Soft coal seam hole collapse early warning method and device, electronic equipment and storage medium

Abstract

The application provides a soft coal seam hole collapse early warning method and device, electronic equipment and storage medium, and belongs to the technical field of mine monitoring. The method comprises the following steps: acquiring standardized time sequence data of a drill bit in a drilling process in a current period; performing space-time alignment and feature enhancement processing on the standardized time sequence data of the drill bit in the drilling process in the current period to obtain feature enhancement data; extracting space-time correlation features of different monitoring dimensions in the feature enhancement data through a space-time attention convolution network to generate a multi-field coupling feature tensor; inputting the multi-field coupling feature tensor into a double-branch adversarial time sequence prediction model to obtain a hole collapse risk precursor feature set in a future period; generating a hole collapse early warning level based on the hole collapse risk precursor feature set in the future period; and performing risk early warning on a hole collapse of a soft coal seam based on the hole collapse early warning level. The application can improve the accuracy of mine risk early warning and ensure the safety of mine operation.

Description

Technical Field

[0001] This application belongs to the field of mine monitoring technology, and more specifically, relates to a method and device for early warning of collapse in soft coal seams, electronic equipment, and storage medium. Background Technology

[0002] Soft and fractured coal seams are a typical type of low-strength coal seam in coal mining. They are characterized by a loose and fragmented coal structure, weak interparticle cohesion, and often accompanied by high moisture content and high ground stress. During drilling operations such as gas drainage, coal seam water injection, and geological exploration, drilling disrupts the original stress balance of the coal seam. The soft and fractured coal around the borehole wall is prone to collapse and spalling under stress, resulting in borehole collapse. Borehole collapse not only blocks the borehole passage, leading to decreased gas drainage efficiency and interruption of water injection operations, but may also induce coal dust ejection, gas outbursts, and other safety accidents, seriously threatening the lives of underground workers and hindering efficient and safe coal mine production. Therefore, real-time and accurate early warning of borehole collapse risks during drilling in soft and fractured coal seams is a crucial link in ensuring the smooth progress of drilling operations and mine safety.

[0003] Existing borehole collapse early warning methods do not fully consider the multi-factor coupling effect under complex geological conditions of fractured and soft coal seams. They only focus on a single or a few factors (such as coal seam hardness and burial depth) and fail to fully recognize the intertwined effects of various factors such as coal seam moisture content, ground stress distribution, and mining impact on borehole collapse, resulting in low early warning accuracy. Summary of the Invention

[0004] The purpose of this application is to provide a method and device for early warning of borehole collapse in soft and fractured coal seams, an electronic device, and a storage medium to improve the prediction accuracy of borehole collapse in soft and fractured coal seams.

[0005] A first aspect of this application provides a method for early warning of borehole collapse in soft coal seams, comprising: Obtain standardized time-series data of the drill bit during the drilling process in the current period. The standardized time-series data is data of different monitoring dimensions obtained after preprocessing the original time-series data. The original time-series data includes drilling resistance parameter sequence values, slag discharge status parameter sequence values, and borehole environment parameter sequence values. The standardized time-series data of the drill bit during the drilling process in the current period are spatiotemporally aligned and feature-enhanced to obtain feature-enhanced data; Spatiotemporal attention convolutional networks are used to extract spatiotemporal correlation features of different monitoring dimensions in the data to generate multi-field coupled feature tensors. The multi-field coupled feature tensor is input into the bi-branch adversarial time series prediction model to obtain the feature set of precursors of hole collapse risk in the future period. One branch of the bi-branch adversarial time series prediction model is the trend prediction branch, and the other branch is the mutation prediction branch. The feature set of precursors of hole collapse risk is obtained by fusing the output results corresponding to the trend prediction branch and the mutation prediction branch respectively. Based on the feature set of precursors to borehole risk in future time periods, a borehole collapse warning level is generated, and a risk warning for borehole collapse in soft and fractured coal seams is given based on the borehole collapse warning level.

[0006] A second aspect of this application provides a borehole collapse early warning device for soft coal seams, comprising: The data acquisition unit is used to acquire standardized time-series data of the drill bit during the drilling process in the current period. The standardized time-series data are data of different monitoring dimensions obtained after preprocessing the original time-series data. The original time-series data includes drilling resistance parameter sequence values, slag discharge status parameter sequence values, and borehole environment parameter sequence values. The data augmentation unit is used to perform spatiotemporal alignment and feature enhancement processing on the standardized time-series data of the drill bit during the drilling process in the current period to obtain feature-enhanced data; The feature extraction unit is used to extract spatiotemporal correlation features of different monitoring dimensions in the feature enhancement data through a spatiotemporal attention convolutional network, and generate a multi-field coupled feature tensor. The prediction unit is used to input the multi-field coupled feature tensor into the two-branch adversarial time series prediction model to obtain the feature set of precursors of hole collapse risk in the future period. One branch of the two-branch adversarial time series prediction model is the trend prediction branch, and the other branch is the mutation prediction branch. The feature set of precursors of hole collapse risk is obtained by fusing the output results corresponding to the trend prediction branch and the mutation prediction branch respectively. The risk warning unit is used to generate a collapse warning level based on the feature set of collapse risk precursors for future periods, and to provide risk warnings for collapses in soft and fractured coal seams based on the collapse warning level.

[0007] A third aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the above-described method for early warning of collapse in soft coal seams.

[0008] In a fourth aspect of this application, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the above-described method for early warning of collapse in soft coal seams.

[0009] The beneficial effects of the early warning method and device for collapse of soft coal seams, electronic equipment, and storage medium provided in this application are as follows: On the one hand, this embodiment acquires standardized time-series data of the drill bit drilling in the current period, covering multiple dimensions of raw data such as drilling resistance, slag removal status, and borehole environment, and preprocesses them; then, it performs spatiotemporal alignment and feature enhancement processing on the preprocessed data, and then extracts spatiotemporal correlation features of different monitoring dimensions through a spatiotemporal attention convolutional network to generate multi-field coupled feature tensors, which fully explores the influence of the intertwining of various factors on borehole collapse, can more accurately grasp the risk of borehole collapse, effectively overcome the problem of poor accuracy caused by the incomplete consideration of factors in existing methods, and significantly improve the accuracy of borehole collapse early warning.

