A lithology identification method and system applied to oil drilling

Abstract

The application discloses a lithology identification method and system applied to oil drilling, wherein the method steps comprise the following: collecting geophysical data in a work site before the drilling process; preprocessing the geophysical data to obtain processed data; performing feature extraction on the processed data to obtain a training set and a test set; constructing a lithology classification model by using the training set and the test set; completing lithology prediction before the oil drilling process by using the lithology classification model to obtain a prediction result; drilling based on the prediction result; collecting rock size data and rock debris data during the drilling process; and completing lithology identification based on the rock size data and the rock debris data. The application adopts multiple geophysical data, comprehensively applies data processing and artificial intelligence algorithms for analysis and processing, can more accurately distinguish and distinguish geological lithology, effectively improves the efficiency and accuracy of lithology identification, and has wide application prospect and economic value.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of geophysical exploration, in particular to a lithology identification method and system applied to oil drilling. BACKGROUND

[0002] In the process of oil drilling, the prediction and identification of lithology is a very important link, because lithology determines the conditions of source rock, reservoir, cap rock, etc., which is directly related to whether the drilling result can reach industrial oil and gas flow for oil and gas exploration. At the same time, due to the particularity of drilling work, there is no physical data available for research before drilling, and geophysical data is the only data that can be provided for lithology prediction, usually including seismic data, gravity and magnetic data and electromagnetic data collected. There is also great uncertainty in the identification of lithology after drilling, because drilling only takes a small amount of rock debris and rock debris and other physical data, and most of the data is indirect logging data. Both need to be combined to effectively identify the lithology of the whole well section and provide a basis for drilling reservoir and oil and gas bearing property evaluation. The above-mentioned pre-drilling prediction and post-drilling identification are difficult to integrate the above-mentioned research work organically due to the involvement of many disciplines and great work difficulty. Therefore, a prediction and identification method system that comprehensively utilizes various geophysical data and geological data is needed to improve the accuracy of lithology identification. SUMMARY

[0003] To solve the technical problems in the above background, the present application preliminarily predicts the lithology by applying geophysical data in the early stage; at the same time, rock size and rock debris data are collected during drilling, and the preliminary prediction results are combined to improve the accuracy of lithology identification.

[0004] To achieve the above-mentioned purpose, the present application provides a lithology identification method applied to oil drilling, characterized in that the steps include:

[0005] Before the drilling process, collecting geophysical data in the work site;

[0006] Pretreating the geophysical data to obtain processed data;

[0007] Feature extraction is performed on the processed data to obtain a training set and a test set; and a lithology classification model is constructed using the training set and the test set;

[0008] The lithology classification model is used to complete the lithology prediction before the oil drilling process to obtain a prediction result;

[0009] Based on the prediction result, drilling is carried out; and during the drilling process, rock size data and rock debris data are collected; and based on the rock size data and the rock debris data, lithology identification is completed.

[0010] Preferably, the geophysical data comprises seismic data, gravity and magnetic survey data, electromagnetic data and borehole data; wherein the seismic data is collected by using a non-blocking seismic source and a geophone; and the electromagnetic data is collected by using an array of sound sources and receivers.

[0011] Preferably, the method of feature extraction comprises: performing feature selection on the processed data to obtain candidate features; then performing feature extraction on the candidate features to complete the feature extraction; and finally dividing the extracted features into the training set and the test set.

[0012] Preferably, the method of constructing the lithology classification model comprises: based on the extracted features, using a convolutional neural network and a recurrent neural network to construct the lithology classification model.

[0013] Preferably, the lithology identification is completed by using a decision tree algorithm.

[0014] The application also provides a lithology identification system applied to oil drilling, comprising: an acquisition module, a preprocessing module, an extraction module, a construction module and an identification module.

[0015] The acquisition module is configured to acquire geophysical data in a work site before the drilling process.

[0016] The preprocessing module is configured to preprocess the geophysical data to obtain processed data.

