Metamorphic rock buried hill while-drilling rapid layer clamping method and application thereof
By employing a rapid layer identification method for metamorphic buried hills during drilling based on sensitive logging element screening and nonlinear machine learning algorithms, the problems of poor quality of deep seismic data and complex lithology identification were solved. This method enabled accurate identification of buried hill interfaces and depth prediction, thereby improving drilling efficiency and safety.
Patent Information
- Application Number
- CN202511558021.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-29
- Publication Date
- 2026-02-03
AI Technical Summary
Existing buried hill drilling layer-finding technology is difficult to accurately identify complex lithological types and subtle mineral variations when the quality of deep seismic data is poor and the seismic reflection signals of ancient buried hills are chaotic and have strong ambiguity.
A rapid layer identification method for buried hills in metamorphic rocks based on sensitive logging element screening and nonlinear machine learning algorithms is adopted. The logging elements are screened by linear discriminant analysis (LDA), and a lithology identification model is constructed by combining self-supervised self-organizing map neural network (SSOM). A rapid prediction model is established to guide the identification of buried hill interfaces and depth prediction.
It enables rapid identification of complex metamorphic buried hills and the lithology of overlying strata, improving identification speed and accuracy, guiding geophysical velocity correction and buried hill top depth prediction, providing a basis for rapid decision-making, and improving the efficiency of buried hill reservoir development.
Smart Images
Figure CN121451948A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the technical field of oil and gas reservoir description, and particularly relates to a metamorphic rock buried hill drilling while caving method based on sensitive logging element screening and a nonlinear machine learning algorithm and application thereof. BACKGROUND
[0002] Deep and ultra-deep oil and gas reservoirs are regarded as the core driving force for future reserve growth, and have become an important focus of global oil and gas exploration and development. In recent years, in the Bohai Bay Basin of China, a number of billion-ton Proterozoic metamorphic rock buried hill oil and gas reservoirs have been continuously proven, further promoting the metamorphic rock buried hill oil and gas reservoir to become a popular object of current exploration and development. Due to poor quality of medium-deep seismic data, and chaotic and strong multi-solution seismic reflection signals of the ancient buried hill, the buried hill drilling while caving work is facing severe challenges, causing great difficulties for drilling tracking of development wells. Although domestic and foreign scholars have made great efforts in the field of accurate identification of buried hill top surface, such as the application of heavy electromagnetic method and conventional seismic data, the accuracy and reliability of identification are still not high due to the complex lithology and variable seismic velocity of the overlying Paleogene system and the ancient buried hill. Therefore, there is an urgent need to develop an efficient and fast buried hill drilling while caving technology.
[0003] Element logging as a relatively common lithology identification method and means has been successfully applied to buried hill drilling while caving decision-making. For example, Hu Yun et al. published "Application of Rock Debris Composition Analysis in Identification of Bohai 19-6 Buried Hill Interface" in the 32nd volume of China Offshore Oil and Gas magazine in December 2020. However, due to the complex lithology and diverse types of the main body of the ancient buried hill and the overlying strata in the target area BZ oilfield, such as mudstone, volcanic breccia, basalt, sandy conglomerate, carbonate rock and granite gneiss, the linear discriminant analysis (LDA) chart-based method is a linear identification method, which is not suitable for solving the nonlinear problem related to multiple factors such as complex buried hill lithology identification. Therefore, it is difficult to accurately identify the diverse types of buried hill and overlying strata lithology and the subtle changes in mineral lithology. The identification effect of volcanic breccia, basalt and gneiss is poor. The application practice in the area shows that the identification accuracy of the LDA linear analysis chart method is low (only 64.7%).