[0010] On the other hand, this embodiment inputs a multi-field coupled feature tensor into a bi-branch adversarial time-series prediction model. The model processes the data through a trend prediction branch and a mutation prediction branch, and then fuses the outputs to obtain a feature set of precursors to borehole risk in future time periods. This bi-branch design in this embodiment can simultaneously capture both trend changes and abrupt changes in borehole risk, making the assessment of future borehole risk more comprehensive and reliable. Generating a borehole warning level based on this feature set and conducting risk warnings can provide more accurate and timely risk information for underground workers, effectively avoiding safety accidents caused by unreliable warnings, ensuring the smooth progress of drilling operations and mine safety, and enhancing the reliability of the entire borehole warning system. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 A flowchart illustrating a method for early warning of collapse in soft coal seams according to an embodiment of this application; Figure 2 This is a structural block diagram of a pre-collision warning device for soft coal seams provided in an embodiment of this application; Figure 3 This is a schematic block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0013] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0014] It should be noted that the terms "first," "second," etc., used in the specification, claims, and drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in sequences other than those illustrated or described herein.

[0015] To address the aforementioned technical issues, this application provides a method for early warning of borehole collapse in soft and fractured coal seams. By extracting features from data across multiple dimensions and inputting them into a bi-branch adversarial time-series prediction model, prediction data for future periods is obtained, thereby generating a borehole collapse early warning level and providing risk warnings for borehole collapse in soft and fractured coal seams.

[0016] To make the objectives, technical solutions, and advantages of this application clearer, the following description will be provided in conjunction with the accompanying drawings and specific embodiments.

[0017] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating a method for early warning of coal seam collapse in an embodiment of this application. The method may include: S101: Obtain standardized time-series data of the drill bit during the drilling process in the current period. The standardized time-series data are data of different monitoring dimensions obtained after preprocessing the original time-series data. The original time-series data includes drilling resistance parameter sequence values, slag discharge status parameter sequence values, and borehole environment parameter sequence values.

[0018] In this embodiment, when the drill bit performs tasks such as gas extraction, coal seam water injection, or geological exploration in a fractured and soft coal seam, the drilling, cutting, and squeezing action with the fractured and soft coal body may disrupt the original stress balance of the coal body, causing borehole collapse. Therefore, data monitoring is required during this stage to enable timely prediction of borehole collapse. The original time-series data that can be collected during the drilling process include drilling resistance parameter sequence values, slag discharge status parameter sequence values, and borehole environment parameter sequence values.

[0019] Among them, the drilling resistance parameter sequence value refers to the continuous time series value of the resistance parameters generated by the coal body against the drill bit / drill rod when the drill bit drills into the soft and fractured coal seam. It is a key parameter reflecting the mechanical properties of the coal body, the difficulty of drilling, and the stress state of the borehole wall. Specifically, it includes the real-time continuous monitoring sequence value of at least one of the following: drill bit axial pressure, drilling speed, and friction force between the drill rod and the coal body. The parameters are dynamically updated with drilling time to form a continuous data sequence. Drill bit axial pressure and friction force between the drill rod and the coal body can be collected by pressure sensors, and drilling speed can be collected by speed sensors.

[0020] The slag discharge state parameter sequence refers to the continuous time series values ​​of the state parameters during the coal powder discharge process in borehole operations. These are key parameters reflecting the amount of coal powder generated, its smooth transport, and whether the slag discharge within the borehole is blocked. They are directly related to the precursory characteristics of coal mass collapse and rockfall in the borehole wall. Specifically, they include at least one corresponding continuous monitoring sequence value of slag discharge flow rate and coal powder moisture content, forming a continuous data sequence that dynamically changes with drilling time. Slag discharge flow rate can be collected using a flow meter, and coal powder moisture content can be collected using a soil detection sensor.

[0021] The borehole environmental parameter sequence refers to the continuous time series values ​​of environmental state parameters within the borehole space. These directly characterize the environmental factors inducing borehole collapse risk. Specifically, they include at least one corresponding continuous monitoring sequence value from borehole wall temperature, borehole mud pressure, and borehole wall radial displacement, forming a continuous data sequence that dynamically changes with drilling time. Borehole wall temperature can be acquired using temperature sensors, borehole mud pressure can be acquired using pressure sensors, and borehole wall radial displacement can be acquired using displacement sensors.

[0022] The aforementioned drilling resistance parameter sequence, slag discharge status parameter sequence, and borehole environment parameter sequence are data collected during the current time period. The current time period can be understood as the time interval elapsed since the last data collection ended, up to a preset time. Because the raw time series data collected during the current time period may contain noise interference, missing data, or outlier deviations, it is necessary to preprocess the raw time series data to obtain valid data for subsequent calculations.

[0023] In one embodiment of this application, preprocessing the original time series data to obtain standardized time series data includes: Preprocessing the data for each monitoring dimension in the original time series data yields standardized time series data corresponding to the monitoring dimension. Preprocessing includes at least one of the following: outlier identification and removal, missing value imputation, data normalization, and data smoothing and denoising.

[0024] In this embodiment, each type of parameter represents a monitoring dimension. Since the original time-series data includes three types of parameters—drilling resistance parameter sequence values, cuttings discharge status parameter sequence values, and borehole environment parameter sequence values—there are also three corresponding monitoring dimensions. Preprocessing operations are performed on the data for each monitoring dimension. These preprocessing operations include at least one of the following: outlier identification and removal, missing value imputation, data normalization, and data smoothing and denoising.

[0025] The system includes outlier identification and removal, which identifies data exceeding a preset threshold as invalid outliers and removes them. Missing value imputation addresses missing values ​​caused by sensor data transmission interruptions or other reasons. It uses interpolation to assign dynamic attention weights to 30 valid data points before and after the missing time period, with the weights inversely proportional to the distance from the data point to the missing time period. Imputed values ​​are generated through weighted fitting. Data normalization uses an improved Z-score standardization method to normalize the data, eliminating dimensional differences and facilitating subsequent calculations. Data smoothing and denoising removes noise components from the data using wavelet packet decomposition.

[0026] Raw time series data can be preprocessed to obtain standardized time series data without missing values ​​or interference. This standardized time series data is the core input data for subsequent spatiotemporal alignment, feature enhancement, and model prediction.

