Automatic geological disaster monitoring device and monitoring method

Abstract

The invention belongs to the technical field of geological disaster monitoring, and discloses an automatic geological disaster monitoring device and monitoring method. The method comprises the following steps: completing joint monitoring of surface goaf deformation on the ground; for a bent sinking zone, a fissure zone and a caving zone of the goaf, an electrical method system, a micro-seismic monitoring system and a distributed optical fiber sensor are arranged underground, dynamic tracking and monitoring of underground resistivity, fissure zone expansion, caving zone rock mass breaking change and the boundary of the goaf are conducted respectively, and delineation of underground abnormal bodies is completed; and cooperative monitoring: completing cooperative dynamic coupling monitoring of deformation processes of the curved sinking zone, the fissure zone and the caving zone of the goaf. According to the method, the accurate delineation of the underground abnormal body is completed; and dynamic coupling monitoring of ground and underground deformation of the coal mine goaf is realized.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of geological disaster monitoring, and particularly relates to a geological disaster automatic monitoring device and a monitoring method. BACKGROUND

[0002] The field of geological disaster analysis is a cross-disciplinary field that focuses on identifying, assessing, and predicting the likelihood of geological disasters. This field combines knowledge and methods from geology, geophysics, geographic information systems (GIS), remote sensing technology, computer science, statistics, and engineering.

[0003] The goal of the field of geological disaster analysis is to improve understanding of geological disasters, reduce the risk of disasters, and effectively respond and rescue in the event of a disaster. This field is of great importance to the safety of people's lives and property and the promotion of social sustainable development.

[0004] Existing risk assessment methods may only focus on a single factor or a local area, lacking a comprehensive consideration of the entire mine system. In the event of an anomaly in the mine, how to quickly and effectively evacuate personnel is a challenge. In extreme cases, such as when all exits cannot be safely escaped, how to coordinate multiple emergency measures (such as drilling ventilation and excavation rescue) is a complex problem. At the same time, drilling ventilation and excavation rescue are common emergency measures when an anomaly occurs in a mine. However, improper operation may cause secondary collapse, increasing the difficulty and risk of rescue.

[0005] To solve the above problems, the invention patent "Coal mine geological disaster monitoring and early warning method and device" (publication number CN118629165A, publication date 20240910) discloses a highly integrated and intelligent system design that not only ensures real-time monitoring and rapid response of the mine state, but also forms a closely connected emergency management network through comprehensive risk assessment and intelligent evacuation path planning. The construction of this network enables the system to quickly initiate multiple emergency measures, such as identifying safe areas, providing route guidance, drilling ventilation, and excavation rescue, when facing anomalies in the mine.

[0006] For example, the invention patent application "Geological disaster identification method based on multi-source data fusion" (publication number CN110765934B, publication date 20210219) uses InSAR technology to screen out locations with large surface deformation, i.e. potential high-risk areas of geological disasters; then conducts remote sensing investigation of geological disasters in potential high-risk areas of geological disasters to determine high-risk areas of geological disasters; and uses unmanned aerial vehicles to take aerial photographs of high-risk areas of geological disasters to obtain accurate locations of geological disasters. This invention uses a multi-level, multi-method, and multi-precision multi-element data fusion method to provide a large amount of real and reliable data for the selection of geological disaster area management methods and the development trend of disasters.

[0007] Through the above analysis, the problems and defects of the prior art are that the prior art does not combine the coupling monitoring of the deformation of the coal mine goaf on the ground and underground, and the prior art basically monitors the geological disaster high-incidence area, but does not effectively monitor the stress change of the underground resistivity and microseism, and the accuracy of the delineation of the underground abnormal body is poor. Therefore, the collaborative dynamic coupling monitoring effect of the coal mine goaf is poor. SUMMARY

[0008] In order to overcome the problems in the related art, the present application discloses a geological disaster automatic monitoring device and a monitoring method.

[0009] The technical solution is as follows: a geological disaster automatic monitoring method, comprising the following steps:

[0010] S1, for the coal mine goaf, InSAR goaf ground deformation inversion processing is performed by radar satellite InSAR for consecutive months, combined with the on-site location deployment of the ground surface Beidou GNSS, the influence of atmospheric delay error is eliminated, and the ground surface goaf deformation joint monitoring is completed;

[0011] S2, for the curved subsidence zone, fracture zone and caving zone of the goaf, an electrical method system, a microseismic monitoring system and a distributed optical fiber sensor are respectively arranged underground to realize the monitoring of the underground resistivity, fracture zone expansion and caving zone rock mass fracture change, and the dynamic tracking monitoring of the goaf boundary, and the delineation of the underground abnormal body is completed;

[0012] S3, according to the joint monitoring results of the ground surface goaf deformation, combined with the collaborative monitoring results of the delineation of the underground abnormal body, the collaborative dynamic coupling monitoring of the deformation process of the curved subsidence zone, fracture zone and caving zone of the goaf is completed.

[0013] In step S2, the underground abnormal body includes: abnormal fracture zone expansion range, caving zone rock mass fracture degree, dynamic tracking of abnormal curved subsidence zone and goaf boundary, and abnormal underground goaf resistivity;

[0014] The monitoring of the underground resistivity, fracture zone expansion and caving zone rock mass fracture change comprises: arranging a sensor network underground, using a fracture zone expansion identification method to capture rock layer fracture signals in real time, delineating the fracture zone expansion range, and combining a microseismic waveform inversion method to quantify the caving zone rock mass fracture degree.

