Intelligent identification and early warning method and system for delamination water disaster

Abstract

The application discloses a kind of delamination water disaster intelligent identification early warning method and system, belong to mine safety and geological disaster intelligent early warning field, by constructing dynamic updating high-precision three-dimensional geological model, and using the intelligent interpolation algorithm of fusion geological structure and rock occurrence, discrete, heterogeneous geological, mechanical, hydrological data are uniformly converted into continuous, visual three-dimensional parameter field, overcome the defect that traditional method data is isolated, representation is one-sided, comprehensively use multiple machine learning and deep learning algorithm, respectively for delamination development, water supply, aquifer stability Key link is intelligently judged and simulated, reduces artificial experience dependence, substantially improves the accuracy of delamination spatial positioning, water source evaluation and disaster trend prediction, through deep reinforcement learning simulation water pressure dynamic evolution, and combine multi-scale time series model to predict future trend, dynamically capture early signal of disaster gestation, significantly improve the timeliness and foresight of early warning.

Description

Technical Field

[0001] This invention belongs to the field of intelligent early warning for mine safety and geological disasters, and specifically relates to an intelligent identification and early warning method and system for delamination water hazards. Background Technology

[0002] During coal mining, the goaf formed after the coal seam is extracted disrupts the original stress balance of the overlying strata, leading to movement, deformation, and fracturing. Due to significant differences in the thickness, strength, and elastic modulus of different strata, they undergo asynchronous deformation during bending and subsidence, resulting in separation spaces, or "delamination," between the layers. When this delamination space develops beneath a water-rich aquifer and is connected to a water source, it easily accumulates high-pressure water, forming delamination water bodies. Under mine pressure, water pressure, or mining disturbance, this delamination water body may suddenly burst into the mine, causing catastrophic delamination water hazards. These water hazards are characterized by their high concealment, rapid inrush, large volume, and long duration, posing a significant safety threat to deep coal mining.

[0003] Currently, the identification and early warning of delamination water damage mainly rely on the following technical approaches, but all of them have obvious limitations:

[0004] The experience-based threshold method based on single monitoring data: This method mainly acquires single-type data by deploying borehole water level gauges, stress gauges, or microseismic monitoring systems. An alarm is triggered when the monitored value exceeds a fixed threshold set based on historical experience. Its drawbacks are: severe lag: by the time an alarm is triggered, the disaster is often already in the development or even outbreak stage, making early warning impossible. High false alarm rate: threshold setting relies on subjective experience, making it difficult to adapt to complex and changing geological conditions and susceptible to interference from accidental factors. Weak causal correlation: relying solely on a single abnormal indicator cannot accurately determine whether the anomaly is caused by aquifer water or other geological factors, lacking in-depth correlation analysis of the disaster formation mechanism.

[0005] Analytical calculation method based on classical theoretical models: This method mainly applies the "upper three zones" (collapse zone, fracture zone, and bending subsidence zone) theory and the mechanical model of thin plates or beams. By inputting limited rock strata parameters, it calculates the height of the water-conducting fracture zone and then infers the possible location of delamination. Its limitations are: Highly simplified model: The complex three-dimensional geological body is simplified into a homogeneous, isotropic ideal model, which cannot consider the influence of complex structures such as faults and folds, as well as the spatial variability of lithology. Difficulty in parameter selection: The key mechanical parameters required for the calculation are usually obtained from laboratory tests of a small number of boreholes, which are difficult to represent the real situation of the entire working face, resulting in poor reliability of the calculation results. Static analysis: It cannot simulate the dynamic development and closure of delamination and the process of water migration and accumulation during mining.

[0006] Discrimination methods based on expert systems or simple data fusion: These methods attempt to simply overlay multiple monitoring data with some geological data and make risk judgments based on expert experience bases or simple logical rules. Although they are an improvement over the previous two methods, fundamental problems remain: Knowledge acquisition bottleneck: Expert experience is difficult to fully and formally transform into computer rules, and updating and maintaining them is difficult. Shallow fusion level: It is only a simple aggregation at the data level, failing to achieve deep fusion and coupled analysis of geological structure, rock mass parameters, hydrological conditions, and dynamic monitoring data under a unified three-dimensional spatial model. Poor adaptability: It cannot automatically learn and optimize the model from massive amounts of data, making it difficult to cope with new mining areas and new conditions. Summary of the Invention

[0007] To address the shortcomings of the existing technologies, this application provides a method and system for intelligent identification and early warning of delamination water hazards.

[0008] The first aspect of this application proposes an intelligent identification and early warning method for delamination water hazards, including the following steps:

[0009] Integrate borehole, geophysical and mining engineering data of the target area to construct a three-dimensional geological model that includes strata, structure and mining space;

[0010] The physical and mechanical parameters and hydrogeological parameters of the rock strata in the target area are collected. An improved Kriging spatial interpolation algorithm that integrates geological structural constraints and rock strata attitude is used to generate a three-dimensional rock mass mechanical parameter field and hydrogeological parameter field.

[0011] Based on the aforementioned three-dimensional geological model and three-dimensional rock mass mechanical parameter field, an improved random forest algorithm is used to determine the rock strata structure, and the deflection difference of the combined rock strata is calculated based on a multi-layer beam mechanical model. Then, a hybrid time series model is used to integrate microseismic monitoring data to predict the spatiotemporal location and risk index of delamination development.

[0012] Aquifer information is automatically extracted from borehole engineering data based on an image recognition model, and hydrological entities in geological text are parsed based on a natural language processing model. Combined with the hydrological parameter field, a graph neural network model is used to evaluate the risk index of water recharge from the aquifer to the separation space.

[0013] Based on the three-dimensional parameter field, an integrated learning model is used to evaluate the stability probability of the aquitard. At the same time, a deep reinforcement learning environment with water pressure and stress as states is constructed to simulate water pressure dynamics, and a time series model is used to predict the water pressure change trend. By combining the delamination risk index, water supply risk index, aquitard stability probability and water pressure prediction trend, comprehensive risk warning information for delamination water hazards is generated.

