Rock stratum structure surface inclination detection method based on linear multi-point while-drilling sensing

Abstract

The application discloses a rock stratum structure surface inclination detection method based on linear multi-point while-drilling sensing, and comprises the following steps: setting a signal generating node and a signal receiving node, the signal receiving node is arranged behind the signal generating node along the drilling direction, the two are linearly arranged, and the signal is received by the signal receiving node after being reflected by the rock stratum structure surface; the reflection path length of the signal is calculated respectively; a plane geometric model is constructed with the drilling route of a drilling machine and the rock stratum structure surface, the signal generating node and the signal receiving node are marked in the plane geometric model, the signal reflection path is calculated, and the optimal estimation value of the rock stratum structure surface inclination is calculated based on the drilling horizontal angle and the cosine theorem. Through linear multi-point collinear arrangement, dynamic matching of lithology wave velocity based on deep learning, spatial geometric simultaneous solution and Kalman filtering optimization, high-precision, strong-interference-resistant and real-time detection of the rock stratum structure surface inclination is realized, and reliable data support is provided for self-adaptive control of an unmanned drilling machine.

Description

Technical Field

[0001] This invention relates to the field of underground drilling measurement and intelligent detection technology in coal mines, specifically a method for detecting the dip angle of rock strata structure surfaces based on linear multi-point drilling sensing. Background Technology

[0002] The dip angle of the rock strata structure is a core parameter reflecting the underground geological conditions, determining the distribution of drilling resistance and the deformation pattern of the surrounding rock. It is a crucial basis for adjusting parameters such as drilling pressure and rotation speed. If the detection of the rock strata structure dip angle is inaccurate or delayed, it can easily lead to stuck drill pipe, borehole collapse, and in severe cases, damage to the integrity of the surrounding rock and cause safety accidents such as gas leaks. Therefore, accurate and real-time detection of the rock strata structure dip angle is a key requirement for ensuring the progress of pressure relief drilling and improving the level of intelligent gas control in coal mines.

[0003] In recent years, linear multi-point drilling sensing technology has been gradually applied to drilling detection scenarios in oil and gas drilling, tunnel drilling, and other projects due to its advantages of high detection resolution and strong data reliability, effectively solving the technical shortcomings of traditional single-point drilling sensing. However, the existing detection models of linear multi-point drilling sensing technology have not been specifically designed based on the structural characteristics of coal mine pressure relief drilling rigs, complex underground operating conditions, and the geological characteristics of coal mine strata. Problems exist such as the lack of scientific parameters for the deployment of the drilling system, the failure to dynamically match the electromagnetic wave velocity with the rock mass lithology, the coarse calculation model for the dip angle of the rock strata structure, and the lack of effective optimization of detection data. Therefore, it cannot be directly applied to the dip angle detection of coal mine pressure relief drilling rigs, and the detection accuracy and real-time performance do not meet the application requirements of unmanned drilling rigs. Summary of the Invention

[0004] In order to at least partially solve the problems existing in the prior art, the present invention provides a method for detecting the dip angle of rock strata structure based on linear multi-point drilling sensing, the technical solution of which is as follows.

[0005] The method for detecting the dip angle of rock strata structure surfaces based on linear multi-point drilling sensing includes the following steps:

[0006] S1, take the signal generation location during the drilling process of the drilling rig as the signal generation node, and set up ≥2 signal receiving devices on the drill rod of the drilling rig as signal receiving nodes;

[0007] Along the drilling direction, the signal receiving node is located behind the signal generating node and is arranged linearly with the signal generating node. The signal is reflected at the rock stratum structure surface in front of the drilling direction, and the signal receiving node can receive the reflected signal.

[0008] S2, calculate the reflection path length of the signal from the signal generating node to the signal receiving node respectively;

[0009] S3. Construct a planar geometric model based on the drilling route of the drilling rig and the rock strata structure surface. Mark the signal generation node and signal receiving node in the geometric model, draw the signal reflection path, and calculate the dip angle of the rock strata structure surface based on the drilling horizontal angle and the cosine theorem.

[0010] Furthermore, the signal is at least one of sound wave, vibration wave, and electromagnetic wave; the signal generation location is the location where the drill bit grinding vibration occurs, or the installation location of the drilling pipe equipped with a signal generator; the signal receiving device is a drilling pipe equipped with a signal receiver; the signal reflection path is calculated and determined based on the signal propagation speed in different rock masses, and the signal propagation speed is adaptively updated through the following steps:

[0011] S201 uses continuous wavelet transform to map a one-dimensional signal into a two-dimensional time-frequency diagram;

[0012] S202, input the two-dimensional time-frequency map into the MobileNetV3-EACS lithology identification model, and output the rock mass lithology results of the current drilling area;

[0013] S203, write a set of discrete function mappings representing the lithology-wave velocity correspondence into the long-term data storage module of the drilling-while-drilling pipe data processing system. The expression is: In the formula, V represents the set of lithological classification domains; V represents the set of calibration wave velocity value domains. The lithological results output by the MobileNetV3-EACS lithology identification model; To calibrate wave velocity; based on lithological results Automatically map and extract the corresponding calibration wave velocity This refers to the speed at which a signal propagates within the rock mass of the current drilling area.

[0014] Furthermore, mapping a one-dimensional signal to a two-dimensional time-frequency graph using continuous wavelet transform includes the following steps:

[0015] S201-1, taking the signal transmission time as the starting point, intercepts the original reflected signal received by the signal receiving node, obtains a one-dimensional time-domain reflected signal segment of fixed time length, and performs DC component removal and amplitude normalization preprocessing on the one-dimensional time-domain reflected signal segment.

[0016] S201-2, select Morlet wavelet as mother wavelet, perform continuous wavelet transform on the preprocessed one-dimensional time-domain reflection signal segment, and output wavelet coefficient matrix;

[0017] S201-3 maps the modulus of the wavelet coefficient matrix to grayscale or color image pixel values, with the horizontal axis representing time and the vertical axis representing scale, constructing a two-dimensional time-frequency graph, and uniformly adjusting the size of the two-dimensional time-frequency graph to a fixed pixel size and performing normalization processing.

[0018] S201-4, the generated two-dimensional time-frequency map is stored in the data cache module of the drilling pipe data processing system in matrix form for real-time access by the MobileNetV3-EACS lithology identification model.

[0019] Furthermore, the two-dimensional time-frequency map is input into the MobileNetV3-EACS lithology identification model to output the rock mass lithology results of the current drilling area, including the following steps:

[0020] S202-1 converts the two-dimensional time-frequency map into a standardized two-dimensional time-frequency map matrix, which serves as the input data for the MobileNetV3-EACS lithology identification model;

[0021] S202-2, the MobileNetV3-EACS lithology identification model uses MobileNetV3-Large as its backbone network. The MobileNetV3-Large network is composed of multiple stacked Bneck layers. Each Bneck layer contains depthwise separable convolutions, compressed excitation structures, and residual connections.

