Coal mine water disaster prevention and control system and method based on artificial intelligence

Abstract

The invention discloses a coal mine water disaster prevention and control system and method based on artificial intelligence. The system comprises a data acquisition module, a data transmission module, a data preprocessing module, a geological modeling module, a water disaster prediction module, a risk assessment module, an artificial intelligence decision module, an equipment control module, an emergency response module and a data management updating module. Sufficient time is won for prevention and treatment work, and the possibility of accidents is effectively reduced; early warning is realized, and the timeliness and accuracy of disaster prevention and control are remarkably improved; sufficient time is won for prevention and treatment work, and the possibility of accidents is effectively reduced; personnel safety evacuation and smooth rescue work are powerfully guaranteed, and disaster loss is reduced to the maximum extent; the efficiency and success rate of prevention and control work are improved, and economic losses and casualties caused by water disasters are reduced; the control effect and the system performance are continuously improved, and the long-term safety production requirement is met.

Description

Technical Field

[0001] The present invention belongs to the technical field of coal mine safety, and particularly relates to a coal mine water disaster prevention and control system and method based on artificial intelligence. Background Art

[0002] Coal mine water disasters have always been one of the key factors threatening the safe production of coal mines. The accidents caused by them not only result in huge economic losses but also seriously endanger the lives of miners. Traditional coal mine water disaster prevention and control mainly rely on manual experience and conventional geological analysis, which have many drawbacks. In terms of monitoring, it is difficult for manual inspections to achieve real-time and accurate monitoring of many key parts underground, resulting in delayed discovery of potential hazards; the data obtained by simple monitoring tools are limited and of low accuracy, and cannot comprehensively reflect the hydrogeological information under complex geological conditions. Geological analysis methods are mostly based on static data and are difficult to adapt to the dynamic changes of geological conditions and water conditions during coal mine mining, making the prevention and control decisions lack scientificity and timeliness.

[0003] With the rise of artificial intelligence technology, its powerful capabilities in data processing, model construction, and intelligent decision-making have brought new hope for coal mine water disaster prevention and control. However, the current related technologies have not been able to fully integrate the advantages of artificial intelligence to build a complete and efficient prevention and control system, and further exploration and innovation are still needed to meet the growing needs of coal mine safety production. Summary of the Invention

[0004] To solve the above technical problems, the present invention provides a coal mine water disaster prevention and control system and method based on artificial intelligence, aiming to solve the deficiencies of traditional coal mine water disaster prevention and control means in the face of complex geological conditions and dynamic water disaster risks, and improve the safety and sustainability of coal mine production.

[0005] To achieve the above object, the present invention adopts the following technical solutions: A coal mine water disaster prevention and control system based on artificial intelligence includes a data acquisition module, a data transmission module, a data preprocessing module, a geological modeling module, a water disaster prediction module, a risk assessment module, an artificial intelligence decision-making module, an equipment control module, an emergency response module, and a data management and update module;

[0006] The data acquisition module transmits the data information collected at the underground coal mine site to the data preprocessing module through the data transmission module. After cleaning, regularizing, and preprocessing the collected data information, the data preprocessing module transmits it to the geological modeling module and the water hazard prediction module respectively. The geological modeling module constructs a coal mine geological model based on the deep learning architecture and transmits the data to the data management and update module. The water hazard prediction module inputs the preprocessed real-time monitoring data into the water hazard risk prediction model based on ensemble learning. The output structure of the water hazard prediction module is transmitted to the risk assessment module. The risk assessment module quantitatively evaluates and grades the water hazard risk. The risk assessment module and the data management and update module simultaneously transmit the data to the artificial intelligence decision-making module. The artificial intelligence decision-making module constructs an intelligent decision-making system based on the reinforcement learning algorithm, formulates effective and scientific prevention and control plans, and the prevention and control plans are implemented through the equipment control module and / or the emergency response module.

[0007] Furthermore, the data acquisition module deploys a variety of intelligent sensors around the aquifers, faults, goafs, and key node parts of the roadway in the underground coal mine. The variety of intelligent sensors include high-precision water level sensors, water pressure sensors, geological strain sensors, water temperature sensors, water chemistry sensors, and microseismic sensors. At the same time, these sensors are manufactured using advanced microelectromechanical technology (MEMS) and nanomaterial processes, with high sensitivity, high precision, and high stability. They can collect multi-dimensional hydrogeological data such as water level changes, water pressure fluctuations, minor deformations of geological structures, water temperature differences, changes in water chemical components, and microseismic signals that may indicate water inrush in real time, and transmit the data to the ground data processing center through a high-speed and reliable data transmission module.

[0008] The data transmission module is responsible for building a safe and efficient data transmission channel between the underground and the ground, adopting a hybrid transmission method that combines wired and wireless transmission. Among them, wired transmission uses mine optical cables to ensure the stability and high speed of data transmission, and mainly undertakes the transmission task of a large amount of real-time monitoring data. Wireless transmission uses low-power and high-bandwidth wireless communication technologies (such as LoRa or 5G private network technology), as a supplement to wired transmission, to flexibly collect and transmit data of some sensors that are difficult to connect by wired means in the complex underground environment, and to ensure redundant transmission of key data in case of emergencies, avoid data loss, and ensure that the ground can receive the data collected by the data acquisition module in a timely and complete manner.

[0009] Furthermore, the data preprocessing module is located in the ground data processing center, where machine learning algorithms are used to clean, regularize, and preprocess the collected data. First, obvious abnormal data points are identified and removed through outlier detection algorithms (such as the 3σ principle based on statistics or the Isolation Forest algorithm) to avoid interference with subsequent analysis. For a small number of missing values, appropriate data interpolation methods (such as linear interpolation or deep learning-based interpolation models) are used to supplement them according to the time series characteristics and spatial correlation of the data to ensure data continuity. Subsequently, normalization or standardization techniques (such as Min-Max normalization or Z-score standardization) are used to unify the dimension and scale of the data, converting the original data into a high-quality dataset suitable for model training and analysis, providing a good data foundation for subsequent artificial intelligence model processing.

[0010] The geological modeling module constructs a coal mine geological model based on a deep learning architecture. Using the powerful spatial feature extraction ability of the Convolutional Neural Network (CNN), detailed spatial feature information such as stratigraphic structure, aquifer distribution, fault strike and throw, and goaf range and shape is extracted from multi-source geological data including geological exploration reports, 3D geological scan data, and borehole data. At the same time, combining the processing advantages of the Recurrent Neural Network (RNN) for time series data, the evolution law of geological structures over time and the dynamic correlation relationships between different geological factors are analyzed. Through the training of a large amount of historical data and real-time monitoring data, the model parameters are continuously optimized, enabling the model to accurately reproduce the complex underground geological environment and update in real time with the input of new data to maintain an accurate grasp of the geological conditions.

[0011] Furthermore, the water hazard prediction module inputs the preprocessed real-time monitoring data into a water hazard risk prediction model based on ensemble learning. The water hazard risk prediction model integrates multiple machine learning algorithms such as decision trees, support vector machines, and random forests, comprehensively considering the influence of multiple factors such as water level, water pressure change trend, geological strain conditions, abnormal water chemical indicators, and microseismic activity frequency and intensity on the occurrence of water hazards. By training the model to learn the complex mapping relationships between different factor combinations and the probability of water hazard occurrence, type (such as roof water hazard, floor water hazard, old goaf water hazard), possible occurrence time, and influence range, accurate prediction of coal mine water hazards is achieved, and detailed risk assessment results are output.

[0012] The risk assessment module further quantifies and grades the water hazard risk based on the output results of the water hazard prediction module, adopts the risk matrix method or the fuzzy comprehensive evaluation method and multiple criteria assessment method, comprehensively considers the possibility of water hazard occurrence, the degree of harm and the disaster resistance of the mine, and divides the water hazard risk into different levels (such as low risk, medium risk, high risk and extremely high risk). According to different risk levels, corresponding response strategies and plans are formulated to provide a basis for subsequent intelligent decision-making. At the same time, the risk assessment results are stored in the database in real time for subsequent query and analysis, providing data support for the long-term management of coal mine water hazard prevention and control.

[0013] Furthermore, the artificial intelligence decision-making module constructs an intelligent decision-making system based on a reinforcement learning algorithm. The intelligent decision-making system uses the water hazard risk assessment results and the real-time working conditions of the mine (such as the current mining progress, the operating status of the drainage equipment, and the ventilation system) as inputs to generate the optimal prevention and control plan from a pre-set library of multiple prevention and control strategies. The prevention and control strategies include but are not limited to the optimized design of drainage projects (such as the reasonable selection and layout of drainage pumps, and the laying plan of drainage pipes), the formulation of grouting water blocking plans (such as the selection of grouting materials, the determination of grouting hole positions and the control of grouting volume), the reasonable retention of waterproof coal pillars, and the strengthening measures of tunnel support. Virtual reality (VR) / augmented reality (AR) technology is used to visualize the generated plans, allowing technical personnel to simulate the implementation process of the plans in an immersive way, intuitively evaluate the feasibility, risks and benefits of the plans, and further optimize and adjust the plans based on the simulation results to ensure the scientific nature and effectiveness of the prevention and control plans.

[0014] The equipment control module is responsible for converting the optimized prevention and control plan generated by the artificial intelligence decision-making module into specific control instructions for the underground prevention and control equipment, and realizing the interconnection between the underground prevention and control equipment (such as drainage pumps, grouting pumps, valves, and ventilators) and the ground monitoring center through the Internet of Things technology. Ground monitoring personnel can remotely operate the start, stop, and operating parameter adjustment functions of the equipment according to the control instructions; at the same time, the equipment control module also has equipment status monitoring and fault diagnosis functions, using sensors to collect the operating parameters of the equipment (such as temperature, vibration, current, and pressure) in real time, and judge whether the equipment is operating normally through data analysis and machine learning models. Once a potential fault is found, an alarm will be issued in time and fault diagnosis information will be provided to assist maintenance personnel to quickly locate and repair the fault, ensuring the reliable operation of the prevention and control equipment.

[0015] Furthermore, when the water hazard risk reaches the preset emergency threshold or a sudden water hazard accident occurs, the emergency response module is immediately activated. The emergency response module automatically triggers the underground emergency refuge system, such as turning on emergency lighting, activating escape indication signs, and controlling the closing of air doors to prevent the spread of harmful gases. At the same time, detailed emergency instructions and refuge information, including the type of water hazard, dangerous areas, escape routes, and emergency assembly points, are sent to underground workers and ground management personnel through multiple communication channels (such as underground broadcasts, personnel positioning terminals, and large-screen displays in the ground command center), ensuring that personnel can evacuate the dangerous area quickly and orderly, minimizing casualties and property losses. In addition, the emergency response module can also be linked with external rescue forces (such as the mine rescue team and the fire department) to timely transmit accident information and on-site situations, assisting in the efficient development of rescue work.

[0016] The data management and update module is responsible for the data management and model update work of the entire system. It establishes a large-scale coal mine hydrogeological database to store multi-source data such as historical geological data, annual monitoring data, water hazard accident cases, records of the implementation of prevention and control plans, and model training results, and classifies, archives, and backs up the data to ensure the security and traceability of the data. Regularly mine and analyze the data in the database to extract valuable information and knowledge to provide data support for the optimization and improvement of the system. At the same time, according to the newly collected monitoring data and feedback information during the prevention and control process, use online learning or incremental learning techniques to update and optimize the relevant artificial intelligence models of the geological modeling module, water hazard prediction module, and artificial intelligence decision-making module, enabling the system to continuously adapt to the dynamic changes of coal mine geological conditions and water hazard risks and continuously improve the prevention and control effect.

