Intelligent interpretation method and system for complex lithology of buried-hill intrusive reservoirs

Abstract

This application belongs to the field of geophysical well logging technology and relates to a method and system for intelligent interpretation of complex lithologies in buried-hill intrusive reservoirs. The method comprises the following steps: obtaining mineral content data determined by target layer logging curves, XRF element logging, and rock thin sections; constructing XRF and thin section sample sets based on XRF, partial thin section mineral content data, and corresponding depth logging curves; constructing a validation set and a prediction set based on logging curve spectral clustering and the remaining thin sections; training and adjusting parameters of two random forest models using the two sample sets; verifying the performance of the two models using the validation set, and determining the optimal model for different logging curve spectral clustering; using the optimal model to classify and predict the prediction set to obtain different mineral contents; and using the intrusive rock QAP map to project different mineral contents to determine the lithologic category of the intrusive rock. This application enables intelligent and refined batch interpretation of complex lithologies in buried-hill intrusive reservoirs through multi-source data fusion.

Description

Technical Field

[0001] The present application belongs to the field of geophysical well logging technology, and specifically relates to a method and system for intelligent interpretation of complex lithology of buried-hill intrusive rock reservoirs. Background Art

[0002] With the continuous advancement of oil and gas exploration, conventional oil and gas reservoirs are gradually reaching saturation. Unconventional oil and gas reservoirs, represented by buried-hill bedrock oil and gas reservoirs, are becoming a new area for breakthroughs in oil and gas exploration both domestically and internationally due to their enormous oil and gas reserve potential. The spatial distribution characteristics and assemblage patterns of buried-hill oil and gas reservoirs control the distribution of reservoirs and oil and gas reservoirs within the buried-hill. Therefore, accurately identifying the lithology of buried-hill intrusive rocks is extremely important for oil and gas exploration.

[0003] Currently, conventional methods for interpreting the lithology of buried-hill intrusive rocks primarily include conventional well logging crossplots and quantitative mineral calculation using formation element logging (ECS logging). Due to the complex and diverse lithology of intrusive rocks, and the influence of various geological processes such as temperature, weathering and leaching, and hydrothermal dissolution, their mineral content exhibits strong heterogeneity. This results in numerous limitations in the practical application of conventional well logging interpretation methods, making it difficult to accurately identify the lithology of buried-hill intrusive rocks, a major challenge hindering buried-hill reservoir research.

[0004] In recent years, with the rapid development of computer technology, artificial intelligence methods such as support vector machines, random forests, and neural networks have been widely applied to well logging lithology identification, achieving considerable success. For example, the improved multi-granularity cascade forest model for well logging lithology identification has significantly improved recognition accuracy compared to traditional methods. However, these artificial intelligence methods still fall short of achieving ideal recognition results when faced with a small number of sample labels and significant inter-well variability.

[0005] In actual buried-hill lithology identification, due to the deep burial depth of buried hills, problems often arise, such as insufficient core sampling, insufficient labeling, and inaccurate mineral content calculations. These issues severely restrict the accurate identification and evaluation of buried-hill intrusive rock lithology, and thus hinder further understanding of buried-hill reservoirs and oil deposits. Therefore, addressing these issues and developing a more efficient and accurate intelligent interpretation method for complex lithology in buried-hill intrusive reservoirs is of great practical significance and has broad application prospects for advancing buried-hill oil and gas exploration technology. Summary of the Invention

[0006] In response to the problems existing in the prior art, this application provides a method and system for intelligent interpretation of complex lithologies of buried-hill intrusive rock reservoirs. According to the basic laws of magma crystallization chemistry and the logging response characteristics of different mineral contents, an intelligent interpretation model is constructed to achieve intelligent and refined batch interpretation of complex lithologies of buried-hill intrusive rock reservoirs, thereby effectively improving the interpretation accuracy and efficiency of complex lithologies of buried-hill intrusive rock reservoirs.

[0007] In a first aspect, the present application provides a method for intelligently interpreting complex lithology of a buried-hill intrusive reservoir, the steps of which are as follows:

[0008] Acquisition steps: Obtain well logging curves, XRF element logging data, and rock thin section data of the drilled target layer in the study area;

[0009] Sample set construction steps: construct an XRF sample set based on the mineral content calculated from XRF element logging data and its corresponding depth logging curves, and construct a rock thin section sample set based on the mineral content identified from some rock thin sections and their corresponding depth logging curves;

[0010] Clustering step: Spectral clustering is performed on the logging curves of the drilled target layer. The sample points corresponding to the remaining rock thin section depth points in the clustering results are used as a validation set, and the sample points without rock thin sections are used as a prediction set.

[0011] Model training steps: the random forest model is trained on the XRF sample set to obtain a random forest model based on XRF samples, and the random forest model is trained on the rock thin section sample set to obtain a random forest model based on rock thin section samples;

[0012] Model validation steps: The performance of the random forest model based on XRF samples and the random forest model based on rock thin section samples were verified using validation sets. The optimal model for calculating mineral content for different well logging curve categories using spectral clustering was determined based on the performance evaluation indicators of the two models.

