A downhole directional long borehole while-drilling CO concentration prediction and grading early warning method

Abstract

The application discloses a kind of downhole directional long borehole while drilling CO concentration prediction and grading early warning method, it is related to coal mine safety and gas disaster early warning technical field, the method includes: to the speed of drilling machine, drilling speed, drilling time length and drill bit temperature processing, constructs coal temperature and CO concentration double output's multitask least square support vector machine model, searches penalty factor and kernel function width;Introduce multistage CO concentration threshold to carry out multistage discrimination, statistics early warning hit rate and false alarm rate, realize normal, grading evaluation from early warning to high risk.According to CO concentration time variation curve, linear fitting estimation reaches the time and time margin of high risk threshold value.Polynomial regression CO model is established, and parameter space is randomly searched to inverse dangerous parameter combination, and sensitivity analysis gives the parameter range for avoiding CO overrun.The application combines multitask learning, time development analysis and parameter optimization, realizes the whole process quantification management of CO risk from early warning, time prediction to parameter control.

Description

Technical Field

[0001] This invention relates to the field of coal mine safety and gas disaster early warning technology, and in particular to a method for predicting and classifying CO concentration during drilling in underground directional long boreholes. Background Technology

[0002] Directional long-bore drilling for pre-drainage of gas is an effective means of preventing gas outbursts. When drilling in soft, outburst-prone coal seams, dry drilling with pneumatic slag removal is commonly used. This method has advantages such as high hole formation rate and less gas accumulation within the borehole. However, the rapid rotation of the drill bit and continuous friction with the coal body cause a rapid increase in coal temperature. Furthermore, incomplete combustion of coal powder in the borehole releases a large amount of CO gas. With the increasing intensity of coal mining and the frequency of drilling operations, the elevated CO concentration generated during drilling has become a significant hidden danger affecting the safety of workers and the working environment. Existing underground gas and CO monitoring systems in coal mines mostly rely on real-time concentration detection by fixed sensors. An alarm is triggered when the CO concentration reaches or approaches the alarm threshold, representing a passive protection mode that lacks prior warning and relies on in-process alarms. This makes it difficult to provide timely information on CO concentration trends and allow for early intervention.

[0003] In the prior art, Chinese patent application CN202310649608.X discloses a monitoring method for a coal mine directional drilling rig borehole temperature and CO intelligent monitoring system. This method uses multi-parameter sensors for temperature and CO concentration to acquire data such as coal body temperature, temperature change, contrast, and CO concentration during the drilling process. The data is transmitted to the monitoring device via a fiber optic demodulator and a MESH network. The intelligent early warning unit analyzes and compares the collected data. When it is determined that there may be an anomaly or potential accident, it links a camera device to acquire on-site video information, performs comprehensive analysis of the accident point data, identifies the accident type and severity, and issues an alarm. At the same time, it establishes a mapping relationship between directional drilling disaster early warning indicators and disaster severity based on big data methods. Chinese patent application CN202510968015.X discloses a CO and temperature monitoring and control system for directional long boreholes in soft coal seams. The system uses a dedicated acquisition device protection and installation structure to ensure reliable deployment and protection of CO and temperature sensors under drilling conditions. The data acquisition and transmission unit acquires and uploads CO concentration and temperature data in real time. The data analysis and processing unit performs trend prediction, automatic calibration, and fault diagnosis on the monitoring data. Finally, the intelligent linkage slag removal and cooling unit automatically adjusts the air compressor intake and slag removal and cooling intensity according to the temperature gradient change rules.

[0004] However, the aforementioned existing technologies mainly focus on building a monitoring and early warning system that integrates multi-parameter acquisition, network transmission, and video linkage. They establish the correspondence between disaster indicators and accident levels through big data methods, or focus on solving the linkage control problems of physical protection and temperature threshold driven by the drilling monitoring device. They pay attention to the reliability of CO and temperature signal acquisition and temperature control strategies, but do not specifically reveal the quantitative prediction modeling, over-limit time assessment, and parameter optimization mechanism between CO concentration and drilling parameters. They still lack multi-output prediction models for CO concentration, graded early warning criteria, and dangerous working condition inversion methods. They are insufficient in terms of fine prediction of CO concentration, quantitative estimation of over-limit time, and optimization of adjustable parameters, making it difficult to provide a systematic and quantifiable technical approach for CO risk prediction and prevention in directional long borehole construction.

[0005] Therefore, there is an urgent need for a method for predicting and classifying CO concentration during drilling in downhole directional long boreholes. Summary of the Invention

[0006] This invention provides a method for predicting and classifying CO concentration during drilling in downhole directional long boreholes, in order to solve the problems of insufficient CO concentration prediction accuracy, limited early warning information, and lack of hazardous condition inversion and parameter sensitivity analysis in the prior art.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A method for predicting and classifying CO concentration during drilling in downhole directional long boreholes includes:

[0009] S1: Collect data on drilling rig speed, drilling speed, drilling time, drill bit temperature, coal body temperature and CO concentration under different working conditions, preprocess the numerical features, and construct sample feature matrix and labels;

[0010] S2: Using drilling rig speed, drilling speed, drilling time and drill bit temperature as inputs, and coal temperature and CO concentration as dual outputs, a multi-task least squares support vector machine model is established, and the model parameters are optimized to minimize the comprehensive prediction error.

[0011] S3: Compare the CO concentration predicted by the trained model with the set threshold to achieve graded early warning;

[0012] S4: Analyze the CO concentration time series under different operating conditions and calculate the time margin for the CO concentration to reach the preset threshold.

[0013] S5: Construct a regression model with operating parameters as input and CO concentration as output;

[0014] S6: Search within the parameter space for parameter combinations that cause the predicted CO concentration to exceed a preset threshold, and generate a set of hazardous operating condition parameters.

[0015] S7: Normalize and analyze the changes in each control quantity in the set of parameters for hazardous operating conditions to determine the safe range of change of the control quantity.

