A high outburst coal seam drilling interface outburst risk dynamic evaluation method and system

Abstract

The application discloses a high outburst coal seam drilling interface outburst risk dynamic evaluation method and system, relates to the high outburst coal seam drilling safety technical field. The method comprises the following steps: S1, real-time collection of multi-source information, generation of a multi-source original information dataset; S2, data unified pretreatment, generation of a standardized dataset; S3, multi-source feature fusion, generation of an interface mutation joint feature vector; S4, risk quantitative mapping: establishment of a quantitative mapping model; S5, dynamic risk evaluation: construction of a dynamic outburst risk evaluation model fusing physical mechanism constraints and data driving, real-time calculation of a system instability probability based on a sliding time window mechanism, and division of a dynamic risk grade according to the system instability probability; and S6, graded early warning output. The application realizes real-time, accurate and advanced evaluation of the high outburst coal seam drilling interface outburst risk, and significantly improves the drilling construction safety.

Description

Technical Field

[0001] This invention relates to the field of drilling safety technology for high-outburst coal seams, and more specifically, to a method and system for dynamic assessment of outburst risk at the drilling interface in high-outburst coal seams. Background Technology

[0002] Drilling in high-gas, outburst-prone coal seams is a crucial step in coal mine gas extraction and geological exploration. Sudden changes in the coal-rock interface during drilling (such as abrupt changes in hardness gradient, fracture development zones, or gas-enriched zones) can easily trigger coal and gas outburst accidents, seriously threatening the lives of construction workers and the safety of mine production.

[0003] The existing technologies for assessing the risk of coal seam outbursts through boreholes have the following main shortcomings:

[0004] (1) Single information collection: It relies on a single drilling parameter (such as drilling pressure and rotation speed) or a single environmental parameter (such as gas concentration), without fully integrating characteristic signals reflecting coal and rock fractures such as acoustic emission and vibration, resulting in blind spots in risk identification;

[0005] (2) Crude data processing: lack of high-precision synchronization and adaptive denoising mechanism for multi-source signals, and low accuracy of feature extraction due to data heterogeneity;

[0006] (3) Limitations of assessment models: Existing models are mostly based on pure data-driven or pure physical mechanisms. The former has weak generalization ability and lacks geological constraints, while the latter is difficult to capture dynamic mutation characteristics, resulting in delayed or misjudged risk assessment.

[0007] (4) Passive early warning response: The early warning adopts a fixed threshold and does not dynamically adjust in combination with real-time geological conditions. It also lacks the ability to predict the trend of risk evolution and cannot achieve early warning.

[0008] Therefore, there is an urgent need to develop a prominent risk assessment method and system that integrates multi-source information, takes into account both physical mechanisms and data-driven approaches, and possesses dynamic adaptive capabilities, in order to solve the problem of precise risk prevention and control when the borehole interface of high-burst coal seams changes abruptly. Summary of the Invention

[0009] To address the above technical problems, this invention proposes a dynamic assessment method and system for outburst risk at the borehole interface of high-risk coal seams, which can achieve real-time, accurate, and advanced assessment and early warning of outburst risk at the borehole interface of high-risk coal seams.

[0010] To achieve the above technical objectives, the present invention adopts the following technical solution:

[0011] In a first aspect, the present invention provides a method for dynamic assessment of outburst risk at the borehole interface of high-risk coal seams, comprising the following steps:

[0012] S1. During the drilling process of high coal seam outburst, drilling parameter signals, geological environment signals and acoustic emission signals are collected in real time to generate a multi-source raw information dataset.

[0013] S2. Process the multi-source raw information dataset collected in step S1 to generate a standardized dataset;

[0014] S3. Extract and fuse features from the standardized dataset obtained in step S2 to generate a joint feature vector of interface mutation.

[0015] S4. Based on the interface mutation joint feature vector generated in step S3, construct a spatiotemporal graph structure to represent the dynamic interaction relationship of multi-physics fields, use causal relationship verification and feature sensitivity analysis methods to screen sensitive features related to prominent risk events, and then establish a quantitative mapping model between the sensitive features and prominent risk indicators.

[0016] S5. Based on the quantitative mapping model established in step S4, a dynamic protruding risk assessment model integrating physical mechanism and data-driven approach is constructed by embedding physical mechanism constraints and establishing a temporal feature extraction and state estimation framework. Using the dynamic protruding risk assessment model, the system instability probability is calculated in real time based on the sliding time window mechanism, and the dynamic risk level is divided according to the system instability probability.

[0017] S6. Based on the dynamic risk level assessment results, output graded early warning signals.

[0018] Beneficial effects: This invention integrates multi-source information: it integrates three types of signals, namely drilling parameters, geological environment, and acoustic emission, covering multi-dimensional information such as coal and rock mechanics, gas migration, and fracture evolution, and eliminates blind spots in risk identification;

[0019] Dynamic and real-time risk assessment: Based on a sliding time window mechanism, continuous risk assessment is achieved with a response delay of ≤2s, meeting the real-time prevention and control requirements of drilling construction.

