Seismic inversion initial model establishment method and device based on lithofacies probability inverse mapping

Abstract

The invention relates to the field of oil and gas exploration, in particular to a method and a device for establishing a seismic inversion initial model based on lithofacies probability inverse mapping. The invention discloses a lithofacies probability inverse mapping-based seismic inversion initial model establishment method, which comprises the following steps of: collecting and processing data to obtain post-stack seismic data and a lithofacies curve; a plurality of seismic attribute bodies are obtained from the post-stack seismic data, and a pseudo-well curve is obtained in combination with the lithofacies curve; training the neural network model by using the pseudo-well curve, and outputting a lithofacies classification result and a lithofacies probability body; analyzing the lithofacies classification result and the lithofacies probability body, and establishing a posterior probability density space of the physical property parameters; and on the basis of the posterior probability density space and the probability density distribution, utilizing a Monte Carlo algorithm and rejection-acceptance sampling to generate a physical property parameter combination meeting conditions, and obtaining a seismic inversion initial model. According to the method, reservoir characterization can be more accurately carried out, so that the seismic inversion initial model and the well point are better in consistency.

Description

Technical Field

[0001] The invention relates to the field of oil and gas exploration, and in particular to a method and a device for establishing a seismic inversion initial model based on lithofacies probability inverse mapping. Background Art

[0002] The inversion of elastic parameters or physical parameters using prestack seismic data is a key step in the oil and gas exploration and development stage. Its purpose is to provide a reliable basis for the prediction and evaluation of oil and gas reservoirs through accurate estimation of underground structures and physical parameters. Accurate elastic parameter inversion can help explorers identify potential oil and gas reservoirs, optimize well location design, reduce exploration risks, and improve the development efficiency of oil and gas fields.

[0003] In seismic inversion, the construction of the initial model is crucial, especially the construction of the initial model has a decisive influence on the inversion results. The core of seismic inversion is to match the observed seismic data with the forward modeling data through iterative optimization to infer the physical properties of the underground medium. However, the model may be inaccurate due to the failure to consider the lateral changes of the strata, which seriously affects the quality of underground structural imaging and the accuracy and reliability of the inversion results. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method and device for establishing an initial model of seismic inversion based on lithofacies probability inverse mapping.

[0005] The technical solution adopted by the present invention to solve the technical problem is: a method for establishing a seismic inversion initial model based on lithofacies probability inverse mapping, comprising the following steps:

[0006] S1. Collect and process oil test data, seismic data, well logging data and stratigraphic data to obtain post-stack seismic data and lithofacies curves;

[0007] S2, acquiring a plurality of seismic attribute bodies from the post-stack seismic data, and obtaining a pseudo-well curve of each seismic attribute body in combination with the lithofacies curve;

[0008] S3, using the pseudo-well curve to train the neural network model, using the trained neural network model in combination with the seismic data to output lithofacies classification results and lithofacies probability bodies;

[0009] S4, analyzing the lithofacies classification results and the lithofacies probability body to obtain probability density distributions of different physical property parameters and establish a posterior probability density space of the physical property parameters;

[0010] S5. Based on the posterior probability density space and the probability density distribution, the seismic data are traversed, and a combination of physical parameters that meets the conditions is generated by using a Monte Carlo algorithm and rejection-acceptance sampling to obtain the initial seismic inversion model.

[0011] Preferably, the step S1 comprises:

[0012] Collecting the oil testing data, seismic data, well logging data and stratum data in the work area;

[0013] Preprocessing the seismic data, the well logging data and the stratigraphic data to obtain the time-depth relationship after the well earthquake; obtaining the distribution of physical property parameters through the time-depth relationship;

[0014] According to the distribution of the physical property parameters, the lithofacies of the oil well are divided to obtain the lithofacies curve of the upper well;

[0015] The seismic data are offset and stacked to obtain the post-stack seismic data.

[0016] Preferably, the step S2 comprises:

[0017] Based on the post-stack seismic data, obtaining a seismic attribute volume according to preset conditions;

[0018] According to the inherent resolution of the post-stack seismic data, a scale coarsening method is used to match the resolution of the lithofacies curve with the inherent resolution to obtain the pseudo-well curve.

[0019] Preferably, the seismic attribute volume comprises an instantaneous amplitude attribute volume, an instantaneous frequency attribute volume and a sweet spot attribute volume.