[0013] Prediction step: input the prediction set into the optimal model for classification prediction to obtain the content of different minerals;

[0014] Lithologic classification steps: Use the QAP triangle chart of the intrusive rock to cast points with different mineral contents to determine the lithologic category of the intrusive rock.

[0015] In some embodiments, in the sample set construction step, the method for constructing the XRF sample set based on the mineral content calculated by XRF element logging data and its corresponding depth logging curve is: calculating the content of different minerals based on XRF element logging data, using the content of different minerals as labels, and the logging curve corresponding to the XRF calculated mineral depth point as a feature set to construct the XRF sample set.

[0016] In some embodiments, the method for calculating the content of different minerals based on XRF element logging is:

[0017] Data preprocessing steps: Iron adjustment, volatile component processing, and closing processing are performed on XRF element logging data;

[0018] Molecular number calculation: convert the oxide content into molecular number;

[0019] Secondary component processing step: merge the secondary oxides;

[0020] Steps for constructing natural minerals: construct natural minerals in different mineral combinations according to the proportional relationship between cations in natural objects;

[0021] Conversion steps: Multiply the number of standard mineral molecules by the corresponding molecular weight to obtain the standard mineral mass percentage.

[0022] In some embodiments, in the sample set construction step, the method for constructing the rock thin section sample set based on the identification of mineral content of some rock thin sections and their corresponding depth logging curves is: identifying the content of different minerals in each rock thin section, using the identified content of different minerals as labels, and the logging curves corresponding to the rock thin section sampling depth as a feature set to construct the rock thin section sample set.

[0023] In some embodiments, the interpretation method further comprises a pre-processing step:

[0024] The well logging curve is standardized and missing values ​​are filled to obtain the standardized well logging curve after the target layer and the entire well section are filled;

[0025] The XRF element logging data is standardized to obtain standardized XRF element logging data.

[0026] In some embodiments, the well logging curves include a natural gamma ray curve, a caliper curve, a neutron curve, a density curve, a photoelectric absorption cross-section index curve, a deep lateral resistivity curve, and a shallow lateral resistivity curve.

[0027] In some embodiments, in the model verification step, the performance is the accuracy of identifying different categories, the performance evaluation indicator is the root mean square error, and the optimal model for spectral clustering different logging curve categories is determined based on the root mean square error.

[0028] In some embodiments, the method for determining the optimal model for different minerals based on the root mean square error is: determine whether the root mean square error of the random forest model based on the XRF sample is smaller than the root mean square error of the random forest model based on the rock thin section sample; if the root mean square error of the random forest model based on the XRF sample is smaller than the root mean square error of the random forest model based on the rock thin section sample, the random forest model based on the XRF sample is the optimal model; otherwise, the random forest model based on the rock thin section sample is the optimal model.

[0029] In a second aspect of the present application, a system for intelligent interpretation of complex lithology of buried-hill intrusive reservoirs is provided, which is used to implement the method for intelligent interpretation of complex lithology of buried-hill intrusive reservoirs described in the first aspect of the present application, comprising:

[0030] Acquisition module, which obtains the well logging curves, XRF element logging data and rock thin section data of the drilled target layer in the study area;

[0031] The sample set construction module constructs XRF sample sets based on the mineral results calculated from XRF element logging data and their corresponding depth logging curves, and constructs rock thin section sample sets based on the mineral content identified from some rock thin sections and their corresponding depth logging curves;

[0032] The clustering module performs spectral clustering on the target well's logging curves, and forms a validation set with the sample points corresponding to the remaining rock thin section depth points in the clustering results, and a prediction set with the sample points without rock thin sections;

[0033] Model training module, which trains the random forest model with the XRF sample set to obtain a random forest model based on XRF samples, and trains the random forest model with the rock thin section sample set to obtain a random forest model based on rock thin section samples;

[0034] The model validation module verifies the performance of the random forest model based on XRF samples and the random forest model based on rock thin section samples through the validation set, and determines the optimal model for different logging curve categories in the prediction set spectral clustering based on the two model performance evaluation indicators;

[0035] The prediction module inputs the prediction set into the optimal model for classification prediction to obtain the content of different minerals;

[0036] The lithology classification module uses the QAP triangle chart of intrusive rocks to project the content of different minerals and determine the lithology category of the intrusive rocks.

[0037] In some embodiments, the system further includes a preprocessing module, which performs standardization processing on the logging curve and fills in missing values ​​to obtain a standardized logging curve after the target layer is filled in for the entire well section; and performs standardization processing on the XRF element logging data to obtain standardized XRF element logging data.

