Large-caliber horizontal directional drilling device and compensation control method thereof

Abstract

The invention relates to the technical field of directional drilling, and discloses a large-diameter horizontal directional drilling device and a compensation control method thereof, and the method comprises the following steps: collecting drilling data; a coupling coefficient is obtained according to the bit pressure, the drill torque and the feeding speed, and whether longitudinal compensation is conducted or not is judged; when it is judged that longitudinal compensation is carried out, drilling data in a preset time period are collected, the drilling data are analyzed based on a sliding window, a drilling feature vector is obtained, the drilling feature vector is input into a multi-head attention model, a predicted bit pressure sequence is obtained, and a preset control quantity is obtained according to the predicted bit pressure sequence; and collecting real-time vibration acceleration, comparing the vibration acceleration with a material tolerance threshold value, and judging whether to carry out early warning or not. According to the method, the drill bit deviation or clamping stagnation phenomenon is restrained, the mechanical damage risk caused by abnormal vibration of equipment is reduced, and the stability and continuity of drilling operation in the complex stratum environment are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of directional drilling, and in particular to a large-caliber horizontal directional drilling device and a compensation control method thereof. Background Art

[0002] During horizontal directional drilling (HDD) in coal mines, the drill rod is a critical component for ensuring the proper operation of the drilling rig. Its operational stability has a direct impact on equipment safety and production. When the drill rod is subjected to external forces such as the coal seam or excitation, its operational state can be significantly affected. In severe cases, it can even cause drill rod sticking, resulting in rig downtime, significant economic losses, and even personal injury. To reduce the occurrence of drill rod sticking and improve production efficiency, it is crucial to identify drill rod sticking.

[0003] Currently, methods for identifying stuck drill pipe conditions are typically based on deep learning or image recognition, without in-depth consideration of drilling parameters associated with the drill tool, such as torque, thrust pressure, and thrust displacement. This impacts the accuracy of drill pipe stuck state identification and reduces reliability. Furthermore, existing drilling control methods lack real-time monitoring and control of drilling rig operating parameters, making them unable to respond promptly to changes in complex geological conditions. This results in low drilling efficiency and may even pose safety risks.

[0004] Therefore, it is necessary to design a large-caliber horizontal directional drilling device and a compensation control method thereof to solve the problems existing in the current technology. Summary of the Invention

[0005] In view of this, the present invention proposes a large-diameter horizontal directional drilling device and a compensation control method thereof, aiming to solve the problem that the current drilling control method lacks real-time monitoring and regulation of the drilling rig's drilling operation parameters, and cannot respond to changes in complex geological conditions in a timely manner, resulting in low drilling efficiency and prone to safety hazards.

[0006] In one aspect, the present invention provides a compensation control method for a large-caliber horizontal directional drilling device, comprising: Collecting drilling data, including drilling weight, drill bit torque, feed speed, and vibration acceleration; obtaining a coupling coefficient according to the bit weight, drill bit torque and feed speed, and determining whether to perform longitudinal compensation according to the coupling coefficient; When it is determined that longitudinal compensation is to be performed, drilling data within a preset time period is collected, the drilling data is analyzed based on a sliding window to obtain a drilling feature vector, the drilling feature vector is input into a multi-head attention model to obtain a predicted weight-on-bit sequence, and a pre-adjusted control amount is obtained based on the predicted weight-on-bit sequence, the pre-adjusted control amount including a thrust compensation amount; When the preset control amount is in operation, real-time vibration acceleration is collected, the vibration acceleration is compared with the material tolerance threshold, and whether to issue an early warning is determined according to the comparison result, and an audible and visual alarm is issued according to the early warning result.

[0007] Furthermore, when determining whether to perform longitudinal compensation according to the coupling coefficient, the method includes: Comparing the coupling coefficient with a minimum coupling coefficient threshold and a maximum coupling coefficient threshold respectively, and determining whether to perform longitudinal compensation according to the comparison results; When the coupling coefficient is less than a minimum coupling coefficient threshold or the coupling coefficient is greater than a maximum coupling coefficient threshold, determining to perform longitudinal compensation; When the coupling coefficient is greater than or equal to a minimum coupling coefficient threshold and the coupling coefficient is less than or equal to a maximum coupling coefficient threshold, it is determined that longitudinal compensation is not performed.

[0008] Furthermore, when collecting drilling data within a preset period and analyzing the drilling data based on a sliding window to obtain a drilling feature vector, the method includes: Synchronous sampling is performed within a preset time period to collect drilling data, and the collected data is preprocessed. The preprocessing includes missing value filling, normalization, and error value removal. The preprocessed data is divided into a number of sliding time windows. Statistical features are extracted for each time window to form the drilling feature vector. The statistical features include mean, standard deviation, range, and first-order difference mean. The drilling feature vector is an input matrix of T×D dimensions, where T represents the time step and D represents the number of input features.

[0009] Furthermore, after obtaining the drilling feature vector, the method further includes: Performing cluster analysis on the drilling feature vector and the historical compensation vector set, and determining a method for obtaining the predicted weight-on-bit sequence based on the analysis results; the historical compensation vector set includes a plurality of historical drilling feature vectors and a plurality of historical predicted weight-on-bit sequences, and each of the historical drilling feature vectors corresponds to a historical predicted weight-on-bit sequence; Taking the historical drilling feature vectors in the historical compensation vector set as sample points, setting the minimum sample number parameter MinPts and the neighborhood radius ϵ; Calculating the density of the neighborhood of each sample point; The samples are divided into several density clusters according to the density reachability rule. Each density cluster represents a drilling feature, and all points that do not meet the density condition are marked as outliers. The outliers do not participate in the current analysis. Putting the drilling feature vector into the historical density cluster model to obtain a cluster comparison result; When the density cluster containing the drilling feature vector includes the historical drilling feature vector, obtaining the predicted weight-on-bit sequence according to the historical predicted weight-on-bit sequence corresponding to the historical drilling feature vector; When the density cluster containing the drilling feature vector does not include the historical drilling feature vector, the drilling feature vector is input into a multi-head attention model to obtain the predicted weight on bit sequence.

[0010] Furthermore, when obtaining the predicted weight-on-bit sequence according to the historical predicted weight-on-bit sequence corresponding to the historical drilling feature vector, the method includes: When the density cluster containing the drilling feature vector contains only one historical drilling feature vector, the historical predicted weight-on-bit sequence corresponding to the historical drilling feature vector is used as the predicted weight-on-bit sequence; When the density cluster containing the drilling feature vector contains two or more historical drilling feature vectors, the similarity between the drilling feature vector and each of the historical drilling feature vectors is obtained, and the historical predicted weight-on-bit sequence corresponding to the maximum similarity is used as the predicted weight-on-bit sequence.

[0011] Furthermore, the drilling feature vector is input into a multi-head attention model to obtain the predicted weight-on-bit sequence, including: The multi-head attention model adopts a Transformer encoder structure. Each attention head is weighted based on the interactive correlation between drilling pressure, drill torque, feed speed and vibration acceleration, and outputs a fused feature representation.

[0012] Furthermore, the predicted WOB sequence generated by the multi-head attention model is a sequence of predicted WOB values ​​for several time steps in a preset future period. The WOB trend change and WOB gradient are calculated based on the WOB change trend and gradient change in the predicted WOB sequence, and the pre-adjusted control quantity is obtained based on the WOB trend change and WOB gradient.

[0013] Furthermore, when determining whether to issue an early warning based on the comparison results, the following steps are included: When the vibration acceleration is greater than the material tolerance threshold, it is determined to issue an early warning; When the vibration acceleration is less than or equal to the material tolerance threshold, it is determined that no warning is issued.

[0014] Furthermore, when sound and light alarms are issued according to the early warning results, the following steps are included: An excess difference is obtained based on the vibration acceleration and the material tolerance threshold, and the excess difference is the difference between the vibration acceleration and the material tolerance threshold. An alarm level is determined based on the excess difference and an audible and visual alarm is performed. The alarm level is proportional to the excess difference.

