Rayleigh wave double-component fusion method and system based on model prediction
By establishing a quantitative relationship model between the fusion gain and signal-to-noise ratio of Rayleigh wave dual components, the problem of unpredictable fusion effect in existing technologies is solved, enabling accurate prediction and automated control of fusion effect, and improving the signal-to-noise ratio of seismic data.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHANGCHUN UNIV OF TECH
- Filing Date
- 2026-02-10
- Publication Date
- 2026-05-12
AI Technical Summary
Existing technologies lack quantitative research in Rayleigh wave two-component fusion, resulting in unreliable prediction of fusion effects, inability to adaptively optimize, difficulty in embedding into automated processing flows, and the potential for low-quality signals to contaminate high-quality signals.
A quantitative relationship model (Y = -kX + C) between Rayleigh wave two-component fusion gain and signal-to-noise ratio similarity is established. The fusion gain is predicted by calculating the similarity index, and a quality safety threshold is set to select the final effective signal.
It achieves accurate forward prediction of fusion effects, avoids blind operation, improves the reliability and quality of seismic data imaging, and steadily improves the overall signal-to-noise ratio.
Smart Images
Figure CN122017989A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of exploration geophysics and seismic signal processing technology, specifically relating to a model-predicted Rayleigh wave dual-component fusion method and system. Background Technology
[0002] When using seismic background noise for surface wave imaging, Rayleigh wave signals mainly exist in two components: vertical (Z) and radial (R). Current mainstream methods typically process only the Z component data, which has a higher signal-to-noise ratio (SNR), while discarding the R component data, which has a lower SNR. This results in low data utilization and fails to fully exploit the potential of the dual-component data. Although the R component also carries a valid signal, practice shows that if the Z and R components are directly superimposed without discrimination (i.e., blind fusion), noise from the low-quality signal will contaminate the high-quality signal, leading to a decrease in the overall SNR of the fused signal. To overcome the shortcomings of blind fusion, some fusion decision-making methods based on signal quality assessment have been developed in this field. For example, there is a class of methods that determine whether to fuse by comparing the similarity of the SNR of the two components. These methods can, to some extent, avoid worst-case scenarios, elevating the fusion decision from completely random to a rule-based level. However, the applicant's research found that these empirical criterion methods based on the similarity of SNR still have limitations in principle. First, their core decision-making relies on a static, empirical quality safety threshold δ. The determination of this threshold lacks theoretical guidance and typically relies on repeated trials and manual adjustments on specific datasets. This not only leads to inefficient parameter optimization and poor portability, but more fundamentally, it simplifies the continuous scientific decision of "whether to fuse" into a binary "either / or" judgment (falling into or not falling into a fixed range), failing to provide a refined evaluation of cases at the critical juncture. A deeper limitation lies in the lack of systematic quantitative research and theoretical guidance from existing technologies (including the methods mentioned above) on the fundamental question of "how the fusion effect changes with the quality difference between the two components." While practice in the field has recognized that excessive signal-to-noise ratio differences can lead to fusion failure, key questions such as how specific differences quantitatively affect gain and whether there is a predictable inflection point for gains remain at the level of qualitative understanding or experiential intuition. Because a precise quantitative relationship model (Y=F(X)) between fusion gain (Y) and signal-to-noise ratio difference (X) has not yet been established, the existing technology has the following inherent defects: (1) it cannot reliably predict the fusion effect of unprocessed data; (2) it cannot optimize decision parameters in reverse according to the target effect to achieve adaptive control; and (3) it is difficult to embed into automated processing flows that require precise quality control. Therefore, this application aims to provide a more accurate, predictable, and optimizable fusion decision scheme. Summary of the Invention
[0003] To address the aforementioned technical issues, this application, based on extensive experimental data, discovers that in Rayleigh wave two-component fusion, there is a stable linear negative correlation between the fusion gain Y and the similarity index X of the two component signal-to-noise ratios, i.e., Y = -kX + C. Based on this discovery, this application proposes a model-predictive-based Rayleigh wave two-component fusion method and system.
