A Method for Logging Data Correction and Precise Lithology Identification Based on Artificial Neural Networks

Abstract

This invention discloses a method for accurate lithology identification and correction of logging data based on artificial neural networks, comprising the following steps: collecting logging data, laboratory test data, and regional geological background data of the target area from buried hills; preprocessing the collected logging data and laboratory test data to establish a correspondence between the logging data and laboratory test data, forming a dataset; constructing an artificial neural network model; training and optimizing the constructed artificial neural network model based on the dataset; inputting the logging data of the target well to be corrected into the trained artificial neural network model, outputting corrected mineral element content data; and, based on the corrected mineral element content data and combined with the regional geological background, achieving lithology classification and accurate identification of the target well. This invention uses the above method to convert logging data into laboratory-precision data, improving the accuracy of lithology identification under complex geological conditions such as Archean metamorphic buried hills.

Description

Technical Field

[0001] This invention relates to the field of oil and gas exploration logging technology, and in particular to a method for logging data correction and accurate lithology identification based on artificial neural networks. Background Technology

[0002] Well logging data is downhole data collected in real time by logging equipment during the drilling process. It includes engineering parameters (drilling time, drilling speed, depth, etc.), gas parameters (all hydrocarbons, non-hydrocarbon gases, etc.), drilling fluid parameters (flow rate, density, resistivity, etc.), and geological logging data (cuttings logging, core logging, fluorescence logging, etc.). It has the advantages of quick acquisition and low cost, and can provide real-time technical support for drilling operations and mineral exploration.

[0003] In the exploration of ultra-deep buried hills for oil and gas in the Bohai Bay Basin and other areas, logging data is often used for lithological assessment. However, existing logging techniques have significant limitations: the X-ray fluorescence spectroscopy (XRF) and X-ray diffraction (XRD) methods commonly used in field logging are affected by factors such as equipment power (logging equipment is mostly miniaturized and has lower power than laboratory equipment) and sample condition (uneven rock cuttings size, differences in surface smoothness, and fluctuations in water content), resulting in insufficient data accuracy. Specifically, this manifests as a systematic error in mineral element detection, such as "high potassium and low sodium," leading to inaccurate lithological assessment.

[0004] While laboratory analysis data offers high precision, the process is cumbersome (requiring sample transportation, pretreatment, and precision instrument testing) and costly, making it difficult to apply on a large scale in field logging. Furthermore, well logging data is challenging to obtain under complex geological conditions such as ultra-deep buried hills (3800-5000m deep, with most areas exceeding 4000m), making it unsuitable as a routine basis for lithological assessment.

[0005] Therefore, there is an urgent need for a technical method that can effectively correct logging data errors and improve the accuracy of lithology identification. Summary of the Invention

[0006] The purpose of this invention is to provide a method for correcting logging data and accurately identifying lithology based on artificial neural networks, so as to realize the conversion of logging data into laboratory-precise data and improve the accuracy of lithology identification under complex geological conditions such as Archean metamorphic buried hills.

[0007] To achieve the above objectives, this invention provides a method for logging data correction and accurate lithology identification based on artificial neural networks, comprising the following steps:

[0008] S1. Collect logging data, laboratory test data, and regional geological background data of the target area from buried hill drilling.

[0009] S2. Preprocess the logging data and laboratory test data collected in S1, establish a one-to-one correspondence between the logging data and the laboratory test data, and form a dataset.

[0010] S3. Construct an artificial neural network model, wherein the artificial neural network includes an input layer, a hidden layer, and an output layer;

[0011] S4. The artificial neural network model constructed in S3 is trained and optimized based on the dataset formed in S2;

[0012] S5. Input the target well logging data to be corrected into the artificial neural network model trained in S4 to obtain the corrected mineral element content data.

[0013] S6. Based on the corrected mineral element content data and combined with the regional geological background, the lithology classification and precise identification of the target well are completed through preliminary judgment and fine differentiation.

[0014] Preferably, S1 is as follows:

[0015] S11. Using on-site X-ray fluorescence spectroscopy and X-ray diffraction equipment, rock cuttings or core samples from buried hill drilling in the target area are tested to obtain logging data. The logging data is mineral element content data, which includes, but is not limited to, the contents of quartz, plagioclase, potassium feldspar, pyroxene, and amphibole.

