Real-time continuous rock debris sampling and intelligent geological logging method

Abstract

The invention relates to the technical field of petroleum geological exploration, in particular to a real-time continuous rock debris sampling and intelligent geological logging method which comprises the following steps: S1, operating a drilling machine to perform oil and gas drilling operation, enabling mud carrying rock debris to flow to the ground along with a drill rod and a drill bit which rotate at a high speed, and separating the rock debris in drilling fluid by using a vibrating screen; s2, the separated rock debris is conveyed through a spiral conveying belt, a mechanical arm is regularly operated according to the sampling interval T to grab the rock debris to a high-speed centrifugal machine, and centrifugal drying operation of the rock debris is conducted; s3, the dried rock debris is put into a hydraulic machine, and a rock debris abrasive disc is obtained; s4, a mechanical arm is used for transferring the rock debris abrasive disc to a scanner, and a scanning image is obtained; and S5, performing feature fusion analysis on the scanned image to generate a three-dimensional geologic model. According to the real-time continuous rock debris sampling and intelligent geological logging method, through a dynamic sampling algorithm, the limitation of traditional fixed interval sampling is broken through, and the problem of rock debris leakage of rapid drilling and uncovering of a stratum is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of petroleum geological exploration, and in particular to a real-time continuous rock cuttings sampling and intelligent geological logging method. Background Art

[0002] Geological logging is one of the core technologies in petroleum geological exploration. It collects and analyzes real-time information on wellbore return materials such as solids, liquids, and gases during the drilling process, combines it with indirect parameters (such as drilling time and changes in mud properties), and systematically establishes underground geological profiles to accurately locate the position, thickness, and fluid properties of oil and gas layers.

[0003] Traditional cuttings logging technology has the following defects: 1. Sampling discreteness: Manual timed sampling leads to data gaps, making it impossible to capture lithologic changes in rapidly drilled formations; 2. Data lag: The separation of scanning and analysis makes it difficult to provide timely feedback on formation information. Therefore, a real-time continuous cuttings sampling and intelligent geological logging method is proposed to address the above problems. Summary of the Invention

[0004] (1) Technical problems solved

[0005] In response to the shortcomings of the existing technology, the present invention provides a real-time continuous rock cuttings sampling and intelligent geological logging method, which has the advantages of improving the rock cuttings capture rate in complex formations and solving the problem of rock cuttings leakage in rapidly drilling formations.

[0006] (2) Technical solution

[0007] To achieve the above-mentioned purpose of improving the cuttings capture rate in complex formations, the present invention provides the following technical solution: a real-time continuous cuttings sampling and intelligent geological logging method, comprising the following steps:

[0008] S1. Operate a drilling rig to perform oil and gas drilling operations. Mud carrying cuttings flows to the surface along the high-speed rotating drill pipe and drill bit. A vibrating screen is used to separate the cuttings from the drilling fluid.

[0009] S2. The separated cuttings are transported by a spiral conveyor belt, and the cuttings are captured by a robotic arm at a timing of the sampling interval T and centrifuged to dry them in a high-speed centrifuge.

[0010] S3. The dried cuttings are put into a hydraulic press to obtain cuttings grinding discs;

[0011] S4. Use the robotic arm to transfer the rock chip to the scanner to obtain a scanned image;

[0012] S5. Perform feature fusion analysis on the scanned images to generate a three-dimensional geological model.

[0013] Preferably, the dynamic sampling algorithm in step S2 includes the following sub-steps:

[0014] S2.1. Set up sensor equipment at the drilling fluid outlet to monitor the composition of cuttings in real time, combined with the well depth D i Calculate the change rate of rock cuttings composition

[0015] S2.2. Based on the change rate of rock fragment composition Obtain the formation lithology mutation threshold ΔL;

[0016] S2.3. Based on the real-time drilling speed R, the sampling interval T is obtained by combining the weights α and β.

