A method and device for effectively picking up borehole mode waves based on density clustering

Abstract

The present invention discloses a method and device for effectively extracting the dispersion of borehole pattern waves based on density clustering, comprising: extracting the dispersion information of acoustic waves from the waveform information recorded by the acoustic logging instrument through a dispersion extraction algorithm; limiting the range of the obtained original dispersion data; standardizing the data obtained after the limited range; clustering the obtained data using the DBSCAN algorithm to divide the data set into different clusters; performing inverse standardization on the processed data; and judging the quality of the data by the number of data points contained in the main cluster of the data set. The device comprises: a slowness-frequency domain transformation module, a data clustering processing module, and a target main cluster identification module. The present invention can solve the technical problems of traditional dispersion analysis and processing of array acoustic logging data, such as the presence of noise points and pseudo-solution points in the slowness-frequency domain, which in turn affects subsequent dispersion analysis.

Description

Technical Field

[0001] The present invention relates to the technical field of applied geophysics and petroleum exploration and development, and in particular to a method for effectively separating the frequency dispersion of borehole mode waves based on density clustering. Background Art

[0002] Dispersion analysis is a key method for processing array acoustic logging data. Waveforms in actual well logging are often dispersive borehole mode waves. These include, but are not limited to, Stoneley waves and various orders of pseudo-Rayleigh waves excited by monopole sources, flexural waves excited by dipole sources, spiral waves excited by quaternary sources, and P-waves excited by various acoustic sources in soft formations. The dispersion of borehole mode waves can be further used for formation information inversion.

[0003] The acoustic waveform signals recorded by the instrument's receiver array must first be transformed from the space-time domain into the slowness-frequency domain using a specific dispersion extraction algorithm to further extract the waveform's dispersion information. Extensive research has been conducted on dispersion extraction methods, which can be categorized into single-mode extraction algorithms and multi-mode extraction algorithms. The Prony method is the earliest multi-mode extraction method, which was improved by Lang et al. and applied to the processing of sonic logging data (Lang SW, Kurkjian AL, McClellan JH, et al. Estimating slowness dispersion from arrays of sonic logging waveforms [J]. Geophysics, 1987, 52 (4): 530-544.); Ekstrom proposed the Matrix Pencil method, which has improved the noise resistance and computational efficiency compared with the Prony method (Ekstrom MP. Dispersion estimation from borehole acoustic arrays using a modified matrix pencil algorithm [C] / / Conference Record of the Twenty-Ninth Asian Conference on Signals, Systems and Computers. IEEE, 1996, 1: 449-453.). The WSS method proposed by Nolte improves the stability of the calculation results by constructing a spectrally weighted correlation function (Nolte B, Rao R, Huang X. Dispersion analysis of split flexural waves[R]. Massachusetts Institute of Technology. Earth Resources Laboratory, 1997.). This method is very suitable for extracting dispersion from dipole wave trains containing only flexural wave modes and is currently a widely used single-mode dispersion extraction method.

[0004] Although dispersion extraction algorithms have made significant progress, actual logging processes are inevitably affected by noise and the limited operating frequency band of the instrument. This results in the presence of noise and spurious solutions in the slowness-frequency domain, which in turn affects subsequent dispersion analysis. Therefore, it is essential to develop a data processing method that can separate the raw slowness-frequency domain dispersion from this noise and spurious solution. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and device for effectively picking up the dispersion of borehole mode waves based on density clustering, so as to solve the technical problems such as the presence of noise points and pseudo-solution points in the slowness-frequency domain in traditional dispersion analysis and processing of array acoustic logging data, which in turn affects subsequent dispersion analysis.

[0006] The present invention adopts the following processing scheme:

[0007] Step 1: extracting the dispersion information of the acoustic wave from the waveform information recorded by the acoustic logging instrument through a dispersion extraction algorithm, including the Prony method, the Matrix Pencil method, or the weighted spectral coherence method;

[0008] Step 2: Limit the range of the original dispersion data obtained in step 1;

[0009] Step 3: standardize the data within the limited range obtained in step 2;

[0010] Step 4: Use the DBSCAN algorithm to cluster the data obtained in step 3 and divide the data set into different clusters;

[0011] Step 5: Denormalize the data obtained in step 4;

[0012] Step 6: The data quality is judged by the number of data points contained in the main cluster of the dataset. By default, a dataset with a main cluster consisting of less than 100 points corresponds to a dataset with poor data quality, but this standard needs to be adjusted according to the specific data set. For data with good data quality, the main cluster of the dataset is the target dispersion. For data with poor data quality, we use the monotonicity of the dispersion curve to select appropriate clusters from the clustering results. These clusters constitute the target dispersion.