[0014] In some embodiments, the improved Kriging spatial interpolation algorithm performs piecewise interpolation by introducing fault influence factors and corrects the interpolation weights by using the dip angle of the rock strata. Its variogram model is an anisotropic model.

[0015] In some embodiments, the improved random forest algorithm employs a dynamic weighting strategy for feature importance, and its weight calculation integrates the partial derivatives of Gini importance and model accuracy with respect to that feature; the input of the hybrid time series model includes historical microseismic event sequences and geological structural feature vectors.

[0016] In some embodiments, the image recognition model is a U-Net++ network with added geological prior constraints; the natural language processing model is an attention-enhanced BiLSTM-CRF model; and the node features of the graph neural network model include aquifer thickness, hydraulic head, distance from the aquifer, and permeability coefficient.

[0017] In some embodiments, the ensemble learning model is an improved random forest algorithm that employs a dynamic sampling strategy to handle class imbalance; the deep reinforcement learning environment is trained using a proximal policy optimization algorithm, and its reward function is negatively correlated with the predicted water pressure error and the rate of change of water pressure; the time series model is an LSTM-Transformer model that integrates multi-scale features of discrete wavelet transform.

[0018] In some embodiments, the comprehensive risk warning information for delamination water hazards includes multi-level warning rules, which are as follows:

[0019] If both the delamination risk index and the water supply risk index exceed the first threshold, and the water pressure prediction trend is upward, then the first-level warning will be triggered.

[0020] If the probability of the stability of the aquitard is lower than the second threshold, and either the delamination risk index or the water supply risk index exceeds the third threshold, then a level 2 warning is triggered.

[0021] If the separation risk index exceeds the third threshold or the water pressure time series forecast shows abnormal fluctuations, a level 3 warning will be triggered.

[0022] Secondly, this application proposes an intelligent identification and early warning system for delamination water hazards, including a three-dimensional geological model construction module for mines, a three-dimensional parameter field construction module, an intelligent judgment module for delamination space, a water supply condition judgment module, and a comprehensive judgment and early warning module for delamination water disaster conditions.

[0023] The mine three-dimensional geological model construction module is used to integrate drilling, geophysical exploration and mining engineering data of the target area to construct a three-dimensional geological model that includes strata, structure and mining space;

[0024] The three-dimensional parameter field construction module is used to collect the rock strata physical and mechanical parameters and hydrogeological parameters of the target area, and to generate a three-dimensional rock mass mechanical parameter field and hydrogeological parameter field by using an improved Kriging spatial interpolation algorithm that integrates geological structural constraints and rock strata attitude.

[0025] The delamination space intelligent determination module is used to determine the rock strata structure based on the three-dimensional geological model and the three-dimensional rock mass mechanical parameter field, using an improved random forest algorithm, and to calculate the deflection difference of the combined rock strata based on the multi-layer beam mechanical model; then, using a hybrid time series model and integrating microseismic monitoring data, it predicts the spatiotemporal location and risk index of delamination development.

[0026] The water supply condition determination module is used to automatically extract aquifer information from borehole engineering data based on an image recognition model, and to parse hydrological entities in geological text based on a natural language processing model; combined with the hydrological parameter field, a graph neural network model is used to evaluate the risk index of water supply from the aquifer to the separation space.

[0027] The comprehensive judgment and early warning module for delamination water disaster conditions is used to evaluate the stability probability of the aquitard based on the three-dimensional parameter field using an integrated learning model; at the same time, it constructs a deep reinforcement learning environment with water pressure and stress as states to simulate water pressure dynamics, and uses a time series model to predict the water pressure change trend; and generates comprehensive risk early warning information for delamination water disaster by combining the delamination risk index, water source replenishment risk index, aquitard stability probability and water pressure prediction trend.

[0028] In some embodiments, the system also includes a data acquisition and input module for accessing borehole data, geophysical data, mining engineering drawings, real-time microseismic monitoring data, and hydrological monitoring data; and an early warning information output module for visually displaying the three-dimensional model, parameter field, risk distribution, and early warning report.

[0029] Thirdly, this application proposes an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0030] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described above.

[0031] The beneficial effects of this invention are:

[0032] By constructing a dynamically updated high-precision three-dimensional geological model and employing an intelligent interpolation algorithm that integrates geological structure and stratum attitude, discrete and heterogeneous geological, mechanical, and hydrological data are uniformly transformed into a continuous and visualized three-dimensional parameter field. This overcomes the shortcomings of traditional methods, such as isolated data and one-sided representation. By comprehensively utilizing various machine learning and deep learning algorithms, intelligent judgment and simulation are performed on key aspects such as aquifer development, water supply, and aquitard stability. This reduces reliance on human experience and significantly improves the accuracy of aquifer spatial location, water source assessment, and disaster trend prediction. Through deep reinforcement learning, the dynamic evolution of water pressure is simulated, and combined with multi-scale time series models, future trends are predicted. This dynamically captures early signals of disaster formation, significantly improving the timeliness and foresight of early warnings. Attached Figure Description

[0033] Figure 1 This is the overall flowchart of the present invention.

[0034] Figure 2 This is a system principle block diagram of the present invention. Detailed Implementation

[0035] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention may be implemented in various forms and should not be limited to the embodiments set forth herein; rather, these embodiments are provided so that a more thorough understanding of the invention can be achieved and that the full scope of the invention can be conveyed to those skilled in the art.

