Intelligent reservoir logging evaluation method driven by double models

Abstract

The invention discloses a dual-model-driven intelligent reservoir logging evaluation method, which comprises the following steps of: establishing a lithology identification logging data set based on a rock slice identification result and conventional logging data, and constructing a data model-driven lithology identification pre-training model; establishing a porosity prediction data set of different lithology based on a core porosity analysis result, and constructing a dual-model driven porosity prediction model of a data model and a knowledge model under the guidance of a lithology classification result; establishing a permeability prediction data set of different lithology based on a permeability analysis result, and constructing a dual-model driven permeability prediction model of a data model and a knowledge model under the guidance of a lithology classification result; on the basis of different lithologic saturation parameters measured in a laboratory, the water saturation of different lithologies is calculated in combination with physical property prediction results driven by double models of different lithologies and conventional logging data. The method can effectively solve the well logging evaluation problem faced by a complex reservoir.

Description

Technical Field

[0001] The present invention relates to the field of intelligent well logging evaluation, and in particular to an intelligent reservoir well logging evaluation method driven by dual models. Background Art

[0002] Well logging includes different means such as electrical logging, acoustic logging, and natural gamma logging. It can provide parameters such as resistivity, acoustic velocity, and radioactivity of the formation around the well wall. It is an important technology for obtaining reservoir physical parameters, discovering and evaluating oil and gas reservoirs, and predicting oil and gas reserves. Well logging evaluation of reservoirs refers to obtaining the porosity, permeability, saturation, etc. of reservoirs through a series of logging technical means and methods, while evaluating the properties and storage conditions of oil and gas reservoirs, providing a basis for oilfield development. With the deepening of reservoir evaluation research, the importance of accurate evaluation of unconventional reservoirs has gradually increased. However, the complex reservoir structure and strong heterogeneity make it difficult to apply reservoir logging evaluation technology based on the experience of conventional reservoirs in different regions. At the same time, with the rapid development of artificial intelligence technology, how to apply it to the processing, mining and utilization of massive logging data, realize the "value-added" of logging information, and upgrade oil and gas exploration and development services has become an important content in the field of logging applications in the new era. Therefore, combining advanced artificial intelligence technology, making the application of logging data intelligent, and realizing intelligent reservoir logging evaluation is a research with both application value and economic value.

[0003] Traditional well logging evaluation technology faces many challenges in the evaluation of unconventional reservoirs. Only by improving the accuracy and efficiency of well logging evaluation can the advancement of oil and gas exploration and development technology be promoted. In order to solve the problems of low accuracy of lithology identification and parameter evaluation of traditional well logging evaluation methods when facing complex geological conditions, and to combine different well logging data to achieve more accurate analysis of complex reservoir evaluation parameters, the present invention has established a dual-model driven intelligent reservoir logging evaluation method based on data model and knowledge model. An intelligent well logging evaluation database is constructed through rock physical data and well logging data, and reservoir parameter prediction driven by data model is completed. Then, the database is enriched by relying on knowledge model to jointly predict parameters and complete dual-model driven intelligent reservoir logging evaluation of complex reservoirs. Summary of the invention

[0004] The technical problem to be solved by the present invention is to combine the data model and the knowledge model to establish a dual-model driven intelligent reservoir logging evaluation method to solve the problem of accurate evaluation of unconventional reservoirs.

[0005] Explanation of terms involved in the present invention: Data model: is a collection of conceptual tools that describe the relationship between data, data semantics, and data consistency constraints; Knowledge model: A knowledge model is a formalized and structured representation of knowledge, which aims to transform complex and diverse knowledge into a form that is easy to understand, manage and apply; Dual-model drive: combining data model and knowledge model to jointly predict parameters; Z-score: data normalization method; KNN: K-nearest neighbor algorithm; Extreme Random Trees: A common method for intelligent regression.

