Method for constructing reservoir permeability prediction model and medium

Abstract

The application discloses a kind of reservoir permeability prediction model construction method and medium include: S1 obtains core test data and is divided into several kinds of reservoir types, and each reservoir type is assigned permeability calculation model;S2 analyzes the well logging curve value of each core, obtains original well logging response sample point;S3 is formed according to well logging curve value analysis cognitive database, according to the well logging response sample point of cognitive database supplement reservoir type;And / or, according to original well logging response sample point adds gaussian noise and supplements well logging response sample point;S4 is trained by the permeability calculation model of reservoir type corresponding original well logging response sample point and supplemented well logging response sample point, whether training result meets preset condition is judged, yes, model construction is completed;Otherwise, execute S3.The application adds well logging response sample point supplement step in reservoir permeability prediction model construction, increases the training data of reservoir permeability prediction model, improves the accuracy of permeability prediction result.

Description

Technical Field

[0001] This invention relates to the field of petroleum geological exploration technology, and in particular to a method and medium for constructing a reservoir permeability prediction model. Background Technology

[0002] In the field of petroleum geology, a reservoir refers to a rock formation with interconnected pores that allows oil and gas to be stored and seep within it. For a long time, reservoir permeability has been calculated primarily using the following four methods: ① For reservoirs with good homogeneity, permeability and porosity data obtained from core experiments show a strong correlation; a high-precision permeability calculation model can be obtained by directly fitting the porosity-permeability relationship; ② By analyzing the influence of factors such as rock composition, grain size, structure, cement, and reservoir space on reservoir permeability, a permeability interpretation model based on rock physical phases or flow unit index classification is established; ③ Considering that changes in mineral composition affect the pore radius distribution of rocks, thus leading to changes in permeability, a permeability model based on rock mineral composition is established; ④ Believing that pore structure is the main controlling factor of reservoir permeability, and considering the different contributions of different pore sizes to permeability, a reservoir permeability prediction model based on pore throat structure distribution is established.

[0003] The aforementioned methods for calculating reservoir permeability have the following main drawbacks: ① Deep offshore reservoirs exhibit strong heterogeneity; while reservoir porosity may be similar, permeability can vary by 1 to 3 orders of magnitude. Directly fitting the functional relationship between porosity and permeability cannot yield a high-precision reservoir permeability prediction model. ② Reservoir permeability prediction models based on mineral composition and pore structure require elemental logging and nuclear magnetic resonance (NMR) logging data to obtain formation elemental content information and NMR logging data reflecting formation pore distribution. Since well conditions in deep offshore drilling are typically complex, it is impossible to perform elemental logging and NMR logging operations in every well. This limits the widespread application of reservoir permeability prediction models based on rock mineral composition and pore throat structure distribution in deep offshore reservoirs due to certain objective limitations. ③ The reservoir classification-based method first requires classifying core porosity and permeability data according to the flow unit index or the concept of rock physical similarity. A least squares method is then used to fit a reservoir permeability prediction model corresponding to each reservoir type. Next, the logging response values ​​at corresponding depth points for each core sample are read, and deep learning methods are used to identify the reservoir type. This method requires extensive drilling and wellbore coring during the modeling process to ensure that the reservoir types analyzed by the cores cover all reservoir types encountered in the region. Otherwise, insufficient representativeness of the modeling sample points will lead to poor generalization ability of the established permeability model, resulting in divergence when applied to new wells, and the permeability prediction results will fail to meet accuracy requirements.

[0004] Core reservoir classification and deep learning-based reservoir type logging identification is an effective method for predicting permeability in deep offshore reservoirs. However, due to cost control and objective operating conditions, sufficient representative drilling and wellbore coring data cannot be obtained during offshore oil and gas exploration and development. As a result, this method is limited by training data in actual permeability prediction, leading to lower prediction accuracy. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to address at least one defect of the related technologies mentioned in the background: the accuracy of reservoir permeability prediction models is affected by the amount of training data, and to provide a method and medium for constructing reservoir permeability prediction models.