[0004] According to the Rayleigh wave two-component fusion method based on model prediction provided in this application, the method includes: Acquire continuously collected seismic data from stations within the target area; Calculate the signal-to-noise ratio (SNR) of the ZZ cross-correlation function and the SNR of the RR cross-correlation function for each station pair; Calculate the ratio of the signal-to-noise ratio of the RR cross-correlation function to the signal-to-noise ratio of the ZZ cross-correlation function for each station pair; Calculate the similarity index based on the ratio; The similarity index is used to predict the fusion gain; Compare the predicted fusion gain with the quality and safety threshold; Based on the comparison results, the final valid signal is selected and output.
[0005] Furthermore, the similarity index is calculated based on the ratio as follows: , As a similarity index, It is a ratio.
[0006] Furthermore, the similarity index is used to predict the fusion gain, including: The fusion gain is calculated using a quantitative relationship model: ,in, and Let be the fitting constant. For fusion gain.
[0007] Furthermore, the quantitative relationship model is obtained by using multiple station pairs with different signal-to-noise ratios as sample datasets and performing linear regression using the least squares method, including: Calculate the signal-to-noise ratio (SNR) of the ZZ cross-correlation function and the SNR of the RR cross-correlation function for the sample dataset; Calculate the ratio of the signal-to-noise ratio of the RR cross-correlation function to the signal-to-noise ratio of the ZZ cross-correlation function; Calculate the similarity index based on the ratio; The signal-to-noise ratio is calculated by linearly superimposing the ZZ cross-correlation function and the RR cross-correlation function. Calculate the fused signal-to-noise ratio gain based on the superimposed signal-to-noise ratio; Using the fusion signal-to-noise ratio gain as the dependent variable and the similarity index as the independent variable, linear regression was performed using the least squares method.
[0008] Furthermore, the formula for calculating the fused signal-to-noise ratio gain is as follows: , in, To fuse signal-to-noise ratio gains, The signal-to-noise ratio after superposition. Let SNR(RR) be the signal-to-noise ratio of the ZZ cross-correlation function, and SNR(RR) be the signal-to-noise ratio of the RR cross-correlation function. To obtain the maximum value.
[0009] Furthermore, if the predicted fusion gain is greater than or equal to the quality and safety threshold, the ZZ cross-correlation function and RR cross-correlation function of the station pair are superimposed as the final effective signal; if the predicted fusion gain is less than the quality and safety threshold, the signal-to-noise ratio of the ZZ cross-correlation function and the RR cross-correlation function of the station pair are compared, and the cross-correlation function with the larger signal-to-noise ratio is selected as the final effective signal.
[0010] According to the Rayleigh wave two-component fusion system based on model prediction provided in this application, Includes: a prediction module for acquiring continuously acquired seismic data from stations within the target area; calculating the signal-to-noise ratio (SNR) of the ZZ cross-correlation function and the SNR of the RR cross-correlation function for each station pair; calculating the ratio of the SNR of the RR cross-correlation function to the SNR of the ZZ cross-correlation function for each station pair; calculating a similarity index based on the ratio; and using the similarity index to predict the fusion gain. The decision module compares the predicted fusion gain with the quality and safety threshold; based on the comparison result, it selects and outputs the final valid signal.
[0011] Furthermore, the similarity index is calculated based on the ratio as follows: , As a similarity index, It is a ratio.
[0012] Furthermore, the similarity index is used to predict the fusion gain, including: The fusion gain is calculated using a quantitative relationship model: ,in, and Let be the fitting constant. For fusion gain; The quantitative relationship model is obtained by using multiple station pairs with different signal-to-noise ratios as sample datasets and performing linear regression using the least squares method, including: Calculate the signal-to-noise ratio (SNR) of the ZZ cross-correlation function and the SNR of the RR cross-correlation function for the sample dataset; Calculate the ratio of the signal-to-noise ratio of the RR cross-correlation function to the signal-to-noise ratio of the ZZ cross-correlation function; Calculate the similarity index based on the ratio; The signal-to-noise ratio is calculated by linearly superimposing the ZZ cross-correlation function and the RR cross-correlation function. Calculate the fused signal-to-noise ratio gain based on the superimposed signal-to-noise ratio; Using the fusion signal-to-noise ratio gain as the dependent variable and the similarity index as the independent variable, linear regression was performed using the least squares method.
[0013] Furthermore, by setting the target gain and solving the similarity index in reverse, the obtained similarity index is used as the quality and safety threshold.