[0017] An artificial intelligence-based coal mine water hazard prevention and control method is implemented using the described coal mine water hazard prevention and control system, including the following steps:

[0018] S1: Conduct data collection and transmission. Multi-source data is collected by sensors at key locations in the coal mine underground and transmitted to the ground data processing center safely and stably through the data transmission module.

[0019] S2: Data preprocessing and model construction. Clean, regularize, and standardize the data, and use CNN and RNN to build an accurate geological model.

[0020] S3: Conduct water hazard prediction and risk assessment. Analyze the data by integrating multiple algorithms to accurately predict water hazards and quantitatively evaluate their risk levels.

[0021] S4: Intelligent decision-making and plan implementation. Determine the plan through reinforcement learning based on risks and working conditions, and use VR / AR for auxiliary implementation and optimization.

[0022] S5: Conduct emergency handling and data update, ensure safety through emergency response and link up with rescue operations, and optimize system performance through data update.

[0023] Furthermore, in S1, first, various sensors in the data acquisition module continuously collect hydrogeological data at key underground locations and transmit the data to the data transmission module. Near the aquifer, water level sensors and water temperature sensors are closely arranged to constantly monitor the slight fluctuations in water level and the changes in water temperature. These data are crucial for judging the dynamic changes of the aquifer. Around the fault, geological strain sensors and microseismic sensors can sensitively capture the minor deformations of the geological structure and the microseismic activities that may indicate water inrush, providing key evidence for detecting water disaster risks in advance; around the goaf, water pressure sensors are distributed all around to closely monitor the fluctuations in water pressure. Any abnormal change may be a precursor to a water disaster. At the same time, at key nodes of the roadway, water chemistry sensors continuously detect the chemical composition of the water. Once abnormal changes are found in the mineral content and pH index of the water, it may mean the occurrence of a water disaster;

[0024] The data transmission module constructs a secure and efficient hybrid transmission network combining wired and wireless. The wired transmission part uses mine optical cables as the main trunk. Its high bandwidth and low latency characteristics ensure that a large amount of real-time monitoring data can be stably and quickly transmitted, undertaking the main data transmission task. Wireless transmission uses low-power, high-bandwidth wireless communication technologies (such as LoRa or 5G private network technology) as supplementary means. In some complex areas where it is difficult to lay optical cables, wireless sensor nodes use wireless communication technologies to send the collected data to nearby wireless access points, and then transmit the data to the ground through connection with the wired network. During the data transmission process, the data transmission module adopts advanced data verification and error correction algorithms, through CRC verification, FEC error correction, and ARQ retransmission technologies, to ensure the reliable transmission of data in complex environments; in the future, an AI-driven adaptive transmission strategy can be introduced to dynamically optimize the verification, error correction, and retransmission mechanisms according to the network state, further improving the transmission efficiency and reliability, and providing a solid data foundation for subsequent analysis work;

[0025] The method for sensor data acquisition in S1 is as follows: The sensor data acquisition module includes different types of sensors for monitoring key environmental parameters in the coal mine: water level sensor (Wi), water pressure sensor (Psi), geological strain sensor (Si), water temperature sensor (Twi), water chemistry sensor (Ci), and microseismic sensor (Mi); the reading recorded by each sensor at time t is represented as Xi(t), where Xi represents the corresponding sensor type; the overall state of the sensor data at time t is represented in vector form: Xi(t) = [Wi(t), Psi(t), Si(t), Twi(t), Ci(t), Mi(t)]

[0026] These sensors, through advanced microelectromechanical technology (MEMS) and nanomaterial processes, have extremely high sensitivity and precision, can work stably in complex and harsh underground environments, continuously collect multi-dimensional hydrogeological data, and quickly transmit the data to the data transmission module through a high-speed and reliable underground communication network.

[0027] Further, in S2, the method for preprocessing the collected data is as follows: After receiving the data from the data transmission module, the data preprocessing module quickly starts a series of machine learning algorithms to clean and regularize the data. First, a method combining the 3σ principle based on statistics and the isolation forest algorithm is used to identify and remove significantly abnormal data points.

[0028] Specifically, assuming that the data follows a normal distribution, the mean (μ) and standard deviation (σ) of the water level (Wi), water pressure (Psi), geological strain (Si), water temperature (Twi), water chemistry (Ci), and microseismic (Mi) parameters are calculated respectively, and the data points outside the range of [μ - 3σ, μ + �σ] are marked as abnormal. At the same time, isolation trees are constructed by randomly selecting features and split points, and isolation forest algorithms are used to build models for Wi, Psi, Si, Twi, Ci, and Mi respectively. The anomaly score of each data point is calculated, and the data points with an anomaly score higher than the preset threshold are marked as abnormal. If a certain data point is marked as abnormal by both the 3σ principle and the isolation forest algorithm, it is determined as an abnormal data point. Through the above method, abnormal data points can be efficiently identified and removed to ensure the accuracy and reliability of the data.

[0029] In S2, for a small number of missing values, based on the time series characteristics and spatial correlation of the data, a suitable data interpolation method is used for supplementation. Taking the water level data as an example, if there are missing values within a short period of time and the data of surrounding sensors show that the water level changes relatively smoothly, the linear interpolation method is used for supplementation. The calculation formula of the linear interpolation method is:

[0030]

[0031] where w is the water level value corresponding to the missing point x, and (x1, w1) and (x2, w2) respectively represent the positions of the non-missing value points adjacent to the missing value x and their corresponding water level values. Through this method, the missing values can be effectively filled to ensure the continuity and integrity of the data.

[0032] In S2, for geological structure data, if the missing part has a certain similarity to the geological features of the known area, an interpolation model based on deep learning (such as KNN) is used to fill in the missing part by learning the features of the surrounding geological data. Subsequently, the Min-Max normalization or Z-score standardization method is used to unify the dimension and scale of the data, and the original data is transformed into a high-quality data set suitable for model training and analysis, providing a good data basis for subsequent artificial intelligence model processing. The Min-Max normalization formula is:

[0033]

[0034] where X′ i (t) is the normalized sensor data, min(X i ) is the minimum value of the readings of sensor i at all time points, and max(X i ) is the maximum value of the readings of sensor i at all time points;

[0035] The Z-score standardization formula is:

[0036]

[0037] where X″ i (t) is the normalized reading of sensor X i at time t, μ(X i ) is the mean value of the readings of sensor X i over all recorded times, and σ(X i ) is the standard deviation of the readings of sensor X i over all recorded times;

[0038] Then at time t, the normalized vector X′ i of the overall state vector X(t) of the sensor data is:

[0039] X′ i (t) = [W′ i (t), Ps′ i (t), S′ i (t), Tw′ i (t), C′ i (t), M′ i (t)]

[0040] where each element X′ i (t) is calculated according to the above normalization formula for the readings of a single sensor;

[0041] In S2, based on the preprocessed data and rich historical geological data, the geological modeling module constructs and trains a coal mine geological model using a convolutional neural network (CNN) and a recurrent neural network (RNN).

[0042] Among them, the convolutional neural network mainly solves the problem of image recognition. It integrates data from multi-source geological data such as geological exploration reports, 3D geological scanning data, and borehole data to train the model, enabling it to have the ability to extract spatial features such as stratigraphic structure, aquifer distribution, faults, and the scope and shape of goafs.

[0043] Specifically, in the convolutional neural network architecture for geological data processing, the initial convolutional layer uses a three-dimensional convolutional kernel (3×3×C) to perform spatial-channel joint feature extraction on the original multi-channel geological data, where C represents the physical property parameter dimension of the input data (such as density, magnetic susceptibility, resistivity). Spatial dimensionality reduction of the feature map is achieved through cascaded strided convolution and max pooling layers. At the same time, the ReLU activation function is used to introduce non-linear transformation. The deep network gradually restores the spatial resolution through transposed convolution operations, and the final output layer reconstructs the channel dimension through a 1×1 convolutional kernel, completely retaining the multi-physical field coupling characteristics of the geological body.

[0044] The operation of the initial convolutional layer is expressed as:

[0045] F1 = f(W1 * X + b1)

[0046]

[0047] Among them, F1 represents the feature map after the first layer of convolution, X is the input sensor data X(t), W1 and b1 are the three-dimensional convolutional kernel and bias term respectively, * represents the convolution operation, and f is the ReLU activation function.

[0048] The operations of strided convolution and pooling layer are expressed as:

[0049] F2 = f(W2 * F1 + b2)

[0050] F pool = MaxPool(F2)

[0051] Among them, W2 is the strided convolution kernel, and F [[ID=3??]] pool is the max pooling layer;

[0052] The operation of the deep network (transposed convolution) is expressed as:

[0053] F3 = f(W3 * F pool + b3)

[0054] Among them, W3 is the transposed convolution kernel; It should be noted that there seems to be a small error in the numbering in the original text (around line 32 where it says "F " and then jumps to "F " again with a different numbering). This has been maintained as accurately as possible in the translation. Also, the numbering in the original text might need to be double-checked for consistency in the source document.

[0055] The output layer operation is expressed as:

[0056] F CNN = f(W4 * F3 + b4)

[0057] where W4 is a 1×1 convolutional kernel of the output layer, and F CNN is the coupled multi-physical field feature vector of the final output;

[0058] Specifically, in the recurrent neural network architecture for multi-source hydrogeological time series analysis, the initial gated recurrent unit (GRU) performs spatio-temporal-physical joint feature modeling on the original time series data through a two-dimensional state gating mechanism (reset gate rt, update gate zt). The input feature dimensions include dynamic monitoring parameters such as water level (m), water pressure (MPa), and formation displacement rate (mm / d), as well as geological state parameters such as rock creep coefficient and fracture development index. The invention realizes multi-time scale feature extraction through cascaded causal dilated convolution and hierarchical state transfer, and at the same time introduces parameterized non-linear interaction using a gated linear unit (GLU); the deep network dynamically focuses on key time nodes through a temporal attention mechanism, and the final output layer realizes the reconstruction of the coupled multi-physical field response through a differentiable physical constraint layer, strictly following the principles of mass conservation and energy balance;

[0059] The time series data input to the recurrent neural network part is X(t - k), X(t - k + 1), …, X(t), which is expressed as:

[0060] h t = f(W h X t + U h h t-1 + b h )

[0061] where ht is the hidden state at the current moment, W h , U h are weight matrices, b h is the bias term, f is the ReLU activation function, and through the recursive operation of time steps, the RNN generates the time series feature vector F RNN ;

[0062] To further improve the performance of the model, the water hazard prediction module adopts Cross-Validation and Grid Search techniques to optimize the selection of hyperparameters and avoid overfitting problems. In addition, the model also introduces the Stacking method in ensemble learning, taking the prediction results of multiple base learners (such as decision trees, support vector machines, random forests) as the input of the meta-learner (such as logistic regression or gradient boosting tree), and performing secondary learning through the meta-learner, so as to integrate the advantages of each base learner and further improve the accuracy and robustness of the prediction.

[0063] Finally, the water hazard prediction module can output detailed risk assessment results, including the probability, type, time range of water hazard occurrence and key information of the affected area. For example, when the model detects an abnormal combination of water level, water pressure and geological strain data, it can generate a risk assessment report, indicating the possible floor water hazard and its potential impact range, providing a scientific basis for coal mine safety production and water hazard prevention.

[0064] Furthermore, in step S3, the water hazard prediction module inputs the preprocessed real-time monitoring data into the water hazard risk prediction model based on ensemble learning. This model integrates multiple machine learning algorithms such as Decision Tree (DT), Support Vector Machine (SVM), and Random Forest (RF), giving full play to the advantages of each algorithm to achieve accurate prediction of coal mine water hazard risks.

