Processing method of two-dimensional nuclear magnetic resonance logging data

Abstract

The invention provides a processing method of two-dimensional nuclear magnetic resonance logging data, and belongs to the technical field of oil-gas exploration. In order to solve the problem of two-dimensional nuclear magnetic resonance fluid component evaluation, a spectrum peak automatic division method is adopted to process two-dimensional nuclear magnetic resonance logging data, firstly, a spectrum peak signal boundary of each component in a two-dimensional spectrum is determined by adopting an amplitude reduction clustering algorithm, and overlapped fluid components are further divided by utilizing a Gaussian mixture model clustering algorithm; according to the method, the fluid boundary division precision is improved, so that a fluid identification chart of each depth point is formed, the fluid type is determined according to the distribution position of main spectral peaks, the fluid volume of each component is calculated, and identification and content calculation of various fluids in a reservoir are realized. The technical bottleneck of two-dimensional nuclear magnetic resonance fluid component automatic division and quantitative evaluation is overcome, and the method has remarkable application value and popularization prospect in unconventional oil and gas reservoir exploration.

Description

Technical Field

[0001] This invention belongs to the field of oil and gas exploration technology and relates to a method for processing two-dimensional nuclear magnetic resonance logging data. Background Technology

[0002] Nuclear magnetic resonance (NMR) logging is an important geophysical logging technique. Due to the limitations of one-dimensional NMR logging in fluid identification, two-dimensional (2D) NMR logging has been developed. 2D NMR logging includes two observation modes: T2-T1 and T2-D. Compared to one-dimensional T2 NMR logging, 2D NMR logging can provide more information and more effectively identify different types of fluids, playing a crucial role in oilfield exploration and development.

[0003] Currently, there are three main methods for fluid identification in two-dimensional nuclear magnetic resonance logging:

[0004] (1) Constructing a fluid identification chart based on nuclear magnetic resonance experiments. This method utilizes core experimental techniques to measure the two-dimensional nuclear magnetic resonance response of cores under different conditions, including saturated state, after centrifugation, and after drying, to obtain the two-dimensional spectral characteristics of components such as bound water, bound oil, movable water, and movable oil, thus forming a fluid identification chart. This chart is then applied to the two-dimensional spectrum of two-dimensional nuclear magnetic resonance logging to calculate the volume content of various fluid components.

[0005] (2) Fluid identification methods based on clustering algorithms. This type of method mainly uses the Gaussian mixture model (GMM) algorithm to fit a Gaussian mixture model to a single two-dimensional spectrum or a deep superimposed two-dimensional spectrum to obtain the Gaussian distribution of each fluid component. By clustering, a fluid identification map is formed, and then the fluid component type is identified and the volume content of each fluid component is calculated.

[0006] (3) Fluid identification method based on blind source separation. The number of fluid components is obtained by using methods such as Akaike Information Criterion (AIC) and Principal Component Analysis (PCA). The two-dimensional spectral characteristics and mixing coefficients of different fluid components are obtained by using the blind source separation algorithm. Based on the two-dimensional spectral characteristics of different fluid components, the specific fluid type is determined and the volume content of different fluid components is calculated.

[0007] The disadvantages of the existing technology are: (1) In technology 1, nuclear magnetic resonance experimental instruments are mostly high-frequency, while nuclear magnetic resonance logging instruments are low-frequency. In addition, the temperature, pressure and other conditions in the experiment are quite different from those downhole. Therefore, when the fluid identification chart constructed by the nuclear magnetic resonance experiment is applied to the evaluation of nuclear magnetic resonance logging fluid, the application effect is poor. Moreover, the fluid identification chart constructed for the core of one study area may not be applicable to other study areas. Experiments need to be carried out. Therefore, the nuclear magnetic resonance experimental method is limited.

[0008] (2) In Technique 2, the fluid identification method based on clustering algorithm is difficult to determine the number of fluid components at each depth point during the two-dimensional nuclear magnetic resonance logging data processing. Processing the superimposed spectrum usually ignores components with smaller signals, and the fluid volume calculated by this method has low accuracy.

[0009] (3) In Technique 3, the fluid identification method based on blind source separation requires the number of fluid components to be determined in advance by other methods. The number of components is fixed. When the actual number of formation fluid components is greater than the set number of components, the calculation results are inaccurate. Moreover, it is also difficult to accurately separate two fluid components when their volume content is linearly correlated at different depth points.

