Oil and gas development favorable area prediction method based on Winsored particle size mean value curve simulation

Abstract

The invention discloses an oil and gas development favorable area prediction method based on Winsored particle size mean curve simulation, and the method comprises the following steps: based on laser particle size experiment data, employing a Winsored algorithm to process rock particle size data, removing abnormal interference values, and simulating a well hole particle size mean longitudinal curve through inputting logging response parameters by means of an XLSTM-Transform hybrid neural network model; selecting a shear modulus density and a longitudinal and transverse wave velocity ratio as sensitive attribute parameters of a granularity mean value in combination with seismic data, and forming two sets of three-dimensional attribute data volumes in combination with waveform structure inversion; according to the method, model training is carried out on the nonlinear relation between the particle size mean value and the seismic sensitive attribute parameters based on a machine learning method, and finally a Winsored particle size mean value inversion body is formed, so that the spatial distribution rule of a relatively large particle size reservoir can be defined, and effective prediction of a development favorable area is realized.

Description

Technical Field

[0001] This invention belongs to the field of oil and gas field development technology, and in particular relates to a method for predicting favorable oil and gas development areas based on Winsorized particle size mean curve simulation. Background Technology

[0002] In the development of highly heterogeneous offshore gas reservoirs, accurate prediction of favorable development areas is crucial for well placement. Traditional methods mainly rely on well logging curves and seismic data, but for low-permeability reservoirs with grain size-controlled properties, conventional well logging response characteristics are not obvious, resulting in insufficient prediction accuracy for relatively high-permeability reservoirs.

[0003] In existing technologies, particle size analysis is mostly limited to the laboratory scale, making it difficult to achieve continuous prediction across wells, and there is a lack of effective multi-scale data fusion methods. Therefore, there is an urgent need for an innovative prediction method that can integrate core experimental, logging, and seismic data. Summary of the Invention

[0004] The problem this invention aims to solve is to provide a method for predicting favorable oil and gas development areas based on Winsorized particle size mean curve simulation. This method solves the problems of low prediction accuracy and poor data continuity in the prior art.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by this invention is: a method for predicting favorable oil and gas development areas based on Winsorized particle size mean curve simulation, comprising the following steps: S1: Laser particle size analysis was performed on core samples from the borehole to obtain rock particle size distribution data at different depths of the drilled well. S2: The Winsorized mean algorithm is used to process the particle size data and calculate the mean particle size that is resistant to outlier interference. S3: Establish a Winsorized particle size mean sample database; S4: Construct an XLSTM-Transformer hybrid neural network model, using conventional logging curves as input and the Winsorized average particle size as output for training, to obtain a prediction model that can simulate the longitudinal Winsorized average particle size curve of continuous wellbore. S5: Select the geophysical elastic parameters shear modulus density μρ and P-wave / S-wave velocity ratio Vp / Vs that are sensitive to the Winsorized grain size mean. Combine them with the pre-stack 3D seismic data volume and form two elastic parameter inversion 3D attribute data volumes that are closely related to the Winsorized grain size mean through waveform structure inversion. S6: Establish a nonlinear correlation training model between the Winsorized average particle size curve and the shear modulus density μρ and the P-wave / S-wave velocity ratio Vp / Vs. Simultaneously write the inversion volumes of the shear modulus density μρ and the P-wave / S-wave velocity ratio Vp / Vs into the training model to form a Winsorized average particle size inversion volume. Through the Winsorized average particle size inversion volume, clarify the spatial distribution pattern of relatively coarse-grained reservoirs and achieve effective prediction of favorable development areas.

[0006] Furthermore, in S1, the measurement range of the laser particle size test is 0.02 μm to 2000 μm.

[0007] Furthermore, step S2 includes the following steps: S21: Determine the data boundaries for Windsorization, and set the minimum threshold a and the maximum threshold b; S22: Perform Windsorization on the original particle size sample data; S23: Calculate the average Winsorized particle size based on the Winsorized dataset.

[0008] Furthermore, in step S21, the minimum threshold a = d(k+1) and the maximum threshold b = d(k-1) are determined by quantile truncation of the sorted sample data {d1, d2, ..., dn}, where d1 < d2 < ... < dn. The proportion of extreme values ​​truncated accounts for 5% to 15% of the total sample data. Here, a is the minimum threshold; b is the maximum threshold; k is the boundary data point; and d is the particle size, which is obtained from laser particle size distribution experiments to determine the proportion of particles of different sizes.

[0009] Furthermore, in step S22, the data Windsorization process formula is as follows: , Where di is the i-th data point in the original particle size sample data.