[0004] The nonlinear machine learning method (self-organizing mapping neural network (SSOM)) as a relatively important technical means in data mining and pattern classification can train and correct the model through a supervised learning process to establish a fast prediction model. Especially in solving nonlinear problems related to multiple factors, it shows high precision, fast speed and strong universality, and is more suitable for realizing fast identification of complex lithology of metamorphic rock buried hill and overlying strata. SUMMARY
[0005] The present application is to solve the problems in the prior art that the deep seismic data quality is poor, the seismic reflection characteristics of the buried hill are complex and variable and difficult to interpret, and the lithology of the metamorphic rock buried hill and the overlying strata is complex, the seismic wave velocity fluctuates significantly, which leads to the difficulty in drilling and layer sticking of the buried hill, and the existing buried hill interface identification method cannot accurately identify the buried hill interface and predict the depth of the ancient buried hill with complex lithology types and subtle changes in minerals, and the purpose is to provide a metamorphic rock buried hill drilling rapid layer sticking method based on sensitive logging element screening and nonlinear machine learning algorithm and its application.
[0006] The present application is realized by the following technical solutions:
[0007] A metamorphic rock buried hill drilling rapid layer sticking method based on sensitive logging element screening and nonlinear machine learning algorithm, comprising the following steps:
[0008] S1, analyze the overlying Paleogene reservoir and ancient buried hill reservoir structure combination mode, and establish a typical mountain-entering reservoir structure combination mode;
[0009] Specifically, based on the analysis of the lithology combination characteristics, a typical mountain-entering reservoir structure combination mode is established; the typical mountain-entering reservoir structure combination mode is obtained by analyzing which lithology exists in the Paleogene and Archean buried hill strata, and thus summarizing which lithology change mode exists when entering the Archean buried hill from the Paleogene strata;
[0010] The typical mountain-entering reservoir structure combination mode includes mudstone entering Archean metamorphic rock buried hill, mudstone-sand-gravel rock entering Archean metamorphic rock buried hill, mudstone-basalt-sand-gravel rock entering Archean metamorphic rock buried hill, mudstone-Shahejie sandstone-sand-gravel rock entering Archean metamorphic rock buried hill, mudstone-volcanic breccia entering Archean metamorphic rock buried hill, and mudstone entering Paleozoic limestone buried hill, a total of 6 types of overlying Paleogene reservoir and Paleozoic limestone buried hill and Archean metamorphic rock buried hill reservoir structure combination mode.
[0011] S2, screen the sensitive logging elements and perform data dimension reduction processing based on the linear discriminant analysis (LDA) algorithm, and construct a sample data set through core calibration logging;
[0012] The step S2 specifically includes the following steps:
[0013] S21, sensitive logging element screening based on the LDA algorithm:
[0014] The sensitive logging element screening is performed through the LDA algorithm, and the logging element most sensitive to the complex metamorphic rock buried hill lithology is extracted to guide the establishment of the buried hill layer sticking chart;
[0015] S22, feature vector extraction based on LDA data dimension reduction processing:
[0016] In order to avoid introducing too much noise due to too high spatial dimension of input samples, and thus affecting the prediction performance, LDA algorithm is used to implement data dimension reduction, and four feature vectors with the most information contribution are selected from the data dimension reduction;
[0017] The feature vectors extracted in the step S22 are four feature vectors with the first four inter-group variances in the LDA algorithm;
[0018] S23, constructing sample data set by core calibration logging method:
[0019] In constructing the lithology sample data set, the core calibration logging method is used, and the data set is randomly divided, of which 70% is used as training sample, and the remaining 30% is used as verification sample.
[0020] S3, constructing complex metamorphic rock buried hill lithology identification model based on SSOM (Supervised-mode self-organizing-map) algorithm, and selecting test sample to carry out model generalization performance verification;
[0021] The step S3 comprises the following steps:
[0022] S31, grid structure initialization and parameter setting:
[0023] The grid structure of input layer, competition layer and output layer is set, and the neural network parameters are initialized and set;
[0024] S2, constructing complex metamorphic rock buried hill lithology identification model by LDA and SSOM:
[0025] Based on the constructed learning sample set, the SSOM algorithm is used to train the prediction model, and the complex lithology identification model based on SSOM algorithm is developed;
[0026] S33, model generalization performance verification:
[0027] For the complex metamorphic rock buried hill lithology identification model based on SSOM algorithm, learning sample is used for back-judgment test, and test sample is used for prediction test, so as to comprehensively evaluate whether the performance of the constructed model meets the geological requirements; if the model shows excellent performance, the prediction model will be applied to the rapid identification of complex metamorphic rock buried hill lithology of other development wells in the region;
[0028] S4. Based on the sensitive logging element screening result of step S2 and the prediction result of complex metamorphic rock buried hill lithology identification model of step S3, the ancient buried hill into mountain card layer chart of classical reservoir structure combination mode established in step S1 is constructed.