[0022] The two-dimensional time-frequency map is passed through each Bneck layer in sequence to gradually extract the lithological time-frequency features from the bottom texture to the high semantic level, and output a 7×7×48 feature map;

[0023] S202-3, in the last three bridge layers of the MobileNetV3-Large network, an ECA-s attention mechanism module is embedded. The specific implementation of the ECA-s attention mechanism module is as follows: First, the input feature map is subjected to global max pooling and global average pooling respectively to obtain two 1×1×C channel descriptors; the two 1×1×C channel descriptors are added element-wise and fused, and then input into an adaptive one-dimensional convolution with a kernel size of λ for cross-channel interaction; channel attention weights are generated through the Sigmoid activation function; the channel attention weights are multiplied with the feature map channel by channel for feature recalibration; where C is the number of channels, and the value of λ is adaptively determined by the number of channels C.

[0024] S202-4: The feature map enhanced by the ECA-s attention mechanism module is input into the classifier module. The classifier structure is as follows: Convolutional layer 1: 1×1 convolution with 48 output channels; Batch normalization layer; ReLU activation function; Global average pooling layer: compresses the 7×7×48 feature map into a 1×1×48 feature vector; Convolutional layer 2: 1×1 convolution with n output channels, where n is the pre-set total number of lithology categories, mapping the feature vector to the original score of each category; Softmax layer: converts the original score into a probability distribution, outputting an N-dimensional probability vector, where each component represents the probability that the current drilling rock mass belongs to the corresponding lithology category; The category index corresponding to the maximum value in the probability vector is taken as the lithology result of the current drilling area and output to the data cache module built into the drilling pipe for use in step S203.

[0025] Furthermore, the method for calculating the drilling horizontal angle is as follows:

[0026] S301, acquire the raw triaxial data collected by the 3-axis fluxgate sensor and the 3-axis MEMS accelerometer in the directional sensor of the drilling pipe data processing system, and use wavelet filtering algorithm to perform digital filtering processing on the raw X, Y, and Z axes of the 3-axis fluxgate sensor and the 3-axis MEMS accelerometer to obtain the correction values ​​of the 3-axis fluxgate sensor and the 3-axis MEMS accelerometer.

[0027] S302, using sensor data that has been smoothed and denoised by wavelet filtering, calculates the drilling horizontal angle according to the following formula. ,in, , , These are the triaxial correction values ​​for the accelerometer.

[0028] Furthermore, the method also includes the following steps:

[0029] S4, in the data cache module of the drilling pipe data processing system, uses a hierarchical and associative storage method based on a discrete algebra structure. Its specific data structure definition is as follows:

[0030] S401, the continuous sampling data of the sensor is defined as a set of discrete time sequence vectors that monotonically increase with time. ,in For discrete timestamps, and They represent The three-dimensional spatial vectors of the fluxgate and accelerometer at a given moment; Indicates the index number; Represents the set of positive integers;

[0031] S402, the results of wavelet filtering and the calculated drilling horizontal angle are defined as a multidimensional ordered tuple structure. Sequential construction and storage are performed, among which, This serves as a unique identifier for each drilling pipe. To drill at a horizontal angle;

[0032] S403, the storage addressing and calling logic in the data cache module of the drilling pipe data processing system adopts a timestamp-based injective function.

[0033] Furthermore, as the drill bit continues to approach the rock strata surface, multiple consecutive rock strata surface dip angles can be obtained. The optimal dip angle of the rock strata surface is then estimated using a Kalman filter model, and the optimal approximate value of the rock strata surface dip angle is output. The steps are as follows:

[0034] S501: The calculated dip angle values ​​of multiple consecutive rock strata are formed into a sequence. The arithmetic mean of the sequence is calculated. If the absolute value of the deviation between the calculated dip angle value of a certain rock strata in the sequence and the average value exceeds the preset threshold, the value is determined to be an outlier and is removed. The remaining valid rock strata dip angle values ​​after removal are arranged in chronological order to form a new input sequence for the Kalman filter model to call.

[0035] S502, the dip angle of the rock strata structure surface is used as the state variable of the Kalman filter model, and the measured value of the dip angle of the rock strata structure surface obtained by a single calculation is used as the observation.

[0036] In the S503 Kalman filter model, the range of values ​​for the process noise covariance matrix is ​​set to be... The range of values ​​for the observation noise covariance matrix is: The update step size for the Kalman gain is 1;

[0037] S504. Following the prediction and update iteration steps of Kalman filtering, multiple sets of measured values ​​of rock stratum structure surface dip angles are filtered and calculated to obtain the optimal approximate value of the rock stratum structure surface dip angle.

[0038] During the iterative calculation process, the storage of the filtered intermediate data adopts a dynamic finite sequence structure based on first-in-first-out logic. The data is overwritten in the data cache module with the iteration step number as the sequential index. Finally, only the optimal approximate estimate of the dip angle state of the rock stratum structure surface is retained and output.

[0039] Furthermore, the Kalman filter formula and filter parameters are pre-stored in the long-term data storage module of the drilling pipe data processing system, and read into the data cache module during drilling pipe initialization;

[0040] At each moment , Kalman gain and intermediate filtering values ​​are stored in the data cache module. The data storage structure adopts a queue structure and is stored sequentially according to the iteration order.

[0041] The intermediate data of the filtering is stored in the data cache module, and only the optimal approximate value of the dip angle of the rock stratum structure surface is retained;

[0042] The preset threshold is ±10°;

[0043] The optimal approximation of the dip angle of the rock stratum structure plane is stored in the data cache module, and the optimal approximation stored in the data cache module is used for real-time retrieval.

[0044] The optimal approximation of the dip angle is output in real time through the data transmission line of the drilling pipe and simultaneously transmitted to the drilling rig main control system, the downhole data acquisition terminal, and the ground monitoring center. The main control system uses this optimal approximation as the basis for adaptive adjustment of drilling parameters, and the ground monitoring center uses this optimal approximation to realize real-time display, historical query, and anomaly warning of the dip angle parameters of the rock strata structure.

[0045] Furthermore, the spacing between adjacent signal receiving nodes is set to 2m-3m; the deviation angle between the linear arrangement axis of the signal receiving nodes and the horizontal drilling direction of the drill bit is <10°.

[0046] Furthermore, the process of calculating the dip angle of the rock stratum structure surface based on the drilling horizontal angle and the cosine theorem in step S3 is implemented by the following geometric solution program based on sequential storage:

[0047] In the data cache module of the drilling pipe data processing system, a one-dimensional array is allocated to store intermediate results sequentially according to the calculation order.

[0048] Store the reflection path lengths of each signal receiving node calculated in step S2 into an array;

[0049] Based on the principle of mirror reflection, determine the distances AP' and BP' between the symmetrical point P' of the signal generating node P with respect to the rock stratum structure surface and the signal receiving nodes A and B, and store them in an array;

[0050] Based on triangle ABP', calculate ∠P'BA using the law of cosines and store the result in an array;

[0051] Based on ∠P'BA and the fixed distance PB between the signal generating node P and the signal receiving node B, the position parameters PD of the signal reflection point D corresponding to the signal receiving node B and its symmetrical point distance DP' are solved and stored in an array. The same method is used to solve the position parameters CP of the signal reflection point C corresponding to the signal receiving node A and its symmetrical point distance CP', and stored in an array.