[0014] Compared with the prior art, the advantages of this application are as follows: Fundamental Breakthrough and Engineering Application: This application discovers and establishes a quantitative relationship model (Y = -kX + C) between the fusion gain and signal-to-noise ratio similarity of Rayleigh wave two-components, transforming the fusion decision from a qualitative judgment relying on experience into a precise technical process that is calculable, predictable, and optimizable based on clear physical and mathematical relationships. This is not only a cognitive breakthrough but also provides an engineering solution that can be directly embedded into automated processes.
[0015] The prediction and decision-making modules constructed in this application system can be directly embedded into existing data processing workflows to achieve accurate and forward-looking prediction of fusion effects, avoiding blind operations; adaptive and optimized fusion strategy parameters replace inefficient manual trial and error; and automatic intelligent control of the processing workflow systematically solves the problem of blind decision-making.
[0016] Verification through examples (such as comparative experiments) shows that this method can consistently improve the overall signal-to-noise ratio of the fused data (e.g., an average gain of 7.90%) and effectively avoid negative gain fusion, directly improving the reliability and quality of seismic data imaging. This core model can be further used as an optimization module and widely applied in seismic data processing software systems. Attached Figure Description
[0017] Figure 1 A general flowchart of the method provided in the embodiments of this application; Figure 2 A scatter plot and a fitted curve diagram illustrating the quantitative relationship between the fusion gain Y and the similarity index X = |R - 1| provided for embodiments of this application; Figure 3 To provide the residual analysis charts provided by the system in this application. Detailed Implementation
[0018] The following examples demonstrate how the quantitative patterns discovered above can be applied to practical fusion decision-making and process control.
[0019] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. To further clarify the objectives, technical solutions, and advantages of this invention, specific embodiments will be described in detail below with reference to the accompanying drawings. The embodiments described herein are for illustrative purposes only and do not constitute a limitation thereof.
[0020] See Figure 1 As shown, a Rayleigh wave two-component fusion method based on model prediction is proposed, which includes: Acquire continuously collected seismic data from stations within the target area; Calculate the signal-to-noise ratio (SNR) of the ZZ cross-correlation function and the SNR of the RR cross-correlation function for each station pair; Calculate the ratio of the signal-to-noise ratio of the RR cross-correlation function to the signal-to-noise ratio of the ZZ cross-correlation function for each station pair; Calculate the similarity index based on the ratio; The similarity index is used to predict the fusion gain; Compare the predicted fusion gain with the quality and safety threshold; Based on the comparison results, the final valid signal is selected and output.
[0021] In one embodiment, acquiring continuously acquired seismic data from stations within the target area includes: acquiring three-component background noise data, including vertical, east-west, and north-south components, continuously acquired from stations within the target area, and performing standardized preprocessing on the three-component background noise data.
[0022] A seismic station refers to a fixed observation point used for earthquake monitoring, equipped with seismographs, data acquisition devices, communication equipment, etc., to acquire continuous background noise data in three components: vertical (Z), north-south (N), and east-west (E). The background noise data undergoes standardized preprocessing, including resampling, bandpass filtering, time-domain normalization, and spectral whitening.
[0023] ZZ cross-correlation function calculation: Using data collected by the stations, calculate the ZZ cross-correlation function for all station pairs, where Z is the vertical component; A station pair refers to pairing up all stations. For example, if there are four stations, ABCD, the possible station pairs are AB, AC, AD, BC, BD, and CD. Calculating the ZZ cross-correlation function for all station pairs means calculating the cross-correlation function of the vertical components of the data collected by the two stations in each pair, and obtaining the vertical component cross-correlation function, which is represented by the ZZ cross-correlation function.
[0024] RR cross-correlation function calculation: Using the preprocessed east-west and north-south components, the RR cross-correlation function of all station pairs is calculated, where R is the radial component; The RR cross-correlation function cannot be obtained by directly taking the cross-correlation of the collected vertical (Z), north-south (N), and east-west (E) components. Instead, it is obtained by cross-correlation of the north-south (N) and east-west (E) components, applying the coordinate rotation formula to transform to the radial-tangential coordinate system, and then extracting the RR cross-correlation function from the cross-correlation function.
[0025] Calculate the signal-to-noise ratio (SNR) of the ZZ cross-correlation function and the SNR of the RR cross-correlation function for each station pair.