[0017] Preferably, the raw materials are in the following proportions by weight:

[0018] S2.1.1. Define well depth D i The jth component content of the rock fragments at location C i,j ;

[0019] S2.1.2. Using the moving average method to calculate C i,j Smoothing is performed to obtain C i,j ′:

[0020]

[0021] Where ω is the window size, which indicates the number of samples involved in smoothing;

[0022] S2.1.3. For each component j, calculate the rate of change of the component between adjacent depths.

[0023]

[0024] Where ΔD i =D i+1 -D i is the depth difference between two measurement points, j = 1, 2, 3, ···, m;

[0025] S2.1.4. Use the weighted average method to combine the change rates of all components

[0026]

[0027] Among them, ω j is the weight of the jth component.

[0028] Preferably, the step S2.2 further includes the following sub-steps:

[0029] S2.2.1. Construct a database of formation lithology mutation thresholds. The database stores the composition change rate set Cd, well depth H oldAnd the corresponding lithologic mutation threshold ΔL old ;

[0030] S2.2.2. Based on the composition change correction coefficient δ, the range of rock cutting composition change rate Ran is obtained:

[0031] Ran-δ≤Ran≤Ran+δ;

[0032] S2.2.3. Traverse the lithologic mutation threshold database. When Cd min ≤Ran≤Cd max If yes, go to S2.2.4. Otherwise, go to S2.2.5.

[0033] S2.2.4. When H old -θ≤H old +θ, where θ is the well depth correction coefficient, ΔL=ΔL old , otherwise jump to S2.2.5;

[0034] S2.2.5. Based on the well depth weight factor WD i , the formation lithology mutation threshold ΔL is calculated.

[0035] Preferably, the step S2.3 further includes the following sub-steps:

[0036] S2.3.1. Construct a weight system database, which stores the rotation speed R, the rock cutting composition change rate, and the and sampling efficiency Y;

[0037] S2.3.2. Use the Pearson correlation coefficient to calculate the weight coefficients α and β:

[0038] Let feature X1 = R, For any feature X i The correlation with the target variable Y, whose Pearson correlation coefficient is defined as:

[0039]

[0040] Among them, Cov(X i ,Y) is feature X i Covariance with the target variable Y, σX i and σY are the features X i The standard deviation from the target variable Y;

[0041] Calculate the weight coefficient Weight coefficient

[0042] Where α+β=1;

[0043] S2.3.3. Calculate the sampling interval T:

[0044]

[0045] Among them, k is a proportional constant with a value of 2 to 3, which is used to adjust the overall dimension of the formula.

[0046] Preferably, the step S2.5 further includes the following sub-steps:

[0047] S2.5.1. Perform image preprocessing on the scanned image;

[0048] S2.5.2. Based on the multimodal Transformer model, multimodal feature fusion is performed to obtain the fused feature F f ;

[0049] S2.5.3. Based on F f The lithology classification and oil and gas probability are obtained through analysis, and a lithology histogram and a heat map of oil and gas display probability are produced.

[0050] Preferably, the step S2.5.1 further comprises the following steps:

[0051] S2.5.1.1. The scanned images are RGB image img1 (3 × H × W) and polarized image img2 (4 × H × W).

[0052] S2.5.1.2. Normalize the pixel values of image img1 to a distribution with a mean of 0.5 and a standard deviation of 0.5.

[0053] S2.5.1.3. Enhance high-contrast regions of the polarized image img2 using an adaptive sigmoid function.

[0054] Preferably, the step S2.5.2 further comprises the following steps:

[0055] S2.5.2.1. Use the pre-trained convolutional neural network (CNN) to extract image features of the RGB image img1:

[0056]

[0057] S2.5.2.2. Use the convolutional neural network (CNN) to extract polarization features of the polarized light image img2:

[0058]

[0059] S2.5.2.3. Combine RGB and polarization features into image modality features:

[0060]

[0061] Among them, Concat(·) represents the vector concatenation operation;

[0062] S2.5.2.4. Use the one-dimensional convolution layer Conv1D to extract local spectral features from the debris XRF spectrum data S:

[0063] H C =Conv1D(S)

