A method and device for predicting seismic reservoir physical property parameters

Abstract

The application provides a seismic reservoir physical property parameter prediction method and device, and the seismic reservoir physical property parameter prediction method comprises the following steps: grouping training data based on data features of the pre-generated training data to generate a plurality of grouped data; learning data features of the plurality of grouped data by using a plurality of correlation vector regressors to generate a plurality of seismic reservoir physical property parameter learning models; and predicting seismic reservoir physical property parameters of a target block according to the plurality of seismic reservoir physical property parameter learning models. The application uses correlation vector regression as an expert, and obtains an expert-weighted learning model through a gating function. The proposed method decomposes a complex large prediction problem into several relatively simple problems, and predicts one by one, thereby solving the problems of poor applicability and low precision of a current single prediction model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of oil and gas field exploration and development, in particular to the technical field of seismic data processing and interpretation, and more particularly to a seismic reservoir physical property parameter prediction method and device. BACKGROUND

[0002] With the continuous development of the global economy, the demand for oil and natural gas of each country is also rising, which requires oil companies to increase exploration and development efforts in the upstream sector to continuously discover new geological reserves and maximize the use of existing reserves. In this process, fine reservoir characterization is the key. Fine reservoir characterization refers to the use of multi-disciplinary information to quantitatively predict and describe reservoirs and predict the spatial distribution of reservoir physical property parameters (including elastic properties, lithofacies, porosity, permeability, etc.). Fine reservoir characterization is the basis for comprehensive geological analysis and evaluation and is a key link in the later exploration and development stages. Fine reservoir parameter prediction is a fundamental and most important step in reservoir evaluation, which requires the establishment of a systematic and widely applicable reservoir physical property parameter prediction method.

[0003] Further, fine reservoir parameter prediction mainly predicts the spatial distribution of reservoir parameters (including elastic properties, lithofacies, porosity, permeability, etc.) in the interwell area by using limited known well data, seismic data, and other geological information. In actual work, due to the complexity of underground geology, the relationship between reservoir physical property parameters and seismic data is affected by many factors including lithology, porosity, diagenesis, fluid, temperature, etc. A single model cannot be suitable for all data, especially for high-dimensional and complex problems. It is difficult to derive a deterministic relationship between reservoir parameters and seismic data, and directly obtain reservoir physical property parameters from seismic data. Currently, seismic reservoir physical property parameter prediction mainly uses rock physics models as a link, but there are two problems:

[0004] (1) Due to different measurement scales, there is inevitably some error in directly matching the core analysis results measured by high-frequency experiments with low-frequency seismic data;

[0005] (2) Rock physics applications require a large number of experimental bases for the rock physics properties of the study area, with many assumptions. The physical properties of rocks in different regions or different locations in the same region are complex and variable, and all have different seismic propagation laws. The rock physics models and empirical formulas based on simple theoretical relationships and empirical knowledge often lead to large prediction errors. There is currently no systematic and universal evaluation process and method for reservoir physical property parameter prediction.

[0006] Machine learning is a widely used technology in the field of artificial intelligence, which uses a large amount of data to train the model, and is often used to solve the complex nonlinear relationship between data and target. It plays an increasingly important role in oil and gas exploration and development. The most commonly used machine learning algorithms in geophysics are support vector regression and neural network. However, these two algorithms are very sensitive to parameters such as hyperparameters and learning rate. The selection of kernel function and parameters is greatly influenced by human beings and has certain randomness. Inappropriate parameters will lead to incorrect prediction results and amplify errors. Additional cross-validation operations are often required to find suitable parameters, but cross-validation operations are often computationally intensive and time-consuming. SUMMARY

[0007] In view of the problems in the prior art, the seismic reservoir physical parameter prediction method and device provided by the present application decompose a complex large prediction problem into several relatively simple problems and predict them one by one, solving the problem of poor applicability and low precision of the current single prediction model. This method does not need to assume that the data is stationary, nor does it need to establish the relationship between elastic parameters and physical parameters by means of rock physics model. It can achieve good results even in seismic reservoir property prediction involving a large amount of data, and is suitable for fine prediction of reservoir parameters in various complex formations. In summary, the present application can quickly and accurately predict the seismic parameters sensitive to the reservoir and reflecting the essential characteristics of the reservoir among a large number of seismic parameters (seismic attributes) that are originally independent of each other, and then predict the reservoir development in the target work area using the predicted seismic parameters, providing a reliable basis for oilfield exploration and development.

[0008] To solve the above technical problems, the present application provides the following technical solutions:

[0009] In a first aspect, the present application provides a seismic reservoir physical parameter prediction method, comprising:

[0010] Based on the data characteristics of the pre-generated training data, the training data is grouped to generate a plurality of grouped data;

[0011] Learning the data characteristics of the plurality of grouped data using a plurality of correlation vector regressors to generate a plurality of seismic reservoir physical parameter learning models;

[0012] According to the plurality of seismic reservoir physical parameter learning models, the seismic reservoir physical parameters of the target block are predicted.