[0026] In one embodiment, the ZZ cross-correlation function of all station pairs is calculated using the preprocessed vertical component, where Z is the vertical component, including: The continuous data of the vertical component (Z) of each station pair is divided into multiple time data segments; for each time data segment, its ZZ cross-correlation function is calculated; the ZZ cross-correlation functions of all time data segments are linearly superimposed to obtain the final ZZ cross-correlation function of the station pair. That is, it is the ZZ cross-correlation function of the vertical component of one station within a station pair with the vertical component of another station within the same time data segment, obtained by superimposing multiple time data segments.
[0027] In one embodiment, the RR cross-correlation function for all station pairs is calculated, including: The data from each station for the continuous east-west component (E) and north-south component (N) are divided into multiple time data segments; For each time data segment, the EE, NN, EN, and NE cross-correlation functions are calculated, where E represents the east-west component and N represents the north-south component. The cross-correlation function results of the same type of component pairs in all time data segments are superimposed to obtain the final EE, NN, EN, and NE cross-correlation functions for that station pair. Based on the azimuth angle between station pairs, the EE, NN, EN, and NE cross-correlation functions are transformed to a radial-tangential coordinate system to obtain the corresponding TT, RR, TR, and RT cross-correlation functions, where R represents radial and T represents tangential. The RR cross-correlation function is extracted from the transformation results.
[0028] The conversion formula is: , in the formula This indicates the azimuth angle between the first and second stations relative to the center of the station. This indicates the azimuth angle from the second station to the first station.
[0029] The calculation process for the signal-to-noise ratio (SNR) of the ZZ cross-correlation function and the SNR of the RR cross-correlation function includes: setting the surface wave signal velocity window. The lower limit of the surface wave signal velocity window The minimum velocity of the cross-correlation function, and the upper limit of the surface wave signal velocity window. The maximum velocity of the cross-correlation function; Determine the signal arrival time window based on the surface wave signal velocity window. ; After setting the noise window to the signal window for a certain period of time seconds, in one example, , , The station spacing is set; the noise window is set to a duration of 150 seconds after the signal window, i.e. .
[0030] The signal-to-noise ratio of the cross-correlation function is obtained by calculating the ratio of the maximum amplitude within the signal window to the root mean square amplitude within the noise window using the cross-correlation function.
[0031] For example, the signal-to-noise ratio (SNR) of the ZZ cross-correlation function is calculated, denoted by SNR(ZZ), and the surface wave signal velocity window is used. Lower limit of the velocity window for mid-surface wave signals for The lower limit of the surface wave signal velocity window The minimum velocity of the ZZ cross-correlation function, and the upper limit of the velocity window for the surface wave signal. The maximum velocity of the ZZ cross-correlation function is the upper limit of the velocity window for the surface wave signal. The maximum velocity of the ZZ cross-correlation function is given. The signal-to-noise ratio of the RR cross-correlation function is calculated using the same method and represented by SNR(RR).
[0032] The signal-to-noise ratio (SNR) of the linear superposition of the ZZ cross-correlation function and the RR cross-correlation function is represented by SNR(ZZ+RR).
[0033] Calculate the ratio of the signal-to-noise ratio (SNR) of the RR cross-correlation function to the SNR of the ZZ cross-correlation function to obtain the SNR ratio: .
[0034] The cross-correlation function in this application is calculated using conventional methods, such as obtaining the ZZ cross-correlation function by cross-correlation of the two Z vertical components of a station pair.
[0035] In one embodiment, if the quality and safety threshold is too small, there will be too few station pairs that meet the fusion conditions, resulting in insufficient data potential mining; if it is too large, some data pairs with large signal-to-noise ratio differences will enter the fusion process, introducing the risk of degradation. This application uses the similarity index obtained by setting a target gain to solve inversely as the quality and safety threshold.
[0036] To accurately quantify the impact of the signal-to-noise ratio (SNR) difference between the two components on the fusion effect, this application first defines the SNR ratio R = SNR(RR) / SNR(ZZ). When R = 1, it indicates that the two components are of completely equal quality; the further the R value deviates from 1, the greater the difference. To characterize this difference, a "similarity index" with clear physical meaning needs to be defined.
[0037] Based on different physical considerations, this application proposes and compares two of the most representative candidate index forms: Candidate metric 1 (linear difference): X1 = |R - 1|. This metric directly measures the absolute difference in signal-to-noise ratio, has an intuitive physical meaning, and aligns with the general understanding of "similarity".
