Reservoir permeability prediction method based on flow unit index classification
Through the index classification of flow unit and the bidirectional long and short-term memory recurrent neural network model, the problem of unclear classification of flow unit is solved, the accuracy of permeability prediction is improved, and the efficiency and safety of oil and natural gas extraction are promoted.
Patent Information
- Application Number
- CN202411109350.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-13
- Publication Date
- 2024-11-15
AI Technical Summary
The existing technology lacks reasonable classification standards in the classification of flow units, and cannot clearly explain the normal distribution characteristics of flow units, resulting in low calculation accuracy of permeability model and affecting the effective development of reservoir oil and gas resources.
By analyzing the experimental data of core porosity and permeability at depth points, the flow unit index is calculated, the permeability logging interpretation model is constructed, and the flow unit type is predicted using the bidirectional long and short-term memory recurrent neural network model, and the porosity model is calculated. Permeability.
It improves the accuracy of permeability prediction and the efficiency of reservoir oil and gas resources, reduces costs and drilling risks, and ensures energy security.
Smart Images

Figure CN118966003A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of petroleum well logging interpretation, and in particular to a reservoir permeability prediction method based on flow unit index classification. Background Art
[0002] At present, there are two methods to solve the calculation of reservoir permeability: statistical methods and machine learning methods. Among statistical methods, it is usually based on rock physics experiments to construct regression models of permeability and porosity. Some are linear, and some are nonlinear, such as the Timur model and the SDR model. In addition, permeability prediction can also use machine learning methods. For example, the use of neural networks and other methods can better capture the nonlinear relationship between reservoir permeability and other logging responses. In addition, in order to better describe the reservoir characteristics, seepage characteristics and heterogeneity characteristics of the reservoir, the flow unit (FU) is introduced, and statistical methods such as cluster analysis and discriminant analysis are used to identify the type of reservoir flow units. The flow unit index (FZI) is introduced as an intermediate transition parameter for permeability calculation. These are all key steps to improve the accuracy of permeability.
[0003] However, tight sandstone reservoirs have small porosity, complex pore structure, and tiny pore throats, so the logging interpretation and accurate prediction of their permeability is a challenging research task. Reservoir permeability is generally correlated with porosity, and a regression method is usually used to establish a permeability interpretation model. However, the permeability results obtained by this method usually deviate significantly from the actual test, and this type of method cannot fully capture the relationship between the seepage law and the complex pore throat structure. Machine learning methods, such as neural networks, usually require a large amount of labeled data for training, while experimental data such as reservoir permeability are expensive to obtain and the amount of data is limited, which may limit the performance of the neural network model; secondly, when the training data is limited, traditional neural networks are prone to overfitting, that is, the model performs well on the training data, but performs poorly on unseen data. This may lead to insufficient generalization ability for reservoir permeability prediction. This may include the introduction of complex machine learning models, such as deep learning networks, recurrent neural networks, etc., to adapt to complex reservoir characteristics and geological sedimentary characteristics.
[0004] The existing technical solutions do not provide a reasonable classification standard for flow unit classification, do not comprehensively consider geological characteristics and pore characteristics, are unclear when interpreting the normal distribution characteristics of flow units, and the conventional FZI cumulative probability map drawing method cannot correctly classify flow units. The optimal flow unit classification cannot be determined, and the given flow unit type cannot be distinguished by well logging data. Therefore, the calculation accuracy of the constructed permeability model is low, which affects the effective development of reservoir oil and gas resources. In the face of this challenge, it is imperative to study new permeability prediction methods and models in well logging interpretation. Summary of the invention
[0005] The technical problem to be solved by the present invention is that no reasonable classification standard is given for the classification of flow units, the normal distribution characteristics of flow units are not explained clearly, and the calculation accuracy of the constructed permeability model is low.