[0027] S102: Perform spatiotemporal alignment and feature enhancement processing on the standardized time-series data of the drill bit during the drilling process in the current period to obtain feature-enhanced data.

[0028] In this embodiment, because the sensors collecting data are installed in different locations, there are problems such as time asynchrony, spatial misalignment, and the masking of effective features. Therefore, it is necessary to first perform spatiotemporal alignment on the standardized time-series data, and then perform feature enhancement processing so that the enhanced data can highlight the abrupt changes in the precursory features of hole collapse. The specific steps are as follows: First, for standardized time-series data across different monitoring dimensions, using the time and spatial axes of drilling operations as dual benchmarks, asynchronous acquisition biases in time and blind spots in monitoring locations in space are eliminated. This achieves unified regularization and synchronous matching of multi-dimensional data in the spatiotemporal dimensions, ensuring spatiotemporal continuity and consistency. Second, based on spatiotemporal alignment, the data undergoes effective feature amplification, invalid noise suppression, and extraction of precursor features of sudden changes. This strengthens core features strongly correlated with the risk of borehole collapse in soft coal seams, weakens irrelevant interference features, and enhances the data's ability to characterize precursors of borehole collapse, ultimately outputting feature-enhanced data.

[0029] S103: Extract spatiotemporal correlation features of different monitoring dimensions in the data by using a spatiotemporal attention convolutional network to generate multi-field coupled feature tensors.

[0030] In this embodiment, the spatiotemporal attention convolutional network can identify spatial correlation features and temporal evolution features that contribute highly to the risk of hole collapse, and extract these features to obtain a multi-field coupled feature tensor.

[0031] The spatiotemporal attention convolutional network comprises a spatial attention module, a temporal attention module, a deep convolutional feature extraction module, and a feature fusion module. The spatial attention module extracts the spatial coupling features of data from different monitoring dimensions, while the temporal attention module extracts the temporal coupling features. The order in which these two modules process the data is not specified. That is, feature-enhanced data can be processed first through the spatial attention module and then through the temporal attention module, or vice versa. The deep convolutional feature extraction module and the feature fusion module extract and fuse features from the data processed by the attention modules to obtain a multi-field coupled feature tensor.

[0032] Multi-field coupling feature tensor is a high-dimensional structured feature data carrier. It is a tensor data that integrates three types of physical field features: drilling resistance, slag removal status, and borehole environment. Its dimension is "time step × number of monitoring dimensions × number of feature dimensions". It fully preserves the coupling relationship, evolution law and risk orientation of spatiotemporal correlation features, and serves as the feature data for subsequent input into the bi-branch adversarial time series prediction model.

[0033] S104: Input the multi-field coupled feature tensor into the bi-branch adversarial time series prediction model to obtain the feature set of precursors of hole collapse risk in the future period. One branch of the bi-branch adversarial time series prediction model is the trend prediction branch, and the other branch is the mutation prediction branch. The feature set of precursors of hole collapse risk is obtained by fusing the output results corresponding to the trend prediction branch and the mutation prediction branch respectively.

[0034] In this embodiment, the dual-branch adversarial time series prediction model is a composite time series prediction model that includes two independent and collaborative prediction branches: a trend prediction branch and a mutation prediction branch. By predicting in parallel with the dual branches, it can simultaneously capture the long-term evolution trend and short-term mutation characteristics of hole collapse risk, thus solving the shortcomings of traditional models that make single predictions and miss mutation risks.

[0035] In this embodiment, by inputting the multi-field coupled feature tensor into the bi-branch adversarial time series prediction model, the gradual trend characteristics and sudden mutation characteristics of the risk of hole collapse in the future period can be obtained, that is, the feature set of precursors of hole collapse risk. This can predict the probability of hole collapse in the future period and make early warning preparations.

[0036] S105: Generate a collapse warning level based on the feature set of potential collapse risks in future time periods, and provide risk warnings for collapses in soft and fractured coal seams based on the collapse warning level.

[0037] In this embodiment, a risk assessment index system for borehole collapse in fractured and soft coal seams is constructed. The index system uses the trend evolution characteristics and abrupt change probability characteristics of the precursor features of borehole collapse risk as core evaluation indicators. The trend evolution characteristics cover the slope and amplitude change rate of continuous changes in various monitoring parameters of drilling resistance, slag removal status, and borehole environment over future periods; the abrupt change probability characteristics cover the probability of sudden increases, sudden decreases, and excessive fluctuations of various monitoring parameters. This embodiment uses the entropy weight-analytic hierarchy process (AHP) combined weighting method to determine the weight coefficients of each evaluation index. Based on these weight coefficients, a weighted calculation is performed on each evaluation index to obtain the comprehensive borehole collapse risk assessment value. Since the entropy weight-analytic hierarchy process is existing technology, this embodiment only uses this method to process data without improving the method itself; therefore, the working principle of this method will not be elaborated here. Four levels of early warning threshold intervals—no risk, low risk, medium risk, and high risk—are preset. The comprehensive borehole collapse risk assessment value is matched one by one with each of the four levels of early warning threshold intervals, generating a unique borehole collapse early warning level after matching the corresponding interval.

[0038] In this embodiment, different collapse warning levels have different warning methods, thus risk warnings can be issued for collapses in soft coal seams based on the collapse warning level. For example, when there is no risk level, a green warning signal is output, maintaining normal drilling operation parameters, continuously collecting monitoring data and updating the collapse risk precursor feature set; when the risk level is low, a blue warning signal is output, triggering a voice prompt warning, automatically adjusting drilling and wall protection parameters, reducing drilling speed by 5%~10%, increasing mud wall protection pressure by 10%~15%, and enhancing slag removal efficiency; when the risk level is medium, a yellow warning signal is output, triggering an audible and visual alarm, immediately suspending drilling operations, initiating the high-pressure injection program for borehole wall reinforcement, activating the high-frequency monitoring mode for gas concentration, and notifying safety management personnel for on-site verification and handling; when the risk level is high, a red emergency warning signal is output, triggering a full-area underground audible and visual alarm and emergency linkage command, immediately executing an emergency drill pipe retraction operation, activating the borehole sealing device, cutting off the power supply within a preset range around the borehole, organizing the evacuation of workers to a safe area, and simultaneously uploading the warning information to the mine safety management platform.