[0036] Firstly, this application proposes an intelligent identification and early warning method for delamination water hazards, such as... Figure 1 As shown, it includes the following steps:

[0037] S100: Integrate borehole, geophysical and mining engineering data of the target area to construct a three-dimensional geological model that includes strata, structure and mining space;

[0038] The specific implementation process is as follows: First, the system receives borehole data from the mine, including parameters such as borehole coordinates, depth, lithology, and thickness of the strata, and constructs a stratigraphic profile using the borehole data. Second, it integrates surface geophysical data (such as seismic exploration and electromagnetic exploration) and underground geophysical data (such as channel wave seismic and transient electromagnetic data) to obtain spatial information such as stratigraphic structure and structural anomalies. Third, it combines data such as mining engineering plans and tunnel sketches to establish a three-dimensional model of the mining space. In the data processing stage, the system uses the Delaunay triangulation algorithm to spatially interpolate discrete borehole data to generate continuous stratigraphic interfaces; it uses the Marching Cubes algorithm to convert geological body data into a three-dimensional mesh model; finally, it uses OpenGL or WebGL technology to achieve three-dimensional visualization rendering. The system also supports dynamic update functionality, allowing for real-time adjustment of model parameters when new drilling or mining data is acquired.

[0039] Taking a coal mine working face as an example, the system first imports data from 50 geological boreholes, each containing 20 strata boundary points. The system automatically identifies the coordinates (X, Y, Z) of each borehole and establishes a spatial distribution model. Then, based on the strata information revealed by the boreholes, the system uses Kriging interpolation to calculate the stratum thickness in un-drilled areas, generating contour maps of the coal seam's roof and floor. Next, seismic exploration data is imported, identifying three fault structures, which are then integrated with the borehole data to create a geological model incorporating the faults. Finally, combining the mining engineering plan, the mined areas and tunnel systems are overlaid onto the geological model to form a complete three-dimensional geological model of the mine. This model can be rotated and scaled arbitrarily and supports profile cutting functions, facilitating observation and analysis by geologists from different angles.

[0040] S200: Collect the physical and mechanical parameters and hydrogeological parameters of the rock strata in the target area, and use the improved Kriging spatial interpolation algorithm that integrates geological structural constraints and rock strata attitude to generate a three-dimensional rock mass mechanical parameter field and hydrogeological parameter field.

[0041] In some embodiments, the improved Kriging spatial interpolation algorithm performs piecewise interpolation by introducing fault influence factors and corrects the interpolation weights by using the dip angle of the rock strata. Its variogram model is an anisotropic model.

[0042] Six core physical and mechanical parameters of each coal (rock) seam were obtained through geological borehole sampling:

[0043]

[0044] Where h is the thickness of the rock strata (m), For elastic modulus, For tensile strength, It is of high density (kN / m³). The coefficient of expansion is denoted as 'splitting coefficient'. For density.

[0045] Three key parameters were obtained from hydrogeological boreholes:

[0046]

[0047] The thickness of the aquifer. For the thickness of the waterproof layer, This is the pressure head value.

[0048] The hybrid Kriging algorithm based on mutation function optimization is adopted, and the specific process is as follows:

[0049] Construct an anisotropic variogram model:

[0050]

[0051] in Value of a nugget. For sill values, For variable range, This is the lag distance.

[0052] Introducing geological structural constraints, a piecewise interpolation strategy is adopted in fault zones:

[0053]

[0054] in As the fault influence factor, when and The value is 1 when it is on the same side of the fault and 0 when it is on the opposite side;

[0055] By incorporating strata attitude parameters, a stratigraphic dip weight is added during the interpolation process:

[0056]

[0057] in, The angle between the direction of the line connecting the sampling point and the point to be estimated and the dip of the strata.

[0058] Taking the Permian coal seam in a certain mining area as an example, the system first collected strata parameter data from 32 geological boreholes, including:

[0059] Sandstone roof of coal seam #3: E = 12.5 ± 2.3 GPa, σ_t = 8.7 ± 1.2 MPa

[0060] Mudstone aquitard: h=6.8±1.5m, K=1.25

[0061] Ordovician limestone aquifer: H = +85~+112m

[0062] During interpolation calculations, the system automatically identifies NW-trending normal faults and processes the data in partitions. For elastic modulus interpolation, the range is set to a = 350m, nugget value c0 = 0.8, and sill value c = 4.2. A regular 500×500m grid is generated on the north side of the fault, and the estimated value for each grid node is calculated, ultimately generating a three-dimensional distribution cloud map of the elastic modulus. The results show a clear parameter gradient zone near the fault, with a 92% agreement with the field-measured fracture zone of the rock strata. The system can update the parameter field within 5 seconds of receiving each new borehole data point.

[0063] S300: Based on the three-dimensional geological model and the three-dimensional rock mass mechanical parameter field, an improved random forest algorithm is used to determine the rock strata structure, and the deflection difference of the combined rock strata is calculated based on the multi-layer beam mechanical model; then, a hybrid time series model is used to integrate microseismic monitoring data to predict the spatiotemporal location and risk index of delamination development.

[0064] In some embodiments, the improved random forest algorithm employs a dynamic weighting strategy for feature importance, and its weight calculation integrates the partial derivatives of Gini importance and model accuracy with respect to that feature; the input of the hybrid time series model includes historical microseismic event sequences and geological structural feature vectors.

[0065] Intelligent classification of rock strata: An improved random forest algorithm is used to determine rock strata types. Key improvements include:

[0066] Dynamic weighting of feature importance:

[0067]

[0068] in, For the importance of traditional gins, For adjustment coefficients, This represents the model's accuracy.

[0069] A constraint pruning strategy based on prior geological knowledge is introduced to delete decision paths that do not conform to the laws of rock strata mechanics;

[0070] Analysis of deflection differences in composite slabs: Establishment of a multi-layer beam mechanical model:

[0071]

[0072] in, Let i be the deflection of the i-th layer. , Interlayer contact stiffness;

[0073] The parameters are optimized using an adaptive differential evolution algorithm:

[0074]

[0075] in, Dynamically adjusts with the number of iterations: ;

[0076] Delamination interface prediction:

[0077] Constructing a hybrid LSTM-Transformer time series model:

[0078]

[0079] in This is a geological structural feature vector. Indicates feature splicing;

[0080] Delamination risk index calculation:

[0081]

[0082] For the Sigmoid function, Assigning weights to each factor These are the normalized eigenvalues;

[0083] Taking the mining of coal seam #3 in a certain mine as an example, the following input data is processed when determining the separation space:

[0084] The mining height is 2.8m and the working face length is 150m.