[0013] The standardization processing algorithm in step 3 is:

[0014] X stand =(X-μ) / σ (1)

[0015] Where, X stand is the standardized data, X is the original data, μ and σ are the mean (mean) and standard deviation (std) of the sample data respectively.

[0016] DBSCAN (Density-Based Spatial Clustering of Application with Noise) in step 4 is a classic density clustering method. The algorithm is based on the ε neighborhood and the minimum number of data points contained in the ε neighborhood, MinPts, and divides the relationship between data points into core objects, density direct, density reachable, and density connected. For a given data set D = {x1, x2, x3, ..., x m},

[0017] (1) If ||x j -x i ||2≤ε, then x j Located at x i In the ε neighborhood of ;

[0018] (2) If x i The ε neighborhood of x contains at least MinPts data points, then x is called i As the core object;

[0019] (3) If x j Located in the core object x i In the ε neighborhood of j From x i directly density-reachable;

[0020] (4) If there is a sample sequence (p1, p2, p3, ..., p n ), and p i+1 It is from p i (i=1,2,3,…n-1) directly density accessible, then it is called p n is density-reachable by p1. Obviously, the sample sequence (p1, p2, p3, ..., p n ) only p n Can be a non-core object;

[0021] (5) If there is a core object x i , so that O1 and O2 are both represented by x i If the density is reachable, then O1 and O2 are said to be density-connected;

[0022] The DBSCAN algorithm is described as follows: first, all core objects in the dataset are determined based on the ε neighborhood specified by a set of parameters (ε, MinPts), and then any core object x is selected. i Find all the iDensity-reachable objects, the collection of these objects is a cluster, and finally repeat the previous step for the unclassified core objects until all core objects are classified.

[0023] The present invention also provides a density-clustered borehole mode wave effective dispersion processing device, comprising:

[0024] The slowness-frequency domain transformation module transforms the waveform in the space-time domain into the slowness-frequency domain and uses the dispersion extraction algorithm to extract the dispersion information of the sound wave;

[0025] The data clustering processing module performs dispersion normalization processing and uses the DBSCAN algorithm to draw different clusters;

[0026] The target main cluster identification module sets the main cluster of the data set as the target dispersion for data with good data quality.

[0027] The present invention has the following advantages:

[0028] Based on a density clustering algorithm called DBSCAN and incorporating acoustic dispersion characteristics, this paper developed a data processing method for effectively separating the dispersion of borehole mode waves. Field data application demonstrates the advantages of this method: it achieves efficient separation of the dispersion of commonly used monopolar Stoneley waves and dipole flexural waves, regardless of data quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 This is a flow chart of the effective dispersion separation algorithm of borehole mode waves based on density clustering of the present invention.

[0030] Figure 2 A structural diagram of the density-clustered borehole mode wave effective dispersion picking and processing device provided by the present invention.

[0031] Figures 3(a), (b), (c), and (d) show the separation effect of the algorithm of the present invention on the dipole bending wave dispersion.

[0032] Figure 4 (a) and (b) show the monopole Stoneley wave dispersion separated by the algorithm of the present invention.

[0033] Figures 5(a) and (b) show the dipole bending wave dispersion separated from field data of poor quality by the algorithm of the present invention. DETAILED DESCRIPTION

[0034] The method of the present invention is further described below with reference to the accompanying drawings and specific processing examples so that those skilled in the art can better understand the method and implement it. However, the examples are not intended to limit the present invention.

[0035] This paper proposes an effective dispersion separation algorithm for borehole mode waves based on density clustering, see Figure 1 As shown, the workflow is as follows:

[0036] Step 1: The acoustic logging instrument records waveform information, from which the dispersion information of the acoustic wave needs to be extracted using a dispersion extraction algorithm such as the Prony method, the Matrix Pencil method, or the weighted spectral coherence method.

[0037] The weighted spectral coherence method was used to extract the dispersion of a field dipole acoustic logging data. The results are as follows: Figure 2 Figure 3(a) shows a flexural wave waveform, and Figure 3(b) shows the dispersion extracted from it. It can be seen that in addition to the flexural wave dispersion, the weighted spectral coherence method also extracts many erroneous modes, which creates an obstacle to the dispersion analysis of acoustic wave data.

[0038] Step 2: Field acoustic wave data is affected by instrumentation, wellbore collapse, and formation heterogeneity. The waveform signal contains a significant amount of noise, and the extracted dispersion will contain some noise points far beyond the normal range. This noise will significantly affect data standardization and, in turn, the noise reduction effect. Therefore, only after limiting the range of field data can the density clustering algorithm be used for dispersion separation.