[0038] Compared with the prior art, the advantages and positive effects of this application are:

[0039] The intelligent interpretation method and system for complex lithology of buried-hill intrusive reservoirs provided in this application comprehensively considers the problems of insufficient rock thin section labels and inaccurate mineral content calculations from XRF elemental logging. Based on the basic laws of magma crystallization chemistry and the logging response characteristics of different mineral contents, two different random forest models are trained using a constructed XRF sample set and a rock thin section sample set. Based on the clustering results of different characteristics of the logging data, a validation set and a prediction set are constructed. The performance of the two models is verified using the validation set, and the optimal model is determined based on performance evaluation indicators. The optimal model is used to predict the content of different minerals based on the prediction set. Finally, the QAP triangle map of the intrusive rock is used to project different mineral contents and determine the lithology category of the intrusive rock. This application can perform intelligent and refined batch interpretation of complex lithology of buried-hill intrusive rocks, effectively improving the interpretation accuracy and efficiency of complex lithology of buried-hill intrusive rocks, and providing a theoretical basis for subsequent analysis of the main controlling factors of buried-hill reservoir formation and prediction of favorable reservoirs. Compared with a single machine model, the greatest advantage of this application is the fusion of multi-source data and the joint characterization of multiple models. Different prediction models are used for different logging response classifications, thus compensating for the prediction error caused by insufficient samples or the measurement error of XRF logging itself when a single model is used alone, and realizing intelligent and detailed lithologic interpretation under dual-model control. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 This is a flow chart of the intelligent interpretation method for complex lithology of buried hill intrusive reservoirs described in an embodiment of the present application;

[0041] Figure 2 This is a structural block diagram of the intelligent interpretation system for complex lithology of buried hill intrusive rock reservoirs described in an embodiment of the present application;

[0042] Figure 3 The random forest model based on XRF samples described in the embodiment of this application is used to predict the mineral content of a single well and interpret the lithology of the pre-Paleogene buried hill intrusive reservoir in a certain study area.

[0043] Figure 4 The random forest model based on rock thin section samples described in the embodiment of this application is used to predict the single-well mineral content and lithologic interpretation results of the pre-Paleogene buried-hill intrusive rock reservoir in a certain study area.

[0044] Figure 5 The present invention provides a single-well lithologic identification result diagram of the pre-Paleogene buried-hill intrusive reservoir in a certain study area using the complex lithologic intelligent interpretation method and system for buried-hill intrusive reservoirs described in the embodiments of the present application.

[0045] In the figure, 1. Acquisition module, 2. Sample set construction module, 3. Clustering module, 4. Model training module, 5. Model verification module, 6. Prediction module, 7. Lithology classification module, 8. Preprocessing module. DETAILED DESCRIPTION

[0046] The present application will be described in detail below with reference to exemplary embodiments in conjunction with the accompanying drawings. However, it should be understood that elements, structures, and features in one embodiment may also be beneficially combined in other embodiments without further description.

[0047] The present application provides a method and system for intelligent interpretation of complex lithology of buried hill intrusive rock reservoirs. The method and system respectively train random forest models by using an XRF sample set constructed based on XRF element logging data and a rock thin section sample set constructed based on rock thin sections to obtain two random forest models with different labels. The performance of the two models is respectively verified by using a validation set constructed by spectral clustering based on logging curves. The optimal model for spectral clustering of different logging curve categories is determined according to the performance evaluation index; different mineral contents are predicted by the optimal model; and the QAP triangle map of the intrusive rock is used to project different mineral contents to determine the lithology category of the intrusive rock. The method and system for intelligent interpretation of complex lithology of buried hill intrusive rock reservoirs of the present application can perform intelligent and refined batch interpretation of complex lithology of buried hill intrusive rock reservoirs. The above-mentioned method and system for intelligent interpretation of complex lithology of buried hill intrusive rock reservoirs are described in detail below in conjunction with the accompanying drawings.

[0048] See also Figure 1 The first embodiment of the present application provides a method for intelligent interpretation of complex lithology of buried-hill intrusive reservoirs, the steps of which are as follows:

[0049] S1. Acquisition step: Acquire the well logging curves, XRF element logging data and rock thin section data of the drilled target layer in the study area.

[0050] In the examples of the present application, a variety of different types of data sources are integrated to provide rich basic information for subsequent lithologic interpretation, avoiding the limitations that may be brought about by a single data source and making the interpretation results more reliable and accurate.

[0051] In some embodiments of the present application, the logging curves include a natural gamma ray curve, a caliper curve, a neutron curve, a density curve, a photoelectric absorption cross-section index curve, a deep lateral resistivity curve, and a shallow lateral resistivity curve.

[0052] S2. Sample set construction steps: construct an XRF sample set based on the mineral content calculated from XRF element logging data and its corresponding depth logging curve, and construct a rock thin section sample set based on the mineral content identified from some rock thin sections and their corresponding depth logging curves.

[0053] In some embodiments of the present application, the method for constructing an XRF sample set based on the mineral content calculated by XRF element logging data and its corresponding depth logging curve is as follows: the content of different minerals is calculated based on the XRF element logging data, the content of different minerals is used as a label, and the XRF calculated mineral depth point corresponding to the depth logging curve is used as a feature set to construct the XRF sample set.

[0054] In the embodiment of the present application, the mineral content calculated based on the XRF element logging data is used as a label, and the logging curve is used as a feature set to construct an XRF sample set. This combination can closely link the mineral composition information with the logging response characteristics, provide samples with clear physical meaning for subsequent model training, and help the model better learn the intrinsic relationship between mineral content and logging curves, thereby improving the prediction accuracy of mineral content.