[0064] Modeling global spectral sequence relationships through Transformer encoder:

[0065] F spe =TransformerEncoder(H C );

[0066] S2.5.2.5. Map the physical property parameters P = [H, D] to a high-dimensional space through a fully connected layer:

[0067] F Phys =W P P+b P

[0068] in, b P is bias;

[0069] S2.5.2.6. Project each modal feature to a unified dimension D i :

[0070] E img =W img F img ,

[0071] E spe =W spe F spe ,

[0072] E Phys =W Phys F Phys

[0073] Among them, W img , W spe , W Phys is the projection matrix;

[0074] S2.5.2.7. Add modal type code (T img ,T spe ,T Phys ∈R D ) and position encoding (PE(i)∈R D ):

[0075] E t =[E img+T img +PE(1);E spe +T spe +PE(2);E Phys +T Phys +PE(3)];

[0076] S2.5.2.8. Modeling inter-modality correlation using multi-head self-attention:

[0077]

[0078] Where Q = E t W Q , K=E t W K , V=E t W V ;

[0079] S2.5.2.9. Generate fusion features F through feedforward network f .

[0080] Preferably, the step S2.5.3 further includes the following sub-steps:

[0081] S2.5.3.1. Fusion feature F f As input, a fully connected layer is used, whose output dimension is the number of lithology categories N class :

[0082] S2.5.3.2. Apply the softmax function to F f Converted into a probability distribution so that the score of each category can be interpreted as the probability that the sample belongs to the corresponding category

[0083] y li =softmax(F f +W class +b class )

[0084] Among them, W class and b class Represent the weight matrix and bias term respectively;

[0085] S2.5.3.3. According to the well depth D during drilling i Information, stack the lithology classification probability y of each sample in sequence li , forming a two-dimensional array;

[0086] S2.5.3.4. Based on the stacked probability values, create a heat map to show the probability of occurrence of each lithology at different depths. For each sample segment, select the category with the highest probability as the final lithology classification result and create a lithology histogram based on this result.

[0087] S2.5.3.5. Use another fully connected layer to fusion feature F f Mapping to the probability of oil and gas existence, the Sigmoid function is applied to limit the output value between 0 and 1, representing the probability of oil and gas existence;

[0088] The fully connected layer is mapped to the oil and gas probability value y li :

[0089] p li =σ(W hc F f +b hc )

[0090] Among them, σ is the sigmoid function, W hc and b hc are the weight matrix and bias term respectively;

[0091] S2.5.3.6. Set the oil and gas probability value p li Mapped to color space, combined with lithologic histograms, the oil and gas probability distribution is overlaid to display the results.

[0092] (3) Beneficial effects

[0093] Compared with the existing technology, the present invention provides a real-time continuous rock cuttings sampling and intelligent geological logging method, which has the following beneficial effects:

[0094] 1. This real-time continuous cuttings sampling and intelligent geological logging method, through a dynamic sampling algorithm, breaks through the limitations of traditional fixed-interval sampling and solves the problem of missed cuttings during rapid drilling.

[0095] 2. This real-time continuous rock cuttings sampling and intelligent geological logging method can more comprehensively capture the characteristics of underground rocks and fluids by integrating data from multiple modalities, thereby improving the accuracy and reliability of lithology and oil and gas display predictions.

[0096] 3. This real-time continuous rock cuttings sampling and intelligent geological logging method greatly reduces the need for manual intervention and improves work efficiency through a fully automatic analysis process from raw data to final results.

[0097] 4. This real-time continuous rock cuttings sampling and intelligent geological logging method automatically generates lithologic histograms and oil and gas display probability heat maps, providing petroleum geologists with an intuitive and detailed description of the subsurface conditions, helping to make exploration and development decisions faster and more accurately, reducing risks and optimizing investment returns. BRIEF DESCRIPTION OF THE DRAWINGS

[0098] Figure 1 This is a flow chart of a real-time continuous cuttings sampling and intelligent geological logging method proposed by the present invention. DETAILED DESCRIPTION

[0099] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0100] Example 1: A real-time continuous rock cuttings sampling and intelligent geological logging method, comprising the following steps:

[0101] S1. Operate a drilling rig to perform oil and gas drilling operations. Mud carrying cuttings flows to the surface along the high-speed rotating drill pipe and drill bit. A vibrating screen is used to separate the cuttings from the drilling fluid.