[0020] As a further preferred embodiment of the dynamic assessment method for outburst risk at the borehole interface of high-outburst coal seams in this invention, step S3, the generation of the joint feature vector of interface mutation, specifically includes the following sub-steps:

[0021] S31. Basic Feature Extraction: Extract time-domain features, frequency-domain features, and time-frequency-domain features from the standardized dataset to form a basic feature set;

[0022] S32. Deep Feature Learning: Deep temporal features are automatically learned from standardized datasets through a bidirectional long short-term memory network to capture long-range dependencies of multi-source signals. At the same time, local mutation features are extracted through a convolutional neural network to construct a deep feature set.

[0023] S33. Attention Fusion: A hierarchical attention fusion mechanism is introduced to assign weights to the basic feature set obtained in step S31 and the deep feature set obtained in step S32; the weighted features are concatenated into a fused feature vector, and attention is calculated again to generate a joint feature vector of interface mutation, calculated as follows:

[0024] ,

[0025] in, To create a query matrix that integrates feature vectors, To fuse the key matrix of eigenvectors, To fuse the value matrix of the eigenvectors, Its vector dimension is determined to balance expressive power and computational efficiency.

[0026] Beneficial effects: By automatically selecting and weighting the key features most relevant to interface mutations through a hierarchical attention mechanism, redundant information is effectively compressed, and a joint feature vector with high representational power is generated, providing a reliable data foundation for subsequent accurate risk mapping and significantly enhancing the model's ability to capture and generalize complex risk precursors.

[0027] As a further preferred embodiment of the dynamic assessment method for outburst risk at the borehole interface of high-outburst coal seams in this invention, the outburst risk indicators include gas expansion energy, coal deformation energy release rate, and the probability of instability of the drill bit-coal-rock system.

[0028] As a further preferred embodiment of the dynamic assessment method for outburst risk at the borehole interface of high-outburst coal seams in this invention, step S5 includes the following sub-steps: S51. Dynamic outburst risk assessment model architecture design: The critical criterion for instability of the drill string-coal rock system based on the Mohr-Coulomb strength theory is embedded as a physical constraint into the loss function of the quantitative mapping model described in step S4. A dynamic outburst risk assessment model is constructed by combining the bidirectional long short-term memory network temporal feature extraction module and the convolutional neural network spatial feature mining module; its loss function is defined as:

[0029] ,

[0030] in, The mean squared error between the model's predicted instability probability and the true label; This is a label representing the actual instability state of the system. Predict the probability of system instability; These are physical constraint terms, constructed based on the Mohr-Coulomb strength theory;

[0031] S52. State variable estimation: Using the joint feature vector of interface mutation generated in step S3 as input, the Kalman filter algorithm is used to perform real-time and dynamic optimal estimation of key intrinsic physical state variables that cannot be directly measured at the borehole interface; the estimation results are fed into the dynamic protruding risk assessment model constructed in step S51 to enhance the model's ability to perceive the real physical state of the system.

[0032] S53. Set the sliding time window length and step size, segment the interface mutation joint feature vector obtained in step S3 and the key intrinsic physical state variables estimated in step S52, and calculate the system instability probability within each time window through the dynamic prominent risk assessment model obtained in step S51; S54. Risk level classification: classify the risk level into Level I safety, Level II concern, Level III early warning and Level IV emergency according to the system instability probability, where instability probability ≤ 10% corresponds to Level I safety, 10% < instability probability ≤ 30% corresponds to Level II concern, 30% < instability probability ≤ 60% corresponds to Level III early warning and instability probability > 60% corresponds to Level IV emergency.

[0033] Beneficial effects: Dual-drive collaborative model architecture: It combines the advantages of physical mechanism constraints and data-driven approaches, ensuring the physical rationality of the model while improving the ability to capture dynamic mutations and exhibiting strong generalization ability.

[0034] As a further preferred embodiment of the dynamic assessment method for outburst risk at the borehole interface of high-outburst coal seams in this invention, the physical constraint term... The specific expression is:

[0035] ,

[0036] In the formula, The equivalent shear stress of the i-th sample is expressed in MPa and is calculated from the real-time torque and shear action area.

[0037] The effective normal stress of the i-th sample is in MPa and is calculated from the drilling pressure, surrounding rock stress, and pore pressure.

[0038] c represents the cohesion of the coal and rock mass, in MPa;

[0039] The internal friction angle of the coal and rock mass, in degrees;

[0040] Critical shear stress, in MPa;

[0041] N is the number of samples in the batch, when When this occurs, it indicates that the system is in an unstable state, and the constraint terms... When the system is stable, this value is zero.

[0042] Beneficial effects: By embedding the physical instability criterion based on the Mohr-Coulomb strength theory as a constraint term into the model loss function, the model prediction is forced to conform to the laws of rock mechanics, effectively avoiding the physically unreasonable predictions that may be generated by pure data-driven approaches, and significantly improving the reliability and generalization ability of risk assessment.