[0006] The technical solution adopted by the present invention to solve the technical problem is: a reservoir permeability prediction method based on flow unit index classification, comprising:
[0007] Step S1, obtaining flow unit indexes of flow units at the plurality of depth points according to core porosity experimental data and permeability experimental data, and classifying the flow units according to the flow unit indexes to obtain a plurality of groups of flow unit types, wherein the plurality of depth points are distributed in a target well section, and each group of flow unit classifications includes a plurality of different flow unit types;
[0008] Step S2, constructing a permeability logging interpretation model for each group of flow unit classifications according to the multiple groups of flow unit classifications, and selecting the best flow unit classification and the permeability logging interpretation model for the best flow unit classification from the multiple groups of flow unit classifications according to a screening criterion;
[0009] Step S3, using the logging data of several depth points to train a bidirectional long short-term memory recurrent neural network model, using the trained bidirectional long short-term memory recurrent neural network model to predict the flow unit index of the target well section of the reservoir, and judging the flow unit type of each depth section of the target well section of the reservoir according to the optimal flow unit classification;
[0010] Step S4, using the logging data of the target well section of the reservoir, and calculating the porosity logging interpretation result of the target well section of the reservoir through the porosity model;
[0011] Step S5, using the flow unit type of each depth segment of the target well section of the reservoir, determine the permeability logging interpretation model corresponding to each depth segment from the permeability logging interpretation model classified by the optimal flow unit, and predict the permeability of the target well section of the reservoir through the porosity logging interpretation result.
[0012] Preferably, in step S1, the flow units are classified according to the flow unit index to obtain multiple groups of flow unit types including:
[0013] Calculate the logarithm of the flow unit index for all depth points;
[0014] Draw a normal cumulative probability plot between the logarithm of the flow unit index and the cumulative frequency according to the logarithm of all flow unit indices;
[0015] Classify the data points on the normal cumulative probability graph multiple times according to a preset classification algorithm;
[0016] According to different classification results, all flow units are classified differently to obtain the multiple groups of flow unit classifications.
[0017] Preferably, the expression of the flow unit index in step S1 is as follows:
[0018]
[0019] Where FZI is the flow unit index;
[0020] τ is the tortuosity of the porous medium;
[0021] F s is the pore shape factor;
[0022] S gv is the specific surface area;
[0023] φ z is the normalized porosity index,
[0024] RQI is the reservoir quality factor,
[0025] φ e is the effective porosity;
[0026] K is the permeability.
[0027] Preferably, the step of selecting the best flow unit classification from the multiple groups of flow unit classifications according to the screening criteria in step S2, and the step of forming a permeability logging interpretation model of the best flow unit classification comprises:
[0028] Calculate the predicted permeability value of the coring section of the permeability logging interpretation model of each group of flow unit classification based on the core porosity experimental data of the coring section;
[0029] The permeability error of the permeability logging interpretation model of each group of flow unit classification is calculated according to the permeability experimental data of the coring section and the permeability prediction value of the coring section of the permeability logging interpretation model of each group of flow unit classification;
[0030] The permeability logging interpretation model with the smallest permeability error is used as the permeability logging interpretation model for the optimal flow unit classification;
[0031] The flow unit classification corresponding to the permeability logging interpretation model of the best flow unit classification is used as the best flow unit classification.
[0032] Preferably, in step S3, using the well logging data to train the bidirectional long short-term memory recurrent neural network model comprises:
[0033] Conduct correlation analysis on the core flow unit index at several depth points and the well logging data at several depth points, and select the well logging data whose correlation meets the preset requirements as sensitive well logging data;
[0034] A training set is constructed using the sensitive logging data to train the bidirectional long short-term memory recurrent neural network model.
[0035] Preferably, the performing correlation analysis on the core flow unit index at a plurality of depth points and the well logging data at a plurality of depth points, and selecting the well logging data whose correlation meets preset requirements as sensitive well logging data comprises:
[0036] Each type of parameter in the logging data is intersected with the flow unit index, and regression analysis is performed based on the intersection results; the regression fitting coefficient corresponding to each type of parameter is calculated, and all parameters whose regression fitting coefficients meet the preset requirements are used as sensitive logging data.