[0102] S2. The separated cuttings are transported by a spiral conveyor belt, and the cuttings are captured by a robotic arm at a timing of the sampling interval T and centrifuged to dry them in a high-speed centrifuge.

[0103] S3. The dried cuttings are put into a hydraulic press to obtain cuttings grinding discs;

[0104] S4. Use the robotic arm to transfer the rock chip to the scanner to obtain a scanned image;

[0105] S5. Perform feature fusion analysis on the scanned images to generate a three-dimensional geological model.

[0106] In this embodiment, the dynamic sampling algorithm in step S2 includes the following sub-steps:

[0107] S2.1. Set up sensor equipment at the drilling fluid outlet to monitor the composition of cuttings in real time, combined with the well depth D i Calculate the change rate of rock cuttings composition

[0108] Well deep D i Steps to obtain:

[0109] Fix the laser rangefinder in a stable position on the drilling rig mast, aiming it at the traveling block or hook (the part that moves as the drill pipe is lowered), and ensure that the laser path is unobstructed;

[0110] Setting D i The initial value is 0, and the real-time distance value of the laser rangefinder is collected at a fixed frequency (10Hz), and the timestamp t is recorded. i The corresponding measured value CE(t i );

[0111] Calculate the drill pipe movement increment ΔCE:

[0112] ΔCE=CE(t i )-CE(t i-1 );

[0113] According to the pulley ratio θ, the movement increment ΔCE is converted into the actual well depth increment:

[0114] ΔD i =ΔCE·θ;

[0115] Update well depth:

[0116] D i =D(t i-1 )+ΔD i .

[0117] S2.2. Based on the change rate of rock fragment composition Obtain the formation lithology mutation threshold ΔL;

[0118] S2.3. Based on the real-time drilling speed R, the sampling interval T is obtained by combining the weights α and β.

[0119] Steps to obtain drilling speed R:

[0120]

[0121] In this embodiment, the following raw materials are included in the following weight ratios:

[0122] S2.1.1. Define well depth D i The jth component content of the rock fragments at location C i,j ;

[0123] S2.1.2. Using the moving average method to calculate C i,j Smoothing is performed to obtain C i,j ′:

[0124]

[0125] Where ω is the window size, which indicates the number of samples involved in smoothing;

[0126] S2.1.3. For each component j, calculate the rate of change of the component between adjacent depths.

[0127]

[0128] Where ΔD i =D i+1 -D i is the depth difference between two measurement points, j = 1, 2, 3, ···, m;

[0129] S2.1.4. Use the weighted average method to combine the change rates of all components

[0130]

[0131] Among them, ω j is the weight of the jth component.

[0132] In this embodiment, step S2.2 further includes the following sub-steps:

[0133] S2.2.1. Construct a database of formation lithology mutation thresholds. The database stores the composition change rate set Cd, well depth H old And the corresponding lithologic mutation threshold ΔL old ;

[0134] S2.2.2. Based on the composition change correction coefficient δ, the range of rock cutting composition change rate Ran is obtained:

[0135] Ran-δ≤Ran≤Ran+δ;

[0136] S2.2.3. Traverse the lithologic mutation threshold database. When Cd min ≤Ran≤Cd max If yes, go to S2.2.4. Otherwise, go to S2.2.5.

[0137] S2.2.4. When H old -θ≤H old +θ, where θ is the well depth correction coefficient, ΔL=ΔL old , otherwise jump to S2.2.5;

[0138] S2.2.5. Based on the well depth weight factor WD i , the formation lithology mutation threshold ΔL is calculated.