[0038] Candidate index two (squared difference): X2 = (R - 1)². In signal processing, the squared term is related to energy or variance, and can better reflect the impact of difference on system performance from the perspective of "energy loss".
[0039] To scientifically determine which indicator form best reveals the intrinsic relationship between X and fusion gain Y, this application conducted a systematic empirical comparison of the two. Table 1 lists the coefficients of determination (R²) of the linear relationship between fusion gain Y and the two candidate indicators. The comparison results are shown in Table 1: Table 1 Comparison of the coefficients of determination (R²) for different similarity indices:
[0040] Based on the above results, this application determines X = |R - 1| as the optimal signal-to-noise ratio similarity index for constructing a quantitative relationship model.
[0041] See Figure 2 The diagram shown is a scatter plot and a fitted curve illustrating the quantitative relationship between the fusion gain Y and the similarity index X = |R - 1| provided in an embodiment of this application. The horizontal axis represents the similarity index X = |R - 1|, and the vertical axis represents the gain. Figure 2 The red line represents the fitted straight line, and the curve represents the actual gain. It can be seen that there is a linear relationship between the similarity index X and the fusion gain Y.
[0042] In one embodiment, predicting fusion gain using a similarity index includes: The fusion gain is calculated using a quantitative relationship model: ,in, and Let be the fitting constant. For fusion gain.
[0043] The quantitative relationship model was obtained by using multiple station pairs with different signal-to-noise ratios as sample datasets and performing linear regression using the least squares method, including: Calculate the signal-to-noise ratio (SNR) of the ZZ cross-correlation function and the SNR of the RR cross-correlation function for the sample dataset; Calculate the ratio of the signal-to-noise ratio of the RR cross-correlation function to the signal-to-noise ratio of the ZZ cross-correlation function; Calculate the similarity index based on the ratio; The signal-to-noise ratio is calculated by linearly superimposing the ZZ cross-correlation function and the RR cross-correlation function. Calculate the fused signal-to-noise ratio gain based on the superimposed signal-to-noise ratio; Using the fusion signal-to-noise ratio gain as the dependent variable and the similarity index as the independent variable, linear regression was performed using the least squares method.
[0044] The formula for calculating the fused signal-to-noise ratio gain is: , in, To fuse signal-to-noise ratio gains, The signal-to-noise ratio after superposition. Let SNR(RR) be the signal-to-noise ratio of the ZZ cross-correlation function, and SNR(RR) be the signal-to-noise ratio of the RR cross-correlation function. To obtain the maximum value.
[0045] In one example, the fused signal-to-noise ratio gain was calculated based on 19 sample data. Similarity index Pearson correlation coefficient This indicates a very strong linear negative correlation. Based on this, using the similarity index... As independent variables, fused signal-to-noise ratio gain Using the least squares method as the dependent variable, linear regression was performed to obtain the quantitative relationship model described in this application: .
[0046] in By fusing signal-to-noise ratio gain This is the form of the dependent variable after fitting.
[0047] In this embodiment, the specific parameters of the fitted quantitative relationship model are as follows: , Therefore, the quantitative relationship model expression is: ,in, The rate of decrease of fusion gain as the difference in signal-to-noise ratio increases was quantitatively characterized; the intercept C represents the maximum relative fusion gain (approximately 32.35%) predicted by the quantitative relationship model when the signal-to-noise ratios of the two components are completely equal (X=0).
[0048] It should be noted that the above quantitative relationship model The core of this study lies in revealing a general linear negative correlation between the fusion gain Y and the signal-to-noise ratio similarity index X. Specifically, the slope... With intercept As a fitting constant, its specific value is obtained by performing linear regression on a specific sample dataset. This does not diminish the value of the model; on the contrary, it reflects its scientific rigor—it demonstrates that the model parameters originate from real-world data and can reflect the statistical characteristics of a specific dataset. When applying this method to different regions or datasets, it can be refitted based on local data to obtain a suitable model. and The value remains the same, while the core relationship that "Y and X are linearly negatively correlated" remains unchanged. Those skilled in the art, based on the teachings of this invention, can obtain specific model parameters suitable for their target scenarios through the described modeling method.
[0049] Coefficient of determination in quantitative relationship model Pearson correlation coefficient (r = -0.9769) and coefficient of determination ( Statistically, this together demonstrates that the fusion gain Similarity index to signal-to-noise ratio There is a stable and reliable linear negative correlation between them, and the quantitative relationship model has a very high goodness of fit.