[0016] Furthermore, step S1 includes:

[0017] S11: During the drilling process, collect real-time data on drilling speed, drilling speed, drilling time, drill bit temperature, coal body temperature and CO concentration, and set working condition numbers according to drilling speed, coal sample type and loading conditions.

[0018] S12: Identify outliers in real-time data, fill missing sample points with time series interpolation, and replace isolated outliers with the mean of a time window.

[0019] S13: Standardize continuous variables other than working condition number by subtracting the mean from the original value and then dividing by the standard deviation. Divide the training set and test set according to a preset ratio to construct a sample feature matrix and corresponding coal body temperature and CO concentration labels.

[0020] Furthermore, step S2 includes:

[0021] S21: Based on a standardized sample set, the input vectors are set as drilling rig speed, drilling speed, drilling time and drill bit temperature, and the output vectors are coal temperature and CO concentration. A nonlinear mapping relationship is established.

[0022] S22: Using radial basis functions as kernel functions, an optimization objective function containing penalty factors and random errors is constructed, and Lagrange multipliers are introduced to transform the optimization problem into a system of linear equations for solution;

[0023] S23: The whale optimization algorithm is used to iteratively search within the range of penalty factor γ (10 to 1000) and kernel function width σ (0.01 to 1.0) to minimize the comprehensive error of coal temperature and CO concentration prediction, and generate a trained dual-output prediction model.

[0024] Furthermore, in step S2, the fitness function of the whale optimization algorithm adopts the weighted sum of the root mean square errors of coal body temperature and CO concentration predictions. The weights are set according to the importance of the two outputs, so that the weight of CO concentration prediction is not lower than the weight of coal body temperature prediction.

[0025] Furthermore, step S3 includes:

[0026] S31: Based on the dual-output prediction model, the test set is predicted to obtain the predicted coal temperature and CO concentration values;

[0027] S32: Set the warning threshold and the poisoning risk threshold, calculate the corresponding value of the CO concentration threshold in the standardized space, compare the predicted CO concentration with the warning threshold and the poisoning risk threshold respectively, and mark the samples that exceed the warning threshold as warning samples and the samples that exceed the poisoning risk threshold as high-risk samples.

[0028] S33: Based on the measured CO concentration, count the number of true positive, false positive, false negative and true negative samples under different operating conditions and different threshold levels, calculate the early warning hit rate according to the ratio of true positive to the sum of true positive and false negative, calculate the false alarm rate according to the ratio of false positive to the sum of false positive and true negative, and generate graded early warning performance indicators.

[0029] Furthermore, the graded early warning discrimination in S3 includes: when the model predicts that the CO concentration exceeds the early warning threshold but is below the poisoning risk threshold, an early warning signal is output and the corresponding working condition number and drilling parameters are recorded; when the predicted CO concentration exceeds the poisoning risk threshold, a poisoning risk alarm signal is output and the recording and subsequent analysis of the dangerous parameter combination are triggered.

[0030] Furthermore, step S4 includes:

[0031] S41: Extract the CO concentration time series under different operating conditions to confirm that the CO concentration in the monitoring interval shows a monotonically increasing and approximately linear change over time.

[0032] S42: A first-order linear function was fitted to the relationship between CO concentration and time using the least squares method. ,in The slope Using the intercept, calculate the theoretical time for CO concentration to reach the poisoning risk threshold under this linear trend assumption;

[0033] S43: Calculate the time difference between the theoretical time and the current drilling end time as the time margin. Sort the theoretical arrival time and time margin obtained for different working conditions in ascending order. Mark the working condition with the smallest time margin as the priority dangerous trend working condition and generate the time margin ranking result.

[0034] Furthermore, step S5 includes:

[0035] S51: Construct a feature matrix with working condition number, drilling speed, drilling speed, drilling time and drill bit temperature as input features and CO concentration as output target;

[0036] S52: Perform second-order polynomial expansion on the feature matrix to generate an expanded feature matrix containing the original features, quadratic terms, and cross terms. Use linear regression to fit the relationship between the expanded features and CO concentration. The polynomial order is 2. The polynomial expansion does not include bias terms to avoid overfitting caused by parameter redundancy.

[0037] S53: Verify the fitting results and generate a parameter space search calculation model.

[0038] Furthermore, step S6 includes:

[0039] S61: Based on the multinomial regression model, keeping the working condition number unchanged, and on the basis of the global maximum value of historical data for different working conditions, the upper limit values ​​of drilling rig speed, drilling speed and drill bit temperature are increased by a preset multiple, and the upper limit value of drilling time is increased by a preset additional drilling time to cover the potentially dangerous operating range and generate the boundary of the four-dimensional parameter space.

[0040] S62: Generate parameter combinations in the four-dimensional parameter space according to a preset step size or a random method. Substitute each parameter combination into a multinomial regression model to calculate the predicted CO concentration. When the predicted CO concentration reaches or exceeds the poisoning risk threshold, record the parameter combination.

[0041] S63: The difference between each control quantity in the parameter combination and the corresponding control quantity in the original working condition is used as the change quantity. When the number of samplings reaches the preset upper limit or when at least one set of parameter combinations that makes the CO concentration exceed the poisoning risk threshold is found in all working conditions, the search stops and a dangerous working condition parameter set is generated.

[0042] Furthermore, step S7 includes:

[0043] S71: Extract the changes in hazardous operating condition parameters, including drilling rig speed, drilling speed, drilling time, and drill bit temperature, and normalize them according to their global range in historical data.

[0044] S72: Calculate the sum of squares of the normalized changes of the four control variables under each working condition as a comprehensive parameter change index from the current working condition to the dangerous working condition. Compare the sum of squares and the normalized changes of each individual variable under different working conditions to quantify the overall parameter adjustment range required for different working conditions to reach the warning threshold.

[0045] S73: Based on the comparison results, determine the maximum permissible range of each control quantity required to keep the CO concentration below the warning threshold, and generate the safe range of control quantity changes.