[0015] Further, using the fracture zone expansion identification method to capture rock layer fracture signals in real time and delineate the fracture zone expansion range comprises:

[0016] S201, sub-region division is performed on the fracture zone extension range saliency image, and a long short-term memory neural network is used to extract multi-scale deep rock fracture visualization features of the fracture zone extension range saliency image;

[0017] S202, the extracted multi-scale deep rock fracture visualization feature vector is input into a pre-trained fracture zone extension range saliency normalization model, the saliency scores of each sub-region of the fracture zone extension range saliency image are normalized, and the original fracture zone extension range saliency image is weighted using the fracture zone extension range global saliency image;

[0018] S203, a fracture zone extension displacement attribute database is established and used as a fracture zone extension displacement attribute category to detect the fracture zone extension displacement attribute of each sub-region of the fracture zone extension range saliency image, and a saliency fracture zone extension displacement feature is used to initialize a network;

[0019] S204, the fracture zone extension displacement attribute is used to weight the fracture zone extension range saliency image feature;

[0020] S205, a long short-term memory neural network is used to decode the saliency fracture zone extension displacement attribute feature to generate a fracture zone extension range saliency image description.

[0021] Step S202 specifically includes:

[0022] (a) Pre-training model: the fracture zone extension range saliency normalization model is a neural network composed of two fully connected layers and an output layer, more than 80% of the pixel points in a sub-region of the fracture zone extension range saliency image have the same saliency label, the sub-region is selected as a training sample and the saliency score is set to 1 as a whole, otherwise it is set to 0;

[0023] (b) input all sub-regions of the fracture zone extension range saliency image into the trained fracture zone extension range saliency normalization model to obtain multiple saliency maps at multiple segmentation levels, and obtain a saliency map D after weighting and averaging the saliency maps at each segmentation scale smap , the original fracture zone extension range saliency image D is weighted with a fusion parameter f, and the expression is:

[0024] D vis = (1-f) x D smap +f x D

[0025] The obtained fracture zone extension range saliency image D vis is used as the input of an end-to-end fracture zone extension range saliency image description model for subsequent training and testing.

[0026] Step S203 specifically includes:

[0027] (1) Statistics are made on all descriptions of fracture zone expansion displacement distance in the training set of fracture zone expansion range database, multiple data with abnormal distance are selected, and a fracture zone expansion displacement attribute database is established; 92% of the data in the training set of fracture zone expansion range data are in the fracture zone expansion displacement attribute database, which contains the current fracture zone expansion position and the next fracture zone expansion displacement position; based on multiple attributes in the fracture zone expansion displacement attribute database, the fracture zone expansion range saliency image D vis is predicted for attribute prediction;

[0028] (2) The fracture zone expansion range saliency image D vis is adjusted to a square of 128x128 pixels and input into the fracture zone expansion displacement attribute detection network, and a 6x6 pixel size multi-dimensional coarse spatial response map L8 is generated, each point in the spatial response map L8 is directly convolved on D vis .

[0029] Step S205 specifically includes:

[0030] According to the threshold b, the top N attributes with high probability ranking are selected {g 1 ,g 2 …g N}, and their respective positions on the spatial response map L8 are found; from the spatial response map L7 to the spatial response map L8 layer, the dimension transformation mapping of the fracture zone expansion range saliency image feature is selected, the mapping weight connected with the spatial response map of {g 1 ,g 2 …g N} is selected, and the importance vector h of the same dimension is obtained by adding up in each dimension of the spatial response map L7; after taking the average of each position of the fracture zone expansion range saliency image feature, the importance weighting in each dimension is performed, and the expression is:

[0031] D vis-attr =h⊙L7

[0032] In the formula, D vis-attr is the fracture zone expansion displacement feature, L7 is the spatial response map, h is the vector, and is the Hadamard product;

[0033] The weighted fracture zone expansion displacement feature D vis-attr is input into the subsequent long short-term memory neural network to generate the fracture zone expansion range description of the rock stratum rupture.

[0034] In step S3, the delineation result of the downhole abnormal body is combined with the monitoring to complete the cooperative dynamic coupling monitoring of the deformation process of the curved subsidence zone, the fracture zone and the caving zone, including:

[0035] Step one, obtain the curved subsidence zone and goaf boundary measurement data, the light irradiation of the distributed optical fiber sensing obtains the measurement data of the curved subsidence zone and goaf boundary target, and the light flux density γ of the stress redistribution of the rock stratum is obtained m ;

[0036] Step two, obtain the curved subsidence zone and goaf boundary profile structure information of the reconstructed object, and obtain the optical fiber optical characteristic parameters

[0037] Step three, use the imaging software to discretize the imaging target, and obtain the curved subsidence zone and goaf boundary grid structure information

[0038] Step four, based on the light transmission model and the finite element theory, the curved subsidence zone and goaf boundary profile structure information and the optical fiber optical characteristic parameters of the reconstructed target are taken as prior information, the surface finite angle curved subsidence zone and goaf boundary data and the linear relationship of the rock stratum stress redistribution of the internal curved subsidence zone and goaf boundary target distribution of the reconstructed target are established

[0039] Step five, the rock stratum stress redistribution linear relationship is converted into a 1 / 2 norm minimization problem, and the expression is:

[0040]

[0041] In the formula, η is a regularization parameter, G is a system matrix, E is a three-dimensional stress distribution of the curved subsidence zone and goaf boundary target to be solved, γ m is the light flux density of the rock stratum stress redistribution, |||2 is the absolute value of the 2 norm, and ||| 1 / 2 is the absolute value of the 1 / 2 norm

[0042] Step six, the rock stratum stress redistribution linear relationship is solved by using a half-threshold iteration technique, a half-threshold operator is introduced, and the expression is:

[0043]

[0044] The half-threshold iteration technique is represented as:

[0045]

[0046] F(E)=E+G T (γ-GE)

