A method for characterizing the occurrence of the wetting phase fluid in a subterranean reservoir

Abstract

The application relates to a kind of underground reservoir wetting phase fluid hosting state characterization method, the method is targeted to adjust the acquisition means of core sample pore radius distribution data, and selects core nuclear magnetic resonance longitudinal relaxation spectrum T1 as the characterization index of pore structure;In the aspect of data analysis processing, the matching relationship of pore radius and T1 is established using the equal cumulative frequency matching method, the matching relationship is fitted using the self-adaptive lifting algorithm, the T1 spectrum of nuclear magnetic resonance logging is screened and projected using the method based on feature mask, and the pore radius distribution atlas of underground reservoir finally obtained can effectively and clearly present the pore structure characteristics of wetting phase fluid hosting, on the basis, the spectrum peak shape, radius interval of distribution atlas are used as the characterization basis of wetting phase fluid composition and flow state, and the accuracy of underground reservoir wetting phase fluid hosting state characterization is improved.

Description

Technical Field

[0001] This invention relates to the technical field of methods for identifying the state of wetted phase fluids in underground reservoirs, and specifically to a method for characterizing the occurrence state of wetted phase fluids in underground reservoirs. Background Technology

[0002] The formation of oil and gas reservoirs typically involves hydrocarbons entering the reservoir from the source rock, continuously displacing the original formation water, and eventually enriching into reservoirs within suitable traps. Since oil and gas are non-wetting fluids, capillary pressure cannot completely displace the original formation water from the pore space. Therefore, formation water will inevitably remain in the pore space. The state of this residual formation water—the wetting phase—is a crucial factor determining the development success of oil and gas reservoirs: high hydrocarbon saturation and the presence of original formation water only in small pores result in high relative permeability and good development outcomes; conversely, poor permeability leads to poor development. Therefore, accurately calculating the pore structure characteristics of underground reservoirs and characterizing the state of original formation water is essential for oil and gas reservoir exploration and development. Furthermore, due to overbalanced drilling, some mud filtrate (also a wetting phase fluid) will infiltrate the reservoir during the drilling process. This mud filtrate will occupy part or all of the pore space previously occupied by the non-wetting phase fluid, disrupting the original fluid distribution balance. To estimate the proportion of the infiltrated non-wetting phase fluid, characterizing the occurrence state of this wetting phase fluid—the mud filtrate—is equally important.

[0003] The characterization of the occurrence state of wetting phase fluids in underground reservoirs mainly involves two steps. First, the pore structure of the wetting phase fluids in underground reservoirs is calculated. Then, the distribution characteristics of the pore structure are analyzed to obtain the occurrence characteristics of wetting phase fluids such as formation water and mud filtrate.

[0004] The main steps of existing technologies for calculating the pore structure of fluids in underground reservoirs are as follows:

[0005] ① Use conventional mercury intrusion porosimetry, centrifugation, or semi-permeable diaphragm method to determine the core capillary radius distribution;

[0006] ② Nuclear magnetic resonance experiments were conducted on water-saturated cores to obtain the transverse relaxation time T2 spectrum of the cores;

[0007] ③ Establish the correspondence between capillary radius and T2 spectrum, and use one-segment / two-segment / three-segment fitting methods to establish the conversion relationship between T2 and capillary radius;

[0008] ④ Using this conversion relationship, the reservoir nuclear magnetic resonance logging T2 spectrum is converted into a pore radius distribution that varies continuously with depth, thus completing the characterization of the pore structure of the fluid in the underground reservoir.

[0009] The aforementioned technical solution establishes a bridge between well logging parameters (T2 NMR spectrum) and pore structure parameters (capillary radius) through core experiments, achieving the goal of continuously characterizing reservoir pore structure using well logging curves. However, existing methods have some technical issues that potentially affect the accuracy of characterization:

[0010] ① Core NMR T2 spectra characterize the pore radius distribution of rocks under natural conditions. However, in step 1 of existing technologies, the capillary radius distribution of cores is obtained using conventional mercury intrusion porosimetry, centrifugation, or semi-permeable septum methods. The capillary radius distribution obtained using these methods is the pore space volume distribution controlled by the throat during displacement, reflecting the dynamic characteristics of pore radius distribution. That is, the capillary radius distribution obtained by the above three methods is actually significantly controlled by the throat scale and is not the pore radius distribution under natural conditions of rocks, thus differing somewhat from the physical meaning of NMR T2 spectra.

[0011] The established transformation relationship may lead to bias in the characterization results;

[0012] ② In step 3 of the existing technology, the one-segment / two-segment / three-segment fitting method used to establish the conversion relationship between T2 and capillary radius requires too much human intervention to determine the specific segmentation mode and connection points used for different core samples, which affects the modeling efficiency. Moreover, the more segments there are, the more non-differentiable points there are in the fitted curve, which may cause numerical rationality problems such as curve discontinuity and jumps when the model is applied.

[0013] ③ In steps 2 and 3 of the existing technology, the T2 spectrum of nuclear magnetic resonance of 100% water-saturated core is used in the laboratory to establish a relationship with the core capillary radius. However, when this conversion relationship is applied to actual reservoirs, since the reservoir pores are not 100% water-saturated, it is necessary to remove the non-wetting phase signal in the reservoir T2 spectrum before conversion. However, due to the influence of diffusion relaxation effect, the T2 value range of non-wetting phase is likely to overlap with the T2 value range of wetting phase. If T2 is used as the conversion index, it is difficult to remove the non-wetting phase, making the pore radius distribution obtained by conversion easily affected by the non-wetting phase signal.

[0014] In addition, existing technologies mainly calculate the pore structure characteristics of underground reservoir fluids, which is only the first step in characterizing the occurrence state of underground reservoir wetting phase fluids. They have not developed a characterization technology for the occurrence state of formation water, mud filtrate, and other wetting phase fluids. Therefore, in terms of both calculation accuracy and characterization process, there is currently no method that can accurately characterize the occurrence state of underground reservoir wetting phase fluids. Summary of the Invention

[0015] The purpose of this invention is to propose a new method that provides high accuracy and can intuitively characterize the occurrence state of wetted phase fluids in underground reservoirs.

[0016] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0017] A method for characterizing the occurrence state of wetting phase fluid in underground reservoirs includes the following steps:

[0018] Step 1: Obtain the longitudinal relaxation spectrum T1 distribution of nuclear magnetic resonance of a core sample saturated with formation water, and obtain the pore radius distribution data of the same core sample; wherein, the pore radius distribution data is the volume frequency distribution data of the pore radius of a cast thin section or the pore radius distribution data of constant-rate mercury intrusion porosimetry.

