 It is a Gardener formula fit of velocity and density, and the fitting result is: Figure 4This is a schematic diagram of the optimal wavelet in the simulation data. The upper left figure is a comparison of the reference corridor waveform and the synthesized waveform. The dashed line is the synthesized waveform, and the solid line is the reference corridor waveform. The upper right figure shows the similarity curves between the waveform synthesized by wavelets of different frequencies and the reference corridor waveform. The lower left figure shows the similarity curves between the waveform synthesized by wavelets of different phases and the reference corridor waveform. The lower right figure shows the optimal wavelet obtained, the -0.7° wavelet at 30Hz, which is basically consistent with the wavelet used in the forward modeling reference corridor waveform. Figure 5 The left image shows a comparison of the inverted pre-drilling velocity, and the right image shows a comparison of the inverted waveforms. Figure 6 The updated speed model is shown in Figure 1. (1) is the real model, (2) is the initial model, and (3) is the model updated based on the inverted speed. Figure 7 This is the fitted first arrival curve. The area in the middle of the dashed line represents the data used in the fitting, and the area to the right of the right dashed line represents the fitted pre-drilling first arrival curve. Figure 8 This is a schematic diagram of the Gardener formula, which is based on the fitting of known segment velocity and density. Figure 9 This is a schematic diagram of the optimal wavelet in the actual VSP data. The upper left image is a comparison of the corridor waveform and the synthesized waveform. The dashed line is the synthesized waveform, and the solid line is the reference corridor waveform. The upper right image is the similarity curve between the waveform synthesized by wavelets of different frequencies and the reference corridor waveform. The lower left image is the similarity curve between the waveform synthesized by wavelets of different phases and the reference corridor waveform. The lower right image is the obtained optimal wavelet, the -15° wavelet at 25Hz. Figure 10 The left image shows a comparison between the inversion velocity and the VSP velocity, while the right image shows a comparison between the inversion waveform and the corridor waveform. Figure 11 The offset-driven velocity update is shown in Figure (1), which is the initial model; Figure (2) is the reference model structure; and Figure (3) is the velocity model after the velocity update is based on the inverted velocity. Detailed Implementation Example

[0016] To facilitate understanding of the technical content of this invention by those skilled in the art, the following description, in conjunction with the accompanying drawings, further illustrates the invention.

[0017] This embodiment uses the marmousi2 model for experiments, such as Figure 1 As shown, the method of the present invention includes the following steps: Since this embodiment uses the Marmousi synthetic velocity model, there is no need to calculate the velocity by picking the first arrival time; the true velocity information can be directly extracted from the model. This experiment selects velocities from traces 3000-5000 (4500m-26250m, 300-1750 lines, sampling interval 15m) in the Marmousi2 model and smooths them to create a ground seismic model. The velocity from trace 4000 is then used as the VSP and the actual pre-drilling velocity. For waveform synthesis, the complete wave impedance curve is convolved with a 30Hz zero-phase Ricker wavelet to generate the corresponding synthetic seismic record (i.e., corridor waveform), which serves as the "reference corridor waveform" data in this embodiment for verification in subsequent pre-drilling prediction and inversion processes.

[0018] The first 750 points of velocity at track 4000 are selected as VSP velocities. First arrival predictions are performed on these velocities, and constraints are applied using multi-wave intersection points. Figure 2 As shown, the predicted drilling speed is obtained; Based on the known density rho and velocity V of the well section, an empirical velocity-density formula is fitted, and the Gardner formula is selected: (1) Establish the relationship between velocity and density to obtain the coefficient. ,like Figure 3 As shown; Pick up the VSP data first arrival time. Based on the actual first arrival time. By fitting the initial arrival curve before drilling, the fitted initial arrival time can be obtained. ; according to Calculate the drilling speed ; splicing by depth and As the initial model ; The nonlinear relationship between seismic records and velocities is transformed into a linear relationship. Seismic records are generally composed of wavelets. Convolutional reflection coefficient Therefore, for convolution operations, a Toeplitz matrix can be constructed, assuming the wavelet... Then the corresponding Toeplitz matrix can be expressed as: ; This transforms the convolution of the reflection coefficient with the wavelet into a matrix multiplication of the reflection coefficient with the W matrix; the reflection coefficient can be expressed as: ; in Z is the reflection coefficient, obtained by multiplying the formation density and velocity, where It can be represented as Then there are: ; Using Taylor expansion: ,make Retaining the first-order terms, we have: To establish a linear relationship between seismic records and velocities, we need to use the Gardener formula fitted in the previous steps. Therefore, the wave impedance can be expressed as: Therefore, the reflection coefficient can be expressed as: ; For exponential functions, there is a Taylor expansion: ; make The reflection coefficient, retaining the first-order term of the Taylor expansion, can be expressed as: ; Combining this with the Toeplitz matrix mentioned earlier, the seismic record can be represented as: ; This allows us to establish a linear relationship between seismic records and the logarithm of velocity.