[0052] Based on CP', DP' and ∠AP'B, calculate the spatial distance CD between signal reflection points C and D and store it in an array;

[0053] Calculate ∠P'DC based on triangle P'DC;

[0054] Using the geometric relationship ∠P'DC = α + ∠P'BA, the dip angle α of the rock stratum structure surface is calculated and stored in an array;

[0055] The α output is fed into the Kalman filter model in step S5.

[0056] Beneficial effects

[0057] 1. This invention adopts a linear multi-point deployment scheme, which establishes the physical basis for multi-point drilling detection by arranging the signal generator and receiver collinearly along the drill pipe (see step S1). By optimizing the deployment parameters, the distance between the signal generator and the drill bit is set to 1m-8m, the distance between adjacent signal receivers is set to 2m-3m, and the deviation angle tolerance of the collinear arrangement is controlled to <10°. This ensures that the reflected signal has sufficient time resolution and signal-to-noise ratio, while also fully considering actual working conditions such as downhole drill pipe bending and installation tolerances, making the technical solution described in this invention highly feasible in the field.

[0058] 2. To address the problem of large ranging errors caused by the fixed wave velocity in traditional methods, this invention introduces Continuous Wavelet Transform (CWT) and the MobileNetV3-EACS lithology identification model (see steps S201-S202) to convert the one-dimensional reflection signal into a two-dimensional time-frequency diagram and output lithology classification in real time; dynamic matching of wave velocity is achieved through a pre-stored set of "lithology-wave velocity" discrete function mappings (see step S203). This method enables the signal propagation speed to be adaptively updated according to the density of the rock mass ahead, reducing the ranging error caused by medium inhomogeneity from the physical source.

[0059] 3. To address the unsteady high-frequency strong vibration interference generated during rock-breaking drilling by the pressure relief drilling rig, this invention introduces a wavelet filtering algorithm (see step S301) at the front end of the drill bit attitude angle calculation to perform time-frequency domain denoising on the original triaxial data of the fluxgate and accelerometer. The filtered sensor data effectively smooths the attitude angle baseline, solves the problem of structural surface inclination angle calculation distortion caused by severe sensor vibration, and significantly improves the data availability under strong vibration conditions.

[0060] 4. This invention constructs a discrete-time series vector set and a multi-dimensional ordered tuple structure in the data caching module (see step S4). Difference calculation and angle calculation are achieved through a timestamp-based injective function, and the calculation results are stored, accessed by address, and invoked by the program. This design ensures a strict correspondence between the original feature vector and the derived angle tuple under the same time slice, greatly reducing the problem of excessive asynchronous data error in multi-node data interaction scenarios. Simultaneously, it reduces the data access time complexity to O(1), lowering the memory overhead of the embedded system.

[0061] 5. This invention inputs the rock strata dip angle values ​​obtained from multiple consecutive calculations into a Kalman filter model (see steps S501-S503), smoothing instantaneous measurement noise through state prediction and iterative updates; simultaneously, outlier removal preprocessing (see step S500) filters out obvious abnormal data, and utilizes a dynamic finite sequence structure to continuously store intermediate filtered values. The final output optimal approximate dip angle effectively smooths out measurement fluctuations while maintaining a response delay of one-tenth of a second, providing reliable geological feedforward data for the adaptive adjustment of drilling parameters of unmanned pressure relief drilling rigs.

[0062] 6. Comparative Analysis of the Effects of the Examples

[0063] The dip angle of the coal and rock structural surface was used to conduct laboratory tests on the model to determine the dip angle detection accuracy. The method of this invention was compared with the traditional two-time single-point constant wave velocity detection method and the single-time unfiltered multi-point detection method. Qualitative and quantitative analyses were performed to obtain three core evaluation indicators: mean absolute error of dip angle, data availability under strong vibration environment, and single-time solution response delay. The comparison results are shown in the table below:

[0064]

[0065] In the table: (1) Mean absolute error of tilt angle is used to measure the accuracy of the model in calculating the tilt angle of the structural surface. The unit is degrees. The closer the value is to 0, the better the detection effect. (2) Data availability under strong vibration environment measures the model's ability to resist noise interference under complex working conditions. The range is from 0% to 100%. The closer the value is to 100%, the better the anti-interference effect. (3) Single solution response delay is used to measure the model's ability to process data in real time. The unit is milliseconds. The smaller the value, the stronger the real-time performance.

[0066] As can be seen from the table, the data availability rate of the method of the present invention under strong vibration environment is significantly higher than that of the traditional single-point constant wave velocity detection method and the unfiltered multi-point detection method, and the mean absolute error of the tilt angle is also significantly smaller; the mean absolute error of the tilt angle is reduced by 5.20° compared with the traditional single-point constant wave velocity detection method and by 2.87° compared with the original unfiltered multi-point detection method; the data availability rate under strong vibration environment is improved by 36.3% compared with the traditional single-point constant wave velocity detection method and by 26.9% compared with the original unfiltered multi-point detection method; although the single solution response delay is slightly increased compared with the traditional method, the delay of 210ms fully meets the engineering requirements of millisecond-level adaptive control of unmanned pressure relief drilling rig, which fully demonstrates the effectiveness of the method of the present invention. Attached Figure Description

[0067] Figure 1 This is a schematic diagram of the overall process of the method of the present invention;

[0068] Figure 2 This is a schematic diagram of the spatial geometry principle of reflection ranging in multi-point drilling detection; in the figure: point P is the location where the signal is generated; points A and B are the locations of the signal receivers, respectively; point O is the intersection of the extension line of the drill bit's drilling direction and the structural surface; points C and D are the reflection points corresponding to the strongest reflected signals received by signal receivers A and B;

[0069] Figure 3 This is a flowchart of lithology identification and dynamic wave velocity matching based on the MobileNetV3-EACS model. Detailed Implementation

[0070] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0071] The following description of at least one exemplary embodiment is merely illustrative and is not intended to limit the invention or its application or use in any way.

[0072] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0073] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0074] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0075] In all examples shown and discussed herein, any specific value should be interpreted as merely exemplary and not as a limitation; therefore, other examples of exemplary embodiments may have different values.

[0076] Unless otherwise specified, the products or equipment described in this embodiment are all from commercial channels.

[0077] During horizontal directional drilling with a pressure-relief drilling rig, the angle between the rock mass structural plane and the drilling direction of the drill bit is defined as the dip angle of the rock stratum structural plane. Figure 2 An illustration is provided, in the diagram. Dip angle of rock strata structure plane.

[0078] This embodiment describes a method for detecting the dip angle of rock strata structure surfaces based on linear multi-point drilling sensing. Figure 1 The flowchart of this method is given, which includes the following steps.

[0079] S1, When the drilling rig is drilling, the drill bit grinds the rock and / or coal seam, which can emit sound wave signals and / or vibration wave signals. In this embodiment, the location where the drill bit grinding vibration occurs is defined as the signal generation node.

[0080] Preferably, a drilling pipe is installed on the drill pipe, which is equipped with a signal generator that can emit electromagnetic wave signals. Therefore, in this embodiment, the location of the drilling pipe can also be defined as a signal generation node.