[0006] A dual-model driven intelligent reservoir logging evaluation method, the method comprising the following steps: Step 1: Establish a lithology identification logging data set based on rock thin section identification results and conventional logging data, and select an intelligent classification algorithm to build a data model-driven lithology identification pre-training model; Step 2: Based on the core porosity analysis results, a porosity prediction data set of different lithologies is established, and a suitable intelligent regression algorithm is selected to construct a dual-model driven porosity prediction model of data model and knowledge model under the guidance of lithology classification results; Step 3: Based on the permeability analysis results, a permeability prediction data set for different lithologies is established. At the same time, a permeability formula library is established, and a suitable intelligent regression algorithm is selected to construct a dual-model driven permeability prediction model of data model and knowledge model under the guidance of lithology classification results; Step 4: Based on the saturation parameters of different lithologies measured in the laboratory, combined with the physical property prediction results driven by the dual model of different lithologies and conventional logging data, the water saturation of different lithologies is calculated.

[0007] As a further technical solution of the present invention, the steps of establishing a lithology identification logging data set based on the rock lithology identification results and conventional logging data in step 1, and selecting an intelligent classification algorithm to construct a data model-driven lithology identification pre-training model include: Using the rock thin section identification results as labels, conventional logging data and element logging data were extracted to establish a lithology identification logging data set. Conventional logging data included acoustic wave time difference, natural gamma, compensated density, compensated neutron, radioactive potassium, deep resistivity and shallow resistivity at the corresponding depth of the rock thin section. Element logging data included sodium, magnesium, aluminum, silicon, potassium, calcium, titanium and iron. A preliminary lithology identification logging data set was established. In order to make the data distribution more balanced, it is necessary to take the logarithm of the deep resistivity and shallow resistivity curves, and then apply Z-score standardization to process different logging data. The calculation formula of Z-score standardization is: Formula 1: ; in, is the standardized data, is the original data, is the mean of the original data, is the standard deviation of the original data; Standard Deviation The calculation formula is: Formula 2: ; in, is the sample size; is the original data of each sample; Afterwards, different lithologies are represented by different numerical values, and the actual lithology labels are converted into numerical labels to complete the digitization of the lithology identification logging data set, thus establishing an important data foundation for building a data model-driven lithology identification pre-training model. Taking the K-nearest neighbor algorithm as an example, the lithology identification logging data set is divided into a training set and a test set in a ratio of 8:2. In order to ensure the balance and stability of the samples, each lithology is divided in the same ratio by random sampling, and then the training and verification of the lithology identification pre-training model are carried out; in the verification process, the accuracy between the results obtained by re-predicting the training set and the core results is the back-judgment accuracy, and the accuracy obtained by predicting the test set by the same method is the test accuracy; Formula 3: ; Formula 4: ; In the formula, To judge the accuracy, is the total number of samples in the training set, The number of correct results obtained by re-predicting the training set, To test accuracy, is the total number of test set samples, The number of correct results obtained for predictions on the test set.

[0008] As a further technical solution of the present invention, in step 2, based on the core porosity analysis results, a porosity prediction data set of different lithologies is established, and an intelligent regression algorithm is selected to construct a dual-model driven porosity prediction model of a data model and a knowledge model under the guidance of the lithology classification results, including: According to the rock thin section and other lithology information and core porosity measurement results of the area to be processed, porosity logging data are extracted, and based on the lithology classification results of this area, a preliminary porosity prediction data set of different lithologies is established; then, based on the porosity logging data and the corresponding porosity calculation formula, the knowledge model porosity prediction results of different lithologies are obtained, and the knowledge model-driven porosity prediction results are used as new features to form a complete data set for porosity prediction together with the porosity logging data; the porosity calculation formula based on the porosity logging data is as follows: Formula 5: ; Formula 6: ; Formula 7: ; Formula 8: ; In the formula, is the mud content; , , , , , , Respectively The porosity calculated by time-of-sound logging, the porosity calculated by compensated neutron logging, the porosity calculated by compensated density logging, the natural gamma logging value, the time-of-sound logging value, the compensated neutron logging value and the compensated density logging value of each sample point; and represent the natural gamma values ​​of pure sandstone and mudstone, respectively; and They are the acoustic time difference logging value and the compensated density logging value of the formation skeleton respectively; and They are the acoustic time difference logging value and the compensated density logging value of the formation fluid respectively; , and They are the acoustic time difference logging value, compensated neutron logging value and compensated density logging value of the mud in the formation; The compaction coefficient of the formation is usually 1; After implementing different intelligent regression algorithms such as linear regression, ridge regression, and decision tree regression, the core data is used as the standard, and the method that best matches the core results is selected as the method suitable for the area to be treated. The porosity prediction data set can be trained to obtain a dual-model driven porosity prediction model under the guidance of different lithology classifications. After that, it is applied in blind wells, and the accuracy of the method is judged by the average absolute error between the prediction results and the porosity of the blind well core. The average absolute error calculation formula is: Formula 9: ; In the formula, is the mean absolute error of the porosity prediction results; For the blind well core data points; For the The porosity prediction result of the corresponding depth of each core data point; For the Core porosity measurements; is the total number of core porosity data points in the blind well.