[0037] Preferably, the step of predicting the flow unit index of the target well section of the reservoir comprises:
[0038] The sensitive logging data of the target well section is input into the bidirectional long short-term memory recurrent neural network model to calculate the flow unit index of the target well section of the reservoir.
[0039] Preferably, the sensitive logging data include: natural gamma, compensated density, compensated neutron, shallow resistivity, porosity and T2 geometric mean.
[0040] Preferably, the step of constructing a permeability logging interpretation model for each group of flow unit classifications comprises:
[0041] Classifying the plurality of depth points according to the classification of each group of flow unit classifications;
[0042] Based on the preset permeability logging interpretation model formula, the core porosity experimental data and permeability experimental data of each type of depth point are fitted respectively to obtain multiple porosity-permeability relationship equations as the permeability logging interpretation model for each group of flow unit classification.
[0043] Preferably, the formula of the permeability logging interpretation model is:
[0044]
[0045] Where K is the permeability;
[0046] φ e is the effective porosity;
[0047] S gv is the specific surface area;
[0048] τ is the tortuosity of the porous medium;
[0049] F s is the pore shape factor;
[0050] F s τ 2 is the Kozeny constant.
[0051] The implementation of the present invention has the following beneficial effects: by taking core samples from a target well section, conducting experiments on the samples, measuring the flow unit index of the core samples, classifying the flow units, optimizing the optimal number of flow units, and constructing an optimal permeability interpretation model; using a bidirectional long short-term memory network to predict the flow unit index of the target well section; using a porosity model to calculate the porosity of the target well section, combining the optimal permeability interpretation model, flow unit index, and porosity of the target well section, the permeability of the target well section can be more accurately predicted, which is of great significance for improving the efficiency of oil and natural gas extraction, reducing costs, reducing drilling risks, protecting the environment, and ensuring energy security. BRIEF DESCRIPTION OF THE DRAWINGS
[0052] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:
[0053] Figure 1 is a flow chart of a first embodiment of the present invention;
[0054] Figure 2 is a flow unit index frequency distribution histogram and a normal cumulative probability graph according to an embodiment of the present invention;
[0055] Figure 3 is a flow unit classification diagram according to an embodiment of the present invention;
[0056] Figure 4 is a permeability error graph according to an embodiment of the present invention;
[0057] Figure 5 is a permeability prediction result diagram of an embodiment of the present invention;
[0058] Figure 6 It is a schematic diagram of predicting a flow unit index using a bidirectional long short-term memory recurrent neural network model according to an embodiment of the present invention. DETAILED DESCRIPTION
[0059] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention. However, it should be clear to those skilled in the art that the present invention may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known methods are omitted to prevent unnecessary details from obstructing the description of the present invention.
[0060] like Figure 1 As shown, in the first embodiment of the present invention, a reservoir permeability prediction method based on flow unit index classification is provided, which is applicable to reservoirs such as tight sandstone, shale, and carbonate. The method comprises:
[0061] Step S1, obtaining flow unit indexes of flow units at several depth points according to core porosity experimental data and permeability experimental data at several depth points, classifying the flow units according to the flow unit indexes to obtain multiple groups of flow unit types, wherein the several depth points are distributed in the target well section range, and each group of flow unit classifications includes multiple different flow unit types.
[0062] In this step, it should be noted that core samples at several different depths are selected from the target well section, and the core samples are analyzed through core porosity experiments and permeability experiments. The flow unit index of the core samples at each depth point is calculated based on the experimental result data and the expression of the flow unit index, so as to classify the flow units at each depth point according to the flow unit index. Among them, these depth points need to be distributed in various depth intervals of the target well section, so as to improve the prediction results of various depth intervals of the target well section. The target well section can be a partial well section of the reservoir, or it can be the entire well section of the reservoir.