[0139]

[0140] Among them, r is a positive parameter

[0141] Cuttings composition change rate after introducing well depth weight factor

[0142]

[0143] Calculation of rock cuttings composition change rate The mean and standard deviation of :

[0144]

[0145] Where n is the total number of data points, μ is the mean, and σ is the standard deviation;

[0146] Calculate the formation lithology mutation threshold ΔL:

[0147] ΔL=μ+K·σ.

[0148] In this embodiment, step S2.3 further includes the following sub-steps:

[0149] S2.3.1. Construct a weight system database, which stores the rotation speed R, the rock cutting composition change rate, and the and sampling efficiency Y;

[0150] S2.3.2. Use the Pearson correlation coefficient to calculate the weight coefficients α and β:

[0151] Let feature X1 = R, For any feature X i The correlation with the target variable Y, whose Pearson correlation coefficient is defined as:

[0152]

[0153] Among them, Cov(X i ,Y) is feature X i Covariance with the target variable Y, σX i and σY are the features X i The standard deviation from the target variable Y;

[0154] Specifically, the covariance can be expressed as:

[0155]

[0156] Where N is the number of samples, X i,j and Y j are the features X in the jth sample i and the value of the target variable Y, and They are feature X i and the mean of the target variable Y.

[0157] Calculate the weight coefficient Weight coefficient

[0158] Where α+β=1;

[0159] S2.3.3. Calculate the sampling interval T:

[0160]

[0161] Among them, k is a proportional constant with a value of 2 to 3, which is used to adjust the overall dimension of the formula.

[0162] In this embodiment, step S2.5 further includes the following sub-steps:

[0163] S2.5.1. Perform image preprocessing on the scanned image;

[0164] S2.5.2. Based on the multimodal Transformer model, multimodal feature fusion is performed to obtain the fused feature F f ;

[0165] S2.5.3. Based on F f The lithology classification and oil and gas probability are obtained through analysis, and a lithology histogram and a heat map of oil and gas display probability are produced.

[0166] In this embodiment, step S2.5.1 further includes the following steps:

[0167] S2.5.1.1. The scanned images are RGB image img1 (3 × H × W) and polarized image img2 (4 × H × W).

[0168] S2.5.1.2. Normalize the pixel values of image img1 to a distribution with a mean of 0.5 and a standard deviation of 0.5.

[0169] Assume that the pixel values of image img1 are three-dimensional array I(x,y,c), where x,y represent the spatial position of the image and c∈{R,G,B} represents the RGB channel;

[0170] Calculate the mean and standard deviation for each channel separately:

[0171]

[0172] Where N is the total number of pixels in the image, that is, N = H × W, μ c is the mean of the cth channel, σ c is the standard deviation of the cth channel;

[0173] Normalized to a distribution with a mean of 0.5 and a standard deviation of 0.5:

[0174] I′(x,y,c)=0.5+0.5·I(x,y,c)

[0175] S2.5.1.3. Enhance high-contrast regions of the polarized image img2 using an adaptive sigmoid function.

[0176] Assume that the pixel value of the polarized light image img2 is a two-dimensional array I(x,y),

[0177] Calculate the local mean μ:

[0178]

[0179] Calculate the local standard deviation σ:

[0180]

[0181] Among them, ω xy is a local window centered at (x,y), and N is the total number of pixels in the local window

[0182] Get the steepness κ and the inflection point position θ:

[0183] κ=κ0·σ(x,y)

[0184] θ=μ(x,y)

[0185] Among them, κ0 is the global proportional coefficient, κ0 = 0.5

[0186] Input the pixel value into the Sigmoid function to obtain the enhanced polarized light pixel value:

[0187]

[0188] In this embodiment, step S2.5.2 further includes the following steps:

[0189] S2.5.2.1. Use the pre-trained convolutional neural network (CNN) to extract image features of the RGB image img1:

[0190]

[0191] Input layer: The input is the RGB image img1;

[0192] Convolutional layer: Each convolutional layer applies a set of learnable filters (weight matrices) to the input image to generate a series of feature maps that capture local patterns at different scales, such as basic shape features such as edges and corners;

[0193] Pooling layer: A pooling layer follows the convolutional layer. The pooling operation reduces the spatial size of the feature map.

