An intelligent sand body seismic prediction method based on RF-GAM

Abstract

The present invention discloses a sand body seismic intelligent prediction method based on RF-GAM, comprising the following steps: S1: collecting drilling data and seismic data of a target block, extracting sand-to-formation ratio information from the drilling data, and extracting seismic attribute information from the seismic data; S2: analyzing the correlation between the two types of information and screening relevant factors of the sand-to-formation ratio; S3: using a multivariate linear regression analysis method to predict sand bodies and obtain preliminary sand body prediction results; S4: based on the preliminary sand body prediction results, screening locations where the relevant factors match the sand-to-formation ratio, and using the seismic attribute information at these locations as influencing factors of the sand-to-formation ratio; S5: using a random forest to screen and obtain important influencing factors; S6: using a generalized additive model to perform attribute fusion, and visualizing the results of the attribute fusion to obtain the final sand body prediction results. The present invention can predict more accurate sand body prediction results, providing technical support for oil and gas field exploration and development.

Description

Technical Field

[0001] The present invention relates to the field of oil and gas field development geological technology, and in particular to a sand body seismic intelligent prediction method based on RF-GAM. Background Art

[0002] Predicting subsurface sand bodies is a key task in oil and gas exploration and development. However, for reservoirs with sparse well networks, such as those in the exploration or early development stages, and offshore reservoirs, well spacing is often large due to limitations in reservoir understanding and operational costs. Accurately predicting the distribution of subsurface sand bodies based solely on drilling data is difficult. To address this issue, traditional methods primarily rely on seismic data, utilizing techniques such as seismic sedimentology, combined with data from drilling coring and well logging, to characterize the planar distribution of sedimentary bodies using seismic attributes within the constraints of well data. However, different seismic attributes reflect different geological properties. When encountering geological conditions with large variations in sand body thickness, strong physical heterogeneity, and complex oil, gas, and water distribution, a single seismic attribute can be incomplete and uncertain. Simple manual interpretation of multiple attributes requires significant human and material resources, and the prediction process is also influenced by subjective experience, resulting in multiple solutions and inaccurate characterization of subsurface sand bodies. Therefore, a sand body prediction method that can more accurately characterize subsurface sand bodies is urgently needed. Summary of the Invention

[0003] In response to the above problems, the present invention aims to provide a sand body seismic intelligent prediction method based on RF-GAM, which can obtain more accurate sand body prediction results.

[0004] The technical solutions of the present invention are as follows:

[0005] An intelligent sand body seismic prediction method based on RF-GAM includes the following steps:

[0006] S1: collecting drilling data and seismic data of the target block, and extracting sand-to-soil ratio information from the drilling data, and extracting seismic attribute information from the seismic data;

[0007] S2: analyzing the correlation between the seismic attribute information and the sand-to-ground ratio information, and screening relevant factors of the sand-to-ground ratio according to the analysis results of the correlation;

[0008] S3: Based on the relevant factors, a multivariate linear regression analysis method is used to predict the sand body to obtain a preliminary prediction result of the sand body;

[0009] S4: Based on the preliminary prediction result of the sand body, a location where the relevant factors match the sand-to-formation ratio is selected, and the seismic attribute information at the location is used as an influencing factor of the sand-to-formation ratio;

[0010] S5: Based on the influencing factors, use random forest to screen and obtain important influencing factors;

[0011] S6: Based on the important influencing factors, a generalized additive model is used to perform attribute fusion, and the result of the attribute fusion is visualized to obtain a final prediction result of the sand body.

[0012] Preferably, in step S1, the sand-to-land ratio information is the sand-to-land ratio information of sand bodies above fine sand.

[0013] Preferably, in step S2, when analyzing the correlation between the seismic attribute information and the sand-to-ground ratio information, the Pearson correlation coefficient is used for correlation analysis.

[0014] Preferably, before visualizing the result of the attribute fusion, the method further includes filtering the result of the attribute fusion to eliminate noise.

[0015] Preferably, during filtering, any one or more filtering algorithms selected from the group consisting of median filtering, mean filtering, bilateral filtering, fuzzy filtering, and majority filtering are used for filtering.

