An automatic seismic wave data interpretation method with prior knowledge embedding

Abstract

The application discloses a prior knowledge embedded automatic interpretation method of seismic wave data, and relates to the technical field of geological exploration; the method not only completes the value range calibration of feature data belonging to each classification label through a normal distribution mathematical statistics mode, and realizes the fusion of a data decision layer by using a D-S theory, so that secondary classification interpretation of seismic wave data on the basis of clustering is automatically realized by taking a KNN algorithm and the D-S evidence theory as cores; the embedding of important prior knowledge of real rock mass quality is realized by taking the data analysis result of the excavated scale data as a reference and combining the calculation of a clustering center distance, so that the mapping from the clustering label to the rock mass quality label is completed, and the rationality and reliability of label classification are fully ensured; the method significantly improves the interpretation efficiency of seismic wave data, greatly saves time and manpower cost, and guarantees the accuracy and reliability of the interpretation result, and is suitable for large-scale application in the interpretation of seismic wave data in geological exploration.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geological exploration, in particular to the interpretation technology of seismic wave data, and particularly relates to a priori knowledge embedded seismic wave data automatic interpretation method. BACKGROUND

[0002] The current highway tunnel construction is faced with the trend that the engineering geological conditions of the tunnel address region are increasingly complex, and adverse geological conditions such as fracture zones, karst caves, sudden mud gushing and the like are still the focus of attention in construction, which seriously threatens the safety of construction. As an important means of ascertaining geological information at a location during tunnel construction, the accuracy of advanced geological prediction directly affects the development of safety work in the construction process. Among them, the seismic wave detection method can timely predict the quality of the rock-soil mass in front of the excavation due to its characteristics of long prediction distance and non-destructive detection, and is one of the common advanced geological means in the tunnel field, and the research on the data interpretation thereof has been the focus of research in this field.

[0003] Seismic wave detection is essentially a geophysical detection method, that is, a method for predicting the geological conditions and rock property changes in front of the tunnel excavation face by using the propagation signal characteristics of elastic waves in different media. According to the difference in principle details, it is specifically divided into TSP, TST, TGS and the like, but in the data interpretation process, various methods currently mainly perform quantitative calculation on the propagation signal data to obtain equivalent rock physical and mechanical parameters including Young's modulus, Poisson's ratio, transverse wave and longitudinal wave velocity, and then realize the judgment of rock quality and various adverse geological conditions through subjective experience judgment of technical personnel. This interpretation method has strong subjectivity on the one hand, is prone to misjudgment and omission by technical personnel, and the interpretation standards are not unified, on the other hand, the seismic wave data is relatively large, and the realization of high reliability interpretation by relying on manual work may lead to high cost and long construction time. SUMMARY

[0004] The present application aims to overcome the problems of poor accuracy and reliability and long interpretation time of the existing seismic wave data interpretation method, and proposes a priori knowledge embedded seismic wave data automatic interpretation method.

[0005] In order to achieve the above-mentioned application purpose, the present application provides a priori knowledge embedded seismic wave data automatic interpretation method, comprising the following steps:

[0006] (1) Data preprocessing: preprocessing the original data of seismic wave detection, and dividing the preprocessed data into excavated scale data and unexcavated scale data;

[0007] (2) Data analysis: The mean value of the excavated scale data is processed, and the mean value of the excavated scale data is taken as the rock mass quality characteristic value combined with the directly observed rock mass quality, and the rock mass quality label (quantitative representation of prior knowledge) is assigned;

[0008] The K nearest neighbor clustering algorithm is used to cluster the unexcavated scale data, and the data clustering results of the unexcavated scale data under unsupervised learning and the clustering centers of the clustered data are obtained.