[0063] The expression of the flow unit index is as follows:
[0064]
[0065] Where FZI is the flow unit index; τ is the tortuosity of the porous medium; F s is the pore shape factor;
[0066] S gv is the specific surface area; φ z is the normalized porosity index, φ e is the effective porosity;
[0067] K is permeability; RQI is reservoir quality factor,
[0068] In a specific embodiment, classifying flow units includes the following steps: S11, calculating the logarithmic value of the flow unit index of all depth points; S12, drawing a normal cumulative probability diagram between the logarithmic value of the flow unit index and the cumulative frequency based on the logarithmic value of all flow unit indices; S13, classifying the data points on the normal cumulative probability diagram multiple times according to a preset classification algorithm; S14, classifying all flow units differently according to different classification results to obtain multiple groups of flow unit classifications.
[0069] It should be noted that different types of flow units represent different lithology, physical properties and pore structure characteristics. Flow units are distinguished and divided by the flow unit index (FZI). When dividing, it is necessary to draw a frequency distribution histogram between the logarithm of the flow unit index and the frequency. The FZI of the same type of flow units should obey the normal distribution ( Figure 2 a, blue line), that is, the FZI frequency of reservoirs in the same geological sedimentary environment follows a normal distribution. Figure 2 It can be seen from the above that when there are multiple inhomogeneous flow units, the FZI distribution characteristics generally do not conform to a normal distribution, but are composed of several normal distributions. It is difficult to analyze the FZI normal distribution characteristics and distinguish the flow units only by frequency distribution histogram. Therefore, the present invention introduces the cumulative probability diagram ( Figure 2 b). Convert the frequency distribution histogram of the flow unit index into a cumulative frequency graph. If the FZI frequency distribution has normal characteristics, its cumulative probability form is approximately a straight line, while different normal distribution data points are represented by straight lines with different slopes. Straight lines with different slopes correspond to different flow unit types. Therefore, the combination of the normal characteristics of FZI in the frequency distribution histogram and its straight line form in the cumulative probability graph can be used as the basis and standard for flow unit classification.
[0070] In this embodiment, after the cumulative probability graph is constructed, clustering or other classification algorithms are used to classify the data points in the normal cumulative probability graph, and then the flow units are indirectly classified according to the classification results, such as K-means algorithm, linear discriminant analysis algorithm, K nearest neighbor algorithm, etc. By implementing the technical solution of this embodiment, the flow units can be classified to obtain multiple groups of flow unit classifications with different classification standards, so as to screen out the best flow unit classification according to the calculation requirements.
[0071] For example, Figure 3 The following table shows the classification of Figure 2 The normal cumulative probability diagram shown is classified into 3 types of flow unit classification, 4 types of flow unit classification, 5 types of flow unit classification, and 6 types of flow unit classification. As shown in the figure, after classification according to different standards, the cumulative frequency of the FZI logarithm of the same type of flow units is approximately a straight line in the coordinate system, and the slope of the straight line is different for different classifications.
[0072] Step S2: construct a permeability logging interpretation model for each group of flow unit classifications according to the multiple groups of flow unit classifications, and select the best flow unit classification and the permeability logging interpretation model for the best flow unit classification from the multiple groups of flow unit classifications according to the screening criteria.
[0073] In this step, it should be noted that it is necessary to construct a permeability logging interpretation model for each group of flow unit classifications, and compare the permeability simulated by the permeability logging interpretation model of each group of flow unit classifications with the permeability calculated by the core sample experimental data, screen out the simulated permeability with the smallest error from the actual permeability, and select the flow unit classification under this classification standard as the best flow unit classification. Among them, one classification result corresponds to a group of flow unit classifications, and a group of flow unit classifications includes multiple different types of flow units. Each type of flow unit represents a different lithology, physical property and pore structure characteristics, and a permeability logging interpretation model is established separately for each type of flow unit.
[0074] In a specific embodiment, the step of constructing a permeability logging interpretation model for each group of flow unit classifications includes: classifying a number of depth points according to the classification of each group of flow unit classifications; based on a preset formula of the permeability logging interpretation model, fitting the core porosity experimental data and permeability experimental data of each type of depth point respectively, and obtaining multiple porosity-permeability relationship equations as the permeability logging interpretation model for each group of flow unit classifications.
[0075] The formula of the permeability logging interpretation model is:
[0076] Where K is the permeability; φ e is the effective porosity; S gv is the specific surface area; τ is the tortuosity of the porous medium; F s is the pore shape factor; F s τ 2 is the Kozeny constant.