[0085] Top strata sequence: 2.5m mudstone (direct top) + 5.8m sandstone (basic top);

[0086] Parameters: Mudstone E=8.2GPa, Sandstone E=15.6GPa, Blast coefficient K=1.18;

[0087] Implementation process: 1. The ERF model first determines the immediate top and the basic top, and the importance of features is ranked as follows: coefficient of dilatation (0.35) > thickness (0.28) > elastic modulus (0.22).

[0088] 2. Calculate the deflection of each layer: mudstone layer D1=32.7mm, sandstone layer D2=18.4mm, difference rate ΔD=43.7%.

[0089] 3. The LSTM-Transformer model takes into account the microseismic monitoring data (number of events, energy release) from the previous 10 days and predicts that the maximum delamination height will be 1.2m in the next 3 days, with the location 85-92m away from the cut-in eye.

[0090] 4. The overall output delamination risk index R=0.76 (threshold 0.6), triggering a yellow alert. Field verification showed that the actual delamination development height was 1.05m, with a prediction error of 12.5%.

[0091] S400: Automatically extracts aquifer information from borehole engineering data based on an image recognition model, and parses hydrological entities in geological text based on a natural language processing model; combined with the hydrological parameter field, uses a graph neural network model to evaluate the risk index of water recharge from the aquifer to the separation space.

[0092] In some embodiments, the image recognition model is a U-Net++ network with added geological prior constraints; the natural language processing model is an attention-enhanced BiLSTM-CRF model; and the node features of the graph neural network model include aquifer thickness, hydraulic head, distance from the aquifer, and permeability coefficient.

[0093] Intelligent extraction of aquifer features:

[0094] Image segmentation of borehole column plots in borehole engineering data based on an improved U-Net++ network:

[0095]

[0096] in, The network architecture adds a geological prior constraint layer to force the identification of typical aquifer characteristics, such as sandstone porosity > 15%;

[0097] Geological texts are processed using an attention-enhanced BiLSTM-CRF model:

[0098]

[0099] in Vectors are embedded into the geological dictionary to identify key entities such as aquifers and impermeable layers;

[0100] Spatial interpolation of hydrological parameters: Constructing an adaptive anisotropic Kriging model:

[0101]

[0102] Weight Determined through an improved variogram:

[0103]

[0104] in Variable range in primary and secondary directions The strike angle of the geological structure;

[0105] Water supply risk assessment: Constructing a graph neural network assessment model:

[0106]

[0107] The node features include: aquifer thickness, hydraulic head, and distance from the aquifer; the edge weight is the hydraulic conductivity coefficient.

[0108] Supply risk index calculation:

[0109]

[0110] in, To replenish the flux, Darcy's law is used for calculation:

[0111]

[0112] Taking the hydrogeological assessment of a certain mining area as an example, the input data is as follows:

[0113] Borehole image: 4000×6000 resolution, containing 5 aquifer markers;

[0114] Text report: "A Permian sandstone aquifer with a thickness of 8.2m and a hydraulic head pressure of 2.3MPa is developed in the roof of coal seam No. 3".

[0115] Observational data: Hydraulic head values ​​from 12 hydrological wells (22.4-26.8m);

[0116] Implementation process:

[0117] The U-Net++ model segmented the aquifer region, and the average thickness was measured to be 8.15±0.3m, with an error of <5% compared to manual measurement.

[0118] The BiLSTM-CRF model identified the "sandstone aquifer" entity and its properties with an accuracy of 92.7%.

[0119] Kriging interpolation generates head pressure distribution maps, with varying range parameters. (along the coal seam strike) .

[0120] GNN model evaluation shows:

[0121] Node characteristics: aquifer thickness 8.2m (weight 0.4), hydraulic head 2.3MPa (0.3), distance 35m (0.2), permeability coefficient 1.2m / d (0.1);

[0122] Calculate supply flux Critical value ;

[0123] Output supply risk index (Threshold 0.8), judged as moderate recharge conditions. Field verification showed an actual inflow of 0.92 m³ / d, with a prediction error of 7.6%.

[0124] S500: Based on the three-dimensional parameter field, an integrated learning model is used to evaluate the stability probability of the aquitard; at the same time, a deep reinforcement learning environment with water pressure and stress as the state is constructed to simulate the dynamics of water pressure, and a time series model is used to predict the trend of water pressure change; by combining the delamination risk index, water source replenishment risk index, aquitard stability probability and water pressure prediction trend, a comprehensive risk warning information for delamination water hazards is generated.

[0125] In some embodiments, the ensemble learning model is an improved random forest algorithm that employs a dynamic sampling strategy to handle class imbalance; the deep reinforcement learning environment is trained using a proximal policy optimization algorithm, and its reward function is negatively correlated with the predicted water pressure error and the rate of change of water pressure; the time series model is an LSTM-Transformer model that integrates multi-scale features of discrete wavelet transform.

[0126] In some embodiments, the comprehensive risk warning information for delamination water hazards includes multi-level warning rules, which are as follows:

[0127] If both the delamination risk index and the water supply risk index exceed the first threshold, and the water pressure prediction trend is upward, then the first-level warning will be triggered.

[0128] If the probability of the stability of the aquitard is lower than the second threshold, and either the delamination risk index or the water supply risk index exceeds the third threshold, then a level 2 warning is triggered.

[0129] If the separation risk index exceeds the third threshold or the water pressure time series forecast shows abnormal fluctuations, a level 3 warning will be triggered.