[0047] In the formula, O η,1 / 2 (·) is a half-threshold iteration function, C is a rock stratum stress value, η is a regularization parameter, is a downward recursion, F(E) is a three-dimensional stress distribution linear value of the curved subsidence zone and goaf boundary target to be solved, and [F(E)] iis the linear set of the three-dimensional stress distribution of the curved subsidence zone and the goaf boundary target of the i-th to be solved, γ is the light flux density, and T is the transposed matrix;

[0048] Step seven, the dynamic tracking method of introducing the goaf boundary reduces the iteration number, and F(E)≈E, and each iteration has the following process:

[0049] Z n+1 =sup(p n+1,1 ),p n+1,1 =O(F(E n ))

[0050]

[0051] In the formula, Z n+1 is the goaf boundary value of n+1 iteration, sup() is a dynamic tracking function, p n+1,1 is the node value of the dynamic tracking of the goaf boundary between n+1 iteration and the first iteration, O() is an iteration result function, F() is a linear function of the three-dimensional stress distribution of the curved subsidence zone and the goaf boundary target to be solved, E n is the current distribution of the three-dimensional stress distribution of the curved subsidence zone and the goaf boundary target to be solved after n iterations, E n+1 is the current distribution of the three-dimensional stress distribution of the curved subsidence zone and the goaf boundary target to be solved after n+1 iterations, p n+1,k+1 is the node trajectory of the dynamic tracking of the goaf boundary between n+1 iteration and the k+1 iteration, p n+1,k is the node trajectory of the dynamic tracking of the goaf boundary between n+1 iteration and the k iteration, t n+1,k is the time used for n+1 iteration and the k iteration;

[0052] Step eight, through the iteration process, the reconstruction result E n of the n-th step is obtained, when ||γ-GE n || / ||γ||≤1e -5 , or ||E n -GE n || / ||γ||≤1e -8 , the iteration is stopped; wherein, e is the Reynolds index;

[0053] Step nine, the imaging software is used to disperse the imaging target, and a grid with a finer grid size is obtained;

[0054] Step ten, the reconstruction result and the curved subsidence zone and the goaf boundary profile structure of the imaging target are image fused, and the result is displayed.

[0055] Step two specifically comprises:

[0056] (2.1) reconstructing the curved subsidence zone and the goaf boundary profile structure information of the object; using the man-machine interactive semi-automatic segmentation method in the 3DMED software, the curved subsidence zone and the goaf are segmented and organized to obtain the curved subsidence zone and the goaf boundary profile structure of the imaging target;

[0057] (2.2) obtaining the optical characteristic parameters; based on the region-based diffuse optical tomography, the optical characteristic parameters of the stress nodes of each rock layer in the curved subsidence zone and the goaf boundary of the imaging target are obtained.

[0058] Step four specifically comprises:

[0059] (4.1) a light transmission model, using a diffusion approximation equation to describe the transmission process of light in the imaging target;

[0060] (4.2) according to the finite element theory, combining the curved subsidence zone and the goaf boundary profile structure information of the reconstructed target and the optical characteristic parameters, the diffusion approximation equation is discretized, the linear relationship between the measurement data of the surface and the distribution of the curved subsidence zone and the goaf boundary target in the internal of the reconstructed target is constructed, and the expression is:

[0061] γm=GE.

[0062] Another purpose of the present application is to provide a geological disaster automatic monitoring device, which implements the geological disaster automatic monitoring method, and the device comprises:

[0063] The uphole surface goaf deformation joint monitoring module is used for the coal mine goaf, uses radar satellite InSAR to perform InSAR goaf surface deformation inversion processing for continuous months, combines the on-site location deployment of the uphole surface Beidou GNSS, eliminates the influence of the atmospheric delay error, and completes the uphole surface goaf deformation joint monitoring;

[0064] The downhole abnormal body delineation module is used for the curved subsidence zone, the fractured zone and the caving zone of the goaf, respectively sets the electrical method system, the microseismic monitoring system and the distributed optical fiber sensing in the downhole, realizes the downhole resistivity, the fractured zone expansion and the caving zone rock mass crushing change monitoring, and the dynamic tracking monitoring of the goaf boundary, and completes the delineation of the downhole abnormal body;

[0065] The dynamic coupling monitoring module is used for the curved subsidence zone, the fractured zone and the caving zone of the goaf, respectively sets the electrical method system, the microseismic monitoring system and the distributed optical fiber sensing in the downhole, realizes the downhole resistivity, the fractured zone expansion and the caving zone rock mass crushing change monitoring, and the dynamic tracking monitoring of the goaf boundary, and completes the delineation of the downhole abnormal body;

[0066] In combination with all the technical solutions described above, the present application has the following beneficial effects:

[0067] The present invention sets up electrical systems, microseismic monitoring systems and distributed optical fiber sensors in the curved and subsided zones, fracture zones and caving zones of the goaf respectively, and conducts dynamic tracking monitoring of the downhole resistivity, the expansion of the fracture zone and the rock fragmentation changes in the caving zone, as well as the boundaries of the goaf, to complete the accurate delineation of abnormal bodies underground.

[0068] This study targets the goaf of the 5-20303 working face of a certain temple coal mine. This method employs combined downhole resistivity and microseismic monitoring within a 350m wide and 1000m long area. This system, combined with surface GNSS and radar satellite InSAR technology, dynamically couples the deformation processes of three zones above the goaf (the curved subsidence zone, the fracture zone, and the caving zone). Leveraging the temporal and spatial complementarity and process synergy of these multiple methods, this method achieves dynamic coupled monitoring of both aboveground and underground deformation within the goaf. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure;

[0070] Figure 1 1 is a diagram of a method for automatic monitoring of geological disasters provided by an embodiment of the present invention;

[0071] Figure 2 This is a flow chart of using a fracture zone expansion identification method provided by an embodiment of the present invention to capture rock fracture signals in real time and accurately delineate the fracture zone expansion range. DETAILED DESCRIPTION

[0072] To make the above-mentioned objects, features, and advantages of the present invention more readily apparent, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings. The following description sets forth numerous specific details to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways than those described herein, and those skilled in the art may make similar modifications without departing from the scope of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0073] Example 1, as Figure 1 As shown, the automatic geological disaster monitoring method provided by the embodiment of the present invention includes:

[0074] S1, for coal mine goaf, uses radar satellite InSAR to conduct InSAR goaf surface deformation inversion processing for several consecutive months. Combined with the deployment of BeiDou GNSS on the surface of the mine, it eliminates the influence of atmospheric delay error and completes the joint monitoring of goaf deformation on the mine surface.