[0010] In current unconventional oil and gas reservoirs such as shale oil, the reservoir porosity is low, the fluid composition is complex, the quality of two-dimensional nuclear magnetic resonance (NMR) logging signals is poor, the resolution of the obtained two-dimensional spectrum is low, and different fluid components overlap, making it difficult to determine fluid boundaries, separating components is challenging, and calculating the content of each component is inaccurate. Therefore, there is an urgent need to develop a two-dimensional NMR logging fluid identification and quantitative evaluation method that can accurately delineate fluid components. Summary of the Invention

[0011] The purpose of this invention is to address the aforementioned problems in existing technologies by proposing a method for processing two-dimensional nuclear magnetic resonance logging data.

[0012] The overall concept of this invention is as follows: In a two-dimensional nuclear magnetic resonance (NMR) spectrum, the responses of each fluid component are peak-shaped signals. Different fluid components are distributed at different locations in the two-dimensional spectrum. Based on the positional characteristics of different component signals in the two-dimensional spectrum, different fluids such as clay-bound water, capillary-bound water, and mobile oil can be identified. By accumulating the signals of each fluid, the fluid volume can be quantitatively calculated. Due to the influence of noise, the signals of different fluid components may partially or even completely overlap. Calculating the fluid volume requires first determining the distribution range of each fluid component, i.e., the boundary delineation of different spectral peak signals. Based on the characteristics of the two-dimensional NMR spectrum, this invention proposes and develops an amplitude descent clustering (ADC) algorithm to delineate the distribution boundaries of non-overlapping and partially overlapping fluid component signals. Then, a Gaussian mixture model (GMMC) clustering algorithm is used to further delineate the overlapping signals. Finally, the fluid type is determined using the automatically delineated distribution locations of each component, and the volume of each component is quantitatively calculated using the distribution range of the delineated fluid components.

[0013] The objective of this invention can be achieved through the following technical solution: a method for processing two-dimensional nuclear magnetic resonance logging data, specifically including the following steps:

[0014] S1 Data Input: Input two-dimensional nuclear magnetic resonance logging data S and the number of depth samples N.d The noise cutoff parameter τ; the two-dimensional nuclear magnetic resonance logging data S includes two-dimensional spectral data from multiple sampling points at different depths;

[0015] S2 Automatic Peak Division Method Processing: According to the depth index order, the two-dimensional spectral data of each depth sampling point is processed using the automatic peak division method to divide the distribution range of the spectral peak signals of each fluid and obtain the fluid identification map of each depth sampling point;

[0016] In the automatic peak segmentation method, the two-dimensional spectral data processing process for any depth sampling point is as follows: First, the amplitude descent clustering algorithm is used to initially segment the fluid signal boundary of the two-dimensional spectral data to obtain an initial fluid identification map; the initial fluid identification map is then detected, and when a fluid in the initial fluid identification map is detected to meet the overlapping fluid condition, the Gaussian mixture model clustering algorithm is used to further subdivide the fluids that meet the overlapping fluid condition.

[0017] S3 fluid type identification and quantitative calculation: Based on the fluid identification chart at each depth sampling point, the fluid type is determined according to the distribution position of the spectral peaks of each fluid component in the fluid identification chart, and the fluid volume is quantitatively calculated.

[0018] The specific process of using the amplitude descent clustering algorithm to perform preliminary boundary delineation of fluid signal data in two-dimensional spectral data is as follows: A1 Signal truncation processing: Based on the noise cutoff parameter τ, the two-dimensional spectral data is truncated to obtain the truncated two-dimensional spectral matrix F. t ;

[0019] A2 Pre-processing: Peak point extraction: For F t Peak point extraction is performed to generate a peak point coordinate matrix P with n1 peak points. c Peak points form independent clusters: Let C be a cluster, and let be a set of coordinate points. i For the i-th cluster, i = 1, 2, ..., n1, under the initial conditions, C i Includes the coordinates P of the i-th peak point ci Sorting: F t All non-zero signal points are sorted in descending order of signal amplitude value to generate an ordered coordinate matrix P. s ;

[0020] A3 clustering assignment: Traverse P in descending order s Each coordinate point in the data is assigned to the cluster closest to it based on the principle of proximity.

[0021] Preferably, when P s A coordinate point P in sj When two or more clusters are equidistant, the coordinate point P is... sjIt is assigned to the cluster with the largest peak signal amplitude.