[0130] The specific implementation process includes classifying the stability of the waterproofing layer:

[0131] Constructing an ensemble learning model based on feature engineering:

[0132] Input features include: thickness of the waterproof layer ,tensile strength Elastic modulus Water pressure gradient Crack development index ;

[0133] Using the improved random forest algorithm, the feature importance weights are calculated as follows:

[0134]

[0135] in For decision trees The set of split nodes, For nodes The splitting path is determined by setting a dynamic sampling strategy and weighting the samples that break through the class imbalance problem.

[0136] Dynamic simulation of water pressure:

[0137] Establish a deep reinforcement learning environment:

[0138] state space Includes current water pressure Permeation rate Stress field

[0139] Action space Adjust the simulation step size Boundary conditions

[0140] Reward function design:

[0141]

[0142] in This is the adjustment coefficient;

[0143] The agent is trained using the PPO algorithm, and the policy update formula is as follows:

[0144]

[0145] Time Series Early Warning: Constructing an LSTM-Transformer Hybrid Model

[0146]

[0147] Multi-scale feature fusion:

[0148]

[0149] DWT stands for Discrete Wavelet Transform Decomposition Layer.

[0150] Taking a water inrush warning at a certain working face as an example, the handling process is as follows:

[0151] 1. Input data preparation:

[0152] Waterproofing layer parameters: thickness 15.6m, tensile strength 2.4MPa, crack index 0.35;

[0153] Monitoring data sequence: Water pressure data for the past 72 hours [2.1, 2.3, 2.6, 2.9, 3.2] MPa;

[0154] Stress field data: Maximum principal stress 28.7 MPa, direction N45°E;

[0155] 2. Model calculation process:

[0156] Random forest model calculation:

[0157] Importance of features: thickness (0.32), tensile strength (0.28), water pressure gradient (0.25);

[0158] Output stability probability: 78.6% (threshold 70%);

[0159] Reinforcement learning simulation:

[0160] Initial state: P = 3.2 MPa ;

[0161] After 500 iterations, it is predicted that the pressure will reach 3.8 MPa after 6 hours.

[0162] LSTM-Transformer prediction:

[0163] Extracted periodic characteristics: 24-hour periodic amplitude 0.4 MPa;

[0164] Predicted 12-hour pressure change trend: 3.2→3.5→3.9MPa;

[0165] 3. Comprehensive Risk Assessment:

[0166] The warning engine aggregates all intermediate results: R layer =0.76 (high), R water =0.71 (medium), water-resistant layer stability probability 78.6% (basically stable), water pressure trend (accelerating upward).

[0167] Logical judgment is made based on preset multi-level early warning rules. This example meets the condition that "the separation risk index exceeds the threshold and the water pressure prediction trend is upward".

[0168] The final comprehensive disaster risk index is calculated to be R=0.65, and an orange (level 2 warning) warning report is generated.

[0169] The early warning report includes: risk location (85-92m section), core criteria (sufficient delamination development and accelerated rise in water pressure), trend prediction (water pressure will approach the critical level in the next 6-12 hours), and specific handling suggestions (intensified monitoring, strengthened patrols, and guaranteed drainage).

[0170] The report is automatically pushed to relevant management personnel through the system platform and mobile terminals, guiding them to take targeted prevention and control measures on site, thus forming a complete intelligent early warning closed loop from "risk identification" to "measure response".

[0171] Secondly, this application proposes an intelligent identification and early warning system for delamination water damage, such as... Figure 2As shown, it includes a three-dimensional geological model construction module for mines, a three-dimensional parameter field construction module, an intelligent judgment module for delamination space, a water supply condition judgment module, and a comprehensive judgment and early warning module for delamination water disaster conditions.

[0172] The mine three-dimensional geological model construction module is used to integrate drilling, geophysical exploration and mining engineering data of the target area to construct a three-dimensional geological model that includes strata, structure and mining space;

[0173] The three-dimensional parameter field construction module is used to collect the rock strata physical and mechanical parameters and hydrogeological parameters of the target area, and to generate a three-dimensional rock mass mechanical parameter field and hydrogeological parameter field by using an improved Kriging spatial interpolation algorithm that integrates geological structural constraints and rock strata attitude.

[0174] The delamination space intelligent determination module is used to determine the rock strata structure based on the three-dimensional geological model and the three-dimensional rock mass mechanical parameter field, using an improved random forest algorithm, and to calculate the deflection difference of the combined rock strata based on the multi-layer beam mechanical model; then, using a hybrid time series model and integrating microseismic monitoring data, it predicts the spatiotemporal location and risk index of delamination development.

[0175] The water supply condition determination module is used to automatically extract aquifer information from borehole engineering data based on an image recognition model, and to parse hydrological entities in geological text based on a natural language processing model; combined with the hydrological parameter field, a graph neural network model is used to evaluate the risk index of water supply from the aquifer to the separation space.

[0176] The comprehensive judgment and early warning module for delamination water disaster conditions is used to evaluate the stability probability of the aquitard based on the three-dimensional parameter field using an integrated learning model; at the same time, it constructs a deep reinforcement learning environment with water pressure and stress as states to simulate water pressure dynamics, and uses a time series model to predict the water pressure change trend; and generates comprehensive risk early warning information for delamination water disaster by combining the delamination risk index, water source replenishment risk index, aquitard stability probability and water pressure prediction trend.

[0177] In some embodiments, the system also includes a data acquisition and input module for accessing borehole data, geophysical data, mining engineering drawings, real-time microseismic monitoring data, and hydrological monitoring data; and an early warning information output module for visually displaying the three-dimensional model, parameter field, risk distribution, and early warning report.

