Intelligent tunneling support and gas prevention and control system for air oxidation zone unstable coal seam

Abstract

The application provides an intelligent tunneling support and gas prevention and control system for wind oxidation unstable coal seams, which collects microseismic monitoring, multi-point gas concentration time series, geological and equipment working condition data of a tunneling working face; a topological graph is constructed based on a microseismic event, the topological graph structure entropy is calculated, and a first characteristic time series representing crack evolution is obtained; multiple fractal analysis is performed on the gas concentration data, a generalized Hurst index is extracted, and a second characteristic time series representing outburst is obtained; the transfer entropy between the two time series is calculated as a weight, and a comprehensive disturbance feature vector is obtained by weighted fusion; the comprehensive disturbance feature vector and the geological and working condition parameters are input into a pre-trained capsule network and long short-term memory network hybrid model, and a future gas risk level is output, and tunneling support measures are controlled according to the risk level.

Description

Technical Field

[0001] This application belongs to the field of prevention and control, and in particular relates to an intelligent tunneling support and gas control system for unstable coal seams in the wind oxidation zone. Background Technology

[0002] Gas risk prediction methods are based on single information sources, such as extrapolating trends from gas concentration monitoring data alone, or using geological parameters for assessment. These methods struggle to reflect the response process of the "rock-gas" two-phase medium under tunneling disturbances. Multi-source information fusion only provides a simple analysis of the frequency and macroscopic statistical indicators of microseismic events, failing to represent the complexity of the surrounding rock fracture network changes inherent in microseismic events. Furthermore, the analysis of gas emission data ignores the multifractal characteristics exhibited by gas seepage in porous media. Multi-source information fusion cannot reflect the relationships and causal connections between different disturbance sources. The dominant factors leading to increased gas risk may change at different stages of tunneling operations, and the fusion strategy cannot adjust the importance of each feature. In addition, the use of statistical models or shallow machine learning models in predictive model construction is insufficient for processing high-dimensional, time-series tunneling big data, making it difficult to learn the relationships and deep-seated characteristics in the gas risk evolution process. Therefore, how to deeply mine the disturbance features closely related to the gas disaster mechanism from multiple monitoring data and build an intelligent prediction model that can integrate multi-dimensional features is a technical bottleneck that urgently needs to be solved. Summary of the Invention

[0003] This invention proposes an intelligent tunneling support and gas control system for unstable coal seams in wind-oxidized zones. It addresses the problem that existing methods fail to deeply mine disturbance features closely related to gas-causing hazards from multi-physics monitoring data and construct intelligent prediction models capable of fusing multi-dimensional features. The system includes:

[0004] The acquisition module is used to acquire microseismic monitoring data of the tunneling face and surrounding rock, multi-point gas concentration time series data, tunneling geological parameters, and equipment operating condition parameters; The calculation module is used to select microseismic events with energy higher than a preset threshold as nodes based on the microseismic monitoring data, construct a microseismic event topology map according to the proximity relationship, and calculate the structural entropy of the topology map to obtain a first perturbation feature time series representing the complexity of fracture evolution; and to perform multifractal detrending fluctuation analysis on the multi-point gas concentration time series data to calculate the generalized Hearst exponent and obtain a second perturbation feature time series representing the gas emission characteristics. The fusion module is used to calculate the transfer entropy from the first perturbation feature time series to the second perturbation feature time series, and generate weights based on the transfer entropy to perform weighted fusion of the first perturbation feature time series and the second perturbation feature time series to obtain a comprehensive perturbation feature vector. The output module is used to take the integrated disturbance feature vector, the tunneling geological parameters and the equipment operating condition parameters as combined inputs, input a pre-trained hybrid model composed of capsule networks and long short-term memory networks, output the future gas risk level, and adjust the tunneling plan according to the risk level.