[0029] S5. The ancient buried hill advancing hill card layer template based on step S4 is constructed to guide geophysical velocity correction and development well advancing buried hill depth prediction.
[0030] The application of a metamorphic rock buried hill while-drilling quick card layer method based on sensitive logging element screening and nonlinear machine learning algorithm in oil and gas reservoir description.
[0031] The beneficial effects of the present application are:
[0032] The present application provides a metamorphic rock buried hill while-drilling quick card layer method based on sensitive logging element screening and nonlinear machine learning algorithm and application thereof, a lithology identification model is constructed by selecting linear discriminant analysis (LDA) algorithm and nonlinear SSOM neural network algorithm, and the model has the advantages of fast identification speed and small error in solving the problem that the buried hill while-drilling card layer is difficult due to poor quality of middle-deep seismic data and chaotic and strong multi-solution seismic reflection signals of ancient buried hill; the evaluation method can be used for oil and gas reservoir description, can effectively guide geophysical velocity correction and depth prediction of the top surface of the buried hill, and provides a quick decision basis for geologists to track while drilling, and has important practical significance for efficient development of buried hill reservoirs. BRIEF DESCRIPTION OF DRAWINGS
[0033] Figure 1 It is a method flowchart of the present application;
[0034] Figure 2 It is a Paleogene sandstone and mudstone and Paleozoic and Archean buried hill reservoir structure combination type diagram provided by example 1 of the present application (a is Donger 2 lower mudstone advancing Archean metamorphic rock buried hill, b is Donger 2 lower mudstone-sandstone advancing Archean metamorphic rock buried hill, c is Donger 2 lower mudstone-basalt-sandstone advancing Archean metamorphic rock buried hill, d is Donger 2 lower mudstone-Shahejie sandstone-sandstone advancing Archean metamorphic rock buried hill, e is Sha 3 mudstone-volcanic breccia advancing Archean metamorphic rock buried hill, and f is Donger 2 lower mudstone advancing Paleozoic limestone buried hill);
[0035] Figure 3 It is a different lithofacies proportion diagram in sample data provided by example 1 of the present application;
[0036] Figure 4 It is a sensitive logging element screening and lithology identification diagram based on LDA algorithm provided by example 1 of the present application (a is a lithology identification diagram based on feature vectors F1, F2 and F3, and b is projection of logging elements on feature vector F1 and feature vector F2);
[0037] Figure 5 It is a Donger 2 lower mudstone advancing metamorphic rock buried hill mode diagram provided by example 1 of the present application (BZ1 well);
[0038] Figure 6 This is a diagram of the buried hill model of mudstone-basalt-conglomerate in the lower section of Dong'er (well BZ2) provided in Embodiment 1 of the present invention.
[0039] Figure 7 This is a schematic diagram of the mudstone confluence with Paleozoic limestone buried hill in the lower section of the Dong'er Formation (well BZ3) provided in Embodiment 1 of the present invention.
[0040] For those skilled in the art, other related figures can be obtained from the above figures without any creative effort. Detailed Implementation
[0041] To enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.
[0042] Example 1
[0043] like Figure 1 As shown, a rapid layer-finding method for metamorphic buried hills based on sensitive logging element screening and nonlinear machine learning algorithms includes the following steps:
[0044] S1. Analyze the structural combination model of the overlying Paleogene reservoir and the ancient buried hill reservoir, and establish a typical structural combination model of the intrusive hill reservoir.