[0043] As a further preferred embodiment of the dynamic assessment method for outburst risk at the borehole interface of high-outburst coal seams in this invention, the key intrinsic physical state variables include coal seam strength gradient and gas expansion power.

[0044] Beneficial effects: By estimating key intrinsic variables such as coal seam intensity gradient and gas expansion power in real time, the model can more accurately perceive the true physical state of the system, enhance the ability to capture risk precursors, and thus improve the accuracy of dynamic risk assessment and the reliability of early warning.

[0045] As a further preferred embodiment of the dynamic assessment method for outburst risk at the borehole interface of high-risk coal seams of the present invention, step S6 outputs a graded early warning signal based on the dynamic risk level assessment results, specifically including the following sub-steps:

[0046] S61. Threshold Adaptive Optimization: Pre-set the original risk level classification threshold applicable to typical geological conditions, acquire geological parameters and state estimation variables at the current borehole location in real time, and calculate the adjustment coefficient by comparing with the baseline conditions. ,

[0047] ,

[0048] in, , , These are the current gas concentration, surrounding rock stress, and gas expansion power, respectively. , , This is the baseline value for the corresponding parameter; , , This is the weighting factor; the original risk level classification threshold is multiplied by the adjustment factor. The system obtains a dynamic early warning threshold applicable to the current geological conditions; if the threshold is exceeded, the original threshold is activated and the system log is recorded.

[0049] S62. Warning signal generation: When the risk level reaches Level II or above, generate an audible and visual warning signal of the corresponding level;

[0050] S63. Trend Report Output: Simultaneously output a comprehensive risk situation report that includes the extrapolation of the risk evolution trend in the next 1.5 to 2 seconds, the current risk level, and the contribution of the dominant sensitive characteristics, providing a decision-making window for implementing proactive regulation;

[0051] S64. Online Model Optimization: Establish an online incremental learning mechanism based on a sliding time window, and adopt the online gradient descent method. Based on real-time drilling data and its corresponding system state labels, the weight parameters of the dynamic protruding risk assessment model are dynamically updated with the goal of minimizing the prediction loss within the sliding time window.

[0052] Beneficial effects: The early warning mechanism is intelligent and adaptive. This invention dynamically adjusts the early warning threshold and predicts risk trends, realizing hierarchical early warning, reducing false alarm rate and missed alarm rate, and providing a scientific basis for construction decision-making.

[0053] As a further preferred embodiment of the dynamic assessment method for outburst risk at the borehole interface of high-burst coal seams of the present invention, in step S1, the drilling parameter signals include drilling pressure, rotation speed, torque and propulsion speed.

[0054] The geological environment signals include gas concentration, coal temperature, and borehole surrounding rock stress.

[0055] The acoustic emission signal is an elastic wave signal generated by coal and rock fracturing.

[0056] Beneficial effects: By simultaneously collecting three types of signals—drilling mechanics, geological environment, and acoustic emission from coal and rock fractures—multi-dimensional information perception covering construction disturbance, geological response, and fracture precursors is achieved, thereby solving the problem of perception blind spots caused by relying on a single parameter and providing a complete data foundation for subsequent accurate identification of multi-coupled risks.

[0057] As a further preferred embodiment of the dynamic assessment method for outburst risk at the borehole interface of high-outburst coal seams of the present invention, step S2, the data processing of the multi-source raw information dataset collected in step S1 includes: performing adaptive denoising, time-series synchronization, and standardization processing, wherein,

[0058] The adaptive denoising method employs wavelet thresholding and adaptively adjusts the threshold based on the signal-to-noise ratio. Its threshold processing function... Defined as:

[0059] ,

[0060] in, These are the wavelet coefficients obtained from each level of decomposition;

[0061] The adaptive threshold is dynamically calculated based on the signal energy and noise variance estimation of each sub-band.

[0062] The smoothing parameter controls the shape of the threshold transition region;

[0063] For symbolic functions, it means The plus or minus sign;

[0064] The timing synchronization employs a dynamic time warping algorithm to achieve high-precision timing alignment of multi-source signals, with a synchronization error ≤5ms; the standardization process uses the Z-score normalization method to convert the data to a unified scale of [0,1].

[0065] Beneficial effects: Precise and efficient data processing: Through wavelet threshold denoising, dynamic time warping synchronization and standardization, data quality and consistency are improved, laying the foundation for feature extraction.