[0077] In this embodiment, a permeability logging interpretation model is established separately for each type of flow unit. By carefully classifying the reservoir and establishing a corresponding permeability model, the permeability of the reservoir can be predicted more accurately.
[0078] In a specific embodiment, screening the best flow unit classification from multiple groups of flow unit classifications according to the screening criteria includes the following steps:
[0079] S21. Calculate the predicted permeability value of the coring section of the permeability logging interpretation model of each group of flow unit classification based on the core porosity experimental data of the coring section. Specifically, obtain the core porosity experimental data of the new depth point, calculate the flow unit index of each point, determine the permeability logging interpretation model corresponding to each point according to the flow unit index division range of the group of flow unit classification, and substitute it into the calculation to obtain the predicted permeability value. Among them, these new depth points need to be distributed in multiple different types of flow units of a group of flow unit classifications.
[0080] S22, calculating the permeability error of the permeability logging interpretation model of each group of flow unit classification according to the permeability experimental data of the coring section and the permeability prediction value of the coring section of the permeability logging interpretation model of each group of flow unit classification. The error values of the permeability logging interpretation model corresponding to each type of flow unit are calculated by using a variance method, a standard deviation method or other methods, and the permeability error of the permeability logging interpretation model of each group of flow unit classification is obtained by adding them.
[0081] S23. The permeability logging interpretation model with the smallest permeability error is used as the permeability logging interpretation model for the best flow unit classification.
[0082] S24, taking the flow unit classification corresponding to the permeability logging interpretation model of the best flow unit classification as the best flow unit classification, that is, the group of flow unit classifications with the smallest permeability error after addition is the best flow unit classification. Figure 4 As shown, the left figure shows the relationship between the permeability value predicted by the permeability logging interpretation model for each group of flow unit classification and porosity after the flow units are classified, and the right figure shows the permeability value predicted by the permeability logging interpretation model (i.e., calculated permeability) and the permeability experimental data measured based on core experimental analysis (i.e., core analysis permeability).
[0083] The intelligently predicted flow unit index itself has a certain error, and its error will be transmitted to the permeability calculation. If the error of the calculated permeability needs to be minimized, the flow units need to be reasonably classified, and at the same time, it is necessary to ensure that the flow unit types have normal distribution characteristics. Secondly, the flow unit types divided by the core experiment flow unit index also need to be sensed and identified by the well logging data. Too many flow unit classifications will lead to the inability to distinguish the well logging, so that the flow units that meet the geological sedimentary characteristics lose their classification significance. Implementing the technical solution of this embodiment and analyzing the calculated permeability errors of different numbers of flow unit classifications can screen out the flow unit classification with the smallest error, ensure that the well logging data can distinguish the flow unit types, select the best flow unit classification, and establish the final permeability logging interpretation model.
[0084] Step S3, using the logging data of several depth points to train a bidirectional long short-term memory recurrent neural network model, using the trained bidirectional long short-term memory recurrent neural network model to predict the flow unit index of the target well section of the reservoir, and judging the flow unit type of each depth section of the target well section of the reservoir according to the optimal flow unit classification.
[0085] In this step, it should be noted that the recurrent neural network (RNN) helps to process sequence-related data. The sample at this moment can be analyzed through the output before this moment. This chain structure helps it to effectively use the information between data at different locations in data with serialized characteristics, and well logging data is serialized data.
[0086] The RNN forward propagation formula can be expressed as:
[0087]
[0088]
[0089] Where t is the sequence time, s t is the output of the hidden layer at time t, o t is the output of the output layer at time t, f and g represent the activation functions of the hidden layer and the output layer respectively, x t is the output of the input layer at time t, U is the weight parameter from the input layer to the hidden layer, W is the weight parameter of the hidden layer from time t-1 to time t, V is the weight matrix from the hidden layer to the output layer, and b and c are biases.