[0009] As a further technical solution of the present invention, in step 3, based on the permeability analysis results, a permeability prediction data set of different lithologies is established, and at the same time, a permeability formula library is established, and a suitable intelligent regression algorithm is selected to construct a dual-model driven permeability prediction model of a data model and a knowledge model under the guidance of the lithology classification results, including the following steps: According to the lithology information and core permeability measurement results of the area to be processed, the core permeability logging data and dual-driven porosity prediction results are extracted. At the same time, based on the lithology classification results of this area, a preliminarily established permeability prediction data set of different lithologies is established. Then, based on the dual-driven porosity prediction results and the applicable permeability calculation formulas for different regions, the knowledge model-driven permeability prediction results of different lithologies are obtained. The knowledge model-driven permeability prediction results are used as new features together with the core permeability logging data to form a complete data set for permeability prediction. After implementing different intelligent regression algorithms such as linear regression, ridge regression, and decision tree regression, the core data is used as the standard, and the method that best matches the core results is selected as the method suitable for the area to be treated. The permeability prediction data set can be trained to obtain a dual-model driven permeability prediction model under the guidance of different lithology classifications. After that, it is applied in blind wells, and the accuracy of the method is judged by the average absolute error between the prediction results and the blind well core permeability. The average absolute error calculation formula is: Formula 10: ; In the formula, is the mean absolute error of the permeability prediction results, For the blind well Core data points, For the The permeability prediction results of the corresponding depth of the core data points are: For the The core permeability measurement results are is the total number of core permeability data points in the blind well.

[0010] As a further technical solution of the present invention, in step 4, based on the saturation parameters of different lithologies measured in the laboratory, combined with the physical property prediction results driven by the dual model of different lithologies and conventional logging data, the step of calculating the water saturation of different lithologies includes: The saturation parameters of different lithologies are determined by rock physics experiments. The saturation parameters are saturation index n, cementation index m and lithology index ; Calculation of water saturation of different lithologies: Based on the empirical relationship between resistivity and water saturation established by Archie, which is the basis for calculating formation water saturation, Archie's formula is as follows: Formula 11: ; In the formula, is water saturation; is the saturation index; is the lithology index; is the cementation index; is the formation water resistivity; is the formation resistivity; is the porosity; Formation water resistivity The saturation parameters measured in the laboratory can be obtained during the logging process by substituting them into the Archie formula, combining the dual-drive predicted porosity results, and selecting the deep lateral resistivity logging curve as the formation resistivity. , the continuous evaluation of water saturation in the well can be completed.

[0011] Beneficial effects achieved by the present invention: The dual-model driven intelligent reservoir logging evaluation method based on data model and knowledge model proposed in the present invention can effectively solve the logging evaluation problems faced by complex reservoirs; The present invention adds the physical property prediction results driven by the knowledge model as one of its features, which has higher interpretability than the reservoir evaluation method driven by the data model; In the technical solution designed by the present invention, the intelligent algorithm of each link is replaceable and can flexibly adapt to the actual conditions of different regions; The dual-model driven intelligent reservoir logging evaluation method based on data model and knowledge model proposed in the present invention not only ensures the calculation accuracy, but also improves the interpretation efficiency of complex reservoirs and releases human resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 Schematic diagram of well data extraction according to an embodiment of the present invention.

[0013] Figure 2 A flowchart for constructing a lithology identification pre-training model according to an embodiment of the present invention.

[0014] Figure 3 A flow chart for constructing a porosity prediction model according to an embodiment of the present invention.

[0015] Figure 4 A flow chart for constructing a permeability prediction model according to an embodiment of the present invention.