[0009] (3) Interpretation label division: Based on the clustering centers of the clustered data, the distances from the clustering centers to the rock mass quality characteristic values are calculated, and different rock mass quality labels are assigned to the clustered unexcavated data according to the size of the distance; wherein the clustered data with the smallest distance from the clustering center to the rock mass quality characteristic value is assigned the same label as the excavated rock mass quality, and other distances are assigned other rock mass quality labels based on this as a reference;

[0010] (4) Range calibration: The data in different rock mass quality labels are fitted with normal distribution to obtain the distribution characteristics of the data of the rock mass quality label, and the value range of the data of the rock mass quality label is obtained according to the set confidence interval;

[0011] (5) Data fusion interpretation: Based on the value range of the data of different rock mass quality labels, the D-S evidence theory is used to fuse and interpret the seismic wave data to obtain the automatic interpretation result of the seismic wave detection data.

[0012] The prior knowledge embedded seismic wave data automatic interpretation method of the application not only completes the value range calibration of the feature data belonging to each classification label by the normal distribution mathematical statistics, and realizes the fusion of the data decision layer by using the D-S theory, so as to automatically realize the secondary classification interpretation of the seismic wave data based on clustering by taking the KNN algorithm and the D-S evidence theory as the core; also, through the prior knowledge embedding concept, the data analysis result of the excavated scale data is taken as a reference, combined with the calculation of the clustering center distance, the embedding of the important prior knowledge of the real rock mass quality is realized, so as to complete the mapping of the clustering label to the rock mass quality label, and the rationality and reliability of the label classification are fully guaranteed; the automatic interpretation method of the application significantly improves the interpretation efficiency of the seismic wave data, greatly saves the time and labor cost, and guarantees the accuracy and reliability of the interpretation result, and is suitable for large-scale application in the interpretation of seismic wave data in geological exploration.

[0013] In step (1), the preprocessing includes data thinning, feature engineering, correlation analysis and data standardization.

[0014] In the seismic wave detection, the data density is usually too large because the elastic wave is continuously recorded during data acquisition. In order to ensure the efficiency of interpretation and the effectiveness of data, it is necessary to perform thinning processing on the original acquisition data; the data thinning refers to a processing mode that divides the original data according to a limited distance and takes the mean value of the divided paragraph as the representative value of the paragraph; which can be represented by the following formula: In the formula, n is the number of samples in the divided paragraph, x n is the sample value, and x is the average value (thinned value).

[0015] In the seismic wave detection result, the equivalent rock physical and mechanical parameters collected are slightly different according to different devices and technical means (TSP, TRT, TGS, etc.), and basically include several common and important characteristic indexes such as Young's modulus, Poisson's ratio, shear wave velocity, longitudinal wave velocity, and wave velocity ratio, and the fluctuation (dispersion) of the above data is also an important basis for subjective interpretation; in order to reflect the dispersion of the characteristics, the standard deviation of the data in each divided paragraph can be calculated on the basis of equidistant division in data thinning, that is, the characteristic engineering refers to a processing mode that calculates the standard deviation of the data in each divided paragraph based on equidistant division in data thinning.

[0016] In the field of machine learning, in order to speed up data fitting and reduce data dimension, it is usually necessary to perform feature correlation analysis to eliminate features with high correlation; preferably, the correlation analysis is a Pearson feature correlation (Pearson correlation coefficient) analysis algorithm, which is a statistical method for measuring the strength and direction of the linear relationship between two variables; the specific formula is as follows: In the formula, a i and b i are the sample values of the two features, and are the sample means of the two features, p is the Pearson correlation coefficient, and n is the sample number of the feature; p is between -1 and +1, which respectively represents the degree of negative correlation and positive correlation, and the closer to 0 indicates the weaker correlation.

[0017] The data standardization is a general process in the field of mathematical statistics, which refers to scaling the values of different dimensionless indexes to the interval range of (0, 1) to eliminate the influence of dimension.

[0018] The excavated scale data is the seismic wave detection data that has been verified by excavation or other means that can directly observe the rock mass quality; the unexcavated scale data is the seismic wave detection data that cannot or temporarily cannot be verified by directly observing the rock mass quality.

[0019] In step (2), the method of calculating the feature mean value is the mean value method.