[0020] Preferably, the step S3 comprises:

[0021] Acquire the pseudo-well curves of different seismic attribute bodies at the well locations of the oil wells to obtain a training set;

[0022] The neural network model is trained using a supervised learning algorithm and the training set; the seismic data is classified using the trained neural network model to obtain the lithofacies classification result; the lithofacies classification result includes several lithofacies classifications; each lithofacies classification is calculated in combination with an inverse distance weighted algorithm to obtain the lithofacies probability body.

[0023] Preferably, the step S4 comprises:

[0024] Establish rock physics modeling according to the lithofacies classification results, obtain well curves, and determine target parameters of the seismic inversion initial model;

[0025] According to the lithofacies curve and the target parameters, statistical rock physical analysis is performed on different lithofacies to obtain the histogram distribution and the probability density distribution of different physical property parameters in different lithofacies;

[0026] The posterior probability density space is established according to the histogram distribution and the probability density distribution.

[0027] Preferably, the probability density distribution comprises an N-dimensional probability density distribution;

[0028] The posterior probability density space includes an N-dimensional physical property parameter space, covering the valid value ranges of N physical property parameters;

[0029] Wherein N is the number of physical property parameters involved in the statistical rock physics analysis.

[0030] Preferably, the step S5 comprises:

[0031] Based on the posterior probability density space and according to the probability density distribution, the posterior probabilities of different lithofacies for the physical property parameter combinations corresponding to each point in the posterior probability density space are calculated and normalized;

[0032] Traversing the posterior probability of each sampling point in the seismic data to obtain the lithofacies probability ratio of each sampling point in the seismic data;

[0033] Generate M physical property parameter combinations that meet the preset parameter distribution according to the Monte Carlo algorithm and the probability density function;

[0034] According to the rejection-acceptance sampling algorithm and the lithofacies probability ratio of the current sampling point in the seismic data, m physical property parameter combinations that meet the conditions are selected from the M physical property parameter combinations within a preset tolerance;

[0035] The m physical property parameter combinations are used as an initial model for direct inversion of physical property parameters.

[0036] Preferably, each sampling point in the seismic data includes each point recorded in the seismic trace, and the sampling rate is 2 ms.

[0037] A device for establishing an initial model for seismic inversion based on inverse mapping of lithofacies probability, the device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the method for establishing an initial model for seismic inversion based on inverse mapping of lithofacies probability as described in any one of the above items are implemented.

[0038] The implementation of the present invention has the following beneficial effects:

[0039] The present invention is based on oil test data, seismic data, well logging data and stratigraphic data, obtains post-stack seismic data and lithofacies curves for seismic inversion to obtain pseudo-well curves, and then combines the neural network model to obtain lithofacies classification results and lithofacies probability bodies. Further analysis is performed to establish the posterior probability density space of physical property parameters. Finally, the Monte Carlo algorithm and rejection-acceptance sampling are used to generate a combination of physical property parameters that meet the conditions, and the seismic inversion initial model is obtained. The present invention can more accurately characterize the reservoir through lithofacies probability inverse mapping, so that the consistency between the initial seismic inversion model and the reservoir characterization at the well location is better, and the spatial variation law of the parameters is consistent with the actual lithofacies distribution, which has more practical geological significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0041] Figure 1 It is a flowchart of a method for establishing a seismic inversion initial model based on lithofacies probability inverse mapping in one embodiment;

[0042] Figure 2 is a two-dimensional probability density analysis result between the same phase and different physical property parameters in one embodiment;

[0043] Figure 3 is the probability density difference of different lithofacies under a certain combination of physical property parameters in an embodiment;

[0044] Figure 4 is a probability volume of different lithofacies in one embodiment;

[0045] Figure 5 An initial model of porosity obtained by seismic inversion of lithofacies probability inverse mapping in one embodiment;

[0046] Figure 6 An initial seismic inversion model of pore structure obtained by lithofacies probability inverse mapping in one embodiment;

[0047] Figure 7 The initial model of physical property parameters (porosity) established according to the conventional interpolation method;

[0048] Figure 8 FIG. 1 is an initial seismic inversion model (porosity) established according to the lithofacies probability inverse mapping method in one embodiment. DETAILED DESCRIPTION

[0049] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, specific embodiments of the present invention are now described in detail with reference to the accompanying drawings.

[0050] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention. However, it should be clear to those skilled in the art that the present invention may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present invention.