[0006] The technical solution adopted by this invention to solve its technical problem is: a method for constructing a reservoir permeability prediction model, which includes the following steps: S1: Obtain core experimental data, classify the core experimental data into several reservoir types according to a preset classification method, and assign a permeability calculation model to each reservoir type; the reservoir types are at least two. S2: Analyze the logging curve values ​​corresponding to each core based on the core test data to obtain the original logging response sample points; S3: Analyze the combination characteristics of logging curves for each reservoir type based on the logging curves corresponding to each core, form a cognitive database, and supplement the logging response sample points for different types of reservoirs based on the cognitive database; on this basis, add Gaussian noise to the original logging response sample points to supplement the logging response sample points for different types of reservoirs. S4: Train the original logging response sample points and the supplemented logging response sample points through the permeability calculation model corresponding to each reservoir type, and determine whether the training results meet the preset training termination conditions. If yes, the reservoir permeability prediction model is completed; otherwise, step S3 is executed again.

[0007] In some embodiments, step S3, supplementing logging response sample points for different types of reservoirs based on the cognitive database, includes: Based on the reservoir productivity evaluation parameters of the sampled sand bodies in the cognitive database, the reservoir type is matched, and the logging curve values ​​of the sampled sand bodies in the cognitive database are supplemented as logging response sample points for the corresponding reservoir type.

[0008] In some embodiments, the reservoir productivity evaluation parameter is the relationship between porosity and permeability data; the sampled sand bodies are the same set of sampled sand bodies; Reservoir types are matched based on the relationship between porosity and permeability data of the same set of sampled sand bodies, including: Obtain core data from the same set of sampled sand bodies, and determine the relationship between porosity and permeability data of the sampled sand bodies based on the core data. Then, determine the reservoir type that matches the reservoir where the same set of sampled sand bodies is located based on the relationship between porosity and permeability data.

[0009] In some embodiments, the reservoir productivity evaluation parameter is permeability; the sampled sand body is a sampled sand body at the same depth point; Reservoir type is matched based on the permeability of sand bodies sampled at the same depth point, including: Based on the pressure measurement results and flow rate data obtained from the pressure measurement operation, sampled sand bodies at the same depth point are obtained; The permeability is obtained by substituting the mobility data into the permeability calculation model corresponding to various reservoir types. By comparing pressure measurement results and permeability, the reservoir type matching the reservoir where the sand body sampled at the same depth point is located can be obtained.

[0010] In some embodiments, the reservoir productivity evaluation parameter is productivity data; the sampled sand body is a single sampled sand body; Reservoir type is matched based on the productivity data of individual sampled sand bodies, including: The measured productivity of a single sampled sand body is obtained based on drill pipe testing; the logging response characteristics of the reservoir in which the single sand body is located are consistent. The permeability of the reservoir where the individual sampled sand body is located is calculated by substituting the logging response characteristics of the individual sampled sand body into various reservoir types. Calculate the model production capacity of the reservoir containing a single sampled sand body based on the permeability of the reservoir containing the single sampled sand body; By comparing the measured production capacity with the model production capacity, the reservoir type that matches the reservoir where a single sampled sand body is located is obtained.

[0011] In some embodiments, step S3 involves adding Gaussian noise to supplement logging response sample points for different reservoir types based on the original logging response sample points, prior to which the following steps are included: The original well logging response sample points are divided into training set and test set according to a preset ratio.

[0012] In some embodiments, in step S3, the step of adding Gaussian noise to supplement logging response sample points for different reservoir types based on the original logging response sample points includes: The training set of original well logging response sample points is normalized, and Gaussian noise with a normal distribution is added according to the type of original well logging response sample points in the training set.

[0013] In some embodiments, the normally distributed Gaussian noise follows a normal distribution with small mean and small variance.

[0014] In some embodiments, the types of original logging response sample points in the training set include at least resistivity; Adding normally distributed Gaussian noise to the resistivity includes: adding normally distributed Gaussian noise to the resistivity using a logarithmic space method.

[0015] By implementing this invention, the following beneficial effects are achieved: This invention acquires core experimental data, categorizes it into several reservoir types, assigns a permeability calculation model to each type, and analyzes the logging curve values ​​corresponding to each core to obtain original logging response sample points. Then, it analyzes the combination characteristics of logging curves for each reservoir type based on the logging curves corresponding to each core, forming a cognitive database. Based on this cognitive database, it supplements logging response sample points for different reservoir types, and / or adds Gaussian noise to the original logging response sample points to supplement them. Finally, it trains the original logging response sample points and the supplemented logging response sample points using the permeability calculation model corresponding to each reservoir type, and determines whether the training results meet the preset training termination conditions. If yes, the reservoir permeability prediction model is completed; otherwise, the step of supplementing logging response sample points is repeated. This application adds a logging response sample point supplementation step to the reservoir permeability prediction model construction, increasing the training data for the model and improving the accuracy and generalization ability of the prediction model. Attached Figure Description