[0015] Compared with the existing technology, the beneficial effects of the present invention are: by real-time collection of drilling data, including parameters such as drilling pressure, drill bit torque, feed speed and vibration acceleration, real-time monitoring of the drilling operation status of the drilling rig is realized, and changes under complex geological conditions can be responded to in a timely manner, effectively improving drilling efficiency and reducing the occurrence of safety hazards; the coupling coefficient is obtained according to the drilling pressure, drill bit torque and feed speed, and whether longitudinal compensation is performed is judged according to the coupling coefficient. Combined with the comparison mechanism of the highest threshold of the coupling coefficient, accurate identification of the drill rod stuck state is achieved, and the accuracy and reliability of identification are improved; the sliding window technology is used to analyze the drilling data, extract the drilling feature vector, and input it into the multi-head attention model, thereby realizing intelligent control of the drilling process and avoiding the problems of parameter and structure optimization of artificial intelligence and deep learning algorithms in traditional methods; the dynamic adjustment effect of drilling parameters is improved and the risk of stuck drill is reduced.

[0016] On the other hand, the present application also provides a large-diameter horizontal directional drilling device, which is used to apply the above-mentioned large-diameter horizontal directional drilling device compensation control method, including: frame; A drilling device is fixed to the upper surface of the frame, and the drilling device includes a drive motor, a drill rod and a guide sleeve; A drilling compensation device is fixed below the drilling device, the drilling compensation device is fixedly connected to the drilling device, and the drilling compensation device includes a first cylinder; a clamping device, disposed between the drive motor and the guide sleeve; Sensor module, including vibration sensor, speed sensor and pressure sensor; a drill rod adjustment device fixed between the frame and the drilling device, the drill rod adjustment device comprising a second cylinder; an air pump connected to the first cylinder and the second cylinder; A supporting device is arranged around the frame, and the supporting device includes a hydraulic lifting rod; A control device is electrically connected to the drilling device, the drilling compensation device, the sensor module, the drill rod adjustment device and the support device. The control device is used to collect drilling data and control the first cylinder and the second cylinder to perform longitudinal compensation of the drill rod according to the drilling data.

[0017] It is understandable that the above-mentioned large-diameter horizontal directional drilling device and compensation control method thereof have the same beneficial effects, and will not be described in detail here. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present invention. The same reference symbols are used throughout the drawings to represent the same components. In the drawings: Figure 1 A flow chart of a compensation control method for a large-caliber horizontal directional drilling device provided by an embodiment of the present invention; Figure 2 A schematic structural diagram of a large-caliber horizontal directional drilling device provided in an embodiment of the present invention; Figure 3 for Figure 2 A magnified view in the middle A direction; Figure 4 A front view of a large-diameter horizontal directional drilling device provided in an embodiment of the present invention.

[0019] Among them, 100, frame; 110, drilling device; 111, drive motor; 112, drill rod; 113, guide sleeve; 120, drilling compensation device; 121, first cylinder; 130, drill rod adjustment device; 131, second cylinder; 140, support device; 141, hydraulic lifting rod; 150, control device; 160, air pump; 170, clamping device. DETAILED DESCRIPTION

[0020] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art. It should be noted that, unless there is a conflict, the embodiments of the present disclosure and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

[0021] In conventional coal mine horizontal directional drilling systems, the heterogeneous and dynamic coupling characteristics of multi-source sensor data make parameter correlation analysis difficult and introduce lags in drilling process control. Parameters such as weight on bit, torque, and feed rate collected by the sensor network exhibit nonlinear correlations in the time and frequency domains. Conventional single-variable threshold determination methods are unable to accurately characterize the multi-physics coupling effects, resulting in an error rate in longitudinal compensation decisions exceeding the allowable range of industrial standards. Under dynamic drilling conditions, the vibration acceleration signal and the resonant modes of the mechanical transmission system overlap in frequency bands, making it difficult for traditional frequency-domain filtering methods to effectively separate effective vibration characteristics from the inherent vibration noise of the equipment.

[0022] For example, during horizontal drilling in limestone formations, torque fluctuations reached 2.3 times the steady-state value when the drill bit contacted a fractured zone, while the WOB sensor detected periodic impact loads. The existing system used mean filtering with a fixed time window, resulting in a mismatch between the feature extraction window and the impact load period and a loss of phase information in the feature vector. Linear regression prediction based on a historical experience database failed to capture the nonlinear dynamic characteristics of the sudden WOB changes, causing a delay of 380 milliseconds in the thrust compensation command and causing the drill pipe lateral swing to exceed the safety threshold.

[0023] If these issues are not addressed, dynamic drill pipe instability will accelerate the axial wear rate of the guide sleeve by 3.8 times, shortening the service life of key components. Compensation control lag will exacerbate contact stress oscillations between the drill bit and the rock formation, increasing the probability of drill pipe buckling to dangerous levels. Misjudgment of vibration characteristics can mask early signs of mechanical failure, increasing the risk of unplanned downtime and directly impacting the safety and economic efficiency of continuous drilling operations underground in coal mines.

[0024] See Figure 1 As shown, in this regard, the present application proposes a compensation control method for a large-caliber horizontal directional drilling device, comprising: S100: Collect drilling data, including drilling weight, drill bit torque, feed speed and vibration acceleration; S200: obtaining a coupling coefficient according to the drilling pressure, the drill torque, and the feed speed, and determining whether to perform longitudinal compensation according to the coupling coefficient; S300: When it is determined that longitudinal compensation is to be performed, drilling data within a preset period is collected, the drilling data is analyzed based on a sliding window to obtain a drilling feature vector, the drilling feature vector is input into a multi-head attention model to obtain a predicted weight-on-bit sequence, and a pre-adjusted control variable is obtained based on the predicted weight-on-bit sequence, the pre-adjusted control variable including a thrust compensation amount; S400: When running based on the pre-adjusted control quantity, real-time vibration acceleration is collected, and the vibration acceleration is compared with the material tolerance threshold. Based on the comparison result, it is determined whether to issue an early warning, and an audible and visual alarm is issued based on the early warning result.

[0025] Specifically, S100 can be implemented using a combination of pressure sensors, torque sensors, speed sensors, and acceleration sensors, achieving multi-dimensional perception of drilling status through synchronous acquisition of multiple types of sensors. S200 can be implemented using a threshold comparison algorithm, and abnormal working condition identification can be achieved by establishing a correlation model between drilling pressure, torque, and feed speed. S300 can be implemented using a sliding window algorithm combined with a time series prediction model, and future drilling pressure changes can be predicted by analyzing historical data trends. A sliding window refers to an algorithm that dynamically intercepts continuous time data. Specifically, it can be used to divide data into segments using fixed time intervals, and capture working condition change patterns by extracting statistical features within the window. A multi-head attention model refers to a neural network structure that processes time series data. Specifically, it can be implemented using a parallel attention mechanism, and improves prediction accuracy by capturing the nonlinear relationship between multiple parameters such as drilling pressure, torque, and vibration. S400 can be implemented using a threshold trigger mechanism, and triggers protective actions by real-time monitoring of the relationship between vibration intensity and material tolerance.

[0026] This application constructs a closed-loop control system that integrates multi-source sensor data, dynamic working condition judgment, trend prediction compensation and real-time safety monitoring. By combining sliding window feature extraction with a multi-head attention model, it solves the compensation lag problem caused by traditional methods ignoring parameter correlation. At the same time, it establishes a direct mapping relationship between vibration monitoring and material tolerance threshold, realizing full process control from anomaly identification, trend prediction to active protection.

[0027] The working process and principle of the present application are as follows: drilling data is collected, including drilling pressure, drill bit torque, feed speed and vibration acceleration. A coupling coefficient is obtained based on the drilling pressure, drill bit torque and feed speed, and whether longitudinal compensation is performed is determined based on the coupling coefficient. When it is determined that longitudinal compensation is required, drilling data within a preset time period is collected, and the drilling data is analyzed based on a sliding window to obtain a drilling feature vector. The drilling feature vector is input into a multi-head attention model to obtain a predicted drilling pressure sequence, and a pre-adjusted control amount is obtained based on the predicted drilling pressure sequence, including a thrust compensation amount. When operating based on the pre-adjusted control amount, real-time vibration acceleration is collected, and the vibration acceleration is compared with the material tolerance threshold. Based on the comparison result, it is determined whether to issue an early warning, and an audible and visual alarm is issued based on the early warning result.