[0013] In an embodiment, the seismic reservoir physical parameter prediction method further comprises generating training data, wherein the generating training data comprises the following steps:

[0014] Generating preliminary training data according to the well logging data of the target block;

[0015] The preliminary training data is subjected to depth correction and environmental correction to generate the training data.

[0016] In an embodiment, the training data is grouped based on the data features of the pre-generated training data to generate a plurality of grouped data, including:

[0017] The training data is grouped into the plurality of grouped data based on input-output relationship by using a k-means algorithm.

[0018] In an embodiment, the seismic reservoir property parameter of the target block is predicted according to the plurality of seismic reservoir property parameter learning models, including:

[0019] The plurality of seismic reservoir property parameter learning models are weighted by using a gating function to generate a seismic reservoir property parameter prediction model.

[0020] The seismic reservoir property parameter of the target block is predicted according to the seismic reservoir property parameter prediction model.

[0021] In a second aspect, the present application provides a seismic reservoir property parameter prediction device, which comprises:

[0022] A training data grouping module is configured to group the training data based on the data features of the pre-generated training data to generate a plurality of grouped data.

[0023] A plurality of learning model generation modules are configured to learn the data features of the plurality of grouped data by using a plurality of correlation vector regressors to generate a plurality of seismic reservoir property parameter learning models.

[0024] A property parameter prediction module is configured to predict the seismic reservoir property parameter of a target block according to the plurality of seismic reservoir property parameter learning models.

[0025] In an embodiment, the seismic reservoir property parameter prediction device further comprises a training data generation module configured to generate training data, and the training data generation module comprises:

[0026] A preliminary training data generation unit is configured to generate preliminary training data according to the logging data of the target block.

[0027] A training data generation unit is configured to subject the preliminary training data to depth correction and environmental correction to generate the training data.

[0028] In an embodiment, the training data grouping module specifically groups the training data into the plurality of grouped data based on input-output relationship by using a k-means algorithm.

[0029] In an embodiment, the parameter prediction module comprises:

[0030] a model weighting unit configured to weight the plurality of seismic reservoir property parameter learning models using a gating function to generate a seismic reservoir property parameter prediction model;

[0031] a property parameter prediction unit configured to predict the seismic reservoir property parameters of the target block according to the seismic reservoir property parameter prediction model.

[0032] In a third aspect, the present application provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method for predicting seismic reservoir property parameters when executing the program.

[0033] In a fourth aspect, the present application provides a computer readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method for predicting seismic reservoir property parameters.

[0034] As can be seen from the above description, the method and device for predicting seismic reservoir property parameters provided by the embodiments of the present application first group the training data based on the data features of the pre-generated training data to generate a plurality of grouped data; then learn the data features of the plurality of grouped data using a plurality of correlation vector regressors to generate a plurality of seismic reservoir property parameter learning models; and finally predict the seismic reservoir property parameters of the target block according to the plurality of seismic reservoir property parameter learning models. The present application decomposes a complex reservoir property parameter prediction problem into several simple problems, and the method and device do not need to assume that the data is stationary, nor do they need to establish the relationship between the elastic parameters and the property parameters by means of rock physics models, and can achieve good results even in seismic reservoir property prediction involving a large amount of data, and are suitable for fine prediction of reservoir parameters of various complex formations and have strong operability. BRIEF DESCRIPTION OF DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0036] Figure 1 Flowchart of the method for predicting seismic reservoir property parameters in the embodiments of the present application Figure 1 ;

[0037] Figure 2 Flowchart of the method for predicting seismic reservoir property parameters in the embodiments of the present application Figure 2 ;

[0038] Figure 3 Flowchart for step 400 in an embodiment of the present application;

[0039] Figure 4 Flowchart for step 100 in an embodiment of the present application;

[0040] Figure 5 Flowchart for step 200 in an embodiment of the present application;

[0041] Figure 6 Flowchart for the method for predicting seismic reservoir physical parameters in a specific application example of the present application;

[0042] Figure 7 Flowchart for the method for predicting seismic reservoir physical parameters in a specific application example of the present application;

[0043] Figure 8 is the measured curve of Well A in an embodiment of the present application: (a) P-wave velocity; (b) S-wave velocity; (c) density; (d)-(f) are respectively porosity, shale content and water saturation;

[0044] Figure 9 is the predicted result of Well A data in an embodiment of the present application: (a) porosity; (b) shale content; (c) water saturation; (d)-(f) are respectively the residual error between the predicted value and the true value of porosity, shale content and water saturation;

[0045] Figure 10 is the measured curve of Well B in an embodiment of the present application: (a) P-wave velocity; (b) S-wave velocity; (c) density; (d)-(f) are respectively porosity, shale content and water saturation;

[0046] Figure 11 is the predicted result of Well B data in an embodiment of the present application: (a) porosity; (b) shale content; (c) water saturation; (d)-(f) are respectively the residual error between the predicted value and the true value of porosity, shale content and water saturation.