[0046] Compared with the prior art, the present invention has the following advantages:

[0047] This invention provides a method for predicting and classifying CO concentration during drilling in downhole directional long boreholes. Based on fully utilizing multi-source monitoring information such as drilling rig speed, drilling speed, drilling time, drill bit temperature, and coal body temperature under the working condition number, this method comprehensively employs multi-task least squares support vector machine, whale optimization algorithm, CO concentration time linear extrapolation, and multiple regression and parameter space search to construct a full-chain, quantifiable, multi-level CO risk assessment system that includes "numerical prediction → threshold early warning → time margin assessment → hazardous working condition inversion and parameter sensitivity analysis".

[0048] This invention aims to improve the accuracy and stability of joint prediction of CO concentration and coal temperature by combining multi-task LSSVM and WOA. By setting warning thresholds and poisoning thresholds referencing the "Occupational Exposure Limits for Hazardous Factors in the Workplace" (GBZ 2.1-2019), the model prediction results are graded and judged, and the hit rate and false alarm rate under different thresholds are quantified, realizing a systematic evaluation of CO warning performance. By linearly fitting the measured CO concentration time curves under different working conditions, the theoretical time for CO to reach the poisoning threshold and the time margin from the current moment to the poisoning threshold are obtained, reserving time for risk disposal on site. Furthermore, a polynomial regression CO model is constructed in the original parameter space, and by using parameter space search and normalized change analysis, the combination of hazardous parameters leading to CO exceeding the limit and the sensitivity of each control quantity are obtained, providing a basis for determining the safe control range of drilling rig speed, drilling speed, drilling time, and drill bit temperature. Attached Figure Description

[0049] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0050] Figure 1 This is a flowchart of a method for predicting and classifying CO concentration during drilling in a downhole directional long borehole, as described in an embodiment of the present invention.

[0051] Figure 2 This is a schematic diagram of the integrated model for CO borehole concentration prediction, early warning, and prevention in an embodiment of the present invention;

[0052] Figure 3 This is a schematic diagram showing the original characteristic distribution of variables such as drilling speed, drilling speed, drilling time, coal temperature and CO concentration in an embodiment of the present invention;

[0053] Figure 4 This is a convergence curve of the fitness of the multi-task WOA LSSVM model as a function of the number of iterations in an embodiment of the present invention.

[0054] Figure 5This is a curve comparing the measured value of the coal body temperature with the predicted value of the multi-task WOA LSSVM in an embodiment of the present invention;

[0055] Figure 6 This is a curve comparing the measured CO concentration values ​​of the test set with the predicted values ​​of the multi-task WOA LSSVM in an embodiment of the present invention.

[0056] Figure 7 This is a schematic diagram illustrating the relationship between the over-limit warning hit rate and false alarm rate under different CO warning thresholds in this embodiment of the invention;

[0057] Figure 8 This is a schematic diagram showing the evolution trend of CO concentration over time under different operating conditions and the linear extrapolation to 50 ppm in the embodiments of the present invention. The linear fitting curves in the figure all meet the goodness of fit requirement of R²≥0.95.

[0058] Figure 9 This is a bar chart comparing the sensitivity of normalized changes in drilling rig speed, drilling speed, drilling time and drill bit temperature under different working conditions when reaching 50ppm in the embodiments of the present invention.

[0059] Figure 10 This is a schematic diagram of the overall technical route for the joint prediction of coal body temperature and CO concentration using multi-task WOA LSSVM, as well as the multi-threshold early warning and hazard parameter inversion in this embodiment of the invention. Detailed Implementation

[0060] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0061] The embodiments of the present invention provide, as follows Figure 1 As shown, a method for predicting and classifying CO concentration during drilling in downhole directional long boreholes includes:

[0062] S1: Collect data on drilling rig speed, drilling speed, drilling time, drill bit temperature, coal body temperature and CO concentration under different working conditions, preprocess the numerical features, and construct sample feature matrix and labels;

[0063] S2: Using drilling rig speed, drilling speed, drilling time and drill bit temperature as inputs, and coal temperature and CO concentration as dual outputs, a multi-task least squares support vector machine model is established, and the model parameters are optimized to minimize the comprehensive prediction error.

[0064] S3: Compare the CO concentration predicted by the trained model with the set threshold to achieve graded early warning;

[0065] S4: Analyze the CO concentration time series under different operating conditions and calculate the time margin for the CO concentration to reach the preset threshold.

[0066] S5: Construct a regression model with operating parameters as input and CO concentration as output;

[0067] S6: Search within the parameter space for parameter combinations that cause the predicted CO concentration to exceed a preset threshold, and generate a set of hazardous operating condition parameters.

[0068] S7: Normalize and analyze the changes in each control quantity in the set of parameters for hazardous operating conditions to determine the safe range of change of the control quantity.

[0069] The following is a detailed description with reference to specific embodiments.

[0070] like Figure 2 As shown, this invention provides a method for predicting and classifying CO concentration during drilling in downhole directional long boreholes, comprising the following steps:

[0071] Step S1: Data Acquisition and Preprocessing

[0072] S11: Real-time acquisition of multi-source data

[0073] During the directional long borehole drilling operation, speed sensors, displacement sensors, temperature sensors, and gas sensors are deployed near the drilling rig, drill bit, and borehole opening to collect real-time data on drilling rig speed n, drilling speed v, drilling time t, drill bit temperature T, and coal seam temperature. and CO concentration Six time-series data items were collected. The tests were numbered according to drilling speed, coal sample type, and loading conditions, and operating condition numbers were set as identifiers for combinations of different drilling speeds, coal sample types, and loading conditions. This embodiment set up 11 typical operating condition types, with operating condition numbers corresponding one-to-one with test numbers. A total of 760 valid samples were obtained in the experiment. Figure 3 The diagram shows the original characteristic distribution of variables such as drilling rig rotation speed, drilling speed, drilling time, coal temperature, and CO concentration.