[0022] The specific process of signal truncation in step A1 is as follows: The signal amplitude values ​​of signal points in the two-dimensional spectral data smaller than τ×fmax are set to zero, resulting in the truncated two-dimensional spectral matrix F. t fmax is the maximum signal amplitude among all signal points.

[0023] Preferably, the value of τ is no greater than 0.01.

[0024] If the ratio of the major axis to the minor axis of a fluid in the initial fluid identification chart is not less than a preset threshold, the fluid is identified as an overlapping component fluid. During the subdivision process, the peak points of the overlapping component fluid are replaced with a fitted Gaussian mixture model.

[0025] Preferably, the fluid types that can be identified include clay-bound water, capillary-bound water, mobile water, bituminous substances, bound oil, and mobile oil.

[0026] This invention addresses the field of fluid identification and quantitative evaluation in two-dimensional nuclear magnetic resonance (NMR) logging. It proposes a method for processing two-dimensional NMR logging data. By introducing the amplitude descent clustering (ADC) algorithm and the Gaussian mixture model clustering (GMMC) algorithm, this invention overcomes the shortcomings of existing technologies and significantly improves the accuracy, robustness, and applicability of fluid identification. Specific beneficial effects are as follows:

[0027] 1. Effectively solves the problem of poor versatility of experimental charts, achieving universal applicability across research areas: Existing technologies (based on nuclear magnetic resonance experiments to construct fluid identification charts) are limited by the differences in experimental and logging instrument conditions (such as frequency, temperature, and pressure), resulting in poor chart application effects and requiring repeated experiments for different research areas, leading to low efficiency. This invention eliminates the reliance on external experimental charts, directly performing automatic peak segmentation based on the logging data itself. It adaptively determines fluid component boundaries through an ADC algorithm and identifies fluid types by combining actual peak positions, achieving universal processing without core experiments. This significantly reduces operating costs and ensures the applicability of the method in different geological regions (such as conventional and unconventional reservoirs).

[0028] 2. Significantly Improves Fluid Component Identification Accuracy and Volume Calculation Accuracy: Existing technologies (fluid identification methods based on clustering algorithms) suffer from difficulties in determining the number of fluid components through depth-point processing, neglecting small signal components in superimposed spectra, and low volume accuracy. This invention uses an ADC algorithm to traverse all signal points in descending amplitude order, combines peak point initialization with dynamic allocation of boundary points, ensuring that all components (including smaller signal components) at each depth point are captured. Subsequently, the GMMC algorithm is used to further subdivide overlapping components, solving the boundary ambiguity problem caused by partial or complete signal overlap.

[0029] 3. Adaptive handling of fluid component number variations and linear correlation issues, enhancing algorithm robustness: Existing technologies (fluid identification methods based on blind source separation) require pre-fixing the number of components. When the actual number of components exceeds the preset limit or the fluid volume is linearly correlated, the separation results are inaccurate. This invention automatically initializes clusters (based on the number of peak points) using an ADC algorithm and dynamically expands the boundaries, eliminating the need for pre-setting the number of components. Furthermore, for multiple clusters of equidistant points, a strategy of "prioritizing the cluster with the largest peak amplitude" is adopted, avoiding errors caused by a fixed number of components. When processing fluids with linearly correlated volumes (such as bound oil and movable oil), the GMMC algorithm uses a Gaussian distribution model for fitting and subdivision, effectively separating overlapping signals. This solves the separation problem in existing technologies.

[0030] In summary, this invention, through an innovative method for processing two-dimensional nuclear magnetic resonance (NMR) logging data, addresses the challenges of complex fluid composition and severe signal overlap in unconventional reservoirs such as shale oil. By synergistically applying the amplitude descent clustering (ADC) algorithm and the Gaussian mixture model clustering (GMMC) algorithm, it achieves fully automated boundary delineation of two-dimensional NMR spectral peaks. Simultaneously, it effectively solves the core defects of existing technologies (such as strong experimental dependence, inaccurate component identification, and fixed group number limitations), realizing fully automated, high-precision fluid identification and quantitative evaluation. This provides reliable technical support for oil and gas exploration and development, and has significant application value and promising prospects in unconventional oil and gas reservoir exploration. Attached Figure Description

[0031] Figure 1 This is a diagram showing the results of automatic peak division in the T2-T1 spectrum of two-dimensional nuclear magnetic resonance.

[0032] Figure 2 The T2-T1 two-dimensional spectral model diagrams for six fluids are shown.