[0065] Specifically, the decision tree algorithm constructs a tree structure and classifies and judges different hydrogeological factors based on indicators such as Information Gain or Gini Index, and can quickly identify the key factor combinations that may lead to water hazards. The support vector machine maps the data to a high-dimensional space through a kernel function (such as the radial basis function RBF) to handle non-linear classification problems, and can find the optimal classification boundary between the occurrence and non-occurrence of water hazards in a complex data space. The random forest algorithm integrates multiple decision trees, uses the Bagging (Bootstrap Aggregating) method to sample and vote on the data, improves the stability and accuracy of the prediction, and identifies key influencing factors through Feature Importance analysis.

[0066] During the model training process, a large number of historical water disaster cases and corresponding hydrogeological data are used to enable the model to learn the complex mapping relationships between different combinations of factors and the probability of water disasters, types (such as roof water disasters, floor water disasters, goaf water disasters), possible occurrence times, and affected areas; for example, when the water level rises sharply within a short period and the water pressure exceeds a certain threshold, and at the same time the geological strain data shows obvious deformation near the fault, the model can predict a relatively high probability of a floor water disaster occurring and roughly estimate the occurrence time and the possible affected area, so as to achieve accurate prediction of coal mine water disasters and output detailed risk assessment results;

[0067] To further improve the performance of the model, the water disaster prediction module adopts cross-validation and grid search techniques to optimize the selection of hyperparameters and avoid overfitting problems; in addition, the model also introduces the Stacking method in ensemble learning, taking the prediction results of multiple base learners (such as decision trees, support vector machines, random forests) as the input of the meta-learner (such as logistic regression or gradient boosting tree), and performing secondary learning through the meta-learner, so as to integrate the advantages of each base learner and further improve the accuracy and robustness of the prediction;

[0068] Based on the output results of the water disaster prediction module, the risk assessment module further uses a multi-criteria evaluation method combining the risk matrix method (RiskMatrix) and the fuzzy comprehensive evaluation method (Fuzzy Comprehensive Evaluation) to quantitatively evaluate and classify the water disaster risks. The risk matrix method constructs a matrix based on the two dimensions of the possibility and harm degree of water disasters, and divides the risks into different grade intervals. For example, the possibility of occurrence is divided into three levels: low, medium, and high, and the harm degree is divided into three levels: minor, severe, and catastrophic, forming nine risk levels through combination; the fuzzy comprehensive evaluation method takes into account the fuzzy and uncertain factors in water disaster risk assessment and uses fuzzy mathematics methods to comprehensively process multiple evaluation indicators; it takes multiple factors such as the trend of water level change, water pressure fluctuation, degree of geological strain, abnormality degree of water chemical indicators, and frequency and intensity of microseismic activities as evaluation indicators, and calculates the comprehensive risk value by determining the weights and membership functions of each indicator; combining the results of the two methods, the water disaster risks are finally divided into different levels of low risk, medium risk, high risk, and extremely high risk; for different risk levels, corresponding response strategies and plans are formulated to provide a scientific basis for subsequent intelligent decision-making; at the same time, the risk assessment results are stored in the database in real time for subsequent query and analysis, providing data support for the long-term management of coal mine water disaster prevention and control;

[0069] In S4, the artificial intelligence decision-making module constructs an intelligent decision-making system based on the reinforcement learning algorithm; the intelligent decision-making system takes the water hazard risk assessment results and the real-time working conditions of the mine (such as the current mining progress, the operating status of drainage equipment, and the ventilation system situation) as inputs, and generates the optimal prevention and control plan from a variety of pre-set prevention and control strategy libraries; in the framework of reinforcement learning, the intelligent decision-making system regards the water hazard prevention and control process as a dynamic decision-making process, continuously interacts with the environment (i.e., the actual situation of the mine), learns the effects of different decision-making strategies in different situations, and optimizes the decision-making strategies according to the reward mechanism (such as the quality of prevention and control effects and the level of costs); for example, when the risk assessment is high-risk roof water hazard and the mine drainage equipment is operating at full load, the system will preferentially select the prevention and control strategy of grouting to block water, and use the optimization algorithm to determine the best layout of grouting holes and the reasonable control range of grouting volume according to the specific geological conditions of the roof and the water hazard situation; at the same time, use virtual reality (VR) / augmented reality (AR) technology to visually present the generated plan, enabling technicians to simulate the plan implementation process as if they were on the spot; in the VR / AR environment, technicians can intuitively see the positions of grouting holes, the laying paths of drainage pipes, and the operating conditions of prevention and control equipment, can evaluate the feasibility, risks, and benefits of the plan in real time, and further optimize and adjust the plan according to the simulation results to ensure the scientificity and effectiveness of the prevention and control plan;

[0070] The equipment control module is responsible for converting the optimized prevention and control plan generated by the artificial intelligence decision-making module into specific control instructions for underground prevention and control equipment, and realizing the interconnection and interoperability between underground prevention and control equipment (such as drainage pumps, grouting pumps, valves, and ventilators) and the ground monitoring center through Internet of Things technology. After receiving the control instructions, the equipment control module first parses and verifies the instructions to ensure the accuracy and security of the instructions; then, through the underground automation control system, remotely operates the start, stop, and operation parameter adjustment functions of the equipment according to the instruction requirements; for example, for the drainage pump, accurately control the rotation speed and drainage volume of the drainage pump according to the water hazard risk and the mine water inflow to achieve efficient drainage; at the same time, the equipment control module uses sensors to collect the operation parameters of the equipment in real time (such as temperature, vibration, current, pressure), and judges whether the equipment is operating normally through data analysis and machine learning models; adopt the fault diagnosis method based on principal component analysis (PCA), reduce the dimensionality of multiple operation parameters, and compare them with the characteristic model in the normal operation state. Once it is found that the equipment operation parameters deviate from the normal range, immediately issue an alarm and provide fault diagnosis information to assist maintenance personnel in quickly locating and repairing the fault to ensure the reliable operation of the prevention and control equipment;

[0071] In S5, when the water hazard risk reaches the preset emergency threshold or a sudden water hazard accident occurs, the emergency response module is immediately activated. The emergency response module is closely connected to the underground emergency shelter system and automatically triggers the underground emergency shelter function;

[0072] Specifically, quickly turn on the emergency lighting to ensure that underground workers can see the escape route clearly in the dark environment; activate the escape indicator signs, and through flashing and arrow indications, guide personnel to evacuate to the safe area; control the air door to close to prevent harmful gases from spreading to the personnel escape route and ensure the safety of personnel during the evacuation process; at the same time, send detailed emergency instructions and shelter information to underground workers and ground management personnel through various communication channels (such as underground broadcasts, personnel positioning terminals, and large-screen displays in the ground command center), including the type of water hazard, dangerous areas, escape routes, and emergency assembly points; the underground broadcast is looped in a high-volume and multi-band manner to ensure that personnel can clearly hear it in the noisy underground environment; the personnel positioning terminal not only receives emergency information but also feeds back the real-time position information of personnel to the ground command center. Ground commanders can thus master the progress of personnel evacuation and conduct timely dispatching and command for possible congestion or detention situations, such as adjusting the air volume and direction of the ventilation system to guide personnel to avoid dangerous areas and ensure that personnel can quickly and orderly evacuate from the dangerous area, minimizing casualties and property losses to the greatest extent; in addition, the emergency response module can also be linked with external rescue forces (such as mine rescue teams and fire departments), and transmit accident information and on-site conditions in a timely manner through a dedicated communication link to assist in the efficient development of rescue work. For example, provide the mine rescue team with detailed maps of underground roadways, the location of water hazards, and information on trapped personnel, providing key support for rescue operations;

[0073] During the operation of the entire system, the data management and update module is responsible for establishing and maintaining a large-scale coal mine hydrogeological database, storing multi-source data such as historical geological data, annual monitoring data, water disaster accident cases, implementation records of prevention and control plans, and model training results. It classifies, archives, and backs up the data to ensure data security and traceability, and adopts distributed storage technology and data encryption algorithms to prevent data loss and leakage. Regularly, it mines and analyzes the data in the database, and uses data mining algorithms (such as association rule mining, clustering analysis) to extract valuable information and knowledge, such as discovering potential laws of water disasters under different geological conditions and the effectiveness of a certain prevention and control strategy in specific situations, providing data support for system optimization and improvement. At the same time, according to newly collected monitoring data and feedback information during the prevention and control process, it uses online learning or incremental learning technology to update and optimize the relevant artificial intelligence models of the geological modeling module, water disaster prediction module, and artificial intelligence decision-making module; for example, when new data shows that the geological structure in a certain area has changed, the geological modeling module will automatically adjust the model parameters and re-learn and adapt to the new geological situation. By continuously updating and optimizing the model, the system can continuously adapt to the dynamic changes of coal mine geological conditions and water disaster risks, and continuously improve the prevention and control effect;

[0074] After the system has been running for some time, the data management and update module will conduct a comprehensive analysis of the collected data. For example, by comparing hydrogeological data from different time periods and different mining areas, it is found that during the mining process of a certain specific coal seam, when the water pressure of the surrounding aquifer rises at a rate exceeding a certain threshold within a short period of time and is accompanied by a significant increase in the content of specific mineral components in the water, the probability of roof water disaster occurrence increases significantly; based on this analysis result, the algorithm of the water disaster prediction module is adjusted to increase the weight of this specific index combination, thereby improving the prediction accuracy.

[0075] Adopting the above technical solution, compared with the prior art, it has the following technical effects:

[0076] 1) By deploying high-precision and multi-type intelligent sensors at key underground positions and combining advanced data processing and analysis technologies, it can collect hydrogeological data in real time and accurately, achieve precise prediction of coal mine water disasters, detect water disaster risks in advance, gain sufficient time for prevention and control work, and effectively reduce the possibility of accidents.

[0077] 2) By using a deep learning architecture to build a coal mine geological model, including convolutional neural network (CNN) and recurrent neural network (RNN), analyzing and modeling the collected data, it can accurately predict potential disaster risks, achieve early warning, and significantly improve the timeliness and accuracy of disaster prevention and control.

[0078] 3) An intelligent decision-making system based on reinforcement learning and a variety of prevention and control strategy libraries generate the optimal prevention and control plan by combining with the real-time working conditions of the mine, and use VR / AR technology for visual presentation and optimization, ensuring the scientificity and effectiveness of the prevention and control decision-making, improving the efficiency and success rate of the prevention and control work, and reducing the economic losses and casualties caused by water disasters.

[0079] 4) The equipment control module realizes the remote precise control and real-time status monitoring of the underground prevention and control equipment, and uses the fault diagnosis function to timely discover and solve the potential equipment faults, ensuring the stable and reliable operation of the prevention and control equipment during the water disaster prevention and control process, and maintaining the continuity and effectiveness of the prevention and control work.

[0080] 5) The emergency response module can be quickly activated in case of emergency, automatically trigger the underground emergency shelter system, send accurate emergency information to personnel through multiple channels, and efficiently link with external rescue forces, effectively ensuring the safe evacuation of personnel and the smooth progress of the rescue work, and minimizing the disaster losses to the greatest extent.