[0016] Figure 5 It is a specific flow chart of an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The technical solution of the present invention is described in detail below in conjunction with the specific drawings.

[0018] This embodiment provides a dual-model driven intelligent reservoir logging evaluation method, the method comprising the following steps: Step 1: Establish a lithology identification logging data set based on rock thin section identification results and conventional logging data, and select an intelligent classification algorithm to build a data model-driven lithology identification pre-training model; Step 2: Based on the core porosity analysis results, a porosity prediction data set of different lithologies is established, and a suitable intelligent regression algorithm is selected to construct a dual-model driven porosity prediction model of data model and knowledge model under the guidance of lithology classification results; Step 3: Based on the permeability analysis results, a permeability prediction data set for different lithologies is established. At the same time, a permeability formula library is established, and a suitable intelligent regression algorithm is selected to construct a dual-model driven permeability prediction model of data model and knowledge model under the guidance of lithology classification results; Step 4: Based on the saturation parameters of different lithologies measured in the laboratory, combined with the physical property prediction results driven by the dual model of different lithologies and conventional logging data, the water saturation of different lithologies is calculated.

[0019] In step 1 of this embodiment, the steps of establishing a lithology identification logging data set based on rock lithology identification results and conventional logging data, and selecting an intelligent classification algorithm to construct a data model-driven lithology identification pre-training model include: Using the rock thin section identification results as labels, conventional logging data and element logging data were extracted to establish a lithology identification logging data set. Conventional logging data included acoustic wave time difference, natural gamma, compensated density, compensated neutron, radioactive potassium, deep resistivity and shallow resistivity at the corresponding depth of the rock thin section. Element logging data included sodium, magnesium, aluminum, silicon, potassium, calcium, titanium and iron. A preliminary lithology identification logging data set was established. According to the identification results of thin sections of different lithologies in the area to be processed, such as Figure 1 The acoustic time difference (AC, in μs / ft), natural gamma (GR, in GAPI), and compensated density (DEN, in g / cm2) of the extracted rock slices at corresponding depths are shown. 3 ), compensated neutron (CNL, in %), radioactive potassium (K40, in %), deep resistivity (RLLD, in OHMM) and shallow resistivity (RLLS, in OHMM) and other conventional logging data, as well as element data such as sodium (Na, in w / w), magnesium (Mg, in w / w), aluminum (Al, in w / w), silicon (Si, in w / w), potassium (K, in w / w), calcium (Ca, in w / w), titanium (Ti, in w / w), and iron (Fe, in w / w), were used to preliminarily establish a lithology identification logging data set; In order to make the data distribution more balanced, it is necessary to take the logarithm of the deep resistivity and shallow resistivity curves, and then apply Z-score standardization to process different logging data. The calculation formula of Z-score standardization is: Formula 1: ; in, is the standardized data, is the original data, is the mean of the original data, is the standard deviation of the original data; Standard Deviation The calculation formula is: Formula 2: ; in, is the sample size; is the original data of each sample; Afterwards, different lithologies are represented by different numerical values, and the actual lithology labels are converted into numerical labels to complete the digitization of the lithology identification logging data set, thus establishing an important data foundation for building a data model-driven lithology identification pre-training model. Data model driven means organizing data and forming information, then integrating and refining the information, and forming an automated decision model through training and fitting based on the data. Its core lies in making decisions and actions based on data. Different intelligent classification algorithms such as K-nearest neighbor (KNN) can be trained to obtain a data model driven lithology identification model. In practical applications, after selecting the algorithm with the smallest error as the applicable intelligent classification algorithm for the area based on core data, the lithology identification logging data set is divided into a training set and a test set in a ratio of 8:2. In order to ensure the balance and stability of the samples, a random sampling method is used to divide each lithology in the same proportion, and then the training and verification of the lithology identification pre-training model are carried out; in the verification process, the accuracy between the results obtained by re-predicting the training set and the core results is the back-judgment accuracy, and the accuracy obtained by predicting the test set by the same method is the test accuracy; Formula 3: ; Formula 4: ; In the formula, To judge the accuracy, is the total number of samples in the training set, The number of correct results obtained by re-predicting the training set, To test accuracy, is the total number of test set samples, The number of correct results obtained for predictions on the test set.