[0194] Fully connected layer: The output of the last convolutional layer is flattened and then fed into a fully connected layer to convert the high-dimensional features into a fixed-size vector, which serves as the overall feature representation of the image.

[0195] Output features: the final feature vector

[0196] S2.5.2.2. Use the convolutional neural network (CNN) to extract polarization features of the polarized light image img2:

[0197]

[0198] Input layer: The input is the polarized light image img2;

[0199] Convolutional layer: The first layer is a convolutional layer with 32 filters of size 3×3, followed by a series of convolutional layers and ReLU activation functions. Each layer deepens the network's understanding of image features.

[0200] Pooling layer: a series of max-pooling layers to reduce the spatial size of the feature map while retaining the most important feature information;

[0201] Fully connected layer: The output of the last convolutional layer is flattened and then fed into a fully connected layer to convert the high-dimensional features into a fixed-size vector, which serves as the overall feature representation of the image.

[0202] Output features: the final feature vector

[0203] S2.5.2.3. Combine RGB and polarization features into image modality features:

[0204]

[0205] Among them, Concat(·) represents the vector concatenation operation;

[0206] S2.5.2.4. Use the one-dimensional convolution layer Conv1D to extract local spectral features from the debris XRF spectrum data S:

[0207] H C =Conv1D(S)

[0208] Modeling global spectral sequence relationships through Transformer encoder:

[0209] F spe =TransformerEncoder(H C );

[0210] S2.5.2.5. Map the physical property parameters P = [H, D] to a high-dimensional space through a fully connected layer:

[0211]

[0212] in, b P is bias;

[0213] S2.5.2.6. Project each modal feature to a unified dimension D i :

[0214] E img =W img F img ,

[0215] E spe =W spe F spe ,

[0216] E Phys =W Phys F Phys

[0217] Among them, W img , W spe , W Phys is the projection matrix;

[0218] S2.5.2.7. Add modal type code (T img ,T spe ,T Phys ∈R D ) and position encoding (PE(i)∈R D ):

[0219] E t =[E img +T img +PE(1);E spe +T spe +PE(2);E Phys +T Phys +PE(3)];

[0220] S2.5.2.8. Modeling inter-modality correlation using multi-head self-attention:

[0221]

[0222] Where Q = E t W Q , K=E t W K , V=E t W V ;

[0223] S2.5.2.9. Generate fusion features F through feedforward network f .

[0224] In this embodiment, step S2.5.3 further includes the following sub-steps:

[0225] S2.5.3.1. Fusion feature F f As input, a fully connected layer is used, whose output dimension is the number of lithology categories N class :

[0226] S2.5.3.2. Apply the softmax function to F f Converted into a probability distribution so that the score of each category can be interpreted as the probability that the sample belongs to the corresponding category

[0227] y li =softmax(F f +W class +b class )

[0228] Among them, W class and b class Represent the weight matrix and bias term respectively;

[0229] S2.5.3.3. According to the well depth D during drilling i Information, stack the lithology classification probability y of each sample in sequence li , forming a two-dimensional array;

[0230] S2.5.3.4. Based on the stacked probability values, create a heat map to show the probability of occurrence of each lithology at different depths. For each sample segment, select the category with the highest probability as the final lithology classification result and create a lithology histogram based on this result.

[0231] S2.5.3.5. Use another fully connected layer to fusion feature F f Mapping to the probability of oil and gas existence, the Sigmoid function is applied to limit the output value between 0 and 1, representing the probability of oil and gas existence;

[0232] The fully connected layer is mapped to the oil and gas probability value y li :

[0233] p li =σ(W hc F f +b hc )

[0234] Among them, σ is the sigmoid function, W hc and b hc are the weight matrix and bias term respectively;

[0235] S2.5.3.6. Set the oil and gas probability value p li Mapped to color space, combined with lithologic histograms, the oil and gas probability distribution is overlaid to display the results.