[0066] Secondly, this invention further discloses a dynamic assessment system for outburst risk at the borehole interface of high-risk coal seams. The system utilizes the aforementioned dynamic assessment method for outburst risk at the borehole interface of high-risk coal seams, comprising: a multi-source information acquisition module for real-time acquisition of drilling parameter signals, geological environment signals, and acoustic emission signals; a data preprocessing module for adaptive denoising, time-series synchronization, and standardization of the acquired signals; a feature fusion module for extracting and fusing multi-source features to generate a joint feature vector of interface mutations; a risk mapping module for constructing a quantitative risk mapping model and establishing a quantitative relationship between features and risk indicators; a dynamic assessment module for real-time calculation of risk levels based on a sliding time window and a dynamic outburst risk assessment model; and an early warning output module for dynamically outputting graded early warning signals and trend reports based on the risk assessment results. Attached Figure Description

[0067] Figure 1 This is a flowchart of a method for dynamic assessment of outburst risk at the borehole interface in high-outburst coal seams, as described in an embodiment of the present invention.

[0068] Figure 2 This is an architecture diagram of the dynamic assessment system for outburst risk at the borehole interface of high-burst coal seams, as described in an embodiment of the present invention.

[0069] Figure 3 This is a comparison chart of the real-time risk level output curve and the actual abnormal event in an embodiment of the present invention;

[0070] Figure 4 This is a diagram showing the correspondence between risk level classification and early warning signals in an embodiment of the present invention;

[0071] Explanation of reference numerals in the attached diagram: 1-Multi-source information acquisition module, 2-Data preprocessing module, 3-Feature fusion module, 4-Risk mapping module, 5-Dynamic assessment module, 6-Early warning output module. Detailed Implementation

[0072] The technical solutions of the embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] This embodiment uses the drilling construction of a high-gas coal seam in a coal mine as an application scenario. The coal seam has a high gas content, medium coal hardness, standard borehole diameter, and a large drilling depth. The specific implementation process is as follows:

[0074] 1. System Deployment:

[0075] 1-1. Installation of multi-source sensing components: Install the drill pressure sensor, speed sensor, and torque sensor on the drill rig power head, the feed speed sensor on the feed cylinder, the gas concentration sensor on the borehole sealing device, the temperature sensor and stress sensor on the borehole wall surrounding rock, the vibration sensor on the drill rig frame, and the acoustic emission sensor adsorbed on the coal body around the borehole opening.

[0076] 1-2. Data Acquisition Card Configuration: A high-performance data acquisition card is used, with a sampling frequency that meets the requirements. The signals from each sensor are connected to the acquisition card and then connected to the industrial computer via data cables.

[0077] 1-3. Software Deployment: Install the evaluation system software based on Python on the industrial computer, integrating functional modules such as data preprocessing, feature fusion, risk assessment, and early warning output.

[0078] 2. Method Implementation:

[0079] (1) Multi-source information acquisition: During the drilling process, the multi-source information acquisition module collects drilling pressure, rotation speed, torque, propulsion speed, gas concentration, temperature, vibration signal and acoustic emission signal in real time to generate a multi-source raw information dataset;

[0080] (2) Data preprocessing:

[0081] 2-1. Denoising Processing: The db4 wavelet is used to decompose the signal of each channel. An improved wavelet thresholding method is then used to denoise and remove abnormal pulse signals generated by the drilling rig cylinder impact. The core of this improved wavelet thresholding method is the use of an adaptive semi-soft thresholding function for denoising. Specifically, for each layer of decomposition, the wavelet coefficients are... Its threshold processing function is defined as:

[0082] ,

[0083] in, The adaptive threshold is dynamically calculated based on the signal energy and noise variance estimation of each sub-band. To smooth the parameters, the shape of the threshold transition zone is controlled. This method effectively preserves the weak abrupt change characteristics by smoothing the transition zone, and adaptively adjusts the threshold intensity according to the local signal-to-noise ratio of the signal, achieving an optimal balance between suppressing pulse noise such as drilling rig cylinder impact and preserving effective signals of coal and rock fracture.

[0084] 2-2. Timing Synchronization: The dynamic time warping algorithm is used to align multi-source signals. Taking the drilling pressure signal as the reference, the time axes of other signals are aligned with the reference signal, and the synchronization error is controlled at a low level.

[0085] 2-3. Standardization: The synchronized multi-source data is transformed to the [0,1] scale using the Z-score normalization formula to generate a standardized dataset;

[0086] (3) Multi-source feature fusion:

[0087] 3-1. Basic Feature Extraction: Extract time-domain features from the standardized dataset, including mean, variance, kurtosis, peak factor, and frequency-domain features, including spectral peak, centroid frequency, mean square frequency, and time-frequency-domain features, including wavelet packet energy entropy and wavelet entropy.