[0019] Step 2: Use the equal cumulative frequency matching method to obtain the discrete matching relationship between the longitudinal relaxation spectrum T1 distribution and the pore radius distribution data in Step 1, and use the adaptive boosting algorithm to perform regression fitting on the discrete matching relationship;

[0020] Step 3: Obtain nuclear magnetic resonance logging data of the underground reservoir, then determine the boundary value of the feature mask of the light non-wetting phase signal, use the feature mask to filter out the light non-wetting phase signal in the nuclear magnetic resonance logging data, and then project the remaining wetting phase signal onto T1 to obtain the T1 relaxation spectrum of the underground reservoir nuclear magnetic resonance that contains only the wetting phase signal.

[0021] Step 4: Using the regression fitting curve relationship obtained in Step 2, i.e. the conversion relationship between longitudinal relaxation T1 and pore radius data, the T1 relaxation spectrum of the underground reservoir nuclear magnetic resonance obtained in Step 3, which only contains wetting phase signal, is converted to obtain the pore radius distribution map of the underground reservoir wetting phase fluid.

[0022] Step 5: Use the peak shape of the pore radius distribution spectrum obtained in Step 4 as the basis for the composition of the wetting phase fluid; use the radius range of the pore radius distribution spectrum obtained in Step 4 as the basis for judging the flow state of a certain wetting phase fluid, thereby obtaining the occurrence state of the wetting phase fluid in the underground reservoir.

[0023] This application characterizes the occurrence state of wetting phase fluids in underground reservoirs by obtaining the pore radius distribution. The acquisition methods for pore radius distribution data from core samples were specifically adjusted, and the longitudinal relaxation spectrum T1 of core nuclear magnetic resonance (NMR) was selected as the characterization index of pore structure. In terms of data analysis and processing, the pore radius and T1 matching relationship were established using the equal cumulative frequency matching method; the matching relationship was fitted using an adaptive lifting algorithm; and the NMR logging T1 spectrum was screened and projected using a feature mask-based method. This ensures that the final obtained underground reservoir pore radius distribution map can effectively and clearly present the pore structure characteristics of the wetting phase fluids. Based on this, the peak morphology and radius range of the distribution map are used as the characterization basis for the composition and flow state of the wetting phase fluids, improving the accuracy of characterizing the occurrence state of wetting phase fluids in underground reservoirs.

[0024] Furthermore, in step 1, during the process of obtaining the longitudinal relaxation spectrum T1 distribution of the nuclear magnetic resonance of the core sample saturated with formation water, the echo interval value and magnetic field gradient G value selected in the nuclear magnetic resonance experiment of the core sample are matched with the nuclear magnetic resonance logging instrument; TW≥8s.

[0025] Furthermore, in step 2, the corresponding relationship between T1 and pore radius is found by using the equal cumulative frequency matching method: based on step 1, the cumulative frequency of T1 and the cumulative frequency of pore radius are obtained by accumulating the frequency point by point. When the cumulative frequencies of the two are the same, the pore radius value and the T1 value under the cumulative frequency have a matching relationship, which are used as sample points for fitting the relationship between T1 value and pore radius value.

[0026] Furthermore, in step 2, the base model of the adaptive boosting algorithm is a log-binomial model, and the number of base models is ≥10.

[0027] Furthermore, the ensemble learner obtained by the adaptive boosting algorithm is:

[0028] Where t is the number of training rounds, T is the total number of training iterations, and h t (x) is the mathematical expression for each base learner, α t The contribution of each base learner to the overall model;

[0029] α t The value is calculated using the following formula:

[0030]

[0031] Where, ∈ t The training error for each base model.

[0032] Furthermore, in step 3, the boundary values ​​of the light, unwetting phase signal feature mask are determined by the following method:

[0033] Step a: Select a typical well section with obvious light non-wetting phase signal, and superimpose its two-dimensional spectrum. Delineate the initial boundary of the non-wetting phase signal in the obtained superimposed spectrum to ensure that all non-wetting phase signals are included.

[0034] Step b: Project the signal within the initial boundary onto the T2 and T1 axes, and fit the projected data using a normal distribution function;

[0035] Step c: Select two-sided confidence interval quantiles with a confidence level of 99.9% on the T2 axis as the left and right boundaries of the feature mask, and select one-sided confidence interval quantiles with a confidence level of 99.9% on the T1 axis as the lower boundary of the feature mask. This determines the boundary values ​​of the feature mask for the light unwetting phase signal.

[0036] Furthermore, in step 3, the remaining wetting phase signal is projected onto T1. Specifically, the two-dimensional wetting phase signal obtained after sieving is projected onto the Y direction by summing the signals.

[0037]

[0038] Among them, T1 j Let represent the value of the j-th bin in the longitudinal relaxation time, i represent the bin in the transverse relaxation time, and n represent the total number of bins in the transverse relaxation time. These are the values ​​at the i-th bin of the horizontal relaxation time and the j-th bin of the vertical relaxation time.

[0039] Furthermore, in step 5, the wetting phase fluid includes one or more components such as formation water and mud filtrate; in step 5, the flow state of the wetting phase fluid refers to the relative ease or difficulty of the flow of the wetting phase fluid during the oil and gas extraction process.

[0040] Furthermore, the method for determining the composition of wettable fluids is as follows: when the pore radius distribution exhibits a bimodal or multimodal characteristic, with obvious intervals between different peaks, the wettable fluid is considered to consist of both formation water and mud filtrate. When the pore radius distribution exhibits a unimodal distribution, the wettable fluid is considered to be dominated by formation water. The method for determining the flow state of a certain wettable fluid is as follows: when its pore radius is in the low value range and significantly lower than the pore radius of the mud filtrate in the reservoir, the fluid is considered to have weak flowability during oil and gas extraction, and the smaller the pore radius, the weaker the flowability. When the pore radius is in the medium to high value range and close to the pore radius of the mud filtrate in the reservoir, the fluid is considered to have relatively strong flowability during oil and gas extraction.

[0041] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0042] 1. This invention specifically adjusts the means of obtaining pore radius distribution data of core samples, overcoming the physical mismatch between capillary radius obtained by methods such as mercury intrusion porosimetry and nuclear magnetic resonance relaxation spectrum;

[0043] 2. This invention uses the equal cumulative frequency matching method to establish the matching relationship between pore radius and T1, and uses an adaptive boosting algorithm to fit the matching relationship, which meets the need to fit the nonlinear relationship between nuclear magnetic resonance relaxation spectrum and pore radius, improves the efficiency of modeling and the objectivity and accuracy of the characterization results of pore radius of wetted phase fluid.