[0075] It can be understood that the InSAR goaf surface deformation inversion processing of radar satellite InSAR for continuous months can adopt the relatively classic SBAS-InSAR (small baseline subset interference); by selecting a plurality of groups of SAR image pairs with short space-time baselines, the decorrelation problem is suppressed; combined with singular value decomposition (SVD) to invert the deformation time series. Successfully applied to monthly deformation tracking of the upper three zones of the goaf (curved subsidence zone, crack zone and caving zone).

[0076] In the process of eliminating the influence of atmospheric delay error on the surface of the well, atmospheric delay is mainly divided into two categories: troposphere and ionosphere. The classic solution of ionosphere delay focuses on double-frequency correction method, model method and difference method;

[0077] Tropospheric delay is caused by neutral atmospheric refraction, which is non-dispersive (irrelevant to frequency), and can be calculated by the following classic models or measured data; including:

[0078] Empirical model method: Hopfield / Saastamoinen model: Based on the elevation, pressure and temperature of the station, the zenith delay formula is established, and the elevation mapping is realized through the mapping function (such as GMF, NMF).

[0079] Simplified model (such as UNB3m): Input station position and annual accumulated day, output meteorological parameter estimated delay, suitable for the scene without meteorological data.

[0080] Meteorological data fusion method: GPT model (Global Pressure and Temperature): Global gridded meteorological model, provides temperature, pressure, combined with UNB3m to calculate water vapor pressure, improves the precision of dry / wet delay separation.

[0081] Numerical weather prediction (NWP) data: Fusion of high-resolution meteorological data to optimize wet component modeling.

[0082] Parameter estimation method: PPP technology: In the positioning solution, the zenith tropospheric delay (ZTD) is estimated as an unknown parameter, combined with a random process (random walk) for dynamic correction.

[0083] In the process of completing the joint monitoring of goaf deformation on the surface of the well, the method of eliminating atmospheric delay error on the surface of the well with Beidou GNSS and SBAS-InSAR (small baseline subset interference) is usually used to complete the joint monitoring of goaf deformation on the surface of the well. The classic method includes:

[0084] Data layer fusion, model layer joint calculation and error complementary mechanism. For example, data layer fusion can expand GNSS point data to surface data through spatial interpolation, while SBAS-InSAR provides high-density spatial information. The model layer may need to establish a deformation model, taking into account both data sources, such as Kalman filtering. In terms of error complementarity, GNSS can correct InSAR atmospheric delay, while InSAR supplements the spatial details of GNSS.

[0085] For example, coal mine goaf monitoring may require the establishment of a monitoring network, combined with real-time GNSS and periodic InSAR observations, to capture dynamic changes through time series analysis. At the same time, multi-source data fusion algorithms, such as least squares configuration or machine learning, may be used to integrate information from different data sources.

[0086] Generally, the method of eliminating the error of Beidou GNSS atmospheric delay: The error of Beidou GNSS mainly comes from ionospheric and tropospheric delay, which needs to be corrected by the following methods:

[0087] Dual-frequency observation method: using the dispersion characteristics of Beidou dual-frequency signals (B1 / B2), ionospheric delay error can be directly eliminated by linear combination. This method is particularly effective in complex terrain of coal mine goaf.

[0088] Model correction method: Klobuchar model: a global empirical model suitable for ionospheric delay, which is corrected in real time through satellite broadcast parameters.

[0089] Hopefield model: based on meteorological parameters (temperature, pressure, humidity) to calculate tropospheric delay, which needs to be combined with ground meteorological station data to improve accuracy.

[0090] Real-time kinematic (RTK): through the synchronous observation of reference station and rover station, common errors (including atmospheric delay) are eliminated, and centimeter-level positioning accuracy is achieved.

[0091] Generally, SBAS-InSAR technology includes: small baseline set processing, selecting SAR image combination with short space-time baseline, reducing time decorrelation effect, and enhancing the ability to extract deformation signal.

[0092] Atmospheric phase correction: using GNSS atmospheric delay data (such as ZTD) to invert InSAR atmospheric phase, and eliminating atmospheric noise in InSAR through spatial interpolation.

[0093] Combined with meteorological models (such as ERA5) to separate tropospheric delay phase.

[0094] Generally, the classic method of GNSS and SBAS-InSAR joint monitoring includes:

[0095] (1) Data layer fusion: time and space complementary mechanism;

[0096] Spatial fusion: combine GNSS point-wise high-precision displacement data with SBAS-InSAR surface deformation field, generate fusion deformation map through Kriging interpolation or least squares configuration.

[0097] Temporal fusion: use GNSS high-frequency time series (minute level) to supplement InSAR low-frequency observations (weeks to months), capture dynamic deformation processes.

[0098] (2) Model layer joint solution.

[0099] Kalman filter / particle filter: construct deformation dynamics model, fuse observation equations of two data sources, and iteratively optimize deformation parameters.

[0100] Deformation field decomposition model: separate the vertical component of subsidence (GNSS dominant) and the horizontal component (InSAR sensitive area) of the mined-out area, and analyze the deformation mechanism.