[0019] A cyclically generated Ricker wavelet w(t) with a phase of 15° from 5Hz to 50Hz is used to synthesize seismic records according to formula (8). Convolving the Ricker wavelets with different dominant frequencies and phase combinations with the reflection coefficient sequence generates corresponding synthetic seismic records. To ensure the amplitude level of the synthetic record matches the corridor waveform, energy matching is performed: first, the ratio of the average amplitude of the synthetic record to that of the corridor waveform is calculated; then, this ratio is used as a scaling factor on the wavelet, ensuring that the overall amplitude of the synthetic record falls within a range similar to that of the corridor waveform. For example... Figure 4 As shown, similarity analysis of all synthetic records and corridor waveforms revealed that the highest correlation coefficient was obtained when the dominant wavelet frequency was 30 Hz and the phase was –0.7°. The optimal wavelet shape obtained under this condition was almost identical to the 30 Hz zero-phase Ricker wavelet used to generate the "real corridor data," indicating that this frequency-phase combination effectively reflects the dominant characteristics of the corridor waveform. This method can effectively reconstruct the wavelet. Constructing the W matrix from the wavelet involves shifting the wavelet sample by sample downward along the time axis to form each row, making it a diagonal strip matrix, thus achieving linear convolution operation. Constructing the forward matrix Where W is based on the wavelet The D matrix is ​​constructed by using a linear operator to perform first-order differences on the model , i.e., by placing -1 on the main diagonal and +1 on the upper (or lower) diagonal. Difference operations.

[0020] The process begins by constructing time-depth constraints for multiple waves. In the Walkaway VSP observation system, although P-wave reflections and PS-converted reflections have different travel time characteristics in the time domain, under the condition of satisfying Snell's law, the two wave types can correspond to the same subsurface reflection interface and share similar reflection point locations. Based on this spatial geometric consistency, P–P and P–S reflection events can be used as joint constraints on the location of subsurface structures. This further limits the extrapolation uncertainty of the pre-drilling regional velocity model and improves the geological rationality of the predicted velocity model, based on the large-scale velocity trend dominated by first arrival information. First, the travel time lines of the P-wave and converted wave reflections are extracted and extended (i.e., fitted curves). The time of the intersection of the two extended lines is then extracted. With depth We obtain the temporal depth relationship at that point.

[0021] Construct the inversion objective function: ; The first term is the square of the L2 norm of the difference between the corridor waveform and the inverted waveform; the second term is TV regularization, which preserves the characteristics of abrupt velocity changes without forcing smooth velocity variations. The first term is the regularization parameter, which controls the strength of TV regularization; the second term is the multi-wavelength time-depth relationship constraint. It is the travel time of the inversion curve at depth, derived from the time-depth conversion formula: The time is then calculated and compared with the time for the corresponding depth obtained in the previous step. Find the L2 norm by taking the difference. It is a regularization parameter that controls the strength of multi-wave depth constraints.

[0022] After setting reasonable initial parameters, the BFGS optimization method was used to iteratively invert the velocity model while keeping the known velocities in the well constant. To obtain higher resolution formation structure, a localized, piecewise inversion was performed in the pre-drilling region using a sliding window of length 200 with a step size of 100, reducing the cumulative error caused by simultaneous inversions over a large area. A Hanning window was introduced between adjacent windows for smooth weighting, ensuring continuous and stable window transitions. By progressively updating the pre-drilling velocity within each window, a refined pre-drilling velocity model was finally obtained.

