Reservoir porosity and permeability prediction method based on dynamic committee integration model

Abstract

The invention relates to a reservoir porosity and permeability prediction method based on a dynamic committee integration model, and the method comprises the following steps: obtaining an original data set which comprises shale content, porosity, permeability, GR, AC, CNL, DEN, RT, RXO, SP, CALC, CALI, depth and lithologic labels; performing data enhancement on the original data set; determining main control factors influencing the porosity and the permeability; constructing a dynamic committee integration model; inputting the main control factors into a dynamic committee integration model; and utilizing the trained dynamic committee integration model to respectively predict the porosity and the permeability. According to the method, the logging data is processed by adopting a machine learning method, reservoir parameters can be efficiently and accurately predicted, and favorable support can be provided for oil-gas exploration and development; the dynamic committee integrated model constructed by the invention can dynamically adjust the weight of each model according to different geological conditions and data features, can more flexibly adapt to the geological condition of a research area compared with a single model, and improves the prediction precision and generalization of the model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geological exploration, and particularly relates to a reservoir porosity and permeability prediction method based on a dynamic committee integrated model. BACKGROUND

[0002] Accurate prediction of reservoir geological parameters can provide a reliable basis for reservoir evaluation. The content of silt and clay in the sediment is characterized, and this parameter is usually closely related to the lithological characteristics. Studies have shown that there is a correlation between reservoir properties, that is, the higher the overall trend, the worse the physical properties, which increases the internal heterogeneity of the reservoir, affects the fluid flow path, makes it difficult to form an effective fracture network in the reservoir, and leads to poor fracturing effect. Porosity directly determines the size of the fluid storage space in the rock, and different porosity distribution conditions will increase or decrease the fluid flow channel, thereby affecting the production. Accurate evaluation of porosity helps to identify high-potential reservoir sections, evaluate the connectivity and heterogeneity of the reservoir, guide the selection of fracturing design and completion method, and optimize the fracturing design.

[0003] For the Upper Wuerhe Formation in the M block of the Junggar Basin, the stratum is mainly lacustrine deposition, and a large number of conglomerate and sandstone-mudstone interbeds are developed. This sedimentary environment leads to a large heterogeneity of reservoir properties. Permeability reflects the ability of fluid to flow through rock. Due to the diversity of lithology distribution and the complexity of micro-pore structure in the study area, the reservoir flow unit is not easy to divide, and permeability is also related to the connectivity between pores. Accurate prediction of permeability helps to determine the fluid transmission capacity of the reservoir, and improving the connectivity between reservoirs helps to improve the overall fluid flow efficiency. The physical properties of different rock types in the block are quite different, and the region has experienced multiple tectonic movements, forming a complex tectonic pattern, with local faults and folds, enhancing the complexity of reservoir evaluation and productivity prediction. SUMMARY

[0004] The present application provides a reservoir porosity and permeability prediction method based on a dynamic committee integrated model to solve the above technical problems.

[0005] The present application is implemented by the following technical solutions:

[0006] The reservoir porosity and permeability prediction method based on the dynamic committee integrated model provided by the present application comprises the following steps:

[0007] A dynamic committee integrated model is constructed, the main control factors affecting porosity and permeability are input into the dynamic committee integrated model, the dynamic committee integrated model is trained, and the trained dynamic committee integrated model is used to predict porosity and permeability.

[0008] Optionally, before the model training, the following steps are further included: obtaining an original data set, wherein the original data includes shale content, porosity, permeability, GR, AC, CNL, DEN, RT, RXO, SP, CALC, CALI, depth, and lithology label; performing data enhancement on the shale content, porosity, permeability, GR, AC, CNL, DEN, RT, RXO, SP, CALC, CALI; and determining the main control factors affecting the porosity and the permeability.

[0009] Preferably, the main control factors for the porosity prediction are DEN, Depth, CALC, AC, CNL, and shale content.

[0010] Preferably, the main control factors for the permeability prediction are DEN, CNL, SP, lithology, GR, and porosity.

[0011] Optionally, the method for determining the main control factors affecting the porosity and the permeability is as follows: using PCCs and two rank correlation coefficients based on Copula function to calculate the linear and nonlinear correlation degrees between the porosity, the permeability, and the corresponding influencing parameters; taking absolute values of the three correlation coefficient calculation values and performing weighted average; and if the absolute value weighted average value of the correlation coefficient of one of the influencing parameters is higher than the average value of the absolute value weighted average values of the correlation coefficients of all the influencing parameters, the influencing parameter is considered as the main control factor.

[0012] Preferably, the CWGAN-GP network is used for data enhancement on the original data set.

[0013] Optionally, the CWGAN-GP data enhancement method includes the following steps: in the discriminator and generator models of the WGAN-GP, the lithology label, skewness, and kurtosis corresponding to the sample are taken as constraint conditions; and the lithology label and the kurtosis and skewness of the data set are taken as a condition vector to be input into the CWGAN-GP model for auxiliary training.

[0014] Optionally, the method for constructing the dynamic committee integrated model includes the following steps:

[0015] The fuzzy C-means clustering is taken as a gate network of dynamic integration, the data set is divided into K clusters, and the optimal matching relationship between the expert models and the sub-data sets is determined by the fuzzy C-means clustering method;

[0016] A weight is assigned to each expert model in each cluster;

[0017] For each sample in the data set, the membership degree of the sample in each cluster is calculated.

[0018] Based on the membership degree of the sample in each cluster and the model weight of the corresponding cluster, the membership degree of the sample and the model weight value of the corresponding cluster are multiplied together to obtain the weight allocation result of each model to each sample in each cluster. Furthermore, the weight values ​​of each model to a certain sample in all clusters are accumulated to obtain the dynamic weight of the expert model for that sample.

[0019] The dynamic weights of each expert model on the samples are assigned, and the prediction results of each expert model are weighted and integrated to form a dynamic committee ensemble model.

[0020] Preferably, four expert models—BP neural network, LightGBM, GRNN, and Dilateformer—are used to construct the dynamic committee ensemble model.

[0021] Compared with the prior art, this application has at least the following beneficial effects:

[0022] 1. This application constructs a dynamic committee ensemble model, which can efficiently and accurately predict porosity and permeability, providing favorable support for oil and gas exploration and development; furthermore, the model can dynamically adjust the weights of each model according to different geological conditions and data characteristics, which can improve the prediction accuracy of the model; in addition, compared with a single model, the dynamic committee ensemble model of this application can more flexibly adapt to the geological conditions of the study area, improving the generalization of the model.

[0023] 2. The dynamic committee ensemble model of this application integrates four network structures: BP, LightGBM, GRNN, and Dilateformer. It still has high prediction performance even when the data is noisy and uncertain. Attached Figure Description

[0024] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is an example diagram of the LightGBM histogram algorithm in the embodiment;

[0026] Figure 2 The diagram shows the BP neural network structure in the embodiment.