[0005] This invention constructs a microseismic event topology map and calculates structural entropy to represent the complexity of changes in the surrounding rock fracture network under tunneling disturbance. Simultaneously, it employs a multifractal detrended fluctuation analysis method to represent the inherent nonlinearity and long-range correlation of the gas outburst process. By utilizing transfer entropy, a causal relationship between the two physical processes of rock mass fracturing and gas migration is established, and based on this, feature fusion is performed to generate a comprehensive index reflecting the disaster-causing mechanism. By combining a hybrid model of capsule networks and long short-term memory networks, the relationships between high-dimensional features can be fully learned, thereby improving the reliability of gas risk prediction for tunneling in unstable coal seams in aeolian oxidation zones and providing decision support for mine disaster prevention and control. Attached Figure Description

[0006] Figure 1 This is a schematic diagram of time-series data on gas concentration at multiple points. Figure 2 This is a schematic diagram illustrating multi-source data acquisition. Figure 3 This is a schematic diagram for detrended volatility analysis; Figure 4 This is a schematic diagram of a hybrid model of capsule network and LSTM. Detailed Implementation

[0007] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0008] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0009] In the first embodiment, the present invention proposes an intelligent tunneling support and gas control system for unstable coal seams in wind-oxidized zones, comprising: The acquisition module is used to acquire microseismic monitoring data of the tunneling face and surrounding rock, multi-point gas concentration time series data, tunneling geological parameters, and equipment operating condition parameters; Multiple microseismic sensors deployed on both sides of the tunneling roadway collect microseismic signals in real time, calculating the occurrence time, three-dimensional spatial coordinates, and energy magnitude of each microseismic event to obtain a microseismic monitoring dataset. Methane sensors are installed at key locations on the tunneling face, including the return airflow, upper corner, and tunneling machine body, to continuously collect methane concentration data at second- or minute-level frequencies, resulting in a multi-point methane concentration time-series dataset. Figure 1 Geological parameters such as coal seam thickness, dip angle, and geological structure (distance from faults, coal permeability) are obtained from geological exploration reports and advanced borehole exploration. Operating parameters such as cutting motor current, traction speed, oil pump pressure, and machine position are retrieved from the tunneling machine's programmable logic controller (PLC) or central control system. Figure 2 .

[0010] The calculation module is used to select microseismic events with energy higher than a preset threshold as nodes based on the microseismic monitoring data, construct a microseismic event topology map according to the proximity relationship, and calculate the structural entropy of the topology map to obtain a first perturbation feature time series representing the complexity of fracture evolution; and to perform multifractal detrending fluctuation analysis on the multi-point gas concentration time series data to calculate the generalized Hearst exponent and obtain a second perturbation feature time series representing the gas emission characteristics. An energy threshold, for example, 100 J, is used to filter out microseismic events with energies exceeding this threshold, thus eliminating environmental noise interference. Each filtered microseismic event is defined as a network node, with attributes including occurrence time t and spatial coordinates x, y, z. A spatial proximity radius R and a temporal proximity window T are set, for example, R is 50 m and T is 24 h. If the temporal difference between two microseismic event nodes i and j is less than T and the spatial Euclidean distance is less than R, an edge is established between the two nodes, connecting all node pairs that meet the conditions, thereby constructing an undirected and unweighted microseismic event topology graph. The degree of each node in the graph, i.e., the number of edges connected to that node, is statistically analyzed, and the degree distribution probability of the entire graph is calculated. The structural entropy of this degree distribution is calculated based on the Shannon entropy formula, representing the complexity of the microseismic event network. When the fracture evolution path changes from centralized to discrete, the disorder of the microseismic event topology structure will significantly increase. The calculation process is repeated within a sliding time window, for example, every hour, thereby generating a time-varying structural entropy sequence, i.e., the first perturbation characteristic time series.

[0011] The average or maximum value of gas concentration data from multiple measuring points at the same time is taken to form a univariate gas concentration time series. Multifractal detrending fluctuation analysis is performed on this series. The cumulative deviation of the series is calculated, and then the cumulative deviation series is divided into multiple non-overlapping sub-intervals of length s. In each sub-interval, the least squares method is used to perform polynomial fitting to determine the local trend, and the root mean square fluctuation after removing the trend is calculated. The fluctuations of all sub-intervals are averaged by order q to obtain the order q fluctuation function. The slope, i.e., the generalized Hearst exponent h(q), is solved by the linear relationship between the order q fluctuation function and the sub-interval length s under double logarithmic coordinates. The generalized Hearst exponent h(2) when q=2 is selected as the index representing the long-range correlation of the gas concentration series. This process is also repeated within the sliding time window to generate an h(2) exponent series that varies with time, i.e., the second perturbation characteristic time series.

[0012] To preserve microseismic activity associated with stress adjustment and potential geological risks, in one optional embodiment, the step of screening microseismic events with energy exceeding a preset threshold as nodes based on the microseismic monitoring data includes: Set energy preset threshold And filter out those with energy higher than the preset energy threshold. Microseismic events.