[0047] Figure 12 is the structure diagram of the device for predicting seismic reservoir physical parameters in an embodiment of the present application Figure 1 ;

[0048] Figure 13 is the structure diagram of the device for predicting seismic reservoir physical parameters in an embodiment of the present application Figure 2 ;

[0049] Figure 14 is the structure diagram of the training data generation module 40 in an embodiment of the present application;

[0050] Figure 15 A structural schematic diagram of a parameter prediction module 30 in an embodiment of the present application;

[0051] Figure 16 A structural schematic diagram of an electronic device in an embodiment of the present application. DETAILED DESCRIPTION

[0052] In order to make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0053] It should be noted that the terms “include” and “have” and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover the non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units need not be limited to those clearly listed steps or units, but can include other steps or units not clearly listed or inherent to the process, method, product or device.

[0054] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the drawings and in combination with the embodiments.

[0055] Embodiments of the present application provide a specific implementation of a seismic reservoir physical property parameter prediction method, referring to Figure 1 The method specifically includes the following contents:

[0056] Step 100: Group the training data based on the data features of the pre-generated training data to generate a plurality of grouped data.

[0057] Specifically, the training data is clustered and grouped to construct different correlation vector experts (machine learning models): for the grouped data features, the training data is divided into a plurality of data sets based on the input-output relationship using a k-means algorithm.

[0058] Step 200: Learn the data features of the plurality of grouped data using a plurality of correlation vector regressors to generate a plurality of seismic reservoir physical property parameter learning models;

[0059] It can be understood that the support vector machine (SVM) is a better method which realizes the idea of structural risk minimization. Its machine learning strategy is the principle of structural risk minimization, in order to minimize the expected risk, the empirical risk and the confidence range should be minimized at the same time, the basic idea of support vector machine method is:

[0060] (1) The learning machine is specially designed for the case of limited samples, which realizes the structural risk minimization: a compromise is sought between the approximation accuracy of the given data and the complexity of the approximation function in order to obtain the best generalization ability;

[0061] (2) The final solution is a convex quadratic programming problem, theoretically, the global optimal solution will be obtained, which solves the local extremum problem that cannot be avoided in neural network method;

[0062] (3) The actual problem is converted to a high-dimensional feature space through nonlinear transformation, and a linear decision function is constructed in the high-dimensional space to realize the nonlinear decision function in the original space, which skillfully solves the dimension problem and ensures good generalization ability, and the algorithm complexity is independent of the sample dimension.

[0063] Specifically, by upgrading, a linear decision function is constructed in the high-dimensional space to realize linear regression, and when using e insensitive function, the basis is mainly e insensitive function and kernel function algorithm. If the fitted mathematical model is expressed as a curve in a multi-dimensional space, the result obtained according to the e insensitive function is an "e pipe" including the curve and the training points. Among all the sample points, only the part of the sample points distributed on the "pipe wall" determines the position of the pipe. This part of the training sample is called "support vector". In order to adapt to the nonlinearity of the training sample set, the traditional fitting method usually adds high-order terms behind the linear equation. This method is indeed effective, but the adjustable parameters increased by this method increase the risk of overfitting. The support vector regression algorithm uses kernel function to solve this contradiction. Using kernel function instead of linear term in linear equation can make the original linear algorithm "nonlinear", that is, it can do nonlinear regression. At the same time, the introduction of kernel function achieves the purpose of "dimension increasing", and the adjustable parameters increased by overfitting can still be controlled.

[0064] Step 300: predicting the seismic reservoir property parameters of the target block according to the plurality of seismic reservoir property parameter learning models.

[0065] Specifically, a specific model is selected to process a group of data, several models can be learned from part of the data, and a final learning model is generated by weighting through a gating function to predict the corresponding output.

[0066] From the above description, the method for predicting seismic reservoir physical parameters provided by the embodiment of the present application first groups the training data based on the data features of the pre-generated training data to generate a plurality of grouped data; then learns the data features of the plurality of grouped data by using a plurality of correlation vector regressors to generate a plurality of seismic reservoir physical parameter learning models; and finally predicts the seismic reservoir physical parameters of the target block according to the plurality of seismic reservoir physical parameter learning models. The present application decomposes a complex reservoir physical parameter prediction problem into several simple problems. The method and device do not need to assume that the data is stationary, nor do they need to establish the relationship between the elastic parameters and the physical parameters by means of rock physics models. Even in the prediction of seismic reservoir properties involving a large amount of data, good results can be achieved. The method and device are suitable for fine prediction of reservoir parameters of various complex formations and have strong operability.