[0027] Figure 3 The GRNN network structure is shown in the example.

[0028] Figure 4 This is a schematic diagram of multi-scale extended attention MSDA in the embodiment;

[0029] Figure 5 Structure diagram of multi-scale expansion attention mechanism in the embodiment;

[0030] Figure 6 Flow chart of CWGAN-GP training in the embodiment;

[0031] Figure 7 Comparison chart of various statistical indexes before and after porosity enhancement in the embodiment;

[0032] Figure 8 Comparison chart of various statistical indexes before and after permeability enhancement in the embodiment;

[0033] Figure 9 Porosity lithology label clustering result chart in the embodiment;

[0034] Figure 10 Permeability lithology label clustering result chart in the embodiment;

[0035] Figure 11 Porosity correlation relationship chart in the embodiment;

[0036] Figure 12 Permeability correlation relationship chart in the embodiment;

[0037] Figure 13 Porosity and permeability prediction model training flow chart in the embodiment;

[0038] Figure 14 Comparison chart of porosity prediction results of various models in the embodiment;

[0039] Figure 15 Comparison chart of permeability prediction results of various models in the embodiment;

[0040] Figure 16 Dynamic committee integrated model prediction flow chart in the embodiment;

[0041] Figure 17 Framework chart for predicting porosity and permeability based on the dynamic committee integrated model in the embodiment;

[0042] Figure 18 Result chart of predicting reservoir porosity and permeability by the dynamic committee integrated model in the embodiment;

[0043] Figure 19 Parameter prediction result chart before data enhancement in the embodiment;

[0044] Figure 20 X1 well reservoir geology parameter logging comprehensive interpretation chart in the embodiment;

[0045] Figure 21The well reservoir geology parameter logging comprehensive interpretation chart of the embodiment. DETAILED DESCRIPTION

[0046] In order to make the objects, 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.

[0047] 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. It should be noted that each embodiment in the present specification adopts a progressive manner for description, and each embodiment mainly describes the differences from other embodiments, and the same and similar parts between each embodiment can be referred to each other.

[0048] Embodiment 1

[0049] The reservoir porosity and permeability prediction method based on the dynamic committee integrated model disclosed in the embodiment comprises the following steps

[0050] 1. The present embodiment calculates and compares the shale content, porosity and permeability collected this time by using the reservoir geology parameter calculation model established in the prior art based on the analysis of core data and logging response characteristics in the same study area.

[0051] Since the neutron, density and shale content are correlated, the shale content V is calculated based on the neutron and density by using the multivariate regression method. sh

[0052] V sh = 3.62 * CNL + 208.12 * DEN - 575.932

[0053] In the above formula, V sh is the shale content, CNL is the compensated neutron, and DEN is the formation density.

[0054] In view of the correlation between multiple logging curves and porosity, but the calculation effect of a single logging curve on porosity is poor, the porosity POR is calculated by using the principal component analysis method.

[0055] POR = 1.098AC - 0.14CNL - 1.6DEN - 0.129V sh

[0056] In the above formula, POR is the porosity, V sh is the shale content, CNL is the compensated neutron, and DEN is the formation density.

[0057] The permeability K is calculated according to the lithology. ​

[0058] Sandstone: K = 0.0006e 0.544φ

[0059] Fine conglomerate: K = 0.0082e 0.863φ

[0060] Medium conglomerate: K = 0.1079e 0.393φ

[0061] In the above formula, K is the permeability, is the porosity under different lithology.

[0062] 2. Constructing a machine learning model

[0063] (1) LIGHTGBM is a machine learning algorithm improved based on gradient boosting tree (GBDT), which sequentially fits new decision trees in the negative direction of the decision tree loss function. By histogram algorithm, each continuous feature value is divided into a limited number of histogram box diagrams, as shown in Figure 1 , the optimal split point is selected from the candidate set of each feature, and the corresponding box is selected as the gain of the split point to accelerate model training. And in gradient one-side sampling (GOSS), only the samples with large gradient are calculated for information gain, and only part of the samples with small gradient are randomly reserved, so as to retain the distribution of the original data as much as possible. By mutual exclusive feature binding, the mutual exclusive features are combined to improve the training efficiency, ensure good generalization ability, and improve the accuracy of the model.

[0064] Compared with traditional machine learning models such as linear regression and SVM, LIGHTGBM is more flexible, can better capture the complex mapping relationship in the data, and can automatically select features that affect the reservoir, logging data based on the dynamic committee integrated model of reservoir porosity and permeability prediction method, reducing the need for manual intervention. Compared with convolutional neural network (CNN), recurrent neural network (RNN), and long short-term memory network (LSTM), LIGHTGBM is more lightweight and suitable for resource-limited environments. Deep learning models often have black box nature, and the internal decision-making process is difficult to explain, while LIGHTGBM can better reveal the influence of internal features on reservoir geological parameters, and can more intuitively analyze feature importance to guide further logging evaluation research.

[0065] (2) The basic structure of BP neural network is composed of input layer, hidden layer and output layer structure. As shown in Figure 2 . Because BP neural network has strong non-linear modeling ability, it can capture the complex non-linear relationship in reservoir geological parameters. This is very useful for understanding and predicting complex geological features.

[0066] (3) GRNN neural network, i.e., generalized regression neural network. The network is based on non-parametric kernel regression for neural network training. The correlation density function of independent variables and dependent variables is calculated from the training samples, so as to calculate the regression value of the dependent variable with respect to the independent variable. The GRNN neural network includes four layers of neurons, the first layer is the input layer, the number of neurons in the layer is the same as the dimension of the input vector in the learning sample. That is, each neuron can be regarded as a simple distributed neural network. The second layer is the pattern layer, the number of neurons in the layer is the same as the number of training samples. The results of all training samples are summed in the sum layer, and the output of the predicted result is realized in the output layer. The structure diagram of the GRNN neural network is shown in Figure 3 .

[0067] Since the GRNN can quickly adapt to the data distribution and does not require a complex parameter adjustment process, it is suitable for complex and variable logging and geological data. By adjusting the smoothing factor of the network, smooth prediction results can be provided, which is suitable for prediction tasks of continuous variables with depth sequence. Compared with traditional machine learning methods, GRNN is more flexible and good at capturing complex relationships in data. Compared with deep learning models, GRNN is simpler to adjust and has higher computational efficiency.