[0013] Specifically, the determination of this threshold is based on the geological background of the mining area, the statistical distribution of historical microseismic data, and expert experience. For example, for a specific coal mine working face, the energy threshold can be set as follows: J. Distinguish between conventional vibrations caused by equipment operation and real microseismic events generated by rock mass fracturing.

[0014] Assuming the microseismic monitoring system records 1000 events within a 24-hour monitoring period, the energy data constitute a set. The energy of each event will be compared one by one. With preset threshold .like If the microseismic event is valid, it is considered a valid event and retained as a node in the topology graph construction; otherwise, if... If the event is not selected, then the event will be discarded. For example, an event with an energy of 500J will be filtered out, while an event with an energy of 1500J will be selected.

[0015] In an optional embodiment, constructing a microseismic event topology map based on proximity relationships includes: When the spatial distance between any two microseismic event nodes is less than the spatial proximity threshold And the time interval between occurrences is less than the time proximity threshold. When the two nodes are connected, a connecting edge is established.

[0016] Set two key thresholds: spatial proximity threshold and time proximity threshold The selection of the threshold is based on prior knowledge of the fracture propagation rate and stress transmission range of the rock mass in the mining area. For example, it can be set based on experience. m, If two microseismic events occur within 50 meters of each other and are no more than 12 hours apart, they are considered to be physically related.

[0017] The process of constructing the connecting edges is as follows: For any two distinct nodes in the filtered set of microseismic event nodes, node i and node j have coordinates as follows: and Calculate the spatial Euclidean distance between them. and time interval Make a judgment: if and If both conditions are met, an undirected edge is created between node i and node j. For example, if node A occurs at point 10 with position (100, 150, 20) and node B occurs at point 15 with position (120, 160, 30), the calculated... m, h. Since 24.5m < 50m and 5h < 12h, a connection is established between A and B. This process is completed by traversing all node pairs, resulting in a topology graph where edges represent potential triggering or linkage relationships between microseismic events.

[0018] In an optional embodiment, the step of performing multifractal detrending fluctuation analysis on the multi-point gas concentration time-series data and calculating the generalized Hearst exponent includes: The cumulative deviation sequence of gas concentration time series data is divided at different scales s and local trends are eliminated by k-order polynomial fitting. Calculate the q-th order wave function The q-order wave function and the scale s satisfy a power-law relationship. ; The generalized Hearst exponent h(q) is determined by the slope of the linear fit of the relationship in a double logarithmic coordinate system.

[0019] Specifically, for the original gas concentration time series, for example, data containing 10,000 sampling points, the mean of the data is calculated, and a cumulative deviation series is generated. This series is then divided into multiple non-overlapping sub-intervals of length s. For example, when the scale s=100, the series is divided into 100 sub-intervals. Within each sub-interval, a k-th order polynomial, such as a quadratic polynomial with k=2, is used to fit the data to obtain the local trend of that interval. The fitted trend is subtracted from the original data to obtain the detrended series. This eliminates the interference of non-stationary trends on fluctuation analysis, such as... Figure 3 .

[0020] For different q values, the q-order fluctuation function is obtained by taking the q-order average of the fluctuations over all sub-intervals. The parameter q can take a wide range of values, for example, from -5 to 5, to detect fluctuations of different magnitudes in the sequence. Repeating the entire process above, using a series of different scales s, for example from 10 to 1000, yields a series of corresponding values. Value. According to multifractal theory, and There exists a linear relationship between them. Therefore, for each fixed q value, in the double logarithmic coordinate system, for... Linear regression analysis of the data points yields a straight line whose slope is the generalized Hearst exponent h(q) corresponding to that q value. By calculating h(q) for different q values, a complete h(q) spectrum can be obtained, representing the dynamic characteristics of the gas concentration sequence.