[0090] Based on the recurrent neural network, the Long Short-Term Memory (LSTM) recurrent neural network cleverly adds forget gates, memory gates, and output gates, so that the weight of the self-loop is variable, avoiding the problem of gradient vanishing or gradient expansion.
[0091] Reservoir flow units are closely related to the geological sedimentary environment, and the sedimentary time series is bidirectional. The flow unit characteristics at a certain depth point are related to the upper and lower strata. The recurrent neural network using a single memory flow cannot effectively establish the time series association of the flow unit index. Therefore, the technical solution of the present application constructs a bidirectional long short-term memory network (BidirectionalLong Short-Term Memory, Bi-LSTM), connecting two recurrent neural networks with opposite time series to the same output. The input of each time node will be transmitted to the forward and reverse long short-term memory network (Long Short-Term Memory, LSTM) units respectively, and the output node can obtain historical and future information at the same time.
[0092] Bi-LSTM can improve the prediction accuracy of flow unit index. There is usually a certain correlation between logging data and flow unit index. Before intelligent prediction, it is necessary to select sensitive logging parameters as input data through correlation analysis.
[0093] In a specific embodiment, constructing a bidirectional long short-term memory recurrent neural network model using well logging data at a plurality of depth points includes:
[0094] S31. Perform correlation analysis on the core flow unit index at several depth points and the well logging data at several depth points, and select the well logging data whose correlation meets the preset requirements as sensitive well logging data.
[0095] Specifically, step S31 includes: S311, intersecting each type of parameter in the logging data with the flow unit index, and performing regression analysis based on the intersection results; S312, calculating the regression fitting coefficient corresponding to each type of parameter, and taking all parameters whose regression fitting coefficients meet the preset requirements as sensitive logging data. In step S311, each type of parameter in the logging data can be matched with the flow unit index, which may involve data merging, aligning time series or spatial positions, etc. The sensitive logging data obtained in this embodiment include: natural gamma, compensated density, compensated neutron, shallow resistivity, porosity and T2 geometric mean. Of course, in other embodiments, the sensitive logging data may also include other parameters.
[0096] S32. Use sensitive logging data to build a training set and train the bidirectional long short-term memory recurrent neural network model.
[0097] Then, if Figure 6 As shown, the sensitive logging data of the target well section is input into the bidirectional long short-term memory recurrent neural network model, and the flow unit index of the target well section of the reservoir is calculated.
[0098] Step S4, using the logging data of the target well section of the reservoir, calculate the porosity logging interpretation results of the target well section of the reservoir through the porosity model. In this step, it should be noted that the porosity model can be constructed according to the existing technology, such as the Kozeny-Carman equation, the four-parameter model, the nine-parameter model, etc., so that the porosity logging interpretation results of the full range of the target well section can be calculated.
[0099] Step S5, using the flow unit type of each depth section of the target reservoir section, determine the permeability logging interpretation model corresponding to each depth section from the permeability logging interpretation model classified by the best flow unit, and predict the permeability of the target reservoir section through the porosity logging interpretation result.
[0100] The technical solution of this embodiment is implemented by taking core samples from the target well section, conducting experiments on the samples, measuring the flow unit index of the core samples, classifying the flow units, optimizing the optimal number of flow units, and constructing an optimal permeability interpretation model; using a bidirectional long short-term memory network to predict the flow unit index of the target well section; using a porosity model to calculate the porosity of the target well section, and combining the optimal permeability interpretation model, flow unit index, and porosity of the target well section to more accurately predict the permeability of the target well section, which is of great significance for improving the efficiency of oil and natural gas extraction, reducing costs, reducing drilling risks, protecting the environment, and ensuring energy security.