[0016] The beneficial effects of the present invention are:

[0017] The present invention uses machine learning to perform fusion prediction of seismic attributes that have a significant impact on the channel-sand ratio and visualizes the prediction results, which can highlight the planar distribution of the channel sand body and provide technical support for the exploration and development of oil and gas fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 Schematic diagram of the flow of the sand body seismic intelligent prediction method based on RF-GAM of the present invention;

[0020] Figure 2 This is a schematic diagram of preliminary prediction results of sand bodies by multiple linear regression according to a specific embodiment of the present invention;

[0021] Figure 3 This is a schematic diagram of sand body prediction results of generalized addition according to a specific embodiment of the present invention;

[0022] Figure 4 This is a schematic diagram of sand body prediction results after mode filtering in a specific embodiment of the present invention. DETAILED DESCRIPTION

[0023] The present invention is further described below in conjunction with the embodiments and drawings. It should be noted that, in the absence of conflict, the embodiments in this application and the technical features in the embodiments can be combined with each other. It should be pointed out that, unless otherwise specified, all technical and scientific terms used in this application have the same meanings as those commonly understood by those of ordinary skill in the art to which this application belongs. The words "including" or "comprising" and similar words used in the present invention are intended to mean that the elements or objects appearing before the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects.

[0024] like Figure 1 As shown, the present invention provides a sand body seismic intelligent prediction method based on RF-GAM, comprising the following steps:

[0025] S1: collecting drilling data and seismic data of the target block, and extracting sand-to-soil ratio information from the drilling data, and extracting seismic attribute information from the seismic data.

[0026] In a specific embodiment, the sand-to-formation ratio information refers to the sand-to-formation ratio of sand bodies with fine sand or greater. It should be noted that this ratio is more accurate and convenient for predicting river channel reservoirs, and the results obtained are more conducive to depicting river channel morphology. Selecting medium sand or coarse sand will yield less accurate results, but these can also be used for prediction, depending on the user's desired accuracy.

[0027] S2: Analyze the correlation between the seismic attribute information and the sand-to-ground ratio information, and screen relevant factors of the sand-to-ground ratio according to the analysis results of the correlation.

[0028] In a specific embodiment, when analyzing the correlation between the seismic attribute information and the sand-to-ground ratio information, the Pearson correlation coefficient is used for correlation analysis. The calculation formula of the Pearson correlation coefficient is as follows:

[0029]

[0030] Where: x is the sand-to-soil ratio information; y is the seismic attribute information.

[0031] The strength of the correlation between the seismic attribute information and the sand-to-ground ratio information is judged according to the absolute value of the Pearson correlation coefficient: when Pearson is between 0.8-1.0, there is an extremely strong correlation; when Pearson is between 0.6-0.8, there is a strong correlation; when Pearson is between 0.4-0.6, there is a moderate correlation; when Pearson is between 0.2-0.4, there is a weak correlation; when Pearson is between 0.0-0.2, there is an extremely weak correlation or no correlation.

[0032] S3: Based on the relevant factors, a multivariate linear regression analysis method is used to predict the sand body and obtain preliminary prediction results of the sand body.

[0033] It should be noted that the multiple linear regression analysis method is an existing technology, and the specific steps of the multiple regression analysis will not be repeated here.

[0034] S4: Based on the preliminary prediction result of the sand body, a position where the relevant factors match the sand-to-formation ratio is screened, and the seismic attribute information at the position is used as an influencing factor of the sand-to-formation ratio.

[0035] S5: Based on the influencing factors, use random forest to screen and obtain important influencing factors.

[0036] It should be noted that the random forest method is prior art. In the present invention, a random forest model is established using the preferred influencing factors. Through model training, the importance of each preferred influencing factor to the sand-to-land ratio is obtained, expressed as the average mean square error. The larger the average mean square error, the more important the corresponding preferred influencing factor, thereby screening the important influencing factors from the preferred influencing factors.

[0037] S6: Based on the important influencing factors, a generalized additive model is used to perform attribute fusion, and the result of the attribute fusion is visualized to obtain a final prediction result of the sand body.

[0038] It should be noted that the generalized additive model is a prior art. The generalized additive model can be used to combine regression models for linear or nonlinear fitting, fit the data in a way with a higher degree of freedom, and add the response variable and each explanatory variable to form a GAM. The calculation formula is as follows:

[0039]

[0040] Where f can be any function, for each individual x in (the nth variable of the i-th group of data) calculate f, and finally unify the results for prediction.

[0041] In the process of fusion and unification, the use of spline functions can make the segmentation process of each piecewise function smooth and have a certain smoothness at the intersection of each segment, making the overall value stable and convergent. Therefore, the spline function is used as a parameter of the optimization model. It should be noted that the spline function is only a preferred function, and the present invention can also use other functions for generalized addition.

[0042] In a specific embodiment, the RF-GAM-based sand body seismic intelligent prediction method of the present invention further includes filtering the attribute fusion results to remove noise before visualizing them. This can make the planar distribution characteristics of the river channel sand bodies more distinct.

[0043] Optionally, during filtering, any one or more filtering algorithms selected from median filtering, mean filtering, bilateral filtering, fuzzy filtering, and majority filtering are used for filtering.