[0017] The extraction module is configured to perform feature extraction on the processed data to obtain a training set and a test set; and use the training set and the test set to construct a lithology classification model.

[0018] The construction module is configured to use the lithology classification model to complete lithology prediction before the oil drilling process to obtain a prediction result.

[0019] The identification module is configured to perform drilling based on the prediction result; and acquire rock size data and rock debris data during the drilling process; and complete lithology identification based on the rock size data and the rock debris data.

[0020] Preferably, the geophysical data comprises seismic data, gravity and magnetic survey data, electromagnetic data and borehole data; wherein the working process of the acquisition module comprises: collecting the seismic data by using a non-blocking seismic source and a geophone; and collecting the electromagnetic data by using an array of sound sources and receivers.

[0021] Preferably, the workflow of the extracting module comprises: performing feature selection on the processed data to obtain candidate features; then performing feature extraction on the candidate features to complete the feature extraction; and finally segmenting the extracted features into the training set and the test set.

[0022] Preferably, the workflow of the constructing module comprises: based on the extracted features, constructing the lithology classification model by using a convolutional neural network and a recurrent neural network.

[0023] Preferably, the workflow of the identifying module comprises: performing the lithology identification by using a decision tree algorithm.

[0024] Compared with the prior art, the application has the following beneficial effects:

[0025] The application adopts various geophysical data, and comprehensively applies data processing and artificial intelligence algorithms for analysis and processing, which can more accurately distinguish and distinguish geological lithology, effectively improves the efficiency and accuracy of lithology identification, and has wide application prospect and economic value. BRIEF DESCRIPTION OF DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of the application, the drawings used in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0027] Figure 1 The figure is a method flowchart of the embodiment of the application.

[0028] Figure 2 The figure is a system structure diagram of the embodiment of the application. DETAILED DESCRIPTION

[0029] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only some of the embodiments of the application, not all. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0030] In order to make the above-mentioned purposes, features and advantages of the application more obvious and easy to understand, the application will be further described in detail below with reference to the drawings and specific embodiments.

[0031] Embodiment one

[0032] As shown in the figure, the figure is a method flowchart of the embodiment, and the steps include: Figure 1 ​

[0033] S1. Collect geophysical data at the site before drilling.

[0034] First, collect geophysical data from the construction site, including seismic data, gravity and magnetic measurement data, electromagnetic data, and borehole data.

[0035] In this implementation, precise site surveying and exploration data planning are required before seismic data acquisition. Seismic data is acquired using an unobstructed ground source and geophone. Electromagnetic data is acquired using an array of sound sources and receivers. Gravity and magnetic measurement data, as well as borehole data, can be acquired using existing instruments.

[0036] S2. Preprocess the geophysical data to obtain the processed data.

[0037] The collected geophysical data is then preprocessed to obtain processed data. The steps include noise removal, amplitude balancing, and time correction.

[0038] High-pass filtering is used to remove low-frequency noise or DC components, filtering out the small-amplitude, slowly changing parts of the real signal. Low-pass filtering is used to remove high-frequency noise, i.e., high-amplitude, short-duration waveforms, while retaining the low-frequency components of the signal. Since there are differences in data amplitude between different sensors, amplitude balancing is necessary to enable effective comparison of data from different sensors. A normalization-based balancing method is used to balance the amplitude. However, due to the influence of equipment and media, the time of data reception may have some offset in the acquired geophysical data. To accurately obtain data characteristics, a bilateral correction method is used for time correction of the data.

[0039] S3. Extract features from the processed data to obtain training and test sets; and use the training and test sets to construct a lithology classification model.

[0040] First, feature selection is performed on the processed data to obtain candidate features. The chi-square test is then used to determine whether the feature significantly distinguishes different categories.