[0051] The method for establishing a seismic inversion initial model based on lithofacies probability inverse mapping provided by an embodiment of the present invention comprises the following steps:

[0052] S1. Collect and process oil test data, seismic data, logging data and stratigraphic data to obtain post-stack seismic data and lithofacies curves.

[0053] Collect and process oil test data, seismic data, well logging data and stratigraphic data to obtain post-stack seismic data and lithofacies curves. Oil test data processing includes analyzing basic well data and oil test results. Seismic data processing uses deconvolution, stacking and migration imaging techniques to improve resolution and signal-to-noise ratio to obtain post-stack seismic data. Well logging data processing involves data cleaning and correction to solve geological problems. Stratigraphic data processing accurately determines the strata through comprehensive calibration. Ultimately, lithofacies curves are generated from these data to provide key geological information for oil and gas exploration.

[0054] S2. Obtain several seismic attribute bodies from the post-stack seismic data, and obtain a pseudo-well curve for each seismic attribute body in combination with the lithofacies curve.

[0055] First, through the seismic attribute analysis technology, key attributes such as amplitude and frequency are extracted from the seismic data. These attributes reflect the underground geological structure and rock physical properties. Secondly, the lithofacies curves in the logging data are used. These curves show the physical properties of rocks, such as porosity and permeability, and provide geological basis for the seismic attributes. Finally, through the seismic trace matching technology, the seismic attributes and lithofacies curves are combined to generate the corresponding pseudo-well curves, ensuring the accuracy and reliability of the seismic attributes.

[0056] S3. Use pseudo-well curves to train the neural network model, use the trained neural network model in combination with seismic data, and output lithofacies classification results and lithofacies probability bodies.

[0057] First, the neural network is trained using pseudo-well curves to enable the model to learn the relationship between seismic signals and lithofacies characteristics; second, the model extracts seismic data features and performs deep learning to capture the temporal and spatial characteristics of the data; finally, the model outputs lithofacies classification and probability bodies, providing the possibility of lithofacies occurrence and providing decision support for oil and gas exploration. By integrating seismic data processing and machine learning technology, the accuracy and efficiency of lithofacies identification are improved.

[0058] S4. Analyze the lithofacies classification results and lithofacies probability bodies, obtain the probability density distribution of different physical property parameters, and establish the posterior probability density space of physical property parameters.

[0059] The analysis of lithofacies classification results and lithofacies probability bodies involves establishing a lithofacies probability distribution model, loading a three-dimensional structural model and importing lithofacies data, and combining wave impedance data to determine the probability value of lithofacies data. Using the Bayesian theoretical framework and Metropolis algorithm, combined with geological statistical prior information, the physical property parameters are inverted to obtain the posterior probability density space.

[0060] Furthermore, the Hamiltonian Monte Carlo method, Markov chain Monte Carlo sampling, ensemble learning method, Bayesian network, multi-point geostatistical simulation method, SNESIM algorithm, GENESIM algorithm, back propagation neural network, support vector machine regression (R-SVM), multiple regression analysis (MRA), support vector machine classification (C-SVM), Bayesian stepwise discriminant (BAYSD) and other frameworks and algorithms can be used to obtain the posterior probability density space.

[0061] S5. Based on the posterior probability density space and probability density distribution, the seismic data are traversed, and the Monte Carlo algorithm and rejection-acceptance sampling are used to generate a combination of physical parameters that meet the conditions to obtain the initial model of seismic inversion.

[0062] Through the Monte Carlo algorithm and rejection-acceptance sampling, qualified physical parameter combinations are extracted from seismic data. The Monte Carlo method simulates the distribution of underground rock physical parameters through random sampling, while rejection-acceptance sampling selects samples from the proposed distribution to ensure that they meet the target distribution. This process combines seismic data with geological statistical characteristics to generate possible physical parameter combinations, construct an initial underground rock physical property model, and provide a starting point for seismic inversion. This method can effectively handle high-dimensional and nonlinear problems and improve inversion accuracy and efficiency.

[0063] Specifically, the initial seismic inversion model is a set of data sets. Displaying the initial seismic inversion model in the form of images can intuitively display the underground structure and rock properties, thereby better analyzing the geological structure and geological resource distribution.