[0074] S12: Outlier Identification and Handling

[0075] Outlier identification and missing value handling were performed on monitoring sequences under different operating conditions. For short-term missing sample points, time series smoothing imputation was used to estimate the missing values, specifically using triple exponential smoothing. For isolated outliers, the mean within the time window was used for replacement to ensure data continuity and reliability.

[0076] S13: Data Standardization and Sample Set Construction

[0077] To eliminate the influence of differences in the units and magnitudes of different variables, Z-score standardization was applied to continuous variables other than the operating condition number. For any original sample x, the standardization result was calculated using the following formula:

[0078] (1)

[0079] Where x represents the original data, μ is the mean of the dataset, and σ is the standard deviation of the dataset, measuring the dispersion of the data. After cleaning and standardization, the 760 samples are divided into training and testing sets according to a preset ratio. In this embodiment, 532 sets are used for training and 228 sets are used for testing, constructing a sample feature matrix and corresponding coal body temperature and CO concentration labels.

[0080] Step S2: Multi-task least squares support vector machine modeling

[0081] S21: Establish nonlinear mapping relationships

[0082] Based on a standardized sample set, the input vector x is defined as the drilling rig rotation speed n, drilling speed v, drilling time t, and drill bit temperature T, and the output vector is the coal body temperature. and CO concentration Establish a nonlinear mapping relationship. Let the nonlinear eigenmapper be φ(x), and the regression function of the temperature or CO sub-model be:

[0083] (2)

[0084] Where w is the weight vector and b is the bias.

[0085] S22: Constructing the optimization objective function

[0086] The radial basis function (RBF) is used as the kernel function, as the RBF kernel function performs excellently in the prediction field due to its robustness. Its mathematical expression is as follows:

[0087] (3)

[0088] in The kernel width determines the regression characteristics and generalization ability of the model.

[0089] Construct a system containing a penalty factor γ and random error The optimization objective function is:

[0090] (4)

[0091] In the formula: γ is the penalty factor; This is to address random error. This is achieved by introducing Lagrange multipliers. :

[0092] (5)

[0093] The optimization problem is transformed into a system of linear equations to be solved.

[0094] S23: Parameter Optimization and Model Training

[0095] In multi-task modeling, the root mean square error of coal body temperature prediction is used. and the root mean square error of CO concentration prediction A combined fitness function is constructed by incorporating both coal body temperature and CO concentration prediction errors into the objective function. A dual-output combined fitness function is designed, incorporating these errors in a weighted manner into the objective function.

[0096] (6)

[0097] Where γ and σ are the penalty factor and kernel width of the least squares support vector machine, respectively. The whale optimization algorithm is used to iteratively search within the range of penalty factor γ from 10 to 1000 and kernel width σ from 0.01 to 1.0 to minimize equation (6), thereby improving the prediction accuracy of the two output variables, coal body temperature and CO concentration, under a unified parameter combination.

[0098] like Figure 4 The figure shows the convergence curve of the fitness of the multi-task model as a function of the number of iterations. The final global optimal parameters are γ=500.0 and σ=0.10, generating a trained dual-output prediction model.

[0099] To quantitatively evaluate the model's prediction accuracy, the root mean square error (RMSE) and coefficient of determination (R²) are calculated for both the training and test sets. In this embodiment, RMSE and R² are defined as follows:

[0100] Goodness of fit: (7)

[0101] Root mean square error: (8)

[0102] in, These are measured values. For predicted values, is the sample mean, and N is the sample size.

[0103] Table 1 below (Table 1 is a statistical table of the prediction performance of the multi-task WOA-LSSVM on the training and test sets in this embodiment of the invention, including TrainRMSE, TestRMSE, Train R², and Test R²) summarizes the RMSE and R² indices of the multi-task model on the training and test sets. The results show that the Train RMSE for coal temperature is approximately 0.0015, and the Test RMSE is approximately 0.1330, corresponding to a Train R² of approximately 1.00000 and a Test R² of approximately 0.98265; the Train RMSE for CO concentration is approximately 0.0026, and the Test RMSE is approximately 0.0994, corresponding to a Train R² of approximately 0.99999 and a Test R² of approximately 0.99041.

[0104] Table 1

[0105] index variable Train RMSE Test RMSE Train R2 Test R2 Multi-tasking WOA-LSSVM coal temperature 0.0015 0.1330 1.00000 0.98265 Multi-tasking WOA-LSSVM CO concentration 0.0026 0.0994 0.99999 0.99041

[0106] like Figure 5 and Figure 6 As shown, the temperature and CO prediction curves output by the multi-task model are basically consistent with the measured curves in terms of peak position and trend, with only slight deviations near a few sharp peaks. A comparison of the bars in Table 2 (a comparison of RMSE and R² indices between single-task WOA-LSSVM and multi-task WOA-LSSVM in coal temperature prediction in this embodiment of the invention), Table 3 (a comparison of RMSE and R² indices between single-task WOA-LSSVM and multi-task WOA-LSSVM in CO concentration prediction in this embodiment of the invention), and the figures shows that, under the same search space and data partitioning, the multi-task model and the two single-task models almost overlap in RMSE and R², indicating that the multi-task framework achieves coordinated optimization of both temperature and CO outputs while ensuring prediction accuracy.

[0107] Table 2

[0108] Model type Dataset RMSE R2 Single task Train 0.0015 1.00000 Multitasking Train 0.0015 1.00000 Single task Test 0.1330 0.98265 Multitasking Test 0.1330 0.98265

[0109] Table 3

[0110] Model type Dataset RMSE R2 Single task Train 0.0026 0.99999 Multitasking Train 0.0026 0.99999 Single task Test 0.0994 0.99041 Multitasking Test 0.0994 0.99041

[0111] Step S3: Graded early warning judgment based on CO concentration threshold

[0112] S31: Model Prediction

[0113] The test set is predicted using a dual-output prediction model to obtain the predicted coal temperature and CO concentration values.