[0178] The following describes a detailed embodiment of the technical solution of this invention using a complete application scenario. Application scenario overview: The target working face is the 15201 fully mechanized longwall face, mining the No. 3 coal seam. The working face is 1500m long, 200m wide, and has an average mining height of 2.8m. Geological exploration shows that the overlying strata sequence of the coal seam is: the immediate roof is a 2.5m thick mudstone, the main roof is a 5.8m thick sandstone, above which is an 8.2m thick sandstone aquifer (confined water), and further upwards is a 15.6m thick mudstone aquitard. A NW-trending normal fault F1 with a drop of approximately 12m is developed in the middle of the working face. This system will provide dynamic intelligent early warning of the risk of delamination water hazards during the mining process of this working face.

[0179] S1. Three-dimensional geological dynamic modeling of the mine:

[0180] 1. Data Input: The system receives and integrates three types of data:

[0181] Drilling data: Import the coordinates, depths, and boundary data of each rock layer (coal seam, mudstone, sandstone, etc.) of 50 geological boreholes around the working face.

[0182] Geophysical interpretation results: Spatial distribution model of fault F1 (polygon data) obtained by interpreting ground 3D seismic exploration data.

[0183] Mining Engineering Data: Import mining engineering plan and tunnel sketches that are dynamically updated according to the mining progress.

[0184] 2. Model building and visualization:

[0185] The system uses borehole data as hard control points and employs Kriging interpolation to generate spatial surfaces of multiple key strata, such as the top and bottom plates of the coal seam, the immediate top surface, and the basic top surface.

[0186] Fault F1 is introduced as a structural constraint, and the Delaunay triangulation algorithm with fault constraints is used in the 3D modeling engine to ensure that continuous ground planes are generated on both sides of the fault, which correctly reflects the faulting relationship of the strata.

[0187] Using the Marching Cubes voxelization algorithm, the above-mentioned surface is converted into a three-dimensional volume mesh model, and a three-dimensional geological body containing various strata entities and fault entities is constructed.

[0188] By overlaying mining data onto the geological body, the spatial location of mined-out areas and tunnel systems can be dynamically displayed.

[0189] Finally, a high-fidelity, arbitrarily rotatable, scalable, and cross-sectional 3D geological model was generated using OpenGL / WebGL technology, providing a unified spatiotemporal framework for subsequent analysis.

[0190] S2. Multi-source heterogeneous data fusion and intelligent construction of three-dimensional parametric fields;

[0191] 1. Parameter Acquisition and Correlation: The system collects and correlates two types of parameters to their corresponding positions in the S1 model:

[0192] Rock mass physical and mechanical parameters set: obtained from borehole rock sample test reports, including the thickness, elastic modulus, tensile strength, unit weight, fragmentation coefficient, and density of each rock layer. For example, the immediate top mudstone has an E value of 8.2 GPa, and the basic top sandstone has an E value of 15.6 GPa.

[0193] Hydrogeological parameter set: obtained from hydrological observation wells, including aquifer thickness, aquitard thickness, and confined head. In this example, the aquifer head is between +22.4 and +26.8 meters.

[0194] 2. Improved Spatial Interpolation Calculation: To generate an accurate three-dimensional parameter distribution field, the system calls the anisotropic Kriging interpolation algorithm and makes two key improvements:

[0195] Geological structural constraints: The algorithm identifies fault F1 and employs a piecewise interpolation strategy. When calculating the parameter values ​​of grid nodes, only sample point data located on the same side of the fault as the node are used, effectively characterizing the obstruction effect of the fault on the continuity of parameter space.

[0196] Stratigraphic attitude correction: The algorithm reads the average dip information of the strata. When calculating the weight of the sample points, the weights are corrected directionally (i.e., multiplied by the cosine of the angle) based on the angle between the line connecting the sample point and the point to be estimated and the dip of the strata, so that the interpolation results are more consistent with the original distribution pattern of the strata deposition.

[0197] 3. Parameter Field Generation: The system generates a dense, regular grid in three-dimensional space, performs the interpolation calculations described above on each grid node, and finally outputs a series of parameter data volumes, including the three-dimensional elastic modulus field, the three-dimensional tensile strength field, and the three-dimensional hydraulic head pressure field. These parameter fields provide a quantitative spatial data foundation for subsequent mechanical calculations and risk assessments.

[0198] S3. Intelligent Prediction of Delamination Space Development

[0199] 1. Intelligent classification of rock strata structure:

[0200] For the current mining location (e.g., 80m from the cut), the system extracts the rock strata parameter sequence of the vertical profile from the three-dimensional mechanical parameter field generated by S2, and inputs it into a random forest model with dynamic weighting of feature importance. This model can not only output the rock strata type (e.g., mudstone, sandstone), but also dynamically evaluate the importance weight of each parameter (e.g., brecciation coefficient, elastic modulus) to the classification results, thereby intelligently identifying the "soft-hard" rock strata combination that is prone to delamination (in this example, the immediate top of mudstone and the basic top of sandstone).

[0201] 2. Combined Rock Strata Mechanical Analysis: Based on the classification results and the precise mechanical parameters provided by S2, a multi-layer beam mechanical model was systematically constructed to simulate the bending deformation of combined rock strata under the influence of mining. An adaptive differential evolution algorithm was used to inversely optimize parameters that are difficult to measure directly, such as interlayer contact stiffness. The calculation results show that the deflection of the immediate top mudstone (32.7 mm) is significantly greater than that of the basic top sandstone (18.4 mm), with a deflection difference rate of 43.7%, confirming the high risk of delamination development from a mechanical perspective.

[0202] 3. Spatiotemporal dynamic prediction:

[0203] The system inputs the aforementioned deflection difference characteristics and lithological combination characteristics, along with real-time monitored microseismic event time-series data (frequency and energy over the past 10 days), into an LSTM-Transformer hybrid neural network model. This model introduces a geological structural feature vector representing fault F1, which is deeply fused with the time-series features through an attention mechanism.