[0041] First, we need to calculate the expected and actual values ​​of each feature under different categories. Specifically, for a dataset with N features and M categories, the calculation steps are as follows:

[0042] Calculate the number of samples K in each category. i And the number of samples K for each feature value i,j ; Calculate the number c of each feature value in the entire dataset. j ; Calculate the number of samples M for each category in the entire dataset.i Calculate the expected number of each feature value in each category

[0043] E i,j = (K i × c j ) / M

[0044] Calculate the actual number of each feature value in each category O i,j ; Calculate the chi-square value χ2 according to the expected value and the actual value χ2 = Σ (O i,j - E i,j )2 / E i,j

[0045] Then, the critical value of the chi-square value needs to be set according to the significance level. In general, the significance level is 0.05, that is, the p-value is 0.05. Calculate the critical value of the chi-square value according to the significance level and the degree of freedom.

[0046] Finally, for each feature, calculate its chi-square value, if the chi-square value is less than the critical value, the feature has no significant impact on the classification and can be deleted; otherwise, it means that the feature has a huge contribution to the classification and can be retained.

[0047] After that, feature extraction is performed on the alternative features;

[0048] Use principal component analysis (PCA) to reduce high-dimensional data to low-dimensional data, so that the data can be more easily analyzed and displayed. For lithology identification, PCA can be used for feature extraction, and the basic steps are as follows:

[0049] Before PCA, the original data needs to be standardized, that is, the original data is standardized to have a mean of 0 and a variance of 1, to reduce the influence between different data models. After that, determine the number of principal components; determine the number of principal components to be retained by the cumulative contribution rate. The contribution rate of the principal component refers to the percentage of the original data information that each principal component can express. Generally speaking, a cumulative contribution rate of more than 80% can be considered to be a good principal component. Calculate the covariance matrix; calculate the covariance matrix of the standardized data to obtain the eigenvalues and eigenvectors of the covariance matrix. Finally, calculate the coefficients of each principal component to obtain the data feature information contained in each principal component, that is, to achieve the purpose of extracting features from the original data.

[0050] Finally, the extracted features are divided into training set and test set according to 7:3.

[0051] Use the method of fusing convolutional neural network (CNN) and recurrent neural network (RNN) to build a model based on the extracted features.

[0052] The CNN and RNN are used to extract the spatio-temporal features in the data, and realize the lithology classification. First, the features extracted from the geophysical data are processed and converted into a Mel-frequency spectrogram, and then the CNN is used to extract the spatial features and the RNN is used to extract the time sequence features.

[0053] Specifically, the CNN performs convolution in the time dimension of the seismic data to extract the features of the data in time and space. For the RNN, the long short-term memory network and the gated recurrent unit network are used to model the time sequence of the data and automatically learn the rules in the unlabeled time data. For example, the sequence data is input into the network using the LSTM, the information is continuously transmitted to different time steps, and the final classification result is obtained. Before establishing the deep learning model, the hyperparameters need to be adjusted according to the characteristics of the data set, such as filter size, stride, activation function and loss function, etc. Then the training set is used to train the model, and the validation set is used to evaluate and adjust the model to obtain better generalization performance.

[0054] Specifically, the CNN basic structure includes: convolution layer, pooling layer and full connection layer. Among them, the convolution layer is used to extract the features of the geophysical data in time and space, the pooling layer is used to reduce the dimension of the CNN, and the full connection layer is used to classify the output of the feature extractor and the target. Then, convolution is performed in the time dimension of the seismic data to extract the spatial features of the data. The convolution layer controls the effect of feature extraction by setting hyperparameters such as filter size, stride, activation function, etc. The pooling layer is used to reduce the dimension of the CNN, which reduces the operation time and eliminates redundant data while retaining important feature information of the original data. Finally, the pooling layer is used to reduce the dimension of the CNN, which reduces the operation time and eliminates redundant data while retaining important feature information of the original data.

[0055] And the RNN basic structure is the hidden layer and the cycle layer. The hidden layer is used to store and transmit time sequence information to perform calculations from the previous time step to the current time step, and calculate the output according to the current input. The cycle layer mainly includes: long short-term memory network and gated recurrent unit network.