[0055] In some embodiments of the present application, the method for calculating the content of different minerals based on XRF element logging data is:

[0056] S21. Data preprocessing: Perform iron adjustment, volatile component processing, and closure processing on the XRF element logging data. These processes ensure data accuracy and integrity.

[0057] Specifically, iron adjustment involves converting Fe₂O₃ to FeO or other forms to standardize calculations. Volatile processing involves removing volatile components (such as H₂O and CO₂) from the sample to ensure calculation accuracy. Closing ensures that the sum of all oxides is 100% to meet rock chemical composition requirements.

[0058] S22. Molecular number calculation: convert the oxide content into molecular number.

[0059] Specifically, the content (i.e., mass percentage) of each oxide is divided by its molecular weight to obtain the corresponding number of molecules. This step is to convert the mass percentage to the number of molecules to facilitate subsequent chemical conversions.

[0060] S23, minor component processing step: merge the minor oxides.

[0061] For example, considering the isomorphic substitution relationship between Mn and Ni and Fe, the molecular numbers of MnO and NiO are added to FeO, that is, MnO and NiO are merged into FeO. Another example: Sr and Ba have similar chemical properties to Ca and often undergo isomorphic substitution with Ca. The molecular numbers of SrO and BaO are added to CaO, that is, SrO and BaO are merged into CaO.

[0062] S24. Steps for constructing natural minerals: Construct natural minerals in different mineral combinations according to the proportional relationship between cations in natural objects.

[0063] Illustratively, natural minerals include accessory minerals and primary minerals.

[0064] The formation of accessory minerals is as follows:

[0065] Apatite is formed based on the ratio of CaO to P2O5. For example, apatite is formed by combining 3.33 times the amount of CaO and P2O5. If Cl and F are present in XRF elemental logging data, they first participate in the formation of apatite. If the amount of Cl is greater than 0.66 of the amount of P2O5, Cl is first subtracted from the amount of apatite (Cl' = Cl - 0.666 of P2O5), and the remaining Cl (i.e., Cl') forms NaCl. If the amount of Cl is less than 0.666 of the amount of P2O5, all Cl enters the apatite, and the remaining Cl is supplemented by F. Ultimately, the remaining F forms fluorite.

[0066] Ilmenite is formed based on the ratio of FeO to TiO2. If FeO is greater than TiO2, equal amounts of FeO and TiO2 form ilmenite. If FeO is less than TiO2, the excess TiO2 and the same amount of CaO (limited to the remaining CaO after the formation of anorthite) first form sphene. If there is still excess TiO2, rutile forms.

[0067] Pyrite is formed according to the ratio of S to FeO, and SO3 combines with an equal amount of CaO to form anhydrous gypsum.

[0068] Chromite is formed according to the ratio of Cr2O3 and FeO. If FeO is insufficient, MgO is added to form magnesium chromite.

[0069] The main minerals are formed as follows:

[0070] Orthoclase is formed according to the ratio of K2O and Al2O3, for example: K2O and equal amounts of Al2O3 form orthoclase.

[0071] If Al₂O₃ is excessive, it combines with an equal amount of Na₂O to form albite. If Al₂O₃ is insufficient and Na₂O is excessive, Na₂O combines with Al₂O₃ to form albite, and the excess Na₂O is distributed to anorthite, leaving no anorthite (An). If there is any excess Na₂O, it combines with Fe₂O₃ to form aegirine. If excess Na₂O remains after aegirine is formed, sodium silicate is formed. If Na₂O is insufficient and Fe₂O₃ is excessive, the remaining Fe₂O₃ combines with FeO to form magnetite. If Fe₂O₃ remains after Fe₂O₃ combines with FeO to form magnetite, the remaining Fe₂O₃ forms hematite. If Fe₂O₃ is insufficient and FeO is excessive, MgO combines with the remaining FeO to form F₃ (F₃ = MgO + FeO). The molecular weight of F₃ and the corresponding minerals are calculated based on the ratio of MgO to the remaining FeO. F₃ is then used to generate wollastonite, pyroxene, and olivine.

[0072] If there is still Al2O3 left, the remaining Al2O3 combines with an equal amount of CaO to form anorthite. If Al2O3 is insufficient and CaO is in excess, after CaO combines with the remaining Al2O3 to form anorthite, the remaining CaO combines with an equal amount of Fm to form diopside. If there is still CaO left and Fm is insufficient, the remaining CaO forms wollastonite. If Fm remains and CaO is insufficient, the remaining Fm constitutes pyroxene and olivine. It should be noted that wollastonite does not coexist with pyroxene and olivine. It should also be noted that the appearance of diopside and wollastonite is an indicator mineral of unsaturated Al cations.

[0073] If Al2O3 still remains, corundum is formed.