[0067] In an embodiment, referring to Figure 2 , the method for predicting seismic reservoir physical parameters further comprises:

[0068] Step 400: generating training data, further referring to Figure 3 , the step 400 comprises:

[0069] Step 401: generating preliminary training data according to the well logging data of the target block;

[0070] Step 402: performing depth correction and environmental correction on the preliminary training data to generate the training data.

[0071] In steps 401 and 402, the known well logging curves are used to calculate the synthetic condition attributes as training data, and the depth correction, environmental correction and other preprocessing are performed. The condition attributes include one or more of the P-wave velocity, S-wave velocity and density.

[0072] In an embodiment, referring to Figure 4 , the step 100 further comprises:

[0073] Step 101: grouping the training data into the plurality of grouped data based on the input-output relationship by using the k-means algorithm.

[0074] It should be noted that the k-means clustering algorithm is an iterative solution clustering algorithm, and the steps are as follows: the data is divided into K groups in advance, K objects are randomly selected as initial cluster centers, the distance between each object and each seed cluster center is calculated, and each object is assigned to the cluster center closest to it. The cluster center and the object assigned to it represent a cluster. The cluster center of the cluster is recalculated according to the existing objects in the cluster after each sample is assigned. This process will be repeated until a certain termination condition is met. The termination condition can be that no (or a minimum number of) objects are reassigned to different clusters, no (or a minimum number of) cluster centers change, or the sum of squared errors is locally minimized.

[0075] In an embodiment, referring to Figure 5 , step 200 further comprises:

[0076] Step 201: weighting the plurality of seismic reservoir property parameter learning models using a gating function to generate a seismic reservoir property parameter prediction model;

[0077] Step 202: predicting the seismic reservoir property parameters of the target block according to the seismic reservoir property parameter prediction model.

[0078] From the above description, it can be seen that the embodiment of the present application provides a seismic reservoir property parameter prediction method, which can dynamically divide data using multiple experts and make predictions respectively, and is more targeted than the traditional method of using a single rock physics model or feature learning model. Therefore, it can predict the relationship between reservoir properties and elastic properties and produce more accurate results. Using relevant vector regression as an expert, it does not rely on human experience to set penalty parameters and can provide a stable sparse learning model that only retains the most important vectors and speeds up the calculation time of test data. The multi-expert model can combine multiple learning models according to the data structure and then specify different experts to solve different problems. The method has the following four advantages:

[0079] (1) Using machine learning methods to mine the complex nonlinear relationship between reservoir parameters and seismic data: machine learning algorithms can fully exploit the internal relationship between reservoir parameters and seismic data, establish a nonlinear mapping relationship between reservoir property parameters and seismic data by learning known data information, and realize automatic reservoir prediction;

[0080] (2) Using relevant vector regression as an expert, filtering redundant samples, and enhancing sparsity to improve prediction efficiency, and without relying on human experience or repeated trials to set hyperparameters, while ensuring the accuracy of the prediction results, greatly saving the calculation time spent in the optimization process or cross-validation;

[0081] (3) This method does not require establishing a physical relationship between reservoir parameters and seismic data for specific lithology or specific study areas. Instead, it uses known data to establish a relationship model between reservoir parameters and seismic elastic parameters, and applies the model to seismic data in unknown areas between wells to achieve the purpose of reservoir parameter prediction. It has a wide range of applications.

[0082] (4) This method does not rely solely on a single learning model for reservoir prediction. Instead, it decomposes a complex reservoir parameter prediction problem into several relatively simple sub-problems, taking into account the characteristics of seismic data. This divide-and-conquer approach allows multiple experts to be trained on different parts of the data, thereby improving prediction accuracy. If the data is very complex, the hybrid expert model can be extended to a hierarchical hybrid expert model to progressively solve multiple sub-problems. However, the data and parameters applicable to hybrid expert systems and related vector regression are assumed to follow a Gaussian distribution. Gaussian distributions are smooth and sensitive to outliers, making them unsuitable for data with unstable components. When the data does not follow a Gaussian distribution, the prediction will be significantly affected.

[0083] For further explanation of this plan, please refer to [link / reference]. Figure 6 The present invention also provides a specific application example of the method for predicting seismic reservoir physical parameters.

[0084] S1: Training data acquisition and preprocessing.

[0085] Synthetic conditional attributes are calculated using known well logging curves as training data, and preprocessing such as depth correction and environmental correction is performed. The conditional attributes include one or more of P-wave velocity, S-wave velocity, and density.

[0086] S2: Data initialization.

