CN-121982234-A - Intelligent recognition and early warning method and system for water disaster of separation layer

CN121982234ACN 121982234 ACN121982234 ACN 121982234ACN-121982234-A

Abstract

The invention discloses an intelligent recognition early warning method and system for water damage of a separation layer, which belong to the field of intelligent early warning of mine safety and geological disasters, wherein discrete, heterogeneous geological, mechanical and hydrologic data are integrally converted into continuous and visual three-dimensional parameter fields by constructing a dynamically updated high-precision three-dimensional geological model and adopting an intelligent interpolation algorithm fusing geological structures and rock stratum shapes, the defects of isolated data and single-sided representation of the traditional method are overcome, various machine learning and deep learning algorithms are comprehensively utilized, intelligent judgment and simulation are respectively carried out on key links such as separation layer development, water source replenishment, water barrier stability and the like, the dependence of artificial experience is reduced, the accuracy of separation layer space positioning, water source evaluation and disaster trend prediction is greatly improved, the dynamic evolution of water pressure is simulated through deep reinforcement learning, the future trend is predicted by combining a multi-scale time sequence model, early signals of disaster inoculation are dynamically captured, and the timeliness and the early warning performance are remarkably improved.

Inventors

JIAO JINBAO
GAO LINSHENG
WANG XIAOPING
Gao Tianwa
ZHANG CHENGJIAN
HE HAIPING
CHEN HONGYU
HUA RUI
GAO WENXING
WANG QIONG

Assignees

中煤新集能源股份有限公司
华北科技学院(中国煤矿安全技术培训中心)

Dates

Publication Date: 20260505
Application Date: 20260119

Claims (10)

1. An intelligent recognition and early warning method for water damage of a separation layer is characterized by comprising the following steps: Integrating drilling, geophysical prospecting and mining engineering data of a target area, and constructing a three-dimensional geological model comprising stratum, structure and mining space; Acquiring rock stratum physical mechanical parameters and hydrogeologic parameters of the target area, and generating a three-dimensional rock stratum mechanical parameter field and a hydrogeologic parameter field by utilizing an improved Kriging spatial interpolation algorithm fusing geologic structure constraint and rock stratum attitude; Judging a rock stratum structure by adopting an improved random forest algorithm based on the three-dimensional geological model and the three-dimensional rock mass mechanical parameter field, and calculating the deflection difference of the combined rock stratum based on the multi-layer beam mechanical model; The aquifer information in the drilling engineering data is automatically extracted based on the image recognition model, and hydrologic entities in the geological text are analyzed based on the natural language processing model; The method comprises the steps of estimating stability probability of a water-resisting layer by adopting an integrated learning model based on the three-dimensional parameter field, constructing a deep reinforcement learning environment simulation water pressure dynamic state taking water pressure and stress as states, predicting a water pressure change trend by adopting a time sequence model, and generating comprehensive risk early warning information of water damage of the water-resisting layer by integrating the water-resisting layer risk index, the water source supply risk index, the stability probability of the water-resisting layer and the water pressure prediction trend.
2. The method of claim 1, wherein the modified Kriging spatial interpolation algorithm performs piecewise interpolation by introducing fault influencing factors and corrects the interpolation weights by formation dip angles, and the variation function model is an anisotropic model.
3. The method of claim 2, wherein the improved random forest algorithm adopts a feature importance dynamic weighting strategy, the weight calculation of the improved random forest algorithm fuses the importance of the base and the partial derivative of the model accuracy to the feature, and the input of the hybrid time sequence model comprises a historical microseismic event sequence and a geological structure feature vector.
4. The method of claim 3, wherein the image recognition model is a U-Net++ network with a geological prior constraint layer added, the natural language processing model is a BiLSTM-CRF model with enhanced attention, and the node characteristics of the graphic neural network model comprise aquifer thickness, water head value, separation distance and permeability coefficient.
5. The method of claim 4, wherein the ensemble learning model is an improved random forest algorithm employing a dynamic sampling strategy to handle class imbalance, the deep reinforcement learning environment is trained using a near-end strategy optimization algorithm, and the reward function is inversely related to predicted hydraulic error and hydraulic rate of change, and the time series model is an LSTM-transducer model incorporating discrete wavelet transform multi-scale features.
6. The method of claim 5, wherein the comprehensive risk of water damage pre-warning information comprises a multi-stage pre-warning rule, wherein the multi-stage pre-warning rule is: if the separation layer risk index and the water source replenishment risk index both exceed a first threshold value, and the water pressure prediction trend is rising, triggering a first-stage early warning; if the stability probability of the water-resisting layer is lower than a second threshold value, and any one of the separation layer risk index and the water source replenishment risk index exceeds a third threshold value, triggering a second-stage early warning; and if the separation layer risk index exceeds a third threshold value or the water pressure time sequence prediction is abnormal fluctuation, triggering a third-stage early warning.
7. The intelligent recognition early warning system for the water disaster of the separation layer is characterized by comprising a mine three-dimensional geological model construction module, a three-dimensional parameter field construction module, an intelligent separation layer space judgment module, a water source supply condition judgment module and a comprehensive judgment and early warning module for the water disaster of the separation layer; The mine three-dimensional geological model construction module is used for integrating drilling, geophysical prospecting and mining engineering data of a target area and constructing a three-dimensional geological model comprising stratum, structure and mining space; The three-dimensional parameter field construction module is used for acquiring rock stratum physical mechanical parameters and hydrogeologic parameters of the target area and generating a three-dimensional rock mechanical parameter field and a hydrogeologic parameter field by utilizing an improved Kriging spatial interpolation algorithm for fusing geologic structure constraints and rock stratum production; The separation layer space intelligent judging module is used for judging a rock stratum structure by adopting an improved random forest algorithm based on the three-dimensional geological model and the three-dimensional rock mechanical parameter field, calculating the deflection difference of the combined rock stratum based on the multi-layer beam mechanical model, and then predicting the space-time position and the risk index of separation layer development by utilizing a mixed time sequence model and fusing microseismic monitoring data; The water source replenishment condition judging module is used for automatically extracting aquifer information in drilling engineering data based on an image recognition model, analyzing hydrologic entities in geological texts based on a natural language processing model, and evaluating a water source replenishment risk index of the aquifer on the separation layer space by utilizing a graph neural network model in combination with the hydrologic parameter field; The comprehensive judging and early warning module is used for estimating the stability probability of the water-resisting layer by adopting an integrated learning model based on the three-dimensional parameter field, constructing a deep reinforcement learning environment with water pressure and stress as states to simulate the water pressure dynamic state, adopting a time sequence model to predict the water pressure change trend, and synthesizing the separation risk index, the water source replenishment risk index, the stability probability of the water-resisting layer and the water pressure prediction trend to generate comprehensive risk early warning information of the separation water damage.
8. The system of claim 7, further comprising a data acquisition and input module for accessing borehole data, geophysical prospecting data, mining engineering drawings, real-time microseismic monitoring data, and hydrological monitoring data, and an early warning information output module for visually displaying a three-dimensional model, a parameter field, risk distribution, and an early warning report.
9. An electronic device comprising a memory, a processor and a computer program stored in the memory and running on the processor, characterized in that the processor implements the steps of the method according to any of claims 1-6 when executing the computer program.
10. A computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program which, when executed by a processor, implements the steps of the method according to any of claims 1-6.