[0039] In summary, on the one hand, this embodiment acquires standardized time-series data of the drill bit drilling in the current period, covering multiple dimensions of raw data such as drilling resistance, slag removal status, and borehole environment, and preprocesses them; then, it performs spatiotemporal alignment and feature enhancement processing on the preprocessed data, and then extracts spatiotemporal correlation features of different monitoring dimensions through a spatiotemporal attention convolutional network to generate multi-field coupled feature tensors. This fully explores the influence of the intertwining of various factors on borehole collapse, enabling a more accurate grasp of borehole collapse risk, effectively overcoming the problem of poor accuracy caused by incomplete consideration of factors in existing methods, and significantly improving the accuracy of borehole collapse early warning.

[0040] On the other hand, this embodiment inputs a multi-field coupled feature tensor into a bi-branch adversarial time-series prediction model. The model processes the data through a trend prediction branch and a mutation prediction branch, and then fuses the outputs to obtain a feature set of precursors to borehole risk in future time periods. This bi-branch design in this embodiment can simultaneously capture both trend changes and abrupt changes in borehole risk, making the assessment of future borehole risk more comprehensive and reliable. Generating a borehole warning level based on this feature set and conducting risk warnings can provide more accurate and timely risk information for underground workers, effectively avoiding safety accidents caused by unreliable warnings, ensuring the smooth progress of drilling operations and mine safety, and enhancing the reliability of the entire borehole warning system.

[0041] In one embodiment of this application, standardized time-series data of the drill bit during the drilling process in the current time period is spatiotemporally aligned and feature-enhanced to obtain feature-enhanced data, including: A spatiotemporal coordinate system is constructed with the drilling time of the drill bit as the time reference axis and the drilling depth as the spatial reference axis. The standardized time series data corresponding to each monitoring dimension are spatiotemporally aligned using the spatiotemporal coordinate system to obtain a spatiotemporally continuous monitoring dataset. Determine the abnormal factor value of each data point in the monitoring dataset, and mark the data points with abnormal factor values ​​greater than a preset threshold as mutation feature points. The preset thresholds are different for data points of different monitoring dimensions. Calculate the energy entropy of the characteristic frequency band corresponding to the abrupt change feature point, and fuse the energy entropy with the original standardized time series data to obtain feature-enhanced data.

[0042] In this embodiment, in order to enhance the features of the standardized time-series data corresponding to each monitoring dimension to highlight the data related to the precursors of borehole collapse, spatiotemporal alignment is first required. The specific implementation steps are as follows: A unified spatiotemporal coordinate system is constructed with drill bit drilling time as the time reference axis and borehole depth as the spatial reference axis. This coordinate system provides a unified time-depth correlation benchmark for multi-dimensional data. Based on this coordinate system, standardized time-series data of each monitoring dimension are correlated and matched to eliminate time deviations caused by differences in sampling frequencies of different sensors. At the same time, it fills in the spatial blind spots caused by insufficient monitoring points, enabling the originally scattered multi-dimensional data to achieve a precise correspondence of "the same time node - the same borehole depth", resulting in a spatiotemporally continuous and uninterrupted monitoring dataset.

[0043] Secondly, feature enhancement is performed on the spatiotemporally aligned monitoring dataset. The main steps are as follows: To determine the degree to which each data point in the monitoring dataset deviates from the normal data distribution, it is necessary to identify the outlier factors. Since different monitoring dimensions contain data with different units, such as the drilling resistance parameter dimension which includes drill bit axial pressure and drilling speed (pressure is measured in kN, speed in m / min), preset thresholds need to be set for each dimension. For example, the preset threshold for drill bit axial pressure is 90 kN, and the preset threshold for drilling speed is 0.8 m / min.

[0044] Data points with abnormal factor values ​​exceeding a preset threshold are marked as abrupt change feature points strongly correlated with precursors of borehole collapse, thus filtering precursor abrupt change signals. Then, wavelet packet decomposition is used to extract the feature bands corresponding to each abrupt change feature point, and the energy entropy of these feature bands is calculated to quantify the intensity and identifiability of the abrupt change features. Finally, a feature-weighted concatenation method is used to fuse the energy entropy with the original standardized time-series data. This preserves the temporal evolution patterns of the original data while enhancing the salience of precursor abrupt change features, ultimately yielding feature-enhanced data that combines spatiotemporal continuity, feature salience, and risk orientation. This provides high-quality data support for subsequent extraction of multi-field coupled feature tensors and borehole risk prediction.

[0045] This embodiment uses a feature-weighted concatenation method to fuse energy entropy with the original standardized time series data. An example is shown below: Original standardized time series data: After preprocessing, it has been normalized to the [0,1] interval. Taking drill bit axial pressure (the core dimension of drilling resistance) as an example, the standardized time series data collected in the current period is as follows: X = [0.2, 0.3, 0.4, 0.7, 0.6, 0.5], a total of 6 time step data points, with 0.7 being the identified abrupt change feature point, corresponding to a slight precursor to borehole wall collapse. Energy entropy calculation was performed on the characteristic frequency band of the above abrupt change feature point (0.7), yielding an energy entropy value E = 0.85 (a larger energy entropy value indicates a more significant precursor to borehole collapse). Calibrated according to the condition of a fractured and soft coal seam, the energy entropy weighting coefficient w1 = 0.6, and the original data weighting coefficient w2 = 0.4. Then, only the marked abrupt change feature point (0.7) was weighted using energy entropy, while non-abrupt change points retained their original standardized values.

[0046] Mutation point weighting = original mutation value × w2 + energy entropy value × w1 = 0.7 × 0.4 + 0.85 × 0.6 = 0.79.

[0047] The weighted and enhanced mutation point values ​​replace the corresponding points in the original data, while the original values ​​of the remaining non-mutation points are retained. The data are then directly spliced ​​together to form the fused feature-enhanced data. This approach does not amplify dimensions but only enhances mutation features, thus meeting the continuity requirements of time-series data.

[0048] Original standardized time series data: X=[0.2,0.3,0.4,0.7,0.6,0.5]; Weighted splicing and fusion of feature-enhanced data: X=[0.2,0.3,0.4,0.79,0.6,0.5].