[0044] 3. This invention uses a feature mask-based method to screen and project the T1 spectrum of nuclear magnetic resonance logging, overcoming the problem of pore radius inversion distortion caused by the overlap of wetted and non-wetting phases in gas-bearing reservoirs, effectively reducing the interference of non-wetting phase fluid signals, and improving the accuracy of pore radius calculation for wetted phase fluid occurrence.

[0045] 4. This invention predicts the composition of the wetting phase fluid (formation water, mud filtrate) and judges the relative ease or difficulty of the wetting phase fluid's flow during oil and gas extraction, filling the gap in the existing technology. Attached Figure Description

[0046] Figure 1 A flowchart illustrating a method for characterizing the occurrence state of wetted phase fluid in underground reservoirs, provided in an embodiment of the present invention.

[0047] Figure 2 In this embodiment of the invention, the same rock core was selected, and the volumetric frequency distribution of the pore radius of the cast thin sheet and the frequency distribution of the pore radius of the constant-rate mercury intrusion were compared through experimental analysis of the pore characteristics of the cast thin sheet and constant-rate mercury intrusion experiment, respectively.

[0048] Figure 3 This is a diagram showing the pore radius and cumulative frequency matching of the T1 relaxation spectrum of three samples from the same region and layer in an embodiment of the present invention.

[0049] Figure 4 The figures show the results of fitting the three-dimensional pore radius and nuclear magnetic resonance T1 relaxation spectrum of three samples from the same region and layer in this embodiment of the invention using logarithmic, log-binomial (equivalent to an adaptive algorithm with 1 base learner), adaptive boosting algorithm (10 base learners), and adaptive boosting algorithm (100 base learners).

[0050] Figure 5 This is a demonstration diagram of the effect of the feature mask boundary value determination method based on normal confidence interval in an embodiment of the present invention, and a comparison diagram of the initial boundary and the final obtained feature mask.

[0051] Figure 6 This is a flowchart illustrating the process and effects of obtaining the T1 distribution of the wetting phase through screening and projection using a two-dimensional feature mask in an embodiment of the present invention.

[0052] Figure 7 The original nuclear magnetic resonance relaxation spectrum T1 spectrum in the embodiments of the present invention, the T1 spectrum after screening and projection by feature mask and used for pore radius calculation, and the calculated reservoir pore radius distribution map.

[0053] Figure 8 This is an embodiment of the invention, showing the characterization results of the calculated reservoir wetting phase fluid occurrence state, and the verification diagram of the reservoir wetting phase fluid occurrence pore radius calculated by the method of the present invention using the constant velocity mercury intrusion frequency distribution data of two core samples.

[0054] Figure 9 The diagram shows the effect of fitting the relationship between T1 and pore radius using the existing method before the improvement of this invention in Comparative Example 1.

[0055] Figure 10 In Comparative Example 1, the pore radius distribution was obtained using conventional mercury intrusion capillary pressure curves, and a two-segment conversion relationship between pore radius and T1 was established. Using this conversion relationship, a pore radius distribution map of the reservoir wetting phase fluid was obtained.

[0056] Figure 11 The T2 spectrum of the reservoir without using the feature mask method for non-wetting phase fluid signal correction in Comparative Example 2, and the pore radius distribution map of the reservoir wetting phase fluid obtained by using the T2 spectrum as a conversion index. Detailed Implementation

[0057] The present invention will now be described in detail with reference to the accompanying drawings.

[0058] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0059] Example 1

[0060] like Figure 1 As shown in the figure, this embodiment provides a method for characterizing the occurrence state of wetted phase fluid in underground reservoirs. The specific steps of this method are as follows:

[0061] Step 1: Obtain the longitudinal relaxation spectrum T1 distribution of the nuclear magnetic resonance of a core sample saturated with formation water, and obtain the pore radius distribution data of the same core sample; wherein, the pore radius distribution data is the volume frequency distribution data of the pore radius of a cast thin section or the pore radius distribution data of constant-rate mercury intrusion porosimetry.

[0062] Specifically, regarding the analysis data from the two-dimensional nuclear magnetic resonance experiment and pore radius test mentioned in step 1, this includes: matching the echo interval selected for the core two-dimensional nuclear magnetic resonance experiment with the two-dimensional nuclear magnetic resonance logging instrument, and taking T... E = 0.6ms, waiting time T W The maximum value of the sequence should fully excite the fluid signal; take T. W For ≥8s, a low-field magnetic field matching the logging instrument is used, with a magnetic field strength of G = 2MHz;

[0063] Regarding the echo interval T during the experiment E The choice of T depends primarily on the compatibility of the actual well logging with the two-dimensional nuclear magnetic resonance logging instrument. E When there is inconsistency, on the one hand, the two methods will differ in their ability to characterize small pores; on the other hand, due to T... E The magnitude of T determines the rate of diffusion relaxation, therefore different T values ​​will affect the rate of diffusion relaxation. E This will cause an overall shift in the relaxation spectrum. In summary, if the core nuclear magnetic resonance T... E and nuclear magnetic resonance logging T E When the differences are significant, the measured relaxation spectra will show obvious discrepancies, and the core NMR relaxation spectra cannot be used to calibrate the NMR logging response. This is because the T... (The sentence is incomplete and requires further context to translate accurately.) E The value is generally around 0.6ms, so T was chosen for this experiment. E =0.6ms.

[0064] Since two-dimensional nuclear magnetic resonance experiments use a waiting-time series method to excite the core, the basic requirement for the waiting-time series is to ensure the maximum T0. W A value that allows for the excitation of pore fluids at various scales without compromising experimental efficiency is chosen. Based on this principle, T is selected. W ≥8s.

[0065] The selection principle of magnetic field gradient G and T E Similarly, the experimental values ​​should be as close as possible to the values ​​from the logging instrument; in this case, G = 2MHz was chosen.

[0066] The experimental data on pore radius should characterize the pore distribution of the core under natural conditions. Data on the pore radius distribution of constant-rate mercury intrusion or the volume frequency distribution of pore radius of cast thin sections can be selected.

[0067] The selection of experimental data for pore radius analysis is a crucial step, as different experimental methods yield pore radii with varying physical meanings, requiring careful selection. The capillary pressure curves obtained by mercury intrusion porosimetry, centrifugation, and the semi-permeable septum method all reflect the pore space volume distribution under throat control during displacement, embodying the dynamic characteristics of pore radius distribution. In other words, the capillary radius distribution obtained by these three methods is actually strongly correlated with the throat scale. This form of pore radius distribution differs from the nuclear magnetic resonance relaxation spectrum, which reflects the pore radius distribution under natural conditions. Therefore, it cannot be used as pore structure calibration data characterizing the occurrence state of the wetting phase fluid.