[0045] This embodiment establishes six typical reservoir structure combination modes, such as Figure 2 As shown, it specifically includes:
[0046] ①The buried hill of Archean metamorphic rocks in the lower mudstone of the East II;
[0047] ②The lower mudstone-conglomerate buried hill of Archean metamorphic rocks in the East II;
[0048] ③The buried hill of Archean metamorphic rocks in the lower mudstone-basalt-conglomerate of the East II;
[0049] ④ The lower section of the East Second Formation is a buried hill of Archean metamorphic rocks consisting of mudstone, Shahejie sandstone, and conglomerate.
[0050] ⑤ Sha-3 Member mudstone-volcanic breccia subduction Archean metamorphic buried hill;
[0051] ⑥ Dong Erxia mudstone enters Paleozoic limestone buried hill.
[0052] S2. Based on the Linear Discriminant Analysis (LDA) algorithm, sensitive logging elements are screened and data dimensionality is reduced, and a sample dataset is constructed by core calibration logging.
[0053] The X-ray element logging (XRF) in the study area can detect 17 elements, including: Al (aluminum), Ba (barium), Ca (calcium), Cl (chlorine), Fe (iron), K (potassium), Mg (magnesium), Mn (manganese), Na (sodium), Ni (nickel), P (phosphorus), S (sulfur), Si (silicon), Sr (strontium), Ti (titanium), V (vanadium) and Zr (zirconium).
[0054] During the training of the lithofacies prediction model, the input features need to be carefully selected to avoid including redundant or low correlation logging elements into the model. This not only does not help to improve the prediction accuracy, but also may introduce noise due to the high input dimension, which may interfere with the learning effect of the model. Therefore, reasonable selection of input features and construction of a concise and efficient input set are key steps to ensure the training quality of the lithofacies prediction model.
[0055] Linear discriminant analysis (LDA) is a widely used statistical method for pattern recognition and data dimensionality reduction. Unlike principal component analysis (PCA), which selects the maximum variance projection direction unsupervisedly, LDA is a supervised dimensionality reduction technique relying on prior information. The core of the LDA method is to find linear combinations (i.e. projection vectors) that can maximize the description and discrimination of prior classes. The goal is to maximize the dispersion between prior classes and minimize the dispersion within classes when high-dimensional samples with class labels are projected onto these vectors. Therefore, the dimensionality reduction problem of high-dimensional samples is transformed into the process of finding optimal projection vectors, which can be expressed by a specific mathematical formula:
[0056]
[0057]
[0058] where: x1, x2, …, x n is the logging data, independent variable; k is the class of the output variable, dimensionless; m k is the number of samples belonging to class k, individual; u k is the sample mean of class k, logging unit; u is the overall mean of all samples, logging unit; x d is the sample belonging to class k, dimensionless; S b is the sample inter-class dispersion matrix, dimensionless; S w is the sample intra-class dispersion matrix, dimensionless;
[0059] W opt is the projection vector of the sample, dimensionless;
[0060] The step S2 specifically includes the following steps:
[0061] S21, sensitive logging element screening based on LDA algorithm:
[0062] Based on the projection of the logging elements on the feature vector F1 and the feature vector F2, the sensitivity of each logging element in the classification process can be ranked, and after obtaining the ranking, the most sensitive logging element for complex lithology recognition can be extracted to guide the establishment of the buried hill card layer chart.
[0063] S22, feature vector extraction based on LDA data dimension reduction processing:
[0064] By singular value decomposition of the variance-covariance matrix constructed by the input data, five feature vectors are successfully extracted, each of which represents five different projection dimensions (Table 1); these feature vectors all represent a specific projection direction, and the mathematical expression of the feature vector can be constructed according to the discriminant function:
[0065] F1 = 0.135*Al + 0.077*Ba - 0.026*Ca + 0.002*Cl + 0.221*Fe - 0.017*K - 0.110*Mg + 0.040*Mn - 0.080*Na + 0.061*Ni - 0.010*P + 0.015*S + 0.448*Si + 0.241*Sr - 0.207*Ti + 0.014*V + 0.064*Zr;
[0066] "Eigenvalue" and "information" represent the amount and proportion of information captured by each feature vector to distinguish lithology classification; from F1 to F5, the eigenvalue decreases and the information decreases; in the LDA algorithm, the first four feature vectors represent the maximum inter-group variance, which can cover 87.6% of the lithology grouping information and can most effectively distinguish the prior group; mapping the lithofacies characteristics to these feature vectors can better identify different lithofacies Figure 4 a and 4b); therefore, the first four feature vectors are selected as the input features of the machine learning prediction model.