[0016] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings: Figure 1 A flowchart of one embodiment of the method for constructing the reservoir permeability prediction model of the present invention is shown; Figure 2 The cumulative frequency distribution of flow unit index values ​​is shown in an embodiment of the method for constructing the reservoir permeability prediction model of the present invention. Figure 3 The diagram illustrates the functional relationship between permeability and porosity in an embodiment of the reservoir permeability prediction model construction method of the present invention. Figure 4 This diagram illustrates an embodiment of the reservoir permeability prediction model construction method of the present invention, which characterizes the same reservoir based on porosity and permeability data. Figure 5 This diagram illustrates an embodiment of the reservoir permeability prediction model construction method of the present invention, which supplements similar reservoirs based on well logging response characteristics. Figure 6A flowchart illustrating an embodiment of the reservoir permeability prediction model construction method of the present invention is shown, which matches the reservoir type based on the permeability of sand bodies sampled at the same depth point. Figure 7 The flowchart illustrates an embodiment of the method for constructing the reservoir permeability prediction model of the present invention, showing a process of matching reservoir type based on the productivity data of a single sampled sand body; Figure 8 This diagram illustrates the accuracy of the training results before supplementary logging response sample points in one embodiment of the reservoir permeability prediction model construction method of the present invention. Figure 9 This diagram illustrates the accuracy of training results after supplementing logging response sample points in one embodiment of the reservoir permeability prediction model construction method of the present invention. Detailed Implementation

[0017] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0018] It should be noted that the flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0019] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0020] like Figure 1 As shown, some embodiments of the present invention disclose a method for constructing a reservoir permeability prediction model. Taking the process of expanding the well logging data sample points of a deep permeability prediction model in an oilfield A in a certain basin as an example, the present invention is illustrated in detail. The method includes the following steps: S1: Obtain core experimental data, classify the core experimental data into several reservoir types according to a preset classification method, and assign a permeability calculation model to each reservoir type; the reservoir types are at least two.

[0021] The core experimental data were analyzed to obtain data on porosity, permeability, rock grain size, thin sections of cast bodies, mercury porosimetry, and X-ray diffraction, which were then used to classify the core experimental data into at least two reservoir types.

[0022] In some embodiments, the preset classification method is the flow unit index or the rock physical phase.

[0023] In some embodiments, the formula for calculating the flow unit index is: in, The standard porosity index is ROI, which is the reservoir quality factor.

[0024] The formula for calculating the standardized porosity index is: ,in, Porosity.

[0025] The formula for calculating the reservoir quality factor is: Where K is the permeability and π is the mathematical constant pi.

[0026] The formula for calculating penetration rate is: ,in, For shape factor, For the degree of curvature, Specific surface area per unit particle volume.

[0027] Substituting the permeability calculation formula into the reservoir quality factor calculation formula, we obtain... Substituting the reservoir quality factor calculation formula into the flow unit index calculation formula, we obtain... (Unit conversion constants not considered) ).

[0028] In some embodiments, the method for assigning a permeability calculation model to each reservoir type is as follows: a mathematical function that can effectively characterize the variation law between porosity and permeability in each reservoir type is selected as the permeability calculation model for each reservoir type.

[0029] like Figure 2 As shown, for example, by plotting the cumulative rate distribution of the flow unit index values, and based on the characteristics of the normal distribution of the flow unit index and the principle of minimizing error propagation, such as... Figure 3 As shown in Table 1, the deep reservoirs of Oilfield A in a certain basin are divided into 5 reservoir types. An exponential function is selected, and the least squares method is used to calculate the functional relationship between core permeability and porosity for each reservoir type. This allows for the allocation of appropriate permeability calculation models for each reservoir type. Table 1. Five reservoir types and their permeability calculation models in Oilfield A of a certain basin. S2: Analyze the logging curve values ​​corresponding to each core based on the core test data to obtain the original logging response sample points.

[0030] For example, the logging curve values ​​corresponding to each depth point of 558 core samples from oilfield A in a certain basin were read. These logging curve values ​​include natural gamma (GR), compensated neutron (TNPH), compensated density (ZDEN), P-wave transit time (DT), deep resistivity (RD), shallow lateral resistivity (RS), and photoelectric absorption cross section index (PE), etc., to obtain 558 original logging response sample points.