[0087] Clustering and grouping to construct different relevance vector experts: For data features, the k-means algorithm is used to divide the training data into multiple datasets based on the input-output relationship. The number of datasets depends on the number of experts. For each expert, a radial basis function is selected to construct the kernel matrix. The model is applied for validation. Different numbers of relevance vector regression experts are used to calculate the target attribute. The appropriate number of experts is selected by comparing the performance of the mean absolute derivative and mean absolute percentage error between the predicted and true values.

[0088] S3: Establish a predictive vector machine expert.

[0089] Calculate the posterior probability of the sample, and calculate the posterior probability of the nth sample based on the existing parameters.

[0090]

[0091] where I denotes the number of experts, x (n) denotes the input elastic properties (density, P-wave and S-wave velocities, etc.), r (n) is the target reservoir property, which can represent porosity, saturation or shale content, etc. P i (r | x) denotes the probability of the data corresponding to the expert prediction, which is assumed to follow a Gaussian distribution with parameter σ:

[0092]

[0093] where represents the mean of the regression problem, w i denotes the weight of different samples in each expert, σ is the corresponding variance value. g i (x, v) denotes the gating subsystem, i.e., the probability of the ith expert for the data X, to ensure the non-negativity of the probability, it can be represented as a normalized exponential function with parameter v:

[0094]

[0095] According to the posterior probability Combined with the existing elastic properties (density, P-wave and S-wave velocities, etc.), the correlation vector regression is used to solve the maximization problem to obtain the expert parameter set to generate the corresponding expert output, and the maximization problem to be solved is:

[0096]

[0097] The equation can be transformed into:

[0098] P(α i ,σ i |r i |)∝P(r i |α i ,σ i )P(r i )P(σ i )

[0099] (α i ,σ i )=arg max P(r i |α i ,σ i )

[0100] The partial derivatives of the calculation formula with respect to and σ i are calculated and set to zero, so that the following update formula can be given:

[0101]

[0102]

[0103]

[0104] Most corresponding tends to zero, that is, a weight vector can be obtained in which most elements are zero, and the corresponding samples will also be removed from the relevant vector regression learning model. Finally, only a few samples remain in the model, called the relevant vector, thus achieving sparsity. In obtaining and After that, the prediction posterior probability of the ith expert for the unknown data point X * is given by:

[0105]

[0106] where and denote the prediction mean.

[0107] S4: Calculate the gating function and combine the expert prediction results.

[0108] The gating function determines the corresponding weight for each selected expert, and the maximization problem of the gating function is as follows:

[0109]

[0110] The gating function V = {v i , i = 1, …, I} is obtained.

[0111] The following process is used to solve the above equation,

[0112]

[0113] Based on the boundary optimization and the maximization of the surrogate function, the update equation of V = {v i , i = 1, …, I} can be obtained as:

[0114] V new = V old - He -1 g(V old )

[0115] where represents the gradient of Q g , and denotes the Kronecker product.

[0116]

[0117]

[0118] The lower bound of the Hessian matrix representing Qg is irrelevant to V, and T is a matrix with the size of N*N and the elements of 1.

[0119] S5: iteration until convergence, and a learning model is obtained.

[0120] Steps S2-S4 are repeated until convergence, and relevant parameters are output, and the output of each of the selected experts is weighted to generate a learning model, which can be represented by the expected value of formula 1, that is, the regression variable is calculated:

[0121]

[0122] S6: input test data and make a prediction.

[0123] Specifically, according to the seismic pre-stack inversion technology, the condition attribute of the to-be-identified region is taken as to-be-identified data, and is input into the learning model to predict a reservoir physical property parameter prediction result.

[0124] Based on the above method, the specific application example of the present application also provides a seismic reservoir physical property parameter prediction method taking a specific work area as an example.

[0125] Specifically, the specific application example of the present application takes high-quality seismic data of a certain well area as input, uses the learning model to predict the reservoir, and analyzes the results. The reservoir in this area mainly develops in the channel sand body. Due to the complex reservoir distribution characteristics, it is very challenging to predict the interwell reservoir development using the known well information in this area. Figure 7 Figure 8 is the measured curve of an embodiment A well of the present application: (a) P-wave velocity; (b) S-wave velocity; (c) density; (d)-(f) are respectively porosity, shale content and water saturation; Figure 9 is the prediction result of the A well data using three relevant vector regression experts of the present application: (a) porosity; (b) shale content; (c) water saturation; (d)-(f) are respectively the residual error of the predicted value and the true value of porosity, shale content and water saturation;

[0126] Figure 10 and Figure 11 are respectively the measured data of B well and the single-well prediction result using three relevant vector regression experts, it can be seen that the prediction results of A well and B well have high consistency with the measured data, and the residual error basically fluctuates around zero, which can effectively predict the reservoir physical property parameters, and fully illustrates the effectiveness of the method.