[0068] In contrast, constant-rate mercury intrusion (CMI) uses extremely low rates to displace the core. The MCI process is quasi-static; by monitoring pressure fluctuations during MCI, pores and throats can be distinguished, and the frequency distributions of pore radius and throat radius can be calculated. The pore radius frequency distribution can be approximated as being uncontrolled by the throats, representing a natural distribution, consistent with the pore radius characterized by NMR relaxation spectroscopy. This can be used as pore radius calibration data to characterize the presence of the wetting phase fluid. Furthermore, the pore characteristic experiment of cast thin sections involves injecting dyed resin or liquid glue into the thin section. After casting, the dyed portion is analyzed and statistically analyzed to obtain the main ranges and morphological characteristics of the pore radius. The resulting analysis—the volumetric frequency distribution of the pore radius in the cast thin section—also represents the pore distribution under natural core conditions. Therefore, the volumetric frequency distribution of the pore radius in cast thin sections or the pore radius frequency distribution data obtained from constant-rate MCI experiments can be used as pore radius calibration data to characterize the presence of the wetting phase fluid.

[0069] Figure 2 The figure shows a rock core used in this embodiment. After a thin slice was cut from the end face of the rock core to make a cast thin slice, the pore radius volume frequency distribution (dashed line) was obtained after pore observation and data analysis. The pore radius frequency distribution (solid line) was obtained after a constant rate mercury intrusion test was performed on the remaining part of the rock core. Both distributions are representations of the pore radius distribution of the rock core in its natural state. As can be seen from the figure, since they were taken from the same rock core, the two distributions are highly similar.

[0070] In this embodiment, the reservoir segment to be characterized underwent both cast thin-section porosity characteristic experiments and constant-rate mercury intrusion porosimetry (NMR) experiments. During actual operation, one set of data can be used for modeling, while the other is used for result verification to ensure the rationality of the method and the accuracy of the results. Since the pore radius distributions obtained by the two methods have the same physical meaning and the results are similar, the results are considered equivalent regardless of which distribution is chosen for modeling or which data is used for verification. In the modeling process of this embodiment, considering the larger number of core blocks in the cast thin-section experiments, to ensure the accuracy of the model, the volumetric frequency distribution data of the cast thin-section pore radius is matched with the nuclear magnetic resonance relaxation spectrum. During verification, the constant-rate mercury intrusion porosimetry pore radius distribution data is used to verify the calculated reservoir pore radius.

[0071] Step 2: Use the equal cumulative frequency matching method to obtain the discrete matching relationship between the longitudinal relaxation spectrum T1 distribution and the pore radius distribution data in Step 1, and use the adaptive boosting algorithm to perform regression fitting on the discrete matching relationship;

[0072] Specifically, regarding the discrete matching relationship between the longitudinal relaxation spectrum and pore radius data obtained using the equal cumulative frequency matching method described in step 2 above, the following explanation is provided:

[0073] Based on the pore radius distribution curve and longitudinal relaxation spectrum obtained in step 1, the cumulative frequency curve of pore radius and the cumulative frequency distribution of longitudinal relaxation time can be obtained by point-by-point frequency accumulation. The basic idea of ​​the equal cumulative frequency matching method is that when the longitudinal relaxation time T1 and the cumulative frequency of pore radius are equal, the two parameters have a matching relationship. That is, it is assumed that T1 and capillary radius R at this time reflect the same type of capillaries with approximately the same shape and radius in real rocks. In other words, when the cumulative frequencies of pore radius and T1 are the same, the pore radius value and T1 value at that cumulative frequency have a matching relationship and can be used as sample points for fitting the relationship between T1 value and pore radius value. The matching method is described by mathematical formulas:

[0074] T1 = F -1 (G(r))

[0075] stF(T1)≈G(r)

[0076] In the formula, r represents the pore radius (μm); G(r) is the cumulative frequency of the pore radius; F(T1) is the cumulative frequency distribution of the longitudinal relaxation time T1. -1 It is its inverse function.

[0077] Table 1 shows a core sample used for modeling in this embodiment, and a set of corresponding T1 values ​​and pore radius values ​​obtained using the equal cumulative frequency matching method.

[0078] Table 1. Corresponding values ​​of T1 and pore radius for a representative modeling core.

[0079]

[0080] See Figure 3 The figure shows the equal cumulative frequency matching diagram of three samples from the same region and the same stratum in this embodiment. The horizontal axis is the longitudinal relaxation time T1, and the vertical axis is the pore radius. It can be seen from the figure that the matching relationship of samples from the same region and the same stratum has a high degree of consistency.

[0081] It should be noted that since the equal cumulative frequency matching method can directly find the corresponding T1 value and pore radius value through the same cumulative frequency, it is convenient for computer automatic processing and can significantly improve the efficiency of data analysis.

[0082] Specifically, regarding the adaptive boosting algorithm described in step 2 above for regression fitting of discrete matching relationships, the following explanation is provided:

[0083] See Figure 4 The diagram illustrates the exploration of the fitting model in this embodiment. Considering that the data points exhibit an approximately logarithmic distribution, a logarithmic model is first used to fit (a), R. 2 The value is only about 86%, which is insufficient to meet the requirements for high-precision fitting. A large class of algorithms in machine learning is called ensemble learning algorithms. Their basic principle is that for data points whose distribution patterns cannot be accurately fitted by a single simple model, multiple simple models (called base learners) are used, and a weighted sum is applied to obtain a complex model to represent the patterns of the data points. Adaptive boosting algorithms are a branch of ensemble learning algorithms. They assign different weights to each sample and then fit different base learners with strong dependencies. Finally, the base learners are weighted and sequentially summed to obtain the final model.

[0084] The ensemble learner obtained by the adaptive boosting algorithm is:

[0085]

[0086] Where t is the number of training rounds, T is the total number of training iterations, and h t (x) is the mathematical expression for each base learner, α t The contribution of each base learner to the overall model is represented by the following value:

[0087]

[0088] Where, ∈ t The training error for each base model.

[0089] Specifically, the base model of the adaptive boosting algorithm in this embodiment uses a log-binomial model. See [link / reference] Figure 4The figure shows the fitting of the relationship between longitudinal relaxation time and pore distribution using different training iterations (i.e., the number of base learners) in this embodiment. It was found that as the number of base learners increased from 1 to 10 (compared to...), the relationship was improved. Figure 4 b and Figure 4 c), R 2 The value has increased significantly, from 10 to 100 (compared to...). Figure 4 c and Figure 4 d), R 2 Since the value does not change much, the number of base models is set to ≥10.

[0090] Step 3: Obtain nuclear magnetic resonance logging data of the underground reservoir, then determine the boundary value of the feature mask of the light non-wetting phase signal, use the feature mask to filter out the light non-wetting phase signal in the nuclear magnetic resonance logging data, and then project the remaining wetting phase signal onto T1 to obtain the T1 relaxation spectrum of the underground reservoir nuclear magnetic resonance that contains only the wetting phase signal.