[0039] Step 3: Because the distance between data points is a key factor affecting the effectiveness of the DBSCAN algorithm, and frequency and velocity have different dimensions and ranges, this can cause the distance calculation to rely on velocity while relatively ignoring frequency, which in turn affects the noise reduction effect. Furthermore, the distance distribution between the original data points is relatively large, which makes the clustering results insensitive to changes in the ε neighborhood parameter. Therefore, before clustering the data, it is necessary to normalize the data to eliminate dimensionality effects and improve the algorithm's performance.

[0040] Step 4. Clustering algorithm belongs to unsupervised learning algorithm. It is a data analysis method that divides the data set into different subsets according to specific standards, and calls each subset a "cluster", so that the similarity of data in the same cluster and the difference between data in different clusters are as large as possible. For clusters with irregular shapes, clustering algorithms such as k-means that assume that clusters are convex cluster structures are no longer applicable. Density clustering algorithms regard clusters as high-density areas separated by low-density areas. They can divide clusters of arbitrary shapes and do not need to set the number of clusters a priori. It has great advantages in processing data sets with irregular clusters. DBSCAN (Density-Based Spatial Clustering of Application with Noise) is a classic density clustering method. Based on the ε neighborhood and the minimum number of data points MinPts contained in the ε neighborhood, the algorithm divides the relationship between data points into core objects, density direct, density reachable and density connected. For a given data set D = {x1, x2, x3, ..., x m},

[0041] (1) If ||x j -x i ||2≤ε, then x j Located at x i within the ε neighborhood of .

[0042] (2) If x i The ε neighborhood of x contains at least MinPts data points, then x is called i For the core object.

[0043] (3) If x j Located in the core object x i In the ε neighborhood of j From x i directly density-reachable.

[0044] (4) If there is a sample sequence (p1, p2, p3, ..., p n ), and p i+1 It is from p i (i=1,2,3,…n-1) directly density accessible, then it is called p n is density-reachable by p1. Obviously, the sample sequence (p1, p2, p3, ..., p n ) only p n Can be a non-core object.

[0045] (5) If there is a core object x i , so that O1 and O2 are both represented by xi If the density is reachable, then O1 and O2 are said to be density-connected.

[0046] The DBSCAN algorithm is described as follows: first, all core objects in the dataset are determined based on the ε neighborhood specified by a set of parameters (ε, MinPts), and then any core object x is selected. i Find all the i Density-reachable objects, the collection of these objects is a cluster, and finally repeat the previous step for the unclassified core objects until all core objects are classified.

[0047] Figure 3(c) shows the effect of using the DBSCAN algorithm to denoise the synthetic dispersion data in Figure 3(b). It can be seen that the entire data set consists of 11 clusters represented by different symbols, as well as noise points represented by “×” that do not belong to any cluster.

[0048] Step 5: Remove the noise data and the cluster starting at 0Hz from the dataset. The remaining cluster with the most data points is called the main cluster of the dataset. Denormalize the clustered data after noise removal. It is important to keep the parameters of the denormalization process consistent with the parameters of the normalization process.

[0049] Step 6: Judge the data quality by the number of data points contained in the main cluster of the dataset. We assume that the data quality of the dataset corresponding to the main cluster consisting of less than 100 points is poor, but this standard needs to be adjusted according to the specific situation of the dataset. For data with good data quality, the main cluster of the dataset is the target dispersion. For data with poor data quality, we use the monotonicity of the dispersion curve to select appropriate clusters from the clustering results. These clusters constitute the target dispersion.

[0050] Based on the same inventive concept, an embodiment of the present invention further provides a density-clustered borehole pattern wave effective dispersion processing device, as described in the following embodiment. Since the principle of the device to solve the problem is similar to the density-clustered borehole pattern wave effective dispersion processing method, the implementation of the device can refer to the implementation of the density-clustered borehole pattern wave effective dispersion processing method, and the repeated parts will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, implementation in hardware, or a combination of software and hardware, is also possible and conceived. Please refer to Figure 2 , is a composition structure diagram of the density clustered borehole mode wave effective dispersion extraction device of the embodiment of the present application. The device may include: a slowness-frequency domain transformation module, a data clustering processing module, and a target main cluster identification module. The structure is described in detail below.

[0051] The slowness-frequency domain transformation module transforms the waveform in the space-time domain into the slowness-frequency domain and uses the dispersion extraction algorithm to extract the dispersion information of the sound wave;

[0052] The data clustering processing module performs dispersion normalization processing and uses the DBSCAN algorithm to draw different clusters;

[0053] The target main cluster identification module sets the main cluster of the data set as the target dispersion for data with good data quality.