[0016] S12. For rock cuttings or core samples from the same well in S11, standard X-ray fluorescence spectroscopy and X-ray diffraction methods are used to detect them and obtain laboratory test data.

[0017] S13. Collect regional geological background data of buried hill drilling in the target area, including but not limited to the distribution pattern of strata lithology and the characteristics of metamorphism.

[0018] Preferably, S2 is as follows:

[0019] S21. Using the box plot method, outliers in the logging data obtained in S1 caused by instrument malfunction, sample contamination, or abnormal rock cutting particle size are removed.

[0020] S22. Unify the data format of the logging data processed in S21 and the laboratory test data obtained in S1;

[0021] S23. Unify the indicator dimensions of the logging data processed in S22 with the laboratory test data.

[0022] Preferably, S3 is as follows:

[0023] S31. The input layer receives the mineral element content parameters from the preprocessed logging data, and obtains the feature vector of the input layer by linear combination of weight calculations.

[0024] S32. Input the feature vector output from S31 into multiple hidden layers in sequence, and use activation functions to extract and transform features to obtain the final output feature vector of the hidden layers.

[0025] S33. The output layer receives the final output features of the hidden layer obtained in S32, processes the final output features of the hidden layer, and outputs the predicted values ​​corresponding to the laboratory test data.

[0026] Preferably, the hidden layer uses a non-linear activation function and has 10 to 20 neurons to ensure that the artificial neural network model learns the complex mapping relationship between logging data and laboratory test data.

[0027] Preferably, S4 is as follows:

[0028] S41. Divide the dataset constructed in S2 into a training set, a validation set, and a test set in an 8:1:1 ratio;

[0029] S42. Train the artificial neural network model based on the training set;

[0030] S43. Adjust the hyperparameters of the artificial neural network model using the validation set, evaluate the generalization ability of the artificial neural network model using the test set, and ensure the prediction accuracy of the artificial neural network model on new data.

[0031] Preferably, S42 is as follows:

[0032] S421. Input the training set into the artificial neural network model constructed in S3, and calculate the predicted value of the laboratory test data through forward propagation;

[0033] S422. Compare the predicted value with the actual value of the laboratory test data, and calculate the prediction error using the mean square error.

[0034] S423. The weight parameters of the artificial neural network model are iteratively adjusted through the backpropagation algorithm to minimize the prediction error; the number of iterations and the learning rate are dynamically set during the training process until the artificial neural network model converges.

[0035] Preferably, S6 is as follows:

[0036] S61. Based on mineral element content data, and referring to the quartz-plagioclase-potassium feldspar three-phase diagram, complete the preliminary lithological determination according to the preset lithological classification standard; among which, the lithological classification standard is: tonalite and granodiorite correspond to TTG gneiss, and monzogranite corresponds to non-TTG lithology.

[0037] S62. Analyze the content of dark minerals such as pyroxene, amphibole, and biotite to distinguish metamorphic rocks from intrusive rocks and achieve accurate lithological identification.

[0038] Therefore, the present invention employs the above-mentioned method for logging data correction and accurate lithology identification based on artificial neural networks, and the beneficial effects are as follows:

[0039] (1) This invention establishes a mapping relationship between logging data and laboratory test data through an artificial neural network model, effectively correcting the systematic error of "potassium-rich and sodium-poor" logging data. The accuracy of the corrected data is significantly improved, approaching the level of laboratory test (e.g., the fit R between the corrected mineral data and laboratory data of well BZ19-6-M11 is over 0.66).

[0040] (2) This invention solves the problem of vague lithology judgment and broad classification of Archean metamorphic buried hills in traditional logging data. It can accurately refine the broad category of "granitic gneiss" into specific types such as TTG gneiss (tonnite, granodiorite), and improve the accuracy of distinguishing between intrusive rocks and metamorphic rocks (e.g., the misjudgment rate of intrusive rocks in well BZ19-6-C12 is reduced by more than 70%).