[0049] As can be seen from the above, this embodiment achieves multi-dimensional data spatiotemporal alignment by constructing a spatiotemporal coordinate system, ensuring data continuity; by marking abrupt change feature points and calculating their characteristic frequency band energy entropy, and then fusing them with the original data, key features can be accurately captured, enhancing data expressiveness and providing strong support for subsequent extraction of effective spatiotemporal correlation features and accurate early warning of hole collapse risks.

[0050] In one embodiment of this application, a spatiotemporal correlation feature of different monitoring dimensions in the feature enhancement data is extracted using a spatiotemporal attention convolutional network to generate a multi-field coupled feature tensor, including: The feature enhancement data is sequentially input into the spatial attention module, temporal attention module, deep convolutional feature extraction module, and feature fusion module of the spatiotemporal attention convolutional network to obtain a multi-field coupled feature tensor.

[0051] In one implementation, feature enhancement data is sequentially input into the spatial attention module, temporal attention module, deep convolutional feature extraction module, and feature fusion module of a spatiotemporal attention convolutional network to obtain a multi-field coupled feature tensor, including: The feature-enhanced data is input into the spatial attention module to calculate the contribution weights of data from different monitoring dimensions to the collapse early warning, thus obtaining a spatially weighted feature matrix. Input the spatial weighted feature matrix into the temporal attention module to calculate the attention weights of data at different times and obtain the spatiotemporal correlation feature matrix; The spatiotemporal correlation feature matrix is ​​input into the deep convolution feature extraction module to extract local correlation features. The spatiotemporal correlation features and local correlation features are then input into the feature fusion module for fusion to obtain a multi-field coupled feature tensor with time step, monitoring dimension, and feature dimension.

[0052] In this embodiment, the spatiotemporal attention convolutional network has a structure of 3 convolutional layers + 2 pooling layers + 1 fully connected layer. The deep convolutional feature extraction module extracts local correlation features from the spatiotemporal weighted feature matrix through 3 convolutional layers with a kernel size of 3×3 and ReLU activation function. It then reduces the feature dimensionality through 2 max pooling layers with a kernel size of 2×2. Finally, the fully connected layer outputs the local correlation features.

[0053] In this embodiment, spatiotemporal correlation features and local correlation features are input into the feature fusion module, and multi-field coupled feature tensors are obtained by element-wise weighted addition or channel-dimension weighted splicing.

[0054] The dimensions of the spatiotemporal correlation features are [6×3×16] (6 time steps, 3 monitoring dimensions, and 16-dimensional global spatiotemporal coupling features). The dimensions of the local correlation features are [6×3×8] (6 time steps, 3 monitoring dimensions, and 8-dimensional local fine correlation features).

[0055] If the fusion is performed by weighted addition of elements, the final fusion generates a multi-field coupled feature tensor with a dimension of [6×3×16]. The feature dimension is consistent with the original spatiotemporal correlation feature, and there is no need to expand the dimension, which greatly reduces the computational load of the subsequent model and is suitable for practical scenarios with limited downhole computing power.

[0056] If the channel dimension weighted splicing method is used for fusion, the final fusion generates a multi-field coupled feature tensor with a dimension of [6×3×24]. This tensor integrates global spatiotemporal correlation features and local fine correlation features to achieve deep coupling of multi-field features of drilling mechanical field, slag removal and transport field and borehole environment field.

[0057] As can be seen from the above, this embodiment first calculates the contribution weights of different monitoring dimensions through a spatial attention module, and then obtains the attention weights at different times through a temporal attention module. Next, a deep convolutional feature extraction module is used to mine local correlation features, fusing spatiotemporal and local correlation features to obtain a multi-field coupled feature tensor. The data processing in this embodiment is progressively layered, enabling comprehensive extraction of key features, improving the ability to perceive the risk of borehole collapse, and enhancing the accuracy of early warnings.

[0058] In one embodiment of this application, a multi-field coupled feature tensor is input into a bi-branch adversarial time series prediction model to obtain a set of precursor features of hole collapse risk for future time periods, including: The multi-field coupled feature tensor is divided into multiple feature sub-tensors according to the time step, and each feature sub-tensor corresponds to a continuous monitoring period. Multiple feature sub-tensors are input in parallel to the trend prediction branch and the mutation prediction branch of the two-branch adversarial time series prediction model to obtain the smooth trend value sequence of each monitoring parameter in the future period and the mutation probability distribution matrix. The mutation probability distribution matrix contains probability values ​​in three dimensions: parameter sudden increase probability, parameter sudden decrease probability, and parameter fluctuation amplitude exceeding the limit. The smoothed trend value sequence of each monitoring parameter in the future period and the mutation probability distribution matrix are fused to obtain the feature set of early signs of hole collapse risk.

[0059] As can be seen from the above, by dividing the multi-field coupled feature tensor into parallel inputs to the dual-branch model according to the time step, the trend prediction branch can capture the smooth trend of the data, and the mutation prediction branch can obtain the mutation probability distribution. The feature set of the early warning of hole collapse risk obtained by fusing the two can take into account both the trend and mutation situation, comprehensively reflect the risk probability of hole collapse, and improve the accuracy and reliability of the early warning.

[0060] In this embodiment, the smooth trend value sequence represents the output of the trend prediction branch of the dual-branch adversarial time series prediction model. It is a continuous and smooth numerical sequence of the changes of various monitoring parameters such as drilling resistance, slag removal status, and borehole environment with time step in the future period, reflecting the long-term gradual evolution law of each parameter.

[0061] The mutation probability distribution matrix is ​​a three-dimensional matrix with dimensions of [number of future prediction periods × number of monitoring dimensions × 3]. The "3" corresponds to three core dimensions: the probability of a sudden increase in parameter value, the probability of a sudden decrease in parameter value, and the probability of parameter fluctuation exceeding limits. Each value in the matrix represents a probability value between 0 and 1 (the larger the value, the higher the probability of the corresponding mutation type occurring). Specifically, the probability of a sudden increase in parameter value refers to the probability that the value of a monitored parameter will suddenly rise above the normal threshold range within a short period (≤ preset time threshold), such as a sudden increase in drilling torque, a sudden increase in gas concentration, and a sudden increase in borehole radial displacement. This is a core precursor probability indicator for sudden changes in coal body stress and initial borehole wall collapse in soft coal seams. The probability of a sudden decrease in parameter value refers to the probability that the value of a monitored parameter will suddenly decrease below the normal threshold range within a short period (≤ preset time threshold), such as a sudden decrease in slag discharge flow rate, a sudden decrease in mud wall pressure, and a sudden decrease in drilling speed. This is a core precursor probability indicator for coal dust accumulation in boreholes, drill pipe jamming, and borehole wall blockage in soft coal seams. The probability of parameter fluctuation exceeding the limit refers to the probability that the value of a certain monitoring parameter will fluctuate beyond the normal fluctuation range in a short period of time (≤ preset time threshold), such as large fluctuations in drill pipe vibration acceleration. It is a core probability indicator that is a precursor to the deterioration of coal body stability and borehole wall instability in soft coal seams.