[0081] 6) The data management and update module continuously collects and analyzes data, and uses online learning or incremental learning technology to update the artificial intelligence model, enabling the system to dynamically adapt to the changes in the coal mine geological conditions and water disaster risks, continuously improving the prevention and control effect and system performance, and meeting the long-term safe production requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 It is the connection block diagram of the coal mine water disaster prevention and control system based on artificial intelligence in the present invention;

[0083] Figure 2 It is the schematic diagram of the method steps of the coal mine water disaster prevention and control method based on artificial intelligence in the present invention;

[0084] Figure 3 It is the design flow chart of the coal mine water disaster data transmission in the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0085] The present invention will be described in detail below in conjunction with the drawings and specific embodiments. At the same time, it should be noted here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments. For some well-known technologies, those skilled in the art can also adopt other alternative methods for implementation; and the drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0086] As Figure 1 shown, a coal mine water disaster prevention and control system based on artificial intelligence of the present invention includes a data acquisition module, a data transmission module, a data preprocessing module, a geological modeling module, a water disaster prediction module, a risk assessment module, an artificial intelligence decision-making module, an equipment control module, an emergency response module and a data management and update module;

[0087] The data acquisition module transmits the data information collected from the underground coal mine site to the data preprocessing module through the data transmission module. After the data preprocessing module cleans, regularizes, and preprocesses the collected data information, it transmits the data to the geological modeling module and the water hazard prediction module respectively. The geological modeling module constructs a coal mine geological model based on the deep learning architecture and transmits the data to the data management and update module. The water hazard prediction module inputs the preprocessed real-time monitoring data into the water hazard risk prediction model based on ensemble learning. The output structure of the water hazard prediction module is transmitted to the risk assessment module. The risk assessment module quantitatively evaluates and classifies the water hazard risk. The risk assessment module and the data management and update module simultaneously transmit the data to the artificial intelligence decision-making module. The artificial intelligence decision-making module constructs an intelligent decision-making system based on the reinforcement learning algorithm, formulates effective and scientific prevention and control plans, and the prevention and control plans are implemented through the equipment control module and / or the emergency response module.

[0088] Furthermore, the data acquisition module deploys a variety of intelligent sensors around the aquifers, faults, goafs, and key nodes of the roadway in the underground coal mine. The variety of intelligent sensors include high-precision water level sensors, water pressure sensors, geological strain sensors, water temperature sensors, water chemistry sensors, and microseismic sensors. At the same time, these sensors are manufactured using advanced microelectromechanical technology (MEMS) and nanomaterial processes, and have high sensitivity, high precision, and high stability. They can collect multi-dimensional hydrogeological data such as water level changes, water pressure fluctuations, minor deformations of geological structures, water temperature differences, changes in water chemical components, and microseismic signals that may indicate water inrush in real time, and transmit the data to the ground data processing center through a high-speed and reliable data transmission module.

[0089] The data transmission module is responsible for building a safe and efficient data transmission channel between the underground and the ground, and adopts a hybrid transmission method combining wired and wireless. Among them, wired transmission uses mine optical cables to ensure the stability and high speed of data transmission, and mainly undertakes the transmission tasks of a large amount of real-time monitoring data; wireless transmission uses low-power and high-bandwidth wireless communication technologies (such as LoRa or 5G private network technology) as a supplement to wired transmission, and is used to flexibly collect and transmit sensor data that is difficult to connect by wired means in the complex underground environment, and to ensure redundant transmission of key data in case of emergencies, avoid data loss, and ensure that the ground can receive the data collected by the data acquisition module in a timely and complete manner.

[0090] Furthermore, the data preprocessing module is located in the ground data processing center, where machine learning algorithms are used to clean, regularize, and preprocess the collected data. First, obvious abnormal data points are identified and removed through outlier detection algorithms (such as the 3σ principle based on statistics or the isolation forest algorithm) to avoid interference with subsequent analysis. For a small number of missing values, appropriate data interpolation methods (such as linear interpolation or deep learning-based interpolation models) are used to supplement them according to the time series characteristics and spatial correlation of the data to ensure data continuity. Subsequently, normalization or standardization techniques (such as Min-Max normalization or Z-score standardization) are used to unify the dimension and scale of the data, converting the original data into a high-quality dataset suitable for model training and analysis, providing a good data foundation for subsequent artificial intelligence model processing.

[0091] The geological modeling module constructs a coal mine geological model based on a deep learning architecture. Using the powerful spatial feature extraction ability of convolutional neural networks (CNNs), detailed spatial feature information such as stratigraphic structure, aquifer distribution, fault strike and throw, and goaf range and shape is extracted from multi-source geological data including geological exploration reports, 3D geological scan data, and borehole data. At the same time, combining the processing advantages of recurrent neural networks (RNNs) for time series data, the evolution law of geological structures over time and the dynamic correlation between different geological factors are analyzed. Through the training of a large amount of historical data and real-time monitoring data, the model parameters are continuously optimized, enabling the model to accurately reproduce the complex underground geological environment and update in real time with the input of new data to maintain an accurate grasp of the geological conditions.

[0092] Furthermore, the water hazard prediction module inputs the preprocessed real-time monitoring data into a water hazard risk prediction model based on ensemble learning. The water hazard risk prediction model integrates multiple machine learning algorithms such as decision trees, support vector machines, and random forests, comprehensively considering the impacts of multiple factors such as water level, water pressure change trend, geological strain conditions, abnormal water chemical indicators, and microseismic activity frequency and intensity on the occurrence of water hazards. By training the model to learn the complex mapping relationship between different factor combinations and the probability of water hazard occurrence, type (such as roof water hazard, floor water hazard, old goaf water hazard), possible occurrence time, and influence range, accurate prediction of coal mine water hazards is achieved, and detailed risk assessment results are output.

[0093] The risk assessment module further quantitatively evaluates and classifies the water hazard risk based on the output results of the water hazard prediction module. It adopts multi-criteria evaluation methods such as the risk matrix method or the fuzzy comprehensive evaluation method, comprehensively considering the possibility of water hazard occurrence, the degree of harm, and the disaster resistance ability factors of the mine. The water hazard risk is divided into different levels (such as low risk, medium risk, high risk, and extremely high risk). For different risk levels, corresponding coping strategies and plans are formulated to provide a basis for subsequent intelligent decision-making. At the same time, the risk assessment results are stored in the database in real-time for subsequent query and analysis, providing data support for the long-term management of coal mine water hazard prevention and control.

[0094] Furthermore, the artificial intelligence decision-making module constructs an intelligent decision-making system based on the reinforcement learning algorithm. The intelligent decision-making system takes the water hazard risk assessment results and the real-time working conditions of the mine (such as the current mining progress, the operating status of drainage equipment, and the ventilation system) as inputs, and generates the optimal prevention and control plan from a variety of pre-set prevention and control strategy libraries. The prevention and control strategies include, but are not limited to, the optimization design of drainage and dewatering projects (such as the reasonable selection and layout of drainage pumps, the laying plan of drainage pipelines), the formulation of grouting water-blocking plans (such as the selection of grouting materials, the determination of grouting hole positions, and the control of grouting volumes), the reasonable setting of waterproof coal pillars, and the strengthening measures for roadway support. The generated plan is visually presented using virtual reality (VR) / augmented reality (AR) technology, enabling technicians to simulate the implementation process of the plan as if they were on the scene, intuitively evaluating the feasibility, risks, and benefits of the plan, and further optimizing and adjusting the plan according to the simulation results to ensure the scientificity and effectiveness of the prevention and control plan.

[0095] The equipment control module is responsible for converting the optimized prevention and control plan generated by the artificial intelligence decision-making module into specific control instructions for underground prevention and control equipment. Through the Internet of Things technology, the underground prevention and control equipment (such as drainage pumps, grouting pumps, valves, and ventilators) is interconnected with the ground monitoring center. Ground monitoring personnel can remotely operate the start, stop, and operation parameter adjustment functions of the equipment according to the control instructions. At the same time, the equipment control module also has the functions of equipment status monitoring and fault diagnosis. It uses sensors to collect the operation parameters of the equipment in real-time (such as temperature, vibration, current, and pressure), and judges whether the equipment is operating normally through data analysis and machine learning models. Once a fault hidden danger is found, an alarm is issued in a timely manner and fault diagnosis information is provided to assist maintenance personnel in quickly locating and repairing the fault, ensuring the reliable operation of the prevention and control equipment.

[0096] Further, when the water hazard risk reaches the preset emergency threshold or a sudden water hazard accident occurs, the emergency response module is immediately activated. The emergency response module automatically triggers the underground emergency refuge system, such as turning on emergency lighting, activating escape indication signs, and controlling the closing of air doors to prevent the spread of harmful gases. At the same time, detailed emergency instructions and refuge information, including the type of water hazard, dangerous areas, escape routes, and emergency assembly points, are sent to underground workers and ground management personnel through multiple communication channels (such as underground broadcasts, personnel positioning terminals, and large-screen displays in the ground command center), ensuring that personnel can quickly and orderly evacuate from the dangerous area, minimizing casualties and property losses to the greatest extent. In addition, the emergency response module can also be linked with external rescue forces (such as mine rescue teams and fire departments), timely transmit accident information and on-site conditions, and assist in the efficient development of rescue work.

[0097] The data management and update module is responsible for the data management and model update work of the entire system. It establishes a large-scale coal mine hydrogeological database to store multi-source data such as historical geological data, annual monitoring data, water hazard accident cases, implementation records of prevention and control plans, and model training results, and classifies, archives, and backs up the data to ensure the security and traceability of the data. Regularly mine and analyze the data in the database, extract valuable information and knowledge, and provide data support for the optimization and improvement of the system. At the same time, according to the newly collected monitoring data and feedback information during the prevention and control process, use online learning or incremental learning techniques to update and optimize the relevant artificial intelligence models of the geological modeling module, water hazard prediction module, and artificial intelligence decision-making module, enabling the system to continuously adapt to the dynamic changes of coal mine geological conditions and water hazard risks, and continuously improve the prevention and control effect.

[0098] As Figure 2 shown, the present invention also provides an artificial intelligence-based coal mine water hazard prevention and control method, which is implemented using the described coal mine water hazard prevention and control system, and includes the following steps:

[0099] S1: Conduct data collection and transmission. Multisource data is collected by sensors at key locations in the coal mine underground and safely and stably transmitted to the ground data processing center through the data transmission module.

[0100] S2: Data preprocessing and model construction. Clean, regularize, and standardize the data, and use CNN and RNN to construct an accurate geological model.

[0101] S3: Conduct water hazard prediction and risk assessment. Analyze the data by integrating multiple algorithms, accurately predict water hazards, and quantitatively evaluate their risk levels.

[0102] S4: Intelligent decision-making and plan implementation. Determine the plan based on risk and working conditions through reinforcement learning, and VR / AR is used for assistance in implementation and optimization.

[0103] S5: Conduct emergency handling and data update, ensure safety through emergency response and link up for rescue, and update data to optimize system performance.

[0104] Furthermore, in S1, first, various sensors in the data acquisition module continuously collect hydrogeological data at key underground locations and transmit the data to the data transmission module. Near the aquifer, the water level sensor and water temperature sensor are closely arranged to constantly monitor the slight fluctuations of the water level and the changes in water temperature. These data are crucial for judging the dynamic changes of the aquifer. Around the fault, the geological strain sensor and microseismic sensor can sensitively capture the minor deformations of the geological structure and the microseismic activities that may indicate water inrush, providing a key basis for detecting water hazard risks in advance; around the goaf, the water pressure sensors are distributed all around to closely monitor the fluctuations of the water pressure. Any abnormal change may be a precursor to water hazards. At the same time, at the key nodes of the roadway, the water chemistry sensor detects the chemical composition of the water in real time. Once abnormal changes are found in the mineral content and pH index of the water, it may mean the occurrence of water hazards;

[0105] As Figure 3 shown, it is a design flow chart of coal mine water hazard data transmission based on artificial intelligence. Among them, the data transmission module constructs a hybrid transmission network combining safe and efficient wired and wireless networks. The wired transmission part takes the mine optical cable as the main trunk road. Its characteristics of high bandwidth and low latency ensure that a large amount of real-time monitoring data can be transmitted stably and quickly, undertaking the main data transmission task. The wireless transmission uses low-power and high-bandwidth wireless communication technologies (such as LoRa or 5G private network technology) as supplementary means. In some complex areas where it is difficult to lay optical cables, the wireless sensor nodes use wireless communication technology to send the collected data to the nearby wireless access points, and then transmit the data to the ground through the connection with the wired network. During the data transmission process, the data transmission module adopts advanced data verification and error correction algorithms. Through CRC verification, FEC error correction, and ARQ retransmission technologies, it ensures the reliable transmission of data in complex environments; in the future, an AI-driven adaptive transmission strategy can be introduced to dynamically optimize the verification, error correction, and retransmission mechanisms according to the network status, further improving the transmission efficiency and reliability, and providing a solid data foundation for subsequent analysis work;

[0106] The sensor data acquisition method in S1 is as follows: the sensor data acquisition module includes different types of sensors for monitoring key environmental parameters in the coal mine: water level sensor (Wi), water pressure sensor (Psi), geological strain sensor (Si), water temperature sensor (Twi), water chemistry sensor (Ci), and microseismic sensor (Mi). The reading recorded by each sensor at time t is represented as Xi(t), where Xi represents the corresponding sensor type. The overall state of the sensor data at time t is represented as a vector: Xi(t) = [Wi(t), Psi(t), Si(t), Twi(t), Ci(t), Mi(t)].