[0236] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A real-time continuous cuttings sampling and intelligent geological logging method, characterized in that: The following steps are involved: S1. Operate a drilling rig to perform oil and gas drilling operations. Mud carrying cuttings flows to the surface along the high-speed rotating drill pipe and drill bit. A vibrating screen is used to separate the cuttings from the drilling fluid. S2. The separated cuttings are transported by a spiral conveyor belt, and the cuttings are captured by a robotic arm at a timing of the sampling interval T and centrifuged to dry them in a high-speed centrifuge. S3. The dried cuttings are put into a hydraulic press to obtain cuttings grinding discs; S4. Use the robotic arm to transfer the rock chip to the scanner to obtain a scanned image; S5. Perform feature fusion analysis on the scanned images to generate a three-dimensional geological model.

2. A real-time continuous cuttings sampling and intelligent geological logging method according to claim 1, characterized in that: The dynamic sampling algorithm in step S2 includes the following sub-steps: S2.

1. Set up sensor equipment at the drilling fluid outlet to monitor the composition of cuttings in real time, combined with the well depth D i Calculate the change rate of cuttings composition S2.

2. Based on the change rate of rock fragment composition Obtain the formation lithology mutation threshold ΔL; S2.

3. Based on the real-time drilling speed R, the sampling interval T is obtained by combining the weights α and β.

3. A real-time continuous cuttings sampling and intelligent geological logging method according to claim 2, characterized in that: The invention comprises the following raw materials in proportion by weight: S2.1.

1. Define well depth D i The jth component content of the rock fragments at location C i,j ; S2.1.

2. Using the moving average method to calculate C i,j Smoothing is performed to obtain C i,j ′: Where ω is the window size, which indicates the number of samples involved in smoothing; S2.1.

3. For each component j, calculate the rate of change of the component between adjacent depths. Where ΔD i =D i+1 -D i is the depth difference between two measurement points, j = 1, 2, 3, ···, m; S2.1.

4. Use the weighted average method to combine the change rates of all components Among them, ω j is the weight of the jth component.

4. A real-time continuous cuttings sampling and intelligent geological logging method according to claim 3, characterized in that: The step S2.2 further includes the following sub-steps: S2.2.

1. Construct a database of formation lithology mutation thresholds. The database stores the composition change rate set Cd, well depth H old And the corresponding lithologic mutation threshold ΔL old ; S2.2.

2. Based on the composition change correction coefficient δ, the range of rock cutting composition change rate Ran is obtained: Ran-δ≤Ran≤Ran+δ; S2.2.

3. Traverse the lithologic mutation threshold database. When Cd min ≤Ran≤Cd max If yes, go to S2.2.

4. Otherwise, go to S2.2.

5. S2.2.

4. When H old -θ≤H old +θ, where θ is the well depth correction coefficient, ΔL=ΔL old , otherwise jump to S2.2.5; S2.2.

5. Based on the well depth weight factor WD i , the formation lithology mutation threshold ΔL is calculated.

5. The method of real-time continuous cuttings sampling and intelligent geological logging according to claim 2, characterized in that: The step S2.3 further includes the following sub-steps: S2.3.

1. Construct a weight system database, which stores the rotation speed R, the rock cutting composition change rate, and the and sampling efficiency Y; S2.3.

2. Use the Pearson correlation coefficient to calculate the weight coefficients α and β: Let feature X1 = R, For any feature X i The correlation with the target variable Y, whose Pearson correlation coefficient is defined as: Among them, Cov(X i ,Y) is feature X i Covariance with the target variable Y, σX i and σY are the features X i The standard deviation from the target variable Y; Calculate the weight coefficient Weight coefficient Where α+β=1; S2.3.

3. Calculate the sampling interval T: Among them, k is a proportional constant with a value of 2 to 3, which is used to adjust the overall dimension of the formula.

6. The method of real-time continuous rock cuttings sampling and intelligent geological logging according to claim 1, characterized in that: The step S2.5 further includes the following sub-steps: S2.5.