[0068] (4) Dilateformer is a multi-scale dilated attention mechanism neural network. In general, the sliding window in the convolutional neural network structure can only process fixed scale features, while the Dilateformer network introduces dilated convolution operation in the multi-head attention mechanism module to capture the dependency between features at different scales, so as to better understand the complex structure and pattern of input features, and improve the running efficiency of the model through sparse selection. The Dilateformer model generally presents a four-layer pyramid structure, and the structure is shown in Figure 5 . The first two layers are multi-scale dilated attention (MSDA), which can be understood as performing a sliding window operation in the multi-head attention mechanism, and the execution process is shown in Figure 4 . The last two layers are ordinary multi-head self-attention (MHSA). An overlap marker is used in each layer, a plurality of overlapping 3x3 convolution modules are used for zero padding, and the resolution of the output is adjusted by alternately controlling the step length of the convolution kernel to be 1 or 2. And use a convolutional overlap downsampler with an overlap kernel size of 3 and a stride of 2 to make each layer module adapt to input of different resolutions. In addition, when the feature information is input into MSDA or MHSA, the conditional embedding (CPE) is used to positionally encode the input and the logging curve, so that the model can better adapt to the feature changes at different depth positions.

[0069] In MSDA, the sliding window dilated attention operation (SWDA) is adopted, that is, the keys and values of each head are sparsely selected in a sliding window centered on the red query patch, and then self-attention calculation is performed on these representative patches, and the calculation formula is as follows:

[0070] X = SWDA (Q, K, V, r)

[0071] In the above formula, X is the calculated self-attention score; Q, K, and V represent query, key, and value matrices, respectively, and each row of the three matrices represents a query / key / value feature vector; r is the dilated rate corresponding to different sliding windows; when querying at the (i, j) position of the original feature map, SWDA sparsely selects keys and values in a sliding window centered on (i, j) with a size of w x w for self-attention calculation, i is the horizontal coordinate value corresponding to the center point of the sliding window, j is the vertical coordinate value corresponding to the center point of the sliding window, and w is the length and width size of the rectangular sliding window.

[0072] Since the reservoir geology parameters usually have a complex depth sequence relationship, the multi-head attention mechanism in the Dilateformer model can effectively capture the long-distance dependency relationship between the logging data, which helps to understand the potential relationship between different positions in the reservoir. And by adjusting the dilated rate, the dilated convolution can extract features at different scales, expand the receptive field of the model, and make the model cover a larger depth sequence area, which is difficult to achieve by traditional machine learning models such as linear regression, support vector machine, etc. Unlike models such as LSTM and RNN, Dilateformer does not rely on the order information of the depth sequence, so it can more flexibly process logging data with longer depth sequence.

[0073] 3. Reservoir porosity and permeability prediction process

[0074] 3.1 Introduction of data set and data preprocessing

[0075] Considering the correlation between the shale content and the porosity in the study area, 9 common logging parameters I and the depth, lithology label are used for the calculation of porosity, shale content and permeability in this embodiment, and the shale content is added to the features affecting the porosity to evaluate the calculation results of the porosity.

[0076] The 9 common logging parameters are GR, AC, CNL, DEN, RT, RXO, SP, CALC, and CALI, wherein:

[0077] GR represents natural gamma, reflecting the total content of radioactive elements in the formation; AC represents acoustic time difference, indicating the difference between the propagation speed of acoustic waves in rock and the propagation speed in standard medium. It is commonly used to identify the lithology and thickness of the formation and identify oil and gas bearing layers; CNL represents compensated neutron, which is the thermal neutron flux caused by the neutron source measured along the well profile, used to estimate the hydrogen index and porosity of the formation; DEN represents formation density, which is the density value of different formations corresponding to the lithology, used to determine the thickness and lithology of the formation; RT represents deep resistivity, which is the resistivity of the formation without being invaded by drilling fluid, used to judge the oil content of the formation; RXO represents the flushing zone resistivity, which is the resistivity in the flushing zone after the drilling fluid invades the formation, used to judge the influence degree of the drilling fluid on the formation; SP represents the natural potential, which is the potential difference generated by the resistivity of the drilling fluid adjacent to the pore formation, used to divide the permeable layer, judge the lithology and the nature of the formation water; CALC: represents the difference between the actual hole diameter and the standard hole diameter. It is used to identify the irregularity of the wellbore and provide a basis for formation evaluation; CALI represents the hole diameter, which usually refers to the diameter of the measured wellbore, used to judge the lithology, check the casing condition, etc.

[0078] To verify the nonlinear relationship between the physical properties caused by the heterogeneity of the glutenite reservoir, the Spearman correlation coefficient is used to calculate the nonlinear relationship between the porosity and the permeability of the study area, as shown in formula (1). The correlation between the porosity and the permeability of the study area is 0.762, indicating that there is a high correlation between the porosity and the permeability of the study area. Therefore, on the basis of nine logging parameters, depth and lithology label, the porosity is also considered as a parameter affecting the permeability, and the porosity information extracted by the model is used to assist the prediction of the permeability, so as to obtain the data set of the permeability prediction.

[0079]

[0080] In the above formula, r is the calculated value of the Spearman correlation coefficient, POR i is the porosity of the core sample, PORM i is the permeability of the core sample, is the average value of the porosity, is the average value of the permeability, i is the serial number of the core sample, and n is the total number of the statistical core samples.

[0081] The experimental data set in this embodiment is from the measured core sample data of the Upper Wuerhe Formation in the M block of the study area. Among them, 213 core samples contain porosity data, 160 core samples record permeability data, and 87 core samples record shale content data. Combined with the lithology label of the corresponding depth point, the logging parameter value constitutes the reservoir geological parameter sample data set of this embodiment. The partial data of the porosity POR and the permeability PERM prediction of the C well are shown in Table 1.

[0082] Table 1: PERM, POR prediction part parameter table

[0083]

[0084] The data set for predicting porosity and permeability is divided into a training set and a test set in a ratio of 7:3. Due to measurement errors and differences in evaluation indicators in the data acquisition process, a maximum and minimum value normalization method is used to make the distribution of each feature data within a reasonable range.

[0085] 3.2, using improved CWGAN-GP to enhance data

[0086] As shown in Figure 6 , the CWGAN-GP generative adversarial network (GAN) is composed of two neural networks of generator and discriminator, and after multiple rounds of adversarial training, a distribution similar to the original sample is generated. WGAN-GP is an improved generative adversarial network. In the generative adversarial network (GAN) model, JS divergence is used to measure the difference between distributions. The discriminator needs to satisfy the Lipsehitz condition to ensure the effectiveness of the wasserstein distance and the stability of the training. The purpose of network training is to maximize the probability of the discriminator judging the real sample (D(x)) to better distinguish real samples from generated samples. Minimize the generator loss function so that the generated sample is closer to the real sample, thereby cheating the discriminator. To further improve the quality of sample generation, the embodiment sets the conditional constraints and the extraction ability of the features. However, if the constraint condition only contains the label information of the sample, it is difficult to learn the internal distribution rule of the feature data in the adversarial training process, which limits the further optimization iteration of the generator. Therefore, in addition to the lithology label corresponding to the sample as a constraint condition, additional dimensionless skewness and kurtosis feature indicators are added to the original model of WGAN-GP in the discriminator and generator model to realize conditional generation, which constitutes the CWGAN-GP network structure of the embodiment. The skewness and kurtosis calculation formulas are shown in equations (2) and (3) to stably reflect and limit the distribution of geological and engineering fracturing data.