[0114] S32: Threshold Setting and Comparison

[0115] To ensure that the model output directly correlates with human health risks, based on mine safety regulations and the limits of CO toxicity to humans, this embodiment uses a CO concentration of 50 ppm as the poisoning risk threshold and 24 ppm as the warning threshold. Let the CO concentration threshold be... The mean of the original CO concentration in the training samples is The standard deviation is Then the corresponding value of this threshold in the standardized space is:

[0116] (9)

[0117] Equation (9) allows for direct threshold discrimination within the standardized model output space. The predicted CO concentration is compared with both the warning threshold and the poisoning risk threshold. Samples exceeding the warning threshold are designated as warning samples, while samples exceeding the poisoning risk threshold are designated as high-risk samples.

[0118] When the model predicts that the CO concentration exceeds the warning threshold of 24 ppm but is below the poisoning risk threshold of 50 ppm, a warning signal is output and the corresponding working condition number and drilling parameters are recorded; when the predicted CO concentration exceeds the poisoning risk threshold of 50 ppm, a poisoning risk alarm signal is output and the recording and subsequent analysis of dangerous parameter combinations are triggered.

[0119] S33: Early Warning Performance Evaluation

[0120] Based on measured CO concentration, the number of true positive (TP), false positive (FP), false negative (FN), and true negative (TN) samples was counted under different operating conditions and threshold levels. The early warning hit rate (Recall) was defined as:

[0121] (10)

[0122] False Alarm Rate (FAR) is defined as:

[0123] (11)

[0124] By calculating equations (10) and (11) under different warning thresholds, the impact of threshold selection on sensitivity and reliability can be analyzed, and a multi-level warning level suitable for field application can be determined accordingly, generating graded warning performance indicators.

[0125] At a threshold of 50 ppm, the statistical results of the 228 test samples are shown in Table 4 (Table 4 is the overall warning confusion matrix statistics table for the 50 ppm CO concentration threshold in this embodiment of the invention, including TP, FP, FN, and TN). TP=0, FP=0, FN=0, TN=228, and the overall false alarm rate is 0. This indicates that within the operating conditions of this experiment, the CO concentration is generally within a safe range and has not yet reached a level that poses an acute poisoning hazard to the human body.

[0126] Table 4

[0127] Statistical results meaning True case TP=0 Both measured and predicted limits were exceeded. False positives FP=0 Actual measurement did not exceed the limit, but the predicted limit was exceeded. False negatives FN=0 Actual measurement exceeded the limit, but prediction did not. True negative example TN=228 Neither the actual measurement nor the prediction exceeded the limit.

[0128] At the 24 ppm threshold, the statistical results are shown in Table 5 (Table 5 is a statistical table of the Recall and False Alarm Rate for CO concentration exceeding the limit at two thresholds of 24 ppm and 50 ppm in this embodiment of the invention): TP=12, FP=0, FN=1, TN=215, corresponding to a Recall of approximately 0.923 and a FAR of 0. Figure 7 The relationship between hit rate and false alarm rate shown indicates that at this threshold level, the multi-task model can capture most CO concentration increase conditions with virtually no false alarms, making it suitable as an alert-level warning threshold.

[0129] Table 5

[0130] ThresholdPPM TP FP FN TN Recall FalseAlarmRate 24 12 0 1 215 0.9230769230769231 0 50 0 0 0 228 No samples exceeding the limit 0

[0131] Step S4: Linear extrapolation analysis of CO concentration over time

[0132] S41: Time Series Extraction and Validation

[0133] The CO concentration time series under different operating conditions were extracted, confirming that the CO concentration in the monitoring interval showed a monotonically increasing trend over time and an approximately linear change.

[0134] S42: Linear Fitting and Theoretical Time Calculation

[0135] For the original CO concentration time series under different operating conditions, the least squares method is used to fit a first-order linear function to the relationship between CO concentration and time. Let the approximate linear relationship between CO concentration and time under a certain operating condition be:

[0136] (12)

[0137] Where a is the slope and b is the intercept. Under this linear trend assumption, the theoretical time t_{50} for the CO concentration to reach the poisoning risk threshold of 50 ppm is calculated using the following formula:

[0138] (13)

[0139] S43: Time Margin Calculation and Hazardous Condition Identification

[0140] If the current drilling end time is The time margin is calculated as the time difference between the theoretical time and the current drilling end time. :

[0141]

[0142] The theoretical arrival time and time margin obtained for different working conditions are sorted in ascending order, and the working condition with the smallest time margin is marked as the priority dangerous trend working condition, generating the time margin ranking result.

[0143] In this embodiment, based on equations (12) and (13), the CO concentration-time curves for 11 operating conditions are linearly fitted and extrapolated. The time margin results are listed in Table 6 (Table 6 shows the extrapolated time to reach 50 ppm under different operating conditions obtained based on CO-time linear fitting in the embodiments of the present invention). and remaining time margin (Statistical table), see corresponding curve examples. Figure 8 .like Figure 8 The figure shows the evolution trend of CO concentration over time under different operating conditions and the linear extrapolation to 50 ppm. The linear fitting curves in the figure all meet the goodness of fit requirement of R²≥0.95.