[0064] The present invention combines statistical rock physics analysis with seismic lithofacies identification results based on machine learning, and applies the lithofacies probability inverse mapping method to realize the allocation of different physical property parameters corresponding to different lithofacies, thereby adding lateral change information of the strata to the seismic inversion initial model, and realizing the establishment of a high-precision seismic inversion initial model.

[0065] Specifically, lithofacies probability inverse mapping is a method for constructing an initial model of reservoir lithofacies. This process combines lithofacies classification information with probabilistic statistical techniques to infer the distribution of elastic parameters and physical property parameters corresponding to each lithofacies type, thereby constructing a physical model of the reservoir. This method can more accurately characterize the spatial distribution of reservoir lithofacies and their uncertainty, and provide a reliable initial model for seismic data interpretation and reservoir evaluation, thereby improving the accuracy of reservoir feature recognition and characterization.

[0066] In some executable embodiments, step S1 includes:

[0067] Collect oil testing data, seismic data, well logging data and stratigraphic data within the work area.

[0068] "Working area" refers to a specific geographical area for oil and gas exploration and development, including oil and gas accumulation in a single trap. It is not only a combination of reservoirs with the same geological structural background, but also a specific unit for management and operation. The oil and gas reservoirs in the working area share the same pressure system, and a professional team is responsible for their exploration, development and production activities. Collecting various types of data and information on a working area basis facilitates global oil, gas and water well production dynamic analysis and reduces the problem of information islands.

[0069] Furthermore, it is also convenient for data sharing, replication and merging, optimizes seismic data acquisition and interpretation, and improves the accuracy of velocity field modeling. The fine calibration of joint well-seismic data enhances the accuracy of stratigraphic interpretation. Big data governance and intelligent logging interpretation applications reduce reliance on expert experience and simplify data processing procedures. The construction of a unified logging database realizes the integrated application of data and improves data security and access efficiency. In addition, the standardized processing of seismic attributes and the improvement of inter-well reservoir prediction accuracy provide strong data support for reservoir management and decision-making, reduce exploration risks, save costs, and enhance the scientificity and effectiveness of oil and gas exploration.

[0070] Preprocess the seismic data, well logging data and stratigraphic data to obtain the time-depth relationship after well seismic. The distribution of physical property parameters is obtained through the time-depth relationship.

[0071] Specifically, the logging data and the formation data are converted from data described by depth to data described by time, so that the logging data and the formation data can be associated with the seismic data, and the distribution of physical property parameters can be further obtained.

[0072] Understandably, preprocessing of seismic data, well logging data, and stratigraphic data can significantly improve data quality, ensure consistency, and accurately establish the time-depth relationship after well-seismic. This process removes noise and outliers, improves data accuracy, and enhances the accuracy of seismic and structural interpretations. Accurate time-depth relationships are crucial for seismic data interpretation, especially in clastic sedimentary formations, which can greatly improve the accuracy and precision of well-seismic calibration. In addition, preprocessed data is more suitable for geological analysis and reservoir evaluation, enhancing data availability, promoting seismic-geological integration, and improving work efficiency. The automated preprocessing process speeds up data processing, providing a solid data foundation for seismic inversion and reservoir modeling, thereby improving the efficiency and success rate of oil and gas exploration.

[0073] According to the distribution of physical property parameters, the lithofacies of the oil wells in the work area are divided and the lithofacies curve of the upper well is obtained.

[0074] Specifically, according to the oil test data report and the distribution of physical parameters such as porosity, water saturation, and shale content on the well, the lithofacies are reasonably divided to obtain the lithofacies curve of the well. Well refers to the field work and activities directly related to the oil well.

[0075] Furthermore, physical parameters refer to the physical properties of rocks, such as porosity, permeability, density, acoustic velocity, etc. The distribution of these parameters reflects how the physical characteristics of underground rocks vary with space. Lithofacies refers to rock units with the same or similar petrological characteristics. By analyzing physical parameters, the rocks under the oil well can be divided into different lithofacies, such as sandstone, mudstone, limestone, etc.

[0076] The seismic data are offset and stacked to obtain post-stack seismic data.

[0077] In some executable embodiments, step S2 includes:

[0078] Based on the post-stack seismic data, the seismic attribute volume is obtained according to the preset conditions.

[0079] Specifically, seismic attributes such as instantaneous amplitude, instantaneous frequency, relative impedance, trace integral, sweet spot and frequency division are extracted based on post-stack seismic data.