[0107] These sensors, through advanced micro-electromechanical technology (MEMS) and nanomaterial processing, have extremely high sensitivity and precision. They can work stably in complex and harsh underground environments, continuously collect multi-dimensional hydrogeological data, and quickly transmit the data to the data transmission module through a high-speed and reliable underground communication network.

[0108] Furthermore, in S2, the method for preprocessing the collected data is as follows: after receiving the data from the data transmission module, the data preprocessing module quickly starts a series of machine learning algorithms to clean and regularize the data. First, a method based on the statistical 3σ principle combined with the isolation forest algorithm is used to identify and eliminate significantly abnormal data points;

[0109] Specifically, assuming that the data obeys a normal distribution, the mean (μ) and standard deviation (σ) of the water level (Wi), water pressure (Psi), geological strain (Si), water temperature (Twi), hydrochemistry (Ci) and microseismic (Mi) parameters are calculated respectively, and data points outside the range of [μ-3σ, μ+3σ] are marked as abnormal. At the same time, an isolation tree is constructed by randomly selecting features and split points, and the isolation forest algorithm is used to construct models for Wi, Psi, Si, Twi, Ci, and Mi respectively. The anomaly score (anomaly score) of each data point is calculated, and the data points with an anomaly score higher than the preset threshold are marked as abnormal. If a data point is marked as abnormal by both the 3σ principle and the isolation forest algorithm, it is determined to be an abnormal data point. Through the above method, abnormal data points can be efficiently identified and eliminated to ensure the accuracy and reliability of the data.

[0110] In S2, for a small number of missing values, appropriate data interpolation methods are used to supplement them based on the time series characteristics and spatial correlation of the data. Taking water level data as an example, if there are missing values in a short period of time and the surrounding sensor data show that the water level changes are relatively stable, linear interpolation is used to supplement them. The calculation formula of the linear interpolation method is:

[0111]

[0112] Among them, w is the water level value corresponding to the missing point x, and (x1, w1) and (x2, w2) respectively represent the positions of the non-missing value points adjacent to the missing value x before and after and their corresponding water level values. Through this method, the invention can effectively fill in the missing values and ensure the continuity and integrity of the data.

[0113] In step S2, for geological structure data, if the missing part has a certain similarity to the geological characteristics of the known area, an interpolation model based on deep learning (such as KNN) is used to fill in the missing part by learning the characteristics of the surrounding geological data. Subsequently, the Min-Max normalization or Z-score standardization method is used to unify the dimension and scale of the data, and the original data is transformed into a high-quality data set suitable for model training and analysis, providing a good data basis for subsequent artificial intelligence model processing. The Min-Max normalization formula is:

[0114]

[0115] Among them, X′ i (t) is the normalized sensor data, min(X i ) is the minimum value of the readings of sensor i at all time points, and max(X i ) is the maximum value of the readings of sensor i at all time points;

[0116] The Z-score standardization formula is:

[0117]

[0118] Among them, X″ i (t) is the normalized reading of sensor X i at time t, μ(X i ) is the mean value of the readings of sensor X i over all recorded times, and σ(X i ) is the standard deviation of the readings of sensor X i over all recorded times;

[0119] Then at time t, the normalized vector X′ i of the overall state vector X(t) of the sensor data is:

[0120] X′ i (t) = [W′ i (t), Ps′ i (t), S′ i (t), Tw′ i (t), C′ i (t), M′ i (t)]

[0121] where each element X′ i (t) is calculated according to the normalization formula of the above single sensor reading;

[0122] In S2, based on the preprocessed data and rich historical geological data, the geological modeling module constructs and trains a coal mine geological model using a convolutional neural network (CNN) and a recurrent neural network (RNN);

[0123] Among them, the convolutional neural network mainly solves the problem of image recognition. It integrates data from multi-source geological data such as geological exploration reports, 3D geological scanning data, and borehole data to train the model, enabling it to have the ability to extract spatial features such as stratigraphic structure, aquifer distribution, faults, and the range and shape of goaf areas;

[0124] Specifically, in the convolutional neural network architecture for geological data processing, the initial convolutional layer performs spatial-channel joint feature extraction on the original multi-channel geological data through a three-dimensional convolutional kernel (3×3×C), where C represents the physical property parameter dimension of the input data (such as density, magnetic susceptibility, resistivity). Spatial dimensionality reduction of the feature map is achieved through cascaded strided convolution and max pooling layers. At the same time, the ReLU activation function is used to introduce non-linear transformation. The deep network gradually restores the spatial resolution through transposed convolution operations, and the final output layer reconstructs the channel dimension through a 1×1 convolutional kernel, completely retaining the multi-physical field coupling characteristics of the geological body;

[0125] The operation of the initial convolutional layer is expressed as:

[0126] F1 = f(W1 * X + b1)

[0127]

[0128] where F1 represents the feature map after the first layer of convolution, X is the input sensor data X(t), W1 and b1 are the three-dimensional convolutional kernel and the bias term respectively, * represents the convolution operation, and f is the ReLU activation function;

[0129] The strided convolution and pooling layer operations are expressed as:

[0130] F2 = f(W2 * F1 + b2)

[0131] F pool = MaxPool(F2)

[0132] where W2 is the strided convolution kernel, and F pool is the max pooling layer;

[0133] The operation of the deep network (transposed convolution) is expressed as:

[0134] F3 = f(W3 * F pool + b3)

[0135] where, W3 is the transposed convolutional kernel;

[0136] The output layer operation is expressed as:

[0137] F CNN = f(W4 * F3 + b4)

[0138] where, W4 is the 1×1 convolutional kernel of the output layer, and F CNN is the coupled multi-physical field feature vector of the final output;

[0139] Specifically, in the recurrent neural network architecture for multi-source hydrogeological time series analysis, the initial gated recurrent unit (GRU) performs spatio-temporal-physical joint feature modeling on the original time series data through a two-dimensional state gating mechanism (reset gate rt, update gate zt). The input feature dimensions include dynamic monitoring parameters such as water level (m), water pressure (MPa), and formation displacement rate (mm / d), as well as geological state parameters such as rock creep coefficient and fracture development index. The invention realizes multi-time scale feature extraction through cascaded causal dilated convolution and hierarchical state transfer, and at the same time introduces parameterized non-linear interaction by using gated linear unit (GLU); the deep network dynamically focuses on key time nodes through the temporal attention mechanism, and the final output layer realizes the reconstruction of the coupled multi-physical field response through a differentiable physical constraint layer, strictly following the principles of mass conservation and energy balance;

[0140] The time series data input to the recurrent neural network part is X(t - k), X(t - k + 1), …, X(t), which is expressed as:

[0141] h t = f(W h X t + U h h t-1 + b h )

[0142] where, ht is the hidden state at the current moment, W h , U h are weight matrices, b h is the bias term, f is the ReLU activation function, and through the recursive operation of time steps, the RNN generates the time series feature vector F RNN ;

[0143] To further improve the performance of the model, the water hazard prediction module adopts cross-validation and grid search techniques to optimize the selection of hyperparameters and avoid overfitting problems. In addition, the model also introduces the Stacking method in ensemble learning, taking the prediction results of multiple base learners (such as decision trees, support vector machines, and random forests) as the input of the meta-learner (such as logistic regression or gradient boosting tree), and performing secondary learning through the meta-learner, so as to integrate the advantages of each base learner and further improve the accuracy and robustness of the prediction.

[0144] Finally, the water hazard prediction module can output detailed risk assessment results, including the probability, type, time range of water hazard occurrence, and key information of the affected area. For example, when the model detects an abnormal combination of water level, water pressure, and geological strain data, it can generate a risk assessment report, indicating the possible floor water hazard and its potential impact range, providing a scientific basis for coal mine safety production and water hazard prevention and control.

[0145] Furthermore, in step S3, the water hazard prediction module inputs the preprocessed real-time monitoring data into the water hazard risk prediction model based on ensemble learning. This model integrates multiple machine learning algorithms such as decision tree (DT), support vector machine (SVM), and random forest (RF), giving full play to the advantages of each algorithm to achieve accurate prediction of coal mine water hazard risks.

[0146] Specifically, the decision tree algorithm constructs a tree structure and classifies and judges different hydrogeological factors based on indicators such as information gain or Gini index, and can quickly identify the key factor combinations that may lead to water hazards. The support vector machine maps the data to a high-dimensional space through a kernel function (such as the radial basis function RBF) to handle non-linear classification problems and can find the optimal classification boundary between the occurrence and non-occurrence of water hazards in a complex data space. The random forest algorithm integrates multiple decision trees, samples and votes on the data using the Bagging (Bootstrap Aggregating) method to improve the stability and accuracy of the prediction, and identifies key influencing factors through feature importance analysis.

[0147] During the model training process, a large number of historical water disaster cases and corresponding hydrogeological data are used to enable the model to learn the complex mapping relationships between different combinations of factors and the probability of water disasters, types (such as roof water disasters, floor water disasters, goaf water disasters), possible occurrence times, and affected ranges. For example, when the water level rises sharply within a short period and the water pressure exceeds a certain threshold, and at the same time the geological strain data shows obvious deformation near the fault, the model can predict a relatively high probability of a possible floor water disaster and roughly estimate the occurrence time and the possible affected area, thus achieving accurate prediction of coal mine water disasters and outputting detailed risk assessment results.

[0148] To further improve the performance of the model, the water disaster prediction module adopts cross-validation and grid search techniques to optimize the selection of hyperparameters and avoid overfitting problems. In addition, the model also introduces the Stacking method in ensemble learning, taking the prediction results of multiple base learners (such as decision trees, support vector machines, random forests) as the input of the meta-learner (such as logistic regression or gradient boosting trees), and performing secondary learning through the meta-learner, so as to integrate the advantages of each base learner and further improve the accuracy and robustness of the prediction.

[0149] Based on the output results of the water disaster prediction module, the risk assessment module further uses a multi-criteria evaluation method that combines the risk matrix method (RiskMatrix) and the fuzzy comprehensive evaluation method (Fuzzy Comprehensive Evaluation) to quantitatively evaluate and classify the water disaster risks. The risk matrix method constructs a matrix based on the two dimensions of the possibility and harm degree of water disasters, and divides the risks into different grade intervals. For example, the possibility of occurrence is divided into three levels: low, medium, and high, and the harm degree is divided into three levels: minor, severe, and catastrophic, forming nine risk levels through combination. The fuzzy comprehensive evaluation method takes into account the fuzzy and uncertain factors in the water disaster risk assessment and uses fuzzy mathematics methods to comprehensively process multiple evaluation indicators. It takes multiple factors such as the water level change trend, water pressure fluctuation, geological strain degree, abnormal degree of water chemical indicators, and microseismic activity frequency and intensity as evaluation indicators, and calculates the comprehensive risk value by determining the weights and membership functions of each indicator. Combining the results of the two methods, the water disaster risks are finally divided into different levels: low risk, medium risk, high risk, and extremely high risk. For different risk levels, corresponding response strategies and plans are formulated to provide a scientific basis for subsequent intelligent decision-making. At the same time, the risk assessment results are stored in the database in real time for subsequent query and analysis, providing data support for the long-term management of coal mine water disaster prevention and control.