[0074] SiO2 is distributed among sphene, pyroxene, orthoclase, albite, anorthite, diopside, wollastonite, or pyroxene. The remaining SiO2 forms quartz. If SiO2 is insufficient, it is deducted from the SiO2 that forms pyroxene. If SiO2 is still insufficient, it is supplemented by the SiO2 released from the conversion of albite to nepheline. If the amount of SiO2 does not reach twice the amount of Na2O, all Na2O forms nepheline.

[0075] S25. Conversion steps: Multiply the number of standard mineral molecules by the corresponding molecular weight to obtain the standard mineral mass percentage.

[0076] It should be noted that the above method for calculating the content of different minerals basically covers all types of minerals and their variants in granite.

[0077] In some embodiments of the present application, a method for constructing a rock thin section sample set based on the mineral content of a portion (e.g., 80%) of rock thin sections and their corresponding depth logging curves is as follows: the content of different minerals in each rock thin section is identified, and the rock thin section sample set is constructed using the identified content of different minerals as a label and the corresponding depth logging curves of the rock thin section sampling as a feature set.

[0078] In the embodiment of the present application, by identifying the mineral content in rock thin sections and combining the corresponding logging curves to construct a rock thin section sample set, the rock thin sections can intuitively reflect the microstructure and mineral composition of the rock. The rock thin section sample set provides more intuitive and accurate mineral content information for subsequent model training, further enriches the model's training samples, and enhances the model's ability to recognize lithologic characteristics.

[0079] S3. Clustering step: Spectral clustering is performed on the logging curves of the target well, and the sample points corresponding to the remaining (for example, 20%) rock thin section depth points in the clustering results are used as a validation set, and the sample points without rock thin sections are used as a prediction set.

[0080] In this embodiment, spectral clustering is performed on the well logging curves. After a portion (e.g., 80%) of the rock thin sections are used to construct a rock thin section sample set, the remaining (e.g., 20%) rock thin section depth points are used to form a validation set, while the sample points without rock thin sections form a prediction set. This partitioning method fully utilizes the known information of the rock thin sections. The validation set can be used to accurately evaluate the model's performance, ensuring the reliability of the model on samples with known lithology. The prediction set, on the other hand, provides the model with prediction space for samples with unknown lithology, enabling the model to predict lithology in well sections without rock thin sections, thus expanding the scope of lithology interpretation.

[0081] S4. Model training step: training the random forest model with the XRF sample set to obtain a random forest model based on XRF samples, and training the random forest model with the rock thin section sample set to obtain a random forest model based on rock thin section samples.

[0082] In this example, random forest models were trained using XRF and rock thin section sample sets, resulting in two random forest models: one based on XRF samples and the other on rock thin section samples. The XRF sample set focuses on the relationship between mineral composition and well logs, while the rock thin section sample set focuses on the relationship between rock mineral composition, microstructure, and well logs. Each model has its own advantages, allowing for different perspectives on lithology interpretation. This provides multiple options for selecting the optimal model, enhancing the flexibility and accuracy of lithology interpretation.

[0083] In some embodiments of the present application, a K-fold cross-validation method is used to train random forest models. Taking the 10-fold cross-validation method as an example, during each model training process, 70% of the sample data is randomly selected as the training set and 30% of the sample data is randomly selected as the test set. Model hyperparameters are optimized through grid search and 10-fold cross-validation, respectively, to obtain random forest models based on XRF samples and rock thin section samples.

[0084] S5. Model validation step: The performance of the random forest model based on XRF samples and the random forest model based on rock thin section samples are respectively verified through the validation set, and the optimal model corresponding to different logging curve categories in the prediction set is determined according to the two model performance evaluation indicators.

[0085] In the embodiment of the present application, the performance of the two random forest models is verified through a validation set, and the optimal model for different logging curve categories of spectral clustering is determined based on performance evaluation indicators (for example, root mean square error, etc.). This can accurately select a model that is more accurate in predicting mineral content under current data conditions, thereby achieving accurate classification interpretation.

[0086] In some embodiments of the present application, the performance is the accuracy of identifying different mineral contents, and the performance evaluation indicator is the root mean square error. The optimal model for different logging curve categories is determined based on the root mean square error.

[0087] The root mean square error (RMS) can intuitively reflect the deviation between the model's predicted value and the actual value. In the examples of this application, by comparing the RMS errors of two models, the model that more accurately predicts mineral content under the current data conditions can be accurately selected, thereby ensuring the high reliability of the model used in subsequent prediction steps and further improving the accuracy of lithologic interpretation.

[0088] In some embodiments of the present application, the method for determining the optimal model for different logging curve categories based on the root mean square error is: determine whether the root mean square error of the random forest model based on XRF samples is smaller than the root mean square error of the random forest model based on rock thin section samples; if the root mean square error of the random forest model based on XRF samples is smaller than the root mean square error of the random forest model based on rock thin section samples, then the random forest model based on XRF samples is the optimal model; otherwise, the random forest model based on rock thin section samples is the optimal model.

[0089] S6. Prediction step: Input the prediction set into the optimal model for classification prediction to obtain the content of different minerals.