[0062] In this embodiment, the working principle of inputting the multi-field coupled feature tensor into the bi-branch adversarial time series prediction model to obtain the precursor feature set of hole collapse risk for future periods is as follows: First, the time step is determined based on the borehole depth. The high-dimensional structured multi-field coupled feature tensor is divided into multiple feature sub-tensors. Each feature sub-tensor independently corresponds to a continuous monitoring period, realizing the block decoupling of the high-dimensional tensor. This preserves the integrity of the multi-field coupled features of drilling resistance, slag removal status, and borehole environment within a single period, while ensuring the temporal continuity of features in each period. At the same time, the block division of sub-tensors reduces the computational load of parallel model calculations, adapting to the engineering requirements of real-time downhole early warning.

[0063] Secondly, in this embodiment, the multiple feature sub-tensors after division are synchronously and in parallel input into the two core branches of the dual-branch adversarial time series prediction model. The trend prediction branch can obtain the long-term gradual evolution law of the hole collapse risk, perform time series trend fitting and smoothing on the spatiotemporal coupling characteristics of each monitoring parameter in the feature sub-tensor, filter out local random noise, and output the smooth trend value sequence of each monitoring parameter over time in the future period, accurately representing the slow development trend of the hole collapse risk. The mutation prediction branch focuses on the short-term sudden abnormal characteristics of the hole collapse risk, identifies and probabilistically quantifies the mutation signals of parameters strongly correlated with the hole collapse in the feature sub-tensor, and outputs a mutation probability distribution matrix containing the probability of parameter sudden increase, parameter sudden decrease, and parameter fluctuation amplitude exceeding the limit, representing the possibility of abnormal mutation of each monitoring parameter in the future period.

[0064] In one embodiment of this application, an m% temporal overlap region is provided between multiple feature sub-tensors. ; Early warning methods for collapse in soft coal seams also include: The time step is determined based on the borehole depth.

[0065] In this embodiment, an m% time overlap region is set between multiple feature sub-tensors to avoid the loss of feature information across time periods, which would affect the analysis results. The emergence and development of borehole collapse risk in soft coal seams is a continuous temporal evolution process. Core borehole collapse precursor features such as sudden changes in drilling resistance and a sharp drop in slag discharge flow often appear at the time boundary between two feature sub-tensors. Without a time overlap region, such continuous abrupt changes across time steps would be interrupted. Setting an m% (preferably 20%~30%) time overlap region allows adjacent feature sub-tensors to share some time-series data, ensuring that borehole collapse precursor features across time steps can be completely included in at least one feature sub-tensor. At the same time, by associating the time-series data in the overlap region, the feature connectivity between adjacent sub-tensors is strengthened, avoiding time-series feature discontinuities. This ensures that the subsequent bi-branch adversarial time-series prediction model can accurately capture the continuous evolution law of borehole collapse risk and effectively reduce the missed detection rate of borehole collapse precursor features.

[0066] In this embodiment, to ensure efficient data collection and analysis in both shallow and deep borehole sections, the time step needs to be adaptively adjusted based on the borehole depth. For example, when the borehole depth is ≤100m, the coal body structure of the fractured and soft coal seam is relatively stable, and the frequency of pre-collapse abrupt changes is low. A larger time step (e.g., 1min / step) is set accordingly to reduce data redundancy and computing power consumption while ensuring the integrity of the temporal features. When the borehole depth is >100m and ≤200m, the coal body looseness and ground stress are significantly increased, and the frequency of pre-collapse abrupt changes increases. A medium time step (e.g., 30s / step) is set accordingly to increase the sampling density of temporal features. When the borehole depth is >200m, the coal body is in a highly fractured and soft state with high stress. The pre-collapse abrupt change features exhibit high frequency and short duration characteristics. A smaller time step (e.g., 10s / 15s / step) is set accordingly to achieve accurate capture of millisecond-level and short-duration abrupt change precursors.

[0067] The above can be seen from the fact that by dynamically narrowing the time step with the drilling depth, and taking into account the needs of collecting collapse risk characteristics in different borehole sections, it is possible to avoid the defects of missing high-frequency abrupt changes in deep borehole sections and data redundancy in shallow borehole sections when the time step is fixed. This adapts to the coupled evolution law of borehole depth and collapse risk in fractured and soft coal seams.

[0068] Based on the same inventive concept, this application also provides a soft coal seam collapse early warning device for implementing the aforementioned soft coal seam collapse early warning method. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more embodiments of the soft coal seam collapse early warning device provided below can be found in the limitations of the soft coal seam collapse early warning method described above, and will not be repeated here.

[0069] This application provides an early warning device for the collapse of a hole in a soft coal seam, such as... Figure 2 As shown, the early warning device 20 for collapse of soft coal seam includes: a data acquisition unit 21, a data enhancement unit 22, a feature extraction unit 23, a prediction unit 24, and a risk warning unit 25.

[0070] The data acquisition unit 21 is used to acquire standardized time-series data of the drill bit during the drilling process in the current period. The standardized time-series data are data of different monitoring dimensions obtained after preprocessing the original time-series data. The original time-series data includes drilling resistance parameter sequence values, slag discharge status parameter sequence values, and borehole environment parameter sequence values.

[0071] The data augmentation unit 22 is used to perform spatiotemporal alignment and feature enhancement processing on the standardized time-series data of the drill bit during the drilling process in the current period to obtain feature-enhanced data.

[0072] The feature extraction unit 23 is used to extract spatiotemporal correlation features of different monitoring dimensions in the feature enhancement data through a spatiotemporal attention convolutional network, and generate a multi-field coupled feature tensor.