[0010] Furthermore, in step S23, the formula for calculating the average Winsorized particle size is as follows: ; Where n is the total number of samples; Let be the value of the i-th sample after Windsorization. If the value of the sample is less than a, then replace it with a; if the value of the sample is greater than b, then replace it with b. This is the Winsorized average value.

[0011] Furthermore, in S4, the conventional logging curve includes at least one of the natural gamma curve, neutron porosity curve, and density curve; the XLSTM network in the XLSTM-Transformer hybrid neural network model is used to capture the local trend and periodic changes of the vertical grain size mean, and the Transformer network is used to capture the long-range dependencies in the formation sequence.

[0012] Furthermore, in S5, a combined Theil-Sen regression and Spearman correlation analysis method is used to screen out the shear modulus density μρ and the P-wave / S-wave velocity ratio Vp / Vs from a variety of pre-stack seismic elastic parameters that have the strongest correlation with the mean Winsorized particle size.

[0013] Furthermore, in S6, the training model is established using an artificial neural network learning method based on Petrel software. The artificial neural network learning method is a self-learning process that establishes a nonlinear mapping relationship between the mean Winsorized particle size and the shear modulus density μρ and the P-wave / S-wave velocity ratio Vp / Vs through error backpropagation and gradient optimization.

[0014] Furthermore, the present invention also provides a system for predicting favorable oil and gas development areas based on Winsorized particle size mean curve simulation, which runs the above-mentioned method for predicting favorable oil and gas development areas based on Winsorized particle size mean curve simulation.

[0015] The advantages and positive effects of this invention are: 1. This invention is based on laser particle size experimental data. It uses the Winsorized algorithm to process rock particle size data, removes abnormal interference values, and uses an XLSTM-Transformer hybrid neural network model to simulate the longitudinal curve of the mean particle size in the wellbore by inputting well logging response parameters. Combined with seismic data, it selects the shear modulus density and the ratio of P-wave to S-wave velocity as sensitive attribute parameters of the mean particle size. Combined with waveform structure inversion, it forms two sets of three-dimensional attribute data volumes. Based on machine learning methods, it trains a model on the nonlinear relationship between the mean particle size and the seismic sensitive attribute parameters, and finally forms a Winsorized mean particle size inversion volume, which can clarify the spatial distribution law of relatively coarse-grained reservoirs and realize the effective prediction of favorable development areas.

[0016] 2. This invention provides an innovative prediction method that integrates core experiments, well logging, and seismic data, resulting in high prediction accuracy and strong data continuity. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall process of an embodiment of the present invention.

[0018] Figure 2 Example D: Rock particle size distribution obtained from laser particle size analysis of the gas field; Figure 3 : Rock grain size distribution map after processing by the Winsorized mean algorithm in the example; Figure 4 : Vertical distribution diagram of the mean Winsorized particle size curve of a single well in the example; Figure 5 : Relationship between mean Winsorized particle size and shear modulus density in the example; Figure 6 : Relationship between mean Winsorized particle size and P-wave / S-wave velocity ratio in the embodiment; Figure 7 Longitudinal inversion profile of the Winsorized mean particle size of the main gas group in the D gas field.

[0019] Figure 8 Winsorized mean particle size inversion plan view of the main gas group in the D gas field. Detailed Implementation

[0020] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] The embodiments of the present invention will be further described below with reference to the accompanying drawings: like Figure 1 As shown, the method for predicting favorable oil and gas development areas based on Winsorized particle size mean curve simulation includes the following steps.

[0022] S1: Based on borehole core samples, laser grain size analysis is performed to obtain rock grain size distribution data at different drilled depths. The borehole core samples can accurately locate the depth of gas-bearing target layers under complex marine conditions. Specifically, accurately locating the depth of gas-bearing target layers overcomes the depth errors caused by conventional core placement and better matches well logging and seismic depth information.

[0023] The aforementioned laser particle size analysis is a particle size distribution measurement technique based on the principle of laser diffraction. The measurement range is typically 0.02μm-2000μm, which can cover the particle size of reservoir rocks such as sandstone and mudstone.

[0024] Rock particle size distribution data are obtained through laser particle size analysis. The resulting particle size distribution curves are classified as single-peak, double-peak, and multi-peak. The skewness of the particle size distribution curve can reflect the main coarseness of the rock particles. A left skewness of the distribution curve reflects that the rock is dominated by relatively fine particles, while a right skewness reflects that the rock is dominated by relatively coarse particles.