[0081] At least two down-the-hole pipes are installed on the drill rod of the drilling rig, in a direction away from the drill bit. Each down-the-hole pipe is equipped with a signal receiver, and the location of the signal receiver is defined as a signal receiving node. That is, in this embodiment, at least two signal receiving nodes are set on the drill rod of the drilling rig, including a first signal receiving node and a second signal receiving node, etc., and the signal receiving nodes and the signal generating nodes are arranged linearly in the drilling direction of the drilling rig.

[0082] The signal emitted by the signal generating node is reflected after reaching the rock stratum surface located in front of the drilling direction and is received by the signal receiving node. In this embodiment, to obtain better signal reception, the following is implemented:

[0083] ① The distance between the signal generator and the drill bit in the drilling pipe located at the signal generation node is set to 1m-8m, preferably 3m.

[0084] Because a strong near-field induction field is generated at the moment of signal transmission, the signal receiver at the signal receiving node may initially fail to effectively identify the valid reflected echo from the rock interface. Placing the signal generator 1m-8m behind the drill bit ensures that the signal source is in close contact with the rock mass being probed, effectively avoiding direct coupling interference from the near field of the signal receiver.

[0085] Meanwhile, a distance of 1m-8m ensures that the processing system in the drill pipe has sufficient time to complete at least one full cycle of the "signal transmission → reflection reception → data processing" process before the drill bit contacts the rock structure surface. If the distance is too small (e.g., <1m), the physical space left for the processing system to perform calculations and for the drilling rig to execute adaptive control is too short, making it difficult to effectively avoid the risk of stuck drill bit; if the distance is too large (e.g., >8m), the signal attenuation will increase over long distances, and the dip angle of the rock structure surface calculated subsequently may be distorted due to the uneven changes in lithology over long distances.

[0086] ② The spacing between adjacent signal receiving nodes is set to 2m-3m, preferably 3m.

[0087] Too small a distance (<2m) will result in a small time difference between the two receiving points, which is easily overwhelmed by sampling errors, reducing ranging accuracy. Furthermore, too small a distance (<2m) will cause discrepancies in the calculations between the two points. Figure 2 The corresponding points A and B approximately coincide, causing multiple points to degenerate into a single point, making it impossible to constrain the spatial position of the reflection point.

[0088] If the spacing is too large (>3m), it will destroy the collinearity premise and introduce model errors. A spacing of 2m-3m can construct a set of nonlinear equations with good performance and achieve stable solution of the dip angle of the rock stratum structure.

[0089] ③ This embodiment does not require the signal receiving node and the signal generating node to be arranged in a strict linear manner. Instead, it requires that their axes be approximately collinear with the horizontal drilling direction of the drill bit, with a deviation angle of 0-20 degrees, preferably <10°, to meet the objective operational needs.

[0090] The reason for setting this angle is that the drill rod will inevitably bend and deform under stress. Strictly requiring absolute collinearity is impossible to achieve in actual operation. The range of 0-20° covers the axis deflection that may occur under normal drilling conditions, ensuring that the solution described in this embodiment is engineering feasible.

[0091] Meanwhile, the tilt angle calculation model of the present invention is constructed based on the collinearity assumption (see below). When the deviation angle is <10°, the model error introduced by non-collinearity is less than 0.1°, and its impact on the final tilt angle calculation result is negligible. If the deviation angle exceeds 20°, the non-collinearity error will be amplified to an unacceptable level, causing the calculation to fail.

[0092] ④ The data processing system of the drilling pipe mainly includes sensor components, a long-term data storage module, and a data cache module, all integrated and packaged inside the drilling pipe casing. The sensor components include electromagnetic / acoustic transceivers for transmitting and receiving detection signals. The long-term data storage module persistently stores initialization data and computational programs such as detection calibration parameters, calculation formulas, and deep learning model parameters. The data cache module temporarily stores real-time acquired data and intermediate data from real-time calculations. The drilling pipe also features an orientation sensor for attitude measurement, orthogonally assembled from a 3-axis fluxgate sensor and a 3-axis MEMS accelerometer. The drilling pipe casing is made of a non-magnetic, pressure-resistant alloy material to shield against downhole electromagnetic interference and withstand drilling pressure. This embodiment uses commercially available drilling pipes, such as Halliburton's SlimPulse MWD pipe, Shanghai Shenkai's SK-TMWD wireless measurement-while-drilling pipe, and Taiyuan Siweifu's YSX2000 mining measurement-while-drilling pipe, all of which meet the requirements of this embodiment.

[0093] ⑤ When there is a lithological interface in front of the drill bit, the signal receiver of the signal receiving node collects the reflection signal of the current drilling rock in real time, and further filters out the strongest reflection signal as the reflection signal through a filter.

[0094] Filters are classified into acoustic filters and electromagnetic signal filters according to the type of detected signal. The working principle is briefly described as follows: the filtering frequency range of acoustic filters is 5Hz-50000Hz, and the filtering frequency range of electromagnetic signal filters is 1kHz-6GHz. The strongest reflected signal is extracted using the signal strength screening method, that is, the signal with the largest amplitude among the signals with reflected signal amplitude greater than the average signal strength at a specific frequency is selected as the effective reflected signal. Under similar background noise conditions, the amplitude threshold of the effective reflected signal is 1.01-5 times the amplitude of the rock mass background noise signal. For example, the strongest reflected echo is extracted by setting the amplitude threshold of the electromagnetic signal filter to 2.5 times the background noise. The filtering parameters of the filter are first obtained by rock mass experiments and pre-stored in the long-term data storage module of the drilling pipe. During initialization, they are read into the data cache module and can be adjusted in real time according to the downhole conditions.

[0095] S2, defines the time at which the signal-generating node emits a signal as... The time when the first signal receiving node receives the reflected signal is The time when the second signal receiving node receives the reflected signal is The speed at which a signal propagates in the rock mass is defined as... ,but:

[0096] The length of the reflection path from the signal to the first signal receiving node (i.e.) Figure 2 "in ")for ;

[0097] The length of the reflection path from the signal transmission point to the second signal receiving node (i.e., Figure 2 "in ")for .

[0098] signal propagation speed in rock mass Determine through the following steps:

[0099] S201 uses continuous wavelet transform (CWT) to map the acquired one-dimensional signal into a two-dimensional time-frequency diagram to characterize the nonlinear characteristics of the signal in the time-frequency domain.

[0100] The specific methods or steps are as follows:

[0101] S201-1, based on signal transmission time Starting from the first and second signal receiving nodes, the original reflected signals are intercepted to obtain an effective reflected signal segment of fixed time length; the signal segment is then preprocessed to remove DC components and normalize amplitude.

[0102] S201-2, selecting the Morlet wavelet as the mother wavelet for the preprocessed one-dimensional time-domain signal. The formula for performing continuous wavelet transform is as follows:

[0103]

[0104] in, This is the scale factor (corresponding to frequency information). This is the translation factor (corresponding to time information). It is the conjugate of the mother wavelet.

[0105] By traversing the set scale range Translation range The wavelet coefficient matrix is ​​calculated. .

[0106] S201-3, wavelet coefficient matrix The modulus value is mapped to grayscale or color image pixel values, with the horizontal axis representing time. The vertical axis represents the scale. (Can be converted to frequency) to construct a two-dimensional time-frequency graph; adjust the size of the time-frequency graph to a fixed pixel size (e.g., 224×224) and normalize it to adapt it to the input requirements of subsequent deep learning models.