[0101] (3) Error complementary correction.

[0102] GNSS-assisted InSAR atmospheric correction: use the regional ionospheric / tropospheric delay inverted by GNSS stations to correct the atmospheric error in the InSAR phase map.

[0103] InSAR-constrained GNSS solution: optimize the layout density and solution weight of the GNSS monitoring network through the spatial distribution of the InSAR deformation field.

[0104] Typical application cases.

[0105] Mine deformation monitoring network construction: set up Beidou GNSS continuous operation reference station (CORS) around the mined-out area, combined with the wide-area coverage of SBAS-InSAR, form a "point-surface-body" monitoring system.

[0106] Dynamic early warning system, real-time GNSS data is used to trigger deformation anomaly early warning, SBAS-InSAR provides historical deformation trend analysis, and together supports risk assessment.

[0107] S2, for the curved subsidence zone, fracture zone and caving zone of the mined-out area, respectively set up electrical method system, microseismic monitoring system and distributed optical fiber sensing in the underground, realize the monitoring of underground resistivity, fracture zone expansion and caving zone rock fracture change, and dynamic tracking monitoring of the boundary of the mined-out area, complete the delineation of the underground abnormal body;

[0108] The underground abnormal body includes the abnormal fracture zone expansion range, the caving zone rock fracture degree, the abnormal curved subsidence zone and the dynamic tracking of the mined-out area boundary, and the abnormal underground mined-out area resistivity;

[0109] The microseismic monitoring system is used to monitor the fracture zone expansion range and the rock mass crushing degree of the caving zone, including:

[0110] A sensor network is arranged in the well, a fracture zone expansion identification method is used to capture rock layer rupture signals in real time, the fracture zone expansion range is accurately circled, and a microseismic waveform inversion method is used to quantify the rock mass crushing degree of the caving zone.

[0111] For example, as shown in Figure 2 The fracture zone expansion identification method is used to capture rock layer rupture signals in real time, and the fracture zone expansion range is accurately circled, including:

[0112] S201, sub-regions of the fracture zone expansion range saliency image are divided, and multi-scale deep rock layer rupture visual feature extraction is performed on the fracture zone expansion range saliency image by using a long short-term memory neural network;

[0113] S202, the extracted multi-scale deep rock layer rupture visual feature vector is input into a pre-trained fracture zone expansion range saliency normalization model, the saliency scores of each sub-region of the fracture zone expansion range saliency image are normalized, and the original fracture zone expansion range saliency image is weighted by using a global saliency image of the fracture zone expansion range;

[0114] S203, a fracture zone expansion displacement attribute database is established and used as a fracture zone expansion displacement attribute category to detect the fracture zone expansion displacement attribute of each sub-region of the fracture zone expansion range saliency image, and a salient fracture zone expansion displacement feature is used to initialize the network;

[0115] S204, the fracture zone expansion displacement attribute is used to weight the features of the fracture zone expansion range saliency image;

[0116] S205, the salient fracture zone expansion displacement attribute feature is decoded by using a long short-term memory neural network to generate a fracture zone expansion range saliency image description.

[0117] For example, step S202 specifically includes:

[0118] (a) Pre-training model: the fracture zone expansion range saliency normalization model is a neural network composed of two fully connected layers and an output layer, more than 80% of the pixel points in a sub-region of the fracture zone expansion range saliency image have the same saliency label, the sub-region is selected as a training sample, and the saliency score of the sub-region is set to 1 as a whole, otherwise it is set to 0;

[0119] (b) input all sub-regions of the fracture zone expansion range saliency image into the trained fracture zone expansion range saliency normalization model to obtain multiple saliency maps at multiple segmentation levels, and obtain a saliency map D fused at each segmentation scale by weighting and averaging the saliency maps.smap , the original crack zone extension range saliency image D is weighted by the fusion parameter f, and the expression is:

[0120] D vis =(1-f)×D smap +f×D

[0121] The obtained crack zone extension range saliency image D vis , used as input to the end-to-end crack band extension range saliency image description model for subsequent training and testing.

[0122] Exemplarily, step S203 specifically includes:

[0123] (1) All the descriptions of the fracture zone expansion displacement distances in the fracture zone expansion range database training set were counted, and multiple data with abnormal distances were selected to establish a fracture zone expansion displacement attribute database; 92% of the data in the fracture zone expansion range data training set appeared in the fracture zone expansion displacement attribute database, including the current fracture zone expansion position and the next fracture zone expansion displacement position; based on multiple attributes in the fracture zone expansion displacement attribute database, the fracture zone expansion range significance image D was constructed. vis Make attribute predictions;

[0124] (2) Expand the crack band to the saliency image D vis The image is resized to a 128×128 pixel square and input into the crack zone extension displacement attribute detection network to generate a 6×6 pixel multi-dimensional rough spatial response map L8. Each point in the spatial response map L8 is directly in D vis Perform convolution operation on it.

[0125] Exemplarily, step S204 specifically includes:

[0126] According to the generated attribute, the weight associated with it is selected and accumulated on the corresponding feature dimension as the importance parameter for attribute prediction. Then, the average value of the fracture zone extension range saliency image features at each position in each sub-region is taken to obtain a single feature, and it is weighted on the corresponding dimension by the importance parameter to represent the fracture zone extension displacement attribute feature of the input fracture zone extension range saliency image.