[0025] S2: The Winsorized mean algorithm is used to process the particle size data and calculate a particle size mean that is resistant to outlier interference. The Winsorized mean algorithm is a robust data processing method that reduces the impact of extreme values ​​on statistical parameters such as the mean and variance by replacing extreme values ​​with values ​​corresponding to certain quantiles, while keeping the overall data volume unchanged. The causes of these extreme values ​​are varied. Maximum values ​​are generally due to the presence of cementing particles or gravel in the rock, while minimum values ​​are generally caused by particle fragmentation. Therefore, processing these extreme values ​​can better preserve the component signals of the main rock particle size. Specifically, extreme values ​​in the particle size data generally account for 5%-15% of the total particle size data. Given sample data of {d1, d2, ..., dn}, where d1 < d2 < ... < dn, preferably, S2 includes the following steps: S21: Define the data boundaries to be Windsorized: a=d(k+1), b=d(k-1); S22: Data Windsorization Processing: ; S23: Winsorized average particle size calculation: ; In the formula, di is the i-th data point of the original particle size sample data; n is the total number of samples; k is the boundary data point; a is the minimum threshold; b is the maximum threshold; and is the value of the i-th sample after Windsorization. If the sample is less than a, it is replaced with a; if it is greater than b, it is replaced with b. The value is the Winsorized average, and d is the particle size, which is obtained from laser particle size distribution experiments to determine the percentage of particles of different sizes.

[0026] S3: Establish a Winsorized particle size mean sample database.

[0027] S4: Construct an XLSTM-Transformer hybrid neural network model, using conventional logging curves as input and Winsorized grain size mean as output for training, to achieve continuous simulation of vertically discrete experimental data from the wellbore. Specifically, XLSTM is a Long Short-Term Memory network, capable of capturing local trends and periodic changes in the vertical grain size mean, while Transformer is an attention-based sequence modeler, capable of capturing the correspondence between two sets of sand bodies. The deep learning architecture formed by the combination of these two technologies can calculate a continuous grain size mean curve with smoother data shape and stronger noise resistance.

[0028] Preferably, the conventional logging curves mentioned above include, but are not limited to, curves for natural gamma (GR), neutron porosity (CNCF), and density (DEN). In actual implementation, these curves should be combined with the actual drilling and data collection in the well.

[0029] S5: The geophysical elastic parameters sensitive to the Winsorized grain size mean, shear modulus density μρ and P-wave / S-wave velocity ratio Vp / Vs, are selected. Combined with pre-stack 3D seismic data volumes, waveform structure inversion is used to generate 3D attribute data volumes for two elastic parameters closely related to the Winsorized grain size mean. Specifically, the pre-stack seismic elastic parameters sensitive to the Winsorized grain size mean are identified using a combined Theil-Sen regression and Spearman correlation method, which clarifies that shear modulus density μρ and P-wave / S-wave ratio Vp / Vs are sensitive parameters to the Winsorized grain size mean.

[0030] Among them, the method of combining Theil-Sen regression and Spearman correlation is effective in detecting the monotonicity of data, with Spearman correlation focusing on testing the monotonicity of data and showing good results in detecting whether there is a monotonic relationship (including nonlinearity) between variables. Theil-Sen regression is a linear regression method, which is strong against outliers and noise, and is very suitable for situations where there is limited data from a few wells in the study area. The combination of the two methods can clearly identify the pre-stack parameters that change stably with grain size.

[0031] The shear modulus density parameter μρ is a parameter that can identify the clay content, density variation, and degree of consolidation of a reservoir.

[0032] The longitudinal wave ratio Vp / Vs is the ratio of longitudinal wave velocity to transverse wave velocity, which can indicate porosity, clay content, and fluid characteristics.

[0033] S6: Based on Petrel software, an ANN learning method is used to establish a nonlinear correlation training model between the mean grain size curve and two pre-stack seismic elastic parameters: shear modulus density and P-wave / S-wave velocity ratio. The inversion volumes of the two pre-stack seismic elastic parameters are simultaneously written into the training model to form a Winsorized mean grain size inversion volume. Through this inversion volume, the spatial distribution pattern of relatively coarse-grained reservoirs can be clearly identified. Based on this, favorable areas for economic development can be searched, and effective prediction of favorable development areas can be achieved.

[0034] The preferred method is the ANN (Artificial Neural Network) learning method, which is a nonlinear mapping algorithm that mimics the connection mode of neurons in the human brain. Essentially, it is a self-learning process that automatically establishes a nonlinear relationship between input and output through error backpropagation and gradient optimization of sample data.