[0204] 4. Risk Output:

[0205] The model predicts that within the next three days, when the working face advances to the 85-92m section, the delamination height will reach approximately 1.2m. The system comprehensively calculates the delamination risk index R for this area. layer The value is 0.76, exceeding the preset threshold of 0.6, thus triggering a yellow warning for delamination development, which is then highlighted in the 3D model.

[0206] S4. Multimodal Fusion Assessment of Water Supply Conditions

[0207] 1. Intelligent extraction of multi-source information:

[0208] Image recognition: The system calls the improved U-Net++ image segmentation neural network to automatically identify a large number of borehole columnar section scans, accurately segment the aquifer (sandstone) and impermeable (mudstone) layers, and quantify their thickness (the average thickness measured in this example is 8.15m).

[0209] Text understanding: The system uses an attention-enhanced BiLSTM-CRF natural language processing model to automatically parse the geological survey report text and accurately extract key entities and attributes such as "sandstone aquifer", "thickness 8.2m", and "hydraulic pressure 2.3MPa".

[0210] 2. Hydrogeological field construction: Using the improved spatial interpolation algorithm built in S2, the head data of the hydrological observation wells are interpolated to generate a three-dimensional head pressure distribution field in the study area.

[0211] 3. Supply Risk Graph Assessment: The system constructs a graph neural network model.

[0212] Mapping: Aquifers, the separation space predicted by S3, fault F1, etc., are defined as map nodes. Node characteristics include thickness, hydraulic head, distance from the separation layer, permeability coefficient, etc.

[0213] Definition of relationship: The edges between nodes represent potential hydraulic connections, and their weights are calculated based on Darcy's law, using both the permeability coefficient and the distance.

[0214] Assessment and Calculation: The model simulates the transmission of water pressure information between hydrogeological units through a message passing mechanism, comprehensively assessing the intensity of recharge to the delamination space. Finally, the water recharge risk index R is calculated. water The value is 0.71, which falls under the category of moderate supply conditions.

[0215] S5. Dynamic simulation and integrated early warning of disaster-causing conditions

[0216] 1. Stability assessment of the waterproof layer:

[0217] The system extracts parameters (thickness 15.6m, tensile strength 2.4MPa, current water pressure gradient) of the key aquitard (located between the aquifer and the segregation layer) and inputs them into a random forest ensemble model employing a dynamic sampling strategy. This model effectively addresses the uneven distribution of "stable" and "unstable" samples in the data, assessing that under current conditions, the stability probability of the aquitard is 78.6% (above the 70% safety threshold), and its current state is rated as "Level B: Basically Stable".

[0218] 2. Dynamic simulation and trend prediction of water pressure:

[0219] Reinforcement Learning Simulation: The system constructs a deep reinforcement learning virtual environment with the current delamination water pressure (3.2 MPa) and surrounding rock stress state as the state space. A proximal policy optimization algorithm is used to train the agent, whose reward function is designed to be negatively correlated with the predicted water pressure error and the rate of water pressure rise, driving the agent to learn and simulate the dynamic evolution of water pressure. The simulation predicts that the water pressure may rise to 3.8 MPa in the next 6 hours.

[0220] Accurate Time Series Prediction: The LSTM-Transformer time series prediction neural network model, which integrates the multi-scale analysis capabilities of discrete wavelet transform, performs in-depth analysis on real-time water pressure monitoring data from the past 72 hours, decomposing and identifying long-term trends, periodic fluctuations, and short-term anomalies. The model predicts that the water pressure will show an accelerating upward trend in the next 12 hours: 3.2→3.5→3.9MPa.

[0221] 3. Comprehensive early warning decision-making and issuance:

[0222] The warning engine aggregates all intermediate results: R layer =0.76 (high), Rwater =0.71 (medium), water-resistant layer stability probability 78.6% (basically stable), water pressure trend (accelerating upward).

[0223] Logical judgment is made based on the preset multi-level early warning rules (as described in claim 8). This example meets the condition that "the separation risk index exceeds the threshold and the water pressure prediction trend is upward".

[0224] The system calculates the final comprehensive disaster risk index to be 0.65 and generates an orange (level 2) warning report.

[0225] The early warning report includes: risk location (85-92m section), core criteria (sufficient delamination development and accelerated rise in water pressure), trend prediction (water pressure will approach the critical level in the next 6-12 hours), and specific handling suggestions (intensified monitoring, strengthened patrols, and guaranteed drainage).

[0226] The report is automatically pushed to relevant management personnel through the system platform and mobile terminals, guiding them to take targeted prevention and control measures on site, thus forming a complete intelligent early warning closed loop from "risk identification" to "measure response".

[0227] Thirdly, this application proposes an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0228] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described above.

[0229] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0230] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0231] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this disclosure.

[0232] In the embodiments provided in this disclosure, it should be understood that the disclosed apparatus / computer devices and methods can be implemented in other ways. For example, the apparatus / computer device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. Multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0233] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0234] Furthermore, the functional units in the various embodiments of this disclosure can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0235] If an integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program may include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. A computer-readable medium may include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in a computer-readable medium may be appropriately added to or subtracted according to the requirements of legislation and patent practice in a jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0236] The above are merely preferred embodiments of the present invention. It should be noted that any modifications and improvements made by those skilled in the art without departing from the present technical solution should also be considered to fall within the scope of protection claimed by the present solution.