[0028] A sensor network collects key parameter data during the drilling process in real time. Based on a multi-parameter coupling analysis model, a coupling coefficient reflecting the drilling status is calculated, avoiding the limitations of single-parameter judgment. A sliding window method is used to extract dynamic features from time series data, overcoming the feature loss problem that can occur with fixed time windows. A multi-head attention model captures nonlinear correlations between parameters, improving the accuracy of weight-on-bit predictions. Real-time monitoring of vibration acceleration allows for timely identification of potential risks and ensures the safety of the drilling process.

[0029] It is understandable that the multi-level control strategy can effectively cope with the dynamic changes under complex geological conditions and improve the stability and safety of the drilling process.

[0030] Specifically, the coupling coefficient is used to determine whether to perform longitudinal compensation, comprehensively considering the interplay of multiple parameters and improving accuracy. Feature extraction using a sliding window captures the dynamic characteristics of drilling data, facilitating real-time control. A multi-head attention model is introduced for WOB prediction, which can handle complex nonlinear relationships between parameters and improve prediction accuracy. A vibration acceleration early warning mechanism can promptly detect anomalies and prevent equipment damage and safety accidents.

[0031] As a preferred embodiment, the solution of this application is specifically implemented as follows: The drilling data is collected in real time using pressure sensors, torque sensors, velocity sensors, and acceleration sensors installed on the drilling device. The data acquisition frequency is set to 100 Hz to ensure that high-frequency changes are captured.

[0032] The coupling coefficient is calculated using a multi-parameter fusion algorithm based on fuzzy logic. Specifically, the weight on bit, drill torque, and feed rate are first normalized, and then the coupling coefficient is calculated based on pre-set fuzzy rules. The coupling coefficient ranges from 0 to 1. When the coupling coefficient is greater than 0.8 or less than 0.2, longitudinal compensation is determined to be necessary.

[0033] When longitudinal compensation is determined to be necessary, the most recent 60 seconds of drilling data are collected. The data is segmented using a 10-second sliding window with a step size of 1 second. Statistical features, including mean, standard deviation, kurtosis, and skewness, are extracted from the data within each window to form a drilling feature vector.

[0034] Furthermore, we used an 8-head attention Transformer encoder structure as a multi-head attention model. The model input is a 60×16 feature matrix, where 60 is the time step length and 16 is the feature dimension per time step. The model output is a sequence of WOB predictions for the next 30 seconds.

[0035] Based on the predicted WOB sequence, the WOB change trend and gradient are calculated to generate the thrust compensation. The compensation calculation takes into account the predicted WOB change rate and magnitude to achieve a smooth transition.

[0036] Vibration acceleration data is collected every 0.1 seconds and compared with the preset material tolerance threshold. When the vibration acceleration exceeds the threshold, an early warning mechanism is triggered. The early warning level is divided into three levels, corresponding to different audible and visual alarms, depending on the degree of the threshold exceeded.

[0037] Through the above scheme, this application realizes intelligent compensation control of large-diameter horizontal directional drilling process. Based on the coupled analysis of multi-source sensor data, the accuracy of longitudinal compensation decision-making is improved. The use of sliding window and multi-head attention model for feature extraction and prediction overcomes the limitations of traditional methods in dealing with nonlinear dynamics. Real-time vibration monitoring and early warning mechanism effectively reduce the risk of equipment failure and safety accidents. This method can adapt to dynamic changes in complex geological conditions, improve drilling efficiency and safety, and provide a new solution for the development of horizontal directional drilling technology in coal mines.

[0038] In some of the aforementioned solutions of this application, when determining longitudinal compensation based on the coupling coefficient, the lack of a clear threshold range as a basis for judgment may lead to misjudgments or omissions in compensation decisions, affecting the stability and reliability of the drilling device. The coupling coefficient can be calculated as drill bit torque / (weight on bit × drilling speed). The calculation is preferably performed using normalized and dimensionless data.

[0039] The present application further proposes to compare the coupling coefficient with the minimum threshold value of the coupling coefficient and the maximum threshold value of the coupling coefficient respectively, and determine whether to perform longitudinal compensation based on the comparison results; when the coupling coefficient is less than the minimum threshold value of the coupling coefficient or the coupling coefficient is greater than the maximum threshold value of the coupling coefficient, it is determined to perform longitudinal compensation; when the coupling coefficient is greater than or equal to the minimum threshold value of the coupling coefficient, and the coupling coefficient is less than or equal to the maximum threshold value of the coupling coefficient, it is determined not to perform longitudinal compensation.

[0040] The minimum and maximum coupling coefficient thresholds are set based on historical operational data of the drilling system under various geological conditions, with the boundaries determined through statistical analysis. For example, the minimum threshold can be set to 0.3, and the maximum threshold to 0.8, with the specific values ​​adjusted based on the drill pipe material properties and geological hardness. During the comparison process, the logic operation module compares the real-time calculated coupling coefficient with the preset thresholds, outputting a binary judgment result. If the coupling coefficient exceeds the threshold range, a longitudinal compensation command is triggered; if it is within the threshold range, the current drilling status is maintained.

[0041] Specifically, during the drilling process, real-time measurement data on bit weight, drill bit torque, and feed speed are received, and a dynamic coupling coefficient is generated using a coupling coefficient calculation model. This coefficient reflects the degree of correlation between the internal parameters of the drilling system. If the coupling coefficient is below the minimum threshold, it indicates that the bit weight and torque are not properly matched, which may lead to insufficient drill pipe thrust. If it is above the maximum threshold, it indicates excessive coupling between the parameters, which may cause drill bit overload. By comparing the coupling coefficient with the preset threshold, an accurate compensation decision signal is generated to avoid misjudgments caused by parameter fluctuations or noise interference. For example, when the coupling coefficient is 0.25, compensation is determined to be necessary, and the thrust compensation amount is generated according to the subsequent steps. If the coupling coefficient is 0.5, compensation is determined to be unnecessary, reducing unnecessary control actions. This threshold judgment mechanism effectively improves the accuracy and response efficiency of compensation control.

[0042] As a preferred embodiment, the solution of this application is specifically implemented as follows: The coupling coefficient is compared with the minimum and maximum coupling coefficient thresholds, respectively, and the comparison results determine whether longitudinal compensation is required. Specifically, longitudinal compensation is determined to be performed when the coupling coefficient is less than the minimum coupling coefficient threshold or greater than the maximum coupling coefficient threshold. For example, the minimum coupling coefficient threshold can be set to 0.3 and the maximum coupling coefficient threshold can be set to 0.7. When the detected coupling coefficient is 0.2 or 0.8, longitudinal compensation is determined to be necessary.

[0043] Furthermore, when the coupling coefficient is greater than or equal to the lowest coupling coefficient threshold and less than or equal to the highest coupling coefficient threshold, it is determined that longitudinal compensation is not required. For example, when the detected coupling coefficient is 0.5, since this value is within the range of 0.3 to 0.7, it is determined that longitudinal compensation is not required.

[0044] Through the above technical solution, the present application can accurately determine whether longitudinal compensation is necessary based on the comparison of the actual value of the coupling coefficient with a preset threshold. This avoids unnecessary compensation operations and improves drilling efficiency. Furthermore, when the coupling coefficient deviates from the normal range, timely longitudinal compensation helps maintain drilling stability and reduces the risk of stuck drills. This threshold-based judgment method is simple, effective, easy to implement, and can quickly respond to parameter changes during the drilling process.

[0045] In some of the above-mentioned solutions of the present application, problems such as incomplete data or noise interference may occur when collecting drilling data, resulting in deviations in subsequent analysis results and affecting the accuracy of the predicted bit weight sequence.

[0046] The present application further proposes to synchronously sample and collect drilling data within a preset time period, and preprocess the collected data. The preprocessing includes missing value filling, normalization processing and error value removal. The preprocessed data is divided into several sliding time windows, and statistical features are extracted for each time window to form a drilling feature vector. The statistical features include mean, standard deviation, range and first-order difference mean. The drilling feature vector is an input matrix of T×D dimensions, where T represents the time step and D represents the number of input features.