[0091] Furthermore, in step 3, the boundary values ​​of the light, unwetting phase signal feature mask are determined by the following method ( Figure 5 (This demonstrates the complete process of this step):

[0092] Step a: Select a typical well section with obvious light non-wetting phase signal, and superimpose its two-dimensional spectrum. Delineate the initial boundary of the non-wetting phase signal in the obtained superimposed spectrum to ensure that all non-wetting phase signals are included.

[0093] See Figure 5 As shown, Figure 5 'a' represents the superimposed spectrum signal and the initial boundary. The purpose of using the superimposed spectrum in this process is to determine the range limit of the light unwetting phase signal. If a two-dimensional spectrum at a certain depth point is used, the range obtained may not fully and accurately reflect the light unwetting phase signal due to the poor representativeness of the relaxation spectrum at that depth point. Figure 5 b represents the signal within the initial boundary.

[0094] Step b: Project the signal within the initial boundary onto the T2 and T1 axes, and fit the projected data using a normal distribution function;

[0095] Because the T2 and T1 spectra of light, unwetting phase signals are not affected by surface relaxation, their T2 and T1 spectra are not controlled by pore size and have a wider distribution like those of wetting phase signals. Instead, they are limited to a small range, exhibiting a near-ring-shaped characteristic with strong signals at the center and weak signals at the periphery. (See [reference]). Figure 5 As shown in b; project it onto axes T2 and T1 respectively ( Figure 5 c and Figure 5 d) Its distribution exhibits the characteristics of a normal distribution, and can be fitted using a normal distribution function;

[0096] Step c: Select two-sided confidence interval quantiles with a confidence level of 99.9% on the T2 axis as the left and right boundaries of the feature mask, and select one-sided confidence interval quantiles with a confidence level of 99.9% on the T1 axis as the lower boundary of the feature mask. This determines the boundary values ​​of the feature mask for the light unwetting phase signal.

[0097] The purpose of using confidence level quantiles to determine the boundary is to further refine the aforementioned artificially delineated initial boundary. Since the initial boundary needs to ensure that all non-wetting phase signals are included, a relatively large envelope range was chosen. This means that not only is the probability of including the light non-wetting phase signal 100%, but some wetting phase signals may also be included, which introduces a degree of subjectivity. Based on the normal distribution function, using a higher confidence level to divide its distribution can simultaneously ensure the accuracy and objectivity of filtering out light non-wetting phase signals. Because the T1 spectrum of the light non-wetting phase signal is only affected by volume relaxation, and its value is significantly larger than the T1 value of the wetting phase signal, only one-sided confidence interval quantiles are needed for T1; conversely, for T2, two-sided confidence interval quantiles are used. See [link to relevant documentation] Figure 4 As shown, specifically, in this embodiment, the quantiles of the two-sided confidence interval with a confidence level of 99.9% obtained by fitting the normal probability distribution curve after T2 axis projection are 38ms and 369ms, respectively, which serve as the left and right boundaries of the feature mask. Figure 5 c); The normal probability distribution curve obtained by projecting onto the T1 axis is used to calculate the quantile of the one-sided confidence interval with a confidence level of 99.9% as 1430ms, which is used as the lower boundary of the feature mask. Figure 5 d). Figure 5 e is a comparison image of the final feature mask and the initial boundary.

[0098] See Figure 6 As shown, after determining the boundary values ​​of the feature mask, the feature mask can be used to screen each depth point of the two-dimensional nuclear magnetic resonance logging data: the light unwetting phase signal located inside the feature mask will be filtered out, and the wetting phase signal outside the mask will be retained. Figure 6 The complete process of this step is shown, in which Figure 6 'a' represents the original two-dimensional spectrum at a certain depth point. Figure 6 b is the projection of the two-dimensional spectrum onto T1. Figure 6 c is the feature mask determined by the aforementioned method. Figure 6 d represents the two-dimensional spectrum of the light, unwetting phase signal after screening using a feature mask. Figure 6 e represents the two-dimensional spectrum of the wetting phase signal after sieving. Figure 6 f is the T1 projection of the two-dimensional spectrum of the wetting phase signal.

[0099] Furthermore, in step 3, the remaining wetting phase signal is projected onto T1. Specifically, the two-dimensional wetting phase signal obtained after sieving is projected onto the Y direction by summing the signals.

[0100]

[0101] Among them, T1 j Let represent the value of the j-th bin in the longitudinal relaxation time, i represent the bin in the transverse relaxation time, and n represent the total number of bins in the transverse relaxation time. The values ​​are for the ith bin of the lateral relaxation time and the j-th bin of the longitudinal relaxation time. The projected T1 signal, after removing the signal of the light, unwetting phase, represents the signal of the existing wetting fluid in the reservoir—including mud filtrate and formation water.

[0102] Step 4: Using the regression fitting curve relationship obtained in Step 2, i.e. the conversion relationship between longitudinal relaxation T1 and pore radius data, the T1 relaxation spectrum of the underground reservoir nuclear magnetic resonance obtained in Step 3, which only contains wetting phase signal, is converted to obtain the pore radius distribution map of the underground reservoir wetting phase fluid.

[0103] Specifically, Figure 7 The process of converting the T1 relaxation spectrum of nuclear magnetic resonance in subsurface reservoirs to obtain the pore radius distribution map of the wetted phase fluid in subsurface reservoirs is demonstrated. Figure 7 a represents the T1 spectrum, which varies continuously with formation depth before the removal of non-wetting phase fluids. Figure 7 b represents the T1 spectrum, which varies continuously with formation depth, obtained after step 3. Figure 7 Compared to a, the unwetting phase T1 signal located in the long relaxation time region was removed; Figure 7 c represents the regression fitting curve relationship obtained in step 2. Figure 7 The pore radius distribution map of the wetted phase fluid in the underground reservoir is obtained by converting the data in b.

[0104] Step 5: Use the peak shape of the pore radius distribution spectrum obtained in Step 4 as the basis for the composition of the wetting phase fluid; use the radius range of the pore radius distribution spectrum obtained in Step 4 as the basis for judging the flow state of a certain wetting phase fluid, thereby obtaining the occurrence state of the wetting phase fluid in the underground reservoir.

[0105] Following the aforementioned steps, the pore radius distribution occupied by the wetting phase fluid is calculated. Furthermore, the following two criteria are used to characterize the occurrence state of the wetting phase fluid in subsurface reservoirs:

[0106] The calculated pore radius distribution peak morphology of the wetted phase fluid in the underground reservoir is used as the basis for determining the composition of the wetted phase fluid.