[0087]

[0088] In the above formula, K represents the kurtosis of the data set, S represents the skewness of the data set, x j represents a certain sample in the data set, Std is the standard deviation, M is the sample mean, and N is the total number of samples in the data set.

[0089] Two conditional values, derived from the formulas for kurtosis and skewness, are concatenated with the sample lithology label L to obtain a conditional vector C = [L, K, S]. This conditional vector is then input into the WGAN-GP model to aid training. The generator generates new samples, and the discriminator distinguishes between the generated and real samples. The parameters of the generator and discriminator are updated by calculating the loss function.

[0090] The loss function of the CGAN-GP discriminator is:

[0091]

[0092] In the above formula, L D Z represents the discriminator loss function; P represents the true sample; g The data distribution represents the real samples; x represents the samples generated by the generator; y represents the constraints added during the training of the discriminator; P r The prior distribution representing the noise; Linear interpolation representing real samples and generated samples; The data distribution representing the linear interpolation samples; The expected score of the discriminator on the real data. The expected score of the discriminator on the generated data; Here, λ represents the gradient penalty term of the discriminator, and λ is the coefficient of the gradient penalty term. To introduce a gradient penalty term for linear interpolation, the gradient penalty term is added to the generated linear interpolation samples to ensure that the discriminator applies the gradient penalty uniformly throughout the data space. This ensures that the discriminator satisfies the Lipschitz condition across the entire data distribution, which helps the discriminator to have a smooth gradient not only on real and generated data points, but on all points, thus effectively preventing gradient vanishing or gradient exploding.

[0093] The loss function of the generator is:

[0094]

[0095] In the above formula, L G represents the discriminator loss function; x represents the samples generated by the generator; y represents the constraints added during the generator training process. Let $\frac{ ...

[0096] In this embodiment, an improved CWGAN-GP was used to augment all parameters in the dataset, including the original predicted label data for porosity and permeability, as well as all well logging curve data in the dataset.

[0097] 4. Analysis of Controlling Factors

[0098] Pearson correlation coefficient is a linear correlation function used to analyze the linear correlation between two features. Let there be two feature sets X, Y:

[0099] X = {x1, x2, x3, … x n}, Y = {y1, y2, y3, … y n}, the Pearson correlation coefficient calculation formula is:

[0100]

[0101] In the above formula, and represent the average value of x and y, respectively, and r represents the Pearson correlation coefficient.

[0102] Kendall’s is a rank correlation coefficient used to measure the consistency of rank ordering changes of two variables. For any two adjacent positions i, j in two feature sets X, Y, if the size relationship of Xi and Xj is consistent with that of Yi and Yj, then the data points (Xi, Yi) and (Xj, Yj) are called consistent pairs, otherwise they are called inconsistent pairs. Based on the proportion of consistent pairs and inconsistent pairs in the two data sets, the Kendall’s rank correlation coefficient is calculated to measure the consistency of rank ordering changes of the two variables, and the calculation formula is shown in (7).

[0103]

[0104] In the above formula, τ is the Kendall correlation coefficient; n c is the number of consistent pairs; n b is the number of inconsistent pairs; n0 is the total number of pairs; n1 is the number of pairs in X; and n2 is the number of pairs in Y.

[0105] Spearman correlation coefficient is used to measure the monotonic correlation of two variables, and unlike Pearson correlation coefficient calculation, it does not require data to meet normal distribution. In calculating the Spearman correlation coefficient of two data sets, first, the data in two feature sets X, Y are ranked in descending order, and for each data point i, the difference between its rank in X and Y is calculated, as shown in formula (8), and based on the rank difference, the Spearman correlation coefficient of the two data sets is calculated, as shown in formula (9).

[0106] d i = rank(X i )- rand(Y i ) (8)

[0107]

[0108] In the above formula, rank(X i ), rank(Y i ) are the rank of data point i in the two groups of data; d i is the difference in rank of data i in the two groups of data.

[0109] The value range of the three correlation coefficients is [-1, 1], 1 represents complete positive correlation, -1 represents complete negative correlation, and 0 represents complete irrelevance. In actual research, the correlation strength is usually classified according to the size of the absolute value of the correlation coefficient. That is, 0-0.3 represents weak correlation or negligible correlation; 0.3-0.5 represents low correlation; 0.5-0.9 represents that there is a medium-high correlation between two variables; and 0.9 or more represents that there is a very strong correlation between variables.

[0110] Pearson correlation coefficient and Kendall's and Spearman's two rank correlation coefficients are used to calculate the linear and nonlinear correlation degrees between the respective influencing factors and porosity and permeability. The calculation results of the three correlation coefficient calculation methods are weighted and averaged to combine the performances of the three methods under different data characteristics, so as to obtain more comprehensive and robust correlation evaluation results. The correlation calculation results of the parameters are shown in Figure 11 、 Figure 12 The absolute values of the correlation parameters are taken in the calculation process. If the absolute value of the weighted correlation coefficient of a certain parameter with respect to a certain reservoir geological parameter is greater than the average value of the absolute values of the weighted correlation coefficients of all influencing parameters with respect to the reservoir geological parameter, it is considered that the influence of the parameter on the prediction of the reservoir geological parameter is greater.

[0111] The standard for screening the main control factors of each reservoir geological parameter is that if the absolute value of the weighted correlation coefficient of a certain parameter with respect to a certain reservoir geological parameter is greater than the average value of the absolute values of the weighted correlation coefficients of all influencing parameters with respect to the reservoir geological parameter, it is considered that the influence of the parameter on the prediction of the reservoir geological parameter is greater.

[0112] 5. Evaluation index of calculation model

[0113] In this embodiment, the coefficient of determination (Coefficient of Determination, R 2 ), mean squared error (Mean Squared Error, RMSE), and mean absolute error (Mean Absolute Error, MAE) are selected as evaluation indexes to measure the accuracy of each reservoir geological prediction model. Among them, R 2The fitting degree of the model is represented by MAE, which represents the average of the sum of the absolute values of all prediction errors, and RMSE, which represents the square root of the average of the square of all data prediction errors. The calculation formula of the above evaluation indexes is:

[0114]

[0115] where y i represents the actual value of the reservoir geological parameter, represents the average value of the actual value of the reservoir geological parameter, represents the predicted value of the reservoir geological parameter, i is the sample number, and m is the total number of data set samples.