[0021] The fusion module is used to calculate the transfer entropy from the first perturbation feature time series to the second perturbation feature time series, and generate weights based on the transfer entropy to perform weighted fusion of the first perturbation feature time series and the second perturbation feature time series to obtain a comprehensive perturbation feature vector. The first perturbation feature time series, i.e., the structural entropy sequence, is used as the source sequence X, and the second perturbation feature time series, i.e., the generalized Hearst exponent sequence, is used as the target sequence Y. The reduction in uncertainty of the next state of Y, given the historical information of Y and X, represents the information flow from X to Y; this reduction is the transfer entropy. A larger transfer entropy value indicates a stronger causal driving effect of the fracture evolution represented by microseismic activity on gas outburst characteristics. The transfer entropy is calculated using a sliding time window method, forming corresponding time-varying transfer entropy sequences at different time periods, which are then normalized to generate a time-related weighting function. At each time point, the calculated transfer entropy value, after normalization, is used as the weight of the first perturbation feature, and the weight of the second perturbation feature is obtained by subtracting this weight from 1. Using this pair of time-varying weights, the two perturbation feature values ​​at the same moment are weighted and summed or concatenated to form a two-dimensional comprehensive perturbation feature vector.

[0022] In some embodiments, calculating the transfer entropy from the first perturbation feature time series to the second perturbation feature time series includes: The first and second perturbation feature time series are discretized, and the continuous values ​​of each time series are divided into a preset number of discrete state intervals to obtain a discretized state sequence. Based on the preset embedding dimension k and time delay τ, the phase space of the obtained discretized state sequence is reconstructed to obtain the historical state vector; Based on the historical state vector, the required joint probability distribution is calculated by statistically analyzing the frequency of different state combinations in the entire time series. Specifically, this includes the joint probability of the next state of the second perturbation feature time series and its historical state, as well as the joint probability of the next state of the second perturbation feature time series and its own historical state and the historical state of the first perturbation feature time series. The conditional probability distribution derived from the calculated joint probability distribution is used to calculate the transfer entropy value from the first perturbation characteristic time series to the second perturbation characteristic time series.

[0023] Specifically, let the time series of the first disturbance characteristic be... The second disturbance characteristic time series is The data ranges of both are divided into Q or a different number of discrete intervals, mapping continuous values ​​to discrete state values. This is based on preset embedding dimensions k and l, and time delays. Here, k is the embedding dimension of the second perturbation feature time series, and l is the embedding dimension of the first perturbation feature time series. Optionally, k equals l. The phase space is reconstructed from the discrete sequences to build the historical state vector of the second perturbation feature time series. and the historical state vector of the first perturbation feature time series The joint probability distribution is calculated by statistically analyzing the frequency of each state combination over a time series of length T, using the following formula:

[0024] in, The target state at time t+1 The denominator represents the number of times the state combination occurs in the entire sequence, and the number of valid samples used for statistics is represented by the denominator.

[0025] Based on the joint probability distribution described above, the conditional probability distribution is derived, and the transfer entropy from the first perturbation feature to the second perturbation feature is calculated using the form of Shannon entropy. Specifically, by measuring the known history of gas itself. Under these conditions, microseismic history is introduced. Future state of gas The reduction in forecast uncertainty is achieved, calculated using the following formula:

[0026] In the formula, the conditional probability is calculated by the ratio of the joint probabilities, i.e. , obtained The larger the value, the more significant the driving effect of the fracture evolution information reflected by microseismic monitoring on gas outburst anomalies.

[0027] To adjust the fusion weights of the two types of disturbance features based on the information flow intensity of microseismic activity on gas concentration fluctuations, in an optional embodiment, the weights generated by the transfer entropy are used to weight and fuse the time series of the first and second disturbance features to obtain a comprehensive disturbance feature vector, including: Calculate the transfer entropy T(t) from the first perturbation characteristic time series S(t) to the second perturbation characteristic time series H(t); Based on the transfer entropy T(t), weights corresponding to S(t) are generated through a normalization function. And let the weights corresponding to H(t) be... ; The combined perturbation eigenvector C(t) is obtained using the following weighted summation formula: .

[0028] For example, using a sliding window with a length of 100 time steps, at each time point t, the degree to which historical information of S(t) reduces the prediction uncertainty of the future value of H(t) is calculated using the data within the window. The calculation result yields a time-varying transfer entropy sequence T(t). If the value of T(t) is high, it indicates that at the current moment, changes in microseismic activity have a strong indicative effect on changes in gas concentration.

[0029] The obtained transfer entropy sequence T(t) is converted into weights ranging from 0 to 1 by a normalization function, such as the sigmoid function or a simple linear normalization. For example, if T(t) reaches its recent maximum value at some point, then It may be close to 1; if T(t) is very small, then It may be close to 0. Correspondingly, the weight of the gas disturbance feature H(t) is... Set as Ensure the sum of the two weights is 1. Calculate the combined perturbation characteristic C(t) using the weighted summation formula. For example, at time t... If calculated ,but This indicates that the overall disturbance at this time is mainly dominated by microseismic features, which enables the fusion features to reflect the changes in the dominant risk factors in different periods. Optionally, S(t) and H(t) are normalized before weighted calculation.