[0150] In S4, the artificial intelligence decision-making module constructs an intelligent decision-making system based on the reinforcement learning algorithm. The intelligent decision-making system takes the water hazard risk assessment results and the real-time mine conditions (such as the current mining progress, the operation status of drainage equipment, and the ventilation system) as inputs, and generates the optimal prevention and control plan from a variety of pre-set prevention and control strategy libraries. Under the framework of reinforcement learning, the intelligent decision-making system regards the water hazard prevention and control process as a dynamic decision-making process. By continuously interacting with the environment (i.e., the actual mine situation), it learns the effects of different decision-making strategies in different situations, and optimizes the decision-making strategies according to the reward mechanism (such as the quality of prevention and control effects and the level of costs). For example, when the roof water hazard is assessed as a high-risk and the mine drainage equipment is operating at full load, the system will preferentially select the prevention and control strategy of grouting to block water, and use the optimization algorithm to determine the best layout of grouting holes and the reasonable control range of grouting volume according to the specific geological conditions of the roof and the water hazard situation. At the same time, the virtual reality (VR) / augmented reality (AR) technology is used to visually present the generated plan, enabling technicians to simulate the plan implementation process as if they were on the spot. In the VR / AR environment, technicians can intuitively see the positions of grouting holes, the laying paths of drainage pipes, and the operation conditions of prevention and control equipment, and can evaluate the feasibility, risks, and benefits of the plan in real time, and further optimize and adjust the plan according to the simulation results to ensure the scientificity and effectiveness of the prevention and control plan.

[0151] The equipment control module is responsible for converting the optimized prevention and control plan generated by the artificial intelligence decision-making module into specific control instructions for underground prevention and control equipment, and realizing the interconnection and interoperability between underground prevention and control equipment (such as drainage pumps, grouting pumps, valves, and ventilators) and the ground monitoring center through the Internet of Things technology. After receiving the control instructions, the equipment control module first parses and verifies the instructions to ensure the accuracy and safety of the instructions. Then, through the underground automatic control system, it remotely operates the start, stop, and operation parameter adjustment functions of the equipment according to the instruction requirements. For example, for the drainage pump, according to the water hazard risk and the mine water inflow situation, it precisely controls the rotation speed and drainage volume of the drainage pump to achieve efficient drainage. At the same time, the equipment control module uses sensors to collect the operation parameters of the equipment in real time (such as temperature, vibration, current, and pressure), and judges whether the equipment is operating normally through data analysis and machine learning models. It adopts a fault diagnosis method based on principal component analysis (PCA). After reducing the dimension of multiple operation parameters, it compares them with the characteristic model in the normal operation state. Once it is found that the equipment operation parameters deviate from the normal range, it immediately issues an alarm and provides fault diagnosis information to assist maintenance personnel in quickly locating and repairing faults to ensure the reliable operation of the prevention and control equipment.

[0152] In S5, when the water hazard risk reaches the preset emergency threshold or a sudden water hazard accident occurs, the emergency response module is immediately activated. The emergency response module is closely connected to the underground emergency refuge system and automatically triggers the underground emergency refuge function;

[0153] Specifically, quickly turn on the emergency lighting to ensure that underground workers can see the escape route clearly in the dark environment; activate the escape indicator signs, and through flashing and arrow indications, guide personnel to evacuate to the safe area; control the air door to close to prevent harmful gases from spreading to the personnel escape route and ensure the safety of personnel during the evacuation process; at the same time, send detailed emergency instructions and refuge information to underground workers and ground management personnel through various communication channels (such as underground broadcasting, personnel positioning terminals, and large-screen displays in the ground command center), including the type of water hazard, dangerous areas, escape routes, and emergency assembly points; the underground broadcasting is carried out in a loop in a high-volume and multi-band manner to ensure that personnel can clearly hear in the noisy underground environment; the personnel positioning terminal not only receives emergency information but also feeds back the real-time position information of personnel to the ground command center, and the ground commanders can thus master the personnel evacuation progress and conduct timely scheduling and command for possible congestion or detention situations, such as adjusting the air volume and direction of the ventilation system to guide personnel to avoid dangerous areas and ensure that personnel can quickly and orderly evacuate from the dangerous area to minimize casualties and property losses; in addition, the emergency response module can also be linked with external rescue forces (such as the mine rescue team and the fire department), and transmit accident information and on-site conditions in a timely manner through a dedicated communication link to assist in the efficient development of rescue work. For example, provide the mine rescue team with detailed maps of underground roadways, the location of water hazards, and information on trapped personnel to provide key support for rescue operations;

[0154] The data management and update module is responsible for establishing and maintaining a large-scale coal mine hydrogeological database during the operation of the entire system. It stores multi-source data such as historical geological data, annual monitoring data, water hazard accident cases, implementation records of prevention and control plans, and model training results. It classifies, archives, and backs up the data to ensure data security and traceability, and adopts distributed storage technology and data encryption algorithms to prevent data loss and leakage. It regularly mines and analyzes the data in the database, and uses data mining algorithms (such as association rule mining, clustering analysis) to extract valuable information and knowledge, such as discovering potential laws of water hazard occurrence under different geological conditions and the effectiveness of a certain prevention and control strategy in specific situations, providing data support for the optimization and improvement of the system. At the same time, according to newly collected monitoring data and feedback information during the prevention and control process, it uses online learning or incremental learning technology to update and optimize the relevant artificial intelligence models of the geological modeling module, water hazard prediction module, and artificial intelligence decision-making module. For example, when new data shows that the geological structure in a certain area has changed, the geological modeling module will automatically adjust the model parameters and re-learn and adapt to the new geological situation. By continuously updating and optimizing the model, the system can continuously adapt to the dynamic changes of coal mine geological conditions and water hazard risks, and continuously improve the prevention and control effect.

[0155] After the system has been running for a period of time, the data management and update module will conduct a comprehensive analysis of the collected data. For example, by comparing hydrogeological data in different time periods and different mining areas, it is found that during the mining process of a certain specific coal seam, when the water pressure of the surrounding aquifer rises at a rate exceeding a certain threshold within a short period of time and is accompanied by a significant increase in the content of specific mineral components in the water, the probability of roof water hazard occurrence increases significantly. Based on this analysis result, the algorithm of the water hazard prediction module is adjusted to increase the weight of this specific index combination, thereby improving the prediction accuracy.

[0156] The above embodiments are only used to illustrate rather than limit the technical solutions of the present invention. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that; still can modify the present invention or make equivalent replacements, and any modification or partial replacement without departing from the spirit and scope of the present invention shall be covered by the scope of the claims of the present invention.

Claims

1. An artificial-intelligence-based coal mine water disaster prevention and control system, characterized in that: It includes a data acquisition module, a data transmission module, a data preprocessing module, a geological modeling module, a water hazard prediction module, a risk assessment module, an artificial intelligence decision-making module, an equipment control module, an emergency response module, and a data management and update module; The data acquisition module transmits the data information collected at the underground coal mine site to the data preprocessing module through the data transmission module. After the data preprocessing module cleans, regularizes, and preprocesses the collected data information, it transmits the data to the geological modeling module and the water hazard prediction module respectively. The geological modeling module constructs a coal mine geological model based on a deep learning architecture and transmits the data to the data management and update module. The water hazard prediction module inputs the preprocessed real-time monitoring data into a water hazard risk prediction model based on ensemble learning. The output structure of the water hazard prediction module is transmitted to the risk assessment module. The risk assessment module quantitatively evaluates and grades the water hazard risk. The risk assessment module and the data management and update module simultaneously transmit the data to the artificial intelligence decision-making module. The artificial intelligence decision-making module constructs an intelligent decision-making system based on a reinforcement learning algorithm and formulates effective and scientific prevention and control plans. The prevention and control plans are implemented through the equipment control module and / or the emergency response module.

2. The coal mine water disaster prevention and control system based on artificial intelligence according to claim 1, characterized in that: The data acquisition module deploys a variety of intelligent sensors around the aquifers, faults, goafs, and key node parts of the underground coal mine. The variety of intelligent sensors include high-precision water level sensors, water pressure sensors, geological strain sensors, water temperature sensors, water chemistry sensors, and microseismic sensors. At the same time, these sensors are manufactured using advanced microelectromechanical technology and nanomaterial processes, and have high sensitivity, high precision, and high stability. They can collect multi-dimensional hydrogeological data such as water level changes, water pressure fluctuations, minor deformations of geological structures, water temperature differences, changes in water chemical components, and microseismic signals that may indicate water inrush in real time, and transmit the data to the ground data processing center through a high-speed and reliable data transmission module; The data transmission module is responsible for building a safe and efficient data transmission channel between the underground and the ground, and adopts a hybrid transmission method combining wired and wireless. Among them, wired transmission uses mine optical cables to ensure the stability and high speed of data transmission, and mainly undertakes the transmission tasks of a large amount of real-time monitoring data; Wireless transmission uses low-power and high-bandwidth wireless communication technology as a supplement to wired transmission. It is used to flexibly collect and transmit the data of some sensors that are difficult to connect by wired means in the complex underground environment, and to ensure the redundant transmission of key data in case of emergency, avoid data loss, and ensure that the ground can receive the data collected by the data acquisition module in a timely and complete manner.

3. The coal mine water disaster prevention and control system based on artificial intelligence according to claim 2, characterized in that: The data preprocessing module is located in the ground data processing center, where machine learning algorithms are used to clean, regularize, and preprocess the collected data. First, obvious abnormal data points are identified and removed through outlier detection algorithms (such as the 3σ principle based on statistics or the isolation forest algorithm) to avoid interference with subsequent analysis. For a small number of missing values, appropriate data interpolation methods are used to supplement them according to the time series characteristics and spatial correlation of the data to ensure data continuity. Subsequently, normalization or standardization techniques are used to unify the dimension and scale of the data, and the original data is transformed into a high-quality data set suitable for model training and analysis, providing a good data foundation for subsequent artificial intelligence model processing. The geological modeling module constructs a coal mine geological model based on a deep learning architecture. Using the powerful spatial feature extraction ability of convolutional neural networks, detailed spatial feature information such as stratigraphic structure, aquifer distribution, fault strike and throw, and goaf range and shape is extracted from multi-source geological data including geological exploration reports, 3D geological scan data, and borehole data. At the same time, combining the processing advantages of recurrent neural networks for time series data, the evolution law of geological structures over time and the dynamic correlation between different geological factors are analyzed. Through the training of a large amount of historical data and real-time monitoring data, the model parameters are continuously optimized, enabling the model to accurately reproduce the complex underground geological environment and update in real time with the input of new data to maintain an accurate grasp of the geological conditions.