[0088] 3-2. Deep Feature Learning: Construct a bidirectional long short-term memory network to extract temporal deep features, construct a convolutional neural network to extract spatial local features, and generate a deep feature set;

[0089] 3-3. Attention Fusion: A hierarchical attention mechanism is used to fuse the basic feature set and the deep feature set. First, in the feature-level attention layer, the basic feature set is fused separately. and deep feature sets Calculate attention weights, where For time steps, and This corresponds to the feature dimension. Subsequently, in the fusion attention layer, the basic feature set is... and deep feature sets Calculate the attention weights separately:

[0090] ,

[0091] in, The query matrix of the basic feature set, The key matrix of the basic feature set, The value matrix of the basic feature set, The query matrix of the basic feature set, The key matrix of the basic feature set, The value matrix of the basic feature set is obtained by linear transformation of the feature matrix. The scaling factor is used to concatenate the weighted features into a fused feature vector. Then perform attention calculation again:

[0092] ,

[0093] in, A query matrix that integrates feature vectors. The key matrix that integrates eigenvectors. The value matrix of the fused feature vectors. The final output is... This is the joint feature vector of interface mutation, and its dimension is... Set to 64 to balance expressive power and computational efficiency;

[0094] (4) Quantitative risk mapping:

[0095] 4-1 Spatiotemporal Structure Construction: Using borehole depth as the spatial axis and time as the temporal axis, a spatiotemporal graph of the grid is constructed to represent the information interaction relationship of multiphysics.

[0096] 4-2. Association Analysis: The causal relationship between the joint features of interface mutations and prominent risk events is verified by causal testing. The dependence strength between the joint features of interface mutations and prominent risk indicators is calculated by mutual information analysis to screen out sensitive features.

[0097] 4-3. Establishment of Quantitative Mapping Model: Based on the selected sensitive features, a quantitative mapping model between them and prominent risk indicators is established. This model uses support vector regression with radial basis function kernel. The input is a joint feature vector, and the output is the gas expansion energy, coal deformation energy release rate, and system instability probability. The training data comes from historical monitoring data from multiple high-risk coal seam borehole construction sites. The recording and calculation methods for each risk indicator are as follows:

[0098] ① Gas expansion energy can be monitored in real time by a gas flow sensor installed at the borehole opening to detect the gas volumetric flow rate. The pressure difference between coal seam gas pressure (provided by geological exploration data) and atmospheric pressure. The following model is used to calculate the energy released by gas expansion per unit time, and the result is integrated over time:

[0099] ,

[0100] in, The gas expansion efficiency coefficient was obtained through laboratory calibration. It represents the constant difference between coal seam gas pressure and borehole outlet environmental pressure, taken from borehole geological design data.

[0101] ② The coal deformation energy release rate is calculated using the strain energy density method by monitoring the micro-strain of the coal body through triaxial strain sensors deployed around the borehole, combined with the coal body's elastic modulus and strength parameters:

[0102] ,

[0103] in, , The stress and strain are respectively in the three principal directions. For the infinitesimal volume of the coal body, The sampling interval is denoted as .

[0104] ③ The probability of system instability is a continuous value (0-1) based on the actual outburst events, precursor signal strength, and historical statistical data during drilling. No outburst occurs and there are no precursor signals. Slight warning signs may appear (such as a brief increase in gas concentration). Obvious warning signs appear (such as sudden drop in drilling pressure, increased acoustic emission events): Localized protrusion or strong precursor signals occur. ;

[0105] The support vector regression model with a radial basis function kernel is used for training. The kernel function expression is as follows:

[0106] ,

[0107] During training, grid search and cross-validation are used to optimize hyperparameters and kernel parameters. To minimize the mean squared error of prediction, the model receives the joint feature vector in real time after training and simultaneously outputs the predicted values ​​of three risk indicators. The lightweight model is deployed on industrial computers to meet the real-time evaluation requirements of the drilling process (single inference time < 200ms).

[0108] (5) Dynamic risk assessment:

[0109] 5-1. Model Loading: The critical criterion for instability of the drill string-coal rock system based on the Mohr-Coulomb strength theory is used as a physical constraint and embedded into the loss function of the quantitative mapping model established in step (4). Simultaneously, a dynamic prominence risk assessment model incorporating physical constraints is loaded by combining the temporal feature extraction module of the Bidirectional Long Short-Term Memory Network (Bi-LSTM) and the spatial feature mining module of the Convolutional Neural Network (CNN). The loss function of this model is defined as:

[0110] ,

[0111] in, The mean squared error between the model's predicted instability probability and the true label; This is a label representing the actual instability state of the system. To predict the probability of system instability. The physical constraint term is constructed based on the Mohr-Coulomb strength theory, and its specific expression is as follows:

[0112] ,

[0113] In the formula, The equivalent shear stress (MPa) for the i-th sample is calculated from the real-time torque and the shear action area. denoted as , where is the effective normal stress (MPa) of the i-th sample, calculated from drilling pressure, surrounding rock stress, and pore pressure; c is the cohesion of the coal and rock mass (MPa). The internal friction angle of the coal and rock mass (°); Where is the critical shear stress (MPa); N is the batch sample size. When this occurs, it indicates that the system is in an unstable state, and the constraint terms... This term is zero when the system is stable. By minimizing... The model is constrained to predict the probability of instability under physical instability conditions;

[0114] 5-2. State estimation: Unmeasurable parameters such as coal seam intensity gradient and gas expansion power are estimated using Kalman filtering;

[0115] 5-3 Sliding window calculation: Set the length and step size of the sliding time window, process the joint feature vector of interface mutation in segments, and calculate the probability of system instability in each time window through a dynamic risk assessment model;

[0116] 5-4. Risk Level Classification: The risk level is output based on the probability of instability, where Level I is safe, Level II is of concern, Level III is a warning, and Level IV is an emergency.