[0030] The output module is used to take the integrated disturbance feature vector, the tunneling geological parameters and the equipment operating condition parameters as combined inputs, input a pre-trained hybrid model composed of capsule networks and long short-term memory networks, output the future gas risk level, and adjust the tunneling plan according to the risk level.

[0031] The two-dimensional comprehensive disturbance feature vector generated in the previous step is concatenated with tunneling geological parameters such as coal seam thickness and equipment operating parameters such as cutting motor current characteristics at the same time step to form a high-dimensional combined input feature vector. The time series of this combined input feature vector is then fed into the hybrid prediction model. The model uses a capsule network layer to treat multiple input features as low-order capsules, and a routing algorithm combines these low-order capsules into high-order capsules that express the intrinsic structural relationships between features. The high-order capsule sequence output by the capsule network at consecutive time steps is fed as input into a long short-term memory (LSM) network layer. The LSM network learns and detects the long-term dependencies of the high-order features in the time dimension through its internal gating mechanism. The final layer of the model is a fully connected layer and a Softmax activation function, which maps the output of the LSM network to a preset gas risk level, such as the probability distribution of three levels: safe, warning, and dangerous, and outputs the level with the highest probability as the prediction result for a future time period. This model requires supervised learning training using a large amount of historical monitoring data and corresponding actual risk event records to optimize the network parameters. However, those skilled in the art should know that other models, such as a hybrid model consisting of 1D-CNN and LSTM, can also be used.

[0032] Based on the predicted future gas risk level, a pre-set, graded response-based adjustment plan for tunneling and support, precisely matched to the risk level, is triggered. For example, when the predicted risk changes from low to medium, it is recommended to moderately reduce the tunneling speed and densify conventional support, such as reducing the spacing between anchor bolts by 10%-20%. When the predicted risk level is high, a significant reduction in tunneling speed or suspension of work will be ordered, and a reinforced support scheme will be initiated, such as using longer, larger diameter anchor bolts with higher preload or reinforced anchor cable support, and using high-strength metal mesh and steel ladder beams to enhance the overall integrity of the system support. At the same time, pre-grouting reinforcement will be implemented in front of the tunneling face, filling and solidifying the potential fracture network generated by micro-seismic activity with grout, thereby improving the self-supporting capacity of the fractured surrounding rock and effectively sealing the gas escape channels. When a critical risk is predicted, the system will issue the highest level alarm, requiring the immediate cessation of all tunneling activities, evacuation of personnel, and implementation of emergency sealing and strong support measures, as well as enhanced ventilation, until multi-source monitoring data shows that the risk has been effectively controlled.

[0033] In an optional embodiment, the tunneling geological parameters include coal seam thickness, coal seam dip angle, and geological structure fraction; the equipment operating parameters include tunneling speed, support strength, and fan air volume.

[0034] Tunneling geological parameters represent static or slowly varying variables of the inherent geological environment of the working face. Coal seam thickness, measured in meters (e.g., 3.2 meters), relates to the size of the mined space and the initial gas reserves. Coal seam dip angle, measured in degrees (e.g., 12 degrees), affects stress distribution, the ease of gas migration, and the operating posture of the tunneling equipment. The geological structure score is a comprehensive indicator, typically ranging from 0 to 100 with a value of 80. It assesses the complexity of fault and fold geological structures; a higher score indicates more complex geological conditions and a greater likelihood of stress concentration and gas enrichment. This data is usually derived from geological exploration reports and real-time updates during tunneling.

[0035] Equipment operating parameters are variables reflecting the intensity of tunneling activities and environmental control measures. Tunneling speed, measured in meters per hour (m / h), e.g., 0.5 m / h, represents the rate of disturbance to the coal and rock mass; a faster speed results in faster stress release and gas desorption. Support strength, measured in megapascals (MPa), e.g., 35 MPa, refers to the supporting force of the hydraulic supports on the roof, affecting the stability of the surrounding rock and the redistribution of the stress field. Fan airflow, measured in cubic meters per minute (m³ / min), e.g., 1800 m³ / min, is an engineering measure to control the gas concentration at the working face, determining the efficiency of gas dilution and discharge. These parameters are collected in real-time by sensors integrated into the equipment, forming a data stream representing the impact of engineering disturbances. Optionally, after obtaining the above parameters, they are normalized to eliminate the influence of dimensions.