4. The coal mine water disaster prevention and control system based on artificial intelligence according to claim 3, characterized in that: The water hazard prediction module inputs the preprocessed real-time monitoring data into a water hazard risk prediction model based on ensemble learning. The water hazard risk prediction model integrates multiple machine learning algorithms such as decision trees, support vector machines, and random forests, comprehensively considering the influence of multiple factors such as water level, water pressure change trend, geological strain conditions, abnormal water chemical indicators, and microseismic activity frequency and intensity on the occurrence of water hazards. By training the model to learn the complex mapping relationship between different factor combinations and the probability, type, possible occurrence time, and influence range of water hazards, the accurate prediction of coal mine water hazards is realized, and a detailed risk assessment result is output. The risk assessment module further quantitatively evaluates and classifies the water hazard risk according to the output result of the water hazard prediction module. Using multi-criteria evaluation methods such as the risk matrix method or the fuzzy comprehensive evaluation method, considering factors such as the possibility of water hazard occurrence, the degree of harm, and the disaster resistance ability of the mine, the water hazard risk is divided into different levels. For different risk levels, corresponding response strategies and plans are formulated to provide a basis for subsequent intelligent decision-making. At the same time, the risk assessment results are stored in the database in real time for subsequent query and analysis, providing data support for the long-term management of coal mine water hazard prevention and control.

5. The coal mine water disaster prevention and control system based on artificial intelligence according to claim 4, characterized in that: The artificial intelligence decision-making module constructs an intelligent decision-making system based on the reinforcement learning algorithm. The intelligent decision-making system takes the water hazard risk assessment results and the real-time working conditions of the mine as inputs, and generates the optimal prevention and control plan from a variety of pre-set prevention and control strategy libraries. The prevention and control strategies include, but are not limited to, the optimized design of drainage projects, the formulation of grouting water-blocking plans, the reasonable setting of waterproof coal pillars, and the strengthening measures for roadway support. The generated plan is visually presented using virtual reality / augmented reality technology, enabling technicians to simulate the implementation process of the plan as if they were on the scene, intuitively evaluate the feasibility, risks, and benefits of the plan, and further optimize and adjust the plan according to the simulation results to ensure the scientificity and effectiveness of the prevention and control plan; The equipment control module is responsible for converting the optimized prevention and control plan generated by the artificial intelligence decision-making module into specific control instructions for underground prevention and control equipment, and realizing the interconnection and intercommunication between the underground prevention and control equipment and the ground monitoring center through the Internet of Things technology. Ground monitoring personnel can remotely operate the start, stop, and operation parameter adjustment functions of the equipment according to the control instructions; at the same time, the equipment control module also has the functions of equipment status monitoring and fault diagnosis. It uses sensors to collect the operation parameters of the equipment in real time, and judges whether the equipment is operating normally through data analysis and machine learning models. Once a fault hidden danger is found, an alarm is issued in time and fault diagnosis information is provided to assist maintenance personnel in quickly locating and repairing the fault to ensure the reliable operation of the prevention and control equipment.

6. The coal mine water disaster prevention and control system based on artificial intelligence according to claim 5, characterized in that: When the water hazard risk reaches the preset emergency threshold or a sudden water hazard accident occurs, the emergency response module is immediately activated. The emergency response module automatically triggers the underground emergency refuge system, such as turning on emergency lighting, starting escape indicator signs, and controlling the closing of air doors to prevent the spread of harmful gases; at the same time, detailed emergency instructions and refuge information, including the type of water hazard, dangerous areas, escape routes, and emergency assembly points, are sent to underground workers and ground management personnel through a variety of communication channels to ensure that personnel can quickly and orderly evacuate from the dangerous area, minimizing casualties and property losses to the greatest extent; in addition, the emergency response module can also be linked with external rescue forces, timely transmit accident information and on-site conditions, and assist in the efficient development of rescue work; The data management and update module is responsible for the data management and model update work of the entire system; it establishes a large-scale coal mine hydrogeological database, stores multi-source data such as historical geological data, annual monitoring data, water hazard accident cases, records of the implementation of prevention and control plans, and model training results, and classifies, archives, and backs up the data to ensure the security and traceability of the data; Regularly mine and analyze the data in the database, extract valuable information and knowledge, and provide data support for the optimization and improvement of the system; at the same time, according to the newly collected monitoring data and feedback information during the prevention and control process, use online learning or incremental learning technology to update and optimize the relevant artificial intelligence models of the geological modeling module, water hazard prediction module, and artificial intelligence decision-making module, so that the system can continuously adapt to the dynamic changes of coal mine geological conditions and water hazard risks, and continuously improve the prevention and control effect.

7. A coal mine water disaster prevention and control method based on artificial intelligence, which is implemented by using the coal mine water disaster prevention and control system described in claim 6, and is characterized in that: Including the following steps: S1: Conduct data collection and transmission. Multisource data are collected by sensors at key locations in the coal mine underground and transmitted safely and stably to the ground data processing center through the data transmission module. S2: Data preprocessing and model construction. Clean, regularize, and standardize the data, and build an accurate geological model using CNN and RNN. S3: Conduct water hazard prediction and risk assessment. Analyze the data by integrating multiple algorithms to accurately predict water hazards and quantitatively evaluate their risk levels. S4: Intelligent decision-making and solution implementation. Determine the solution through reinforcement learning based on risks and working conditions, and use VR / AR for assistance in implementation and optimization. S5: Conduct emergency handling and data update. Ensure safety through emergency response and link up with rescue operations, and update the data to optimize the system performance.

8. The method for preventing and controlling coal mine water disasters based on artificial intelligence according to claim 7, characterized in that: In S1, first, various sensors in the data collection module continuously collect hydrogeological data at key underground locations and transmit the data to the data transmission module. Near the aquifer, the water level sensor and water temperature sensor are closely arranged to constantly monitor the slight fluctuations in water level and the changes in water temperature. These data are crucial for judging the dynamic changes of the aquifer. Around the fault, the geological strain sensor and microseismic sensor can sensitively capture the minor deformations of the geological structure and the microseismic activities that may indicate water inrush, providing a key basis for detecting water hazard risks in advance. Around the goaf, the water pressure sensors are distributed all around to closely monitor the fluctuations in water pressure. Any abnormal change may be a precursor to water hazards. At the same time, at the key nodes of the roadway, the water chemistry sensor detects the chemical composition of the water in real time. Once abnormal changes are found in the mineral content and pH index of the water, it may mean the occurrence of water hazards. The data transmission module constructs a safe and efficient hybrid transmission network combining wired and wireless. The wired transmission part uses the mine optical cable as the main trunk. Its high bandwidth and low latency characteristics ensure that a large amount of real-time monitoring data can be transmitted stably and quickly, undertaking the main data transmission task. Wireless transmission uses low-power, high-bandwidth wireless communication technology as a supplementary means. In some complex areas where it is difficult to lay optical cables, the wireless sensor nodes use wireless communication technology to send the collected data to the nearby wireless access points, and then transmit the data to the ground through the connection with the wired network. During the data transmission process, the data transmission module adopts advanced data verification and error correction algorithms. Through CRC verification, FEC error correction, and ARQ retransmission technologies, it ensures the reliable transmission of data in complex environments. In the future, an AI-driven adaptive transmission strategy can be introduced to dynamically optimize the verification, error correction, and retransmission mechanisms according to the network status, further improving the transmission efficiency and reliability, and providing a solid data foundation for subsequent analysis work. The sensor data acquisition method in S1 is as follows: the sensor data acquisition module includes different types of sensors for monitoring key environmental parameters in the coal mine: water level sensor, water pressure sensor, geological strain sensor, water temperature sensor, water chemistry sensor and microseismic sensor; the reading recorded by each sensor at time t is expressed as Xi(t), where Xi represents the corresponding sensor type; the overall state of the sensor data at time t is expressed as a vector: Xi(t) = [Wi(t), Psi(t), Si(t), Twi(t), Ci(t), Mi(t)] These sensors, through advanced micro-electromechanical technology and nanomaterial processing, have extremely high sensitivity and precision. They can work stably in complex and harsh underground environments, continuously collect multi-dimensional hydrogeological data, and quickly transmit the data to the data transmission module through a high-speed and reliable underground communication network.

9. The coal mine water disaster prevention and control method based on artificial intelligence according to claim 8, characterized in that: In S2, the method for preprocessing the collected data is as follows: after receiving the data from the data transmission module, the data preprocessing module quickly starts a series of machine learning algorithms to clean and regularize the data. First, a method based on the statistical 3σ principle combined with the isolation forest algorithm is used to identify and eliminate significantly abnormal data points; Specifically, assuming that the data obeys a normal distribution, the mean μ and standard deviation σ of the water level Wi, water pressure Psi, geological strain Si, water temperature Twi, hydrochemistry Ci, and microseismic Mi parameters are calculated respectively, and data points outside the range of [μ-3σ, μ+3σ] are marked as abnormal. At the same time, an isolation tree is constructed by randomly selecting features and split points, and the isolation forest algorithm is used to construct models for Wi, Psi, Si, Twi, Ci, and Mi respectively. The anomaly score of each data point is calculated, and the data points with an anomaly score higher than a preset threshold are marked as abnormal. If a data point is marked as abnormal by both the 3σ principle and the isolation forest algorithm, it is determined to be an abnormal data point. Through the above method, abnormal data points can be efficiently identified and eliminated to ensure the accuracy and reliability of the data. In S2, for a small number of missing values, appropriate data interpolation methods are used to supplement them based on the time series characteristics and spatial correlation of the data. Taking water level data as an example, if there are missing values in a short period of time and the surrounding sensor data show that the water level changes are relatively stable, linear interpolation is used to supplement them. The calculation formula of the linear interpolation method is: Among them, w is the water level value corresponding to the missing point x, (x1, w1) and (x2, w2) represent the positions of the adjacent non-missing value points before and after the missing value x and their corresponding water level values, respectively. This method can effectively fill in the missing values and ensure the continuity and integrity of the data. In S2, for geological structure data, if the missing part has certain similarities with the geological characteristics of the known area, an interpolation model based on deep learning is used to fill in the missing part by learning the characteristics of the surrounding geological data. Subsequently, the Min-Max normalization or Z-score standardization method is used to unify the dimension and scale of the data, and the original data is transformed into a high-quality data set suitable for model training and analysis, providing a good data basis for subsequent artificial intelligence model processing. The Min-Max normalization formula is: where X' i (t) is the normalized sensor data, min(X i ) is the minimum value of the readings of sensor i at all time points, and max(X i ) is the maximum value of the readings of sensor i at all time points; The Z-score standardization formula is: where X″ i (t) is the normalized reading of sensor X i at time t, μ(X i ) is the mean of the readings of sensor X i over all recorded times, and σ(X i ) is the standard deviation of the readings of sensor X i over all recorded times; Then, at time t, the vector X′(t) after normalization of the overall state vector X(t) of the sensor data is: i (t) is: X′ i (t) = [W′ i (t), Ps′ i (t), S′ i (t), Tw′ i (t), C′ i (t), M′ i (t)] where each element X′ i (t) is calculated according to the normalization formula of the above single sensor reading; In S2, based on the preprocessed data and rich historical geological materials, the convolutional neural network and the recurrent neural network are used to construct and train the coal mine geological model for the geological modeling module. Among them, the convolutional neural network mainly solves the problem of image recognition. It integrates data from multi-source geological materials such as geological exploration reports, three-dimensional geological scanning data, and borehole data to train the model, enabling it to have the ability to extract spatial features such as stratigraphic structure, aquifer distribution, faults, and the scope and shape of goafs. Specifically, in the convolutional neural network architecture for geological data processing, the initial convolutional layer performs spatial-channel joint feature extraction on the original multi-channel geological data through a three-dimensional convolutional kernel (3×3×C), where C represents the physical property parameter dimension of the input data. Spatial dimensionality reduction of the feature map is achieved through cascaded strided convolution and max pooling layers, and at the same time, the ReLU activation function is used to introduce non-linear transformation. The deep network gradually restores the spatial resolution through transposed convolution operations, and the final output layer reconstructs the channel dimension through a 1×1 convolutional kernel, completely retaining the multi-physical field coupling characteristics of the geological body. The operation of the initial convolutional layer is expressed as: F1 = f(W1 * X + b1) Among them, F1 represents the feature map after the first layer of convolution, X is the input sensor data X(t), W1 and b1 are the three-dimensional convolutional kernel and the bias term respectively, * represents the convolution operation, and f is the ReLU activation function. The operations of the strided convolution and pooling layers are expressed as: F2 = f(W2 * F1 + b2) F pool = MaxPool(F2) Among them, W2 is the step convolution kernel, and F pool is the max pooling layer; The operations of the deep network are expressed as: F3 = f(W3 * F pool + b3) Among them, W3 is the transposed convolution kernel; The operation of the output layer is expressed as: F CNN = f(W4 * F3 + b4) Among them, W4 is a 1×1 convolutional kernel of the output layer, and F CNN is the coupled multi-physical field feature vector of the final output; Specifically, in the recurrent neural network architecture for multi-source hydrogeological time series analysis, the initial gated recurrent unit performs spatio-temporal-physical joint feature modeling on the original time series data through a two-dimensional state gating mechanism. The input feature dimension includes dynamic monitoring parameters such as water level (m), water pressure (MPa), and formation displacement rate (mm / d), as well as geological state parameters such as rock creep coefficient and fracture development index. Multi-time scale feature extraction is achieved through cascaded causal dilated convolution and hierarchical state transfer, and at the same time, the gated linear unit is used to introduce parametric non-linear interaction. The deep network dynamically focuses on key time nodes through a time-domain attention mechanism, and the final output layer realizes the reconstruction of multi-physical field coupling response through a differentiable physical constraint layer, strictly following the principles of mass conservation and energy balance. The time series data input to the recurrent neural network part is X(t - k), X(t - k + 1), …, X(t), which is expressed as: h t = f(W h X t + U h h t-1 + b h ) Among them, ht is the hidden state at the current moment, W h , U h are weight matrices, b h is the bias term, f is the ReLU activation function, and through the recursive operation of time steps, the RNN generates the time series feature vector F RNN ; To further improve the performance of the model, the water hazard prediction module adopts cross-validation and grid search techniques to optimize the selection of hyperparameters and avoid overfitting problems. In addition, the model also introduces the Stacking method in ensemble learning, taking the prediction results of multiple base learners as the input of the meta-learner, and performing secondary learning through the meta-learner, so as to integrate the advantages of each base learner and further improve the accuracy and robustness of the prediction. Finally, the water hazard prediction module can output detailed risk assessment results, including the probability, type, time range of water hazard occurrence, and key information of the affected area. For example, when the model detects an abnormal combination of water level, water pressure, and geological strain data, it can generate a risk assessment report, indicating the possible floor water hazard and its potential impact range, providing a scientific basis for coal mine safety production and water hazard prevention and control.