[0020] The K-Nearest Neighbors (KNN) is a basic classification and regression method, that is, the k training samples closest to the target sample in the feature space are found, so that the label of the new sample is determined by the label of the training sample (that is: find the k training samples (neighbors) closest to the new sample; the class with the highest frequency in the k neighbors will be assigned to the label of the new sample); k is a positive integer, representing the number of "training samples"; preferably, the distance calculation formula is: In the formula, x and y represent the horizontal and vertical coordinates of the sample, respectively, x i and y i are the coordinate values of sample x and sample y in the i-th dimension, and n is the dimension number of the sample.

[0021] In step (4), the normal distribution is also called Gaussian distribution, which is a continuous probability distribution; preferably, the probability density function (PDF) is: In the formula, μ is the mean of the distribution, σ is the standard deviation of the distribution, σ 2 is the variance of the distribution, x is the random variable value, and f is the probability density function.

[0022] The confidence interval is an interval used to estimate the unknown value of a parameter. For a normal distribution, the quantile function can be used to calculate the confidence interval at a certain confidence level; preferably, the confidence interval is: In the formula, z is the quantile of the normal distribution, μ is the mean of the distribution, σ is the standard deviation of the distribution, and n is the sample size; for example, calculating a 90% confidence interval means that the confidence interval covers 90% of the data in the middle, excluding 5% of the data at both ends of the normal distribution.

[0023] In step (5), the D-S evidence theory is used to select a suitable BPA calculation function during fusion, and the conflict factor value of the data sample is calculated; according to the value of the conflict factor, the data with high conflict (k is a value between 0 and 1, the closer to 1, the higher the conflict, generally not less than 0.9) is processed to weaken the conflict, and finally the fusion result is determined according to the maximum membership principle (for example: a sample belongs to labels 1, 2 and 3 with probabilities of 0.2, 0.5 and 0.3 respectively, then the classification result is determined as label 2), and the automatic interpretation result of the seismic wave detection data is obtained.

[0024] Preferably, the BPA calculation function is an interval Euclidean distance model, specifically: let be the value of the jth evaluation index, and F i = [f i- f i + ]for the range value of the i-th rock mass quality label, the interval Euclidean distance of the two is: The basic probability (BPA) calculation function is:

[0025] Preferably, the conflict weakening processing refers to using the average evidence method to weaken the conflict degree between data; the specific steps are as follows: assuming that there are n pieces of evidence to be fused m j , j = 1, 2,..., n are highly conflictive, the average evidence body is calculated by using the formula , and the solution of the average evidence body m MEA (A) is completed. After that, the D-S fusion rule is continued to perform the fusion processing, and the conflict weakening processing is repeated for n-1 times, so that the final fusion result can be obtained.

[0026] Preferably, the D-S evidence theory is as follows: assuming that θ = {F1, F2,..., Fn} and φ = {A1, A2,..., A n} are the identification framework and the evidence set of the multi-source information fusion problem respectively, and there is a function m: 2 θ → [0, 1] satisfying the condition In the formula: is an empty set; m(A j (F i )) is the support probability of the evidence A j to the subset F i in the identification framework; it is called that m is a basic probability distribution function of the identification framework θ, and the calculation result is called a basic distribution probability BPA (Basic probability assignment); when the BPA of each evidence body in the evidence set is obtained from multiple information sources, these evidences need to be fused according to certain rules; the conventional D-S fusion rule is as follows:

[0027]

[0028] In the formula, K is called a conflict factor; the conflict factor K is used to describe the conflict between the evidences in the D-S evidence theory, and the greater K is, the greater the conflict between the evidence bodies is, which means that the reliability of the fused decision result is worse.

[0029] Further, the application further proposes a deciphering device, comprising at least one processor, and a memory in communication connection with the at least one processor; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the prior knowledge embedded automatic deciphering method of seismic wave data.

[0030] Further, the application further proposes a computer readable medium, which stores instructions executable by a processor, and the instructions, when executed by the processor, enable the processor to execute the prior knowledge embedded automatic deciphering method of seismic wave data.

[0031] Further, the application further proposes an application of the prior knowledge embedded automatic deciphering method of seismic wave data in constructing a geological model.