[0020] Figure 2 The specific process of building a data model-driven lithology identification pre-training model includes establishing a lithology identification logging data set, logging data standardization, lithology label digitization, training set and test set division, model training and model verification. In practical applications, it is necessary to select a suitable intelligent classification algorithm as the core algorithm for lithology identification based on the actual conditions of different regions.

[0021] In step 2 of this embodiment, the steps of establishing a porosity prediction data set of different lithologies based on the core porosity analysis results, and selecting an intelligent regression algorithm to construct a dual-model driven porosity prediction model of a data model and a knowledge model under the guidance of the lithology classification results include: According to the lithology information such as rock thin sections and core porosity measurement results of the area to be processed, porosity logging data are extracted. The porosity logging data include conventional logging data such as acoustic time difference (AC), natural gamma (GR), compensated density (DEN), compensated neutron (CNL), radioactive potassium (K40), deep resistivity (RLLD) and shallow resistivity (RLLS) at the corresponding depth of core porosity. Based on the lithology classification results of this area, a porosity prediction data set of different lithologies is preliminarily established; then, based on the porosity logging data and the corresponding porosity calculation formula, the knowledge model porosity prediction results of different lithologies are obtained, and the knowledge model-driven porosity prediction results are used as new features to form a complete data set for porosity prediction together with the porosity logging data; the porosity calculation formula based on the porosity logging data is as follows: Formula 5: ; Formula 6: ; Formula 7: ; Formula 8: ; In the formula, is the mud content; , , , , , , Respectively The porosity calculated by time-of-sound logging, the porosity calculated by compensated neutron logging, the porosity calculated by compensated density logging, the natural gamma logging value, the time-of-sound logging value, the compensated neutron logging value and the compensated density logging value of each sample point; and represent the natural gamma values ​​of pure sandstone and mudstone, respectively; and They are the acoustic time difference logging value and the compensated density logging value of the formation skeleton respectively; and They are the acoustic time difference logging value and the compensated density logging value of the formation fluid respectively; , and They are the acoustic time difference logging value, compensated neutron logging value and compensated density logging value of the mud in the formation; The compaction coefficient of the formation is usually 1; The intelligent regression algorithm is an algorithm that uses artificial intelligence technology to solve regression problems. It is usually used to establish a model between characteristic variables and response variables, so as to evaluate the expected value of the response variable. Therefore, accurate formation porosity is the goal of the regression prediction model in this step, and the real core porosity, conventional logging data and porosity calculated by porosity logging are known characteristic variables. After implementing different intelligent regression algorithms such as linear regression, ridge regression, and decision tree regression, the core data is used as the standard, and the method that best matches the core results is selected as the method suitable for the area to be processed. The porosity prediction data set can be trained to obtain a dual-model driven porosity prediction model under the guidance of different lithology classifications. After that, it is applied in blind wells, and the accuracy of the method is judged by the average absolute error between the prediction results and the blind well core porosity. The average absolute error calculation formula is: Formula 9: ; In the formula, is the mean absolute error of the porosity prediction results; For the blind well core data points; For the The porosity prediction result of the corresponding depth of each core data point; For the Core porosity measurements; is the total number of core porosity data points in the blind well.

[0022] Figure 3 The specific process of the dual-model driven porosity prediction model under the guidance of lithology classification results includes establishing preliminary porosity prediction data sets for different lithologies, obtaining knowledge model driven porosity results based on logging data and porosity calculation formulas, establishing a complete porosity prediction data set, selecting appropriate intelligent regression methods, model training and model verification. In practical applications, it is necessary to select appropriate intelligent regression algorithms as the core algorithms for porosity prediction according to the actual conditions of different regions.