[0031] S3: Analyze the combination characteristics of logging curves for each reservoir type based on the logging curves corresponding to each core, form a cognitive database, and supplement the logging response sample points for different types of reservoirs based on the cognitive database; and / or, supplement the logging response sample points for different types of reservoirs by adding Gaussian noise based on the original logging response sample points.

[0032] Based on the well logging curve analysis of each reservoir type, the combined characteristics of well logging curves for each reservoir type are analyzed. That is, by performing superposition analysis, cross plot analysis, morphological analysis or mathematical transformation on multiple different well logging curves, a comprehensive response pattern reflecting a certain geological feature (such as lithology, physical properties, and hydrocarbon content) is extracted to form a cognitive database.

[0033] Training a reservoir permeability prediction model using only 558 original well logging response sample points for deep learning is clearly insufficient. In other words, a reservoir permeability prediction model trained with only 558 original well logging response sample points will have poor accuracy. Furthermore, due to the limited number of well logging response sample points, even a model obtained using these sample points cannot be applied to predicting permeability in other reservoirs, exhibiting insufficient generalization ability. Therefore, it is necessary to supplement the well logging response sample points using different methods to increase the training data for the reservoir permeability prediction model and improve the accuracy of the permeability prediction results.

[0034] In some embodiments, step S3, supplementing logging response sample points for different types of reservoirs according to the cognitive database, includes: matching reservoir types according to the reservoir productivity evaluation parameters of the sampled sand bodies in the cognitive database, and supplementing the logging curve values ​​of the sampled sand bodies in the cognitive database as logging response sample points for the corresponding reservoir types.

[0035] In some embodiments, the reservoir productivity evaluation parameter is the relationship between porosity and permeability data; the sampled sand bodies are the same set of sampled sand bodies; Matching reservoir types based on the relationship between porosity and permeability data of the same set of sampled sand bodies includes: obtaining core data from the wellbore of the same set of sampled sand bodies, obtaining the relationship between porosity and permeability data of the sampled sand bodies based on the core data from the wellbore, and obtaining the reservoir type that matches the reservoir where the same set of sampled sand bodies is located based on the relationship between porosity and permeability data.

[0036] If the logging curves of the same set of sampled sand bodies exhibit essentially consistent response characteristics, and only one or two core samples are obtained from the wellbore within that set of sand bodies, and the porosity and permeability data from these core samples indicate that the cores belong to the same type of reservoir, then the logging curve values ​​of this entire layer can be added to the logging response sample points of that type of reservoir. For example, such as... Figure 4 As shown in the logging response characteristic diagram of three sand bodies (sand bodies 1, 2, and 3) developed in the deep formation of well B in oilfield A, a wellbore core was obtained in sand body 1 (the dot in the porosity log indicates the point where the wellbore core was obtained). Core analysis shows that sand body 1 is a Class IV reservoir. From the logging curves, the formations above and below the wellbore core show differences in natural gamma, compensated neutron (dashed line in the porosity log), and compensated density (…). Figure 4 The porosity logging (solid line in the first column) and nuclear magnetic resonance (NMR logging) show essentially the same characteristics in the transverse relaxation time spectrum (T2 spectrum). Specifically, the compensated density and compensated neutron line shown in the porosity logging are basically straight lines, and the NMR peak locations shown in the NMR logging are basically consistent. Therefore, it is considered that the reservoir type is consistent with that at the wellbore center, and it can be regarded as a logging response sample point for Class IV reservoirs. Similarly, the porosity and permeability analysis at the wellbore center of sand bodies No. 2 and No. 3 indicates that they are Class III reservoirs. The logging curve characteristics of sand bodies No. 2 and No. 3 that are consistent with those at the wellbore center can also be added to the logging response sample points for Class III reservoirs.

[0037] In other embodiments, for reservoir types whose logging response characteristics obtained in step S1 are very clear, and for reservoirs that have not been cored at certain depth points, logging response sample points of that type of reservoir can be supplemented based on the logging curve values ​​of that reservoir.