[0107] S201-4, the generated two-dimensional time-frequency map is temporarily stored in the data cache module of the drilling pipe in matrix form for real-time use by the MobileNetV3-EACS lithology identification model in step S202.

[0108] S202, input the two-dimensional time-frequency map into the pre-trained MobileNetV3-EACS lithology identification model, and accurately output the lithology classification results of the current drilling area.

[0109] The specific methods or steps are as follows:

[0110] S202-1, The standardized two-dimensional time-frequency map matrix with a size of 224×224×3 generated in step S201 is used as the input data for the MobileNetV3-EACS lithology identification model.

[0111] S202-2, the MobileNetV3-EACS lithology identification model uses MobileNetV3-Large as its backbone network, which is composed of multiple stacked bottleneck structures;

[0112] Each Bneck contains depthwise separable convolutions, compressed excitation structures, and residual connections;

[0113] The two-dimensional time-frequency map is passed through each Bneck layer in sequence to gradually extract the lithological time-frequency features from the bottom texture to the high semantic level. The final output feature map size is 7×7×960.

[0114] S202-3 embeds an improved ECA-s (Efficient Multi-channel Attention) attention mechanism module in the last three Bneck structures of the backbone network.

[0115] The ECA-s module is implemented as follows: First, the input feature map is subjected to global max pooling and global average pooling respectively to obtain two 1×1×C channel descriptors; the two are added element-wise and fused, and then input into an adaptive one-dimensional convolution with a kernel size of λ for cross-channel interaction (the value of λ is adaptively determined by the number of channels C); channel attention weights are generated through the Sigmoid activation function; the weights are multiplied with the original input feature map channel by channel to achieve feature recalibration.

[0116] The ECA-s mechanism effectively expands the receptive field by combining the advantages of maximum pooling and average pooling, enhancing the model's sensitivity to subtle differences in time-frequency plots among different lithologies.

[0117] S202-4, input the feature map enhanced by ECA-s into the classifier module. The classifier structure is as follows: (1) Convolutional layer 1: 1×1 convolution, with 48 output channels, used to reduce the feature dimension while retaining key information and avoiding information loss; (2) Batch normalization layer: accelerates training convergence; (3) ReLU activation function: introduces nonlinearity; (4) Global average pooling layer: compresses the 7×7×48 feature map into a 1×1×48 feature vector; (5) Convolutional layer 2: 1×1 convolution, with N output channels (N is the total number of pre-set lithology categories), maps the feature vector to the original score of each category; (6) Softmax layer: converts the original score into a probability distribution, outputs an N-dimensional probability vector, and each component represents the probability that the current drilled rock mass belongs to the corresponding lithology category.

[0118] Lithology determination: The category index corresponding to the maximum value in the probability vector is taken as the lithology classification result for the current drilling area. The data is output to the data cache module for use in step S203.

[0119] S203, In the long-term data storage module of the drilling pipe, a set of discrete function mappings representing the correspondence between "lithology and wave velocity" is pre-stored. ,in, Let V be the set of domains defining lithology classification, and V be the set of calibrated wave velocity value domains. Lithology identified based on the MobileNetV3-EACS lithology identification model. Automatically map and extract the corresponding calibration wave velocity The speed of signal propagation in rock mass .

[0120] For example, if the MobileNetV3-EACS lithology identification model identifies the lithology of the current drilling rock mass as "dense fine sandstone", and the calibration wave velocity corresponding to dense fine sandstone in the discrete mapping set is 3850 m / s, then 3850 m / s will be used as the signal propagation speed in the current rock mass.

[0121] In this embodiment, preferably, when the wave velocity difference in the signal propagation speed in the rock mass exceeds 10%, the wave velocity is recalculated; the calibrated wave velocity is obtained by indoor core wave velocity testing or on-site drilling wave velocity calibration; the on-site drilling wave velocity calibration method performs wave velocity correction once every 0.1m to 10m of drilling.

[0122] S3, calculate the drilling horizontal angle of the drilling rig. Based on the reflection path length of the signal output from step S2 to the first / second signal receiving node, the fixed distance between the signal generating node and the signal receiving node, calculate the dip angle of the rock stratum structure surface according to the cosine theorem.

[0123] The calculation method or steps for the drilling horizontal angle are as follows:

[0124] S301, acquire the raw three-axis data collected by the 3-axis fluxgate sensor and the 3-axis MEMS accelerometer in the orientation sensor; use wavelet filtering algorithm to perform digital filtering processing on the raw X, Y, and Z axis data of the 3-axis fluxgate sensor and the 3-axis MEMS accelerometer respectively to obtain the correction values ​​of the 3-axis fluxgate sensor and the 3-axis MEMS accelerometer.

[0125] S302, using sensor data that has been smoothed and denoised by wavelet filtering, calculates the drilling horizontal angle according to the following formula. ,in, , , These are the triaxial correction values ​​for the accelerometer.

[0126] Calculated drilling horizontal angle The data is stored in the data cache module (see step S4 for the specific data structure) and used as a known parameter for the subsequent geometric calculation of the dip angle of the rock stratum structure surface.

[0127] S4, the data cache module along with the drilling pipe, uses a hierarchical and associative storage method based on a discrete algebra structure. Its specific data structure definition is as follows:

[0128] S401, in the drilling casing, the continuously sampled data from the sensors are defined as a set of discrete-time sequence vectors that are strictly monotonically increasing with time. ,in, Discrete timestamps; and They represent The three-dimensional spatial vectors of the fluxgate and accelerometer at a given moment; This represents the index number, i.e., the sequential number of the discrete sampling points; It represents the set of positive integers (i.e., 1, 2, 3, ...).

[0129] S402, the wavelet filtering and the calculated angle results are defined as a multidimensional ordered tuple structure. Sequential construction and storage are performed, where for This serves as a unique identifier for the signal generating node, the first reflected signal receiving node, and the second reflected signal receiving node. To drill at a horizontal angle.

[0130] In step S403, the storage addressing and retrieval logic in the data caching module adopts a timestamp-based injective function to ensure that the original feature vector and the derived angle tuple under any time slice have strict correspondence and unique reachability in the storage space. In the subsequent step S5, when performing Kalman filtering and equation solving, the main control chip does not need to perform a full table traversal; it only needs to calculate the hash offset of the injective function. When the main control chip performs Kalman filtering iterations in step S5, it uses the current timestamp... The corresponding attitude angle is retrieved directly from the cache using this injective function. and original vector , This ensures the real-time nature of state estimation and data consistency. It allows for the accurate extraction of three-dimensional spatial vectors and angle parameters at a specific drilling moment in O(1) time complexity, completely eliminating the drift problem in structural surface dip angle calculation caused by asynchronous data from multiple nodes.

[0131] Meanwhile, to support the Kalman filter iteration in step S5, the data caching module also constructs a dynamic finite sequence based on first-in-first-out logic for rolling storage of intermediate filter values.

[0132] S5. As the drill bit of the drilling rig continuously approaches the rock structure surface during the dynamic drilling process, the drilling pipe at the first / second signal receiving node continuously and in real time collects data, outputs multiple rock structure surface dip angle values, uses the Kalman filter model to make the optimal estimate, and outputs the optimal approximate value of the rock structure surface dip angle.