[0067] Table 1: Standard discriminant function coefficients of five feature vectors
[0068]
[0069]
[0070] S23, construct a sample data set by core calibration logging:
[0071] The 15055 core sample data of BZ oilfield is used as the learning sample and test sample required for the establishment of the prediction model. Each sample includes 4 input variables (x1, x2, x3, x4) and 1 output variable (y); wherein the input variables correspond to the 4 projection vectors (F1, F2, F3, F4) after LDA algorithm dimensionality reduction; the output variable y is the lithology category label, corresponding to 6 typical lithologies of overlying Paleogene reservoir and buried hill reservoir, represented by y = {I, II, III, IV, V, VI}; wherein I, II, III, IV, V and VI correspond to mudstone, volcanic breccia, basalt, sandy conglomerate, carbonate rock and granite gneiss respectively.
[0072] Under the double considerations of ensuring the accuracy of model training and the effectiveness of prediction results, 70% of the sample data is used as the training set, and the remaining 30% is used as the test set. The training and performance verification of the model are carried out by dividing the data in this way.
[0073] Specifically, 10539 groups of samples are randomly selected to construct the training set, which is used for the training process of the model, and on this basis, the prediction model of lithofacies type is established; at the same time, 4516 groups of samples are selected as the test set to comprehensively test the performance of the prediction model.
[0074] S3, based on SSOM (Supervised-mode self-organizing-map) algorithm to construct complex metamorphic rock buried hill lithology identification model, and select test sample to carry out model generalization performance verification;
[0075] SOM algorithm is a kind of unsupervised self-organizing mapping neural network algorithm proposed by Kohonen in 1982 [Kohonen T. Self-organized formation of topologically correct feature maps. Biol[J]. Cybern, 1982, 43(1): 59-69.]; the algorithm is widely used in data clustering and visualization processing, and the network structure is composed of input layer and competitive layer; SOM neural network algorithm can realize sample classification through a series of processes such as network structure initialization, grid training and weight adjustment (equations 6-8).
[0076]
[0077] W ij = W ij + η1(x i -W ij ) …… (8)
[0078] In the formula, i and j are the neurons in the input layer and the competitive layer respectively, which are dimensionless; D jEuclidean distance, dimensionless; x i input variable; g is the winning neuron node, dimensionless; P(j) is the probability of winning of the competition layer neuron j, decimal; μ is a constant; W ij is the weight between the input layer neuron i and the competition layer neuron j, decimal.
[0079] SOM is a machine learning algorithm in unsupervised mode, which can automatically classify samples according to the similarity between sample data; the pattern recognition process does not depend on user-defined information, but the classification accuracy is significantly lower than that of machine learning algorithms in supervised mode; some scholars improve the neural network structure of SOM by adding a third layer (output layer) grid, and develop a self-organizing mapping (SSOM) neural network algorithm in supervised mode (Wang Y, Yang S C, Lu Y, et al. Low permeability reservoir evaluation method based on well logging petrophysical facies recognition: Taking the upper member of Mengyin Formation in Gaoqing area of Dongying Sag as an example [J]. Journal of China University of Mining & Technology, 2018, 47(6): 1264-1275.); in the output layer, the number of nodes corresponds to the number of output categories, and each node represents a specific output category; SSOM neural network has excellent ability to flexibly adjust the weight relationship between the competition layer and the output layer (see formula 9) based on user-defined information, and thus accurately predict the category of unknown samples.
[0080] W jk = W jk + η2(Y k -W jk ) …… (9)
[0081] In the formula, k is the output layer neuron; Y k is the sample category, dimensionless; η2 is the learning rate, dimensionless; W jk is the weight between the output layer neuron k and the competition layer neuron j, decimal. W jk is the weight between the competition layer neuron j and the input layer neuron k, decimal.