[0073] Prediction unit 24 is used to input the multi-field coupled feature tensor into the dual-branch adversarial time series prediction model to obtain the feature set of precursors of hole collapse risk in the future period. One branch of the dual-branch adversarial time series prediction model is the trend prediction branch, and the other branch is the mutation prediction branch. The feature set of precursors of hole collapse risk is obtained by fusing the output results corresponding to the trend prediction branch and the mutation prediction branch respectively.

[0074] Risk warning unit 25 is used to generate a collapse warning level based on the collapse risk precursor feature set for future periods, and to provide risk warning for collapse of soft coal seams based on the collapse warning level.

[0075] In one embodiment, the data acquisition unit 21 is specifically used for: Preprocessing the data for each monitoring dimension in the original time series data yields standardized time series data corresponding to the monitoring dimension. Preprocessing includes at least one of the following: outlier identification and removal, missing value imputation, data normalization, and data smoothing and denoising.

[0076] In one implementation, the data enhancement unit 22 is specifically used for: A spatiotemporal coordinate system is constructed with the drilling time of the drill bit as the time reference axis and the drilling depth as the spatial reference axis. The standardized time series data corresponding to each monitoring dimension are spatiotemporally aligned using the spatiotemporal coordinate system to obtain a spatiotemporally continuous monitoring dataset. Determine the abnormal factor value of each data point in the monitoring dataset, and mark the data points with abnormal factor values ​​greater than a preset threshold as mutation feature points. The preset thresholds are different for data points of different monitoring dimensions. Calculate the energy entropy of the characteristic frequency band corresponding to the abrupt change feature point, and fuse the energy entropy with the original standardized time series data to obtain feature-enhanced data.

[0077] In one implementation, the feature extraction unit 23 is specifically used for: The feature enhancement data is sequentially input into the spatial attention module, temporal attention module, deep convolutional feature extraction module, and feature fusion module of the spatiotemporal attention convolutional network to obtain a multi-field coupled feature tensor.

[0078] In one implementation, the feature extraction unit 23 is specifically used for: The feature-enhanced data is input into the spatial attention module to calculate the contribution weights of data from different monitoring dimensions to the collapse early warning, thus obtaining a spatially weighted feature matrix. Input the spatial weighted feature matrix into the temporal attention module to calculate the attention weights of data at different times and obtain the spatiotemporal correlation feature matrix; The spatiotemporal correlation feature matrix is ​​input into the deep convolution feature extraction module to extract local correlation features. The spatiotemporal correlation features and local correlation features are then input into the feature fusion module for fusion to obtain a multi-field coupled feature tensor with time step, monitoring dimension, and feature dimension.

[0079] In one implementation, the prediction unit 24 is specifically used for: The multi-field coupled feature tensor is divided into multiple feature sub-tensors according to the time step, and each feature sub-tensor corresponds to a continuous monitoring period. Multiple feature sub-tensors are input in parallel to the trend prediction branch and the mutation prediction branch of the two-branch adversarial time series prediction model to obtain the smooth trend value sequence of each monitoring parameter in the future period and the mutation probability distribution matrix. The mutation probability distribution matrix contains probability values ​​in three dimensions: parameter sudden increase probability, parameter sudden decrease probability, and parameter fluctuation amplitude exceeding the limit. The smoothed trend value sequence of each monitoring parameter in the future period and the mutation probability distribution matrix are fused to obtain the feature set of early signs of hole collapse risk.

[0080] In one implementation, an m% temporal overlap region is provided between multiple feature subtensors. .

[0081] In one embodiment, the soft coal seam collapse early warning device 20 further includes a computing unit; The calculation unit is used to determine the time step based on the borehole depth.

[0082] See Figure 3 , Figure 3 This is a schematic block diagram of an electronic device provided according to an embodiment of this application. Figure 3 The electronic device 300 in this embodiment may include one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The processors 301, input devices 302, output devices 303, and memories 304 communicate with each other via a communication bus 305. The memories 304 store computer programs, including program instructions. The processors 301 execute the program instructions stored in the memories 304. Specifically, the processors 301 are configured to invoke the program instructions to perform the functions of each unit / unit in the above-described device embodiments, for example... Figure 2 The functions of the data acquisition unit 21, data augmentation unit 22, feature extraction unit 23, prediction unit 24, and risk warning unit 25 are shown.

[0083] It should be understood that, in the embodiments of this application, the processor 301 may be a central processing unit (CPU), but it may also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0084] Input device 302 may include a touchpad, a fingerprint sensor (for collecting the user's fingerprint information and fingerprint orientation information), a microphone, etc., and output device 303 may include a display (LCD, etc.), a speaker, etc.

[0085] The memory 304 may include read-only memory and random access memory, and provides instructions and data to the processor 301. A portion of the memory 304 may also include non-volatile random access memory.

[0086] In specific implementations, the processor 301, input device 302, and output device 303 described in the embodiments of this application can execute the implementation method described in the early warning method for collapse of soft coal seams provided in the embodiments of this application, or they can execute the implementation method of the electronic device described in the embodiments of this application, which will not be repeated here.

[0087] In another embodiment of this application, a computer-readable storage medium is provided. This computer-readable storage medium stores a computer program, which includes program instructions. When executed by a processor, the program instructions implement all or part of the processes in the methods described above. Alternatively, the computer program can instruct related hardware to complete the process. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include any entity or device capable of carrying computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0088] The computer-readable storage medium can be an internal storage unit of the electronic device in any of the foregoing embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., provided on the electronic device. Furthermore, the computer-readable storage medium can include both internal and external storage units of the electronic device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0089] Those skilled in the art will recognize that the modules / units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0090] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the electronic devices and units described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0091] In the several embodiments provided in this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of modules / units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules, units, or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces or modules / units, or it may be an electrical, mechanical, or other form of connection.

[0092] The modules / units described as separate components may or may not be physically separate. Similarly, the components shown as modules / units may or may not be physical modules / units; they may be located in one place or distributed across multiple network modules / units. Some or all of the modules / units can be selected to achieve the purpose of the embodiments of this application, depending on actual needs.