[0116] 6. Algorithm simulation process

[0117] 6.1. Model training process

[0118] The process of the reservoir porosity and permeability prediction method based on dynamic committee integrated model based on four kinds of machine learning can be divided as follows:

[0119] (1) For the collected reservoir geological parameter data set, data anomaly and missing value processing is performed. The CWGAN-GP model with conditional constraints is used to enhance the data of the reservoir geological parameter data set. Through statistical index analysis, the t-sne visualization tool verifies the rationality of the expanded sample features and class label data. The data set is standardized to eliminate the dimensional differences between different features.

[0120] (2) After data enhancement, the porosity and permeability data are screened by the main control factors, and then the training set and test set are divided. The training set data is input into the BP, Grnn, LightGBM, and Dilateformer regression models for training. The Bayesian optimization algorithm is used to search the hyperparameter search space iteratively to determine the optimal hyperparameter combination. After determining the optimal hyperparameter combination, the coefficient of determination (R2), root mean square error (RMSE), and other indicators are used to evaluate the training effect, and the generalization ability of each model is evaluated on the test set data. To verify the effectiveness of data enhancement and the generalization ability of the model, the original sample data corresponding to each parameter is applied to the trained optimal model, and the model performance is evaluated. The model training process is shown in Figure 13 .

[0121] 6.2. Model hyperparameter introduction

[0122] (1) In the BP neural network prediction model, the main parameter introduction and optimization range setting are as follows:

[0123] Hidden_layers_sizes: refers to the number of hidden layer neurons in the network. The setting of this parameter will affect the expression ability of the model. Setting too many will increase the model training time, leading to overfitting, and setting too few will cause the model to be unable to fully capture the feature representation between data, which may result in underfitting. The BP neural network hyperparameter optimization range of this embodiment is set to (5, 30).

[0124] Learning_rate: the BP neural network learning rate hyperparameter optimization range of this embodiment is set to (0.001, 0.1).

[0125] Maximum number of iterations: the hyperparameter range of the maximum number of iterations of this embodiment is set to (10, 500).

[0126] Regularization coefficient: this parameter is used to regulate the complexity of the model during training. Setting too large will lead to oversimplified model, resulting in underfitting, and too small will make the model complexity insufficient, resulting in overfitting. The regularization coefficient optimization hyperparameter range of this embodiment is set to (0.0001, 0.01).

[0127] (2) In the LightGBM model, the main hyperparameters and optimization range settings are as follows:

[0128] N_estimators: refers to the number of decision trees, which affects the complexity of the model. Too many trees may result in overfitting, and too few trees are prone to underfitting, making the model unable to fully capture the data features. The decision tree hyperparameter optimization range of this embodiment is set to (0, 300).

[0129] Max Depth: refers to the maximum growth depth of each decision tree. The role is to regulate the complexity of the model during training to prevent overfitting. The decision tree maximum growth depth hyperparameter range of this embodiment is set to (1, 10).

[0130] Min Child Samples: refers to the minimum leaf node tree, which limits the growth of the decision tree and the complexity of the model. The minimum leaf node optimization range of this embodiment is set to (1, 20).

[0131] Lambda_L1, Lambda_L2: refers to the large gradient sample sampling rate and the large gradient random sample sampling rate. Through these two parameters, the model randomly selects part of the samples and features for training during each iteration. Random sampling improves the generalization ability of the model. The optimization range of Lambda_L1 and Lambda_L2 is set to (0, 1).

[0132] (3) The prediction result of the GRNN neural network is related to the smoothing factor. If the parameter is set too large, the prediction result will be close to the mean value of the sample data. If it is too small, the model will be too sensitive to local changes in the input sample, making the prediction curve not smooth. In this embodiment, the smoothing factor optimization range is set to (0.1, 10).

[0133] (4) In the Dilateformer model, the main hyperparameters and the optimization range are set as follows:

[0134] Droupout refers to the technique of randomly shutting down neurons in the neural network during training. That is, during the forward propagation process, some neurons are stopped working according to a certain proportion to avoid the phenomenon of model overfitting. In this paper, the Droupout optimization range is set to (0.1, 1). In order to improve the training efficiency and ensure the prediction accuracy, the network layer range is set to (2, 4), the number of hidden layer neurons is set to (5, 30), and the learning rate range is set to (0.001, 0.1).

[0135] In the CWGAN-GP network, in addition to reasonably setting the generator, discriminator iteration times, generator and discriminator learning rate, iteration times and other hyperparameters. It also needs to consider two hyperparameters: gradient decay factor and square gradient decay factor. These two parameters are used to calculate the gradient norm of the linear interpolation point between the real sample and the generated sample, so as to control the punishment degree of the model to the gradient norm during the training process. Usually, the range of these two parameters is set to (0.1, 10), so as to ensure that the gradient norm is as close to 1 as possible, so as to effectively prevent the phenomenon of gradient explosion.

[0136] 6.3. Data enhancement results

[0137] In the process of data enhancement of porosity and permeability original data set, the average sample sampling rate ASOR is used to evaluate the similarity of the distribution of original sample and generated sample, that is, the average of the proportion of each feature generated data falling within the value range of original data. The calculation formula of ASOR is shown in (13)-(15):

[0138]

[0139] In the above formula, ASOR is the average sample sampling rate, D(G(Z i )) is the generated sample of the adversarial network, n is the number of generated samples, [min, max] is the value range of the original data, d is the total number of features of each sample data set, i is the serial number corresponding to different samples, j is the serial number corresponding to different features, and SOR j is the sample sampling rate corresponding to different features.

[0140] The hyperparameter settings of CWGAN-GP when performing data augmentation are shown in Table 2. The iteration rate of the generator and the discriminator is usually 5 by default to improve the training efficiency of the generator and reduce the risk of mode collapse during training.

[0141] Table 2: CWGAN-GP model parameter setting table

[0142] Model parameters Conditions Model parameters Conditions Discriminator learning rate 0.0008 Generator learning rate 0.003 Number of generator iterations 5000 Gradient decay factor 0.0001 Generator and discriminator iteration rate 5 Squared gradient decay factor 0.2

[0143] The CWGAN-GP is used to perform data augmentation on the porosity and permeability data set, and the data augmentation results of the GAN network on the porosity and permeability data set are compared. The ASOR threshold is set to 95%, and the number of samples in each reservoir geological parameter generated data set is determined. Finally, the number of generated samples of shale content, porosity, and permeability is determined as 3268, 5350, and 4800, respectively. Figures 7-8 The statistical index comparison chart of each feature before and after data augmentation of porosity and permeability is shown in the following table.

[0144] It can be seen that compared with the GAN enhanced data, CWGAN-GP can accurately retain the positive and negative of the skewness of the original data, and can accurately fit the skewness value of each feature of the original data. Under the condition of ensuring that each type of data feature is learned, the diversity of the original data is improved as much as possible.