[0056] In this embodiment, the RNN mainly models the time sequence information in the geophysical data, uses the memory unit to gradually learn the features of the time sequence, and uses the learned features to update and maintain the hidden state information in subsequent calculations. Then, the error back propagation is used in the RNN to capture the time sequence information, and the model effect is further optimized by updating the hidden state information and weight parameters.

[0057] S4. Use the lithology classification model to complete the lithology prediction before the oil drilling process and obtain the prediction results.

[0058] The aforementioned model is used to make a preliminary assessment of the lithology at the site, yielding a result. This aims to facilitate drilling planning before drilling commences, preventing cost waste. Once the planning is complete, subsequent drilling operations can begin.

[0059] S5. Based on the prediction results, conduct drilling; and during the drilling process, collect rock tape data and rock cuttings data; and based on the rock tape data and rock cuttings data, complete lithology identification.

[0060] Based on the above predictions, drilling operations were commenced after the planning was completed. During the drilling process, rock rod data and cuttings data were collected for lithological identification to verify the accuracy of the predictions and provide support for subsequent work.

[0061] The steps include: collecting and cleaning rock ruler data and rock cuttings data; then extracting representative features from the rock physical property data through feature engineering and feature selection.

[0062] Building a decision tree: When classifying, the decision tree recursively divides the dataset according to the order of feature selection until all samples belong to the same category or a preset stopping condition is met.

[0063] Decision tree evaluation: Evaluating the classification accuracy of a decision tree is a crucial step. Cross-validation and confusion matrices are used to assess the classification performance of the decision tree, and the decision trees with higher accuracy are selected for subsequent analysis.

[0064] Classification using decision trees: When classifying, the extracted results are classified through a decision tree to obtain the classification results and complete the identification of lithology.

[0065] Example 2

[0066] like Figure 2 The diagram shown is a schematic representation of the system structure according to an embodiment of this application, including: an acquisition module, a preprocessing module, an extraction module, a construction module, and an identification module; the acquisition module is used to acquire geophysical data at the drilling site before the drilling process; the preprocessing module is used to preprocess the geophysical data to obtain processed data; the extraction module is used to extract features from the processed data to obtain a training set and a test set; the construction module is used to construct a lithology classification model using the training set and the test set; and the identification module is used to complete lithology identification during the oil drilling process using the lithology classification model.

[0067] The following will, in conjunction with this embodiment, explain in detail how this application solves technical problems in real life.

[0068] Before drilling, geophysical data is collected at the site using a data acquisition module. This geophysical data includes seismic data, gravity and magnetic measurement data, electromagnetic data, and borehole data.

[0069] In this implementation, before acquiring seismic data, the acquisition module needs to conduct precise site surveys and plan the exploration data for the exploration area; seismic data is acquired using an unobstructed ground source and detector. In addition, electromagnetic data is acquired using an array of sound sources and receivers. For gravity and magnetic measurement data and borehole data, existing instruments can be used for acquisition.

[0070] The collected geophysical data is preprocessed using a preprocessing module to obtain processed data. The process includes noise removal, amplitude balancing, and time correction.

[0071] High-pass filtering is used to remove low-frequency noise or DC components, filtering out the small-amplitude, slowly changing parts of the real signal. Low-pass filtering is used to remove high-frequency noise, i.e., high-amplitude, short-duration waveforms, while retaining the low-frequency components of the signal. Since there are differences in data amplitude between different sensors, amplitude balancing is necessary to enable effective comparison of data from different sensors. A normalization-based balancing method is used to balance the amplitude. However, due to the influence of equipment and media, the time of data reception may have some offset in the acquired geophysical data. To accurately obtain data characteristics, a bilateral correction method is used for time correction of the data.

[0072] The feature extraction module first performs feature selection on the processed data to obtain candidate features. The chi-square test is then used to determine whether the feature significantly distinguishes different categories.