[0144] Table 6

[0145] Operating conditions slope_ppm_per_s intercept_ppm co_max_ppm t_last_s t_50ppm_linear_s interval_50ppm_linear_s reachable_by_linear_trend 1 0.10440242704354469 -1.335490387 6.2 66 491.7078255771138 425.7078255771138 TRUE 2 0.12540396606334842 -2.286233401 11.7 100 416.94242249718 316.94242249718 TRUE 3 0.1477501564356436 -6.152725743 29.6 200 380.05188689619445 180.05188689619445 TRUE 4 0.07364517188693659 -0.93427479 4.3 66 691.6172979827188 625.6172979827188 TRUE 5 0.08331569547511314 -1.358823793 7.74567 100 616.4363569251447 516.4363569251447 TRUE 6 0.0655178958066395 -3.006706016 13.5746 200 809.0416421910772 609.0416421910772 TRUE 7 0.05477541100076394 -0.730780034 3.4 66 926.1597331128362 860.1597331128362 TRUE 8 0.04957731538461539 -0.801924397 4.60906 100 1024.7009948514956 924.7009948514956 TRUE 9 0.037383474665113575 -1.702870041 8.26181 200 1383.0407821619142 1183.0407821619142 TRUE 10 0.23062284997087945 -6.820903512 42 200 246.38019831562286 46.38019831562286 TRUE 11 0.06744243838089692 -2.304639878 12.1 200 775.5449110883416 575.5449110883416 TRUE

[0146] The results show that the measured peak CO values ​​for all operating conditions were below 50 ppm within the current test duration. However, if the current linear upward trend remains unchanged, CO will eventually cross the 50 ppm threshold, with a time margin ranging from tens of seconds to thousands of seconds. Among them, operating condition 10 has the largest linear slope, and it only requires drilling for about 46 seconds more than the 200-second threshold to extrapolate to 50 ppm. It is one of the potentially dangerous operating conditions that is closest to the poisoning threshold and requires priority adjustment of drilling parameters.

[0147] Step S5: Construction of the Multiple Regression CO Model

[0148] S51: Feature Matrix Construction

[0149] In the original parameter space, a feature matrix is ​​constructed with the working condition number, drilling speed n, drilling speed v, drilling time t, and drill bit temperature T as input features and CO concentration ppm as the output target.

[0150] S52: Polynomial Extension and Regression Fitting

[0151] A second-order polynomial expansion is performed on the feature matrix to generate an expanded feature matrix containing the original features, quadratic terms, and interaction terms. A linear regression method is used to fit the relationship between the expanded features and CO concentration, constructing a second-order polynomial regression model to fit the relationship between CO concentration and multi-source control parameters. The polynomial order is 2, and the polynomial expansion does not include bias terms; this design aims to avoid overfitting due to parameter redundancy. This is implemented using PolynomialFeatures + LinearRegression.

[0152] S53: Model Validation

[0153] The fitting results were validated, and a parameter space search computational model was generated. The model achieved an RMSE of approximately 0.8123 ppm and an R² of approximately 0.98993 on the training set. A comparison of the measured and predicted curves is shown below. Figure 4 and Figure 5 As shown, the two are basically overlapping, verifying the model's fitting effect.

[0154] Step S6: Hazardous operating condition inversion and parameter space search

[0155] S61: Parameter Space Boundary Setting

[0156] Based on a multinomial regression model, for the termination states of different current working conditions, while keeping the working condition number unchanged, the value ranges of drilling rig speed, drilling speed, drilling time, and drill bit temperature are appropriately widened based on existing construction parameters. Based on the global maximum values ​​of historical data for different working conditions, the upper limits of drilling rig speed, drilling speed, and drill bit temperature are amplified by a preset multiple, and the upper limit of drilling time is increased by a preset additional drilling time. This ensures that the upper limits cover a larger combination of working conditions within the allowable range of normal production, thereby covering potentially dangerous operating intervals. This forms a four-dimensional parameter space including speed, feed rate, operating time, and drill bit temperature, and generates the boundary of this four-dimensional parameter space.

[0157] S62: Parameter Search and Hazardous Condition Identification

[0158] Within this four-dimensional space, parameter combinations are generated at a preset step size or using a random search method. Each set of sampled parameters is substituted into a multinomial regression model to calculate the predicted CO concentration. When the predicted CO concentration reaches or exceeds the preset CO poisoning risk threshold of 50 ppm, the corresponding parameter combination is recorded as the hazardous condition parameter set.

[0159] S63: Generation of hazardous operating condition parameter sets

[0160] The difference between each control quantity in the calculated parameter combination and the corresponding control quantity in the original operating condition is taken as the change, and the change in each control quantity compared with the original operating condition is recorded. The cumulative number of random searches is set to a finite upper limit. The search stops when the number of samples reaches the preset upper limit or when at least one set of parameter combinations that causes the CO concentration to exceed the poisoning risk threshold has been found for all operating conditions, and a set of hazardous operating condition parameters is generated.

[0161] Step S7: Parameter sensitivity analysis and determination of safety control range

[0162] S71: Normalization

[0163] The parameters of dangerous working conditions, namely the changes in drilling rig speed, drilling speed, drilling time and drill bit temperature, are extracted and normalized according to their global range in historical data.

[0164] S72: Calculation of Comprehensive Parameter Change Index: For each operating condition, record the first combination that meets the conditions (adj_rpm, adj_speed, adj_time_s, adj_temp) and the corresponding pred_co_ppm. Normalize the changes in the four control variables according to their original ranges to obtain total_change_norm, as shown in Table 7 (Table 7 is a statistical table of dangerous parameter combinations reaching 50ppm under different operating conditions obtained through searching the four control variable parameters in this embodiment of the invention, including adj_rpm, adj_speed, adj_time_s, adj_temp, pred_co_ppm, and total_change_norm). Compare the sum of squares and the normalized changes of each individual item under different operating conditions to quantify the overall parameter adjustment range required for different operating conditions to reach the warning threshold.