[0145] Although GAN is closer to the original data in some features than CWGAN-GP, the distribution of GAN synthetic data is significantly different from the original data for most features. Overall, the ability of GAN to learn the distribution of the original data is not stable, and the quality of the generated data is not as good as CWGAN-GP. The data generated by CWGAN-GP is closer to the corresponding features of the original data in most features. However, there are still some obvious differences in some features. For example, in the shale content and permeability generated sample set, the mean and median of the RT and RXO generated samples are significantly larger, which is mainly because the original sample set has large values and large fluctuations in deep and shallow resistivity. In the shale content and permeability data set, there are obvious errors in the mean and median of the GR and AC generated samples. This is because in the reservoir geological parameter data set, most of the index data is small, and the mean of GR and AC is above 68, which belongs to the large value class. The large value class accounts for a small proportion of the total features, resulting in that CWGAN-GP can learn the overall distribution trend of the data, but the learning ability of the large value class is poor.

[0146] Maximum mean discrepancy (MMD) is a popular criterion to measure the difference between two similar data distributions in mathematical statistics tasks. Two different data distributions are mapped to a reproducing Hilbert space H through a mapping function. In this high-dimensional space, the probability distribution distance between the original sample and the generated sample is evaluated by the maximum mean discrepancy, which is calculated as follows:

[0147]

[0148] In the above formula, X s and X t represent the reservoir geology parameter data sets before and after data enhancement, respectively, and represent the i-th sample in X s and X t , n and m represent the number of data in the two domain data sets, is a mapping function, i.e. mapping data from a low-dimensional space to a high-dimensional space.

[0149] The MMD values of the three reservoir geology parameter distributions before and after CWGAN-GP and GAN data enhancement are shown in Tables 4-5. Compared with the GAN network, the probability distribution of the generated sample and the real sample is closer, and the MMD values of the three parameters are between 0.25-0.5, indicating that the sample generated by CWGAN-GP can be used as an extended sample of the original core sample, which can expand the number of training samples and improve the training ability of the subsequent model and the generalization performance of the model.

[0150] Table 3: Comparison of MMD values of each reservoir geology parameter data before and after data enhancement

[0151] Parameters MMD values for CWGAN-GP MMD values for GAN Porosity 0.28 0.55 Permeability 0.42 1.06 Shale content 0.37 0.72

[0152] The clustering effect of the lithology label before and after the reservoir geology parameter data set is enhanced by the t-SNE visualization tool, and the clustering results are shown in Figure 9 , Figure 10 . For example, in the original data set of porosity, there is a clear clustering of large and medium gravel and coarse gravel, and there is also some overlap between small and medium gravel and fine gravel. After the CWGAN-GP data enhancement, the clustering effect of each reservoir geology parameter data set can further improve the clustering trend of the same type of lithology label, effectively preserve the separation characteristics between different lithologies, and also learn that there is an overlapping relationship between different lithology samples similar to the original sample.

[0153] 7、Experimental results analysis

[0154] 7.1 Main control factor screening results

[0155] The linear and nonlinear correlation degrees between porosity, permeability and corresponding influencing parameters are calculated by Pearson correlation coefficient and Kendall's, Spearman's two rank correlation coefficients respectively. The correlation parameters are all taken as absolute values in the calculation process. Figure 11 , Figure 12 The correlation relationship between porosity, permeability and corresponding influencing factors is shown in FIGS.

[0156] According to the calculation results of Figure 11 , Figure 12 DEN, Depth, CALC, AC, V sh (shale content), CNL have great influence on porosity prediction; DEN, CNL, SP, lithology, GR, POR (porosity) have great influence on permeability prediction, and each of the above parameters is selected as the main control factor of the reservoir porosity and permeability prediction method based on the dynamic committee set integration model.

[0157] 7.2 Porosity prediction

[0158] Porosity data enhancement:

[0159] In this embodiment, the data set after porosity expansion is divided in the ratio of 7:3 according to the ratio of 7:3. The porosity is predicted by four models. Among them, the prediction accuracy of BP neural network is 0.88, the prediction accuracy of LightGBM is 0.86, the prediction accuracy of GRNN is 0.89, and the prediction accuracy of Dilateformer is 0.90. The comparison of prediction results is shown in Figure 14 (a)-(d). Table 4 lists the optimal hyperparameters selected by the four models when predicting porosity.

[0160] Table 4: Porosity prediction hyperparameter optimization results table

[0161]

[0162] 7.3 Permeability prediction

[0163] Permeability data enhancement

[0164] The permeability prediction data set in this section is divided in the same way as in the previous two sections, and the expanded permeability label data is predicted by four models. Among them, the prediction accuracy of BP neural network is 0.83, the prediction accuracy of LightGBM is 0.78, the prediction accuracy of GRNN is 0.84, and the prediction accuracy of Dilateformer is 0.86. The comparison of prediction results is shown in Figure 15 (a)-(d). Table 5 shows the optimal hyperparameters selected by each model when predicting permeability.

[0165] Table 5: Permeability prediction hyperparameter optimization results table

[0166]

[0167] Table 6 shows the best performance indicators of four machine learning models after hyperparameter optimization in the test set to predict porosity and permeability. It can be seen that the Dilateformer neural network shows more accurate prediction performance in prediction, and the prediction accuracy is slightly better than other models; while the GRNN neural network also shows high prediction accuracy in predicting porosity. That is, these two neural network models can accurately fit the nonlinear relationship between porosity and its influencing factors based on the internal complex nonlinear function.

[0168] Table 6: Comparison of parameter prediction evaluation indicators

[0169]

[0170] Table 7 records the prediction evaluation indicators (RMSE, MAE) of the four models for core point porosity and permeability data. And the prediction results of the four models for core point porosity and permeability are compared with the RMSE and MAE calculation results obtained by the porosity multiple regression calculation method and the lithology index fitting formula model for permeability established by the predecessors in the same study area.

[0171] Table 7: Comparison of porosity and permeability core data prediction evaluation indicators

[0172]

[0173] By comparing and analyzing the above evaluation indicators, it can be concluded that the prediction results of the four machine learning models for porosity are better than the calculation results of the traditional method, that is, the selected machine learning model has certain accuracy in porosity prediction. For permeability prediction, due to the poor heterogeneity and poor physical properties of the whole research area, the measured permeability of the collected core samples has certain differences in numerical value, and the overall distribution is uneven. Thus, the prediction effect of the four models for permeability is not as good as the prediction result for porosity. The calculation effect of the lithology-based permeability calculation model for mudstone and coarse gravel is better than the prediction result of the machine learning model, that is, the calculation model also has usability and accuracy in permeability calculation of the Urho group in the research block of the embodiment.