[0033] Figure 3 This is a diagram showing the results of a two-dimensional nuclear magnetic resonance logging numerical simulation of a 50m section containing six fluid components.

[0034] Figure 4 This is a diagram showing the automatic peak division results of two-dimensional nuclear magnetic resonance synthetic logging data (SNR=20).

[0035] Figure 5 This is a comparison chart of the calculated fluid component volume with the preset model.

[0036] Figure 6 This is a diagram showing the processing results for well A1.

[0037] Figure 7 Figure 3550–3600 m shows the quantitative evaluation results of fluid composition from two-dimensional nuclear magnetic resonance logging in well A1.

[0038] Figure 8For result verification - movable oil volume and multi-temperature pyrolysis experiment S1+S 2-1 Relationship diagram. Detailed Implementation

[0039] The following are specific embodiments of the present invention, which are described in conjunction with the accompanying drawings. However, the present invention is not limited to these embodiments.

[0040] Example 1

[0041] A method for processing two-dimensional nuclear magnetic resonance logging data, specifically including the following steps:

[0042] S1 Data Input: Input T2-T1 2D NMR logging data S, with N depth samples. d The noise cutoff parameter τ; the data S includes two-dimensional spectral data from multiple sampling points at different depths;

[0043] S2 Automatic Peak Division Method Processing: According to the index order, the two-dimensional spectral data of each depth sampling point is processed using the automatic peak division method to divide the distribution range of the spectral peak signals of each fluid and obtain the fluid identification map of each depth sampling point;

[0044] In the automatic peak segmentation method, the processing procedure for the two-dimensional spectral data of any depth sampling point is as follows: Amplitude descent clustering algorithm: First, the amplitude descent clustering algorithm is used to initially segment the fluid signal boundary, obtaining the initial fluid identification map for each depth sampling point.

[0045] The specific process is as follows: A1 signal truncation processing: To reduce the influence of minimum signals on the results, let τ be 0.01 or 0.0001, assign zero to the values ​​in the two-dimensional spectral data that are less than τ×fmax, and obtain the two-dimensional spectral matrix F after truncation threshold. t fmax is the maximum signal amplitude among all signal points;

[0046] A2 allocation preprocessing: Obtaining F t The coordinate matrix P corresponding to the peak point in c The size is n1×2, where n1 is the number of peak points and 2 represents the two dimensions T2 and T1;

[0047] Extract the non-zero signal points from the two-dimensional spectral matrix Ft, arrange them in descending order of signal amplitude, and obtain the corresponding coordinate matrix P. s The size is n²×2, where n² is F. t The number of non-zero signal points in the middle, where 2 represents the two dimensions T2 and T1;

[0048] Let C be a cluster, and let be the set of coordinate points. i For the i-th cluster, i = 1, 2, ..., n1, under the initial conditions, C iIncludes the coordinates P of the i-th peak point ci ;

[0049] Let the distance between any two coordinate points P1 and P2 be:

[0050]

[0051] Then P s The j-th coordinate point and the i-th cluster C i The distance between them is:

[0052]

[0053] Traverse P in descending order s For each coordinate point in the array, calculate the distance between it and each cluster, and assign the current coordinate point to the cluster with the smallest distance.

[0054] When P s When a coordinate point in the signal is equidistant from two or more clusters, the coordinate point is assigned to the cluster with the largest peak signal amplitude.

[0055] Gaussian Mixture Model Clustering: For each depth sampling point, the initial fluid identification map is detected. When a fluid in the initial fluid identification map is found to meet the overlapping fluid condition, the Gaussian Mixture Model clustering algorithm is used to subdivide the fluids that meet the overlapping fluid condition. Overlapping fluid condition: When the ratio of the major axis to the minor axis of a fluid in the initial fluid identification map is not less than a preset threshold, the fluid is determined to be an overlapping component fluid. During the subdivision process, the peak points of the overlapping component fluids are replaced with the fitted Gaussian Mixture Model.

[0056] S3 fluid type identification and quantitative calculation: Based on the fluid identification chart at each depth sampling point and the spectral peak distribution position of each fluid, the specific type of each fluid is determined, and the volume of each fluid is quantitatively calculated.