[0107] The radius range of the pore radius distribution peak of the wetted phase fluid in the underground reservoir, obtained from the calculation, is used as the basis for judging the fluidity state of a certain wetted phase fluid.

[0108] Criteria for determining the composition of the wetting phase fluid: Since the capillary radius distribution in rock pores is continuous and uninterrupted, if the pore radius distribution exhibits a bimodal or multimodal characteristic, with distinct intervals between different peaks, then the wetting phase fluid is considered to consist of formation water and mud filtrate, with the intermediate intervals containing signals of the removed light, non-wetting phase. Conversely, it is more likely to consist of only one component.

[0109] Criteria for determining the flowability of wetting phase fluids: For underground reservoirs, under the same production pressure differential, wetting phase fluids in large pores are more likely to enter the wellbore due to less capillary force constraint. Conversely, wetting phase fluids in small pores are more easily confined within the pores due to the balance between production pressure differential and capillary force. Therefore, the relative size of the radius range of the pore distribution spectrum peaks can be used as a basis for determining the flowability of the wetting phase fluid or its ease of flow. During drilling, due to the over-equilibrium effect (mud column pressure greater than formation pressure), mud filtrate often intrudes into the reservoir. The pore radius of this mud filtrate undoubtedly possesses better flowability (because the flow of mud filtrate in the reservoir is already a fait accompli). Therefore, the pore radius of the mud filtrate can be used as a benchmark to judge the flowability of the remaining wetting phase fluids. The specific method for determining the flow state of a certain wetting phase fluid is as follows: when its pore radius is in the low range and is significantly lower than the pore radius of the mud filtrate in the reservoir, it is considered that the fluid of this part of the wetting phase fluid is weak during oil and gas extraction, and the smaller the pore radius, the weaker the fluidity; when the pore radius is in the medium to high range and is close to the pore radius of the mud filtrate in the reservoir, it is considered that the fluid of this part of the wetting phase fluid is relatively strong during oil and gas extraction.

[0110] See Figure 8 As shown, in the specific well section used in this embodiment, the T1 relaxation spectrum corrected by the feature mask is used. Figure 7 (as shown in b) The pore radius distribution of the wetting phase fluid was calculated and its occurrence state was characterized. Figure 8a) During the calculation process, combined with the relevant technical specifications for nuclear magnetic resonance logging interpretation and the actual logging response characteristics of this well section, it was determined that signals with T1 values ​​below 5ms were mainly the response of clay-bound water, while those above 5ms were the response of wetting phase fluids in the reservoir's large and small pores. Therefore, 5ms was taken as the left boundary for calculating the pore radius. The calculation results show that the reservoir in this well section is highly heterogeneous and affected by overbalanced drilling, with significant mud filtrate intrusion. Medium and large pores with good fluid flowability are filled with wetting phase mud filtrate, while micro and small pores that have varying degrees of binding effect on fluid flow are still saturated with the original wetting phase formation water. Figure 8 b and Figure 8 c represents the pore radius distribution obtained from constant-rate mercury intrusion porosimetry experiments on cores at depths of 4854.4 m and 4860.7 m, respectively. These two cores were not used for modeling, and therefore can be used to verify the accuracy of the pore radius calculated by the method of this invention. Figure 8 The calculation results of the corresponding depth points in a are Figure 8 b and Figure 8 The comparison with c shows that the median and left and right ranges of the pore radius calculated from the above depth points are close to the pore radius distribution of the constant-rate mercury intrusion used for verification, which directly confirms the beneficial effects of the method of the present invention.

[0111] It should be noted that the pore radius obtained by constant-rate mercury intrusion is measured under conditions of gradual pressurization of a single-phase fluid to near saturation, reflecting the complete pore radius distribution of the core. However, actual reservoirs contain two-phase fluids. This invention mainly characterizes the pore radius of the wetting phase fluid in actual underground reservoirs, without considering the small amount of light, non-wetting phase fluid present in the near-wellbore zone. Therefore, the calculated pore radius distribution cannot be completely consistent with the results obtained by constant-rate mercury intrusion. The relatively high degree of agreement between the median and the pore radius distribution obtained by constant-rate mercury intrusion is sufficient to demonstrate the beneficial effects achieved by this invention.

[0112] Comparative demonstration of the beneficial effects of the present invention

[0113] The characterization of the occurrence state of wetting phase fluids in underground reservoirs mainly involves two steps. First, the pore structure of the wetting phase fluids in underground reservoirs is calculated. Then, the distribution characteristics of the pore structure are analyzed to obtain the occurrence characteristics of wetting phase fluids such as formation water and mud filtrate.

[0114] The existing technology for calculating the pore radius distribution of fluids in underground reservoirs uses the following technical solution, and its main steps are as follows:

[0115] ① Use conventional mercury intrusion porosimetry, centrifugation, or semi-permeable diaphragm method to determine the core capillary radius distribution;

[0116] ② Nuclear magnetic resonance experiments were conducted on water-saturated cores to obtain the transverse relaxation time T2 spectrum of the cores;

[0117] ③ Establish the correspondence between capillary radius and T2 spectrum, and use one-segment / two-segment / three-segment fitting methods to establish the conversion relationship between T2 and capillary radius;

[0118] ④ Using this conversion relationship, the reservoir nuclear magnetic resonance logging T2 spectrum is converted into a pore radius distribution that varies continuously with depth, thus completing the characterization of the pore structure of the fluid in the underground reservoir.

[0119] The aforementioned technical solution establishes a bridge between well logging parameters (T2 NMR spectrum) and pore structure parameters (capillary radius) through core experiments, achieving the goal of continuously characterizing reservoir pore structure using well logging curves. However, existing methods have some technical issues that potentially affect the accuracy of characterization:

[0120] ① The T2 NMR spectrum of the core represents the pore radius distribution of the rock under natural conditions. However, in step 1 of the existing technology, the capillary radius distribution of the core is obtained by conventional mercury intrusion porosimetry, centrifugation, or semi-permeable septum method. The capillary radius distribution obtained by the above methods is the pore space volume distribution under the throat control during the displacement process, which reflects the dynamic characteristics of the pore radius distribution. That is, the capillary radius distribution obtained by the above three methods is actually significantly controlled by the throat scale and is not the pore radius distribution under natural conditions of the rock. There is a certain difference between the physical meaning of the T2 NMR spectrum and the established conversion relationship may lead to the deviation of the characterization results.

[0121] ② In step 3 of the existing technology, the one-segment / two-segment / three-segment fitting method used to establish the conversion relationship between T2 and capillary radius requires too much human intervention to determine the specific segmentation mode and connection points used for different core samples, which affects the modeling efficiency. Moreover, the more segments there are, the more non-differentiable points there are in the fitted curve, which may cause numerical rationality problems such as curve discontinuity and jumps when the model is applied.