[0073] First, we need to calculate the expected and actual values ​​of each feature under different categories. Specifically, for a dataset with N features and M categories, the calculation process is as follows:

[0074] Calculate the number of samples K in each category. i And the number of samples K for each feature value i,j ; Calculate the number c of each feature value in the entire dataset. j ; Calculate the number of samples M for each category in the entire dataset. i Calculate the expected number of each feature value in each category.

[0075] E i,j =(K i ×c j ) / M

[0076] Calculate the actual number O of each feature value in each category.i,j ; Calculate the chi-square value χ2 according to the expected value and the actual value, χ2 = Σ(O i,j -E i,j )2 / E i,j

[0077] Then, the critical value of the chi-square value needs to be set according to the significance level. In general, the significance level is 0.05, that is, the p-value is 0.05. The critical value of the chi-square value is calculated according to the significance level and the degree of freedom.

[0078] Finally, for each feature, calculate its chi-square value, if the chi-square value is less than the critical value, the feature has no significant impact on classification and can be deleted; otherwise, it means that the feature has a huge contribution to classification and can be retained.

[0079] After that, feature extraction is performed on the alternative features;

[0080] The principal component analysis (PCA) is used to reduce the high-dimensional data to low-dimensional data, so that the data can be more easily analyzed and displayed. For lithology identification, PCA can be used for feature extraction, and the basic process is as follows:

[0081] Before PCA, the original data needs to be standardized, that is, the original data is standardized to have a mean value of 0 and a variance of 1, in order to reduce the influence between different data models. After that, the number of principal components is determined; by cumulative contribution rate, the number of principal components to be retained is determined. The contribution rate of the principal component refers to the percentage of the original data information that each principal component can express. Generally speaking, the cumulative contribution rate of more than 80% can be considered as a good principal component. Calculate the covariance matrix; calculate the covariance matrix of the standardized data to get the eigenvalues and eigenvectors of the covariance matrix. Finally, calculate the coefficients of each principal component to get the data feature information contained in each principal component, that is, the purpose of extracting features from the original data is achieved.

[0082] Finally, the extracted features are divided into training set and test set according to 7:3.

[0083] The construction module uses the method of fusing convolutional neural network (CNN) and recurrent neural network (RNN) to construct the model based on the extracted features.

[0084] The spatial and temporal features in the data are extracted by using CNN and RNN to realize lithology classification. First, the features extracted from the geophysical data are processed to convert them into Mel-frequency spectrogram, and then CNN is used to extract spatial features and RNN is used to extract time series features.

[0085] Specifically, CNN performs convolution in the time dimension of seismic data to extract the features of data in time and space. For RNN, long short-term memory network and gated recurrent unit network are used to model the time series of data and automatically learn the rules in the unlabeled time data. For example, sequence data is input into the network using LSTM, information is continuously passed to different time steps, and the final classification result is obtained. Before building the deep learning model, hyperparameter adjustment is also needed according to the characteristics of the data set, such as filter size, stride, activation function and loss function, etc. Then the model is trained using the training set, and the validation set is used to evaluate and adjust the model to obtain better generalization performance.

[0086] Specifically, the CNN basic structure includes convolution layer, pooling layer and full connection layer. Among them, the convolution layer is used to extract the features of geophysical data in time and space, the pooling layer is used to reduce the dimension of CNN, and the full connection layer is used to classify the output of the feature extractor and the target. Then, convolution is performed in the time dimension of seismic data to extract the spatial features of the data. The convolution layer controls the effect of feature extraction by setting hyperparameters such as filter size, stride, activation function, etc. The pooling layer is used to reduce the dimension of CNN, which reduces the operation time and eliminates redundant data while retaining important feature information of the original data. Finally, the pooling layer is used to reduce the dimension of CNN, which reduces the operation time and eliminates redundant data while retaining important feature information of the original data.