[0090] It should be noted that since the optimal model was strictly screened from the validation set, it can more accurately predict the mineral content in the prediction set, providing a reliable basis for subsequent lithologic classification, achieving efficient and accurate prediction of the complex lithology of the buried-hill intrusive reservoir, and improving the efficiency and quality of lithologic interpretation.

[0091] S7. Lithologic classification steps: Use the QAP triangle chart of the intrusive rock to project the content of different minerals to determine the lithologic category of the intrusive rock.

[0092] It should be noted that the QAP triangle chart is a commonly used lithologic classification tool in geology. The volume fractions of three minerals, Q, A, and P, are determined by using methods such as rock thin sections, XRF elemental logging, and conventional well log interpretation. These data are then projected onto the QAP chart, and the rock name and type are determined based on the location of the projected points. This method is both scientific and authoritative. Q represents the percentage of quartz in the total content of quartz, alkali feldspar, and plagioclase, typically calculated by dividing the volume fraction of quartz by the total volume fraction of quartz, alkali feldspar, and plagioclase. A represents the percentage of alkali feldspar in the total content of alkali feldspar and plagioclase, typically calculated by dividing the volume fraction of alkali feldspar by the total volume fraction of alkali feldspar and plagioclase. P represents the percentage of plagioclase in the total content of alkali feldspar and plagioclase, typically calculated by dividing the volume fraction of plagioclase by the total volume fraction of alkali feldspar and plagioclase. In the embodiment of the present application, by projecting the predicted mineral content on the QAP triangle chart, the lithologic category of the intrusive rock can be intuitively determined, thereby achieving a scientific conversion from mineral content to lithologic category, and providing clear lithologic classification results for the lithologic evaluation of the buried hill intrusive rock reservoir and further geological research.

[0093] In some embodiments of the present application, the interpretation method further includes a preprocessing step:

[0094] The well logging curve is standardized and missing values ​​are filled to obtain the standardized well logging curve after the target layer and the entire well section are filled;

[0095] The XRF element logging data is standardized to obtain standardized XRF element logging data.

[0096] In this application, effective preprocessing of well log and XRF elemental logging data was performed to improve data quality and consistency. Standardization eliminated dimensional and data range differences between different well log and XRF elemental logging data, making the data more suitable for subsequent model training and analysis. Missing value imputation filled gaps in the data, ensuring data integrity.

[0097] The above-mentioned intelligent interpretation method for complex lithology of buried hill intrusive rock reservoirs in this application uses the mineral content results of XRF interpretation to make up for the problem of fewer rock thin sections. At the same time, the accuracy of XRF is reversed using rock thin sections to obtain a double-label sample set. The random forest model of XRF samples and the random forest model based on rock thin section samples are obtained based on the training model of the double-label sample set. The accuracy of the random forest model based on XRF samples and the random forest model based on rock thin section samples are tested by using the calibration curve clustering results with rock thin sections, and the best model is selected. At the same time, the logging response characteristics of different classifications can also be determined. According to the curve clustering results without rock thin section calibration, the best model is selected to predict the mineral content in combination with the logging response characteristics, thereby realizing the interpretation of different lithologies, and realizing the intelligent and refined batch interpretation of complex lithology of buried hill intrusive rock reservoirs.

[0098] See also Figure 2 The second embodiment of the present application provides a system for intelligent interpretation of complex lithology of buried-hill intrusive reservoirs, which is used to implement the method for intelligent interpretation of complex lithology of buried-hill intrusive reservoirs described in the first aspect of the present application, including:

[0099] Acquisition module 1, obtains the well logging curves, XRF element logging data and rock thin section data of the drilled target layer in the study area;

[0100] Sample set construction module 2: constructs an XRF sample set based on the mineral content calculated from XRF element logging data and its corresponding depth logging curve, and constructs a rock thin section sample set based on the mineral content identified from some rock thin sections and their corresponding depth logging curves;

[0101] Clustering module 3 performs spectral clustering on the target well's logging curves, and forms a validation set with the sample points corresponding to the remaining rock thin section depth points in the clustering results, and forms a prediction set with the sample points without rock thin sections;

[0102] Model training module 4, training the random forest model with the XRF sample set to obtain a random forest model based on XRF samples, and training the random forest model with the rock thin section sample set to obtain a random forest model based on rock thin section samples;

[0103] Model validation module 5 verifies the performance of the random forest model based on XRF samples and the random forest model based on rock thin section samples through the validation set, and determines the best model for different logging curve categories in the prediction set based on the two model performance evaluation indicators;

[0104] Prediction module 6, inputs the prediction set into the optimal model for classification prediction to obtain the content of different minerals;

[0105] The lithology classification module 7 uses the QAP triangle chart of the intrusive rock to project the content of different minerals and determine the lithology category of the intrusive rock.

[0106] In some embodiments of the application, see Figure 2 The system also includes a preprocessing module 8, which performs standardization processing on the logging curve and fills in missing values ​​to obtain a standardized logging curve after the target layer and the entire well section is filled; and standardizes the XRF element logging data to obtain standardized XRF element logging data.