[0057] Example 2

[0058] like Figure 1 As shown, in Figure 1 In the diagram, (a) represents the T2-T1 two-dimensional spectral data; (b) shows the peak boundaries delineated using the amplitude descent clustering algorithm; and (c) shows the subdivision using the Gaussian mixture model clustering algorithm. The amplitude descent clustering algorithm was used to process (a), dividing it into four fluid components (defined by the white solid circle) based on the number of peak points. The processing result is shown in (b). In (b), component 1 has a large major axis and minor axis ratio of approximately 1.82, satisfying the overlapping fluid condition. For this case, the Gaussian mixture model clustering (GMMC) algorithm was used to further process component 1, dividing it into two components and replacing the peak points of component 1 with the fitted Gaussian mixture model. The processing result is shown in (c).

[0059] Example 3

[0060] Model Validation: To verify the accuracy of the processing method for two-dimensional nuclear magnetic resonance logging data, T2-T1 two-dimensional spectra for six different fluid types were set up, such as... Figure 2 As shown, (a): clay-bound water; (b): capillary-bound water; (c): mobile water; (d): asphaltenes; (e): bound oil; (f): mobile oil; their peak T2 values ​​are 1 ms, 10 ms, 100 ms, 1 ms, 15 ms, and 150 ms, respectively; their peak T1 values ​​are 1 ms, 10 ms, 100 ms, 25 ms, 75 ms, and 750 ms, respectively; the volume of each component is 3 pu.

[0061] Two-dimensional nuclear magnetic resonance (NMR) logging data was constructed for a 50m well section (T2-T1) containing the six fluid components. The sampling interval was 0.1m, and the total fluid volume at each depth point was 18 pu. The saturation (volume weight) of each fluid component was randomly generated. Using the generated 2D NMR logging data, forward modeling was performed to generate echo data. Gaussian white noise was added to generate echo data with a signal-to-noise ratio (SNR) of 20. The BRD (Butler-Reeds-Dawson) algorithm was then used to invert and obtain the noisy 2D NMR logging data. Figure 3 As shown, the data of the constructed two-dimensional nuclear magnetic resonance logging model are as follows: the first channel is the depth channel, the second channel is the model data without noise, the third to eighth channels are the data of six fluid components: clay-bound water, capillary-bound water, movable water, asphaltene, bound oil and movable oil, respectively, and the ninth channel is the two-dimensional nuclear magnetic resonance model data including noise.

[0062] The proposed SPAS (Spectral Peak Segmentation Method) is used to process the two-dimensional spectral data at each depth point, automatically segmenting the spectral peak signals of each fluid component to form a fluid identification chart. Based on this chart, the fluid type is identified and the volume of each component is calculated. Figure 4 (a) is the two-dimensional spectrum of the 200th depth point in the noise model. Figure 4 (b) is a fluid identification chart at this depth point, which divides the fluid into 7 types. The positions of their spectral peaks are shown in Table 1. The fluid type is identified based on the spectral peaks T1, T2 and T1 / T2 values. Component 1 is asphaltene, component 2 is mobile water, component 3 is clay-bound water, component 4 is mobile oil, components 5 and 7 are bound oil, and component 6 is capillary-bound water. Figure 4(c) shows the distribution locations of the main spectral peaks in the noise model. These peaks can be divided according to the range of T2 values ​​and the T1 / T2 values ​​to determine the different fluid component types at each depth point. T2 < 5ms, T1 / T2 < 5 indicates clay-bound water; T2 < 5ms, T1 / T2 ≥ 5 indicates asphaltene; 5ms ≤ T2 < 50ms, T1 / T2 < 2 indicates capillary-bound water; 5ms ≤ T2 < 50ms, T1 / T2 ≥ 2 indicates bound oil; T2 ≥ 50ms, T1 / T2 < 2 indicates mobile water; T2 ≥ 50ms, T1 / T2 ≥ 2 indicates mobile oil.

[0063]

[0064] Table 1

[0065] A fluid identification map is automatically constructed based on the spectral peaks at each depth point. The fluid type is identified according to the peak distribution location, and the volume content of each fluid component is calculated. The calculated fluid volume is then compared with the model, such as... Figure 5 As shown, (a): clay-bound water; (b): capillary-bound water; (c): mobile water; (d): asphaltene; (e): capillary-bound oil; (f): mobile oil. The volumes of the six fluid components are consistent with the model, with an average relative error of about 25%. The small average relative error indicates that the formed two-dimensional nuclear magnetic resonance spectrum peak automatic division and fluid quantitative evaluation method is relatively accurate and effective.