[0087] And the RNN basic structure is hidden layer and cycle layer. The hidden layer is used to store and transmit time series information to perform calculation from previous time step to current time step, and calculate output according to current input. The cycle layer mainly includes long short-term memory network and gated recurrent unit network.

[0088] In this embodiment, RNN mainly models the time series information in geophysical data, uses memory unit to learn the features of the time series step by step, and uses the learned features to update and maintain hidden state information in subsequent calculations. Then, error back propagation is used in RNN to capture time series information, and the effect of the model is further optimized by updating hidden state information and weight parameters.

[0089] The construction module uses the lithology classification model to complete the lithology prediction before the oil drilling process, and obtains the prediction result. The above model is used to make a preliminary judgment on the lithology of the construction site, and the judgment result is obtained. The purpose of this is to facilitate the planning of the drilling plan before drilling and avoid wasting costs. After planning, subsequent drilling work can be carried out.

[0090] Finally, the last identification module performs drilling based on the prediction result, and collects rock size data and rock debris data during the drilling process, and completes lithology identification based on the rock size data and the rock debris data.

[0091] Based on the above prediction result, after completing the scheme planning, drilling work is carried out. During the drilling process, rock size data and rock debris data are collected, lithology identification is performed, and the accuracy of the above prediction result is verified to support the subsequent work.

[0092] The process includes: collecting and cleaning rock size data and rock debris data; then extracting representative features from rock physical property data through feature engineering and feature selection.

[0093] Building a decision tree: when classifying, the decision tree recursively divides the data set according to the order of feature selection until all samples belong to the same category or the preset stopping condition is reached.

[0094] Decision tree evaluation: evaluating the classification accuracy of the decision tree is an important process. The classification effect of the decision tree is evaluated through cross-validation and confusion matrix, and the decision tree with higher accuracy is selected for subsequent analysis.

[0095] Classify using decision tree: when classifying, the extracted results are classified by the decision tree to obtain the classification results and complete the identification of lithology.

[0096] The above-described embodiments are only descriptions of the preferred modes of the present application and do not limit the scope of the present application. Without departing from the design spirit of the present application, various modifications and improvements to the technical solutions of the present application made by those skilled in the art shall fall within the protection scope determined by the claims of the present application.

Claims

1. A method for lithology identification applied to oil drilling, characterized by the steps of The method comprises the following steps: collecting geophysical data in a work site before a drilling process; preprocessing the geophysical data to obtain processed data; extracting features from the processed data to obtain a training set and a test set; and using the training set and the test set to build a lithology classification model; the step comprises, based on the extracted features, using a convolutional neural network and a recurrent neural network to build the lithology classification model; using a method of fusing a convolutional neural network and a recurrent neural network to build a model based on the extracted features; using CNN and RNN to extract the spatio-temporal features in the data to realize lithology classification; first, the features extracted from the geophysical data are processed and converted into a mel-frequency spectrogram, then CNN is used to extract spatial features and RNN is used to extract time series features; CNN performs convolution in the time dimension of seismic data to extract the spatial and temporal features of the data; for RNN, long short-term memory networks and gated recurrent unit networks are used to model the time series of the data and automatically learn the rules in the unlabeled time data; The CNN basic structure comprises a convolutional layer, a pooling layer and a fully connected layer; the convolutional layer is used to extract the features of the geophysical data in time and space, the pooling layer is used to reduce the dimension of the CNN, and the fully connected layer is used to classify the output of the feature extractor and the target; then, convolution is performed in the time dimension of the seismic data to extract the spatial features of the data; the pooling layer is used to reduce the dimension of the CNN, which reduces the operation time and eliminates redundant data while retaining important feature information of the original data; finally, the pooling layer is used to reduce the dimension of the CNN, which reduces the operation time and eliminates redundant data while retaining important feature information of the original data; The RNN basic structure is a hidden layer and a recurrent layer; the hidden layer is used to store and transmit time series information to perform calculations from previous time steps to the current time step and calculate the output according to the current input; the recurrent layer comprises long short-term memory networks and gated recurrent unit networks; using the lithology classification model to complete lithology prediction before the drilling process to obtain a prediction result; based on the prediction result, drilling is carried out; and during the drilling process, rock size data and rock chip data are collected; and based on the rock size data and the rock chip data, lithology identification is completed.