[0174] 8. Improved committee model based on dynamic committee integrated model for reservoir porosity and permeability prediction

[0175] 8.1 Committee machine prediction model

[0176] The committee machine prediction model is an integrated neural network algorithm, whose core idea is "collecting ideas from all walks of life", that is, a machine learning technology for improving the overall prediction performance by combining multiple experts (i.e. single models). The committee model is divided into static committee model and dynamic committee integration model. The static committee model assigns fixed weights to each expert model by fixedly weighting the prediction results of multiple experts. For data with multiple different patterns or distributions, fixed weights may not fully capture the characteristics of all patterns, and the training performance of each expert is difficult to improve. The dynamic committee divides the input data into multiple sub-tasks through the gate network, and assigns them to the corresponding expert model; and integrates the results of multiple different models to output the final prediction result. Fuzzy C-means clustering is usually used as the gate network for dynamic integration to allocate weights. This method divides the sub-data according to the principle that the difference between classes is large enough and the difference within classes is small enough, and the formula is as follows:

[0177]

[0178] In the above formula, J(U, V: X) is the loss function of fuzzy clustering, which is used to measure the advantages and disadvantages of the current clustering result; U is the membership matrix, V is the clustering centroid, X is the input logging data, M is the number of data, C is the number of clusters, q is the fuzzy coefficient, x i is the i-th data point of the sample data set, v k is the k-th cluster center, is the degree to which data point i belongs to cluster k, i is the serial number of the sample data point, and k is the serial number of the cluster center.

[0179] If the sample is directly combined with the model to obtain the prediction result, the relationship between the model and the sample will be weakened to some extent. Therefore, the dynamic weighting integration method is used to realize weight allocation in this embodiment. First, the model is weighted. For a single sample x, the membership degree u k (x) of each cluster is calculated, and the membership degrees of the same sample in different clusters need to satisfy that the weighted sum of all membership degrees is 1. Then, the optimal weight w i (x) of each expert learner i to the sample x is calculated. Based on the calculated dynamic weight, the optimal matching relationship between each expert and the sub-data set is determined. And each expert model is combined as the final prediction model. The model prediction process is shown in Figure 16 .

[0180] Considering that LightGBM is good at dealing with complex relationships between logging features, BP neural network can learn deep coupling between reservoir geological parameters through hidden layer structure, GRNN is suitable for approximation of complex functions, and its probability density estimation characteristics can effectively reduce overfitting in the case of limited original lithology data calibration. Dilateformer is good at capturing long-distance dependence between logging data, and can effectively capture long-term sedimentary cycles in the vertical direction of logging curves. Due to the complex geological conditions and strong reservoir heterogeneity in M block, the dynamic committee integration model is constructed by using the above four models. Specifically, the construction method of dynamic committee integration model includes the following steps:

[0181] Fuzzy C-means clustering is used as the gate network of dynamic integration. This method divides the data set into K clusters based on the principle that the inter-class difference is large enough and the intra-class difference is small enough. The data set is divided into K clusters by fuzzy C-means clustering. Through the fuzzy C-means clustering method, the optimal matching relationship between each expert model and the sub-data set is determined.

[0182] Assign weights to each expert model: In each cluster after division, the performance of each expert learner is evaluated by weighted mean square error (WeightedMSE). Based on the weighted mean square error value of the model in each cluster, the weight of the model in each cluster is assigned. Generally, the better the model performs, the higher the weight.

[0183] Membership calculation based on samples: For each sample in the data set, calculate its membership in each cluster, i.e. the degree to which the sample belongs to each cluster. For the same sample, the weighted sum of its membership values in each cluster should generally equal 1.

[0184] Dynamic weighted integration method is used to assign weights to each expert model: According to the membership of the sample in each cluster and the model weight of the corresponding cluster, the sample membership and the weight value of each model in the corresponding cluster are multiplied, so as to obtain the weight distribution result of each model in each cluster for each sample. The weight value of each model in all clusters for a certain sample is accumulated to obtain the dynamic weight of the expert model for the sample.

[0185] The dynamic weight of each expert model to the sample is used to integrate and combine the prediction results of each expert model as a dynamic committee integration model. This model can dynamically adjust the weights of each model according to different geological conditions and data characteristics, so as to better adapt to the complex geological conditions of M block.

[0186] The specific steps of implementing dynamic weight distribution using fuzzy C-means clustering are as follows:

[0187] Step 1, feature clustering: Use the features of the training set to divide the data into K clusters through FCM clustering.

[0188] Step 2, evaluate model performance: Within each cluster, evaluate the performance of each base learner by weighted mean square error (WeightedMSE), as shown in Equation (18).

[0189]

[0190] In the above formula, M is the number of samples in cluster K; t is the predicted value of expert model i for sample j. j It is the true value of sample j, u jk Let be the membership degree of sample j to cluster k.

[0191] Step 3, Weight Allocation: Based on the model's MSE within each cluster k,j Assign weights V to the model for each cluster. k,j For example, the calculation formula (19) suggests that the better the model performs, the higher its weight.

[0192]

[0193] In the above formula, ε is a small constant, introduced to avoid the MSE of a certain model being... k,j When ε is 0, the denominator becomes 0, leading to a division-by-zero error. Typically, ε takes values ​​much smaller than the order of magnitude of the predicted MSE, such as when the MSE of the predicted porosity and permeability is on the order of 10. -2 -10 1 When this period is in between, the MSE is at 10. -6 -10 -8 Random numbers are generated between the given values.

[0194] Step 4, Prediction Phase: First, for each test sample x, calculate its membership degree to each cluster, that is, the degree of membership u of the sample to each cluster. k (x), where if the total number of clusters is K, then k takes values ​​from 1 to K, and satisfies the following conditions: Calculate the final weight w for each model based on its membership degree and the model weight of the corresponding cluster. i (x), as in formula (20); the prediction results of the base learner are weighted based on dynamic weights, and the final ensemble prediction value T(x) is obtained by formula (21).

[0195]

[0196] By using a dynamic ensemble approach, the weights of each base expert are dynamically adjusted based on the characteristics of the input samples, the optimal matching relationship between each expert and the subset is determined, and more accurate prediction results are generated.

[0197] When predicting porosity and permeability, the ensemble models that contribute the most to the prediction results are Dilateformer, GRNN, Dilateformer, and BP, respectively. Table 8 shows the dynamic weight allocation for each model. The overall framework of the improved dynamic committee ensemble model for predicting porosity and permeability is as follows: Figure 17 As shown.

[0198] Table 8: Model for Weight Table during Prediction with Different Parameters

[0199]

[0200] 8.2 Model Prediction Results

[0201] like Figure 18 As shown in the figure, the results of reservoir porosity and permeability prediction methods based on the dynamic committee ensemble model are compared. Table 9 records the performance evaluation indicators of the ensemble model for predicting various parameters:

[0202] Table 9: Comparison of Evaluation Indicators for Predicted Parameters in the Dynamic Committee Integrated Model

[0203] Parameters [R 2 ]]> RMSE MAE Porosity 0.94 0.43 0.24 Permeability 0.91 3.16 1.50

[0204] Figure 19 The prediction results of the ensemble learning model for porosity and permeability of the original core data points were visualized.