[0133] The optimization calculation process of the Kalman filter model in this step is as follows:

[0134] S501, Input Data Preprocessing

[0135] The consecutive tilt angle calculation values ​​stored in step S4 are arranged into a sequence, and the arithmetic mean of the sequence is calculated. If the absolute value of the deviation between a certain tilt angle value in the sequence and the average value exceeds a preset threshold (preferably ±10° in this embodiment), the value is determined to be an outlier and is removed. The remaining valid tilt angle values ​​after removal are arranged in chronological order to form a new input sequence for use by the Kalman filter model.

[0136] S502, Model Setting

[0137] The dip angle of the rock strata structure surface is used as the state variable of the Kalman filter model, and the measured value of the dip angle of the rock strata structure surface obtained from a single calculation is used as the observation. The Kalman filter model includes state equations and observation equations, which are as follows:

[0138]

[0139]

[0140] In the formula, for State estimate at time 1; This is the state transition matrix (with a value of 1). for State estimate at time 1; This is process noise; for The observed value at time; The observation matrix (with a value of 1); To observe noise.

[0141] S503, Filter Parameter Settings

[0142] In the Kalman filter model, the process noise covariance matrix Q ranges from 10⁻. 6 The range of the observation noise covariance matrix R is ~10⁻². 5 ~10⁻¹, the update step size of the Kalman gain is 1.

[0143] S504, Solving for the optimal value

[0144] Following the prediction and update iteration steps of Kalman filtering, the measured values ​​of the dip angles of multiple rock strata are filtered and calculated to obtain the optimal approximation of the dip angles of the rock strata.

[0145] During the iterative calculation process, the intermediate filtered data is stored using a dynamic finite sequence structure based on first-in-first-out logic. The data is overwritten in the data cache module with the iteration step number as the sequential index, and finally only the optimal approximate estimate of the dip angle state of the rock strata structure is retained and output.

[0146] By using a dynamic finite sequence rolling over mechanism, the data caching module only needs to retain a finite set of intermediate data required for the current iteration step, without storing all historical data. This significantly reduces memory usage while ensuring filtering accuracy, thus meeting the resource constraints of the embedded drilling pipe.

[0147] In this embodiment,

[0148] ① The relevant formulas and filtering parameters of Kalman filtering are pre-stored in the long-term data storage module of the drilling pipe, and read into the data cache module during the initialization of the drilling pipe;

[0149] ②At each time point , Kalman gain and filter intermediate values ​​are stored in the data cache module. The data storage structure adopts a queue structure and is stored sequentially according to the iteration order. The data type is 32-bit single precision.

[0150] ③ The intermediate data of the filtering is stored in the data cache module, and only the optimal approximate value of the dip angle of the rock stratum structure is retained.

[0151] ④ Before inputting the dip angle values ​​of multiple rock strata structural planes into the Kalman filter model, outlier removal preprocessing is performed. The outlier judgment criterion is that if the deviation of a single dip angle calculation value from the average of all dip angle calculation values ​​exceeds ±10°, it is judged as an outlier and removed. After removing outliers, the remaining valid dip angle values ​​are input into the Kalman filter model for iterative calculation in the order of acquisition time.

[0152] ⑤ The optimal approximate value of the dip angle of the rock strata structure is stored in the data cache module of the drilling system. This optimal approximate value stored in the data cache module is used for real-time retrieval. This optimal approximate value of the dip angle is output in real time through the drilling data transmission line and simultaneously transmitted to the drilling rig main control system, the downhole data acquisition terminal, and the surface monitoring center. The main control system uses this optimal approximate value as the core basis for adaptive adjustment of drilling parameters, and the surface monitoring center uses this optimal approximate value to realize real-time display, historical query, and anomaly warning functions for the dip angle parameters of the rock strata structure.

[0153] ⑥ For every 0.01m to 1m drilled, the second / third drilling pipe collects 1 to 1000 reflection signals and calculates the dip angle of the rock stratum structure. The effective acquisition time of a single acquisition is between 0.0001s and 5s. 1 to 200 sets of reflection signals are acquired during each acquisition process, and the signal reception time of the signal receiver is used as the basis for calculation.

[0154] ⑦ The arrival time data of the collected reflected signals is stored in the data cache module. The calculated values ​​such as the inclination angle, drilling depth, and acquisition time of each group are also stored in the data cache module and output to the main control system of the drilling rig.

[0155] The following is in conjunction with the appendix Figure 2 This embodiment provides a calculation example of the process of calculating the dip angle of the rock stratum structure surface in step S3. Obviously, operators can also use other methods or ideas to calculate the dip angle of the rock stratum structure surface, but the calculation example provided in this embodiment can greatly simplify the calculation process.

[0156] Step 1, in time The signal generator sends a signal or the drill bit begins grinding. The signal receivers located at signal receiving nodes A and B are respectively at... and The reflected signal was received. The distance from the signal emission point P to the signal reception points A and B can be determined based on the signal propagation speed within the rock mass. With time difference and Calculate the signal reflection path length , ;

[0157] Step 2: Based on the principle of mirror reflection, define the point P, the location of the signal generator or drill bit, as the symmetrical point relative to the rock stratum structure surface. The symmetrical point for and The intersection of the extensions; satisfying , ;according to (Read the pre-stored data) , The length is obtained by solving the angle using the trigonometric function cosine theorem. radian , (Pre-stored formula) and and Angular radius;

[0158] Step 3, at an angle , and Based on length, combined The relationship, constructed using the Law of Cosines, contains a Equations with unknown length:

[0159]

[0160]

[0161] The equation is pre-stored in memory; solving the equation yields... Length and The value is obtained using the same method. The value of .

[0162] The specific calculation process is as follows:

[0163] set up ,but ,exist In the middle, by the Law of Cosines, we get Substitute , to obtain A quadratic equation with unknowns Solving the equation yields The value is then obtained. , The numerical value; using the same method, construct the cosine theorem equation and solve it to obtain... , The value.

[0164] Step 4, according to , , (Reading each value) is obtained using the Law of Cosines. Find the value of and then solve using the Law of Cosines. radian, due to Then the tilt angle of the structural surface is obtained. .

[0165] The specific calculation process is as follows:

[0166] Solve Length, in In the middle, by the Law of Cosines, we get:

[0167]

[0168] Then there is,

[0169]

[0170] In the formula, for The radian value, ;

[0171] , The length of the line segment is known.

[0172] Solve radian value ,exist In the middle, by the Law of Cosines, we get:

[0173]

[0174]

[0175] Solving for the dip angle of the rock strata structure Using geometric relationships ,have to

[0176] In the formula, for The radian value, for The radian value was calculated. The value is in radians and can be converted to angles as needed.

[0177] Furthermore, the process of calculating the dip angle of the rock strata structure plane in step S3 is encapsulated into a calculation program that can be directly called by the drilling-while-drilling data processing system. The specific execution flow is as follows:

[0178] S6, Execution flow of geometry solution program based on sequential storage

[0179] S601, allocate a one-dimensional array in the data cache module of the drilling pipe. This array is used to store the intermediate results of each step in the order of calculation. The array elements are of type 32-bit single-precision floating point and the index starts from 1.