[0050] Based on the established quantitative relationship model, this application illustrates its specific effects through the following embodiments.
[0051] Input data: A certain station , .
[0052] Calculate the ratio of signal to noise ratio. .
[0053] Similarity index for calculating signal-to-noise ratio .
[0054] Similarity index Substitute into the quantitative relationship model to predict the fusion gain: .
[0055] Target setting: The fusion operation is required to bring at least [a certain result]. (i.e., a 10% gain) The target fusion gain refers to the fusion gain.
[0056] Substituting the target fusion gain into the quantitative relationship model: The conditions that the signal-to-noise ratio similarity index must meet are derived by inverse solving: .
[0057] Therefore, the optimal quality and safety threshold is determined. This quality and safety threshold has a clear physical meaning and is goal-oriented, enabling adaptive optimization of strategy parameters.
[0058] On the other hand, this application provides a Rayleigh wave dual-component fusion system based on model prediction, comprising: a prediction module for acquiring continuously acquired seismic data from station pairs within a target area; calculating the signal-to-noise ratio (SNR) of the ZZ cross-correlation function and the SNR of the RR cross-correlation function for each station pair; calculating the ratio of the SNR of the RR cross-correlation function to the SNR of the ZZ cross-correlation function for each station pair; calculating a similarity index based on the ratio; and using the similarity index to predict the fusion gain. The decision module compares the predicted fusion gain with the quality and safety threshold; based on the comparison result, it selects the output data.
[0059] This embodiment embeds a Rayleigh wave vertical and radial dual-component fusion system into an automation system to achieve real-time quality control. It realizes a closed loop of "prediction-decision-control," eliminates negative gain fusion, and ensures that the process adaptively outputs the optimal signal.
[0060] The following complete verification loop, progressing step by step from theory to practice, confirms the reliability, universality, and superiority of the laws revealed in this application.
[0061] Statistical goodness-of-fit test: Quantitative relationship model based on all 19 samples The statistical test was performed, and the results are as follows: Coefficient of determination (R²): R² = 0.9543. This indicates that the quantitative relationship model can explain 95.43% of the variation in fusion gain (Y).
[0062] Pearson correlation coefficient (r): r = -0.9769. This confirms a very strong linear negative correlation between the fusion gain Y and the similarity index X.
[0063] Root mean square error (RMSE): RMSE = 0.0433. This indicates that the average deviation between the predicted values and the actual observed values of this quantitative relationship model is small, and the prediction accuracy is high.
[0064] Regression significance test (t-test): The test of the slope coefficient of this quantitative relationship model shows that the corresponding p value is less than 0.001, indicating that the linear negative correlation is highly significant at the significance level of 0.001.
[0065] Conclusion: Internal statistical tests confirm that the quantitative relationship model has extremely high explanatory power, predictive accuracy and statistical significance for the sample data.
[0066] Basic Residual Check (Model Hypothesis Proof): To verify the basic assumptions of the quantitative relationship model, a scatter plot of residuals versus sample numbers was plotted. See [link to relevant documentation]. Figure 3 As shown. Figure 3 All scatter points are randomly and uniformly distributed around the zero reference line (y=0), with no obvious pattern. This result indicates that the residuals conform to the characteristics of randomness and independence, which is consistent with the assumptions of a linear model, further supporting the effectiveness of the quantitative relationship model.
[0067] Cross-validation (generalization capability verification): To verify the universality of the quantitative relationship model and prevent overfitting, the hold-out method was used for cross-validation. (1) Data partitioning: The 19 samples were randomly divided into a training set (14 samples) and an independent test set (5 samples).
[0068] (2) Model reconstruction: Using only the training set data for fitting, a new quantitative relationship model is obtained: Y = -1.0865 X+0.3224; The coefficient of determination R² of this quantitative relationship model on the training set is 0.9481, and the root mean square error RMSE is 0.0400.
[0069] (4) Independent testing: Substitute the X values of the test set into the quantitative relationship model above to obtain the predicted value Y2_pred, and compare it with the true value Y2_true. The root mean square error RMSE on the test set is calculated to be 0.0400.
[0070] Verification results: Key indicators and detailed forecast data are shown in Tables 2 and 3.