[0047] Synchronous sampling ensures data alignment in time, avoiding feature extraction bias caused by asynchronous sampling. Missing value filling complements incomplete data segments through interpolation or the mean of historical data to ensure data continuity. Normalization maps drilling data of different dimensions to a unified interval, eliminating the impact of dimensional differences on model training. Error removal filters out abnormally fluctuating data by setting a threshold. For example, outliers are identified and removed by using the standard deviation threshold within a sliding window. The sliding time window is divided using a fixed or adaptive step size, for example, a window of 5 seconds, with a 50% overlap between windows to capture dynamic data changes. Among the statistical features, the mean reflects the average level of the data within the window, the standard deviation measures the degree of data fluctuation, the range reflects the data range, and the first-order difference mean characterizes the trend of changes between adjacent time points. In the T×D dimensional input matrix, T corresponds to the number of time windows, and D corresponds to the number of statistical features extracted from each window.

[0048] Specifically, synchronous sampling is first used to acquire time series data on drilling pressure, drill torque, feed speed, and vibration acceleration, and the data is then preprocessed. Missing values ​​are filled using linear interpolation to fill short-term missing segments, while long-term missing segments are filled using the mean of historical data from the same period. Normalization uses the maximum-minimum method to scale each parameter to the [0,1] interval. Error value removal sets a dynamic threshold based on the standard deviation of each parameter within the sliding window. For example, when a data point deviates from the window mean by more than three standard deviations, it is identified as an error value and removed. The preprocessed data is divided into multiple sliding windows, and four statistical features are extracted from each window: mean, standard deviation, range, and first-order difference mean. For example, for the drilling pressure parameter, the average value of all sampling points within the window is calculated as the mean feature, the difference between the maximum and minimum values ​​within the window is used as the range feature, and the average of the differences between adjacent sampling points is used as the first-order difference mean feature. Ultimately, each window generates a vector containing four statistical features. Multiple windows are arranged in chronological order to form a T × D dimensional input matrix, where T is the number of windows and D is the number of features per window. Through these steps, the raw time series data is converted into a structured feature matrix, providing high-quality input for subsequent models and improving the accuracy of predicted WOB sequences.

[0049] As a preferred embodiment, the solution of the present application is specifically implemented as follows: synchronous sampling is performed within a preset time period to collect drilling data, and the collected data is preprocessed. Preprocessing includes missing value filling, normalization, and error removal. The preprocessed data is divided into several sliding time windows, and statistical features are extracted for each time window to form a drilling feature vector. Statistical features include mean, standard deviation, range, and first-order difference mean. The drilling feature vector is an input matrix of T×D dimensions, where T represents the time step and D represents the number of input features.

[0050] Specifically, a preset time period of 10 minutes can be set, during which drilling data is collected every 1 second. For the collected data, missing values ​​are first filled using linear interpolation to fill the missing data points. Normalization is then performed to map each data point to a range between 0 and 1. Error removal is then performed, using the 3σ criterion to eliminate outliers. After preprocessing, the data is divided into 20 sliding time windows, each containing 30 seconds of data. For each window, the mean, standard deviation, range, and first-order difference mean are calculated to form a 4×20-dimensional feature matrix, which serves as the drilling feature vector.

[0051] Through the above-mentioned technical solution, this application can effectively extract key feature information from the drilling process, providing reliable input data for subsequent WOB prediction. By preprocessing and extracting features from the raw data, the impact of data noise can be reduced, and the representativeness and discrimination of features can be improved. The sliding window design enables feature extraction to capture the temporal variations of drilling parameters, facilitating more accurate prediction of future WOB trends. As a result, this solution provides a reliable data foundation for real-time monitoring and prediction of the drilling process, which is conducive to improving the accuracy and efficiency of drilling control.

[0052] In some of the aforementioned solutions, drilling data is analyzed using a sliding window to generate drilling feature vectors, which are then directly fed into a multi-head attention model for prediction. However, when faced with complex geological conditions, historical data contains a large number of drilling feature vectors and corresponding prediction results under similar conditions. If model predictions are re-performed each time, this wastes computing resources and causes response delays, impacting the efficiency of real-time compensation control.

[0053] The present application further proposes that after obtaining the drilling feature vector, the drilling feature vector and the historical compensation vector set are clustered and analyzed, and the method for obtaining the predicted drilling pressure sequence is determined based on the analysis results; the historical compensation vector set includes several historical drilling feature vectors and corresponding historical predicted drilling pressure sequences; the historical drilling feature vectors in the historical compensation vector set are used as sample points, and the minimum sample number parameter and the neighborhood radius are set; the density of the neighborhood of each sample point is calculated; the samples are divided into several density clusters according to the density reachability rule, and all points that do not meet the density conditions are marked as outliers; the drilling feature vector is input into the historical density cluster model to obtain the cluster comparison result; when the density cluster where the drilling feature vector is located contains the historical drilling feature vector, the predicted drilling pressure sequence is obtained based on the corresponding historical predicted drilling pressure sequence; when it is not included, the drilling feature vector is input into the multi-head attention model to obtain the predicted drilling pressure sequence.

[0054] Density calculation utilizes a neighborhood search method based on Euclidean distance. The neighborhood radius is dynamically adjusted based on the distribution characteristics of historical data to ensure that drilling feature vectors under different operating conditions are correctly classified. The minimum sample number parameter is set to 5 to avoid interference from noisy data. After outliers are marked, subsequent analysis is performed only on valid density clusters. Once a drilling feature vector is classified into a density cluster, the predicted WOB sequences corresponding to all historical drilling feature vectors within that cluster are prioritized. The optimal historical result is selected through similarity matching to avoid duplicate calculations.

[0055] Specifically, after completing data preprocessing and feature vector generation, the current drilling feature vector is first density clustered with the historical data set. For example, the DBSCAN algorithm is used to cluster historical data offline to form multiple density clusters. When applied online, the real-time feature vector is mapped to the established density cluster model. If the current feature vector is within a density cluster, the prediction sequence corresponding to the most similar historical data in the cluster is directly called; if it is in an outlier area, the multi-head attention model is triggered for real-time prediction. In this way, while ensuring prediction accuracy, the frequency of model calculations is reduced and the response speed is improved. For example, when the similarity between the drilling feature vector and a historical cluster reaches 0.9, the average value of the prediction sequence within the cluster is directly used as the output, shortening the calculation time by 60%.

[0056] As a preferred embodiment, the solution of this application is specifically implemented as follows: Cluster analysis is performed on the drilling feature vectors and the historical compensation vector set. The method for obtaining the predicted WOB sequence is determined based on the analysis results. The historical compensation vector set includes several historical drilling feature vectors and several historical predicted WOB sequences, and each historical drilling feature vector corresponds to a historical predicted WOB sequence.

[0057] The historical drilling feature vectors in the historical compensation vector set are used as sample points. The minimum sample number parameter, MinPts, and the neighborhood radius, ϵ, are set. The density of the neighborhood of each sample point is calculated. The samples are divided into several density clusters based on the density reachability rule. Each density cluster represents a drilling feature. All points that do not meet the density condition are marked as outliers and are excluded from the current analysis.

[0058] The drilling feature vector is fed into the historical density cluster model to obtain cluster comparison results. If the density cluster containing the drilling feature vector contains the historical drilling feature vector, a predicted WOB sequence is obtained based on the historical predicted WOB sequence corresponding to the historical drilling feature vector. If the density cluster containing the drilling feature vector does not contain the historical drilling feature vector, the drilling feature vector is fed into the multi-head attention model to obtain a predicted WOB sequence.

[0059] Specifically, we can use the DBSCAN density clustering algorithm for cluster analysis. First, set MinPts = 5 and ϵ = 0.5. Then, for each sample point, calculate the number of points within the ϵ neighborhood. If the number is greater than or equal to MinPts, mark the point as a core point. Next, assign all core points with the highest density to the same cluster. Finally, assign the boundary points of non-core points that are densely accessible to the cluster to their corresponding clusters, and mark the remaining points as outliers.

[0060] Furthermore, the current drilling feature vector is compared with the historical density cluster model. If the current drilling feature vector falls into an existing density cluster, the historical predicted WOB sequence in that cluster is directly used. If it does not belong to any existing density cluster, it is input into the multi-head attention model to generate a new predicted WOB sequence.