[0117] (6) Tiered early warning output:

[0118] 6-1 Threshold Optimization: Dynamically adjust the early warning threshold based on the current coal seam gas content and coal hardness; pre-set the original risk level classification threshold applicable to typical geological conditions: instability probability ≤10% is safe, 10%-30% is of concern, 30%-60% is of warning, and >60% is of emergency; acquire the geological parameters and state estimation variables of the current borehole location in real time, and calculate the adjustment coefficient by comparing with the baseline conditions. ,

[0119] ,

[0120] in, , , These are the current gas concentration, surrounding rock stress, and gas expansion power, respectively. , , This is the baseline value for the corresponding parameter; , , This is the weighting factor; the original risk level classification threshold is multiplied by the adjustment factor. The system obtains a dynamic early warning threshold applicable to the current geological conditions; if the threshold is exceeded, the original threshold is automatically activated and the system log is recorded.

[0121] 6-2. Warning signal generation: When the risk level reaches Level II, the system issues a yellow audible and visual warning, the indicator light flashes, and the buzzer sounds a low-frequency alarm; when it reaches Level III, it issues an orange audible and visual warning, the indicator light stays on, and the buzzer sounds a medium-frequency alarm; when it reaches Level IV, it issues a red audible and visual warning, the indicator light flashes rapidly, the buzzer sounds a high-frequency alarm, and at the same time triggers the drilling rig emergency stop signal.

[0122] 6-3. Trend Report Output: Simultaneously output a comprehensive risk situation report that includes the extrapolation of the risk evolution trend in the next 2 seconds, the current risk level, and the contribution of the dominant sensitive characteristics, providing a decision-making window for implementing proactive regulation;

[0123] 6-4. Online Model Optimization: An online incremental learning mechanism for parameters is established, employing the sliding window online gradient descent (SW-OGD) algorithm. Based on real-time drilling data, the parameters of the dynamic risk assessment model are continuously updated to improve assessment accuracy. The specific update process is as follows:

[0124] Assume the dynamic risk assessment model is a parameterized function.

[0125] ,

[0126] in, The input feature vector at time t is composed of the joint feature vector of interface mutation and the key intrinsic physical state variables estimated by Kalman filter. For the parameters of the model at time t, This represents the predicted probability of system instability. Each time a new sample is received... (in The label for the system instability state obtained from actual observation or inversion (with a value of 0 or 1), i.e., the label for the model parameters. Perform an incremental update to obtain .

[0127] Use a sliding time window to retain the most recent W samples The window length W is set according to the drilling process and data sampling rate, and is usually taken as... (Corresponding to approximately 5 seconds of data), at time t, define a loss function based on a sliding window:

[0128] ,

[0129] in, For cross-entropy loss, ;

[0130] Parameter updates are performed using the online gradient descent method with momentum:

[0131] ,

[0132] In the formula Learning rate To avoid noise introduced by frequent updates, the update trigger threshold is set to [value]. That is, parameter updates are only performed when the proportion of samples with a prediction error greater than 0.2 within the sliding window exceeds 30%; otherwise, the parameters remain unchanged.

[0133] 3. Implementation Results:

[0134] Based on the practical application of the system described in this invention in drilling operations at a high-risk coal mine, real-time risk assessment data for 60 seconds was obtained through continuous monitoring. This data fully records the occurrence process of three abnormal events and the system's early warning response performance, providing direct evidence for verifying the effectiveness of the method and system of this invention.

[0135] According to monitoring data, the system's early warning response delay for all three abnormal events was controlled within ≤2 seconds, as detailed below:

[0136] First gas anomaly (t=18s): The risk level reaches 3.5 (Level III warning) at t=17s, and a warning signal is issued 1 second in advance;

[0137] The second torque surge event (t=35s): the risk level reached 3.2 (Level III warning) at t=33s, and a warning signal was issued 2 seconds in advance;

[0138] The third gas anomaly (t=51s): The risk level reached 3.2 (Level III warning) at t=49s, and a warning signal was issued 2 seconds in advance.

[0139] Data shows that the system achieved 100% accuracy in identifying the three abnormal events. Specifically, all abnormal events were accurately identified and triggered the corresponding warning levels, with no missed reports, fully meeting the technical requirement of "warning the risk evolution trend in the next 2 seconds." The risk level curve transitioned smoothly without frequent oscillations or misjudgments, demonstrating the system's good stability and reliability. Statistical analysis shows that the present invention has a high risk identification accuracy, short warning response delay, and low false alarm and missed alarm rates, significantly outperforming existing single-parameter evaluation techniques.