[0071] Table 2 Comparison of key indicators for cross-validation:
[0072] Table 3. Detailed prediction results for the test set:
[0073] Conclusion: The quantitative relation model maintains the same prediction accuracy (RMSE 0.0400) on completely independent test sets as it does on the training set, and the predicted values are highly consistent with the true values. This proves that the quantitative relation revealed in this application has good generalization ability and is not a random fit to a specific dataset.
[0074] Decision optimization verification (application effectiveness verification): To verify the superiority of the intelligent decision-making method of this application compared with the traditional "fixed threshold method", the following comparative experiment was designed: (1) Experimental design: Evaluation metric: Overall average gain. For each station pair, if the decision is to merge, its actual gain Y_true is included; otherwise, the gain is recorded as 0. Finally, the average gain is calculated for all station pairs.
[0075] Control group (fixed threshold method): Three static quality and safety thresholds were set (δ = 0.20, 0.30, 0.40), and the decision rule was: fusion was performed when |R-1| ≤ δ.
[0076] Experimental group (method of this application): Employing a quantitative relationship model: Make predictions and perform fusion only if Y_pred > 0.05.
[0077] (2) Verification Results: The comparison of the processing results of the two strategies for the same batch of 19 stations is shown in Table 4: Table 4 Comparison of Decision Optimization Verification Results:
[0078] Conclusion: Experimental results show that, based on this application, a higher overall average gain (7.90%) can be achieved while maintaining a comparable number of fusions (8) to different fixed threshold strategies. This demonstrates that this application can accurately quantify the expected value of each fusion, making an optimal trade-off between "fusion quantity" and "fusion quality," thereby effectively overcoming the shortcomings of the fixed threshold method, such as blind parameter setting and low decision-making efficiency.
[0079] External independent dataset validation (universal validation): To verify the model's universality under the most stringent conditions, a completely independent external dataset was used for testing.
[0080] The validation method and external dataset were derived from background noise data collected in a certain region in 2013, comprising 10 station pairs. This dataset had no overlap with the sample set used to establish the internal quantitative relationship model (see detailed implementation) in terms of spatial location, collection time, and geological background, ensuring strict independence of the validation. The validation strictly followed the modeling method described in this application: first, the signal-to-noise ratio similarity index X and fusion gain Y for each station pair in the external dataset were calculated to form data points; then, with X as the independent variable and Y as the dependent variable, linear regression analysis was performed using the least squares method to independently establish its own quantitative relationship model.
[0081] Verification results: Based on the above 10 independent sample points, the fitted quantitative relationship model is as follows: The key statistical indicators of this model are as follows: Coefficient of determination (R²): 0.9475; Pearson correlation coefficient (r): -0.9734; Root mean square error (RMSE): 0.0818; Detailed sample data, model predictions, and residuals are shown in Table 5: Table 5. Detailed results of validation on external independent datasets:
[0082] The results analysis, using the method described in this application to model a comprehensive independent validation set, yielded a model with a determination coefficient as high as 0.9475, strongly demonstrating the high accuracy and universality of the "linear negative correlation between Y and X" rule. In particular, extreme samples with extremely large signal-to-noise ratio differences (X=0.7924) and severe fusion failures (Y=-0.7715) showed a high degree of agreement with the model predictions, perfectly validating the core theory of this application in practice. Furthermore, the method can stably build high-performance models even with multi-source mixed data, proving its excellent robustness and cross-scenario applicability.
[0083] The verification results conclusively demonstrate that the linear negative correlation law revealed in this application is an objective and stable scientific discovery; and the provided quantitative modeling method is a highly reliable and plug-and-play technological innovation that can provide accurate and optimizable decision support for Rayleigh wave dual-component fusion.
[0084] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.
Claims
1. A Rayleigh wave two-component fusion method based on model prediction, characterized in that, The method includes: Acquire continuously collected seismic data from stations within the target area; Calculate the signal-to-noise ratio (SNR) of the ZZ cross-correlation function and the SNR of the RR cross-correlation function for each station pair; Calculate the ratio of the signal-to-noise ratio of the RR cross-correlation function to the signal-to-noise ratio of the ZZ cross-correlation function for each station pair; Calculate the similarity index based on the ratio; The similarity index is used to predict the fusion gain; Compare the predicted fusion gain with the quality and safety threshold; Based on the comparison results, the final valid signal is selected and output.