[0082] The step S3 includes the following steps:
[0083] S31, grid structure initialization and parameter setting:
[0084] Set the grid structure of the input layer, competition layer and output layer, and initialize and set the neural network parameters;
[0085] The input layer in this training learning is the projection vector of the logging element processed by LDA algorithm, the competition is composed of a 10*10 (100 neurons) topology network, each competition layer neuron node represents a logging response feature, the output layer is composed of 6 neuron nodes, each output node represents the lithofacies type of the cored section; The initial learning rate is set to 0.5, and the threshold learning rate is set to 0.001; This means that during the process of competitive learning, the learning rate will gradually decrease, and once the learning rate is reduced to 0.001 or less, the learning process will be terminated; Otherwise, the learning process will continue until the learning rate meets the threshold requirement, in addition, the maximum number of iterations is set to 12000 times;
[0086] S32, LDA cooperates with SSOM to build complex metamorphic rock buried hill lithology identification model:
[0087] The LDA cooperates with SSOM neural network to carry out the prediction model training process, and establishes the SSOM model for predicting lithofacies type, that is, a self-organizing mapping topology network containing 10*10 (100 neurons), each network unit is a rose diagram composed of normalized logging data variables, and the size of each part in the rose diagram is related to the intensity of the logging element response; Each neuron unit will be excited by a specific logging response combination feature, which means that 100 kinds of logging response combination features will be automatically generated in the competition layer neuron topology diagram;
[0088] S33, model generalization performance verification:
[0089] For the complex metamorphic rock buried hill lithology identification model based on SSOM algorithm, learning samples are used for back-judgment test, and test samples are used for prediction test, so as to comprehensively evaluate whether the performance of the constructed model meets the geological requirements; If the model shows excellent performance, the prediction model will be applied to the rapid identification of complex metamorphic rock buried hill lithology of other development wells in the region;
[0090] Randomly extracted 4516 groups of blind test samples are used to test the generalization ability of the prediction model, the results show that only 664 groups of samples are misjudged, and the prediction accuracy is as high as 85.30% (Table 2). This test result fully proves that the model has good generalization performance. Therefore, we can confirm that the combination of LDA and SSOM neural network can deeply mine the lithofacies information in logging elements, and through the competitive learning and mutual supervision mechanism between neurons, the optimal identification and prediction results are output. This shows that the method of using LDA cooperates with SSOM model to predict lithofacies type is feasible, and provides strong support for the accurate identification of buried hill top surface.
[0091] Table 2: LDA cooperates with SSOM model prediction accuracy statistics
[0092]
[0093] S4. Based on the sensitive logging element screening result of step S2 and the prediction result of the complex metamorphic rock buried hill lithology identification model of step S3, a buried hill mountain-entering carding chart plate of the classical reservoir structure combination mode established in step S1 is constructed;
[0094] The established LDA-SSOM prediction model is applied to the lithology identification of 10 exploratory wells in the region, and the carding chart plate corresponding to the typical mountain-entering reservoir structure mode is established, such as the mudstone-metabasalt-sandstone-entering metamorphic rock buried hill mode chart (BZ2 well; Figure 5 ), the mudstone-entering Paleozoic limestone buried hill mode chart (BZ3 well; Figure 6 ), and the like; the establishment of the buried hill carding chart plate is of great significance for the accurate identification of the buried hill top surface. Figure 7
[0095] S5. The buried hill mountain-entering carding chart plate constructed based on step S4 is used to guide the geophysical velocity correction and the prediction of the buried hill depth of the development well.
[0096] By using the buried hill reservoir structure mode and the identification method provided in the application, the average error of the buried hill interface of 11 development wells in the BZ oilfield is 11.4 m (the maximum is 24.8 m), the buried hill interfaces of the 11 development wells in the region are successfully carded (Table 3), under the premise of ensuring entering the buried hill, the mountain-entering footage is minimized, the safety of drilling construction is ensured, the drilling efficiency is improved, and the drilling cost is reduced; at the same time, the buried hill interface is accurately carded, the drilling coring data of the top of the buried hill is successfully obtained, the formation testing data is fully and accurately obtained, and the foundation for the comprehensive evaluation of the buried hill reservoir is laid.