[0061] Through the above technical solution, this application achieves intelligent classification and prediction of drilling characteristics. A density clustering algorithm can effectively identify drilling states with similar characteristics, improving the accuracy and efficiency of WOB prediction. For newly emerging drilling characteristics, a multi-head attention model is used for prediction. This approach, combining historical data with real-time prediction, can better cope with complex and changing drilling environments and improve the accuracy and stability of drilling control.

[0062] In some of the above-mentioned schemes of the present application, when there are multiple historical drilling feature vectors in the historical compensation vector set, directly using the model prediction method will result in a waste of computing resources, and randomly selecting the historical predicted drilling pressure sequence may lead to deviation in the compensation control amount, affecting the accuracy of the longitudinal compensation of the drilling device.

[0063] The present application further proposes that when the density cluster containing the drilling feature vector contains two or more historical drilling feature vectors, the similarity between the drilling feature vector and each historical drilling feature vector is obtained, and the historical predicted drilling weight sequence corresponding to the maximum similarity is used as the predicted drilling weight sequence.

[0064] Similarity calculations use the cosine similarity algorithm, determining the similarity metric by the ratio of the vector inner product to the product of the module length. Historical drilling feature vectors are stored in a circular buffer, sorted by timestamp, and hash-indexed for fast retrieval. A similarity threshold is set during the maximum similarity screening process. When all similarities fall below the threshold, the system automatically switches to model prediction mode.

[0065] Specifically, after completing the density cluster classification, all historical drilling feature vectors in the cluster are traversed, and the cosine similarity between the current drilling feature vector and the historical vector is calculated one by one. During the calculation process, the time window dimension is weighted, and the recent time period data is given a higher weight coefficient. When multiple similarities are detected exceeding the preset threshold, the historical prediction sequence corresponding to the maximum similarity is selected, and its time step is linearly interpolated to adapt to the current drilling speed. This process achieves fast matching through the pre-stored historical working condition mapping table, avoiding repeated running of the attention model, and reducing the calculation time by 60% while ensuring prediction accuracy. The drilling compensation device generates a thrust compensation amount based on the gradient change of the selected sequence, so that the longitudinal compensation action is executed 150-300 milliseconds in advance.

[0066] As a preferred embodiment, the solution of this application is specifically implemented as follows: When obtaining a predicted WOB sequence based on a historical predicted WOB sequence corresponding to a historical drilling feature vector, the following steps are included: If the density cluster containing the drilling feature vector contains only one historical drilling feature vector, the historical predicted WOB sequence corresponding to that historical drilling feature vector is directly used as the predicted WOB sequence. For example, if the density cluster contains only one historical drilling feature vector [0.5, 0.3, 0.7, 0.2], and its corresponding historical predicted WOB sequence is [100, 105, 110, 115, 120] kN, then that historical predicted WOB sequence is selected as the current predicted WOB sequence.

[0067] When the density cluster containing a drilling feature vector contains two or more historical drilling feature vectors, the similarity between the drilling feature vector and each of the historical drilling feature vectors is first calculated. Similarity can be calculated using methods such as Euclidean distance and cosine similarity. Furthermore, the historical predicted WOB sequence corresponding to the maximum similarity is selected as the predicted WOB sequence. For example, if the density cluster contains three historical drilling feature vectors (A, B, and C), and their similarities to the current drilling feature vector are 0.8, 0.9, and 0.7, respectively, then the historical predicted WOB sequence corresponding to historical drilling feature vector B, which has a similarity of 0.9, is selected as the predicted WOB sequence.

[0068] Through the above-mentioned technical solution, the present application can flexibly select the method for obtaining the predicted WOB sequence based on the distribution of the current drilling feature vector in historical data. When highly similar historical data exists, the historical prediction results are directly utilized, avoiding repeated calculations and improving prediction efficiency. For multiple similar historical data, the closest prediction result is selected through similarity comparison to ensure prediction accuracy. This method fully utilizes historical experience while being adaptable to new drilling situations, thereby improving the accuracy and efficiency of WOB predictions and providing reliable data support for subsequent compensation control.

[0069] In some of the aforementioned solutions, when the density cluster containing the drilling feature vector does not contain historical drilling feature vectors, the drilling feature vector must be input into the model to obtain a predicted WOB sequence. However, when dealing with multi-parameter interactions, traditional models struggle to fully capture the dynamic relationships between WOB, drill bit torque, feed rate, and vibration acceleration, resulting in insufficient accuracy in the predicted WOB sequence, which in turn affects the calculation accuracy of the pre-adjusted control variable.

[0070] This application further proposes a multi-head attention model using a Transformer encoder structure, in which each attention head performs weight allocation based on the interactive correlation between drilling pressure, drill bit torque, feed speed and vibration acceleration, and outputs a fused feature representation.

[0071] The multi-head attention model consists of multiple parallel attention heads, each of which independently calculates the interaction weights for different parameter combinations. For example, one attention head focuses on the correlation between drilling pressure and drill torque, while another analyzes the dynamic impact of feed rate and vibration acceleration. The Transformer encoder gradually extracts parameter correlation features within different time steps by stacking multiple layers of attention mechanisms. The weight assignment process uses a scaled dot-product attention mechanism to dynamically weight the time steps and feature dimensions in the input matrix. The fused feature representation is formed by concatenating the outputs of each attention head and performing a linear transformation to form a feature vector containing multidimensional interaction information.

[0072] Specifically, after the drilling feature vector is input into the multi-head attention model, the model first maps the input matrix into a query vector, a key vector, and a value vector. Each attention head independently calculates the dot product of the query vector and the key vector to generate an attention score matrix, which is then normalized into a weight distribution using the Softmax function. The normalized weights are multiplied by the value vector to produce the output of each attention head. The outputs of multiple attention heads are concatenated and fused through a linear layer to form the final fused feature representation. This representation captures the nonlinear interactions between WOB, drill bit torque, feed rate, and vibration acceleration at different time steps. Based on the fused feature representation, the model generates a sequence of WOB prediction values ​​for multiple future time steps through a fully connected layer. This multi-attention mechanism design simultaneously captures local and global dependencies between different parameters. For example, when the feed rate changes suddenly, the model promptly corrects the WOB prediction trend by adjusting the weight of the vibration acceleration, thereby improving the temporal coherence and physical plausibility of the prediction sequence.

[0073] As a preferred embodiment, the solution of this application is specifically implemented as follows: The multi-head attention model uses a Transformer encoder structure. Each attention head assigns weights based on the cross-correlation between drilling pressure, drill torque, feed rate, and vibration acceleration, and outputs a fused feature representation. Specifically, the multi-head attention model contains multiple parallel attention heads, each of which independently calculates attention weights. First, the input drilling feature vector is mapped to query, key, and value vectors through a linear transformation. Then, the dot product of the query vector and the key vector is calculated, scaled, and softmax normalized to obtain the attention weights. Next, the attention weights are multiplied by the value vector and summed to obtain the weighted feature representation. Finally, the outputs of multiple attention heads are concatenated and linearly transformed to obtain the final fused feature representation. This allows the model to capture the complex relationships between different drilling parameters and extract richer feature information.

[0074] Through the above-mentioned technical solution, this application can fully utilize the multi-head attention mechanism to capture the complex interactions between drilling parameters and improve the effectiveness of feature extraction. Multiple attention heads learn the correlation between parameters from different perspectives, enhancing the model's understanding of the drilling status. The fused feature representation contains more comprehensive and in-depth drilling information, providing high-quality input for subsequent drilling pressure prediction, thereby improving the accuracy and reliability of the prediction. In addition, this method has strong adaptability and can automatically adjust parameter weights according to different drilling scenarios, effectively responding to complex and changing geological conditions.

[0075] In some of the above-mentioned schemes of the present application, the predicted drilling pressure sequence generated by the multi-head attention model can reflect future drilling pressure changes, but it is difficult to accurately capture the dynamic change trend by relying solely on the predicted value sequence itself, resulting in a lag in the calculation of the pre-adjusted control quantity and an inability to respond in a timely manner to the impact of drilling pressure fluctuations on longitudinal compensation.