[0044] The median filter replaces the grayscale value of a pixel with the median of the grayscale values ​​in the pixel area. This method can remove impulse noise and salt and pepper noise while retaining the edge details of the image. The median filter is very effective in removing salt and pepper noise and patch noise.

[0045] The mean filter is similar to the median filter in that it replaces the original pixel value with the average value of the pixels around the pixel point, thereby filtering out the edge information of the image while filtering out the noise.

[0046] Bilateral filtering is a nonlinear filtering method that compromises image proximity and pixel value similarity, removing noise while preserving edge information. Bilateral filtering consists of two functions: one with filter coefficients determined by spatial distance, and the other with filter coefficients determined by pixel differences.

[0047] The so-called "blur" in the blur filter can be understood as taking the average value of the surrounding pixels for each pixel. The "middle point" takes the average value of the "surrounding points". Numerically, this is a kind of "smoothing". Graphically, this is equivalent to producing a "blurred" effect, and the "middle point" loses detail. When calculating the average value, the larger the value range, the stronger the "blurring effect". Unlike the mean filter, it is not a simple average, but a weighted average based on distance. The closer the point, the greater the weight, and the farther the point, the smaller the weight. The weight distribution model adopts a two-dimensional normal distribution (Gaussian function).

[0048] The mode filtering is a method of averaging a target pixel and surrounding pixels and then filling the target pixel back to achieve the filtering purpose.

[0049] The East China Sea Shelf Basin is located on the southeastern edge of the Eurasian Plate, where the Indo-Australian Plate, the Pacific Plate, and the Eurasian Plate converge. It covers an area of ​​approximately 26×10 4 km 2The Huangyan structure in the Xihu Sag is located in the south-central part of the central inversion tectonic belt of the Xihu Sag in the East China Sea Shelf Basin. It is a NE-SW trending anticline with relatively flat strata. The target stratum studied is the lower member of the Oligocene Huagang Formation, a product of sedimentary filling during the Eocene fault-depression transition phase. It primarily develops a shallow lake-deltaic sedimentary system within a continental setting, with a sedimentary thickness ranging from 1,000 to 2,000 meters. The Huagang Formation is generally thinner than the Huashang Formation. Its primary lithologies include sub-feldspathic sandstone, sub-lithic sandstone, feldspathic lithic sandstone, and lithic feldspathic sandstone. Its porosity ranges from 2.1% to 12.5%, primarily concentrated between 8% and 10%, and its permeability ranges from 0.02 mD to 22.72 mD. With the exception of one fracture with high permeability, all other permeabilities are below 2 mD, primarily concentrated between 0.1 mD and 0.4 mD. This represents a typical low-porosity, low-permeability, tight sandstone reservoir. With the deepening of oil and gas exploration and development in the Xihu Depression, a large amount of seismic and well logging data has been accumulated, and further application of this existing data is urgently needed. Taking this block as an example, the sand body seismic intelligent prediction method based on RF-GAM described in the present invention is used to predict the sand body, including the following steps:

[0050] (1) Collect existing drilling data and seismic data, and extract the sand-to-ground ratio information of fine sand and above from the drilling data, and extract seismic attribute information from the seismic data, wherein the seismic attribute information includes half energy, sum of negative amplitudes, number of zero crossings, root mean square amplitude, threshold value, sum of energy, average negative amplitude, average energy, average positive instantaneous frequency, average instantaneous phase, average amplitude, average positive amplitude, time window, sum of amplitudes, positive and negative sampling ratio, sum of positive amplitudes, maximum amplitude, minimum amplitude, amplitude standard deviation, and arc length;

[0051] (2) The Pearson correlation coefficient was used to analyze the correlation between the seismic attribute information and the sand-to-ground ratio information. Some of the correlation results are shown in Table 1:

[0052] Table 1 Pearson correlation coefficient analysis results

[0053]

[0054] According to the Pearson correlation coefficient analysis results in Table 1, it can be seen that the root mean square amplitude is extremely strongly correlated with the sand-to-ground ratio of fine sand and above, the average energy is strongly correlated with the sand-to-ground ratio of fine sand and above, the arc length is moderately correlated with the sand-to-ground ratio of fine sand and above, and the average negative amplitude and the negative amplitude are both weakly correlated with the sand-to-ground ratio of fine sand and above. Therefore, the root mean square amplitude, arc length and average energy are selected as the correlation factors of the sand-to-ground ratio of fine sand and above.

[0055] (1) Preliminary prediction of sand body is performed using multiple linear regression analysis method: Based on the root mean square amplitude, arc length and average energy, multiple linear regression analysis is performed on the sand-to-ground ratio of fine sand and above, and the analysis results are imported into Direct for visualization, and the following are obtained: Figure 2 The preliminary prediction results of the sand body are shown.