[0023] In step 3 of this embodiment, the steps of establishing a permeability prediction data set for different lithologies based on the permeability analysis results, establishing a permeability formula library, and selecting a suitable intelligent regression algorithm to construct a dual-model driven permeability prediction model of a data model and a knowledge model under the guidance of the lithology classification results include: According to the rock thin section and other lithology information and core permeability measurement results of the area to be processed, the core permeability logging data and dual-driven porosity prediction results are extracted. The core permeability logging data are conventional logging data such as acoustic time difference (AC), natural gamma (GR), compensated density (DEN), compensated neutron (CNL), radioactive potassium (K40), deep resistivity (RLLD) and shallow resistivity (RLLS) at the corresponding depth. At the same time, based on the lithology classification results of this area, a preliminarily established permeability prediction data set for different lithologies. Afterwards, based on the dual-driven porosity prediction results and the applicable permeability calculation formulas for different regions, the knowledge model-driven permeability prediction results for different lithologies are obtained, and the knowledge model-driven permeability prediction results are used as new features together with the core permeability logging data to form a complete data set for permeability prediction. Accurate formation permeability is the goal of the regression prediction model in this step, and the measured core permeability, conventional logging data, and permeability calculated based on porosity prediction results are known characteristic variables. After implementing different intelligent regression algorithms such as linear regression, ridge regression, and decision tree regression, the core data is used as the standard, and the method that best matches the core results is selected as the method suitable for the area to be treated. The permeability prediction data set can be trained to obtain a dual-model driven permeability prediction model under the guidance of different lithology classifications. After that, it is applied in blind wells, and the accuracy of the method is judged by the average absolute error between the prediction results and the blind well core permeability. The average absolute error calculation formula is: Formula 10: ; In the formula, is the mean absolute error of the permeability prediction results, For the blind well Core data points, For the The permeability prediction results of the corresponding depth of the core data points are: For the The core permeability measurement results are is the total number of core permeability data points in the blind well.

[0024] Figure 4 The specific process of the dual-drive permeability prediction model under the guidance of lithology classification results includes establishing a preliminary permeability prediction data set for different lithologies, calculating permeability based on porosity prediction results, establishing a complete permeability prediction data set, selecting a suitable intelligent regression method, model training and model verification. In practical applications, it is necessary to select a suitable intelligent regression algorithm as the core algorithm for permeability prediction according to the actual conditions of different regions.

[0025] In step 4 of this embodiment, based on the saturation parameters of different lithologies measured in the laboratory, combined with the physical property prediction results driven by the dual models of different lithologies and conventional logging data, the step of calculating the water saturation of different lithologies includes: The saturation parameters of different lithologies are determined by rock physics experiments. The saturation parameters are saturation index n, cementation index m and lithology index , can be measured through laboratory rock electrical experiments. During the experiment, it is necessary to collect representative rock samples, then pre-process the rock samples by cleaning and drying, and then use the resistivity measuring device to determine the saturation parameters of different rock types; Calculation of water saturation of different lithologies: Based on the empirical relationship between resistivity and water saturation established by Archie, which is the basis for calculating formation water saturation, Archie's formula is as follows: Formula 11: ; In the formula, is water saturation; is the saturation index; is the lithology index; is the cementation index; is the formation water resistivity; is the formation resistivity; is the porosity; Formation water resistivity The saturation parameters measured in the laboratory can be obtained during the logging process by substituting them into the Archie formula, combining the dual-drive predicted porosity results, and selecting the deep lateral resistivity logging curve as the formation resistivity. , the continuous evaluation of water saturation in the well can be completed.

[0026] In order to facilitate those skilled in the art to better understand the technical solution of the present invention, the specific application implementation of the present invention is given as follows: The method of the present invention is applied to a complex lithology reservoir. The accuracy and error of each link are shown in Table 1. The specific flow chart of the method is as follows: Figure 5As shown in the figure: First, the logging data and lithology thin section data are used to form a lithology identification logging data set, and the lithology identification is performed based on the KNN algorithm. The return accuracy and test accuracy are 96.74% and 95.65%, respectively, indicating that the lithology identification results are reliable; secondly, the porosity prediction data set of different lithologies is established, and the data model is used to drive the lithology identification results. Combined with the knowledge model to drive the porosity, core porosity and logging data, the dual-driven porosity intelligent regression of different lithologies is jointly performed. Taking the porosity prediction result of diorite as an example, the dual-driven predicted porosity has a lower average absolute error and better effect; then, different lithologies are established. The permeability prediction dataset of different lithologies is used. The lithology identification results driven by the data model are combined with the dual-driven porosity knowledge model to drive the permeability, core permeability and logging data, and the dual-driven permeability intelligent regression of different lithologies is carried out together. Taking the diorite permeability prediction result as an example, the dual-driven predicted permeability has a lower average absolute error and better effect. Finally, based on the physical property results obtained by the dual-driven prediction and the saturation parameters obtained by the rock physics experimental measurement, the dual-driven saturation prediction of different lithologies can be completed, and the dual-driven intelligent reservoir evaluation of complex lithology reservoirs can be realized. The evaluation results in the diorite reservoir are consistent with the final test conclusion.