[0038] For example, such as Figure 5 As shown, core sampling analysis of multiple wells in Oilfield A at a certain location revealed porosity and permeability in formations exhibiting this logging response characteristic. These formations are all Class I reservoirs, meaning their natural gamma value is less than 80 gAPI. The neutron porosity curve has a left scale of 0.45 v / v and a right scale of -0.15 v / v, while the density curve has a left scale of 1.95 g / cm³ and a right scale of 2.95 g / cm³. Under these conditions, the porosity logging column shows a clear negative crossover characteristic between the compensated neutron curve and the compensated density curve (i.e., the compensated neutron curve is to the left of the compensated density curve). Based on this, the logging response characteristics are clearly understood, and reservoirs exhibiting this characteristic in actual drilling can be expanded into the logging response sample points for Class I reservoirs. After this expansion, the number of reservoir type logging response sample points in Oilfield A increased significantly from 558 to 978.

[0039] In some embodiments, the reservoir productivity evaluation parameter is permeability; the sampled sand body is a sampled sand body at the same depth point; like Figure 6 As shown, the reservoir type is matched based on the permeability of the sand samples taken at the same depth point. This includes: obtaining the pressure measurement results and mobility data of the sand samples taken at the same depth point based on the pressure measurement operation; substituting the mobility data into the permeability calculation model corresponding to various reservoir types to calculate the permeability; and obtaining the reservoir type that matches the reservoir where the sand samples are located by comparing the pressure measurement results and the permeability.

[0040] This embodiment addresses the scenario where only pressure testing was performed at the same depth and reliable mobility data was obtained. In this case, the permeability at a certain point at that depth can be calculated using permeability calculation models assigned to different reservoir types. If the result of the i-th permeability calculation model matches the pressure testing result, the logging curve value of that model is added to the logging response sample point of the i-th reservoir type.

[0041] In some embodiments, the reservoir productivity evaluation parameter is productivity data; the sampled sand body is a single sampled sand body; like Figure 7 As shown, matching reservoir types based on the production capacity data of a single sampled sand body includes: obtaining the measured production capacity of a single sampled sand body obtained from drill pipe testing; ensuring that the logging response characteristics of the reservoir containing the single sand body are consistent; calculating the permeability of the reservoir containing the single sampled sand body by substituting the logging response characteristics of the single sampled sand body into various reservoir types; calculating the model production capacity of the reservoir containing the single sampled sand body based on the permeability of the reservoir containing the single sampled sand body; and obtaining the reservoir type that matches the reservoir containing the single sampled sand body by comparing the measured production capacity and the model production capacity.

[0042] This embodiment focuses on drilling pipe testing (DST testing) of a single sand body to obtain stable production data. Permeability calculation models corresponding to different reservoir types are substituted into the calculation of the reservoir's permeability, and the model's production data is obtained from the permeability. If the model's production capacity calculated by the j-th permeability calculation model matches the measured production capacity, the logging curve value of that reservoir is added to the logging response sample points of the j-th reservoir. Calculating production capacity data through permeability is an indirect calculation, requiring conversion through multiple intermediate data points to ensure that the logging response characteristics of the reservoir are basically consistent. This is to ensure that the reservoir segment belongs to the same reservoir type, thereby guaranteeing the accuracy, representativeness, and generalizability of the production capacity calculation.

[0043] In some embodiments, in step S3, Gaussian noise is added to supplement the logging response sample points of different reservoir types based on the original logging response sample points. Before this, the original logging response sample points are divided into a training set and a test set according to a preset ratio.

[0044] For example, 80% of the original well logging response sample points (which may also include well logging response sample points supplemented by the cognitive database) covering 5 types of reservoirs are selected as the training set and 20% as the test set, and loaded into the preset neural network model for training.

[0045] Classifying the training and test sets before supplementing the logging response sample points with Gaussian noise is to prevent data leakage. For example, if the logging response sample points are supplemented with Gaussian noise before classifying the training and test sets, it is possible to confuse a certain original logging response sample point X with the logging response sample points obtained after Gaussian supplementation. One set was assigned to the training set and the other to the test set, resulting in data leakage and affecting the accuracy of penetration rate prediction results.

[0046] In some embodiments, in step S3, the step of adding Gaussian noise to supplement logging response sample points of different reservoir types based on the original logging response sample points includes: normalizing the training set of the original logging response sample points, and adding normally distributed Gaussian noise according to the type of the original logging response sample points in the training set.

[0047] By adding normally distributed Gaussian noise, the logging response samples in both the training and test sets can be expanded by 3 to 5 times. Normalization is used to eliminate differences in the numerical magnitude and characteristics of different original logging response sample points.