[0036] In an optional embodiment, the data processing flow of the hybrid model is as follows: the combined input is fed into a capsule network to extract deep spatial feature vectors, and the feature vectors output by the capsule network are input into a long short-term memory network in chronological order, and the output layer of the long short-term memory network completes the prediction of risk level.

[0037] The first stage of the model is the capsule network. At each time step t, the combined perturbation features, geological parameters, and operating parameters at that time are combined into a multi-dimensional input vector, for example, a vector containing 7 feature values. This vector is then fed into the capsule network. Through capsule units and routing algorithms, the capsule network can learn the complex nonlinear relationships and inherent hierarchical structure between the input features. It not only focuses on the existence of features but also on the pose and variations of the features, thereby generating an information-rich deep feature vector, such as a 16-dimensional vector. The output vector can be viewed as a highly condensed and abstract representation of the state of the entire mine system at the current time.

[0038] The second stage of the model is a Long Short-Term Memory (LSTM) network. This network takes the deep feature vectors output by the capsule network at multiple consecutive time steps (e.g., a sequence of feature vectors from each hour over the past 24 hours), organizes them chronologically, and uses this sequence as input to the LTM network. The LTM network is a special type of recurrent neural network with internal gating mechanisms, including forget gates, input gates, and output gates, enabling it to learn and remember long-term dependencies in the time series. By processing these feature vector sequences, the network detects the evolution and trends of risk states over time. At the last time step, the hidden state of the LTM network is fed into a fully connected layer and a Softmax activation function, outputting a probability distribution vector. Each element of this vector corresponds to a probability of a preset risk level, thus predicting the future gas risk level.

[0039] The hybrid neural network model is structured by concatenating a capsule network and a long short-term memory network. The capsule network consists of a convolutional layer for initial feature extraction, a main capsule layer with 32 eight-dimensional capsules, and a digital capsule layer connected via a routing algorithm, outputting a 16-dimensional deep feature vector. The long short-term memory network comprises two stacked LSTM layers, each containing 128 hidden units, used to process the time-series features output by the capsule network. The model's input is historical time-series data, such as a sequence containing the past 24 time steps, where each time step's input is a vector of seven feature values, including integrated perturbation features, geological parameters, and operating condition parameters. The model's output is the gas risk level for the next time step, specifically an N-dimensional probability vector, for example, N=4, corresponding to four risk levels. The training set consists of historical monitoring data from the mining area, generating thousands of sample pairs using a sliding window method. Each sample pair contains an input sequence and its corresponding true risk level label. The training process employs the Adam optimizer and a classification cross-entropy loss function. The network weights are iteratively updated in mini-batch over approximately 100 training epochs until the model achieves optimal performance on the validation set. Figure 4 .

[0040] In an optional embodiment, the output of the future gas risk level includes: The gas risk level is divided into N levels, and the output of the model is an identifier that corresponds one-to-one with each of the N levels.

[0041] Specifically, based on safety procedures and on-site management needs, N risk levels are defined. For example, setting N=4, the risk is divided into four levels: Level 1 is safe, corresponding to a methane concentration below 0.5%; Level 2 is alert, corresponding to a methane concentration between 0.5% and 1.0%; Level 3 is dangerous, corresponding to a methane concentration between 1.0% and 1.5%; and Level 4 is high-risk, corresponding to a methane concentration above 1.5%. Each level is assigned a unique numerical identifier, such as 0, 1, 2, and 3, corresponding to levels 1 to 4 respectively.

[0042] The output layer of the hybrid prediction model, the Softmax layer, generates a probability vector containing N elements. For example, with N=4, the model might output a vector like [0.05, 0.15, 0.7, 0.1] at a given time. The four elements of this vector represent the probability that the model predicts a future risk level of 1, 2, 3, or 4. The level corresponding to the element with the highest probability value in the vector is selected as the prediction result. In the example above, the third element, 0.7, is the highest, corresponding to level 3 hazard. Therefore, the model's output is a warning signal indicating either level 2 or level 3 hazard. This level classification and output format facilitates staff's quick understanding of the risk situation and allows them to take appropriate contingency measures.