[0041] (3) Based on field logging data, this invention achieves the goal of eliminating the need for additional sampling and testing costs, is convenient and efficient, and is suitable for scenarios such as ultra-deep buried hills where logging and large-scale laboratory testing are difficult to perform.

[0042] (4) This method can be widely applied to oil and gas exploration in Archean metamorphic buried hills around the North China Craton, such as the Bohai Bay Basin, providing reliable lithological data support for drilling decisions (such as drilling fluid performance adjustment) and reservoir evaluation (such as TTG gneiss reservoir pore distribution prediction), thereby improving the success rate of exploration and development.

[0043] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0044] Figure 1 This is a flowchart of the method for accurate lithology identification and logging data correction based on artificial neural networks according to the present invention;

[0045] Figure 2 This is a forward propagation diagram of a neural network containing two hidden layers in an embodiment of the present invention;

[0046] Figure 3 The three-phase diagrams of the measured mineral data and the mineral data obtained by the logging method of well BZ19-6-M11 in this embodiment of the invention are shown below; (a) is the three-phase diagram of the measured mineral data of well BZ19-6-M11; (b) is the three-phase diagram of the logging mineral data of well BZ19-6-M11.

[0047] Figure 4 These are three-phase diagrams of measured mineral data and mineral data obtained by logging methods in the Matouya area according to embodiments of the present invention; wherein, (a) is a three-phase diagram of measured mineral data in the Matouya area; and (b) is a three-phase diagram of mineral data obtained by logging methods in the Matouya area.

[0048] Figure 5 This is the regression graph of the artificial neural network model training in this embodiment of the invention;

[0049] Figure 6 This is a three-phase diagram of mineral data obtained from model prediction of well BZ19-6-M11 in this embodiment of the invention;

[0050] Figure 7 These are three-phase diagrams of logging mineral data and predicted mineral data of well BZ19-6-C12 in this embodiment of the invention; wherein, (a) is a three-phase diagram of logging mineral data of well BZ19-6-C12; and (b) is a three-phase diagram of predicted mineral data of well BZ19-6-C12.

[0051] Figure 8 These are three-phase diagrams of logging mineral data and predicted mineral data of well BZ19-6-M16 in this embodiment of the invention; wherein, (a) is a three-phase diagram of logging mineral data of well BZ19-6-M16; and (b) is a three-phase diagram of predicted mineral data of well BZ19-6-M16.

[0052] Figure 9 These are well maps comparing the logging lithology and corrected lithology of wells BZ19-6-C12 and BZ19-6-M16 in this embodiment of the invention; wherein, (a) is a well map comparing the logging lithology and corrected lithology of well BZ19-6-C12; and (b) is a well map comparing the logging lithology and corrected lithology of well BZ19-6-M16. Detailed Implementation

[0053] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0054] like Figure 1 As shown, the present invention provides a method for logging data correction and accurate lithology identification based on artificial neural networks, comprising the following steps:

[0055] S1. Collect logging data, laboratory test data, and regional geological background data of the target area from buried hill drilling.

[0056] S2. Preprocess the logging data and laboratory test data collected in S1, establish a one-to-one correspondence between the logging data and the laboratory test data, and form a dataset.

[0057] S3. Construct an artificial neural network model, wherein the artificial neural network includes an input layer, a hidden layer, and an output layer;

[0058] S4. The artificial neural network model constructed in S3 is trained and optimized based on the dataset formed in S2;

[0059] S5. Input the target well logging data to be corrected into the artificial neural network model trained in S4 to obtain the corrected mineral element content data.

[0060] S6. Based on the corrected mineral element content data and combined with the regional geological background, the lithology classification and precise identification of the target well are completed through preliminary judgment and fine differentiation.

[0061] Example

[0062] This embodiment focuses on the Archean metamorphic buried hill of Bozhong 19-6, located in the Bozhong Depression of the Bohai Bay Basin. The buried hill has a depth of 3800-5000m and is classified as an ultra-deep buried hill. The main lithology in the area is TTG gneiss (accounting for more than 85%) and intrusive bodies such as granite porphyry, diorite porphyry, and diabase.