[0127] ​In summary, the seismic reservoir physical property parameter prediction method provided by the specific application instance comprises the following three parts: a plurality of experts for predicting reservoir physical property parameters; a gating function for dividing input data and assigning areas responsible by each expert; and a probability model combining the expert and the gating function.The method uses relevant vector regression as an expert and obtains a weighted learning model of the expert through the gating function.The proposed method decomposes a complex prediction problem into several relatively simple problems and predicts them one by one, solving the problem of poor applicability and low precision of the current single prediction model.The method and device do not need to assume that the data is stationary, nor do they need to establish the relationship between elastic parameters and physical parameters by means of rock physics models, and good results can be achieved even in seismic reservoir property prediction involving a large amount of data.In summary, the present application has the following beneficial effects:

[0128] (1) Using machine learning methods to mine the complex nonlinear relationship between reservoir parameters and seismic data: Machine learning algorithms can fully exploit the internal relationship between reservoir parameters and seismic data, establish a nonlinear mapping relationship between reservoir physical parameters and seismic data by learning known data information, and realize automatic reservoir prediction;

[0129] (2) Using relevant vector regression as an expert, filtering redundant samples, and enhancing sparsity to improve prediction efficiency, and without relying on human experience or repeated trials to set hyperparameters, the accuracy of the prediction results is guaranteed while greatly saving the calculation time spent in the optimization process or cross-validation;

[0130] (3) The method does not need to establish a determined physical relationship between reservoir parameters and seismic data for a specific lithology or a specific research area, but uses known data to establish a relationship model between reservoir parameters and seismic elastic parameters, and applies the model to seismic data in the unknown area between wells to realize the purpose of reservoir parameter prediction, and has a wide range of applications;

[0131] (4) The method is not only used to realize reservoir prediction by using a single learning model, but also can decompose a complex reservoir parameter prediction problem into several relatively simple sub-problems according to the characteristics of seismic data, divide and conquer, and train multiple experts to be responsible for different parts of the data to improve the accuracy of reservoir prediction.

[0132] Based on the same inventive concept, the application further provides a device for predicting seismic reservoir physical parameters, which can be used to implement the method described in the above embodiments, as described in the following embodiments. Since the device for predicting seismic reservoir physical parameters solves the problem by the same principle as the method for predicting seismic reservoir physical parameters, the implementation of the device for predicting seismic reservoir physical parameters can be referred to the implementation of the method for predicting seismic reservoir physical parameters, and the repeated parts will not be described here. The term "unit" or "module" used below can be a combination of software and / or hardware that can implement a predetermined function. Although the system described in the following embodiments is preferably implemented in software, the implementation of hardware or a combination of software and hardware is also possible and is conceived.

[0133] The embodiments of the application provide a specific implementation of a device for predicting seismic reservoir physical parameters, which can implement the method for predicting seismic reservoir physical parameters, as described in Figure 12 , the device for predicting seismic reservoir physical parameters specifically includes the following contents:

[0134] The training data grouping module 10 is configured to group the training data based on the data features of the pre-generated training data, so as to generate a plurality of grouped data.

[0135] The plurality of learning model generation modules 20 are configured to learn the data features of the plurality of grouped data by using a plurality of correlation vector regressors, so as to generate a plurality of seismic reservoir physical parameter learning models.

[0136] The parameter prediction module 30 is configured to predict the seismic reservoir physical parameters of the target block according to the plurality of seismic reservoir physical parameter learning models.

[0137] In an embodiment, as described in Figure 13 , the device for predicting seismic reservoir physical parameters further includes a training data generation module 40, as described in Figure 14 , the training data generation module 40 includes:

[0138] The preliminary training data generation unit 41 is configured to generate preliminary training data according to the well logging data of the target block.

[0139] The training data generation unit 42 is configured to perform depth correction and environmental correction on the preliminary training data, so as to generate the training data.

[0140] In an embodiment, the training data grouping module specifically groups the training data into the plurality of grouped data based on the input-output relationship by using a k-means algorithm.

[0141] In an embodiment, as described in Figure 15 , the parameter prediction module 30 includes:

[0142] The model weighting unit 31 is configured to weight the plurality of seismic reservoir property parameter learning models by using a gating function to generate a seismic reservoir property parameter prediction model.

[0143] The parameter prediction unit 32 is configured to predict the seismic reservoir property parameters of the target block according to the seismic reservoir property parameter prediction model.