[0140] Table 1 Risk Prediction Results

[0141]

[0142] This invention further discloses a dynamic assessment system for outburst risk at the borehole interface of a high-risk coal seam, applied to the aforementioned dynamic assessment method for outburst risk at the borehole interface of a high-risk coal seam, comprising:

[0143] Multi-source information acquisition module 1 is used to acquire drilling parameter signals, geological environment signals and acoustic emission signals in real time;

[0144] Data preprocessing module 2 is used to perform adaptive noise reduction, timing synchronization and standardization processing on the acquired signals;

[0145] Feature fusion module 3 is used to extract and fuse multi-source features to generate a joint feature vector of interface mutations;

[0146] Risk mapping module 4 is used to construct a quantitative mapping model and establish a quantitative relationship between the joint characteristics of interface mutations and prominent risk indicators.

[0147] Dynamic assessment module 5 is used to calculate risk levels in real time based on a sliding time window and a dynamic highlighted risk assessment model;

[0148] The early warning output module 6 is used to dynamically output graded early warning signals and trend reports based on the risk assessment results.

[0149] Any adaptive changes made according to actual needs are within the scope of protection of this invention.

[0150] It should be noted that, for those skilled in the art, it is obvious that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0151] Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. Furthermore, those skilled in the art will recognize that, based on the ideas of this invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A method for dynamic assessment of outburst risk at the borehole interface in high-risk coal seams, characterized in that, Includes the following steps: S1. During the drilling process of high coal seam outburst, drilling parameter signals, geological environment signals and acoustic emission signals are collected in real time to generate a multi-source raw information dataset. S2. Process the multi-source raw information dataset collected in step S1 to generate a standardized dataset; S3. Extract and fuse features from the standardized dataset obtained in step S2 to generate a joint feature vector of interface mutation. S4. Based on the interface mutation joint feature vector generated in step S3, construct a spatiotemporal graph structure to represent the dynamic interaction relationship of multi-physics fields, use causal relationship verification and feature sensitivity analysis methods to screen sensitive features related to prominent risk events, and then establish a quantitative mapping model between the sensitive features and prominent risk indicators. S5 includes the following sub-steps: S51. Dynamic Outburst Risk Assessment Model Architecture Design: The instability critical criterion of the drill string-coal rock system based on the Mohr-Coulomb strength theory is embedded as a physical constraint into the loss function of the quantitative mapping model described in step S4. A dynamic outburst risk assessment model is constructed by combining a bidirectional long short-term memory network temporal feature extraction module and a convolutional neural network spatial feature mining module; its loss function is defined as: ， in, The mean squared error between the model's predicted instability probability and the true label; This is a label representing the actual instability state of the system. Predict the probability of system instability; These are physical constraint terms, constructed based on the Mohr-Coulomb strength theory; S52. State variable estimation: Using the joint feature vector of interface mutation generated in step S3 as input, the Kalman filter algorithm is used to perform real-time and dynamic optimal estimation of key intrinsic physical state variables that cannot be directly measured at the borehole interface; the estimation results are fed into the dynamic protruding risk assessment model constructed in step S51 to enhance the model's ability to perceive the real physical state of the system. S53. Set the sliding time window length and step size, segment the interface mutation joint feature vector obtained in step S3 and the key intrinsic physical state variables estimated in step S52, and calculate the system instability probability in each time window through the dynamic protrusion risk assessment model obtained in step S51. S54. Risk Level Classification: Based on the system instability probability, the risk level is divided into Level I Safety, Level II Attention, Level III Warning, and Level IV Emergency. Specifically, an instability probability ≤ 10% corresponds to Level I Safety, 10% < instability probability ≤ 30% corresponds to Level II Attention, 30% < instability probability ≤ 60% corresponds to Level III Warning, and an instability probability > 60% corresponds to Level IV Emergency. S6. Based on the assessment results of the dynamic risk level, output a graded early warning signal.

2. The method for dynamic assessment of outburst risk at the borehole interface of high-outburst coal seams according to claim 1, characterized in that, Step S3, generating the joint feature vector of interface mutation, specifically includes the following sub-steps: S31. Basic Feature Extraction: Extract time-domain features, frequency-domain features, and time-frequency-domain features from the standardized dataset generated in step S2 to form a basic feature set; S32. Deep Feature Learning: Deep temporal features are automatically learned from standardized datasets through a bidirectional long short-term memory network to capture long-range dependencies of multi-source signals. At the same time, local mutation features are extracted through a convolutional neural network to construct a deep feature set. S33. Attention Fusion: A hierarchical attention fusion mechanism is introduced. Weights are assigned to the basic feature set obtained in step S31 and the deep feature set obtained in step S32. The weighted features are concatenated into a fused feature vector. Attention is calculated again to generate a joint feature vector representing interface mutations, calculated as follows: ， in, To create a query matrix that integrates feature vectors, To fuse the key matrix of eigenvectors, To fuse the value matrix of the eigenvectors, Its vector dimension is determined to balance expressive power and computational efficiency.