[0076] The present application further proposes calculating the WOB trend change amount and WOB gradient based on the predicted WOB trend and gradient change amount in the WOB sequence, and obtaining the pre-adjustment control amount according to the WOB trend change amount and WOB gradient.

[0077] The WOB trend change is determined by the linear regression slope of the predicted values ​​within a sliding window, while the WOB gradient is calculated by the difference between the predicted values ​​of adjacent time steps. The trend change reflects the overall direction of WOB change, while the gradient indicates the intensity of instantaneous fluctuations. Combining these two allows for dynamic adjustment of the thrust compensation amplitude and response speed. For example, when the trend change is positive and the gradient continues to increase, the thrust compensation increases exponentially. When the trend change approaches zero and the gradient fluctuates dramatically, the compensation is adjusted using a gradient weighting method.

[0078] Specifically, the predicted WOB sequence is divided into multiple time windows. Within each window, a trend line is fitted using the least squares method, with its slope used as the trend change. The mean difference between adjacent predicted values ​​within the window is also calculated as the gradient change. The trend change and gradient change are input into a compensation calculation model, which incorporates a piecewise linear function and fuzzy control rules. When the trend change exceeds a set threshold, the main compensation mechanism is activated, while when the gradient change exceeds a critical value, fine-tuning compensation is triggered. Through collaborative analysis of trends and gradients, sudden WOB changes can be identified in advance, and compensation instructions can be generated before the actual WOB change occurs, enabling proactive pre-adjustment control and effectively reducing the risk of excessive vibration acceleration.

[0079] As a preferred embodiment, the solution of this application is specifically implemented as follows: The predicted WOB sequence generated by the multi-head attention model is a sequence of predicted WOB values ​​for a predetermined number of time steps in the future. Based on the WOB trend and gradient changes in the predicted WOB sequence, the WOB trend change and WOB gradient are calculated. The pre-adjusted control variable is then derived from the WOB trend change and WOB gradient.

[0080] Specifically, a predicted WOB sequence might include predicted WOB values ​​every 5 minutes for the next 30 minutes. First, the WOB differences between adjacent time steps are calculated to determine the WOB trend. The average of these differences is then calculated as the WOB trend change. For the WOB gradient, the difference between the WOB at each time step and the initial WOB is calculated and divided by the corresponding time interval. Finally, based on a pre-defined mapping relationship, the WOB trend change and WOB gradient are converted into corresponding pre-adjusted control variables, such as thrust compensation.

[0081] For example, if the WOB trend is positive and the WOB gradient is large, indicating a potential sharp increase in WOB, a large negative thrust compensation is generated to prevent excessive WOB. Conversely, if the WOB trend is negative and the WOB gradient is small, a small positive thrust compensation is generated to maintain appropriate WOB.

[0082] Through the above technical solution, the present application can timely calculate the WOB change trend and gradient based on the predicted WOB sequence generated by the multi-head attention model, thereby accurately determining the required pre-adjustment control amount. This method takes into account the dynamic variation characteristics of WOB and can more accurately perform compensation control during the drilling process, effectively preventing equipment damage or reduced drilling efficiency caused by sudden changes in WOB. At the same time, real-time adjustment of the pre-adjustment control amount also helps maintain the stability of the drilling process and improve the safety and efficiency of drilling operations.

[0083] In some of the above-mentioned solutions of this application, a binary judgment is made only by whether the vibration acceleration exceeds the material tolerance threshold, which cannot distinguish between emergency states of different degrees of over-limit, resulting in a lack of targeted alarm feedback and difficulty in guiding operators to take differentiated emergency measures.

[0084] The present application further proposes obtaining an excess difference based on the vibration acceleration and the material tolerance threshold, determining an alarm level according to the excess difference and performing an audible and visual alarm, wherein the alarm level is proportional to the excess difference.

[0085] The excess difference is calculated by directly subtracting the real-time vibration acceleration value from the material tolerance threshold. A larger difference indicates a more severe deviation from the safe range. Alarm levels are determined based on a preset difference range. For example, a difference of 5-10 m / s² corresponds to a Level 1 alarm, while a difference of 10-20 m / s² corresponds to a Level 2 alarm. The audible and visual alarm system utilizes a multi-color LED light array and a variable-frequency buzzer. Different alarm levels trigger different color flashing frequencies and sound frequency combinations.

[0086] Specifically, when the vibration acceleration sensor detects that the real-time vibration value exceeds the material tolerance threshold, the absolute difference between the current vibration acceleration and the threshold is calculated. This difference is entered into a preset level mapping table, which contains multiple difference intervals and their corresponding alarm mode parameters. For example, when the difference is 8m / s², the yellow light is triggered to flash at a frequency of 1Hz, and the buzzer issues a short alarm at intervals of 0.5 seconds; when the difference is 15m / s², the red light is triggered to flash continuously at a frequency of 3Hz, and the buzzer issues a continuous high-frequency alarm. The alarm level increases linearly with the increase of the difference, ensuring that operators can intuitively judge the current level of equipment abnormality through the combination of sound and light signals and take emergency measures of the corresponding level in a timely manner.

[0087] As a preferred embodiment, the solution of the present application is specifically implemented as follows: the vibration acceleration signal of the drilling compensation device during operation is collected in real time, and the signal is filtered and noise-reduced before being numerically compared with the preset material tolerance threshold. When the vibration acceleration exceeds the material tolerance threshold, the positive deviation between the current vibration acceleration value and the threshold is calculated, and the deviation value is divided into three intervals: 0-5m / s², 5-10m / s², and above 10m / s². The corresponding deviation intervals trigger a combination alarm mode of yellow slow flashing light and low-frequency buzzing sound, orange constant light and medium-frequency alarm sound, and red fast flashing light and high-frequency buzzing sound. The alarm signal is output synchronously through the three-color LED light group and the multi-band speaker installed on the control panel. The larger the deviation value, the higher the light flashing frequency and the sound pulse frequency.

[0088] Through the above technical solution, this application implements a dynamic, graded alarm function based on the degree of vibration exceeding the limit, effectively resolving the problem in existing technologies where a single threshold alarm cannot distinguish between dangerous levels. By establishing a mapping relationship between deviation intervals and sound and light signals, operators can intuitively judge the severity of vibration anomalies and quickly implement differentiated emergency response measures, avoiding response delays or misoperations caused by ambiguous alarm information, thereby improving the safety of underground operations.

[0089] In some of the above-mentioned schemes of this application, an early warning is triggered when the vibration acceleration exceeds the material tolerance threshold, but only a simple binary judgment is performed, and the degree of excess cannot be distinguished, resulting in a lack of targeted alarm response and difficulty for operators to quickly assess the risk level and take corresponding measures.

[0090] This application further proposes obtaining an excess difference based on the vibration acceleration and the material tolerance threshold, where the excess difference is the difference between the vibration acceleration and the material tolerance threshold, determining an alarm level based on the excess difference and performing an audible and visual alarm, and the alarm level is proportional to the excess difference.

[0091] The excess difference is obtained by subtracting the real-time vibration acceleration from the material tolerance threshold. The larger the difference, the more the vibration intensity exceeds the material's tolerance. The alarm level is divided into multiple intervals. For example, a level 1 alarm is defined when the difference is between 0-10% of the threshold, a level 2 alarm is defined when the difference is between 10%-20% of the threshold, and a level 3 alarm is defined when the difference exceeds the 20% threshold. The sound and light alarm uses a multi-level signal output mode. The level 1 alarm corresponds to a low-frequency sound and yellow light, the level 2 alarm corresponds to a medium-frequency sound and orange light, and the level 3 alarm corresponds to a high-frequency sound and red light. The proportional relationship between the alarm level and the excess difference is achieved through linear mapping or piecewise function, ensuring that the alarm intensity increases with the degree of risk.