[0032] Compared with the prior art, the application has the following beneficial effects:

[0033] 1、The automatic deciphering method of seismic wave data completes the value range calibration of the feature data belonging to each clustering label through the normal distribution mathematical statistics, and realizes the fusion of the data decision layer by using the D-S theory, so that the secondary classification deciphering of the seismic wave data on the basis of clustering is automatically realized by taking the KNN algorithm and the D-S evidence theory as the core, the deciphering efficiency of the seismic wave data is significantly improved, and the time and labor cost are greatly saved.

[0034] 2、The automatic deciphering method of seismic wave data realizes the embedding of the important prior knowledge of the real rock mass quality by using the data analysis result of the excavated scale data as a reference and combining the calculation of the clustering center distance, so that the mapping of the clustering label to the rock mass quality deciphering label is completed, the rationality and reliability of the label classification are fully ensured, and the accuracy and reliability of the deciphering result are ensured. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 It is a flowchart of the automatic deciphering method of seismic wave data in the embodiment 1 of the application;

[0036] Figure 2 It is a heat map of the correlation analysis result of the seismic wave data in the embodiment 1 of the application;

[0037] Figure 3 It is a clustering scatter plot of the seismic wave data in the embodiment 1 of the application;

[0038] Figure 4 It is a normal distribution fitting effect diagram of the clustering data deciphering label (feature wave velocity ratio positive) in the embodiment 1 of the application;

[0039] Figure 5 Figure of normal distribution fitting effect of the clustered data interpretation label (characteristic transverse wave velocity) in Embodiment 1 of the present application;

[0040] Figure 6 Figure of normal distribution fitting effect of the clustered data interpretation label (characteristic longitudinal wave velocity) in Embodiment 1 of the present application;

[0041] Figure 7 Section and three-dimensional synthetic model obtained after the automatic interpretation result of the seismic wave data in Embodiment 4 of the present application is used for geological model construction. DETAILED DESCRIPTION

[0042] The present application will be further described in detail below in combination with the drawings and specific embodiments. However, this should not be understood as limiting the scope of the above-mentioned subject matter of the present application to the following embodiments, and any technology realized based on the content of the present application falls within the scope of the present application.

[0043] Embodiment 1

[0044] This embodiment is based on the TGS seismic wave detection data obtained in the construction process of a certain actual tunnel project in Guangxi, and the specific steps include:

[0045] (1) According to the actual situation on site, three types of rock mass quality are interpreted according to the detection data, i.e., three types of interpretation labels (rock mass quality labels) are complete rock mass, broken rock mass and weak interlayer, respectively;

[0046] After data extraction by the TGS matching equipment, the original seismic wave detection data has three indexes, which are transverse wave vs, longitudinal wave vp and wave velocity ratio vp / vs, respectively;

[0047] Among them, in the data preprocessing process, the equidistant segmentation d is selected as 0.5 m when the data is thinned, and the standard deviation within the segmentation paragraph is calculated for the longitudinal wave vp to represent the discrete wave fluctuation degree when the feature engineering is performed, and a new feature vpStd is derived. The correlation analysis result of the four features (rock mass quality labels: complete rock mass, broken rock mass, weak interlayer, vpStd) in total is shown in Table 1. Figure 2 and the correlation result is used to eliminate the characteristic longitudinal wave vp; Figure 2

[0048] Among them, the correlation analysis adopts the Pearson feature correlation (Pearson correlation coefficient) analysis algorithm, and the specific formula is as follows: In the formula, a i and b i are the sample values of the two features, respectively, and ​where is the sample mean of the two features, is the Pearson correlation coefficient, and n is the sample size of the feature. The correlation coefficient,, ranges between -1 and +1, indicating the degree of negative and positive correlation, respectively. The closer to 0, the weaker the correlation.

[0049] When dividing the preprocessed seismic wave data into excavated (tunnel face) scale data and unexcavated scale data, the excavated scale is set to the first 5 m of the detection range, and the remaining part is the unexcavated scale data.