[0205] It can be seen that the constructed dynamic committee ensemble model significantly improves the prediction performance of porosity compared to the single model. The model maintains a good fit between the predicted and actual porosity values, with a relatively uniform error distribution. However, the fit between the actual and predicted permeability values ​​is slightly worse. This is related to the strong overall heterogeneity of the study area, the low variation in permeability values, and the generally poor permeability performance of the reservoir. Achieving more accurate permeability prediction is a direction that the prediction model needs further optimization.

[0206] 8.3 Single-well processing interpretation application examples

[0207] Based on the reservoir porosity and permeability prediction method model established above, the reservoir geological parameters of two verification wells, X1 and X2, in block M, which participated in the model training, were interpreted by comprehensive logging. Figure 20 , Figure 21 The results show that the prediction curves of porosity and permeability by the dynamic committee ensemble model are in high agreement with the core data curves. This indicates that the model still has high predictive performance even when the data contains noise and uncertainty.

[0208] The dynamic committee integration model established in the embodiment integrates four network structures of BP, LightGBM, GRNN and Dilateformer. Based on the model, porosity and permeability are predicted, and the results show that the correlation coefficients R of porosity and permeability prediction reach 0.94 and 0.91 respectively, indicating that the dynamic committee integration model can further improve the prediction accuracy, and compared with a single model, can more flexibly adapt to the geological conditions of the study area and improve the generalization of the model. 2 2 The correlation coefficients R of porosity and permeability prediction reach 0.94 and 0.91 respectively, indicating that the dynamic committee integration model can further improve the prediction accuracy, and compared with a single model, can more flexibly adapt to the geological conditions of the study area and improve the generalization of the model.

[0209] The dynamic committee integration model is applied to the reservoir geological parameter logging interpretation of the verification well, and the results show that the dynamic committee integration model can better predict the reservoir geological parameters, and the model can be applied to more logging parameter prediction.

[0210] The above is only the preferred embodiment of the present application and is not used to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for predicting reservoir porosity and permeability based on a dynamic committee ensemble model, characterized in that, Includes the following steps: Construct a dynamic committee integration model; The main controlling factors affecting porosity and permeability are input into the dynamic committee ensemble model, and the dynamic committee ensemble model is trained. The porosity and permeability were predicted using a trained dynamic committee ensemble model.

2. The method for predicting reservoir porosity and permeability based on a dynamic committee ensemble model according to claim 1, characterized in that, The main controlling factors for porosity prediction are DEN, Depth, CALC, AC, CNL, and clay content.

3. The method for predicting reservoir porosity and permeability based on a dynamic committee ensemble model according to claim 1, characterized in that, The main controlling factors for permeability prediction are DEN, CNL, SP, lithology, GR, and porosity.

4. The method for predicting reservoir porosity and permeability based on a dynamic committee ensemble model according to claim 1, characterized in that, Methods for determining the controlling factors include: The linear and nonlinear correlations between porosity, permeability and the corresponding influencing parameters were calculated using the Pearson correlation coefficient and the two rank correlation coefficients of Kendall's and Spearman, respectively. Take the absolute values ​​of the three correlation coefficients and then perform a weighted average. If the weighted average of the absolute values ​​of the correlation coefficients of one of the influencing parameters is higher than the mean of the weighted average of the absolute values ​​of the correlation coefficients of all the influencing parameters, then that influencing parameter is considered to be the controlling factor.

5. The method for predicting reservoir porosity and permeability based on a dynamic committee ensemble model according to any one of claims 1-4, characterized in that, Before training the model, the following steps are also included: Obtain the raw dataset, which includes clay content, porosity, permeability, GR, AC, CNL, DEN, RT, RXO, SP, CALC, CALI, depth, and lithology labels; Data augmentation was performed on clay content, porosity, permeability, GR, AC, CNL, DEN, RT, RXO, SP, CALC, CALI, depth, and lithological labels. Identify the main controlling factors affecting porosity and permeability.

6. The method for predicting reservoir porosity and permeability based on a dynamic committee ensemble model according to claim 5, characterized in that, The CWGAN-GP network is used to augment the data.

7. The method for predicting reservoir porosity and permeability based on a dynamic committee ensemble model according to claim 6, characterized in that, CWGAN-GP's data augmentation method includes the following steps: In the discriminator and generator models of WGAN-GP, the lithological label, skewness, and kurtosis corresponding to the sample are used as constraints. The lithological labels and the kurtosis and skewness of the dataset are used as conditional vectors to assist in the training of the CWGAN-GP model.

8. The method for predicting reservoir porosity and permeability based on a dynamic committee ensemble model according to claim 1, characterized in that, The method for constructing a dynamic committee integration model includes the following steps: Fuzzy C-means clustering is used as a gate network for dynamic integration to divide the data base into subsets, dividing the dataset into K clusters, and the optimal matching relationship between each expert model and the subset is determined by the fuzzy C-means clustering method. Assign weights to the expert models within each cluster; For each sample in the dataset, calculate its membership degree in each cluster; Based on the membership degree of the sample in each cluster and the model weight of the corresponding cluster, the membership degree of the sample and the model weight value of the corresponding cluster are multiplied together to obtain the weight allocation result of each model to each sample in each cluster. Furthermore, the weight values ​​of each model to a certain sample in all clusters are accumulated to obtain the dynamic weight of the expert model for that sample. The dynamic weights of each expert model on the samples are assigned, and the prediction results of each expert model are weighted and integrated to form a dynamic committee ensemble model.

9. The method for predicting reservoir porosity and permeability based on a dynamic committee ensemble model according to claim 8, characterized in that, The following formula is used to assign weights V to the model of each cluster. k,j ; In the above formula, M is the number of samples in cluster K. t is the predicted value of expert model i for sample j. j It is the true value of sample j, u jk Let ε be the membership degree of sample j to cluster k, where ε is a constant; For each test sample x, calculate its membership degree u to each cluster. k (x), satisfying Based on the membership degree and the model weight of the corresponding cluster, the final weight w of each model is calculated using the following formula. i (x): The prediction results of each expert model are weighted based on dynamic weights to obtain the final integrated prediction value: In the above formula, T(x) is the predicted value of the dynamic committee ensemble model.

10. The method for predicting reservoir porosity and permeability based on a dynamic committee ensemble model according to claim 1, 8, or 9, characterized in that, The dynamic committee ensemble model is constructed using four expert models: BP neural network, LightGBM, GRNN, and Dilateformer.

Images