[0080] According to the intrinsic resolution of post-stack seismic data, the resolution of lithofacies curves is matched with the intrinsic resolution by using the scale coarsening method to obtain pseudo-well curves.

[0081] It should be noted that the inherent resolution of seismic data refers to the ability of the seismic data itself to resolve the smallest detail or structure. This is limited by many factors such as the frequency of seismic waves, the signal-to-noise ratio of seismic data, geological conditions, etc. Scale coarsening can be achieved through a variety of mathematical methods, such as the Backus equivalent averaging method, which processes small-scale well logging curves to make them comparable at the seismic scale. Lithofacies curves are obtained from well logging data and reflect the curves that reflect the changes in the physical properties of underground rocks with depth. The resolution of lithofacies curves refers to the ability of these curves to resolve the smallest changes in rock physical properties.

[0082] Furthermore, the lithofacies curve refers to the lithofacies curve at the well logging resolution (generally 0.125m), and the seismic data refers to the original seismic data and extracted seismic attributes at the seismic resolution (generally 2ms). The scale coarsening method is the process of converting high-resolution well data to match the low-resolution seismic data. This method is a distance-weighted combined weight method. The specific implementation process is: within a fixed-size sliding time window, the lithofacies with the most occurrences are counted, and the lithofacies with the largest number (largest weight) in the window are assigned to the central sampling point; on this basis, the concept of distance weighting is introduced, and the closer the lithofacies are to the center point, the higher the weight is, so as to update the weight, and finally the lithofacies data on the well logging are coarsened to the scale of seismic data, and the lithofacies transition is kept smooth.

[0083] In some executable embodiments, the seismic attribute volume includes an instantaneous amplitude attribute volume, an instantaneous frequency attribute volume, and a sweet spot attribute volume.

[0084] The calculation formulas corresponding to the instantaneous amplitude, instantaneous frequency and sweet spot properties are as follows:

[0085]

[0086]

[0087] Among them, s(t) is the seismic data, h(t) is the Hilbert transform result of the seismic data, A(t), (t) and Sweetness are the calculated instantaneous amplitude, instantaneous frequency and sweet spot attributes respectively. Sweet spot attributes are often used to distinguish reservoirs from non-reservoirs, corresponding to seismic reflection signals with relatively strong amplitude and low frequency. Strong amplitude features correspond to strong reflections between hydrocarbon-bearing reservoirs and overburden (mudstone), and low frequency features correspond to large sets of pure sandstone reflections.

[0088] In some executable embodiments, step S3 includes:

[0089] The pseudo well curves of different seismic attribute bodies at the oil well locations are obtained to obtain a training set.

[0090] Furthermore, the oil well location includes the well point. Pseudo-well curves of different seismic attribute data bodies at the well point are extracted as input data of the training process, that is, the training set.

[0091] The neural network model is trained using a supervised learning algorithm and a training set. The seismic data is classified using the trained neural network model to obtain a lithofacies classification result. The lithofacies classification result includes several lithofacies types. Each lithofacies type is calculated using an inverse distance weighted algorithm to obtain a lithofacies probability body.

[0092] Specifically, a trained neural network model is obtained by training using supervised learning algorithms such as kNN, and the most frequently appearing marker type in these k neurons will be output as the prediction result.

[0093] When using the kNN algorithm, the majority voting method is generally used to determine the prediction results. If the number of markers is consistent, the distance weighting strategy will be used to weight the votes according to the distance between each neighboring unit and the sample to be predicted, so as to avoid the result of the same number of markers. The marker type is all the lithofacies types on the lithofacies curve. The prediction result output is the predicted lithofacies type.

[0094] The trained neural network is used to perform facies classification on the seismic data of the target layer in the entire work area, and the lithofacies probability bodies of different lithofacies predicted by these k neurons are calculated based on the inverse distance weighted algorithm.

[0095] The lithofacies probability body is the predicted probability of the same lithofacies at different locations in the work area, and its dimension is the same as the original seismic data and attribute body. The lithofacies type is determined according to the actual lithology and fluid distribution in the work area. Taking the common sandstone and mudstone formation as an example, the lithofacies type can include oil-bearing sandstone phase, oil-water phase, water-bearing sandstone phase, sand-mud transition phase and mudstone phase.

[0096] In some executable embodiments, step S4 includes:

[0097] According to the lithofacies classification results, rock physics modeling is established, pseudo-well curves are obtained, and target parameters of the initial seismic inversion model are determined.