[0023] The inverted waveform results are as follows Figure 5 As shown on the right, the waveform similarity is very high, reaching 0.99, and the waveform residual reaches... This indicates that the synthetic seismic records generated by the inverted velocity model can reproduce the observation data with high accuracy, verifying the reliability of the inversion results and the stability of the algorithm in different reflection segments; the inverted velocity results are as follows: Figure 5 As shown on the left, overall, the inverted velocity profile is consistent with the real velocity model in terms of structural morphology, inter-layer velocity trends, and large-scale velocity background. It can accurately depict the velocity increase characteristics with depth and the locations of velocity abrupt changes at major stratigraphic interfaces. In most areas, the error between the inverted velocity and the real velocity is small, indicating that the algorithm has achieved stable convergence during multiple iterations and can reflect the stratigraphic velocity structure well. However, near strong reflection interfaces (i.e., areas where the real velocity changes drastically), although the inverted velocity can correctly identify the location of velocity abrupt changes, reflecting the trend of "large velocity changes," the velocity amplitude is deviated, and the real velocity value is not fully recovered. This phenomenon indicates that in areas with large velocity gradients or drastic changes in reflection energy, the inversion algorithm still has certain limitations in its ability to recover high wavenumber components.

[0024] Finally, the inverted single-track velocity is interpolated along the layers of the initial model to update the offset velocity, and the result is as follows. Figure 6 As shown, it can effectively construct an accurate initial model.

[0025] Example 2 Example 2 uses measured data from a well in the Kuqa Depression for verification. The well depth is 6260 m, and the VSP data sampling interval is 20 m; the density data of the known well section is located in the depth range of 6768 m to 7027 m, with a sampling interval of 0.125 m. The overall technical concept of this example is the same as that of Example 1.

[0026] The travel time curves of P-wave reflected waves (PP waves) and P-S converted reflected waves (PS waves) are picked from the seismic profile, and extrapolation fitting is performed on the travel time curves. The time of the intersection point of the extrapolated extension curves of PP waves and PS waves is obtained by calculating the intersection point of the two types of reflection wave extension curves. With depth This establishes the time-depth relationship of the reflective interface.

[0027] The first arrival travel times of the P-wave direct arrivals are collected, and a curve fitting method is used to model the first arrival travel times. Considering the well depth of 6260 m and the strong correlation between the formation structure near the well bottom and the pre-drilling area, first arrival travel time data within the depth range of 4000 m to 6260 m are selected to extrapolate and predict the first arrival travel times in the pre-drilling interval of 6260 m to 8000 m. Simultaneously, the time-depth relationship corresponding to the multi-wave reflection intersection points obtained in the above steps is introduced as a hard constraint into the first arrival travel time fitting process to constrain the rationality of the pre-drilling first arrival prediction results. Figure 7 As shown. Finally, the predicted first arrival travel times are converted into velocity parameters to construct an initial velocity model for the pre-drilling area.

[0028] By splicing the in-well initial velocity model with the pre-drilling initial velocity model, a complete inverted initial velocity model is obtained.

[0029] Using known velocity and density data within the well section, the Gardner empirical formula was parametrically fitted to establish the correlation between velocity and density. The fitting results are as follows: Figure 8 As shown, .

[0030] Since this embodiment uses measured seismic data, the corridor waveform contains significant noise. Therefore, bandpass filtering is first applied to the corridor waveform in the frequency domain. Considering that the seismic signal is mainly concentrated in the low-frequency band, a bandpass filter of 10–50 Hz is selected to filter out extremely low-frequency drift and high-frequency random noise, thereby improving signal quality.

[0031] A 5-50 Hz Lake wavelet with a phase of ±60° was cyclically generated and convolved and energy-matched with VSP velocity models within a depth range of 4000 m to 6260 m to synthesize corresponding seismic records. Among all synthesized seismic records, the correlation coefficient between the wavelet with a dominant frequency of 25 Hz and a phase of −27° and the measured corridor waveform reached its maximum. Therefore, this wavelet was selected as the optimal wavelet for subsequent inversion. Figure 9 As shown.

[0032] An inversion objective function was constructed, and the time-depth relationship corresponding to the aforementioned multi-wave reflection intersection points was introduced into the objective function as a constraint term. A sliding window with a window length of 200 and a step size of 100 was used for the inversion, and a Hann window was used for smooth transition between adjacent windows. The inversion results are as follows: Figure 10 As shown, the waveform above the dashed line on the left represents the wellbore velocity. Given the high accuracy of the wellbore velocity obtained through VSP inversion, a strategy is adopted where wellbore velocity is included in the calculation but not updated during the inversion process. Simultaneously, a sliding window inversion method is used instead of a global inversion, progressively advancing from the bottom of the well towards the pre-drilling area, ensuring that most of the VSP data participates in the inversion calculation, thereby guaranteeing the stability and reliability of the inversion results within the window. The corresponding inversion waveform is shown below. Figure 11 As shown in the right figure, the inversion results show that the waveform fitting effect in the pre-drilling area is good, and the inverted velocity exhibits fluctuations that are consistent with geological understanding within a reasonable range, indicating that the method can converge stably and effectively approximate the real underground strata structure.