[0127] Exemplarily, step S205 specifically includes:

[0128] According to the threshold b, the top N attributes with high probability ranking are selected {g 1 ,g 2 …g N}, find their respective positions on the spatial response map L8; from the spatial response map L7 to the spatial response map L8 layer, the mapping of the crack zone extension range significance image feature is dimensionally transformed, and the mapping weight connected with {g 1 ,g 2 …g N The mapping weight connected with the spatial response map of} is accumulated and added in each dimension of the spatial response map L7 to obtain an importance vector h of the same dimension; after taking the average of each position of the crack zone extension range significance image feature, the importance in each dimension is weighted, and the expression is:

[0129] D vis-attr =h⊙L7

[0130] In the formula, D vis-attr is the crack zone extension displacement feature, L7 is the spatial response map, h is the vector, and is the Hadamard product;

[0131] The weighted crack zone extension displacement feature D vis-attr is input into the subsequent long-short memory neural network to generate the description of the crack zone extension range under the rock stratum rupture.

[0132] S3, according to the joint monitoring result of the surface deformation of the mined-out area on the well, combined with the delineation result of the abnormal body in the well, the deformation process of the curved subsidence zone, the crack zone and the caving zone is dynamically coupled and monitored.

[0133] For example, in step S3, combined with the delineation result of the abnormal body in the well, the deformation process of the curved subsidence zone, the crack zone and the caving zone in the mined-out area is dynamically coupled and monitored, including:

[0134] Step one, obtain the curved subsidence zone and the mined-out area boundary measurement data, and obtain the light flux density γ m of the rock stratum stress redistribution by irradiating the curved subsidence zone and the mined-out area boundary target of the distributed optical fiber sensing to obtain the measurement data.

[0135] Step two, obtain the curved subsidence zone and the mined-out area boundary profile structure information of the reconstruction object, and obtain the optical fiber optical characteristic parameters;

[0136] Step three, use the imaging software to discretize the imaging target to obtain the curved subsidence zone and the mined-out area boundary grid structure information;

[0137] Step four, based on the light transmission model and the finite element theory, the curved subsidence zone and the mined-out area boundary profile structure information and the optical fiber optical characteristic parameters of the reconstruction target are taken as prior information to establish the linear relationship between the curved subsidence zone and the mined-out area boundary data of the surface finite angle and the rock stratum stress redistribution of the curved subsidence zone and the mined-out area boundary target distribution inside the reconstruction target.

[0138] Step five, the linear relationship of the stress redistribution is converted into a 1 / 2 norm minimization problem, expressed as:

[0139]

[0140] where η is a regularization parameter, G is a system matrix, E is a three-dimensional stress distribution of the curved subsidence zone and the boundary target of the goaf to be solved, γ m is the light flux density of the stress redistribution of the rock stratum, |||2 is the absolute value of the 2 norm, and 1 / 2 is the absolute value of the 1 / 2 norm;

[0141] Step six, the linear relationship of the stress redistribution of the rock stratum is solved by using a half-threshold iteration technique, a half-threshold operator is introduced, and the expression is:

[0142]

[0143] The half-threshold iteration technique is expressed as:

[0144]

[0145] F(E) = E + G T (γ - GE)

[0146] where O η,1 / 2 (·) is a half-threshold iteration function, C is the stress value of the rock stratum, η is a regularization parameter, is a downward recursion, F(E) is a linear value of the three-dimensional stress distribution of the curved subsidence zone and the boundary target of the goaf to be solved, and [F(E)] i is a set of the i-th linear value of the three-dimensional stress distribution of the curved subsidence zone and the boundary target of the goaf to be solved, γ is the light flux density, and T is a transpose matrix.

[0147] Step seven, a dynamic tracking method of the boundary of the goaf is introduced to reduce the number of iterations, F(E) is approximately equal to E, and each iteration has the following process:

[0148] Z n+1 = sup(p n+1,1 ), p n+1,1 = O(F(E n ))

[0149]

[0150] where Z n+1 is the boundary value of the goaf of the n+1 iteration, sup() is a dynamic tracking function, p n+1,1is the node value for dynamic tracking of the goaf boundary between the n+1th and 1st iterations, O() is the iteration result function, F() is the linear function of the target three-dimensional stress distribution of the curved subsidence zone and the goaf boundary to be solved, E n The current distribution of the target three-dimensional stress distribution of the curved subsidence zone and the goaf boundary after n iterations is E n+1 The current distribution of the target three-dimensional stress distribution of the curved subsidence zone and the goaf boundary after n+1 iterations is p n+1,k+1 is the node trajectory for dynamic tracking of the goaf boundary of the n+1th and k+1th iterations, p n+1,k is the node trajectory for dynamic tracking of the goaf boundary of the n+1th and kth iterations, t n+1,k The time taken for the +1 and kth iterations;

[0151] Step 8: Through the iterative process, the reconstruction result E of step n is obtained n , when ||γ-GE n || / ||γ||≤1e -5 or ||E n -GE n || / ||γ||≤1e -8 When , the iteration stops; where e is the Reynolds index;

[0152] Step nine, using imaging software to discretize the imaging target to obtain a grid with a finer grid size;

[0153] Step 10: Perform image fusion of the reconstruction result, the curved subsidence zone of the imaging target, and the boundary profile structure of the goaf, and display the result.

[0154] Exemplarily, step 2 specifically includes:

[0155] (2.1) Reconstruct the cross-sectional structural information of the curved subsidence zone and goaf boundary of the object; use the human-computer interactive semi-automatic segmentation method in 3DMED software to perform tissue segmentation on the curved subsidence zone and goaf to obtain the cross-sectional structure of the curved subsidence zone and goaf boundary of the imaging target;

[0156] (2.2) Obtaining optical characteristic parameters: Based on regional diffuse optical tomography, the optical characteristic parameters of each rock stress node at the curved subsidence zone and the boundary of the goaf within the imaging target are obtained.