[0063] A method for logging data correction and accurate lithology identification based on artificial neural networks includes the following steps:

[0064] S1. Collect logging data, laboratory test data, and regional geological background data of the target area from buried hill drilling.

[0065] S11. Using on-site X-ray fluorescence spectroscopy (XRF) and X-ray diffraction (XRD) equipment, rock cuttings or core samples from buried hill drilling in the target area are tested to obtain logging data. The logging data is mineral element content data, which includes the contents of quartz, plagioclase, potassium feldspar, pyroxene, amphibole, etc.

[0066] S12. For cuttings or core samples from the same well in S11, standard XRF and XRD methods are used for testing to obtain laboratory test data, ensuring that the sample sources correspond one-to-one.

[0067] S13. Collect regional geological background data of buried hill drilling in the target area, including the distribution pattern of strata and lithology (e.g., the Archean metamorphic rocks in the North China Craton are mainly TTG gneiss), metamorphic characteristics, etc.

[0068] This embodiment collects logging data from 57 development wells and 17 exploration wells in the Archean metamorphic buried hill area of ​​Bozhong 19-6, including the contents of quartz, plagioclase, potassium feldspar, pyroxene, and amphibole measured by on-site XRD. At the same time, rock cuttings samples were collected from the corresponding wells (one sample every 10m, for a total of more than 1,000 samples), and the mineral content data were obtained by laboratory standard XRD method (instrument model: Bruker D8 Advance).

[0069] Taking well BZ19-6-M11 as an example, the comparison results between logging data and laboratory test data are as follows: Figure 3 As shown, the logging data exhibits a pattern of "more potassium feldspar and less plagioclase," leading to misjudgment of lithology.

[0070] To verify the generalizability of the error, 10 TTG gneiss samples collected in the field at Matouya (a typical TTG gneiss outcrop area of ​​the North China Craton) were simultaneously tested. The results are as follows: Figure 4 As shown, the logging XRD data also exhibits the characteristic of "high potassium and low sodium", misclassifying the tonalite metamorphic rock as granodiorite metamorphic rock, proving that this error is a systematic error of the logging method.

[0071] S2. Preprocess the logging data and laboratory test data collected in S1, establish a one-to-one correspondence between the logging data and the laboratory test data, and form a dataset.

[0072] S21. Using the box plot method, outliers in the logging data obtained in S1 caused by instrument malfunction, sample contamination, or abnormal rock cutting particle size are removed.

[0073] S22. Unify the data format of the logging data processed in S21 and the laboratory test data obtained in S1 to ensure that the mineral content of both types of data is expressed as a mass percentage.

[0074] S23. Unify the index dimensions of the logging data processed in S22 with the laboratory test data to ensure that the mineral types detected by the two types of data after processing in S22 are consistent.

[0075] In this embodiment, the box plot method was used to remove outliers in the logging data caused by uneven rock cuttings size (particle size > 2 mm or < 0.075 mm) and abnormal water content (water content > 5%), resulting in the removal of 86 sets of abnormal data. The percentage index format of mineral content was standardized to ensure that both logging data and laboratory test data were expressed as "mass percentage", and finally 100 sets of valid corresponding data pairs were established (each set contains 3 core indicators: quartz, plagioclase, and potassium feldspar).

[0076] S3. Based on the TensorFlow framework, construct an artificial neural network model. The artificial neural network includes an input layer, hidden layers, and an output layer. The model structure is as follows: Figure 2 As shown.

[0077] S31. The input layer has three neurons, corresponding to the contents of quartz, plagioclase, and potassium feldspar, respectively. After receiving the mineral element content parameters from the preprocessed logging data, this layer performs a linear combination of weights to obtain the feature vector of the input layer, as shown below:

[0078] ;

[0079] in, These are the feature vectors of the input layer; These are the weights of the input layer; These are the mineral element content parameters in the pre-processed logging data.

[0080] S32. The feature vector of the input layer output from S31 is sequentially input into multiple hidden layers. Feature extraction and transformation are achieved through activation functions to generate the final output feature vector of the hidden layers. Each hidden layer is set with 10 to 20 neurons to ensure that the artificial neural network model learns the complex mapping relationship between logging data and laboratory test data.