[0144] As can be seen from the above description, the seismic reservoir property parameter prediction device provided by the embodiment of the present application firstly groups the training data based on the data features of the pre-generated training data to generate a plurality of grouped data; then learns the data features of the plurality of grouped data by using a plurality of correlation vector regressors to generate a plurality of seismic reservoir property parameter learning models; and finally predicts the seismic reservoir property parameters of the target block according to the plurality of seismic reservoir property parameter learning models. The present application decomposes a complex reservoir property parameter prediction problem into several simple problems. The method and device do not need to assume that the data is stationary, nor do they need to establish the relationship between the elastic parameters and the property parameters by means of rock physics models. Even in the prediction of seismic reservoir properties involving a large amount of data, good results can be achieved. The method and device are suitable for fine prediction of reservoir parameters of various complex formations and have strong operability. The device uses correlation vector regression as an expert and obtains an expert-weighted learning model by using a gating function. The proposed method decomposes a complex large prediction problem into several relatively simple problems and predicts them one by one, thereby solving the problem of poor applicability and low precision of the current single prediction model.

[0145] The device, module or unit illustrated in the above embodiments can be specifically implemented by a computer chip or entity, or by a product with certain functions. A typical implementation device is an electronic device, specifically, the electronic device may, for example, be a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an electronic mail device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0146] In a typical example, the electronic device specifically includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the above-mentioned front-end framework-based dynamic burying method when executing the program, and the steps include:

[0147] Step 100: Group the training data based on the data features of the pre-generated training data to generate a plurality of grouped data;

[0148] Step 200: Learn the data features of the plurality of grouped data by using a plurality of correlation vector regressors to generate a plurality of seismic reservoir property parameter learning models;

[0149] Step 300: predicting the seismic reservoir physical property parameters of the target block according to the plurality of seismic reservoir physical property parameter learning models.

[0150] Reference will now be made to the following description Figure 16 which shows a structural diagram of an electronic device 600 suitable for use in implementing embodiments of the present application.

[0151] As shown in Figure 16 , the electronic device 600 includes a central processing unit (CPU) 601, which can perform various appropriate operations and processes according to programs stored in a read only memory (ROM) 602 or loaded into a random access memory (RAM) 603 from a storage section 608. Various programs and data required for the operation of the system 600 are also stored in the RAM 603. The CPU 601, the ROM 602, and the RAM 603 are connected to each other through a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0152] The following components are connected to the I / O interface 605: an input section 606 including a keyboard, a mouse, etc.; an output section 607 including a display such as a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, a modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the I / O interface 605 as necessary. A removable media 611 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc. is attached to the drive 610 as necessary, so that a computer program read therefrom is installed in the storage section 608 as necessary.

[0153] In such embodiments, the computer program can be downloaded and installed from the network via the communication section 609, and / or installed from the removable media 611.

[0154] Computer-readable media includes permanent and non-permanent, movable and non-movable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible by a computing device. According to the definition herein, computer-readable media does not include transitory media such as modulated data signals and carriers.

[0155] For the convenience of description, the above device is described as various units described respectively in function. Of course, in the implementation of the present application, the functions of each unit can be implemented in the same or more software and / or hardware.

[0156] These computer program instructions can also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer readable memory produce a product including instruction devices, which realize the processes Figure 1 one or more processes and / or blocks Figure 1 one or more blocks or multiple blocks specified in the function.

[0157] These computer program instructions can also be loaded into a computer or other programmable data processing device, so that a series of operation steps are performed on the computer or other programmable device to produce computer implemented processes, so that the instructions executed on the computer or other programmable device provide steps for realizing the functions specified in the processes Figure 1 one or more processes and / or blocks Figure 1 one or more blocks or multiple blocks specified in the function.

[0158] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system or a computer program product. Therefore, the present application can be in the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the present application can be in the form of a computer program product implemented on one or more computer usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0159] The above merely provides an example of the present application and is not intended to limit the present application. The present application can have various changes and modifications for those skilled in the art. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the scope of claims of the present application.

Claims

1. A method for predicting seismic reservoir physical parameters, characterized in that, include: Based on the data features of the pre-generated training data, the training data is grouped to generate multiple grouped data. Multiple correlation vector regressors are used to learn the data features of the multiple grouped data to generate multiple seismic reservoir physical parameter learning models; Predict seismic reservoir properties of the target block based on the learning model of the multiple seismic reservoir properties parameters; Generating training data includes the following steps: Preliminary training data is generated based on the well logging data of the target block; The preliminary training data is subjected to depth correction and environmental correction to generate the training data. Specifically, the synthetic conditional attributes are calculated using known well logging curves as training data, and depth correction and environmental correction are performed. The conditional attributes include one or more of P-wave velocity, S-wave velocity and density. The steps for establishing the correlation vector regressor include: Calculate the posterior probability of the sample, and calculate the posterior probability of the nth sample based on the existing parameters. ; Where I represents the number of correlation vector regressors, This represents the input elastic properties, including density, P-wave velocity, and S-wave velocity. These are the target reservoir properties, including porosity, saturation, or shale content; The probability predicted by the correlation vector regressor corresponding to the data follows a Gaussian distribution with parameter σ. in This represents the average value of the regression problem. This represents the weight of different samples in each correlation vector regressor. That is the corresponding variance value. Let represent the probability of the gating subsystem, i.e., the i-th correlation vector regressor, for the data X. To ensure the non-negativity of the probability, it is represented as a parameterized expression. Normalized exponential function: Based on posterior probability Combining existing density, P-wave and S-wave velocities, a maximization problem is solved using correlation vector regression to obtain the parameter set of the correlation vector regressor, thereby generating the corresponding output of the correlation vector regressor. The maximization problem to be solved is: This equation can be transformed into: Calculation formula about and Taking the partial derivatives and setting them equal to zero, we have the following update formula: most When it approaches infinity, the corresponding As the relevance vectors approach zero, a weight vector is obtained where most elements are zero. The corresponding samples are then removed from the relevance vector regression learning model. Finally, only a small number of samples remain in the model, forming the relevance vector, thus achieving sparsity. and Then, the i-th correlation vector regressor performs regression on the unknown data points. The predicted posterior probability is given by the following formula: in and This represents the predicted average.