1. Perform image preprocessing on the scanned image; S2.5.

2. Based on the multimodal Transformer model, multimodal feature fusion is performed to obtain the fused feature F f ; S2.5.

3. Based on F f The lithology classification and oil and gas probability are obtained through analysis, and a lithology histogram and a heat map of oil and gas display probability are produced.

7. A real-time continuous rock cuttings sampling and intelligent geological logging method according to claim 6, characterized in that: The step S2.5.1 further includes the following steps: S2.5.1.

1. The scanned images are RGB image img1 (3 × H × W) and polarized image img2 (4 × H × W). S2.5.1.

2. Normalize the pixel values of image img1 to a distribution with a mean of 0.5 and a standard deviation of 0.

5. S2.5.1.

3. Enhance the high-contrast areas of the polarized light image img2 using an adaptive Sigmoid function.

8. The method of real-time continuous cuttings sampling and intelligent geological logging according to claim 7, characterized in that: The step S2.5.2 further includes the following steps: S2.5.2.

1. Use the pre-trained convolutional neural network (CNN) to extract image features of the RGB image img1: S2.5.2.

2. Use the convolutional neural network (CNN) to extract polarization features of the polarized light image img2: S2.5.2.

3. Combine RGB and polarization features into image modality features: Among them, Concat(·) represents the vector concatenation operation; S2.5.2.

4. Use the one-dimensional convolution layer Conv1D to extract local spectral features from the debris XRF spectrum data S: H C =Conv1D(S) Modeling global spectral sequence relationships through Transformer encoder: F spe =TransformerEncoder(H C ); S2.5.2.

5. Map the physical property parameters P = [H, D] to a high-dimensional space through a fully connected layer: F Phys =W P P+b P Among them, W P ∈R 2×DPh y s , b P is bias; S2.5.2.

6. Project each modal feature to a unified dimension D i : HAVE BEEN img =W img F img , HAVE BEEN spe =W spe F spe , HAVE BEEN Phys =W Phys F Phys Among them, W img , W spe , W Phys is the projection matrix; S2.5.2.

7. Add modal type code (T img ,T spe ,T Phys ∈R D ) and position encoding (PE(i)∈R D ): It is t =[E img +T img +PE(1);E spe +T spe +PE(2);E Phys +T Phys +PE(3)]; S2.5.2.

8. Modeling inter-modality correlation using multi-head self-attention: where Q = E t W Q and K = E t W K and V = E t W V ; S2.5.2.

9. Generate fusion features F through feedforward network f .

9. The method of real-time continuous rock cuttings sampling and intelligent geological logging according to claim 6, characterized in that: The step S2.5.3 further includes the following sub-steps: S2.5.3.

1. Fusion feature F f As input, a fully connected layer is used, whose output dimension is the number of lithology categories N class : S2.5.3.

2. Apply the softmax function to F f Converted into a probability distribution so that the score of each category can be interpreted as the probability that the sample belongs to the corresponding category y li =softmax(F f +W class +b class ) Among them, W class and b class Represent the weight matrix and bias term respectively; S2.5.3.

3. According to the well depth D during drilling i Information, stack the lithology classification probability y of each sample in sequence li , forming a two-dimensional array; S2.5.3.

4. Based on the stacked probability values, create a heat map to show the probability of occurrence of each lithology at different depths. For each sample segment, select the category with the highest probability as the final lithology classification result and create a lithology histogram based on this result. S2.5.3.

5. Use another fully connected layer to fusion feature F f Mapping to the probability of oil and gas existence, the Sigmoid function is applied to limit the output value between 0 and 1, representing the probability of oil and gas existence; The fully connected layer is mapped to the oil and gas probability value y li : p li =σ(W hc F f +b hc ) Among them, σ is the sigmoid function, W hc and b hc are the weight matrix and bias term respectively; S2.5.3.

6. Set the oil and gas probability value p li Mapped to color space, combined with lithologic histograms, the oil and gas probability distribution is overlaid to display the results.

Images