[0081] In this embodiment, two hidden layers are set up. The first hidden layer has 15 neurons and the second hidden layer has 12 neurons. The activation function is ReLU function, the loss function is mean squared error (MSE), and the optimizer is Adam optimizer.

[0082] First, the feature vector obtained from S31 is fed into the first hidden layer (H1). After linear combination processing by the activation function, the output feature vector of the first hidden layer is obtained. ,in, It is the output feature of the first hidden layer; It uses the ReLU activation function; after weighted linear combination, the input feature vector of the second transport layer (H2) is obtained. ,in, It is the input feature vector of the second transport layer; It is the weight from the first hidden layer to the second hidden layer.

[0083] Secondly, the second hidden layer processes the input feature vector using an activation function to obtain the output feature vector. 、 After further weighting and linear combination, the final output feature vector of the hidden layer is obtained. ,in, These are the weights from the second hidden layer to the final output of the hidden layer; It is the final output feature vector of the hidden layer.

[0084] S33. The output layer has three neurons, each corresponding to the content of one of the three minerals tested in the laboratory. This layer receives the final output features from the hidden layer obtained in S32, processes these features, and outputs the predicted values ​​corresponding to the laboratory test data. .

[0085] S4. The artificial neural network model constructed in S3 is trained and optimized based on the dataset in S2.

[0086] S41. Divide the 100 datasets constructed in S2 into training set, validation set and test set in a ratio of 8:1:1.

[0087] S42. Train the artificial neural network model based on the training set.

[0088] S421. Input the training set into the artificial neural network model constructed in S3, and calculate the predicted value of the laboratory test data through forward propagation.

[0089] S422. Compare the predicted value with the actual value of the laboratory test data, and calculate the prediction error using the mean squared error (MSE).

[0090] S423. The model weight parameters are iteratively adjusted using the backpropagation algorithm (gradient descent method) to minimize the prediction error. The number of iterations during training is 800 to 1200, and the learning rate is 0.0005 to 0.002. In this embodiment, it is set to 1000 iterations and the learning rate is set to 0.001 until the model converges.

[0091] S43. Use the validation set to adjust the model hyperparameters (such as the number of neurons in the hidden layer), and use the test set to evaluate the model's generalization ability to ensure the model's prediction accuracy on new data.

[0092] In this embodiment, the model training regression results are as follows: Figure 5 As shown, the training set fit R=0.66319, the validation set R=-0.0013581 (the validation set fluctuated due to the small amount of data), the test set R=-0.59892, and the overall data fit R=0.38511. The model converged overall, and the training set showed the best fit. The logging data from well BZ19-6-M11 was input into the trained model, and the predicted mineral data three-phase diagram is shown below. Figure 6 As shown, the data reverted to the tonalite and granodiorite regions, and are consistent with the lithological results of laboratory testing, demonstrating a significant correction effect.

[0093] S5. Input the target well logging data to be corrected into the trained artificial neural network model, and output the corrected mineral element content data. This data can approximate the accuracy of laboratory test data and eliminate the systematic error of "high potassium and low sodium".

[0094] S6. Based on the corrected mineral element content data and combined with the regional geological background, the lithology classification and precise identification of the target well are completed through preliminary judgment and fine differentiation.

[0095] S61. Based on mineral element content data, and referring to the quartz-plagioclase-potassium feldspar three-phase diagram, complete the preliminary lithological determination according to the preset lithological classification standard; among which, the lithological classification standard is: tonalite and granodiorite correspond to TTG gneiss, and monzogranite corresponds to non-TTG lithology.

[0096] S62. Analyze the content of dark minerals such as pyroxene, amphibole, and biotite to distinguish metamorphic rocks (TTG gneiss) from intrusive rocks (granite porphyry, diorite porphyry, and diabase) and achieve accurate lithological identification.

[0097] Well BZ19-6-C12: (as follows) Figure 7 As shown in (a), the three-phase diagram of the well logging XRD data shows that the lithology is concentrated in the monzogranite area, which is inconsistent with the geological background of the Bohai Central region, which is "mainly composed of TTG gneiss". Inputting this well logging data into the trained model yields the corrected prediction data, and its three-phase diagram is shown below. Figure 7 As shown in (b), the mineral data is concentrated in the granodiorite region and is preliminarily identified as TTG gneiss based on the classification criteria. Only 12% of the data falls in the non-TTG region.