[0048] In some embodiments, the formula for adding Gaussian noise based on the type of the original logging response sample points in the training set is as follows: in, These are the logging response sample points after adding Gaussian noise. These are the original well logging response sample points. These are noise samples independently drawn from a Gaussian distribution. , It follows a normal distribution, where N is a normal distribution. The mean of a normal distribution is . The variance is the variance of the normal distribution.

[0049] For example, the logging response data supplemented with Gaussian noise was expanded by 5 times. The modeling sample points of the A oilfield reservoir type logging data after cognitive database and Gaussian noise supplementation were further expanded to 4890, which is 8.7 times larger than the original 558 core logging response sample points. This greatly enriched the number of logging response sample points in the deep learning training set and test set.

[0050] In some embodiments, the normally distributed Gaussian noise follows a normal distribution with small mean and small variance.

[0051] For example, the mean μ is set to 0 in the calculation formula for adding Gaussian noise, and different variances are set for different logging curves. The requirement for small mean and small variance is to prevent the logging response sample points, which are increased by Gaussian noise, from fluctuating greatly. This would prevent the learning model from learning the patterns of the logging response sample points or from learning incorrect patterns, thus affecting permeability prediction.

[0052] In some embodiments, the types of original logging response sample points in the training set include at least resistivity; Adding normally distributed Gaussian noise to the resistivity includes: adding normally distributed Gaussian noise to the resistivity using a logarithmic space method.

[0053] In some embodiments, the types of original logging response sample points may also include compensation density, compensation neutrons, etc., and physical constraints are set for compensation density and compensation neutrons.

[0054] In other embodiments, the types of logging response sample points also include natural gamma and sonic transit time. By adding Gaussian noise to different types of logging response sample points using different methods, it is ensured that the data after adding Gaussian noise is within a reasonable range, thereby improving the accuracy of subsequent calculation model learning and training of logging response sample points.

[0055] S4: Train the original logging response sample points and the supplemented logging response sample points through the permeability calculation model corresponding to each reservoir type, and determine whether the training results meet the preset training termination conditions. If yes, the reservoir permeability prediction model is completed; otherwise, step S3 is executed again.

[0056] For example, the preset training termination condition can be set as the accuracy of reservoir type identification exceeding 82% as the condition for ending the training of the deep learning model. Figure 8 As shown, before supplementing the well logging response sample points, the result was obtained by directly inputting 558 unexpanded original well logging response sample points into the deep neural network model for training. The accuracy of reservoir type identification was less than 65%. Figure 9 As shown, after supplementing the well logging response sample points, the accuracy of reservoir type identification reached 84%.

[0057] Input the logging response sample points from the test set into the reservoir permeability prediction model. If the accuracy of reservoir type identification exceeds 82%, the model obtained by deep learning training is considered reliable, and the construction of the reservoir permeability prediction model is completed. Otherwise, step 3 needs to be repeated to supplement the logging response sample points, further enrich the logging response sample points of each type of reservoir, and optimize the structure and hyperparameters of the deep learning model.

[0058] The method of cognitive database and Gaussian noise is used to expand the logging response sample points corresponding to each type of reservoir. Then, a reservoir permeability prediction model with strong applicability is constructed through deep learning algorithm, so as to achieve high-precision prediction of permeability of deep offshore reservoirs.

[0059] For example, the permeability of well C in oilfield A of a certain basin was predicted by a reservoir permeability prediction model and then applied to reservoir type identification. A total of 11 well cores were obtained at this depth, and the reservoir type of 9 of them was consistent with the results of deep neural network identification. The accuracy rate of identification was 82%, which can meet the accuracy requirements of current offshore oil and gas exploration and development for permeability prediction.

[0060] The present invention also discloses a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements a method for constructing a reservoir permeability prediction model as described in any of the above embodiments.

[0061] It is understood that the above embodiments only illustrate some implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can freely combine the above embodiments or technical features without departing from the concept of the present invention, and can also make several modifications and improvements, all of which fall within the protection scope of the present invention. That is, the embodiments described "in some embodiments" can be freely combined with any of the preceding and following embodiments. Therefore, all equivalent transformations and modifications made within the scope of the claims of the present invention should be covered by the claims of the present invention.