[0050] (2) The unexcavated scale data is clustered by the K-Nearest Neighbor clustering algorithm (k=3) to obtain the data clustering results (cluster labels are Cluster0, Cluster1, and Cluster2, and the clustering results are shown in Figure 3 ) under unsupervised learning and the cluster centers of each cluster data;

[0051] (3) Based on the cluster centers of the clustered data, the distances from each cluster center to the rock mass quality characteristic value are calculated, and different rock mass quality labels are assigned to the clustered unexcavated data according to the size of the distance. The cluster data with the smallest distance from the cluster center to the rock mass quality characteristic value has the same label as the excavated rock mass quality, and other cluster data with different distances are assigned with other rock mass quality labels (the rock mass quality label is “broken rock mass”, D1=1.91, D2=0.72, and D3=1.44 are calculated, so the rock mass quality interpretation label division results Cluster1 is “broken rock mass”, Cluster2 is “intact rock mass”, and Cluster0 is “soft interlayer”);

[0052] (4) When performing normal distribution mathematical statistics and value range division on the data (unexcavated rock mass quality characteristic values) in different rock mass quality labels, the normal distribution fitting effect is shown in Figure 4 , Figure 5 , Figure 6 The 90% confidence interval is selected, and the final value range calibration results of each rock mass quality label data are shown in Table 1;

[0053] where is the mean of the distribution, is the standard deviation of the distribution, is the variance of the distribution, x is the random variable value, and f is the probability density function. 2

[0054] where is the confidence interval: where z is the quantile of the normal distribution, is the mean of the distribution, is the standard deviation of the distribution, and n is the sample size.

[0055] ​​Table 1 Value range of rock mass quality label data under 90% confidence interval

[0056] Features Cluster2 relatively intact rock mass Cluster1 fractured rock mass Cluster0 weak interlayer v p / v s ]]> -0.07～1.85 -1.92～0.76 -1.43～1.4 v s ]]> -2.19～0.14 -0.81～1.47 0.23～1.69 v p Std]]> -1.10～0.13 -1.09～0.54 -0.21～3.53

[0057] (5) In the fusion process, a suitable BPA calculation function is selected by D-S evidence theory, and the conflict factor value of the data sample is calculated; according to the value of the conflict factor, the data with high conflict (the conflict factor threshold is set to K = 0.90, the proportion of samples corrected by D-S evidence theory accounts for 13% of the total samples, the average conflict factor K of all samples is 0.92, which belongs to the category of high conflict) is subjected to conflict weakening processing, and finally the fusion result is determined according to the maximum membership principle to obtain the automatic interpretation result of the seismic wave detection data.

[0058] wherein the BPA calculation function is an interval Euclidean distance model, and specifically, let be the value of the jth evaluation index, F i i - i + be the range value of the ith rock mass quality label, and the interval Euclidean distance between the two is: The basic probability (BPA) calculation function is:

[0059] wherein the conflict weakening processing specifically comprises: assuming that there are n pieces of evidence to be fused m j , j = 1, 2, …, n, which have high conflict, then the average evidence body is calculated by using the formula , and after the solution of the average evidence body m MEA (A) is completed, the fusion processing is continued by using the D-S fusion rule, and the conflict weakening processing is repeated for n-1 times, so that the final fusion result is obtained.

[0060] Embodiment 2

[0061] An interpretation device comprises at least one processor and a memory in communication connection with the at least one processor; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the priori knowledge embedded seismic wave data automatic interpretation method in embodiment 1.

[0062] Embodiment 3

[0063] A computer readable medium having instructions executable by a processor stored thereon, wherein the instructions, when executed by the processor, cause the processor to execute the priori knowledge embedded seismic wave data automatic interpretation method in embodiment 1.