[0098] Analysis of data from a few non-TTG areas, combined with the analysis of pyroxene content (>5%) and amphibole content (>3%), identified them as granite (6%), diorite porphyry (4%), and diabase (2%).

[0099] Well BZ19-6-M16: (as follows) Figure 8 As shown in (a), the three-phase diagram of the well logging data also leans towards the monzogranite region. After model correction, as shown... Figure 8 As shown in (b), the predicted data are all concentrated in the tonalite and granodiorite areas. According to the classification criteria, they are identified as pure TTG gneiss with no intrusive rocks, which is in complete agreement with the regional geological background. The lithology judgment is accurate after correction.

[0100] The lithology correction results of the two wells were further verified using well maps, such as... Figure 9 As shown, where, Figure 9 In Figure (a), the "Looping Lithology" column is marked with a large number of "monzogranite", while the "Corrected Lithology" column is mainly "TTG metamorphic rock", with a small amount of "granite", "diorite porphyry" and "diabase". Figure 9 In Figure (b), the "Corrected Lithology" column is entirely labeled "TTG Metamorphic Rock," with no other lithological annotations. Figure 9 It can be seen that the corrected lithology is completely consistent with the main lithology of the region (TTG gneiss), and the accurate classification of metamorphic rocks has been achieved, solving the problem of "broad lithology classification" in traditional logging.

[0101] The lithological identification results of wells BZ19-6-C12 and BZ19-6-M16 using this method show that:

[0102] 1. The corrected lithological judgment is completely consistent with the regional geological background (the Bozhong 19-6 buried hill is mainly composed of TTG gneiss), and the lithology such as "monzogranite" misjudged by traditional logging data is accurately corrected to "TTG gneiss".

[0103] 2. The accuracy of metamorphic rock classification has been improved by more than 80% (from "granitic gneiss" to "TTG gneiss"), and the accuracy of intrusive rock identification has been improved by more than 70% (the misjudgment rate of intrusive rocks in well BZ19-6-C12 has been reduced from 35% to 8%).

[0104] 3. It provided reliable lithological data support for subsequent drilling operations in the area. For example, well BZ19-6-M16 adjusted the drilling fluid density based on the corrected lithological data, which reduced the risk of wellbore collapse and improved drilling efficiency by 15%.

[0105] Therefore, the present invention employs the above-mentioned method for logging data correction and accurate lithology identification based on artificial neural networks, and the beneficial effects are as follows:

[0106] (1) This invention establishes a mapping relationship between logging data and laboratory test data through an artificial neural network model, effectively correcting the systematic error of "potassium-rich and sodium-poor" logging data. The accuracy of the corrected data is significantly improved, approaching the level of laboratory test (e.g., the fit R between the corrected mineral data and laboratory data of well BZ19-6-M11 is over 0.66).

[0107] (2) This invention solves the problem of vague lithology judgment and broad classification of Archean metamorphic buried hills in traditional logging data. It can accurately refine the broad category of "granitic gneiss" into specific types such as TTG gneiss (tonnite, granodiorite), and improve the accuracy of distinguishing between intrusive rocks and metamorphic rocks (e.g., the misjudgment rate of intrusive rocks in well BZ19-6-C12 is reduced by more than 70%).

[0108] (3) This invention is based on field logging data, without the need for additional sampling and testing costs. It is convenient to operate and highly efficient, and is suitable for scenarios such as ultra-deep buried hills where logging and large-scale laboratory testing are difficult.

[0109] (4) This method can be widely applied to oil and gas exploration in Archean metamorphic buried hills around the North China Craton, such as the Bohai Bay Basin, providing reliable lithological data support for drilling decisions (such as drilling fluid performance adjustment) and reservoir evaluation (such as TTG gneiss reservoir pore distribution prediction), thereby improving the success rate of exploration and development.