2. The Rayleigh wave dual-component fusion method based on model prediction according to claim 1, characterized in that, The similarity index is calculated based on the ratio: , As a similarity index, It is a ratio.
3. The Rayleigh wave dual-component fusion method based on model prediction according to claim 2, characterized in that, The similarity index is used to predict the fusion gain, including: The fusion gain is calculated using a quantitative relationship model: ,in, and Let be the fitting constant. For fusion gain.
4. The Rayleigh wave dual-component fusion method based on model prediction according to claim 3, characterized in that, The quantitative relationship model is obtained by using multiple station pairs with different signal-to-noise ratios as sample datasets and performing linear regression using the least squares method, including: Calculate the signal-to-noise ratio (SNR) of the ZZ cross-correlation function and the SNR of the RR cross-correlation function for the sample dataset; Calculate the ratio of the signal-to-noise ratio of the RR cross-correlation function to the signal-to-noise ratio of the ZZ cross-correlation function; Calculate the similarity index based on the ratio; The signal-to-noise ratio is calculated by linearly superimposing the ZZ cross-correlation function and the RR cross-correlation function. Calculate the fused signal-to-noise ratio gain based on the superimposed signal-to-noise ratio; Using the fusion signal-to-noise ratio gain as the dependent variable and the similarity index as the independent variable, linear regression was performed using the least squares method.
5. The Rayleigh wave dual-component fusion method based on model prediction according to claim 4, characterized in that, The formula for calculating the fusion signal-to-noise ratio gain is as follows: , in, To fuse signal-to-noise ratio gains, The signal-to-noise ratio after superposition. Let SNR(RR) be the signal-to-noise ratio of the ZZ cross-correlation function, and SNR(RR) be the signal-to-noise ratio of the RR cross-correlation function. To obtain the maximum value.
6. The Rayleigh wave dual-component fusion method based on model prediction according to claim 3, characterized in that, If the predicted fusion gain is greater than or equal to the quality and safety threshold, the ZZ cross-correlation function and RR cross-correlation function of the station pair are superimposed as the final effective signal; if the predicted fusion gain is less than the quality and safety threshold, the signal-to-noise ratio of the ZZ cross-correlation function and the RR cross-correlation function of the station pair are compared, and the cross-correlation function with the larger signal-to-noise ratio is selected as the final effective signal.
7. A Rayleigh wave two-component fusion system based on model prediction. Its features are, Includes: a prediction module for acquiring continuously acquired seismic data from stations within the target area; calculating the signal-to-noise ratio (SNR) of the ZZ cross-correlation function and the SNR of the RR cross-correlation function for each station pair; calculating the ratio of the SNR of the RR cross-correlation function to the SNR of the ZZ cross-correlation function for each station pair; calculating a similarity index based on the ratio; and using the similarity index to predict the fusion gain. The decision module compares the predicted fusion gain with the quality and safety threshold; based on the comparison result, it selects and outputs the final valid signal.
8. The Rayleigh wave dual-component fusion system based on model prediction according to claim 7, characterized in that, The similarity index is calculated based on the ratio: , As a similarity index, It is a ratio.
9. A Rayleigh wave dual-component fusion system based on model prediction according to claim 7, characterized in that, The similarity index is used to predict the fusion gain, including: The fusion gain is calculated using a quantitative relationship model: ,in, and Let be the fitting constant. For fusion gain; The quantitative relationship model is obtained by using multiple station pairs with different signal-to-noise ratios as sample datasets and performing linear regression using the least squares method, including: Calculate the signal-to-noise ratio (SNR) of the ZZ cross-correlation function and the SNR of the RR cross-correlation function for the sample dataset; Calculate the ratio of the signal-to-noise ratio of the RR cross-correlation function to the signal-to-noise ratio of the ZZ cross-correlation function; Calculate the similarity index based on the ratio; The signal-to-noise ratio is calculated by linearly superimposing the ZZ cross-correlation function and the RR cross-correlation function. Calculate the fused signal-to-noise ratio gain based on the superimposed signal-to-noise ratio; Using the fusion signal-to-noise ratio gain as the dependent variable and the similarity index as the independent variable, linear regression was performed using the least squares method.
10. A Rayleigh wave dual-component fusion system based on model prediction according to claim 9, characterized in that, By setting a target gain and then inversely solving for the similarity index, the obtained similarity index is used as a quality and safety threshold.