[0122] ③ In steps 2 and 3 of the existing technology, the T2 spectrum of nuclear magnetic resonance of 100% water-saturated core is used in the laboratory to establish a relationship with the core capillary radius. However, when this conversion relationship is applied to actual reservoirs, since the reservoir pores are not 100% water-saturated, it is necessary to remove the non-wetting phase signal in the reservoir T2 spectrum before conversion. However, due to the influence of diffusion relaxation effect, the T2 value range of non-wetting phase is likely to overlap with the T2 value range of wetting phase. If T2 is used as the conversion index, it is difficult to remove the non-wetting phase, making the pore radius distribution obtained by conversion easily affected by the non-wetting phase signal.

[0123] In addition, existing technologies mainly calculate the pore structure characteristics of underground reservoir fluids, which is only the first step in characterizing the occurrence state of underground reservoir wetting phase fluids. They have not developed a characterization technology for the occurrence state of formation water, mud filtrate, and other wetting phase fluids. Therefore, in terms of both calculation accuracy and characterization process, there is currently no method that can accurately characterize the occurrence state of underground reservoir wetting phase fluids.

[0124] Comparative Example 1

[0125] To illustrate the beneficial effects of the present invention, Comparative Example 1 was provided, which used the following techniques to characterize the pore structure of the wetted phase fluid:

[0126] ① The capillary radius distribution obtained from conventional mercury intrusion porosimetry was used as the pore radius distribution for fitting;

[0127] ② A two-stage fitting method was used to establish the relationship between pore radius and relaxation spectrum;

[0128] ③ The T1 after feature mask correction and projection was used as the conversion index.

[0129] As can be seen from the above techniques, this comparative example uses T1, which is corrected and projected using a feature mask and has the characteristics of this invention, as a transformation index. However, it does not use the experimental method (cast sheet or constant-rate mercury intrusion) reflecting the natural distribution of pore radius as described in this invention to obtain the pore radius distribution, nor does it use an adaptive boosting algorithm for fitting. The latter two methods use existing technologies. The purpose of this comparative example is mainly to illustrate the effectiveness and superiority of this invention in using the experimental method (cast sheet or constant-rate mercury intrusion) reflecting the natural distribution of pore radius and the adaptive boosting algorithm.

[0130] An analysis of the results, problems, and causes of this comparative analysis is conducted.

[0131] 1. Figure 9 This is a fitting graph showing the relationship between pore radius and T1 spectrum obtained using a two-segment fitting method. From a modeling time efficiency perspective, this method requires manually determining the number of segments (in this example, it's divided into two segments based on actual analysis) and the connection points between each segment. During computer programming, further conditional judgments and classification processing are needed, significantly reducing the efficiency of model building. From a modeling effect perspective, the connection points between segments are not differentiable (the slopes to the left and right of the connection point are inconsistent). In actual calculations, this theoretically leads to non-objective changes in the pore radius distribution values. A more easily observable drawback is the potential for issues with the numerical reasonableness of the pore radius values: from... Figure 9As can be seen, compared with the adaptive boosting algorithm, the slope of each segment of this method is fixed. For the first segment of the fitting relationship, when T1 < 10ms, it is easy for the pore radius to be negative, which causes the problem of the rationality of the fitted values.

[0132] 2. Figure 10 The reservoir pore radius distribution map obtained using the above-mentioned technical method shows that the calculated pore radius is significantly higher than that obtained by the method of the present invention. Figure 8 a) Values ​​more than one order of magnitude lower, and negative values ​​appearing in the low-value region: The main reason is that the conventional mercury intrusion capillary radius used as the inversion sample reflects the throat radius rather than the pore radius. The throat radius is generally one to two orders of magnitude lower than the pore radius, and it mainly reflects the connectivity of the pores, but cannot reflect the wetting phase occurrence state required for the study. Because the calculated radius values ​​are very small, and because a more accurate adaptive boosting algorithm was not used but conventional algebraic piecewise fitting was employed, negative values ​​appeared in the low-value region. In addition, comparing the calculated results with the constant-rate mercury intrusion pore radius distribution (… Figure 8 b and Figure 8 c) By comparison, it can also be shown that the pore radius calculated using existing methods is significantly smaller.

[0133] The above analysis and comparison with the effects of the method of the present invention confirm the superiority and beneficial effects of the two techniques of the present invention: using experimental methods (cast thin plates or constant-rate mercury intrusion) that reflect the natural distribution of pore radius to obtain the pore radius distribution, and using an adaptive lifting algorithm for fitting.

[0134] Comparative Example 2

[0135] To illustrate the beneficial effects of the present invention, Comparative Example 2 was provided, which used the following techniques to characterize the pore structure of the wetted phase fluid:

[0136] ① The volumetric frequency distribution of pore radius in the cast sheet was used as the pore radius distribution for fitting;

[0137] ② An adaptive lifting algorithm was used to establish the relationship between pore radius and relaxation spectrum;

[0138] ③ Use the original (uncorrected for feature masking) T2 as the transformation index.

[0139] As can be seen from the above techniques, this comparative example uses the volumetric frequency distribution of the pore radius of the cast sheet, which has the characteristics of this invention, as the pore radius distribution for fitting. It also uses the adaptive boosting algorithm, which has the characteristics of this invention, to establish the relationship between the pore radius and the relaxation spectrum. However, it does not use the T1 spectrum based on feature mask correction and projection as in this invention; instead, it uses the original T2 spectrum as the conversion index, employing existing technical means. The purpose of this comparative example is mainly to illustrate the effectiveness and superiority of the T1 spectrum based on feature mask correction and projection used in this invention as the conversion index.

[0140] An analysis of the results, problems, and causes of this comparative analysis is conducted.

[0141] Figure 11 a is the T2 spectrum used for conversion. Figure 11 b represents the reservoir pore radius distribution obtained after the conversion. Figure 11 As can be seen from a, a strong non-wetting phase interference signal appears at T2: because the T2 and T1 spectra of the light non-wetting phase signal are not affected by surface relaxation, the signal is limited to a small range in the two-dimensional spectrum of a single depth point. Figure 6 d) In a one-dimensional distribution, it exhibits a narrower peak. A characteristic of this distribution on a depth-varying one-dimensional relaxation time map is that it displays a fixed relaxation time value in a certain well section (e.g., Figure 11 a. At 4851m-4854m, the unique T2 value of the interference signal is 250ms; at 4854m-4869m, the unique T2 value of the interference signal is 100ms. Furthermore, it cannot reflect the true pore radius characteristics, thus significantly impacting the judgment of the wetting phase fluid's occurrence characteristics. Specifically, this is manifested in… Figure 11 In b, there are multiple vertically distributed non-true pore radius distribution values.