[0066] Example 4

[0067] The two-dimensional nuclear magnetic resonance logging data processing method of the present invention was applied to the shale oil reservoir of well A1 in a basin in southern China.

[0068] Shale oil is petroleum found in sandstone or carbonate rocks within mudstone and shale formations; shale oil reservoirs are unconventional reservoirs. Well A1 primarily consists of silty mudstone and shale. Two-dimensional nuclear magnetic resonance (NMR) logging data for this well was obtained using a Schlumberger NMR logging instrument in CMR-NG mode, with a measurement depth of 3462–3613 m and a sampling interval of 0.1905 m. The well exhibited a low signal-to-noise ratio in its two-dimensional NMR signal, with overlapping fluid components in the inverted two-dimensional spectrum, making it difficult to determine the boundaries between different fluids, resulting in poor fluid identification accuracy and inaccurate volume calculations. Therefore, this invention was used to process and evaluate this well.

[0069] The well was processed using the method of this invention, and the proposed automatic peak segmentation method was used to perform fluid identification and quantitative calculation of fluid volume on the two-dimensional nuclear magnetic resonance logging data. Figure 6 (a) and Figure 6 (b) Fluid identification results at depths of 3551.37m and 3560.70m, respectively, identified 5 and 6 fluid components. The main peak distribution locations of the fluid identification results in this well are shown below. Figure 6 As shown in (c). Experimental results indicate that the T2 cutoff value between clay-bound water and capillary-bound water is 3 ms, and the T2 cutoff value between capillary-bound water and mobile water is 10 ms. Based on the peak positions, the fluid types are determined as follows: the component with peak T2 < 3 ms and T1 / T2 < 5 is clay-bound water; the component with peak T2 < 3 ms and T1 / T2 ≥ 5 is asphaltene; the component with peak 3 ms ≤ T2 < 10 ms and T1 / T2 < 2 is capillary-bound water; the component with peak 3 ms ≤ T2 < 10 ms and T1 / T2 ≥ 2 is bound oil; the component with peak T2 ≥ 5 ms and T1 / T2 < 2 is mobile water; and the component with peak T2 ≥ 5 ms and T1 / T2 ≥ 2 is mobile oil. Based on the fluid identification results and the determined fluid types, the fluid volumes of the six components—capillary-bound water, mobile water, asphaltene, bound oil, and mobile oil—are calculated. The treatment results for the 3550-3600m section of the well are as follows: Figure 7 As shown, the first channel is the natural gamma; the second channel is the depth channel; the third and fourth channels are the T2 and T1 distributions, respectively; the fifth channel is the T2-T1 spectrum; the sixth channel is the fluid profile; the seventh channel is the NMR porosity, including total porosity and effective porosity; the eighth and ninth channels are the total water volume and total oil volume, respectively.

[0070] To verify the accuracy and effectiveness of this method in processing two-dimensional nuclear magnetic resonance logging data, the calculated oil-bearing volume (the sum of bound oil and movable oil volumes) was compared with the oil-bearing portion (S) from the multi-temperature pyrolysis experiment. 1-1 S 1-2 and S 2-1 The sum of these values ​​is compared. Since the fluid volume obtained from nuclear magnetic resonance (NMR) experiments and the results from multi-temperature pyrolysis experiments are not the same physical quantity, a cross-plot of the calculated oil-bearing volume and the oil-bearing portion from the multi-temperature pyrolysis experiments is plotted, as shown below. Figure 8 As shown, the fitting formula is y = 0.0017·x + 0.0131, R0 2 =0.91, indicating a linear positive correlation between the calculated oil-bearing volume and the oil-bearing portion from multi-temperature pyrolysis, with a good correlation, demonstrating the accuracy and effectiveness of the fluid component volume calculated by this method.

[0071] The key point of this invention is to address the challenge of evaluating fluid components in two-dimensional nuclear magnetic resonance (NMR) spectra. It proposes an amplitude descent clustering algorithm to determine the peak signal boundaries of each component in the two-dimensional spectrum, and uses a Gaussian mixture model clustering algorithm to further divide completely overlapping fluid components, thereby forming a fluid identification map for each depth point. Based on the distribution location of the main spectral peaks, the fluid type is determined, and the fluid volume of each component is calculated. An amplitude descent clustering algorithm is used to delineate the spectral peak signal boundaries of fluid components in a two-dimensional spectrum. This amplitude descent clustering algorithm is an innovative algorithm proposed for the response characteristics of fluid components in two-dimensional nuclear magnetic resonance. To address the problem of multiple components completely overlapping after amplitude descent clustering, a Gaussian mixture model clustering algorithm is used for further separation, improving the accuracy of fluid boundary delineation. The algorithm processes the two-dimensional spectrum at each depth point to generate a fluid boundary delineation map. Fluid identification is performed based on the spectral peak positions, and the volume of all fluid components is calculated. This invention is tested and applied using Schlumberger CMR-NG mode T2-T1 two-dimensional nuclear magnetic resonance logging data as an example. It is also applicable to T2-D two-dimensional nuclear magnetic resonance logging acquisition mode and other two-dimensional nuclear magnetic resonance logging instruments.