[0054] The following further illustrates the application effect of the density-clustered wellbore pattern wave effective dispersion extraction method of the present invention in combination with specific well logging example processing results.

[0055] Figures 4(a), (b) and 5(a), (b) further demonstrate the dispersion separation effect of the density clustering algorithm on the monopole and dipole data at depth X765.881 m in a certain well: the monopole data quality is good, and no additional adjustment of the density clustering algorithm is required to achieve the separation of Stoneley wave dispersion, as shown in Figures 3(a), (b), (c), and (d); the dipole data quality is poor, and the flexural wave dispersion shows an abnormal jump near 3 kHz, which causes the density clustering algorithm to divide it into several different clusters. Based on the monotonicity of the flexural wave dispersion, we select the cluster whose average velocity decreases with increasing frequency to separate the flexural wave dispersion of this poor data quality, as shown in Figures 5(a) and (b).

[0056] In Figure 3(a), the horizontal axis is time (ms) and the vertical axis is source distance (m). In Figures 3(b), (c), (d), Figures 4(a), (b), and Figures 5(a), (b), the horizontal axis is frequency (kHz) and the vertical axis is speed (km / s).

[0057] The above examples are merely preferred examples for fully illustrating the method of the present invention, and the scope of protection of the present invention is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on the present invention are all within the scope of protection of the present invention. The scope of protection of the present invention shall be subject to the claims.

Claims

1. A method for picking effective dispersion of borehole mode waves based on density clustering, comprising the following steps: Step 1: extracting the dispersion information of the acoustic wave from the waveform information recorded by the acoustic logging instrument through a dispersion extraction algorithm, including the Prony method, the Matrix Pencil method, or the weighted spectral coherence method; Step 2: Limit the range of the original dispersion data obtained in step 1; Step 3: standardize the data within the limited range obtained in step 2; Step 4: Use the DBSCAN algorithm to cluster the data obtained in step 3 and divide the data set into different clusters; Step 5: Denormalize the data obtained in step 4; Step 6: The data quality is judged by the number of data points contained in the main cluster of the dataset. By default, a dataset with a main cluster consisting of less than 100 points corresponds to a dataset with poor data quality, but this standard needs to be adjusted according to the specific data set. For data with good data quality, the main cluster of the dataset is the target dispersion. For data with poor data quality, we use the monotonicity of the dispersion curve to select appropriate clusters from the clustering results. These clusters constitute the target dispersion. The standardization processing algorithm in step 3 is: X stand =(X-μ) / σ (1) Where, X stand is the standardized data, X is the original data, μ and σ are the mean (mean) and standard deviation (std) of the sample data respectively; The DBSCAN algorithm in step 4 is based on the ε neighborhood and the minimum number of data points MinPts contained in the ε neighborhood, and divides the relationship between data points into core objects, density direct, density reachable and density connected. For a given data set D = {x1, x2, x3, ..., x m }, (1) If ||x j -x i ||2≤ε, then x j Located at x i In the ε neighborhood of ; (2) If x i The ε neighborhood of x contains at least MinPts data points, then x is called i As the core object; (3) If x j Located in the core object x i In the ε neighborhood of j From x i Direct density accessible; (4) If there is a sample sequence (p1, p2, p3, ..., p n ), and p i+1 It is from p i (i=1,2,3,…n-1) directly density accessible, then it is called p n is density-reachable by p1; obviously, the sample sequence (p1, p2, p3, ..., p n ) only p n Can be a non-core object; (5) If there is a core object x i , so that O1 and O2 are both represented by x i If the density is reachable, then O1 and O2 are said to be density connected; The DBSCAN algorithm is described as follows: first, all core objects in the dataset are determined based on the ε neighborhood specified by a set of parameters (ε, MinPts), and then any core object x is selected. i Find all the i Density-reachable objects, the collection of these objects is a cluster, and finally repeat the previous step for the unclassified core objects until all core objects are classified.

2. A method for picking up effective frequency dispersion of borehole mode waves based on density clustering according to claim 1, and a device for processing effective frequency dispersion of borehole mode waves based on density clustering to implement the above method, characterized in that: include: The slowness-frequency domain transformation module transforms the waveform in the space-time domain into the slowness-frequency domain and uses the dispersion extraction algorithm to extract the dispersion information of the sound wave; The data clustering processing module performs dispersion normalization processing and uses the DBSCAN algorithm to draw different clusters; The target main cluster identification module sets the main cluster of the data set as the target dispersion for data with good data quality.

Images