2. The method for predicting seismic reservoir physical parameters according to claim 1, characterized in that, Based on the data features of the pre-generated training data, the training data is grouped to generate multiple grouped data, including: The training data is grouped into multiple groups based on the input-output relationship using the k-means algorithm.

3. The method for predicting seismic reservoir physical parameters according to claim 1, characterized in that, Predicting seismic reservoir physical parameters based on the aforementioned multiple seismic reservoir physical parameter learning models includes: The multiple seismic reservoir physical parameter learning models are weighted using a gating function to generate a seismic reservoir physical parameter prediction model. The seismic reservoir properties of the target block are predicted based on the seismic reservoir property prediction model.

4. A device for predicting seismic reservoir physical parameters, characterized in that, include: The training data grouping module is used to group the training data based on the data features of the pre-generated training data to generate multiple grouped data. Multiple learning model generation modules are used to learn the data features of the multiple grouped data using multiple correlation vector regressors to generate multiple seismic reservoir physical parameter learning models. The physical property parameter prediction module is used to predict the seismic reservoir physical property parameters of the target block based on the learning model of the multiple seismic reservoir physical property parameters. Training data generation module, used to generate training data, the training data generation module includes: A preliminary training data generation unit is used to generate preliminary training data based on the well logging data of the target block; The training data generation unit is used to perform depth correction and environmental correction on the preliminary training data to generate the training data. Specifically, it uses known well logging curves to calculate synthetic conditional attributes as training data and performs depth correction and environmental correction. The conditional attributes include one or more of P-wave velocity, S-wave velocity and density. The steps for establishing the correlation vector regressor include: Calculate the posterior probability of the sample, and calculate the posterior probability of the nth sample based on the existing parameters. ; Where I represents the number of correlation vector regressors, This represents the input elastic properties, including density, P-wave velocity, and S-wave velocity. These are the target reservoir properties, including porosity, saturation, or shale content; The probability predicted by the correlation vector regressor corresponding to the data follows a Gaussian distribution with parameter σ. in This represents the average value of the regression problem. This represents the weight of different samples in each correlation vector regressor. That is the corresponding variance value. Let represent the probability of the gating subsystem, i.e., the i-th correlation vector regressor, for the data X. To ensure the non-negativity of the probability, it is represented as a parameterized expression. Normalized exponential function: Based on posterior probability Combining existing density, P-wave and S-wave velocities, a maximization problem is solved using correlation vector regression to obtain the parameter set of the correlation vector regressor, thereby generating the corresponding output of the correlation vector regressor. The maximization problem to be solved is: This equation can be transformed into: Calculation formula about and Taking the partial derivatives and setting them equal to zero, we have the following update formula: most When it approaches infinity, the corresponding As the relevance vectors approach zero, a weight vector is obtained where most elements are zero. The corresponding samples are then removed from the relevance vector regression learning model. Finally, only a small number of samples remain in the model, forming the relevance vector, thus achieving sparsity. and Then, the i-th correlation vector regressor performs regression on the unknown data points. The predicted posterior probability is given by the following formula: in and This represents the predicted average.

5. The seismic reservoir physical parameter prediction device according to claim 4, characterized in that, The training data grouping module specifically utilizes the k-means algorithm to group the training data into multiple groups based on the input-output relationship.

6. The seismic reservoir physical parameter prediction device according to claim 4, characterized in that, The parameter prediction module includes: The model weighting unit is used to weight the multiple seismic reservoir physical parameter learning models using a gating function to generate a seismic reservoir physical parameter prediction model. The physical property parameter prediction unit is used to predict the seismic reservoir physical property parameters of the target block according to the seismic reservoir physical property parameter prediction model.

7. An electronic 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, it implements the steps of the seismic reservoir physical parameter prediction method according to any one of claims 1 to 3.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the seismic reservoir physical parameter prediction method according to any one of claims 1 to 3.

Images