[0072] The above specific embodiments further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for processing two-dimensional nuclear magnetic resonance logging data, characterized in that, Specifically, the following steps are included: S1 Data Input: Input two-dimensional nuclear magnetic resonance logging data S and the number of depth samples N. d The noise cutoff parameter τ; the two-dimensional nuclear magnetic resonance logging data S includes two-dimensional spectral data from multiple sampling points at different depths; S2 Automatic Peak Division Method Processing: According to the depth index order, the two-dimensional spectral data of each depth sampling point is processed using the automatic peak division method to divide the distribution range of the spectral peak signals of each fluid and obtain the fluid identification map of each depth sampling point; In the automatic peak segmentation method, the two-dimensional spectral data processing process for any depth sampling point is as follows: First, the amplitude descent clustering algorithm is used to initially segment the fluid signal boundary of the two-dimensional spectral data to obtain an initial fluid identification map; the initial fluid identification map is then detected, and when a fluid in the initial fluid identification map is detected to meet the overlapping fluid condition, the Gaussian mixture model clustering algorithm is used to further subdivide the fluids that meet the overlapping fluid condition. S3 fluid type identification and quantitative calculation: Based on the fluid identification map of each depth sampling point, the specific type of each fluid is determined and the volume of each fluid is quantitatively calculated.

2. The method for processing two-dimensional nuclear magnetic resonance logging data as described in claim 1, characterized in that, The specific process of using the amplitude descent clustering algorithm to perform preliminary boundary delineation of fluid signal data in two-dimensional spectral data is as follows: A1 Signal truncation processing: Based on the noise cutoff parameter τ, the two-dimensional spectral data is truncated to obtain the truncated two-dimensional spectral matrix F. t ; A2 Pre-processing: Peak point extraction: For F t Peak point extraction is performed to generate a peak point coordinate matrix P with n1 peak points. c ; Peak points form independent clusters: Let C be a cluster, and let be a set of coordinate points. i For the i-th cluster, i = 1, 2, ..., n1, under the initial conditions, C i Includes the coordinates P of the i-th peak point ci ; Sort: F t All non-zero signal points are sorted in descending order of signal amplitude value to generate an ordered coordinate matrix P. s ; A3 clustering assignment: Traverse P in descending order s Each coordinate point in the data is assigned to the cluster closest to it based on the principle of proximity.

3. The method for processing two-dimensional nuclear magnetic resonance logging data as described in claim 2, characterized in that, When P s A coordinate point P in sj When two or more clusters are equidistant, the coordinate point P is... sj It is assigned to the cluster with the largest peak signal amplitude.

4. The method for processing two-dimensional nuclear magnetic resonance logging data as described in claim 2, characterized in that, The specific process of signal truncation in step A1 is as follows: The signal amplitude values ​​of signal points in the two-dimensional spectral data smaller than τ×fmax are set to zero, resulting in the truncated two-dimensional spectral matrix F. t fmax is the maximum signal amplitude among all signal points.

5. The method for processing two-dimensional nuclear magnetic resonance logging data as described in claim 1, characterized in that, The value of τ is no greater than 0.

01.

6. The method for processing two-dimensional nuclear magnetic resonance logging data as described in claim 1, characterized in that, Overlapping fluid condition: When the ratio of the major axis to the minor axis of a certain fluid in the initial fluid identification chart is not less than a preset threshold, the fluid is determined to be an overlapping component fluid; during subdivision processing, the peak points of the overlapping component fluid are replaced with the fitted Gaussian mixture model.

7. The method for processing two-dimensional nuclear magnetic resonance logging data as described in claim 1, characterized in that, It can identify fluid types including clay-bound water, capillary-bound water, movable water, bituminous substances, bound oil, and movable oil.

Images