[0033] Finally, the inverted single-track velocity is interpolated along the layers of the initial model to update the offset velocity, and the result is as follows. Figure 11 As shown; Overall, existing inversion methods cannot accurately invert the correct velocities, and the inversion algorithm needs to be optimized to improve the accuracy of the inversion. In addition, given that the existing initial model only has low-frequency information, the method of building the initial model should be updated to include high-frequency information, so as to accurately invert the formation velocity of the strong reflective interface.

[0034] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Various modifications and variations can be made to the invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the invention should be included within the scope of the claims of the invention.

Claims

1. A method for predicting formation velocity and modeling driving velocity before VSP drilling, characterized in that, Includes the following steps: S1. Using the VSP multi-wave intersection point and adjacent channel information, constrain the pre-drilling velocity, predict the pre-drilling velocity by fitting the first arrival curve, and construct a reasonable pre-drilling initial velocity model. S2. Combining known velocity density data from neighboring areas, the relationship between the two is fitted based on a rock physics model. Subsequently, seismic wavelets for inversion of the target segment are extracted using VSP corridor and well logging information, laying the foundation for pre-drilling velocity inversion. S3. Using the corridor waveform as the target, perform the inversion of the drilling speed, and finally update the offset speed.

2. The VSP pre-drilling formation velocity prediction and driving velocity modeling method according to claim 1, characterized in that, Step S1 includes: S11. Extract the first arrival travel time of the zero-biased VSP seismic profile and use curve fitting to fit the first arrival time before drilling. S12, extract the multi-wave intersection point from the zero-biased VSP seismic profile, combine the velocity information of adjacent seismic traces, constrain the fitted first arrival curve, calculate the pre-drilling velocity based on the pre-drilling VSP first arrival, and obtain the pre-drilling initial velocity model.

3. The VSP pre-drilling formation velocity prediction and driving velocity modeling method according to claim 1, characterized in that, Step S2 includes: S21. Based on the known velocity and density data of the well section, establish an empirical formula for velocity-density to establish the relationship between velocity and density. Select the Gardner relation to establish a functional mapping relationship between velocity and density, and transform the relationship between reflection coefficient and wave impedance into a direct relationship between reflection coefficient and velocity. S22. Assuming the wavelets in the VSP corridor seismic data are in the form of Ricker wavelets, reasonably set the Ricker wavelet frequency and phase range, and convolve the Ricker wavelets with different frequency and phase combinations within the traversal range with the reflection coefficients of known well sections to generate seismic records. Subsequently, perform similarity comparison analysis between each synthetic record and the VSP corridor waveform, and select the wavelet with the highest similarity as the wavelet for subsequent velocity inversion based on the principle of the highest correlation coefficient. Finally, determine the wavelet energy coefficient according to the energy ratio of the synthetic record and the corridor waveform to ensure the energy consistency between the synthetic record and the corridor waveform.

4. The VSP pre-drilling formation velocity prediction and driving velocity modeling method according to claim 2, characterized in that, Step S3 includes: The difference between S31 and VSP pre-drilling velocity inversion and surface seismic velocity inversion lies in the fact that the accurate formation velocity in the section near the bottom of the well serves as a hard constraint for pre-drilling velocity inversion. An initial velocity model is constructed to participate in the pre-drilling velocity inversion. The upper part is the actual formation velocity above the bottom of the well, and the lower part is the initial velocity constructed using step S12. In subsequent velocity inversion, only the pre-drilling part is iterated. Then, based on the velocity-density relationship extracted in S21, the reflection coefficient is directly expressed as a function of velocity. S32. The inversion loss function is minimized by combining VSP corridor records and synthetic seismic records. The initial velocity model is used as the starting point for inversion, and a loss function with TV regularization constraints is constructed. The sliding window technique is used to perform piecewise inversion iteration of formation velocity. The inversion is performed by minimizing the loss function, thereby achieving high-precision prediction of pre-drilling velocity. S33. Based on the known stratigraphic structure obtained by interpreting ground seismic data, and combined with the zero-biased VSP velocity and the inverted pre-drilling velocity, the velocity model is updated along the stratigraphic structure to obtain an accurate velocity model that includes the pre-drilling area.

Images