[0043] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0044] The functional modules shown in the above-described block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0045] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0046] The aspects of this application have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by dedicated hardware performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0047] The above description is merely a specific implementation of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope 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 protection scope of this application.

Claims

1. An intelligent tunneling support and gas control system for unstable coal seams in wind-oxidizing zones, characterized in that, Includes the following steps: The acquisition module is used to acquire microseismic monitoring data of the tunneling face and surrounding rock, multi-point gas concentration time series data, tunneling geological parameters, and equipment operating condition parameters; The calculation module is used to select microseismic events with energy higher than a preset threshold as nodes based on the microseismic monitoring data, construct a microseismic event topology map according to the proximity relationship, and calculate the structural entropy of the topology map to obtain the first perturbation feature time series representing the complexity of fracture evolution. Multifractal detrending fluctuation analysis was performed on the multi-point gas concentration time series data to calculate the generalized Hearst exponent and obtain the second perturbation characteristic time series representing the gas emission characteristics. The fusion module is used to calculate the transfer entropy from the first perturbation feature time series to the second perturbation feature time series, and generate weights based on the transfer entropy to perform weighted fusion of the first perturbation feature time series and the second perturbation feature time series to obtain a comprehensive perturbation feature vector. The output module is used to take the integrated disturbance feature vector, the tunneling geological parameters and the equipment operating condition parameters as combined inputs, input a pre-trained hybrid model composed of capsule networks and long short-term memory networks, output the future gas risk level, and adjust the tunneling plan according to the risk level.

2. The system according to claim 1, characterized in that, The calculation of the transfer entropy from the first perturbation feature time series to the second perturbation feature time series includes: The first and second perturbation feature time series are discretized, and the continuous values ​​of each time series are divided into a preset number of discrete state intervals to obtain a discretized state sequence. Based on the preset embedding dimension k and time delay τ, the phase space of the obtained discretized state sequence is reconstructed to obtain the historical state vector; Based on the historical state vector, the required joint probability distribution is calculated by statistically analyzing the frequency of different state combinations in the entire time series. Specifically, this includes the joint probability of the next state of the second perturbation feature time series and its historical state, as well as the joint probability of the next state of the second perturbation feature time series and its own historical state and the historical state of the first perturbation feature time series. The conditional probability distribution derived from the calculated joint probability distribution is used to calculate the transfer entropy value from the first perturbation characteristic time series to the second perturbation characteristic time series.

3. The system according to claim 1, characterized in that, The construction of the microseismic event topology map based on proximity relationships includes: When the spatial distance between any two microseismic event nodes is less than the spatial proximity threshold and the time interval between their occurrences is less than the temporal proximity threshold, a connection edge is established between the two nodes.

4. The system according to claim 1, characterized in that, The step of performing multifractal detrending fluctuation analysis on the multi-point gas concentration time series data and calculating the generalized Hearst exponent includes: The cumulative deviation sequence of gas concentration time series data is divided at different scales s and local trends are eliminated by k-order polynomial fitting. Calculate the q-th order wave function The q-order wave function and the scale s satisfy a power-law relationship. ; The generalized Hearst exponent h(q) is determined by the slope of the linear fit of the relationship in a double logarithmic coordinate system.

5. The system according to claim 2, characterized in that, The step of generating weights based on the transfer entropy, and then weighting and fusing the first perturbation feature time series and the second perturbation feature time series to obtain a comprehensive perturbation feature vector includes: Calculate the transfer entropy from the first perturbation characteristic time series S(t) to the second perturbation characteristic time series H(t); Weights corresponding to S(t) are generated based on the transfer entropy using a normalization function. And let the weights corresponding to H(t) be... ; The comprehensive perturbation eigenvector C(t) is obtained by using a weighted summation formula.

6. The system according to claim 1, characterized in that, The tunneling geological parameters include coal seam thickness, coal seam dip angle, and geological structure fraction; the equipment operating parameters include tunneling speed, support strength, and fan air volume.

7. The system according to claim 1, characterized in that, The data processing flow of the hybrid model is as follows: the combined input is fed into a capsule network to extract deep spatial feature vectors, and the feature vectors output by the capsule network are input into a long short-term memory network in chronological order. The output layer of the long short-term memory network then completes the prediction of the risk level.

8. The system according to claim 1, characterized in that, The output of the future gas risk level includes: The gas risk level is divided into N levels, and the output of the model is an identifier that corresponds one-to-one with each of the N levels.

Images