[0027] Table 1 Analysis of intelligent logging evaluation results ; It should be noted that, in this article, the term "comprises" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of more restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0028] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A dual-model driven intelligent reservoir logging evaluation method, characterized in that: The method comprises the following steps: Step 1: Establish a lithology identification logging data set based on rock thin section identification results and conventional logging data, and select an intelligent classification algorithm to build a data model-driven lithology identification pre-training model; Step 2: Based on the core porosity analysis results, a porosity prediction data set of different lithologies is established, and an intelligent regression algorithm is selected to construct a dual-model driven porosity prediction model of data model and knowledge model under the guidance of lithology classification results; Step 3: Based on the permeability analysis results, a permeability prediction data set for different lithologies is established. At the same time, a permeability formula library is established. An intelligent regression algorithm is selected to construct a dual-model driven permeability prediction model of data model and knowledge model under the guidance of lithology classification results. Step 4: Based on the saturation parameters of different lithologies measured in the laboratory, combined with the physical property prediction results driven by the dual model of different lithologies and conventional logging data, the water saturation of different lithologies is calculated.

2. The dual-model driven intelligent reservoir logging evaluation method according to claim 1, characterized in that: In step 1, the steps of establishing a lithology identification logging data set based on rock lithology identification results and conventional logging data, and selecting an intelligent classification algorithm to build a data model-driven lithology identification pre-training model include: Using the rock thin section identification results as labels, conventional logging data and element logging data were extracted to establish a lithology identification logging data set. Conventional logging data included acoustic wave time difference, natural gamma, compensated density, compensated neutron, radioactive potassium, deep resistivity and shallow resistivity at the corresponding depth of the rock thin section. Element logging data included sodium, magnesium, aluminum, silicon, potassium, calcium, titanium and iron. A preliminary lithology identification logging data set was established. In order to make the data distribution more balanced, it is necessary to take the logarithm of the deep resistivity and shallow resistivity curves, and then apply Z-score standardization to process different logging data. The calculation formula of Z-score standardization is: Formula 1: ; in, is the standardized data, is the original data, is the mean of the original data, is the standard deviation of the original data; Standard Deviation The calculation formula is: Formula 2: ; in, is the sample size; is the original data of each sample; Afterwards, different lithologies are represented by different numerical values, and the actual lithology labels are converted into numerical labels to complete the digitization of the lithology identification logging data set, thus establishing an important data foundation for building a data model-driven lithology identification pre-training model. The intelligent classification algorithm selects the K-nearest neighbor algorithm, and divides the lithology identification logging data set into a training set and a test set at a ratio of 8:

2. In order to ensure the balance and stability of the samples, a random sampling method is used to divide each lithology in the same ratio, and then the training and verification of the lithology identification pre-training model are carried out; in the verification process, the accuracy between the results obtained by re-predicting the training set and the core results is the back-judgment accuracy, and the accuracy obtained by predicting the test set using the same method is the test accuracy; Formula 3: ; Formula 4: ; In the formula, To judge the accuracy, is the total number of samples in the training set, The number of correct results obtained by re-predicting the training set, To test accuracy, is the total number of test set samples, The number of correct results obtained for predictions on the test set.