[0098] Furthermore, rock physics modeling is carried out to obtain pseudo-well curves of physical parameters such as pore structure and permeability, and to determine several target parameters for inversion initial modeling.

[0099] In an executable embodiment, a sandstone-mudstone formation is selected to directly invert the physical parameters such as the rock matrix modulus, fluid modulus and pore structure parameters. The initial model to be constructed includes seven parameters, including matrix bulk modulus, matrix shear modulus, fluid modulus, matrix density, fluid density, porosity and equivalent pore aspect ratio. However, some parameters are coupled with each other. After rock physics modeling and analysis, four independent and uncoupled physical parameters can be determined: mud content, water saturation, porosity and equivalent pore aspect ratio. The initial models of the seven physical parameters to be established can all be directly calculated by these four parameters. Therefore, the initial model of the target physical parameters established by the lithofacies probability inverse mapping method includes four physical parameters, including mud content, water saturation, porosity and equivalent pore aspect ratio.

[0100] According to the lithofacies curves and target parameters, statistical rock physics analysis is performed on different lithofacies types to obtain the histogram distribution and probability density distribution of different physical property parameters in different lithofacies.

[0101] Furthermore, based on the lithofacies curves and physical property parameters on the well, statistical rock physics analysis of different lithofacies was carried out to obtain the histogram distribution and N-dimensional probability density distribution of different physical property parameters in different lithofacies types, where N is the number of physical property parameters involved in statistical rock physics.

[0102] According to the histogram distribution and probability density distribution, the posterior probability density space is established.

[0103] Histogram distribution and probability density distribution are the basis for establishing the posterior probability density space. The histogram helps estimate the probability density function through the characteristics of the facies distribution, while the prior probability density is combined with the likelihood function (describing the relationship between the facies distribution and the physical property parameters) to form the posterior probability density through the Bayesian theorem.

[0104] In some executable embodiments, the probability density distribution includes an N-dimensional probability density distribution.

[0105] The posterior probability density space includes an N-dimensional physical property parameter space, covering the valid value ranges of N physical property parameters.

[0106] Where N is the number of physical parameters involved in statistical rock physics analysis.

[0107] In some executable embodiments, step S5 includes:

[0108] Based on the posterior probability density space and the probability density distribution, the posterior probabilities of different lithofacies for the physical property parameter combinations corresponding to each point in the posterior probability density space are calculated and normalized.

[0109] Specifically, according to the histogram statistics of the physical property parameters, the prior probability value p of each point in the N-dimensional parameter space is assigned priorBased on the obtained N-dimensional parameter space, according to Bayesian theorem and probability density distribution, the posterior probability of different lithofacies corresponding to each parameter combination at each point in the space is calculated and normalized.

[0110] The posterior probability of each sampling point in the seismic data is traversed to obtain the lithofacies probability ratio of each sampling point in the seismic data.

[0111] Specifically, each sampling point is each point recorded in the seismic trace.

[0112] According to the Monte Carlo algorithm and probability density function, M physical property parameter combinations that conform to the preset parameter distribution are generated.

[0113] According to the rejection-acceptance sampling algorithm and the lithofacies probability ratio of the current sampling point in the seismic data, m physical property parameter combinations that meet the conditions are selected from M physical property parameter combinations within a preset tolerance.

[0114] Specifically, the combination formula of M physical property parameters is as follows:

[0115] {x} i ~p(f{x}),fori=1,2,…,

[0116] Among them, M{x} i Indicates that M samples {x} are generated according to the probability density function p(f(x)) i , M represents the total number of samples generated, and p(f{x}) represents the probability density function used to generate samples.

[0117] The m physical property parameter combinations are used as the initial model for direct inversion of physical property parameters, so as to realize the allocation of different physical property parameters corresponding to different lithofacies.

[0118] In one executable embodiment, each sampling point in the seismic data includes each point of the seismic trace record, and the sampling rate is 2 ms.

[0119] In an executable embodiment, taking oil-bearing sandstone and mudstone formations as an example, this embodiment selects formations with a mud content lower than 35% and a water saturation lower than 0.55 as oil-bearing sandstone phases, defines formations with a mud content lower than 35% and a water saturation between 0.55 and 0.7 as oil-water co-layer phases, defines formations with a mud content lower than 35% and a water saturation higher than 0.7 as water-bearing sandstone phases, defines formations with a mud content between 35% and 60% as sand-mud transition phases, and defines formations with a mud content higher than 60% as mudstone phases.