[0101] In a specific application example, in order to demonstrate the effect of the present invention, a tight sandstone reservoir in a certain oil field is taken as an example. In the tight sandstone layer of the region, early inner front channel conglomerate deposits, middle inner front distributary channel conglomerate deposits, and late front mouth bar sandstone deposits are developed. The logging porosity is calculated by the density porosity interpretation model, and the flow unit index (FZI) is predicted by Bi-LSTM. The flow unit classification method is adopted, and the FZI frequency histogram and FZI cumulative probability diagram ( Figure 1 ) are divided into 3 types of flow units, 4 types of flow units, 5 types of flow units, and 6 types of flow units, and the corresponding permeability models are constructed. Figure 4 As shown in the figure, the permeability error is calculated: the prediction error is 0.72 when the flow units are divided into 3 categories, the prediction error is 0.57 when the flow units are divided into 4 categories, the prediction error is 0.51 when the flow units are divided into 5 categories, and the prediction error is 0.53 when the flow units are divided into 6 categories. It can be seen that when the flow units are divided into 5 categories, the permeability prediction error is the smallest, so it is preferred to divide them into 5 categories as the best flow unit classification. The corresponding logging response and physical property characteristics are shown in Table 1 below.
[0102]
[0103] The sensitive logging parameters input into the Bi-LSTM recurrent neural network include: natural gamma, compensated density, deep-shallow resistivity ratio, porosity, T2 geometric mean, and the FZI prediction results are as follows: Figure 5As shown in the figure, the reservoir is classified into five types of flow units (flow unit channels) by FZI. The higher the value (value: 1-5), the better the seepage characteristics. The upper section of the well is mainly distributed in the estuary bar, followed by the distributary bay, and the main sedimentary microfacies in the lower section is underwater distributary channel, interspersed with a small number of distributary bays. The distributary bay has a high natural gamma value, mainly composed of gray-green mudstone and siltstone, with poor seepage capacity, corresponding to the first-level flow unit; the natural gamma curve of the underwater distributary channel is a toothed box type, with large lithology changes, and this microfacies spans multiple flow unit intervals; the natural gamma curve of the estuary bar is funnel-box type, with obvious toothing and good seepage capacity. From the comprehensive interpretation results of permeability, it can be seen that the seepage unit classification is obvious, and the flow unit type is also well related to the sedimentary facies, which shows that the logging data can effectively distinguish the flow unit type, and also proves the feasibility and scientificity of the five-type flow unit classification in the layer section of the region. Based on the permeability interpretation model of different layers, the reservoir porosity results are input to calculate the reservoir permeability (permeability channel). From the logging calculation results, it can be seen that compared with the nuclear magnetic SDR model, the permeability calculated by the flow unit matches the core permeability better and meets the logging evaluation accuracy.
[0104] It can be understood that the above embodiments only express the preferred implementation modes of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the patent scope of the present invention. It should be pointed out that, for ordinary technicians in this field, the above technical features can be freely combined without departing from the concept of the present invention, and several deformations and improvements can be made, which all belong to the protection scope of the present invention. Therefore, all equivalent changes and modifications made to the scope of the claims of the present invention should belong to the coverage of the claims of the present invention.
Claims
1. A reservoir permeability prediction method based on flow unit index classification, characterized in that: include: Step S1, obtaining flow unit indexes of flow units at the plurality of depth points according to core porosity experimental data and permeability experimental data, and classifying the flow units according to the flow unit indexes to obtain a plurality of groups of flow unit types, wherein the plurality of depth points are distributed in a target well section, and each group of flow unit classifications includes a plurality of different flow unit types; Step S2, constructing a permeability logging interpretation model for each group of flow unit classifications according to the multiple groups of flow unit classifications, and selecting the best flow unit classification and the permeability logging interpretation model for the best flow unit classification from the multiple groups of flow unit classifications according to a screening criterion; Step S3, using the logging data of several depth points to train a bidirectional long short-term memory recurrent neural network model, using the trained bidirectional long short-term memory recurrent neural network model to predict the flow unit index of the target well section of the reservoir, and judging the flow unit type of each depth section of the target well section of the reservoir according to the optimal flow unit classification; Step S4, using the logging data of the target well section of the reservoir, and calculating the porosity logging interpretation result of the target well section of the reservoir through the porosity model; Step S5, using the flow unit type of each depth segment of the target well section of the reservoir, determine the permeability logging interpretation model corresponding to each depth segment from the permeability logging interpretation model classified by the optimal flow unit, and predict the permeability of the target well section of the reservoir through the porosity logging interpretation result.