2. The lithology identification method for oil drilling according to claim 1, characterized in that, The geophysical data comprises seismic data, gravity and magnetic measurement data, electromagnetic data and borehole data; the seismic data is collected by using a non-isolation seismic source and a geophone; and the electromagnetic data is collected by using an array of sound sources and receivers.

3. The lithology identification method for oil drilling according to claim 1, characterized in that, The method of feature extraction comprises the following steps: feature selection is performed on the processed data to obtain candidate features; then, feature extraction is performed on the candidate features to complete the feature extraction; finally, the extracted features are divided into the training set and the test set.

4. The lithology identification method for oil drilling according to claim 1, characterized in that, The decision tree algorithm is used to complete the lithology identification.

5. A lithology identification system for use in oil drilling, characterized in that, The method comprises the following steps: a collection module, a preprocessing module, an extraction module, a construction module and an identification module; the collection module is used to collect geophysical data in a work site before a drilling process; The preprocessing module is configured to preprocess the geophysical data to obtain processed data; The extraction module is configured to extract features from the processed data to obtain a training set and a test set; And the training set and the test set are used to build a lithology classification model; The process includes feature selection on the processed data to obtain candidate features; then, the candidate features are extracted to complete the feature extraction; finally, the extracted features are divided into the training set and the test set; The CNN and RNN are used to extract the spatial and temporal features of the data to realize lithology classification. First, the features extracted from the geophysical data are processed to convert them into a mel-frequency spectrogram, and then the CNN is used to extract spatial features and the RNN is used to extract time series features. The CNN performs convolution in the time dimension of the seismic data to extract the spatial and temporal features of the data; for the RNN, the long short-term memory network and the gated recurrent unit network are used to model the time series of the data and automatically learn the rules in the unlabeled time data; The CNN basic structure includes a convolution layer, a pooling layer, and a fully connected layer; the convolution layer is used to extract the spatial and temporal features of the geophysical data, the pooling layer is used to reduce the dimension of the CNN, and the fully connected layer is used to classify the output of the feature extractor with the target; then, the convolution is performed in the time dimension of the seismic data to extract the spatial features of the data; the pooling layer is used to reduce the dimension of the CNN, which reduces the operation time and eliminates redundant data while retaining important feature information of the original data; finally, the pooling layer is used to reduce the dimension of the CNN, which reduces the operation time and eliminates redundant data while retaining important feature information of the original data; The RNN basic structure is a hidden layer and a recurrent layer; the hidden layer is used to store and transmit time series information to perform calculations from previous time steps to the current time step and calculate the output according to the current input; the recurrent layer includes a long short-term memory network and a gated recurrent unit network. The construction module is configured to use the lithology classification model to complete lithology prediction before the oil drilling process to obtain a prediction result. The identification module is configured to drill based on the prediction result, and collect rock size data and rock debris data during the drilling process, and complete lithology identification based on the rock size data and the rock debris data.

6. The lithology identification system for use in oil drilling according to claim 5, characterized in that, The geophysical data includes seismic data, gravity and magnetic survey data, electromagnetic data, and borehole data; the working process of the acquisition module includes acquiring the seismic data using a non-blocking seismic source and a geophone; and acquiring the electromagnetic data using an array of sound sources and receivers.

7. The lithology identification system for use in oil drilling according to claim 5, characterized in that, The working process of the construction module includes using a convolutional neural network and a recurrent neural network to build the lithology classification model based on the extracted features.

8. The lithology identification system for use in oil drilling according to claim 5, characterized in that, The working process of the identification module includes using a decision tree algorithm to complete the lithology identification.

Images