3. The dual-model driven intelligent reservoir logging evaluation method according to claim 2, characterized in that: In step 2, based on the core porosity analysis results, a porosity prediction data set of different lithologies is established, and the steps of selecting an intelligent regression algorithm to construct a dual-model driven porosity prediction model of a data model and a knowledge model under the guidance of the lithology classification results include: According to the lithology information and core porosity measurement results of the area to be processed, porosity logging data are extracted, and based on the lithology classification results of this area, a porosity prediction data set of different lithologies is preliminarily established; then, based on the porosity logging data and the corresponding porosity calculation formula, the knowledge model porosity prediction results of different lithologies are obtained, and the knowledge model-driven porosity prediction results are used as new features to form a complete data set for porosity prediction together with the porosity logging data; the porosity calculation formula based on the porosity logging data is as follows: Formula 5: ; Formula 6: ; Formula 7: ; Formula 8: ; In the formula, is the mud content; , , , , , , Respectively The porosity calculated by time-of-sound logging, the porosity calculated by compensated neutron logging, the porosity calculated by compensated density logging, the natural gamma logging value, the time-of-sound logging value, the compensated neutron logging value and the compensated density logging value of each sample point; and represent the natural gamma values ​​of pure sandstone and mudstone, respectively; and They are the acoustic time difference logging value and the compensated density logging value of the formation skeleton respectively; and They are the acoustic time difference logging value and compensated density logging value of formation fluid respectively; , and They are the acoustic time difference logging value, compensated neutron logging value and compensated density logging value of mud in the formation respectively; is the compaction coefficient of the formation; After implementing different intelligent regression algorithms such as linear regression, ridge regression, and decision tree regression, the core data was used as the standard, and the method that best matched the core results was selected as the method suitable for the area to be treated. The porosity prediction data set was then trained to obtain a dual-model driven porosity prediction model under the guidance of different lithology classifications. Afterwards, it was applied in blind wells, and the accuracy of the method was judged by the mean absolute error between the prediction results and the porosity of the blind well core. The mean absolute error calculation formula is: Formula 9: ; In the formula, is the mean absolute error of the porosity prediction results; For the blind well core data points; For the The porosity prediction result of the corresponding depth of each core data point; For the Core porosity measurements; is the total number of core porosity data points in the blind well.

4. The dual-model driven intelligent reservoir logging evaluation method according to claim 2, characterized in that: In step 3, based on the permeability analysis results, a permeability prediction data set for different lithologies is established, and a permeability formula library is established. The steps of selecting an intelligent regression algorithm to construct a dual-model driven permeability prediction model of a data model and a knowledge model under the guidance of the lithology classification results include: According to the lithology information and core permeability measurement results of the area to be processed, the core permeability logging data and dual-driven porosity prediction results are extracted. At the same time, based on the lithology classification results of this area, a preliminarily established permeability prediction data set of different lithologies is established. Then, based on the dual-driven porosity prediction results and the applicable permeability calculation formulas for different regions, the knowledge model-driven permeability prediction results of different lithologies are obtained. The knowledge model-driven permeability prediction results are used as new features together with the core permeability logging data to form a complete data set for permeability prediction. After implementing different intelligent regression algorithms such as linear regression, ridge regression, and decision tree regression, the core data is used as the standard, and the method that best matches the core results is selected as the method suitable for the area to be treated. The permeability prediction data set can be trained to obtain a dual-model driven permeability prediction model under the guidance of different lithology classifications. After that, it is applied in blind wells, and the accuracy of the method is judged by the average absolute error between the prediction results and the blind well core permeability. The average absolute error calculation formula is: Formula 10: ; In the formula, is the mean absolute error of the permeability prediction results, For the blind well Core data points, For the The permeability prediction results of the corresponding depth of the core data points are: For the The core permeability measurement results are is the total number of core permeability data points in the blind well.

5. The dual-model driven intelligent reservoir logging evaluation method according to claim 1, characterized in that: In step 4, based on the saturation parameters of different lithologies measured in the laboratory, combined with the physical property prediction results driven by the dual models of different lithologies and conventional logging data, the steps of calculating the water saturation of different lithologies include: The saturation parameters of different lithologies are determined by rock physics experiments. The saturation parameters are saturation index n, cementation index m and lithology index ; Calculation of water saturation of different lithologies: Based on the empirical relationship between resistivity and water saturation established by Archie, which is the basis for calculating formation water saturation, Archie's formula is as follows: Formula 11: ; In the formula, is water saturation; is the saturation index; is the lithology index; is the cementation index; is the formation water resistivity; is the formation resistivity; is the porosity; Formation water resistivity The saturation parameters measured in the laboratory were substituted into Archie's formula, and the porosity results were predicted by dual drive. The deep lateral resistivity logging curve was used as the formation resistivity. , the continuous evaluation of water saturation in the well can be completed.

Images