Claims

1. A method for intelligent identification and early warning of delamination water damage, characterized in that, Includes the following steps: Integrate borehole, geophysical and mining engineering data of the target area to construct a three-dimensional geological model that includes strata, structure and mining space; The rock strata physical and mechanical parameters and hydrogeological parameters of the target area are collected. The improved Kriging spatial interpolation algorithm, which integrates geological structural constraints and rock strata attitude, is used to generate a three-dimensional rock mass mechanical parameter field and hydrogeological parameter field. The improved Kriging spatial interpolation algorithm performs piecewise interpolation by introducing fault influence factors and corrects the interpolation weights by rock strata dip angle. Its variogram model is an anisotropic model. Based on the aforementioned three-dimensional geological model and three-dimensional rock mass mechanical parameter field, an improved random forest algorithm is used to determine the rock strata structure, and the deflection difference of the combined rock strata is calculated based on a multi-layer beam mechanical model. Then, a hybrid time series model is used to integrate microseismic monitoring data to predict the spatiotemporal location and risk index of delamination development. The improved random forest algorithm adopts a dynamic weighting strategy for feature importance, and its weight calculation integrates the Gini importance and the partial derivative of the model accuracy with respect to that feature. The input of the hybrid time series model includes historical microseismic event sequences and geological structural feature vectors. Aquifer information is automatically extracted from borehole engineering data based on an image recognition model, and hydrological entities in geological text are parsed based on a natural language processing model. Combining the hydrological parameter field, a graph neural network model is used to evaluate the risk index of water recharge from the aquifer to the separated layer space. The image recognition model is a U-Net++ network with added geological prior constraints; the natural language processing model is an attention-enhanced BiLSTM-CRF model; the node features of the graph neural network model include aquifer thickness, hydraulic head, distance from the separated layer, and permeability coefficient. Based on the three-dimensional parameter field, an ensemble learning model is used to evaluate the stability probability of the aquitard. Simultaneously, a deep reinforcement learning environment is constructed with water pressure and stress as states to simulate water pressure dynamics, and a time series model is used to predict water pressure change trends. By integrating the delamination risk index, water supply risk index, aquitard stability probability, and water pressure prediction trends, comprehensive risk warning information for delamination water hazards is generated. The ensemble learning model is an improved random forest algorithm that employs a dynamic sampling strategy to handle class imbalance. The deep reinforcement learning environment is trained using a proximal policy optimization algorithm, and its reward function is negatively correlated with the predicted water pressure error and the rate of water pressure change. The time series model is an LSTM-Transformer model that integrates multi-scale features of discrete wavelet transform.

2. The method according to claim 1, characterized in that: The comprehensive risk early warning information for delamination water hazards includes multi-level early warning rules, which are as follows: If both the delamination risk index and the water supply risk index exceed the first threshold, and the water pressure prediction trend is upward, then the first-level warning will be triggered. If the probability of the stability of the aquitard is lower than the second threshold, and either the delamination risk index or the water supply risk index exceeds the third threshold, then a level 2 warning is triggered. If the separation risk index exceeds the third threshold or the water pressure time series forecast shows abnormal fluctuations, a level 3 warning will be triggered.

3. A smart identification and early warning system for delamination water damage, characterized in that: It includes a three-dimensional geological model construction module for mines, a three-dimensional parameter field construction module, an intelligent judgment module for delamination space, a water supply condition judgment module, and a comprehensive judgment and early warning module for delamination water disaster conditions; The mine three-dimensional geological model construction module is used to integrate drilling, geophysical exploration and mining engineering data of the target area to construct a three-dimensional geological model that includes strata, structure and mining space; The three-dimensional parameter field construction module is used to collect the physical and mechanical parameters and hydrogeological parameters of the rock strata in the target area. It uses an improved Kriging spatial interpolation algorithm that integrates geological structural constraints and rock strata attitude to generate a three-dimensional rock mass mechanical parameter field and hydrogeological parameter field. The improved Kriging spatial interpolation algorithm performs piecewise interpolation by introducing fault influence factors and corrects the interpolation weights by rock strata dip angle. Its variogram model is an anisotropic model. The delamination spatial intelligent determination module is used to determine the rock strata structure based on the three-dimensional geological model and the three-dimensional rock mass mechanical parameter field using an improved random forest algorithm, and to calculate the deflection difference of the combined rock strata based on a multi-layer beam mechanical model. Then, using a hybrid time series model, it integrates microseismic monitoring data to predict the spatiotemporal location and risk index of delamination development. The improved random forest algorithm adopts a dynamic weighting strategy for feature importance, and its weight calculation integrates the Gini importance and the partial derivative of the model accuracy with respect to that feature. The input of the hybrid time series model includes historical microseismic event sequences and geological structural feature vectors. The water supply condition determination module is used to automatically extract aquifer information from borehole engineering data based on an image recognition model, and to parse hydrological entities in geological text based on a natural language processing model. Combined with the hydrological parameter field, a graph neural network model is used to evaluate the water supply risk index of the aquifer to the separation space. The image recognition model is a U-Net++ network with added geological prior constraints; the natural language processing model is an attention-enhanced BiLSTM-CRF model; the node features of the graph neural network model include aquifer thickness, hydraulic head, distance from the separation layer, and permeability coefficient. The comprehensive judgment and early warning module for delamination water disaster conditions is used to evaluate the stability probability of the aquitard based on the three-dimensional parameter field using an ensemble learning model. Simultaneously, a deep reinforcement learning environment simulating water pressure dynamics is constructed, with water pressure and stress as states, and a time series model is used to predict water pressure change trends. By integrating the delamination risk index, water supply risk index, aquitard stability probability, and water pressure prediction trends, comprehensive risk early warning information for delamination water hazards is generated. The ensemble learning model is an improved random forest algorithm using a dynamic sampling strategy to handle class imbalance. The deep reinforcement learning environment is trained using a near-end policy optimization algorithm, and its reward function is negatively correlated with the predicted water pressure error and the rate of water pressure change. The time series model is an LSTM-Transformer model that integrates multi-scale features of discrete wavelet transform.

4. The system according to claim 3, characterized in that: It also includes a data acquisition and input module for accessing borehole data, geophysical data, mining engineering drawings, real-time microseismic monitoring data, and hydrological monitoring data; It also includes an early warning information output module, which is used to visualize and display the 3D model, parameter field, risk distribution, and early warning report.

5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1-2.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the method as described in any one of claims 1-2.

Images