[0092] Specifically, when the vibration acceleration sensor detects real-time vibration data, it calculates the difference between the current vibration acceleration and the preset material tolerance threshold. If the difference is positive, the alarm level is determined based on the ratio of the absolute value of the difference to the threshold. For example, when the material tolerance threshold is 50m / s² and the real-time vibration acceleration is 60m / s², the excess difference is 10m / s², and the excess ratio is 20%, triggering a level 3 alarm. The sound and light alarm module activates a high-frequency buzzer and a red warning light according to the level 3 alarm instruction. The operator can immediately identify the high-risk state and take shutdown or speed reduction measures. If the excess difference is 5m / s², a level 1 alarm is triggered, and a low-frequency buzzer and a yellow warning light are activated, prompting the operator to adjust the drilling parameters to avoid escalating risks. Through the graded alarm mechanism, differentiated warning signals can be provided at different risk stages to optimize emergency response efficiency.

[0093] As a preferred embodiment, the solution of the present application is specifically implemented as follows: when the vibration acceleration exceeds the material tolerance threshold, the numerical difference between the current vibration acceleration and the material tolerance threshold is calculated, and the difference is divided into three level intervals. When the difference is in the first interval, a yellow warning light and a low-frequency buzzer are triggered; when the difference enters the second interval, it switches to an orange rotating warning light and a medium-frequency intermittent buzzer; when the difference reaches the third interval, a red strobe light and a high-frequency continuous alarm are activated. The alarm device has a built-in priority control module to ensure that high-level alarm signals can cover low-level states. The thresholds for each level are pre-set according to the Young's modulus and fatigue strength parameters of the drilling pipe, and the effectiveness of the grading is verified through field tests.

[0094] Through the above technical solution, this application effectively solves the problem of false alarms or response delays caused by the single threshold triggering of traditional early warning mechanisms. By dynamically quantifying the degree of vibration exceeding the standard and matching it with multi-level alarms, operators can accurately judge the severity of the vibration anomaly and take targeted tiered measures such as deceleration and emergency shutdown. This avoids drill pipe breakage or guide sleeve wear caused by excessive vibration, while also reducing operational interruptions caused by unnecessary downtime.

[0095] In summary, by real-time collection of drilling data, including parameters such as drilling pressure, drill bit torque, feed speed and vibration acceleration, real-time monitoring of the drilling operation status of the drilling rig is achieved, which can respond to changes in complex geological conditions in a timely manner, effectively improve drilling efficiency and reduce the occurrence of safety hazards; the coupling coefficient is obtained according to the drilling pressure, drill bit torque and feed speed, and whether to perform longitudinal compensation is determined based on the coupling coefficient. Combined with the comparison mechanism of the highest threshold of the coupling coefficient, the drill rod stuck state is accurately identified, and the accuracy and reliability of identification are improved; the sliding window technology is used to analyze the drilling data, extract the drilling feature vector, and input it into the multi-head attention model, realizing intelligent control of the drilling process, avoiding the problems of parameter and structure optimization of artificial intelligence and deep learning algorithms in traditional methods; improving the dynamic adjustment effect of drilling parameters and reducing the risk of stuck drill.

[0096] In another preferred embodiment based on the above embodiment, refer to Figure 2-4 As shown, this embodiment provides a large-diameter horizontal directional drilling device 110, which is used to apply the above-mentioned compensation control method of the large-diameter horizontal directional drilling device 110, including: Rack 100; A drilling device 110 is fixed to the upper surface of the frame 100 , and includes a drive motor 111 , a drill rod 112 , and a guide sleeve 113 ; A drilling compensation device 120 is fixed below the drilling device 110. The drilling compensation device 120 is fixedly connected to the drilling device 110 and includes a first cylinder 121. A clamping device 170 is provided between the driving motor 111 and the guide sleeve 113; Sensor module, including vibration sensor, speed sensor and pressure sensor; a drill rod adjusting device 130 , fixed between the frame 100 and the drilling device 110 , the drill rod adjusting device 130 comprising a second cylinder 131 ; an air pump 160 connected to the first cylinder 121 and the second cylinder 131; A supporting device 140 is provided around the frame 100, and the supporting device 140 includes a hydraulic lifting rod 141; The control device 150 is electrically connected to the drilling device 110, the drilling compensation device 120, the sensor module, the drill rod adjustment device 130 and the support device 140. The control device 150 is used to collect drilling data and control the first cylinder 121 and the second cylinder 131 to perform longitudinal compensation of the drill rod according to the drilling data.

[0097] Specifically, the frame 100 serves as the integral support structure, with the drilling device 110 secured to its surface. A drilling compensation device 120 is connected below the drilling device 110. A first cylinder 121 is rigidly connected to the drilling device 110, transmitting longitudinal compensatory thrust. Sensor modules are located at key locations on the drill rod 112. Vibration sensors monitor the vibration acceleration of the drill rod 112, speed sensors collect feed rate data, and pressure sensors measure the weight on bit. The second cylinder 131 of the drill rod adjustment device 130 is located between the frame 100 and the drilling device 110 and adjusts the position of the drill rod 112 through telescopic movement. An air pump 160 is connected to the first and second cylinders 121, 131, providing driving force. The hydraulic lift rods 141 of the support device 140 are located at the four corners of the frame 100 and maintain horizontal stability by adjusting the height. The control device 150 integrates a signal processing module and an actuator drive module, receiving sensor data in real time and generating control commands. A clamping device 170 is provided to further stabilize the drill rod 112.

[0098] Specifically, the vibration sensor, velocity sensor and pressure sensor continuously collect the vibration acceleration, feed speed and drilling pressure data of the drill pipe 112. At the same time, the velocity sensor can also collect vibration acceleration and transmit it to the control device 150. The control device 150 preprocesses the data, including filtering, normalization and outlier removal, and then extracts statistical features based on sliding window analysis. Based on the clustering comparison results of the feature vector and historical data, the control device 150 selects a pre-stored compensation strategy or starts a multi-head attention model to predict the drilling pressure trend. When the predicted drilling pressure exceeds the preset threshold, the control device 150 drives the first cylinder 121 to output the compensating thrust, and the second cylinder 131 can also be manually adjusted to adjust the longitudinal angle of the drill pipe. The hydraulic lifting rod 141 of the support device 140 dynamically adjusts the height according to the inclination of the frame 100 to offset the deflection caused by uneven formations. For example, the first cylinder 121 has a maximum thrust of 50kN and a response time of less than 100ms. The second cylinder 131 has a stroke of ±15° and an adjustment accuracy of 0.1°. The hydraulic lift rod 141 has a lifting speed of 5mm / s and a positioning error of no more than 0.5mm. Through the coordinated control of multiple actuators, longitudinal compensation and attitude adjustment of the drill rod are synchronized, effectively reducing the risk of drill sticking.

[0099] As a preferred embodiment, the solution of this application is specifically implemented as follows: the frame 100 utilizes a high-strength alloy steel welded frame structure with four sets of adjustable support legs mounted at the bottom. The drilling device 110 is fixed to the center of the upper surface of the frame 100 via a flange. The drive motor 111 utilizes a variable-frequency, speed-regulated three-phase asynchronous motor. The output shaft is connected to the drill rod 112 via a planetary reducer. The front end of the drill rod 112 is equipped with a carbide composite drill bit. The guide sleeve 113 is made of wear-resistant ceramic material and is mounted on the outside of the drill rod 112. The drilling compensation device 120 comprises two sets of first cylinders 121 mounted in parallel. The top ends of the cylinder piston rods are connected to the bottom of the drilling device 110 via an articulated mechanism. The cylinder stroke is 200-300 mm. The sensor module includes a triaxial vibration sensor, a photoelectric encoding velocity sensor, and a piezoresistive pressure sensor. The vibration sensor is mounted on the connection flange of the drill rod 112. The velocity sensor is integrated into the output end of the drive motor 111. The pressure sensor is located within the axial thrust bearing of the drill rod 112. The second cylinder 131 of the drill rod adjustment device 130 is vertically mounted between the side wall of the frame 100 and the drilling device 110, with the end of the cylinder piston rod connected to the outer casing of the drilling device 110 via a ball joint. The hydraulic lifting rod 141 of the support device 140 adopts a double-acting cylinder structure and is distributed at the four corners of the frame 100. A pressure-balancing load plate is installed on the top of the cylinder. The control device 150 uses an industrial-grade embedded controller, which is connected to the sensor module via a CAN bus to collect vibration frequency, axial displacement, and pressure data in real time. The control logic generates compensation instructions based on the fuzzy PID algorithm, driving the first cylinder 121 and the second cylinder 131 to work together to achieve longitudinal position compensation of the drill rod.