3. The method for dynamic assessment of outburst risk at the borehole interface of high-outburst coal seams according to claim 1, characterized in that, The key risk indicators include gas expansion energy, coal deformation energy release rate, and the probability of instability of the drill bit-coal-rock system.

4. The method for dynamic assessment of outburst risk at the borehole interface of high-risk coal seams according to claim 1, characterized in that, The physical constraints The specific expression is: ， In the formula, The equivalent shear stress of the i-th sample is expressed in MPa and is calculated from the real-time torque and shear action area. The effective normal stress of the i-th sample is in MPa and is calculated from the drilling pressure, surrounding rock stress, and pore pressure. c represents the cohesion of the coal and rock mass, in MPa; The internal friction angle of the coal and rock mass, in degrees; Critical shear stress, in MPa; N is the number of samples in the batch, when When this occurs, it indicates that the system is in an unstable state, and the constraint terms... When the system is stable, this value is zero.

5. The method for dynamic assessment of outburst risk at the borehole interface of high-risk coal seams according to claim 1, characterized in that, The key intrinsic physical state variables include coal seam strength gradient and gas expansion power.

6. The method for dynamic assessment of outburst risk at the borehole interface of high-outburst coal seams according to claim 1, characterized in that, Step S6 outputs a graded early warning signal based on the dynamic risk level assessment results, specifically including the following sub-steps: S61. Threshold Adaptive Optimization: Pre-set the original risk level classification threshold applicable to typical geological conditions, acquire geological parameters and state estimation variables at the current borehole location in real time, and calculate the adjustment coefficient by comparing with the baseline conditions. , ， in, , , These are the current gas concentration, surrounding rock stress, and gas expansion power, respectively. , , This is the baseline value for the corresponding parameter; , , This is the weighting factor; the original risk level classification threshold is multiplied by the adjustment factor. The system obtains a dynamic early warning threshold applicable to the current geological conditions; if the threshold is exceeded, the original threshold is activated and the system log is recorded. S62. Warning signal generation: When the risk level reaches Level II or above, generate an audible and visual warning signal of the corresponding level; S63. Trend Report Output: Simultaneously output a comprehensive risk situation report that includes the extrapolation of the risk evolution trend in the next 1.5 to 2 seconds, the current risk level, and the contribution of the dominant sensitive characteristics, providing a decision-making window for implementing proactive regulation; S64. Online Model Optimization: Establish an online incremental learning mechanism based on a sliding time window, and adopt the online gradient descent method. Based on real-time drilling data and its corresponding system state labels, the weight parameters of the dynamic protruding risk assessment model are dynamically updated with the goal of minimizing the prediction loss within the sliding time window.

7. The method for dynamic assessment of outburst risk at the borehole interface of high-outburst coal seams according to claim 1, characterized in that, In step S1, the drilling parameter signals include drilling pressure, rotational speed, torque, and feed rate; The geological environment signals include gas concentration, coal temperature, and borehole surrounding rock stress. The acoustic emission signal is an elastic wave signal generated by coal and rock fracturing.

8. The method for dynamic assessment of outburst risk at the borehole interface of high-outburst coal seams according to claim 1, characterized in that, In step S2, the multi-source raw information dataset collected in step S1 is processed, including adaptive denoising, time-series synchronization, and standardization. The adaptive denoising method employs wavelet thresholding and adaptively adjusts the threshold based on the signal-to-noise ratio. Its threshold processing function... Defined as: ， in, These are the wavelet coefficients obtained from each level of decomposition; The adaptive threshold is dynamically calculated based on the signal energy and noise variance estimation of each sub-band. The smoothing parameter controls the shape of the threshold transition region; For symbolic functions, it means The plus or minus sign; The timing synchronization adopts a dynamic time warping algorithm to achieve high-precision timing alignment of multi-source signals with a synchronization error ≤5ms; the standardization process adopts the Z-score normalization method to transform the data to a unified scale of [0,1].

9. A dynamic assessment system for outburst risk at the borehole interface of a high-risk coal seam, applied to the dynamic assessment method for outburst risk at the borehole interface of a high-risk coal seam as described in any one of claims 1-8, characterized in that, include: Multi-source information acquisition module (1) is used to acquire drilling parameter signals, geological environment signals and acoustic emission signals in real time; The data preprocessing module (2) is used to perform adaptive noise reduction, timing synchronization and standardization on the acquired signals; The feature fusion module (3) is used to extract and fuse multi-source features to generate a joint feature vector of interface mutation. The risk mapping module (4) is used to construct a quantitative mapping model and establish a quantitative relationship between the joint characteristics of interface mutations and prominent risk indicators. The dynamic assessment module (5) is used to calculate the risk level in real time based on the sliding time window and the dynamic highlighted risk assessment model; The early warning output module (6) is used to dynamically output graded early warning signals and trend reports based on the risk assessment results.

Images