[0035] The present invention will now be described in detail with reference to specific embodiments: In this embodiment, the D gas field is developed by gravity flow sedimentation, with the lithology mainly consisting of fine sandstone. The reservoir clay content is mostly greater than 15%, which is a typical fine-grained, high-clay, low-permeability reservoir. The reservoir properties in the study area are controlled by the grain size of the sedimentary rocks. The coarser the reservoir grain size, the better the physical properties.

[0036] S1: Based on the core sample of the wellbore, laser particle size analysis was performed to obtain rock particle size distribution data at different depths of the drilled well.

[0037] S2: Rock Grain Size Analysis: Analysis of laser grain size data from four wells in the D gas field revealed that the average grain size distribution of the reservoir ranged from 4.23 to 7.28 φ, with the main distribution range being 4.89 to 6.27 φ. Figure 2 As shown.

[0038] S3: Establishment of the Winsorized Particle Size Mean Sample Database: Based on the particle size mean distribution characteristics of this gas field, the Winsorized algorithm was used to process the laser particle size data. 5% of the minimum and maximum values ​​were selected for Winsorization. The processed particle size mean distribution ranged from 4.54 to 6.87 φ. Based on this, a particle size mean sample database for Gas Field D was established, as follows... Figure 3 As shown.

[0039] S4: Continuous Simulation of Mean Grain Size Data: Statistical analysis of data from the same depth revealed a high correlation between natural gamma, neutron porosity, density, and mean grain size in the D gas field, with relatively complete data collection. Natural gamma characterizes the variation in coarse and fine grain content in sediments; smaller natural gamma values ​​indicate higher coarse grain content. Neutron porosity can characterize the size of reservoir space to some extent; reservoirs with high coarse grain content have lower neutron porosity and larger reservoir space. Density also reflects grain size and reservoir space size; lower density values ​​indicate relatively coarser sediment grains and more developed reservoir space. Using an XLSTM-Transformer hybrid neural network model, the input natural gamma, neutron porosity, and density parameters were trained to obtain the mean grain size curves for four single wells in the gas field, as shown below. Figure 4 As shown.

[0040] S5: Optimization of Geophysical Elastic Parameters: First, the well depth is converted to the time domain, and pre-stack sampling data of the target layer is extracted to ensure that each well depth has matching attribute data. Univariate robust ordination is performed using the Spearman correlation method to identify the elastic parameters with the strongest monotonic relationship to the grain size mean. Then, Theil-Sen regression is used to determine the robust trend slope of the grain size mean, selecting the elastic parameters with the most robust relationship to the grain size mean: shear modulus density μρ and P-wave / S-wave velocity ratio Vp / Vs. These are combined with waveform structure inversion to form corresponding three-dimensional attribute data volumes, such as... Figure 5 , Figure 6 As shown.

[0041] S6: Construction of Winsorized Grain Size Mean Inversion Body: An ANN learning method was used to construct a training model for the grain size mean, shear modulus density, and P-wave / S-wave velocity ratio. The inversion bodies for both parameters were simultaneously input into the existing nonlinear correlation training model to establish the Winsorized grain size mean inversion body. Through the well profile characteristics of this inversion body, it can be found that the relatively coarse-grained reservoirs in Gas Field D are mainly distributed in the lower part of the reservoir. According to the average attribute planar distribution map of the inversion body, the relatively coarse-grained reservoirs are concentrated in the northwest and south of the gas field, which are favorable reservoirs in the gas field. Figure 7 , Figure 8 As shown.

[0042] The advantages and positive effects of this invention are: 1. This invention is based on laser particle size experimental data. It uses the Winsorized algorithm to process rock particle size data, removes abnormal interference values, and uses an XLSTM-Transformer hybrid neural network model to simulate the longitudinal curve of the mean particle size in the wellbore by inputting well logging response parameters. Combined with seismic data, it selects the shear modulus density and the ratio of P-wave to S-wave velocity as sensitive attribute parameters of the mean particle size. Combined with waveform structure inversion, it forms two sets of three-dimensional attribute data volumes. Based on machine learning methods, it trains a model on the nonlinear relationship between the mean particle size and the seismic sensitive attribute parameters, and finally forms a Winsorized mean particle size inversion volume, which can clarify the spatial distribution law of relatively coarse-grained reservoirs and realize the effective prediction of favorable development areas.

[0043] 2. This invention provides an innovative prediction method that integrates core experiments, well logging, and seismic data, resulting in high prediction accuracy and strong data continuity.