[0180] In this embodiment, for (Array) is an abbreviation used in program execution flow. This represents an array containing 13 storage units, used to store intermediate results generated during the calculation process in sequence.

[0181] S602, Read the reflection path length calculated in step S2 (correspond Figure 2 (PC+AC in the middle), save Read the reflection path length (correspond Figure 2 (PD+BD in the middle), store ;

[0182] S603, based on the principle of specular reflection, makes Deposit ; Deposit ;

[0183] Read from a fixed distance from the long-term data storage module and ;

[0184] Calculate the angle using the Law of Cosines formula :

[0185]

[0186] Will deposit .

[0187] S604, from Read ,from Read Read from the long-term data storage module ;

[0188] Calculate using the simplified formula derived in step S3 (Right now ):

[0189]

[0190] Will deposit , i.e., PD;

[0191] calculate Deposit ;

[0192] calculate (Symmetry), store ;

[0193] Calculate using the same method (based on the symmetry of triangle PAC) and :

[0194]

[0195] Will deposit , deposit (in This is a fixed distance from the signal generating node P to the signal receiving node A, pre-stored in the long-term storage module of the drilling pipe data processing system.

[0196] S605, from Read ,from Read Read data from a fixed distance from the long-term data storage module. ;

[0197] Calculate the included angle (Right now (radian value)

[0198]

[0199] Will deposit .

[0200] from Read ,from Read ,from Read ;

[0201] exist Calculation using the Law of Cosines :

[0202]

[0203] Will deposit .

[0204] S606, from Read ,from Read ,from Read ;

[0205] exist In the middle, calculation arc :

[0206]

[0207] Using geometric relationships Calculate the dip angle of the rock strata structure plane. :

[0208]

[0209] Will deposit And convert it to an angle value as needed.

[0210] S607, In The value is sent to the Kalman filter module in step S5 as the calculated tilt angle value at the current moment. After completing one round of calculation, the array can be cleared or overwritten using a first-in-first-out (FIFO) approach for the next round of calculation. Throughout the process, each intermediate result is stored sequentially in the array, and subsequent steps strictly rely on the previously stored values, ensuring the accuracy of the calculation logic and data consistency. Simultaneously, sequential access enables efficient memory management and real-time calculation.

[0211] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A method for detecting the dip angle of rock strata structure surfaces based on linear multi-point drilling sensing, characterized in that, Includes the following steps: S1, take the signal generation location during the drilling process of the drilling rig as the signal generation node, and set up ≥2 signal receiving devices on the drill rod of the drilling rig as signal receiving nodes; Along the drilling direction, the signal receiving node is located behind the signal generating node and is arranged linearly with the signal generating node. The signal is reflected at the rock stratum structure surface in front of the drilling direction, and the signal receiving node can receive the reflected signal. S2, calculate the reflection path length of the signal from the signal generating node to the signal receiving node respectively; S3. Construct a planar geometric model based on the drilling route of the drilling rig and the rock strata structure surface. Mark the signal generation node and signal receiving node in the geometric model, draw the signal reflection path, and calculate the dip angle of the rock strata structure surface based on the drilling horizontal angle and the cosine theorem.

2. The method for detecting the dip angle of rock strata structure surfaces based on linear multi-point drilling sensing according to claim 1, characterized in that, The signal is at least one of sound wave, vibration wave, and electromagnetic wave; the signal generation location is the location where the drill bit grinding vibration occurs, or the installation location of the drilling pipe equipped with a signal generator; the signal receiving device is the drilling pipe equipped with a signal receiver. The signal reflection path is calculated and determined based on the signal propagation velocity in different rock masses, and the signal propagation velocity is adaptively updated through the following steps: S201 uses continuous wavelet transform to map a one-dimensional signal into a two-dimensional time-frequency diagram; S202, input the two-dimensional time-frequency map into the MobileNetV3-EACS lithology identification model, and output the rock mass lithology results of the current drilling area; S203, write a set of discrete function mappings representing the lithology-wave velocity correspondence into the long-term data storage module of the drilling-while-drilling pipe data processing system. The expression is: In the formula, V represents the set of lithological classification domains; V represents the set of calibration wave velocity value domains. The lithological results output by the MobileNetV3-EACS lithology identification model; To calibrate wave velocity; based on lithological results Automatically map and extract the corresponding calibration wave velocity This refers to the speed at which a signal propagates within the rock mass of the current drilling area.

3. The method for detecting the dip angle of rock strata structure surfaces based on linear multi-point drilling sensing according to claim 2, characterized in that, Mapping a one-dimensional signal to a two-dimensional time-frequency graph using continuous wavelet transform includes the following steps: S201-1, taking the signal transmission time as the starting point, intercepts the original reflected signal received by the signal receiving node, obtains a one-dimensional time-domain reflected signal segment of fixed time length, and performs DC component removal and amplitude normalization preprocessing on the one-dimensional time-domain reflected signal segment. S201-2, select Morlet wavelet as mother wavelet, perform continuous wavelet transform on the preprocessed one-dimensional time-domain reflection signal segment, and output wavelet coefficient matrix; S201-3 maps the modulus of the wavelet coefficient matrix to grayscale or color image pixel values, with the horizontal axis representing time and the vertical axis representing scale, constructing a two-dimensional time-frequency graph, and uniformly adjusting the size of the two-dimensional time-frequency graph to a fixed pixel size and performing normalization processing. S201-4, the generated two-dimensional time-frequency map is stored in the data cache module of the drilling pipe data processing system in matrix form for real-time access by the MobileNetV3-EACS lithology identification model.

4. The method for detecting the dip angle of rock strata structure surfaces based on linear multi-point drilling sensing according to claim 2, characterized in that, Input the two-dimensional time-frequency map into the MobileNetV3-EACS lithology identification model, and output the rock mass lithology results of the current drilling area, including the following steps: S202-1 converts the two-dimensional time-frequency map into a standardized two-dimensional time-frequency map matrix, which serves as the input data for the MobileNetV3-EACS lithology identification model; S202-2, the MobileNetV3-EACS lithology identification model uses MobileNetV3-Large as its backbone network. The MobileNetV3-Large network is composed of multiple stacked Bneck layers. Each Bneck layer contains depthwise separable convolutions, compressed excitation structures, and residual connections. The two-dimensional time-frequency map is passed through each Bneck layer in sequence to gradually extract the lithological time-frequency features from the bottom texture to the high semantic level, and output a 7×7×48 feature map; S202-3, in the last three bridge layers of the MobileNetV3-Large network, an ECA-s attention mechanism module is embedded. The specific implementation of the ECA-s attention mechanism module is as follows: First, the input feature map is subjected to global max pooling and global average pooling respectively to obtain two 1×1×C channel descriptors; the two 1×1×C channel descriptors are added element-wise and fused, and then input into an adaptive one-dimensional convolution with a kernel size of λ for cross-channel interaction; channel attention weights are generated through the Sigmoid activation function; the channel attention weights are multiplied with the feature map channel by channel for feature recalibration; where C is the number of channels, and the value of λ is adaptively determined by the number of channels C. S202-4: The feature map enhanced by the ECA-s attention mechanism module is input into the classifier module. The classifier structure is as follows: Convolutional layer 1: 1×1 convolution with 48 output channels; Batch normalization layer; ReLU activation function; Global average pooling layer: compresses the 7×7×48 feature map into a 1×1×48 feature vector; Convolutional layer 2: 1×1 convolution with n output channels, where n is the pre-set total number of lithology categories, mapping the feature vector to the original score of each category; Softmax layer: converts the original score into a probability distribution, outputting an N-dimensional probability vector, where each component represents the probability that the current drilling rock mass belongs to the corresponding lithology category; The category index corresponding to the maximum value in the probability vector is taken as the lithology result of the current drilling area and output to the data cache module built into the drilling pipe for use in step S203.