Claims

1. A method for constructing a reservoir permeability prediction model, characterized in that, The method includes the following steps: S1: Obtain core experimental data, classify the core experimental data into several reservoir types according to a preset classification method, and assign a permeability calculation model to each reservoir type; the reservoir types are at least two. S2: Analyze the logging curve values ​​corresponding to each core based on the core test data to obtain the original logging response sample points; S3: Analyze the combination characteristics of logging curves for each reservoir type based on the logging curves corresponding to each core sample, forming a cognitive database; and supplement the cognitive database with logging response sample points for different types of reservoirs; and / or, Gaussian noise was added to the original logging response sample points to supplement logging response sample points for different reservoir types; S4: Train the original logging response sample points and the supplemented logging response sample points through the permeability calculation model corresponding to each reservoir type, and determine whether the training results meet the preset training termination conditions. If yes, the reservoir permeability prediction model is completed; otherwise, step S3 is executed again.

2. The method for constructing a reservoir permeability prediction model according to claim 1, characterized in that, In step S3, logging response sample points for different types of reservoirs are supplemented based on the cognitive database, including: Based on the reservoir productivity evaluation parameters of the sampled sand bodies in the cognitive database, the reservoir type is matched, and the logging curve values ​​of the sampled sand bodies in the cognitive database are supplemented as logging response sample points for the corresponding reservoir type.

3. The method for constructing a reservoir permeability prediction model according to claim 2, characterized in that, The reservoir productivity evaluation parameter is the relationship between porosity and permeability data; the sampled sand bodies are the same set of sampled sand bodies; Reservoir types are matched based on the relationship between porosity and permeability data of the same set of sampled sand bodies, including: Obtain core data from the same set of sampled sand bodies, and determine the relationship between porosity and permeability data of the sampled sand bodies based on the core data. Then, determine the reservoir type that matches the reservoir where the same set of sampled sand bodies is located based on the relationship between porosity and permeability data.

4. The method for constructing a reservoir permeability prediction model according to claim 2, characterized in that, The reservoir productivity evaluation parameter is permeability; the sampled sand bodies are sand bodies sampled at the same depth point. Reservoir type is matched based on the permeability of sand bodies sampled at the same depth point, including: Based on the pressure measurement results and flow rate data obtained from the pressure measurement operation, sampled sand bodies at the same depth point are obtained; The permeability is obtained by substituting the mobility data into the permeability calculation model corresponding to various reservoir types. By comparing pressure measurement results and permeability, the reservoir type matching the reservoir where the sand body sampled at the same depth point is located can be obtained.

5. The method for constructing a reservoir permeability prediction model according to claim 2, characterized in that, The reservoir productivity evaluation parameters are productivity data; the sampled sand body is a single sampled sand body; Reservoir type is matched based on the productivity data of individual sampled sand bodies, including: The measured productivity of a single sampled sand body is obtained based on drill pipe testing; the logging response characteristics of the reservoir in which the single sand body is located are consistent. The permeability of the reservoir where the individual sampled sand body is located is calculated by substituting the logging response characteristics of the individual sampled sand body into various reservoir types. Calculate the model production capacity of the reservoir containing a single sampled sand body based on the permeability of the reservoir containing the single sampled sand body; By comparing the measured production capacity with the model production capacity, the reservoir type that matches the reservoir where a single sampled sand body is located is obtained.

6. The method for constructing a reservoir permeability prediction model according to claim 1, characterized in that, In step S3, Gaussian noise is added to supplement the logging response sample points for different reservoir types based on the original logging response sample points. This includes the following prior steps: The original well logging response sample points are divided into training set and test set according to a preset ratio.

7. The method for constructing a reservoir permeability prediction model according to claim 6, characterized in that, In step S3, the step of adding Gaussian noise to supplement logging response sample points for different reservoir types based on the original logging response sample points includes: The training set of original well logging response sample points is normalized, and Gaussian noise with a normal distribution is added according to the type of original well logging response sample points in the training set.

8. The method for constructing a reservoir permeability prediction model according to claim 7, characterized in that, The normally distributed Gaussian noise follows a normal distribution with small mean and small variance.

9. The method for constructing a reservoir permeability prediction model according to claim 8, characterized in that, The types of original logging response sample points in the training set include at least resistivity; Adding normally distributed Gaussian noise to resistivity includes: adding normally distributed Gaussian noise to resistivity using a logarithmic space method.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for constructing the reservoir permeability prediction model as described in any one of claims 1-9.

Images