[0044] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for predicting favorable oil and gas development areas based on Winsorized particle size mean curve simulation, characterized in that: Includes the following steps, S1: Laser particle size analysis was performed on core samples from the borehole to obtain rock particle size distribution data at different depths of the drilled well. S2: The Winsorized mean algorithm is used to process the particle size data and calculate the mean particle size that is resistant to outlier interference. S3: Establish a Winsorized particle size mean sample database; S4: Construct an XLSTM-Transformer hybrid neural network model, using conventional logging curves as input and the Winsorized average particle size as output for training, to obtain a prediction model that can simulate the longitudinal Winsorized average particle size curve of continuous wellbore. S5: Select the geophysical elastic parameters shear modulus density μρ and P-wave / S-wave velocity ratio Vp / Vs that are sensitive to the Winsorized grain size mean. Combine them with the pre-stack 3D seismic data volume and form two elastic parameter inversion 3D attribute data volumes that are closely related to the Winsorized grain size mean through waveform structure inversion. S6: Establish a nonlinear correlation training model between the Winsorized average particle size curve and the shear modulus density μρ and the P-wave / S-wave velocity ratio Vp / Vs. Simultaneously write the inversion volumes of the shear modulus density μρ and the P-wave / S-wave velocity ratio Vp / Vs into the training model to form a Winsorized average particle size inversion volume. Through the Winsorized average particle size inversion volume, clarify the spatial distribution pattern of relatively coarse-grained reservoirs and achieve effective prediction of favorable development areas.

2. The method for predicting favorable oil and gas development areas based on Winsorized particle size mean curve simulation according to claim 1, characterized in that: In S1, the measurement range of the laser particle size test is 0.02 μm to 2000 μm.

3. The method for predicting favorable oil and gas development areas based on Winsorized particle size mean curve simulation according to claim 1 or 2, characterized in that: S2 includes the following steps: S21: Determine the data boundaries for Windsorization, and set the minimum threshold a and the maximum threshold b; S22: Perform Windsorization on the original particle size sample data; S23: Calculate the average Winsorized particle size based on the Winsorized dataset.

4. The method for predicting favorable oil and gas development areas based on Winsorized particle size mean curve simulation according to claim 3, characterized in that: In step S21, the minimum threshold a = d(k+1) and the maximum threshold b = d(k-1) are determined by quantile truncation of the sorted sample data {d1, d2, ..., dn}, where d1 < d2 < ... < dn. The proportion of extreme values ​​truncated accounts for 5% to 15% of the total sample data. Here, a is the minimum threshold; b is the maximum threshold; k is the boundary data point; and d is the particle size, which is obtained from laser particle size distribution experiments to determine the proportion of particles of different sizes.

5. The method for predicting favorable oil and gas development areas based on Winsorized particle size mean curve simulation according to claim 4, characterized in that: In step S22, the data Windsorization process formula is as follows: , Where di is the i-th data point in the original particle size sample data.

6. The method for predicting favorable oil and gas development areas based on Winsorized particle size mean curve simulation according to claim 4 or 5, characterized in that: In step S23, the formula for calculating the average Winsorized particle size is as follows: ; Where n is the total number of samples; Let be the value of the i-th sample after Windsorization. If the value of the sample is less than a, then replace it with a; if the value of the sample is greater than b, then replace it with b. This is the Winsorized average value.

7. The method for predicting favorable oil and gas development areas based on Winsorized particle size mean curve simulation according to claim 1 or 2, characterized in that: In S4, the conventional logging curve includes at least one of the natural gamma curve, neutron porosity curve, and density curve; the XLSTM network in the XLSTM-Transformer hybrid neural network model is used to capture the local trend and periodic changes of the vertical grain size mean, and the Transformer network is used to capture the long-range dependencies in the formation sequence.

8. The method for predicting favorable oil and gas development areas based on Winsorized particle size mean curve simulation according to claim 1 or 2, characterized in that: In S5, a combination of Theil-Sen regression and Spearman correlation analysis was used to screen out the shear modulus density μρ and the P-wave / S-wave velocity ratio Vp / Vs from a variety of pre-stack seismic elastic parameters that have the strongest correlation with the mean Winsorized particle size.

9. The method for predicting favorable oil and gas development areas based on Winsorized particle size mean curve simulation according to claim 1 or 2, characterized in that: In S6, the training model is established using the Petrel software and an artificial neural network learning method. The artificial neural network learning method is a self-learning process that establishes a nonlinear mapping relationship between the mean Winsorized particle size and the shear modulus density μρ and the P-wave / S-wave velocity ratio Vp / Vs through error backpropagation and gradient optimization.

10. A system for predicting favorable oil and gas development areas based on Winsorized particle size mean curve simulation, characterized in that: The method for predicting favorable oil and gas development areas based on Winsorized particle size mean curve simulation, as described in any one of claims 1 to 9, is implemented.

Images