5. The method for detecting the dip angle of rock strata structure surfaces based on linear multi-point drilling sensing according to claim 2, characterized in that, The method for calculating the drilling horizontal angle is as follows: S301, acquire the raw triaxial data collected by the 3-axis fluxgate sensor and the 3-axis MEMS accelerometer in the directional sensor of the drilling pipe data processing system, and use wavelet filtering algorithm to perform digital filtering processing on the raw X, Y, and Z axes of the 3-axis fluxgate sensor and the 3-axis MEMS accelerometer to obtain the correction values ​​of the 3-axis fluxgate sensor and the 3-axis MEMS accelerometer. S302, using sensor data that has been smoothed and denoised by wavelet filtering, calculates the drilling horizontal angle according to the following formula. ,in, , , These are the triaxial correction values ​​for the accelerometer.

6. The method for detecting the dip angle of a rock stratum structure surface based on linear multi-point drilling sensing according to claim 5, characterized in that, The method further includes the following steps: S4, in the data cache module of the drilling pipe data processing system, uses a hierarchical and associative storage method based on a discrete algebra structure. Its specific data structure definition is as follows: S401, the continuous sampling data of the sensor is defined as a set of discrete time sequence vectors that monotonically increase with time. ,in For discrete timestamps, and They represent The three-dimensional spatial vectors of the fluxgate and accelerometer at a given moment; Indicates the index number; Represents the set of positive integers; S402, the results of wavelet filtering and the calculated drilling horizontal angle are defined as a multidimensional ordered tuple structure. Sequential construction and storage are performed, among which, This serves as a unique identifier for each drilling pipe. To drill at a horizontal angle; S403, the storage addressing and calling logic in the data cache module of the drilling pipe data processing system adopts a timestamp-based injective function.

7. The method for detecting the dip angle of a rock stratum structure surface based on linear multi-point drilling sensing according to any one of claims 2-6, characterized in that, As the drill bit continues to approach the rock strata surface, multiple consecutive dip angles of the rock strata surface can be obtained. The optimal dip angle of the rock strata surface is then estimated using a Kalman filter model, and the optimal approximate value of the dip angle is output. The steps are as follows: S501: The calculated dip angle values ​​of multiple consecutive rock strata are formed into a sequence. The arithmetic mean of the sequence is calculated. If the absolute value of the deviation between the calculated dip angle value of a certain rock strata in the sequence and the average value exceeds the preset threshold, the value is determined to be an outlier and is removed. The remaining valid rock strata dip angle values ​​after removal are arranged in chronological order to form a new input sequence for the Kalman filter model to call. S502, the dip angle of the rock strata structure surface is used as the state variable of the Kalman filter model, and the measured value of the dip angle of the rock strata structure surface obtained by a single calculation is used as the observation. In the S503 Kalman filter model, the range of the process noise covariance matrix is ​​set to 10⁻. 6 The range of the observation noise covariance matrix is ​​~10⁻². 5 ~10⁻¹, the update step size of the Kalman gain is 1; S504. Following the prediction and update iteration steps of Kalman filtering, multiple sets of measured values ​​of rock stratum structure surface dip angles are filtered and calculated to obtain the optimal approximate value of the rock stratum structure surface dip angle. During the iterative calculation process, the storage of the filtered intermediate data adopts a dynamic finite sequence structure based on first-in-first-out logic. The data is overwritten in the data cache module with the iteration step number as the sequential index. Finally, only the optimal approximate estimate of the dip angle state of the rock stratum structure surface is retained and output.

8. The method for detecting the dip angle of a rock stratum structure surface based on linear multi-point drilling sensing according to claim 7, characterized in that, The Kalman filter formula and filter parameters are pre-stored in the long-term data storage module of the drilling pipe data processing system and read into the data cache module during drilling pipe initialization. At each moment , Kalman gain and intermediate filtering values ​​are stored in the data cache module. The data storage structure adopts a queue structure and is stored sequentially according to the iteration order. The intermediate data of the filtering is stored in the data cache module, and only the optimal approximate value of the dip angle of the rock stratum structure surface is retained; The preset threshold is ±10°; The optimal approximation of the dip angle of the rock stratum structure plane is stored in the data cache module, and the optimal approximation stored in the data cache module is used for real-time retrieval. The optimal approximation of the dip angle is output in real time through the data transmission line of the drilling pipe and simultaneously transmitted to the drilling rig main control system, the downhole data acquisition terminal, and the ground monitoring center. The main control system uses this optimal approximation as the basis for adaptive adjustment of drilling parameters, and the ground monitoring center uses this optimal approximation to realize real-time display, historical query, and anomaly warning of the dip angle parameters of the rock strata structure.

9. The method for detecting the dip angle of a rock stratum structure surface based on linear multi-point drilling sensing according to any one of claims 2-6, characterized in that, The process of calculating the dip angle of the rock stratum structure surface based on the drilling horizontal angle and the cosine theorem in step S3 is implemented by the following geometric solution program based on sequential storage: In the data cache module of the drilling pipe data processing system, a one-dimensional array is allocated to store intermediate results sequentially according to the calculation order. Store the reflection path lengths of each signal receiving node calculated in step S2 into an array; Based on the principle of mirror reflection, determine the distances AP' and BP' between the symmetrical point P' of the signal generating node P with respect to the rock stratum structure surface and the signal receiving nodes A and B, and store them in an array; Based on triangle ABP', calculate ∠P'BA using the law of cosines and store the result in an array; Based on ∠P'BA and the fixed distance PB between the signal generating node P and the signal receiving node B, the position parameters PD of the signal reflection point D corresponding to the signal receiving node B and its symmetrical point distance DP' are solved and stored in an array. The same method is used to solve the position parameters CP of the signal reflection point C corresponding to the signal receiving node A and its symmetrical point distance CP', and stored in an array. Based on CP', DP' and ∠AP'B, calculate the spatial distance CD between signal reflection points C and D and store it in an array; Calculate ∠P'DC based on triangle P'DC; Using the geometric relationship ∠P'DC = α + ∠P'BA, the dip angle α of the rock stratum structure surface is calculated and stored in an array.

10. The method for detecting the dip angle of a rock stratum structure surface based on linear multi-point drilling sensing according to any one of claims 1-6, characterized in that, The spacing between adjacent signal receiving nodes is set to 2m-3m; the deviation angle between the linear arrangement axis of the signal receiving nodes and the horizontal drilling direction of the drill bit is <10°.

Images