[0120] Figure 2The statistical rock physics analysis results based on the above phase classification show that the simple normal distribution model cannot well describe the probability density relationship between most physical parameters, while the probability density estimation based on the kernel function can achieve the description of such complex probability models. Among them, each row represents a type of seismic phase, and each column represents a combination of physical parameters. Based on this analysis, statistical rock physics analysis of different phases can be carried out based on lithofacies curves and physical parameters, and the histogram distribution and N-dimensional probability density distribution of different physical parameters in different phase types can be obtained. Figure 3 The probability density distribution corresponding to one of the attribute combinations (porosity and equivalent pore aspect ratio) shows that there is a certain overlap between different phases, which will increase the uncertainty of the forward use of physical property parameters to identify lithofacies, but it is difficult to have a negative impact on the inverse mapping process.

[0121] After completing the statistical rock physics analysis, the instantaneous amplitude, instantaneous frequency, relative impedance, trace integral and sweet spot are used to participate in the training of the neural network. The lithofacies classification results identified by the trained neural network are as follows: Figure 4 As shown in (the classification results of water-bearing sandstone, oil-bearing sandstone and mudstone are selected). Based on the lithofacies classification results and the corresponding probability body, the inverse mapping of physical property parameters is realized by using steps S4 to S5. The initial models of the two physical property parameters of porosity and pore structure obtained by the lithofacies probability inverse mapping are shown in Figure 5 and Figure 6 As shown, this is the final high-precision low-frequency model.

[0122] Due to the limited number of well points in the work area (the work area contains logging data from 4 wells), it is difficult to use the vertical logging data to obtain the plane variogram. Therefore, it is not very reliable to hastily use geostatistical methods to establish the initial model of physical property parameters. Therefore, the conventional initial model of physical property parameters is established by using trend analysis in the vertical direction and discrete smooth interpolation in the plane. The results are as follows: Figure 7 As shown in the figure, compared with the actual physical property parameters of a well at the well point, it can be seen that there is a large error. The initial model of physical property parameters established by using lithofacies probability inverse mapping is as follows Figure 8 As shown in the figure, the calculated results of the physical property parameters at the well point are in good agreement, which proves the effectiveness of this method.

[0123] The initial model established by the present invention based on the inverse mapping of lithofacies probability not only avoids the serious problems of traditional well-to-well interpolation modeling, but also does not need to rely on accurate variograms to simulate the laws of spatial geological changes. In addition, the vertical variation law of physical property parameters is not as obvious as that of elastic parameters, and the spatial variation law of physical property parameters is more easily controlled by geological units such as lithofacies and sedimentary phases. Therefore, on the basis of establishing a stratigraphic sequence framework, the initial model of physical property parameters obtained by using vertical trends and plane discrete smooth interpolation has large errors at the well points, and the initial model established by the traditional method has low credibility. The initial model of physical property parameters established by using the inverse mapping of lithofacies probability has better consistency with the well points, and the spatial variation law of the parameters is consistent with the lithofacies distribution, which is more practical geological significance.

[0124] The present invention also provides a device for establishing an initial model of seismic inversion based on lithologic probability inverse mapping. The device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the method for establishing an initial model of seismic inversion based on lithologic probability inverse mapping as described above are implemented.

[0125] The above embodiments only express the preferred implementation modes of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the patent scope of the present invention. It should be pointed out that, for ordinary technicians in this field, the above technical features can be freely combined without departing from the concept of the present invention, and several deformations and improvements can be made, which all belong to the protection scope of the present invention. Therefore, all equivalent changes and modifications made to the scope of the claims of the present invention should belong to the coverage of the claims of the present invention.

Claims

1. A method for establishing an initial model for seismic inversion based on lithofacies probability inverse mapping, characterized in that: The following steps are involved: S1. Collect and process oil test data, seismic data, well logging data and stratigraphic data to obtain post-stack seismic data and lithofacies curves; S2, acquiring a plurality of seismic attribute bodies from the post-stack seismic data, and obtaining a pseudo-well curve of each seismic attribute body in combination with the lithofacies curve; S3, using the pseudo-well curve to train the neural network model, using the trained neural network model in combination with the seismic data to output lithofacies classification results and lithofacies probability bodies; S4, analyzing the lithofacies classification results and the lithofacies probability body to obtain probability density distributions of different physical property parameters and establish a posterior probability density space of the physical property parameters; S5. Based on the posterior probability density space and the probability density distribution, the seismic data are traversed, and a combination of physical parameters that meets the conditions is generated using a Monte Carlo algorithm and rejection-acceptance sampling to obtain the seismic inversion initial model.