[0165] Table 7

[0166] Operating conditions reached 50ppm by time only found_solution_with_param_adjust adj_rpm adj_speed adj_time_s adj_temp pred_co_ppm total_change_norm 1 FALSE TRUE 1104.4718023283133 1.5804606308742843 246.8491847758504 128.11916603266076 60.30405385988879 0.16741448332627218 2 FALSE TRUE 1058.074450899857 1.162122298935602 238.19200798505946 138.23027053061475 59.63034431972631 0.15431377905497276 3 FALSE TRUE 1038.1249647732557 0.5761148513873269 318.46099417512625 150.66576968477403 62.0766449471866 0.11256189471619756 4 FALSE TRUE 876.8248149695054 1.5271572708520778 261.83503775441943 127.72110267277473 61.2439794890355 0.27863853147339757 5 FALSE TRUE 815.9304063633274 1.236235073398539 324.6858041619801 131.5631076349025 55.95676285103389 0.2536373966580878 6 FALSE TRUE 774.109043133901 0.8158290672049078 325.18601823082844 152.3038469432618 73.47098311909329 0.35452202471821614 7 FALSE TRUE 635.7678693975481 1.7269037088966237 250.7436460765157 110.93606740852519 52.64837374045401 0.3115914581581831 8 FALSE TRUE 495.5976176431721 1.4403480813783347 213.30843778317256 132.178740744941 60.57481798674197 0.4125394526903691 9 FALSE TRUE 516.0303191857939 0.6757602043383386 233.17279393104235 154.9165144027153 54.335165118902275 0.4390548593552437 10 FALSE TRUE 995.0194014147861 0.531650497342664 396.17114158168 150.86389569351684 50.12635181181425 0.028048166937545527 11 FALSE TRUE 1023.062790996498 0.8176460776783857 355.36004440631956 149.07043622392652 69.19885461135624 0.34532877981640914

[0167] S73: Determining the scope of safety control

[0168] A comparative analysis of parameter sensitivity under different operating conditions is conducted, and the results are presented in the form of bar charts, such as... Figure 9 The chart shows a sensitivity comparison of the normalized changes in drilling rig speed, drilling speed, drilling time, and drill bit temperature under different operating conditions when CO concentration reaches 50 ppm. Based on the comparison results, the safe variation range of control quantities to prevent CO concentration from exceeding the warning threshold is determined under different operating conditions. The maximum allowable variation range of each control quantity required to keep CO concentration below the warning threshold is determined, and the safe variation range of control quantities is generated.

[0169] Taking operating condition 10 as an example, under the current experimental settings, the measured peak CO concentration was approximately 42 ppm. The combination of hazardous parameters given by the multinomial regression model was: drilling rig speed approximately 995 r / min, drilling speed approximately 0.53 cm / s, drilling time approximately 396 s, and drill bit temperature approximately 150.9℃, predicting a CO concentration of approximately 50.13 ppm, corresponding to a total_change_norm of only approximately 0.028. This indicates that under this condition, even a slight adjustment to the drilling parameters could trigger the poisoning threshold; therefore, it is necessary to strictly control the drilling rig speed to not exceed 995 r / min and the drilling time to not exceed 396 s. In contrast, operating conditions with lower initial CO concentrations have significantly higher total_change_norms, requiring a much larger increase in drilling speed, drilling speed, or temperature to reach 50 ppm.

[0170] It can be seen that the dominant control quantity required to reach 50ppm varies under different operating conditions. Some operating conditions are more sensitive to the drilling time, while others are more sensitive to the drilling speed or drill bit temperature. Figure 9 The results of the "hazardous operating condition inversion" are transformed into operable parameter constraints, providing a quantitative basis for actually formulating "which type of control variable change to prioritize restriction".

[0171] Based on the above steps, this embodiment forms a complete technical route of "data acquisition and preprocessing—multi-task prediction—multi-threshold early warning—over-limit time assessment—hazard parameter range back-calculation," and its overall structure is as follows: Figure 10 The diagram shows the overall technical route for joint prediction of coal body temperature and CO concentration, multi-threshold early warning, and hazard parameter inversion.

[0172] The application process of this method at the drilling site includes: real-time acquisition of drilling parameters and CO concentration data → inputting into a multi-task model for short-term prediction and threshold warning → periodically updating the time margin for different working conditions based on the CO concentration-time linear extrapolation results → combining the safety control range given by multiple regression and parameter sensitivity analysis to constrain and optimize the drilling rig speed, drilling speed, drilling time and drill bit temperature.

[0173] When this invention is deployed in an actual mine, the on-site monitoring system will collect n, v, t, T, and other parameters in real time. , Information such as temperature and CO is input into the early warning system. After standardization and working condition identification, it is sent to the multi-task model to calculate the predicted values ​​of temperature and CO. The system automatically gives the warning level based on the 24ppm and 50ppm thresholds, and provides the expected time margin for reaching 50ppm and the suggested range of drilling parameter adjustments based on linear extrapolation and search results of dangerous parameters. This improves the inherent safety level of drilling in deep coal seams while ensuring drilling efficiency.

[0174] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from the spirit and scope of this invention.

Claims

1. A method for predicting and classifying CO concentration during drilling in downhole directional long boreholes, characterized in that, include: S1: Collect data on drilling rig speed, drilling speed, drilling time, drill bit temperature, coal body temperature and CO concentration under different working conditions, preprocess the numerical features, and construct sample feature matrix and labels; S2: Using drilling rig speed, drilling speed, drilling time and drill bit temperature as inputs, and coal temperature and CO concentration as dual outputs, a multi-task least squares support vector machine model is established, and the model parameters are optimized to minimize the comprehensive prediction error. S3: Compare the CO concentration predicted by the trained model with the set threshold to achieve graded early warning; S4: Analyze the CO concentration time series under different operating conditions and calculate the time margin for the CO concentration to reach the preset threshold. S5: Construct a regression model with operating parameters as input and CO concentration as output; S6: Search within the parameter space for parameter combinations that cause the predicted CO concentration to exceed a preset threshold, and generate a set of hazardous operating condition parameters. S7: Normalize and analyze the changes in each control quantity in the set of parameters for hazardous operating conditions to determine the safe range of change of the control quantity.

2. The method for predicting and classifying CO concentration during drilling in downhole directional long boreholes according to claim 1, characterized in that, Step S1 includes: S11: During the drilling process, collect real-time data on drilling speed, drilling speed, drilling time, drill bit temperature, coal body temperature and CO concentration, and set working condition numbers according to drilling speed, coal sample type and loading conditions. S12: Identify outliers in real-time data, fill missing sample points with time series interpolation, and replace isolated outliers with the mean of a time window. S13: Standardize continuous variables other than working condition number by subtracting the mean from the original value and then dividing by the standard deviation. Divide the training set and test set according to a preset ratio to construct a sample feature matrix and corresponding coal body temperature and CO concentration labels.