[0110] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for logging data correction and accurate lithology identification based on artificial neural networks, characterized in that, Includes the following steps: S1. Collect logging data, laboratory test data, and regional geological background data of the target area from buried hill drilling. S2. Preprocess the logging data and laboratory test data collected in S1, establish a one-to-one correspondence between the logging data and the laboratory test data, and form a dataset; S3. Construct an artificial neural network model, which includes an input layer, a hidden layer, and an output layer; S4. Train and optimize the artificial neural network model constructed in S3 based on the dataset formed in S2; S5. Input the logging data of the target well to be corrected into the artificial neural network model trained in S4 to obtain the corrected mineral element content data; S6. Based on the corrected mineral element content data and combined with the regional geological background, complete the lithological classification and accurate identification of the target well through preliminary judgment and fine differentiation; S1 specifically includes: S11, using on-site X-ray fluorescence spectroscopy and X-ray diffraction equipment to test cuttings or core samples from buried hill drilling in the target area and obtain logging data, wherein the logging data is mineral element content data, including the content of quartz, plagioclase, potassium feldspar, pyroxene, and amphibole; S12, using standard X-ray fluorescence spectroscopy and X-ray diffraction methods to test cuttings or core samples from the same drilling in S11 and obtain laboratory test data; S13, collecting regional geological background data of buried hill drilling in the target area, including the distribution pattern of stratigraphic lithology and metamorphic characteristics; S3 specifically consists of: S31, the input layer receives mineral element content data from the preprocessed logging data, and obtains the feature vector of the input layer through linear combination of weights; S32, the feature vector of the input layer output from S31 is sequentially input into multiple hidden layers, and feature extraction and transformation are achieved through activation functions to obtain the final output feature vector of the hidden layer; S33, the output layer receives the final output feature of the hidden layer obtained from S32, processes the final output feature of the hidden layer, and outputs the predicted value corresponding to the laboratory test data; S6 specifically includes: S61. Based on mineral element content data, referring to the quartz-plagioclase-potassium feldspar three-phase diagram, and according to the preset lithology classification standard, complete the preliminary lithology determination; among which, the lithology classification standard is: tonalite and granodiorite correspond to TTG gneiss, and monzogranite corresponds to non-TTG lithology; S62. Analyze the content of dark minerals such as pyroxene, amphibole, and biotite to distinguish metamorphic rocks from intrusive rocks, and achieve accurate lithology identification.

2. The method for logging data correction and accurate lithology identification based on artificial neural networks according to claim 1, characterized in that, S2 specifically includes: S21, using the box plot method to remove outliers in the logging data obtained in S1 caused by instrument malfunction, sample contamination, or abnormal cuttings particle size; S22, unifying the data format of the logging data processed in S21 and the laboratory test data obtained in S1; and S23, unifying the indicator dimensions of the logging data processed in S22 and the laboratory test data.

3. The method for logging data correction and accurate lithology identification based on artificial neural networks according to claim 1, characterized in that, The hidden layer uses a non-linear activation function and has 10 to 20 neurons to ensure that the artificial neural network model learns the complex mapping relationship between logging data and laboratory test data.

4. The method for logging data correction and accurate lithology identification based on artificial neural networks according to claim 3, characterized in that, S4 specifically consists of: S41, dividing the dataset constructed in S2 into a training set, a validation set, and a test set in an 8:1:1 ratio; S42, training the artificial neural network model based on the training set; and S43, adjusting the hyperparameters of the artificial neural network model using the validation set and evaluating the generalization ability of the artificial neural network model through the test set to ensure the prediction accuracy of the artificial neural network model on new data.

5. The method for logging data correction and accurate lithology identification based on artificial neural networks according to claim 4, characterized in that, S42 specifically involves: S421, inputting the training set into the artificial neural network model constructed in S3, and calculating the predicted value of the laboratory test data through forward propagation; S422, comparing the predicted value with the true value of the laboratory test data, and calculating the prediction error using mean squared error; S423, iteratively adjusting the weight parameters of the artificial neural network model through the backpropagation algorithm to minimize the prediction error; dynamically setting the number of iterations and the learning rate during training until the artificial neural network model converges.

Images