[0093] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for early warning of collapse in soft coal seams, characterized in that, include: Obtain standardized time-series data of the drill bit during the drilling process in the current period. The standardized time-series data is data of different monitoring dimensions obtained after preprocessing the original time-series data. The original time-series data includes drilling resistance parameter sequence values, slag discharge status parameter sequence values, and borehole environment parameter sequence values. The standardized time-series data of the drill bit during the drilling process in the current period is spatiotemporally aligned and feature-enhanced to obtain feature-enhanced data; Spatiotemporal correlation features of different monitoring dimensions in the feature enhancement data are extracted by a spatiotemporal attention convolutional network to generate a multi-field coupled feature tensor. The multi-field coupled feature tensor is input into a two-branch adversarial time series prediction model to obtain a feature set of precursors to the risk of hole collapse in the future period. One branch of the two-branch adversarial time series prediction model is a trend prediction branch, and the other branch is a mutation prediction branch. The feature set of precursors to the risk of hole collapse is obtained by fusing the output results corresponding to the trend prediction branch and the mutation prediction branch respectively. A collapse warning level is generated based on the feature set of precursors to collapse risk in future time periods, and a risk warning for collapse in soft coal seams is given based on the collapse warning level.

2. The method for early warning of collapse in soft coal seams as described in claim 1, characterized in that, The raw time series data is preprocessed to obtain standardized time series data, including: The data for each monitoring dimension in the original time series data are preprocessed to obtain the standardized time series data corresponding to the monitoring dimension. The preprocessing includes at least one of the following: outlier identification and removal, missing value imputation, data normalization, and data smoothing and denoising.

3. The method for early warning of collapse in soft coal seams as described in claim 1, characterized in that, The process of performing spatiotemporal alignment and feature enhancement on the standardized time-series data of the drill bit during the drilling process in the current period to obtain feature-enhanced data includes: A spatiotemporal coordinate system is constructed with the drilling time of the drill bit as the time reference axis and the drilling depth as the spatial reference axis. The standardized time series data corresponding to each monitoring dimension are spatiotemporally aligned using the spatiotemporal coordinate system to obtain a spatiotemporally continuous monitoring dataset. Determine the abnormal factor value of each data point in the monitoring dataset, and mark data points with abnormal factor values ​​greater than a preset threshold as mutation feature points. The preset thresholds are different for data points of different monitoring dimensions. Calculate the energy entropy of the characteristic frequency band corresponding to the mutation feature point, and fuse the energy entropy with the original standardized time series data to obtain feature-enhanced data.

4. The method for early warning of collapse in soft coal seams as described in claim 1, characterized in that, The step of extracting spatiotemporal correlation features of different monitoring dimensions from the feature enhancement data through a spatiotemporal attention convolutional network to generate a multi-field coupled feature tensor includes: The feature enhancement data is sequentially input into the spatial attention module, temporal attention module, deep convolutional feature extraction module, and feature fusion module of the spatiotemporal attention convolutional network to obtain a multi-field coupled feature tensor.

5. The method for early warning of collapse in soft coal seams as described in claim 4, characterized in that, The step of sequentially inputting the feature enhancement data into the spatial attention module, temporal attention module, deep convolutional feature extraction module, and feature fusion module of a spatiotemporal attention convolutional network to obtain a multi-field coupled feature tensor includes: The enhanced feature data is input into the spatial attention module to calculate the contribution weights of different monitoring dimensions to the collapse hole early warning, thus obtaining a spatially weighted feature matrix. The spatially weighted feature matrix is ​​input into the temporal attention module to calculate the attention weights of data at different times, thereby obtaining the spatiotemporal correlation feature matrix. The spatiotemporal correlation feature matrix is ​​input into the deep convolution feature extraction module to extract local correlation features. The spatiotemporal correlation features and the local correlation features are then input into the feature fusion module for fusion to obtain a multi-field coupled feature tensor of time step-monitoring dimension-feature dimension.

6. The method for early warning of collapse in soft coal seams as described in claim 1, characterized in that, The step of inputting the multi-field coupled feature tensor into a two-branch adversarial time-series prediction model to obtain a set of precursor features for future hole collapse risks includes: The multi-field coupled feature tensor is divided into multiple feature sub-tensors according to the time step, and each feature sub-tensor corresponds to a continuous monitoring period. The multiple feature sub-tensors are input in parallel to the trend prediction branch and the mutation prediction branch of the two-branch adversarial time series prediction model to obtain the smooth trend value sequence of each monitoring parameter in the future period and the mutation probability distribution matrix. The mutation probability distribution matrix includes probability values ​​in three dimensions: parameter sudden increase probability, parameter sudden decrease probability, and parameter fluctuation amplitude exceeding the limit. The smoothed trend value sequence of each monitoring parameter in the future time period and the mutation probability distribution matrix are fused to obtain the feature set of early signs of hole collapse risk.

7. The method for early warning of collapse in soft coal seams as described in claim 6, characterized in that, An m% temporal overlap region is set between the multiple feature sub-tensors. ; The method further includes: The time step is determined based on the borehole depth.

8. A pre-collision warning device for soft coal seams, characterized in that, include: The data acquisition unit is used to acquire standardized time-series data of the drill bit during the drilling process in the current period. The standardized time-series data is data of different monitoring dimensions obtained after preprocessing the original time-series data. The original time-series data includes drilling resistance parameter sequence values, slag discharge status parameter sequence values, and borehole environment parameter sequence values. The data augmentation unit is used to perform spatiotemporal alignment and feature enhancement processing on the standardized time-series data of the drill bit during the drilling process in the current period to obtain feature-enhanced data; The feature extraction unit is used to extract spatiotemporal correlation features of different monitoring dimensions in the feature enhancement data through a spatiotemporal attention convolutional network, and generate a multi-field coupled feature tensor. The prediction unit is used to input the multi-field coupled feature tensor into the dual-branch adversarial time series prediction model to obtain the feature set of precursors of hole collapse risk in the future period. One branch of the dual-branch adversarial time series prediction model is a trend prediction branch and the other branch is a mutation prediction branch. The feature set of precursors of hole collapse risk is obtained by fusing the output results corresponding to the trend prediction branch and the mutation prediction branch respectively. The risk warning unit is used to generate a collapse warning level based on the feature set of collapse risk precursors for future periods, and to provide risk warnings for collapses in soft coal seams based on the collapse warning level.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images