2. The method for establishing the initial model of seismic inversion based on lithofacies probability inverse mapping according to claim 1, characterized in that: The step S1 comprises: Collecting the oil testing data, seismic data, well logging data and stratum data in the work area; Preprocessing the seismic data, the well logging data and the stratigraphic data to obtain the time-depth relationship after the well earthquake; obtaining the distribution of physical property parameters through the time-depth relationship; According to the distribution of the physical property parameters, the lithofacies of the oil well are divided to obtain the lithofacies curve of the upper well; The seismic data are offset and stacked to obtain the post-stack seismic data.

3. The method for establishing the initial model of seismic inversion based on lithofacies probability inverse mapping according to claim 1, characterized in that: The step S2 comprises: Based on the post-stack seismic data, obtaining a seismic attribute volume according to preset conditions; According to the inherent resolution of the post-stack seismic data, a scale coarsening method is used to match the resolution of the lithofacies curve with the inherent resolution to obtain the pseudo-well curve.

4. The method for establishing the initial model of seismic inversion based on lithofacies probability inverse mapping according to claim 3, characterized in that: The seismic attribute volume includes an instantaneous amplitude attribute volume, an instantaneous frequency attribute volume and a sweet spot attribute volume.

5. The method for establishing the initial model of seismic inversion based on lithofacies probability inverse mapping according to claim 1, characterized in that: The step S3 comprises: Acquire the pseudo-well curves of different seismic attribute bodies at the well locations of the oil wells to obtain a training set; The neural network model is trained using a supervised learning algorithm and the training set; the seismic data is classified using the trained neural network model to obtain the lithofacies classification result; the lithofacies classification result includes several lithofacies classifications; each lithofacies classification is calculated in combination with an inverse distance weighted algorithm to obtain the lithofacies probability body.

6. The method for establishing an initial model for seismic inversion based on lithofacies probability inverse mapping according to claim 1, characterized in that: The step S4 comprises: Establish rock physics modeling according to the lithofacies classification results, obtain well curves, and determine target parameters of the seismic inversion initial model; According to the lithofacies curve and the target parameters, statistical rock physical analysis is performed on different lithofacies to obtain the histogram distribution and the probability density distribution of different physical property parameters in different lithofacies; The posterior probability density space is established according to the histogram distribution and the probability density distribution.

7. The method for establishing the initial model of seismic inversion based on lithofacies probability inverse mapping according to claim 6, characterized in that: The probability density distribution includes an N-dimensional probability density distribution; The posterior probability density space includes an N-dimensional physical property parameter space, covering the valid value ranges of N physical property parameters; Wherein N is the number of physical property parameters involved in the statistical rock physics analysis.

8. The method for establishing an initial model for seismic inversion based on lithofacies probability inverse mapping according to claim 6, characterized in that: The step S5 comprises: Based on the posterior probability density space and according to the probability density distribution, the posterior probabilities of different lithofacies for the physical property parameter combinations corresponding to each point in the posterior probability density space are calculated and normalized; Traversing the posterior probability of each sampling point in the seismic data to obtain the lithofacies probability ratio of each sampling point in the seismic data; Generate M physical property parameter combinations that meet the preset parameter distribution according to the Monte Carlo algorithm and the probability density function; According to the rejection-acceptance sampling algorithm and the lithofacies probability ratio of the current sampling point in the seismic data, m physical property parameter combinations that meet the conditions are selected from the M physical property parameter combinations within a preset tolerance; The m physical property parameter combinations are used as an initial model for direct inversion of physical property parameters.

9. The method for establishing the initial model of seismic inversion based on lithofacies probability inverse mapping according to claim 8, characterized in that: Each sampling point in the seismic data includes each point of the seismic trace record, and the sampling rate is 2ms.

10. A device for establishing an initial model of seismic inversion based on lithofacies probability inverse mapping, the device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the method for establishing a seismic inversion initial model based on lithofacies probability inverse mapping as described in any one of claims 1 to 9 are implemented.

Images