[0107] The above-mentioned intelligent interpretation system for complex lithology of buried hill intrusive rock reservoirs in this application uses the mineral content results of XRF interpretation to make up for the problem of fewer rock thin sections. At the same time, the accuracy of XRF is reversed using rock thin sections to obtain a double-label sample set. The random forest model of XRF samples and the random forest model based on rock thin section samples are obtained based on the training model of the double-label sample set. The accuracy of the random forest model of XRF samples and the random forest model based on rock thin section samples are tested by clustering the calibration curves with rock thin sections, and the best model is selected. At the same time, the logging response characteristics of different classifications can also be determined. According to the clustering results of the calibration curves without rock thin sections, the best model is selected to predict the mineral content in combination with the logging response characteristics, thereby realizing the interpretation of different lithologies, and realizing the intelligent and refined batch interpretation of complex lithology of buried hill intrusive rock reservoirs.

[0108] In order to verify the effectiveness of the intelligent interpretation method and system for complex lithology of buried hill intrusive reservoirs described in the above embodiments of this application, the following specific embodiments are used for illustration.

[0109] Example: Take the pre-Paleogene buried hill intrusive reservoir in a certain study area as an example.

[0110] The well logging curves of the pre-Paleogene buried-hill intrusive reservoir in a certain study area are obtained. The well logging curves include natural gamma ray curve GR, caliper curve CAL, neutron curve CN, density curve DEN, photoelectric absorption cross-section index curve PE, deep lateral resistivity curve RT, and shallow lateral resistivity curve RXO.

[0111] The natural gamma ray curve GR, caliper curve CAL, neutron curve CN, density curve DEN, photoelectric absorption cross section index curve PE, deep lateral resistivity curve RT and shallow lateral resistivity curve RXO are pre-processed and missing values ​​are filled to obtain the standardized logging curve, as shown in the attached figure. Figure 3 、 Figure 4 、 Figure 5 Well logging curve on the left.

[0112] The XRF element logging data of the target well is standardized, and the quartz, plagioclase and potassium feldspar mineral contents calculated according to the basic laws of magma crystallization chemistry are used as labels, as shown in the attached Figure 3 Intermediate mineral XRF calculation curve ( Figure 3 (blue, orange, and gray fill curves in ).

[0113] The well logging curves were spectrally clustered. At the same time, the minerals identified by rock thin sections and the minerals calculated by XRF were double-labeled and the well logging curves were used for model training respectively.

[0114] The prediction results of the dual model are shown in the attached Figure 5 The rightmost mineral dual model prediction curve and the lithology results divided by the mineral QAP triangle map are shown in the attached figure. Figure 5 The rightmost lithology channel. Compared with the existing technology, the intelligent interpretation method and system of complex lithology of buried hill intrusive reservoirs in this application can effectively improve the interpretation accuracy of complex lithology of buried hill intrusive reservoirs. By comparison, it can be found that whether it is XRF calculation of mineral content or thin section prediction of mineral content ( Figure 3 、 Figure 4 The blue, orange and gray filling curves in the figure) and the actual thin section identification mineral content ( Figure 3 、 Figure 4 The blue, orange, and gray bar graphs all have large errors, while the mineral content predicted by the dual model has a better trend of agreement with the mineral content identified by thin sections ( Figure 5 (The blue, orange, and gray filling curves and bar graphs in the figure are shown in Tables 1, 2, and 3, respectively.) The statistical errors of some of the results are shown in Tables 1, 2, and 3. By comparison, the dual model has a higher prediction accuracy for mineral content, and thus, more accurate lithology identification results can be obtained by projecting points on the QAP chart.

[0115] Table 1

[0116]

[0117] Table 2

[0118]

[0119] Table 3

[0120]

[0121] The above embodiments are used to explain the present application rather than to limit the present application. Any modifications and changes made to the present application within the spirit of the present application and the protection scope of the claims shall fall within the protection scope of the present application.

Claims

1. An intelligent interpretation method for complex lithology of buried-hill intrusive reservoirs, characterized by: The steps are: Acquisition steps: Obtain well logging curves, XRF element logging data, and rock thin section data of the drilled target layer in the study area; Sample set construction steps: construct an XRF sample set based on the mineral content calculated from XRF element logging data and its corresponding depth logging curves, and construct a rock thin section sample set based on the mineral content identified from some rock thin sections and their corresponding depth logging curves; Clustering step: Spectral clustering is performed on the logging curves of the drilled target layer. The sample points corresponding to the remaining rock thin section depth points in the clustering results are used as a validation set, and the sample points without rock thin sections are used as a prediction set. Model training steps: the random forest model is trained on the XRF sample set to obtain a random forest model based on XRF samples, and the random forest model is trained on the rock thin section sample set to obtain a random forest model based on rock thin section samples; Model validation steps: The performance of the random forest model based on XRF samples and the random forest model based on rock thin section samples were verified using validation sets. The optimal model for calculating mineral content for different well logging curve categories using spectral clustering was determined based on the performance evaluation indicators of the two models. Prediction step: input the prediction set into the optimal model for classification prediction to obtain the content of different minerals; Lithologic classification steps: Use the QAP triangle chart of the intrusive rock to cast points with different mineral contents to determine the lithologic category of the intrusive rock.