Description

Intelligent recognition and early warning method and system for water disaster of separation layer Technical Field The invention belongs to the field of mine safety and geological disaster intelligent early warning, and particularly relates to an intelligent recognition early warning method and system for water disaster of a separation layer. Background In the coal mining process, a goaf formed after coal seam mining leads to the original stress balance of an overlying strata to be destroyed, and movement, deformation and fracture are generated. Because of the remarkable differences of the physical and mechanical properties such as thickness, strength, elastic modulus and the like of different rock layers, the different rock layers can generate asynchronous deformation in the bending and sinking process, and further a separation space, namely a separation layer, is formed between the layers. When the separation layer space grows below the water-rich aquifer and is communicated with a water source, high-pressure water is extremely easy to accumulate to form separation layer water. Under the action of mine pressure, water pressure or mining disturbance, the delamination water body can suddenly collapse into the well, so that disastrous delamination water damage is caused. The water hazard has the characteristics of strong concealment, rapid water burst, large water quantity and long duration, and is one of the important security threats facing deep coal exploitation. At present, identification and early warning for water damage of a separation layer mainly depend on the following technical paths, but obvious limitations exist: The method is based on an empirical threshold method of single monitoring data, and the single type of data is mainly obtained by arranging a drilling water level gauge, a stress gauge or a microseismic monitoring system. And when the monitored value exceeds a fixed threshold set based on historical experience, an alarm is sent out. The method has the defects of serious hysteresis, and incapability of realizing advanced early warning because disasters are always in a development and even outbreak stage during alarming. The false alarm rate is high, the threshold value setting depends on subjective experience, is difficult to adapt to complex and changeable geological conditions, and is easy to be interfered by accidental factors. The cause association is weak, namely that the abnormality of only a voucher index cannot accurately judge whether the abnormality is caused by water in a separation layer or caused by other geological factors, and the deep association analysis on the disaster formation mechanism is lacking. The analysis calculation method based on the classical theory model mainly applies the theory of 'upper three bands' (collapse band, fracture band and bending sinking band) and the mechanical model of the thin plate or beam, and calculates the height of the water-guiding fracture band by inputting limited rock stratum parameters so as to infer the possible development position of the separation layer. The method has the limitation that the model is highly simplified, namely, a complex three-dimensional geologic body is simplified into a homogeneous and isotropic ideal model, and the influence of complex structures such as faults, folds and the like and lithologic space variability cannot be considered. The parameters are difficult to take value, the key mechanical parameters required for calculation are usually from laboratory tests of a small number of drilling holes, the actual condition of the whole working surface is difficult to represent, and the reliability of the calculation result is poor. Static analysis, namely, the process of dynamic development, closure and water migration and accumulation of a separation layer in the exploitation process cannot be simulated. The method tries to simply superimpose various monitoring data and partial geological data, and carries out risk discrimination based on expert experience library or simple logic rules. Although advanced over the former two methods, there are fundamental problems in that knowledge acquisition bottlenecks are present, that expert experience is difficult to translate into computer rules in a complete and formal manner, and that updating and maintenance are difficult. The fusion layer is shallow, namely, only the simple collection of the data layers is achieved, and the depth fusion and coupling analysis of the geological structure, the rock parameters, the hydrologic conditions and the dynamic monitoring data under the unified three-dimensional space model can not be achieved. The self-adaptive capacity is poor, the model can not be automatically learned and optimized from mass data, and a new mining area and a new condition can not be met. Disclosure of Invention Aiming at the defects of the prior art, the application provides an intelligent recognition and early warning method and system for water damage o