[0157] Exemplarily, step four specifically includes:

[0158] (4.1) Light transmission model, which uses the diffusion approximation equation to describe the transmission process of light within the imaging target;

[0159] (4.2) According to the finite element theory, the curved subsidence zone of the reconstruction target is fused with the profile structure information and optical characteristic parameters of the goaf boundary, the diffusion approximation equation is discretized, and the linear relationship between the measurement data of the surface and the rock stress redistribution of the curved subsidence zone and the goaf boundary target distribution of the reconstruction target is constructed, and the expression is:

[0160] γm=GE

[0161] In the formula, G is a system matrix, E is a three-dimensional stress distribution of the curved subsidence zone and the goaf boundary target to be solved

[0162] It can be understood that the step S3 completes the cooperative dynamic coupling monitoring of the deformation process of the goaf curved subsidence zone, the fracture zone and the caving zone, including the following related contents:

[0163] The time series InSAR technology and three-dimensional displacement monitoring include:

[0164] The distributed scatterer (DS) enhancement technology improves the coherent target density of the goaf by constructing a multi-scattering characteristic DS (M-DS) identification model, and solves the monitoring blind area problem of the traditional InSAR in the low coherence area. Combined with the L1 norm deformation estimation method, the outlying point interference can be effectively suppressed, and it is suitable for monitoring the fracture zone with nonlinear deformation.

[0165] The three-dimensional deformation solving model combines PS (permanent scatterer), DS and M-DS targets, and uses the probability integral method model to invert the three-dimensional displacement field, so as to realize the synchronous monitoring of the vertical subsidence and horizontal movement of the curved subsidence zone. For example, through InSAR data, it is found that there is abnormal subsidence outside a certain kilometer, which is related to deep groundwater pumping.

[0166] The pixel offset technology (POI) is used for the large gradient deformation of the caving zone, and the SAR image intensity tracking technology is used to break through the phase unwrapping limit, so that the millimeter to meter level rapid deformation process can be captured.

[0167] Downhole geophysical monitoring technology.

[0168] Microseismic monitoring system. The sensor network is arranged in the well, and the rock fracture signal is captured in real time by using the improved method of the application, and the expansion range of the fracture zone is accurately circled. Combined with the waveform inversion technology, the rock breaking degree of the caving zone can be quantified.

[0169] Distributed optical fiber sensing (BOTDR / DTS), optical fibers are arranged along the roadway, and the rock stress redistribution caused by deformation and temperature change is monitored by using the improved method of the application, which is especially suitable for dynamic tracking of the curved subsidence zone and the goaf boundary.

[0170] The electrical method system is arranged in the well for hydrological monitoring, including resistivity monitoring.

[0171] For example, the deformation process of the bending subsidence zone, fracture zone and caving zone in the temple gob area is monitored dynamically and cooperatively, which usually includes:

[0172] GIS spatiotemporal database construction, integration of InSAR deformation field, underground microseismic event, hydrological monitoring and other data, and generation of 4D visualization model of "three-zone" development by using spatial interpolation algorithm.

[0173] Numerical simulation and machine learning prediction, using FLAC3D or UDEC software to simulate rock movement, combining LSTM neural network to train time series monitoring data, and realizing dynamic prediction of caving zone height and fracture zone angle. The gob research has verified that the prediction accuracy of this method for deformation response of working face advancing process is more than 90%.

[0174] Cooperative monitoring system design, including:

[0175] Spatial layering control strategy:

[0176] Surface layer: high-frequency InSAR (such as Sentinel-1, 12-day revisit) is used to monitor large-scale deformation of the bending subsidence zone.

[0177] Shallow fracture zone: combined with borehole tiltmeter and high-density resistivity method, to capture 10-50m deep rock layer separation phenomenon.

[0178] Deep caving zone: through underground microseismic and borehole television imaging, to monitor the characteristics of rock mass collapse below 30m.

[0179] Dynamic coupling correction mechanism, Kalman filter fusion algorithm is established, taking InSAR monitoring results as constraint conditions, to correct the local observation error of underground sensor in real time, and to improve the overall accuracy of the system.

[0180] Among them, Kalman filter realizes state estimation through two recursive processes of time update (prediction) and observation update (correction), which specifically includes five key steps:

[0181] State prediction: based on the optimal state estimation of the last time, the system state at the current time is predicted, and the equation is:

[0182]

[0183] In the formula, is the optimal state estimation value at time k, is the optimal state estimation value at time k-1;

[0184] Covariance prediction: update the covariance matrix of the predicted state, which reflects the prediction uncertainty, and the equation is:

[0185]

[0186] wherein, P is the covariance update prediction value at time k k-1 Q is the process noise covariance at time k-1

[0187] Kalman gain calculation: weighing the credibility of the prediction value and the observation value to determine the fusion weight, the equation is:

[0188]

[0189] wherein, K k K is the fusion weight value at time k, H is the observation matrix, T is the transpose matrix, and R is the measurement noise covariance

[0190] The essence of the Kalman filter fusion algorithm is a multi-source data fusion process, and its core mechanisms include:

[0191] Noise suppression: quantifying the uncertainty of prediction and observation through the covariance matrix to suppress Gaussian noise interference

[0192] Dynamic weight distribution: the Kalman gain is self-adaptively adjusted according to the noise level, and when the observation noise is small (R is small), the gain is increased, and the observation value weight is increased; otherwise, the prediction value is relied on

[0193] Recursive iteration: continuously iteratively optimize state estimation to realize real-time dynamic adjustment.