[0056] (2) According to Figure 2 The preliminary prediction results of the sand bodies shown in the figure are as follows: the location where the relevant factors match the sand-to-land ratio is selected (i.e., Figure 2 The seismic attribute information and the sand-to-ground ratio fitting effect are found in the , and the seismic attribute information at this position is used as the label data in the subsequent supervised machine learning, with a total of 83 sample data as the influencing factors of the sand-to-ground ratio.

[0057] (3) A random forest model was established using the selected influencing factors. Through model training, the importance value of each seismic attribute on the sand-to-ground ratio above fine sand was obtained, expressed as the average mean square error; the larger the average mean square error, the more important the factor. After multiple trainings of the random forest, the average value was taken as the importance value of each influencing factor. Based on the average importance value of each influencing factor, six factors, namely root mean square amplitude, average energy, maximum amplitude, average positive amplitude, amplitude standard deviation, and arc length, were selected as important influencing factors affecting the sand-to-ground ratio above fine sand. The results are shown in Table 2:

[0058] Table 2 Results of selection of important influencing factors

[0059]

[0060] (1) The six important influencing factors and the sand-to-ground ratio of fine sand and above obtained in step (5) are used as the modeling data of the generalized additive model to construct a regression model (in this embodiment, the f function of the generalized additive model adopts a spline function). The data points of the entire area are used as the prediction data for constructing the model, and the sand-to-ground ratio of the entire study area is predicted. The results are as follows: Figure 3 shown.

[0061] (2) Use the majority filter algorithm to eliminate some noise points in the river channel sand body in the softened data: according to all the pixels in the image, perform mean processing on the pixels inside and on the edge of the river channel, and make the data around N*M pixels become the values ​​after mean processing, where the kernel size is set to a 5*5 matrix.

[0062] (3) Convert the pixel points in the filtered image into data points of sand-to-ground ratio attributes and import them into Direct software for visualization. The results are as follows: Figure 4 shown.

[0063] according to Figure 2-4It can be seen that the visualization results after the generalized additive model and filtering processing can clearly show the planar distribution of deep-sea river channels, which can provide reliable data for oil and gas field exploration and development. Compared with the existing technology, the present invention has a significant improvement.

[0064] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as a preferred embodiment as above, it is not intended to limit the present invention. Any technician familiar with this profession can make some changes or modifications to equivalent embodiments of the technical contents disclosed above without departing from the scope of the technical solution of the present invention. However, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A sand body seismic intelligent prediction method based on RF-GAM, characterized by: The following steps are involved: S1: collecting drilling data and seismic data of the target block, and extracting sand-to-soil ratio information from the drilling data, and extracting seismic attribute information from the seismic data; S2: analyzing the correlation between the seismic attribute information and the sand-to-ground ratio information, and screening out relevant factors of the sand-to-ground ratio from the seismic attribute information according to the analysis result of the correlation; S3: Based on the relevant factors, a multivariate linear regression analysis method is used to predict the sand body to obtain a preliminary prediction result of the sand body; S4: Based on the preliminary prediction result of the sand body, screening the position where the relevant factors match the sand-to-formation ratio, the matching position being the position where the fitting effect of the seismic attribute information and the sand-to-formation ratio in the preliminary prediction result of the sand body meets the fitting threshold, and using the seismic attribute information at this position as the influencing factor of the sand-to-formation ratio; S5: Based on the influencing factors, use random forest to screen and obtain important influencing factors; S6: Based on the important influencing factors, a generalized additive model is used to perform attribute fusion, and the result of the attribute fusion is visualized to obtain a final prediction result of the sand body.

2. The sand body seismic intelligent prediction method based on RF-GAM according to claim 1 is characterized in that: In step S1, the sand-to-ground ratio information is the sand-to-ground ratio information of sand bodies above fine sand.

3. The sand body seismic intelligent prediction method based on RF-GAM according to claim 1 is characterized in that: In step S2, when analyzing the correlation between the seismic attribute information and the sand-to-ground ratio information, the Pearson correlation coefficient is used for correlation analysis.

4. The sand body seismic intelligent prediction method based on RF-GAM according to any one of claims 1 to 3, characterized in that: Before visualizing the result of the attribute fusion, the method further includes filtering the result of the attribute fusion to eliminate noise.

5. The sand body seismic intelligent prediction method based on RF-GAM according to claim 4 is characterized in that: During filtering, any one or more filtering algorithms among median filtering, mean filtering, bilateral filtering, fuzzy filtering, and majority filtering are used for filtering.

Images