2. The intelligent interpretation method for complex lithology of buried-hill intrusive reservoirs according to claim 1, characterized in that: In the sample set construction step, the method for constructing the XRF sample set based on the mineral content calculated by XRF element logging data and its corresponding depth logging curve is as follows: the content of different minerals is calculated based on the XRF element logging data, the content of different minerals is used as a label, and the depth logging curve corresponding to the XRF calculated mineral depth point is used as a feature set to construct the XRF sample set.

3. The intelligent interpretation method for complex lithology of buried-hill intrusive reservoirs according to claim 2, characterized in that: The method for calculating the content of different minerals based on XRF element logging is: Data preprocessing steps: Iron adjustment, volatile component processing, and closing processing are performed on XRF element logging data; Molecular number calculation: convert the oxide content into molecular number; Secondary component processing step: merge the secondary oxides; Steps for constructing natural minerals: construct natural minerals in different mineral combinations according to the proportional relationship between cations in natural objects; Conversion steps: Multiply the number of standard mineral molecules by the corresponding molecular weight to obtain the standard mineral mass percentage.

4. The intelligent interpretation method for complex lithology of buried-hill intrusive reservoirs according to claim 1, characterized in that: In the sample set construction step, the method for constructing the rock thin section sample set based on the identification of mineral content of some rock thin sections and their corresponding depth logging curves is as follows: identifying the content of different minerals in each rock thin section, using the identified different mineral contents as labels, and the corresponding depth logging curves of rock thin section sampling as a feature set to construct the rock thin section sample set.

5. The intelligent interpretation method for complex lithology of buried-hill intrusive reservoirs according to claim 1, characterized in that: The interpretation method also includes a pre-processing step: The well logging curve is standardized and missing values ​​are filled to obtain the standardized well logging curve after the target layer and the entire well section are filled; The XRF element logging data is standardized to obtain standardized XRF element logging data.

6. The intelligent interpretation method for complex lithology of buried-hill intrusive reservoirs according to claim 1, characterized in that: The logging curves include a natural gamma ray curve, a caliper curve, a neutron curve, a density curve, a photoelectric absorption cross-section index curve, a deep lateral resistivity curve, and a shallow lateral resistivity curve.

7. The intelligent interpretation method for complex lithology of buried-hill intrusive reservoirs according to claim 1, characterized in that: In the model verification step, the performance is the accuracy of prediction of different mineral contents, and the performance evaluation index is the root mean square error. The optimal model of different logging curve categories of spectral clustering is determined based on the root mean square error.

8. The intelligent interpretation method for complex lithology of buried-hill intrusive reservoirs according to claim 7, characterized in that: The method for determining the optimal model for different minerals based on the root mean square error is as follows: determine whether the root mean square error of the random forest model based on XRF samples is smaller than the root mean square error of the random forest model based on rock thin section samples. If the root mean square error of the random forest model based on XRF samples is smaller than the root mean square error of the random forest model based on rock thin section samples, the random forest model based on XRF samples is the optimal model; otherwise, the random forest model based on rock thin section samples is the optimal model.

9. An intelligent interpretation system for complex lithology of buried-hill intrusive reservoirs, used to implement the intelligent interpretation method for complex lithology of buried-hill intrusive reservoirs according to any one of claims 1 to 8, characterized in that: include: Acquisition module, which obtains the well logging curves, XRF element logging data and rock thin section data of the drilled target layer in the study area; The sample set construction module constructs XRF sample sets based on the mineral content calculated from XRF element logging data and its corresponding depth logging curves, and constructs rock thin section sample sets based on the mineral content identified from some rock thin sections and their corresponding depth logging curves; The clustering module performs spectral clustering on the target well's logging curves, and forms a validation set with the sample points corresponding to the remaining rock thin section depth points in the clustering results, and a prediction set with the sample points without rock thin sections; Model training module, which trains the random forest model with the XRF sample set to obtain a random forest model based on XRF samples, and trains the random forest model with the rock thin section sample set to obtain a random forest model based on rock thin section samples; The model validation module verifies the performance of the random forest model based on XRF samples and the random forest model based on rock thin section samples through the validation set, and determines the optimal model for different logging curve categories in the prediction set spectral clustering based on the two model performance evaluation indicators; The prediction module inputs the prediction set into the optimal model for classification prediction to obtain the content of different minerals; The lithology classification module uses the QAP triangle chart of intrusive rocks to project the content of different minerals and determine the lithology category of the intrusive rocks.

10. The intelligent interpretation system for complex lithology of buried hill intrusive reservoirs according to claim 9, characterized in that: The system also includes a preprocessing module, which performs standardization processing on the logging curve and fills in missing values ​​to obtain a standardized logging curve after the target layer and the entire well section is filled; and performs standardization processing on the XRF element logging data to obtain standardized XRF element logging data.

Images