[0142] In summary, the problems encountered using existing methods are mainly due to issues such as inversion samples, fitting methods, computational indices, and non-wetting phase correction. The method of this invention improves upon these aspects, thus achieving significant advantages over existing methods. It should be noted that the equal cumulative frequency is also an important feature of this invention, but its main purpose and advantage lie in improving modeling efficiency rather than model accuracy. This point has been explained in the foregoing examples and is therefore not shown in the comparative examples.

[0143] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements 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 characterizing the occurrence state of wetting phase fluid in underground reservoirs, characterized in that, Includes the following steps: Step 1: Obtain the longitudinal relaxation spectrum T1 distribution of the nuclear magnetic resonance of a core sample saturated with formation water, and obtain the pore radius distribution data of the same core sample; wherein, the pore radius distribution data is the volume frequency distribution data of the pore radius of a cast thin section or the pore radius distribution data of constant-rate mercury intrusion porosimetry. Step 2: Use the equal cumulative frequency matching method to obtain the discrete matching relationship between the longitudinal relaxation spectrum T1 distribution and the pore radius distribution data in Step 1, and use the adaptive boosting algorithm to perform regression fitting on the discrete matching relationship; Step 3: Obtain nuclear magnetic resonance logging data of the underground reservoir, then determine the boundary value of the feature mask of the light non-wetting phase signal, use the feature mask to filter out the light non-wetting phase signal in the nuclear magnetic resonance logging data, and then project the remaining wetting phase signal onto T1 to obtain the T1 relaxation spectrum of the underground reservoir nuclear magnetic resonance that contains only the wetting phase signal. Step 4: Using the regression fitting curve relationship obtained in Step 2, i.e. the conversion relationship between the longitudinal relaxation spectrum T1 and the pore radius data, the T1 relaxation spectrum of the underground reservoir nuclear magnetic resonance obtained in Step 3, which only contains the wetting phase signal, is converted to obtain the pore radius distribution map of the underground reservoir wetting phase fluid. Step 5: Use the peak shape of the pore radius distribution spectrum obtained in Step 4 as the basis for the composition of the wetting phase fluid; use the radius range of the pore radius distribution spectrum obtained in Step 4 as the basis for judging the flow state of a certain wetting phase fluid, thereby obtaining the occurrence state of the wetting phase fluid in the underground reservoir.

2. The method for characterizing the occurrence state of the wetting phase fluid in an underground reservoir according to claim 1, characterized in that, In step 1, during the process of obtaining the longitudinal relaxation spectrum T1 distribution of the nuclear magnetic resonance of the core sample saturated with formation water, the echo interval value and magnetic field gradient G value selected in the nuclear magnetic resonance experiment of the core sample are matched with the nuclear magnetic resonance logging instrument; TW≥8s.

3. The method for characterizing the occurrence state of the wetting phase fluid in an underground reservoir according to claim 1, characterized in that, In step 2, the corresponding relationship between T1 and pore radius is found by using the equal cumulative frequency matching method: Based on step 1, the cumulative frequency of T1 and the cumulative frequency of pore radius are obtained by accumulating the frequency point by point. When the cumulative frequencies of the two are the same, the pore radius value and the T1 value under the cumulative frequency have a matching relationship, which are used as sample points for fitting the relationship between T1 value and pore radius value.

4. The method for characterizing the occurrence state of the wetted phase fluid in an underground reservoir according to claim 3, characterized in that, In step 2, the base model of the adaptive boosting algorithm is a log-binomial model, and the number of base models is ≥10.

5. The method for characterizing the occurrence state of the wetting phase fluid in an underground reservoir according to claim 4, characterized in that, The ensemble learner obtained by the adaptive boosting algorithm is: ;in, t For training rounds, T This represents the total number of training iterations. For each base learner, there is a mathematical expression. The contribution of each base learner to the overall model; The value is calculated using the following formula: in, The training error for each base model.

6. The method for characterizing the occurrence state of the wetting phase fluid in an underground reservoir according to claim 1, characterized in that, In step 3, the boundary values ​​of the light, unwetting phase signal feature mask are determined by the following method: Step a: Select a typical well section with obvious light non-wetting phase signal, and superimpose its two-dimensional spectrum. Delineate the initial boundary of the non-wetting phase signal in the obtained superimposed spectrum to ensure that all non-wetting phase signals are included. Step b: Project the signal within the initial boundary onto the T2 and T1 axes, and fit the projected data using a normal distribution function; Step c: Select two-sided confidence interval quantiles with a confidence level of 99.9% on the T2 axis as the left and right boundaries of the feature mask, and select one-sided confidence interval quantiles with a confidence level of 99.9% on the T1 axis as the lower boundary of the feature mask. This determines the boundary values ​​of the feature mask for the light unwetting phase signal.

7. The method for characterizing the occurrence state of the wetted phase fluid in an underground reservoir according to claim 6, characterized in that, In step 3, the remaining wetting phase signal is projected onto T1. Specifically, the two-dimensional wetting phase signal obtained after sieving is projected onto the Y direction by summing the signals. Let represent the value of the j-th bin in the longitudinal relaxation time, i represent the bin in the transverse relaxation time, and n represent the total number of bins in the transverse relaxation time. These are the values ​​at the i-th bin of the horizontal relaxation time and the j-th bin of the vertical relaxation time.

8. The method for characterizing the occurrence state of the wetted phase fluid in an underground reservoir according to any one of claims 1-7, characterized in that, In step 5, the wetting phase fluid includes one or more components, including formation water and mud filtrate; in step 5, the flow state of the wetting phase fluid refers to the relative ease or difficulty of the flow of the wetting phase fluid during the oil and gas extraction process.

9. The method for characterizing the occurrence state of the wetted phase fluid in an underground reservoir according to claim 8, characterized in that, The method for determining the composition of wettable fluids is as follows: when the pore radius distribution exhibits a bimodal or multimodal characteristic, with obvious intervals between different peaks, the wettable fluid is considered to consist of both formation water and mud filtrate. When the pore radius distribution exhibits a unimodal distribution, the wettable fluid is considered to be dominated by formation water. The method for determining the flow state of a certain wettable fluid is as follows: when its pore radius is in the low value range and significantly lower than the pore radius of the mud filtrate in the reservoir, the fluid is considered to have weak flowability during oil and gas extraction, and the smaller the pore radius, the weaker the flowability. When the pore radius is in the medium to high value range and close to the pore radius of the mud filtrate in the reservoir, the fluid is considered to have relatively strong flowability during oil and gas extraction.

Images