[0097] Table 3: Prediction depth and actual drilling depth statistics of the buried hill interface of the BZ oilfield
[0098]
[0099] Note: Negative numbers represent actual drilling deepening, and positive numbers represent actual drilling shallowing
[0100] The application makes up for the shortcomings of the LDA algorithm, combines the theoretical advantages of the nonlinear machine learning method, establishes a metamorphic rock buried hill while-drilling carding method based on sensitive element screening and nonlinear machine learning algorithm, effectively guides the geophysical velocity correction and the depth prediction of the buried hill top surface in the process of buried hill while-drilling, provides a basis for quick decision-making for geologists in the process of while-drilling, and has important practical significance for the efficient development of buried hill reservoirs.
[0101] The applicant states that the above description is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and it should be understood by those skilled in the art that any changes or replacements within the technical scope disclosed by the present application can be easily thought out by those skilled in the art, and all of them fall within the protection scope and disclosure scope of the present application.
Claims
1. A method for rapid layer detection during drilling of metamorphic buried hills, characterized in that: Includes the following steps: S1. Analyze the structural combination patterns of the overlying Paleogene reservoirs and ancient buried hill reservoirs, and establish a typical structural combination pattern of the intrusive hill reservoir. S2. Based on the LDA algorithm, sensitive logging elements are screened and data dimensionality is reduced, and a sample dataset is constructed by core calibration logging; S3. Construct a lithology identification model for buried hills of complex metamorphic rocks based on the SSOM algorithm, and select test samples to verify the generalization performance of the model; S4. Based on the results of the sensitive logging element screening in step S2 and the prediction results of the complex metamorphic buried hill lithology identification model in step S3, construct the ancient buried hill entry layer map of the classic reservoir structure combination model established in step S1. S5. Based on the ancient buried hill entry layer map constructed in step S4, guide geophysical velocity correction and prediction of the entry depth of development wells into the buried hill.
2. The method for rapid layer detection while drilling metamorphic buried hills according to claim 1, characterized in that: The typical buried mountain reservoir structure combination modes include mudstone buried mountain into Archean metamorphic rocks, mudstone-conglomerate buried mountain into Archean metamorphic rocks, mudstone-basalt-conglomerate buried mountain into Archean metamorphic rocks, mudstone-Shahejie sandstone-conglomerate buried mountain into Archean metamorphic rocks, mudstone-volcanic breccia buried mountain into Archean metamorphic rocks, and mudstone buried mountain into Paleozoic limestone.
3. The method for rapid layer detection while drilling metamorphic buried hills according to claim 1, characterized in that: Step S2 specifically includes the following steps: S21. Sensitive logging element screening based on LDA algorithm: S22. Feature vector extraction based on LDA data dimensionality reduction processing: S23. Construct a sample dataset by core calibration logging.
4. The rapid layer-finding method for buried hills in metamorphic rocks according to claim 3, characterized in that: The feature vectors extracted in step S22 are the four feature vectors with the largest and smallest inter-group variances when sorted from largest to smallest in the LDA algorithm.
5. The rapid layer-finding method for buried hills in metamorphic rocks according to claim 3, characterized in that: In step S23, 70% of the sample dataset is used as training samples, and the remaining 30% is used as validation samples.
6. The method for rapid layer detection while drilling metamorphic buried hills according to claim 1, characterized in that: The specific steps of step S3 include the following: S31. Set the mesh structure of the input layer, competition layer, and output layer, and initialize the neural network parameters; S2. Based on the established learning sample set, the SSOM algorithm is used to train the prediction model, and a complex lithology identification model based on the SSOM algorithm is developed. S33. For the lithology identification model of complex metamorphic buried hills based on the SSOM algorithm, the learning samples are used for back-judgment verification, and the test samples are used for prediction verification to evaluate whether the performance of the constructed model meets the geological requirements.
7. The application of the rapid layer-finding method for metamorphic buried hills as described in any one of claims 1 to 6 in the description of oil and gas reservoirs.