10. A coal mine water disaster prevention and control method based on artificial intelligence according to claim 9, characterized in that: In step S3, the water hazard prediction module inputs the preprocessed real-time monitoring data into the water hazard risk prediction model based on ensemble learning. This model integrates multiple machine learning algorithms such as decision trees, support vector machines, and random forests, giving full play to the advantages of each algorithm to achieve accurate prediction of coal mine water hazard risks. Specifically, the decision tree algorithm constructs a tree structure and classifies and judges different hydrogeological factors based on information gain or Gini coefficient indicators, capable of quickly identifying key factor combinations that may lead to water hazards. The support vector machine maps data to a high-dimensional space through a kernel function to handle non-linear classification problems, and can find the optimal classification boundary between the occurrence and non-occurrence of water hazards in a complex data space. The random forest algorithm integrates multiple decision trees, samples and votes on data using the Bagging method, improves the stability and accuracy of prediction, and identifies key influencing factors through feature importance analysis. During the model training process, a large number of historical water hazard cases and corresponding hydrogeological data are used to enable the model to learn the complex mapping relationship between different factor combinations and the probability, type, possible occurrence time, and affected range of water hazard occurrence. For example, when the water level rises sharply in a short period of time and the water pressure exceeds a certain threshold, and at the same time the geological strain data shows obvious deformation near the fault, the model can predict that the probability of possible floor water hazard is relatively high and roughly estimate the occurrence time and possible affected area range, thus achieving accurate prediction of coal mine water hazards and outputting detailed risk assessment results. To further improve the performance of the model, the water hazard prediction module adopts cross-validation and grid search techniques to optimize the selection of hyperparameters and avoid overfitting problems. In addition, the model also introduces the Stacking method in ensemble learning, taking the prediction results of multiple base learners as the input of the meta-learner, and performing secondary learning through the meta-learner, so as to integrate the advantages of each base learner and further improve the accuracy and robustness of the prediction. Based on the output of the flood disaster prediction module, the risk assessment module further uses a multi-criteria assessment method that combines the risk matrix method and the fuzzy comprehensive evaluation method to quantitatively assess and grade flood disaster risks. The risk matrix method constructs a matrix based on the two dimensions of the possibility of flood disaster occurrence and the degree of harm, dividing risks into different level intervals. For example, the possibility of occurrence is divided into three levels: low, medium, and high, and the degree of harm is divided into three levels: minor, severe, and catastrophic. Through combination, nine risk levels are formed. The fuzzy comprehensive evaluation method takes into account the fuzziness and uncertainty factors in flood disaster risk assessment and uses fuzzy mathematics methods to comprehensively process multiple evaluation indicators. It uses water level change trends, water pressure fluctuations, geological strain, abnormality of hydrochemical indicators, and frequency and intensity of microseismic activity as evaluation indicators. By determining the weight and membership function of each indicator, it calculates the comprehensive risk value. Combining the results of the two methods, the water disaster risk is finally divided into different levels of low risk, medium risk, high risk and very high risk; Develop corresponding response strategies and plans for different risk levels, providing a scientific basis for subsequent intelligent decision-making. At the same time, the risk assessment results are stored in the database in real time for subsequent query and analysis, providing data support for the long-term management of coal mine water hazard prevention and control. In said S4, the artificial intelligence decision-making module builds an intelligent decision-making system based on the reinforcement learning algorithm; An intelligent decision-making system is constructed, taking the flood risk assessment results and the real-time working conditions of the mine as inputs, and generating the optimal prevention and control plan from a library of multiple pre-set prevention and control strategies. Within the framework of reinforcement learning, the intelligent decision-making system regards the flood control process as a dynamic decision-making process. By continuously interacting with the environment, it learns the effects of different decision-making strategies in different situations and optimizes the decision-making strategy based on a reward mechanism. For example, when the risk assessment is high for roof flooding and the mine drainage equipment is operating at full capacity, the system will prioritize the prevention and control strategy of grouting water blocking. Based on the specific geological conditions of the roof and the flood situation, an optimization algorithm is used to determine the optimal layout of grouting holes and the reasonable control range of grouting volume. At the same time, virtual reality / augmented reality technology is used to visualize the generated plan, allowing technicians to simulate the implementation process in an immersive way. In the VR / AR environment, technicians can intuitively see the location of grouting holes, the laying path of drainage pipes, and the operation of prevention and control equipment, and can evaluate the feasibility, risks, and benefits of the plan in real time. Based on the simulation results, the plan can be further optimized and adjusted to ensure the scientific nature and effectiveness of the prevention and control plan. The device control module is responsible for converting the optimized prevention and control plan generated by the artificial intelligence decision-making module into specific control instructions for underground prevention and control equipment. Through Internet of Things technology, it realizes the interconnection and interoperability between underground prevention and control equipment and the ground monitoring center. After receiving the control instructions, the device control module first parses and verifies the instructions to ensure the accuracy and safety of the instructions. Then, through the underground automation control system, it remotely operates the start, stop, and operation parameter adjustment functions of the equipment according to the instruction requirements. For example, for the drainage pump, according to the water hazard risk and the mine water inflow situation, it precisely controls the rotation speed and drainage volume of the drainage pump to achieve efficient drainage. At the same time, the device control module uses sensors to collect the operation parameters of the equipment in real time, and judges whether the equipment is operating normally through data analysis and machine learning models. It adopts a fault diagnosis method based on principal component analysis. After reducing the dimension of multiple operation parameters, it compares them with the characteristic model in the normal operation state. Once it is found that the equipment operation parameters deviate from the normal range, it immediately issues an alarm and provides fault diagnosis information to assist maintenance personnel in quickly locating and repairing faults to ensure the reliable operation of the prevention and control equipment. In S5, when the water hazard risk reaches the preset emergency threshold or a sudden water hazard accident occurs, the emergency response module is immediately activated. The emergency response module is closely connected to the underground emergency refuge system and automatically triggers the underground emergency refuge function. Specifically, it quickly turns on the emergency lighting to ensure that underground operators can see the escape route clearly in the dark environment. It activates the escape indicator signs, and through flashing and arrow indications, guides personnel to evacuate to the safe area. It controls the air doors to close to prevent harmful gases from spreading to the personnel escape route and ensures the safety of personnel during the evacuation process. At the same time, it sends detailed emergency instructions and refuge information to underground operators and ground management personnel through multiple communication channels, including the type of water hazard, dangerous areas, escape routes, and emergency assembly points. The underground broadcast is looped in a high-volume and multi-band manner to ensure that personnel can clearly hear it in the noisy underground environment. The personnel positioning terminal not only receives emergency information but also feeds back the real-time position information of personnel to the ground command center. Ground commanders can thus master the personnel evacuation progress and conduct timely dispatching and command for possible congestion or stagnation situations, such as adjusting the air volume and direction of the ventilation system to guide personnel to avoid dangerous areas and ensure that personnel can quickly and orderly evacuate from the dangerous area, minimizing casualties and property losses. In addition, the emergency response module can also be linked with external rescue forces, and timely transmit accident information and on-site conditions through a dedicated communication link to assist in the efficient development of rescue work. For example, it provides the detailed map of the underground roadway, the location of the water hazard, and the information of trapped personnel to the mine rescue team to provide key support for the rescue operation. During the operation of the entire system, the data management and update module is responsible for establishing and maintaining a large-scale coal mine hydrogeological database, storing multi-source data such as historical geological data, annual monitoring data, water disaster case records, implementation records of prevention and control plans, and model training results. It classifies, archives, and backs up the data to ensure data security and traceability, and adopts distributed storage technology and data encryption algorithms to prevent data loss and leakage. Regularly, it mines and analyzes the data in the database, uses data mining algorithms to extract valuable information and knowledge, such as discovering potential laws of water disasters under different geological conditions and the effectiveness of a certain prevention and control strategy in specific situations, providing data support for system optimization and improvement. At the same time, according to newly collected monitoring data and feedback information during the prevention and control process, it uses online learning or incremental learning technology to update and optimize the relevant artificial intelligence models of the geological modeling module, water disaster prediction module, and artificial intelligence decision-making module. For example, when new data shows that the geological structure in a certain area has changed, the geological modeling module will automatically adjust the model parameters, re-learn, and adapt to the new geological situation. By continuously updating and optimizing the model, the system can continuously adapt to the dynamic changes of coal mine geological conditions and water disaster risks, and continuously improve the prevention and control effect. After the system has been running for a period of time, the data management and update module will conduct a comprehensive analysis of the collected data. For example, by comparing hydrogeological data from different time periods and different mining areas, it is found that during the mining of a certain specific coal seam, when the water pressure of the surrounding aquifer rises at a rate exceeding a certain threshold within a short period of time and is accompanied by a significant increase in the content of specific mineral components in the water, the probability of roof water disaster occurrence increases significantly. Based on this analysis result, the algorithm of the water disaster prediction module is adjusted to increase the weight of this specific combination of indicators, thereby improving the prediction accuracy.

Images