[0100] It is understandable that by real-time collection of drilling data, including parameters such as drilling pressure, drill bit torque, feed speed and vibration acceleration, real-time monitoring of the drilling operation status of the drilling rig is achieved, which can respond to changes under complex geological conditions in a timely manner, effectively improve drilling efficiency and reduce the occurrence of safety hazards; the coupling coefficient is obtained according to the drilling pressure, drill bit torque and feed speed, and whether to perform longitudinal compensation is determined based on the coupling coefficient. Combined with the comparison mechanism of the highest threshold of the coupling coefficient, the drill rod stuck state is accurately identified, and the accuracy and reliability of identification are improved; the sliding window technology is used to analyze the drilling data, extract the drilling feature vector, and input it into the multi-head attention model, realizing intelligent control of the drilling process, avoiding the problems of parameter and structure optimization of artificial intelligence and deep learning algorithms in traditional methods; improving the dynamic adjustment effect of drilling parameters and reducing the risk of stuck drills.

[0101] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0102] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0103] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0104] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0105] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A compensation control method for a large-caliber horizontal directional drilling device, characterized in that: include: Collecting drilling data, including drilling weight, drill bit torque, feed speed, and vibration acceleration; obtaining a coupling coefficient according to the bit weight, drill bit torque and feed speed, and determining whether to perform longitudinal compensation according to the coupling coefficient; When it is determined that longitudinal compensation is to be performed, drilling data within a preset time period is collected, the drilling data is analyzed based on a sliding window to obtain a drilling feature vector, the drilling feature vector is input into a multi-head attention model to obtain a predicted weight-on-bit sequence, and a pre-adjusted control amount is obtained based on the predicted weight-on-bit sequence, the pre-adjusted control amount including a thrust compensation amount; When the preset control amount is in operation, real-time vibration acceleration is collected, the vibration acceleration is compared with the material tolerance threshold, and whether to issue an early warning is determined according to the comparison result, and an audible and visual alarm is issued according to the early warning result.

2. The compensation control method for large-diameter horizontal directional drilling equipment according to claim 1, characterized in that: When determining whether to perform longitudinal compensation according to the coupling coefficient, the method includes: Comparing the coupling coefficient with a minimum coupling coefficient threshold and a maximum coupling coefficient threshold respectively, and determining whether to perform longitudinal compensation according to the comparison results; When the coupling coefficient is less than a minimum coupling coefficient threshold or the coupling coefficient is greater than a maximum coupling coefficient threshold, determining to perform longitudinal compensation; When the coupling coefficient is greater than or equal to a minimum coupling coefficient threshold and the coupling coefficient is less than or equal to a maximum coupling coefficient threshold, it is determined that longitudinal compensation is not performed.

3. The compensation control method for large-diameter horizontal directional drilling equipment according to claim 1, characterized in that: Collecting drilling data within a preset time period, analyzing the drilling data based on a sliding window, and obtaining a drilling feature vector includes: Synchronous sampling is performed within a preset time period to collect drilling data, and the collected data is preprocessed. The preprocessing includes missing value filling, normalization, and error value removal. The preprocessed data is divided into a number of sliding time windows. Statistical features are extracted for each time window to form the drilling feature vector. The statistical features include mean, standard deviation, range, and first-order difference mean. The drilling feature vector is an input matrix of T×D dimensions, where T represents the time step and D represents the number of input features.

4. The compensation control method for large-diameter horizontal directional drilling equipment according to claim 3, characterized in that: Also includes: After obtaining the drilling feature vector, it also includes: Performing cluster analysis on the drilling feature vector and the historical compensation vector set, and determining a method for obtaining the predicted weight-on-bit sequence based on the analysis results; the historical compensation vector set includes a plurality of historical drilling feature vectors and a plurality of historical predicted weight-on-bit sequences, and each of the historical drilling feature vectors corresponds to a historical predicted weight-on-bit sequence; Taking the historical drilling feature vectors in the historical compensation vector set as sample points, setting the minimum sample number parameter MinPts and the neighborhood radius ϵ; Calculating the density of the neighborhood of each sample point; The samples are divided into several density clusters according to the density reachability rule. Each density cluster represents a drilling feature, and all points that do not meet the density condition are marked as outliers. The outliers do not participate in the current analysis. Putting the drilling feature vector into the historical density cluster model to obtain a cluster comparison result; When the density cluster containing the drilling feature vector includes the historical drilling feature vector, obtaining the predicted weight-on-bit sequence according to the historical predicted weight-on-bit sequence corresponding to the historical drilling feature vector; When the density cluster containing the drilling feature vector does not include the historical drilling feature vector, the drilling feature vector is input into a multi-head attention model to obtain the predicted weight on bit sequence.

5. The compensation control method for large-diameter horizontal directional drilling equipment according to claim 4, characterized in that: Obtaining the predicted weight-on-bit sequence according to the historical predicted weight-on-bit sequence corresponding to the historical drilling feature vector includes: When the density cluster containing the drilling feature vector contains only one historical drilling feature vector, the historical predicted weight-on-bit sequence corresponding to the historical drilling feature vector is used as the predicted weight-on-bit sequence; When the density cluster containing the drilling feature vector contains two or more historical drilling feature vectors, the similarity between the drilling feature vector and each of the historical drilling feature vectors is obtained, and the historical predicted weight-on-bit sequence corresponding to the maximum similarity is used as the predicted weight-on-bit sequence.

6. The compensation control method for large-diameter horizontal directional drilling equipment according to claim 5, characterized in that: Inputting the drilling feature vector into the multi-head attention model to obtain the predicted weight on bit sequence includes: The multi-head attention model adopts a Transformer encoder structure. Each attention head is weighted based on the interactive correlation between drilling pressure, drill torque, feed speed and vibration acceleration, and outputs a fused feature representation.

7. The compensation control method for large-diameter horizontal directional drilling equipment according to claim 6, characterized in that: The predicted WOB sequence generated by the multi-head attention model is a sequence of predicted WOB values ​​for a number of time steps in a preset future period. The WOB trend change and WOB gradient are calculated based on the WOB change trend and gradient change in the predicted WOB sequence, and the pre-adjusted control variable is obtained based on the WOB trend change and WOB gradient.

8. The compensation control method for large-diameter horizontal directional drilling equipment according to claim 7, characterized in that: When determining whether to issue an early warning based on the comparison results, the following are included: When the vibration acceleration is greater than the material tolerance threshold, it is determined to issue an early warning; When the vibration acceleration is less than or equal to the material tolerance threshold, it is determined that no warning is issued.

9. The compensation control method for large-diameter horizontal directional drilling equipment according to claim 8, characterized in that: When sound and light alarms are issued based on the early warning results, it includes: An excess difference is obtained based on the vibration acceleration and the material tolerance threshold, and the excess difference is the difference between the vibration acceleration and the material tolerance threshold. An alarm level is determined based on the excess difference and an audible and visual alarm is performed. The alarm level is proportional to the excess difference.

10. A large-diameter horizontal directional drilling device, used for applying the large-diameter horizontal directional drilling device compensation control method according to any one of claims 1 to 9, characterized in that: include: frame; A drilling device is fixed to the upper surface of the frame, and the drilling device includes a drive motor, a drill rod and a guide sleeve; A drilling compensation device is fixed below the drilling device, the drilling compensation device is fixedly connected to the drilling device, and the drilling compensation device includes a first cylinder; a clamping device, disposed between the drive motor and the guide sleeve; Sensor module, including vibration sensor, speed sensor and pressure sensor; a drill rod adjustment device fixed between the frame and the drilling device, the drill rod adjustment device comprising a second cylinder; an air pump connected to the first cylinder and the second cylinder; A supporting device is arranged around the frame, and the supporting device includes a hydraulic lifting rod; A control device is electrically connected to the drilling device, the drilling compensation device, the sensor module, the drill rod adjustment device and the support device. The control device is used to collect drilling data and control the first cylinder and the second cylinder to perform longitudinal compensation of the drill rod according to the drilling data.

Images