3. The method for predicting and classifying CO concentration during drilling in downhole directional long boreholes according to claim 1, characterized in that, Step S2 includes: S21: Based on a standardized sample set, the input vectors are set as drilling rig speed, drilling speed, drilling time and drill bit temperature, and the output vectors are coal temperature and CO concentration. A nonlinear mapping relationship is established. S22: Using radial basis functions as kernel functions, an optimization objective function containing penalty factors and random errors is constructed, and Lagrange multipliers are introduced to transform the optimization problem into a system of linear equations for solution; S23: The whale optimization algorithm is used to iteratively search within the range of penalty factor γ (10 to 1000) and kernel function width σ (0.01 to 1.0) to minimize the comprehensive error of coal temperature and CO concentration prediction, and generate a trained dual-output prediction model.

4. The method for predicting and classifying CO concentration during drilling in downhole directional long boreholes according to claim 1, characterized in that: In step S2, the fitness function of the whale optimization algorithm is a weighted sum of the root mean square errors of coal body temperature and CO concentration predictions. The weights are set according to the importance of the two outputs, so that the weight of CO concentration prediction is not lower than the weight of coal body temperature prediction.

5. The method for predicting and classifying CO concentration during drilling in downhole directional long boreholes according to claim 1, characterized in that, Step S3 includes: S31: Based on the dual-output prediction model, the test set is predicted to obtain the predicted coal temperature and CO concentration values; S32: Set the warning threshold and the poisoning risk threshold, calculate the corresponding value of the CO concentration threshold in the standardized space, compare the predicted CO concentration with the warning threshold and the poisoning risk threshold respectively, and mark the samples that exceed the warning threshold as warning samples and the samples that exceed the poisoning risk threshold as high-risk samples. S33: Based on the measured CO concentration, count the number of true positive, false positive, false negative and true negative samples under different operating conditions and different threshold levels, calculate the early warning hit rate according to the ratio of true positive to the sum of true positive and false negative, calculate the false alarm rate according to the ratio of false positive to the sum of false positive and true negative, and generate graded early warning performance indicators.

6. The method for predicting and classifying CO concentration during drilling in downhole directional long boreholes according to claim 1, characterized in that: The graded early warning discrimination in S3 includes: when the model predicts that the CO concentration exceeds the early warning threshold but is below the poisoning risk threshold, an early warning signal is output and the corresponding working condition number and drilling parameters are recorded; when the predicted CO concentration exceeds the poisoning risk threshold, a poisoning risk alarm signal is output and the recording and subsequent analysis of dangerous parameter combinations are triggered.

7. The method for predicting and classifying CO concentration during drilling in downhole directional long boreholes according to claim 1, characterized in that, Step S4 includes: S41: Extract the CO concentration time series under different operating conditions to confirm that the CO concentration in the monitoring interval shows a monotonically increasing and approximately linear change over time. S42: A first-order linear function was fitted to the relationship between CO concentration and time using the least squares method. ,in The slope Using the intercept, calculate the theoretical time for CO concentration to reach the poisoning risk threshold under this linear trend assumption; S43: Calculate the time difference between the theoretical time and the current drilling end time as the time margin. Sort the theoretical arrival time and time margin obtained for different working conditions in ascending order. Mark the working condition with the smallest time margin as the priority dangerous trend working condition and generate the time margin ranking result.

8. The method for predicting and classifying CO concentration during drilling in downhole directional long boreholes according to claim 1, characterized in that, The S5 steps include: S51: Construct a feature matrix with working condition number, drilling speed, drilling speed, drilling time and drill bit temperature as input features and CO concentration as output target; S52: Perform second-order polynomial expansion on the feature matrix to generate an expanded feature matrix containing the original features, quadratic terms, and cross terms. Use linear regression to fit the relationship between the expanded features and CO concentration. The polynomial order is 2. The polynomial expansion does not include bias terms to avoid overfitting caused by parameter redundancy. S53: Verify the fitting results and generate a parameter space search calculation model.

9. The method for predicting and classifying CO concentration during drilling in downhole directional long boreholes according to claim 1, characterized in that, Step S6 includes: S61: Based on the multinomial regression model, keeping the working condition number unchanged, and on the basis of the global maximum value of historical data for different working conditions, the upper limit values ​​of drilling rig speed, drilling speed and drill bit temperature are increased by a preset multiple, and the upper limit value of drilling time is increased by a preset additional drilling time to cover the potentially dangerous operating range and generate the boundary of the four-dimensional parameter space. S62: Generate parameter combinations in the four-dimensional parameter space according to a preset step size or a random method. Substitute each parameter combination into a multinomial regression model to calculate the predicted CO concentration. When the predicted CO concentration reaches or exceeds the poisoning risk threshold, record the parameter combination. S63: The difference between each control quantity in the parameter combination and the corresponding control quantity in the original working condition is used as the change quantity. When the number of samplings reaches the preset upper limit or when at least one set of parameter combinations that makes the CO concentration exceed the poisoning risk threshold is found in all working conditions, the search stops and a dangerous working condition parameter set is generated.

10. The method for predicting and classifying CO concentration during drilling in downhole directional long boreholes according to claim 1, characterized in that, Step S7 includes: S71: Extract the changes in hazardous operating condition parameters, including drilling rig speed, drilling speed, drilling time, and drill bit temperature, and normalize them according to their global range in historical data. S72: Calculate the sum of squares of the normalized changes of the four control variables under each working condition as a comprehensive parameter change index from the current working condition to the dangerous working condition. Compare the sum of squares and the normalized changes of each individual variable under different working conditions to quantify the overall parameter adjustment range required for different working conditions to reach the warning threshold. S73: Based on the comparison results, determine the maximum permissible range of each control quantity required to keep the CO concentration below the warning threshold, and generate the safe range of control quantity changes.

Images