2. The reservoir permeability prediction method according to claim 1, characterized in that: In step S1, the flow units are classified according to the flow unit index to obtain multiple groups of flow unit types including: Calculate the logarithm of the flow unit index for all depth points; Draw a normal cumulative probability plot between the logarithm of the flow unit index and the cumulative frequency according to the logarithm of all flow unit indices; Classify the data points on the normal cumulative probability graph multiple times according to a preset classification algorithm; According to different classification results, all flow units are classified differently to obtain the multiple groups of flow unit classifications.
3. The reservoir permeability prediction method according to claim 1, characterized in that: The expression of the flow unit index in step S1 is as follows: Where FZI is the flow unit index; τ is the tortuosity of the porous medium; F s is the pore shape factor; S gv is the specific surface area; φ z is the normalized porosity index, RQI is the reservoir quality factor, φ e is the effective porosity; K is the permeability.
4. The reservoir permeability prediction method according to claim 1, characterized in that: The step of selecting the best flow unit classification from the multiple groups of flow unit classifications according to the screening criteria in step S2, and the permeability logging interpretation model of the best flow unit classification comprises: Calculate the predicted permeability value of the coring section of the permeability logging interpretation model of each group of flow unit classification based on the core porosity experimental data of the coring section; The permeability error of the permeability logging interpretation model of each group of flow unit classification is calculated according to the permeability experimental data of the coring section and the permeability prediction value of the coring section of the permeability logging interpretation model of each group of flow unit classification; The permeability logging interpretation model with the smallest permeability error is used as the permeability logging interpretation model for the optimal flow unit classification; The flow unit classification corresponding to the permeability logging interpretation model of the best flow unit classification is used as the best flow unit classification.
5. The reservoir permeability prediction method according to claim 1, characterized in that: The step S3 uses the well logging data to train the bidirectional long short-term memory recurrent neural network model, which includes: Conduct correlation analysis on the core flow unit index at several depth points and the well logging data at several depth points, and select the well logging data whose correlation meets the preset requirements as sensitive well logging data; A training set is constructed using the sensitive logging data to train the bidirectional long short-term memory recurrent neural network model.
6. The reservoir permeability prediction method according to claim 5, characterized in that: The performing correlation analysis on the core flow unit index at a plurality of depth points and the well logging data at a plurality of depth points, and selecting the well logging data whose correlation meets the preset requirements as sensitive well logging data comprises: Each type of parameter in the logging data is intersected with the flow unit index, and regression analysis is performed based on the intersection results; the regression fitting coefficient corresponding to each type of parameter is calculated, and all parameters whose regression fitting coefficients meet the preset requirements are used as sensitive logging data.
7. The reservoir permeability prediction method according to claim 5, characterized in that: The step of predicting the flow unit index of the target well section of the reservoir comprises: The sensitive logging data of the target well section is input into the bidirectional long short-term memory recurrent neural network model to calculate the flow unit index of the target well section of the reservoir.
8. The reservoir permeability prediction method according to claim 5, characterized in that: The sensitive logging data include: natural gamma, compensated density, compensated neutron, shallow resistivity, porosity and T2 geometric mean.
9. The reservoir permeability prediction method according to claim 1, characterized in that: The step of constructing a permeability logging interpretation model for each group of flow unit classifications comprises: Classifying the plurality of depth points according to the classification of each group of flow unit classifications; Based on the preset permeability logging interpretation model formula, the core porosity experimental data and permeability experimental data of each type of depth point are fitted respectively to obtain multiple porosity-permeability relationship equations as the permeability logging interpretation model for each group of flow unit classification.
10. The reservoir permeability prediction method according to claim 1, characterized in that: The formula of the permeability logging interpretation model is: Where K is the permeability; φ e is the effective porosity; S gv is the specific surface area; τ is the tortuosity of the porous medium; F s is the pore shape factor; F s τ 2 is the Kozeny constant.
Citation Information
Cited By
Complex reservoir permeability prediction method and system
CN120492936A