[0064] ​​Example 4

[0065] The final interpretation results obtained in Example 1 were visualized (modeled), including five two-dimensional profiles selected at tunnel levels and one synthetic three-dimensional entity model (as shown in Figure 7

[0066] Finally, it should be noted that the above detailed description of the embodiments is only the preferred practice of the application, and cannot limit the scope of the application. The technical solutions described in the above embodiments can be replaced by equivalents, and the essence of the corresponding technical solutions does not deviate from the scope of the technical solutions of the embodiments of the application. They should be covered in the scope of the claims and description of the application.​

Claims

1. A method for automatically interpreting seismic wave data with prior knowledge embedded, characterized in that: The following steps are involved: (1) Data preprocessing: Preprocess the raw data of seismic wave detection and divide the preprocessed data into excavated scale data and unexcavated scale data; (2) Data analysis: The excavated scale data are processed by averaging, and combined with the directly observed rock mass quality, the mean of the excavated scale data is used as the rock mass quality characteristic value and assigned a rock mass quality label; The unexcavated scale data are clustered using the K-nearest neighbor clustering algorithm to obtain the data clustering results of the unexcavated scale data under unsupervised learning and the cluster centers of each cluster data. (3) Interpretation label division: Based on the cluster center of the cluster data, the distance from each cluster center to the rock mass quality characteristic value is calculated, and different rock mass quality labels are assigned to the unexcavated data after clustering according to the size of the distance. Among them, the cluster data with the smallest distance from the cluster center to the rock mass quality characteristic value is assigned the same label as the excavated rock mass quality, and based on this, the cluster data with other distances are assigned different rock mass quality labels; (4) Range calibration: perform normal distribution fitting on the data in different rock mass quality labels to obtain the distribution characteristics of the rock mass quality label data, and obtain the value range of the rock mass quality label data according to the set confidence interval; (5) Data fusion interpretation: Based on the value range of different rock mass quality label data, the DS evidence theory is used to fuse and interpret the seismic wave data to obtain the automatic interpretation results of the seismic wave detection data.

2. The method for automatic interpretation of seismic wave data according to claim 1, characterized in that: The preprocessing includes data rarefaction, feature engineering, correlation analysis, and data standardization.

3. The method for automatic interpretation of seismic wave data according to claim 1, characterized in that: In step (4), the normal distribution is a kind of continuous probability distribution; the density function of the probability is: Where μ is the mean of the distribution, σ is the standard deviation of the distribution, and 2 is the variance of the distribution, x is the value of the random variable, and f is the density function of the probability.

4. The method for automatic interpretation of seismic wave data according to claim 1, characterized in that: In step (4), the confidence interval is: Where z is the quantile of the normal distribution, μ is the mean of the distribution, σ is the standard deviation of the distribution, and n is the sample size.

5. The method for automatic interpretation of seismic wave data according to claim 1, characterized in that: In step (5), the fusion interpretation includes: selecting a suitable BPA calculation function through DS evidence theory and calculating the conflict factor value of the data sample; performing conflict weakening processing on the data with high conflict according to the value of the conflict factor, and finally determining and interpreting the fusion result according to the maximum membership principle to obtain the automatic interpretation result of the seismic wave detection data.

6. The method for automatic interpretation of seismic wave data according to claim 5, characterized in that: The BPA calculation function is the interval Euclidean distance model, specifically: is the value of the jth evaluation index, F in the index system i =[f i - ,f i + ] is the range value of the i-th rock mass quality label, and the Euclidean distance between the two intervals is: Then the basic probability calculation function is:

7. The method for automatic interpretation of seismic wave data according to claim 5, characterized in that: The conflict weakening process refers to using the average evidence method to weaken the conflict between data. The specific steps are: suppose there are n evidence bodies m to be fused j , j=1,2,…,n has high conflict, then use the formula Calculate the average body of evidence and complete the average body of evidence m MEA After solving (A), continue the DS fusion rule to perform fusion processing, repeat the iterative conflict weakening process n-1 times, and then you can get the final fusion result.

8. An automatic interpretation device for seismic wave data, characterized in that: It includes at least one processor and a memory communicatively connected to the at least one processor; the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method for automatic interpretation of seismic wave data as described in any one of claims 1-7.

9. A computer-readable medium, characterized in that Instructions executable by a processor are stored thereon, and when the instructions are executed by the processor, the processor executes the method for automatically interpreting seismic wave data according to any one of claims 1 to 7.

10. Application of the automatic interpretation method of seismic wave data embedded with prior knowledge according to any one of claims 1 to 7 in constructing a geological model.

Images