[0194] In example 2, the geological disaster automatic monitoring device provided by the present application comprises:

[0195] The above-ground surface goaf deformation joint monitoring module is used for the coal mine goaf, uses radar satellite InSAR to continuously perform InSAR goaf surface deformation inversion processing for several months, combines the on-site location deployment of the above-ground surface Beidou GNSS, eliminates the influence of atmospheric delay error, and completes the above-ground surface goaf deformation joint monitoring

[0196] The underground anomaly body delineation module is used for the curved subsidence zone, the fracture zone and the caving zone of the goaf, respectively sets the electrical method system, the microseismic monitoring system and the distributed optical fiber sensing in the underground, realizes the monitoring of the underground resistivity, the fracture zone expansion and the caving zone rock mass crushing change, and the dynamic tracking monitoring of the goaf boundary, and completes the delineation of the underground anomaly body

[0197] The dynamic coupling monitoring module is used for the above-ground surface goaf deformation joint monitoring result, combined with the delineation result of the underground anomaly body, cooperatively monitors, completes the cooperatively dynamic coupling monitoring of the deformation process of the curved subsidence zone, the fracture zone and the caving zone of the goaf.

[0198] Application example: for the goaf of 5-20303 working face of Dangsi Coal Mine, the combined monitoring of underground resistivity and microseismic is used in the range of 350m wide and 1000m long, combined with GNSS on the ground, radar satellite InSAR technology, to monitor the deformation process of the upper three zones of the goaf (curved subsidence zone, fracture zone and caving zone) in a coordinated and dynamic coupling manner. The spatiotemporal complementarity and process coordination of multiple methods are utilized to realize the dynamic coupling monitoring of the deformation of the goaf in the mine and on the ground.

[0199] According to the needs, for the goaf of Dangsi Coal Mine, ALOS-2, RADARSAT-2 and Sentinel-1 SAR data are used for InSAR goaf surface deformation inversion processing for four consecutive months. In addition, Beidou GNSS is deployed at a suitable location on the site to eliminate the influence of atmospheric delay error and realize high-precision combined monitoring of the deformation of the goaf on the ground. In order to supplement the shortcomings of surface deformation monitoring methods, the electrical method system and the microseismic monitoring system are deployed for underground resistivity and microseismic combined stress change monitoring to realize the delineation of underground abnormal bodies. Combined with the multi-method coordinated monitoring of the mine and on the ground, the deformation process of the upper three zones of the goaf (curved subsidence zone, fracture zone and caving zone) is monitored in a coordinated and dynamic coupling manner. The technical contribution of the present application lies in the improvement of the accuracy of the delineation of underground abnormal bodies.

[0200] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any modification, equivalent replacement and improvement made by those skilled in the art within the technical range disclosed by the present application, as long as it is within the spirit and principles of the present application, should be covered within the protection scope of the present application.

Claims

1. A method for automatic monitoring of geological disasters, characterized in that: The method comprises the following steps: S1, for coal mine goaf, uses radar satellite InSAR to conduct InSAR goaf surface deformation inversion processing for several consecutive months. Combined with the deployment of BeiDou GNSS on the surface of the mine, it eliminates the influence of atmospheric delay error and completes the joint monitoring of goaf deformation on the mine surface. S2: For the curved and subsided zones, fracture zones, and caving zones in the goaf, electrical method systems, microseismic monitoring systems, and distributed fiber optic sensors are installed underground to monitor the underground resistivity, fracture zone expansion, and rock fragmentation changes in the caving zone. This allows for dynamic tracking and monitoring of the goaf boundary, thereby delineating the underground anomalies. S3, based on the joint monitoring results of the surface goaf deformation, combined with the delineation results of the underground anomaly, complete the coordinated dynamic coupling monitoring of the deformation process of the goaf's curved subsidence zone, fracture zone and caving zone.

2. The automatic monitoring method for geological disasters according to claim 1, characterized in that: In step S2, the underground anomaly includes: abnormal fracture zone extension range, rock mass fragmentation degree of caving zone, dynamic tracking of abnormal curved subsidence zone and goaf boundary, abnormal goaf resistivity; Realize the monitoring of underground resistivity, fracture zone expansion and rock fragmentation in the caving zone, including: deploying a sensor network underground, using the fracture zone expansion identification method to capture rock rupture signals in real time, delineating the fracture zone expansion range, and combining the microseismic waveform inversion method to quantify the degree of rock fragmentation in the caving zone.

3. The automatic monitoring method for geological disasters according to claim 2, characterized in that: The fracture zone extension identification method is used to capture rock fracture signals in real time and delineate the fracture zone extension range, including: S201, dividing the fracture zone extension range saliency image into sub-regions, and extracting multi-scale deep rock fracture visualization features from the fracture zone extension range saliency image using a long short-term memory neural network; S202: Inputting the extracted multi-scale deep rock fracture visualization feature vector into a pre-trained fracture zone extension range saliency normalization model to normalize the saliency scores of each sub-region of the fracture zone extension range saliency image, and weighting the original fracture zone extension range saliency image with the fracture zone extension range global saliency image; S203, establishing a fracture zone expansion displacement attribute database and using it as a fracture zone expansion displacement attribute category, performing fracture zone expansion displacement attribute detection on each sub-region of the fracture zone expansion range significance image, and using significant fracture zone expansion displacement feature pairs to initialize the network; S204, using the fracture zone expansion displacement attribute, weighting the fracture zone expansion range saliency image feature; S205, using a long short-term memory neural network to decode the significant fracture zone expansion displacement attribute features, and generate a significant image description of the fracture zone expansion range.

4. The automatic monitoring method for geological disasters according to claim 3, characterized in that: Step S202 specifically includes: (a) Pre-training model: The saliency normalization model for the crack extension range is a neural network consisting of two fully connected layers and one output layer. If more than 80% of the pixels in a subregion of the crack extension range saliency image have the same saliency label, the subregion is selected as a training sample and its saliency score is set to 1 overall, otherwise it is set to 0. (b) All sub-regions of the crack band extension range saliency image are input into the trained crack band extension range saliency normalization model to obtain multiple saliency maps at multiple segmentation levels. The saliency maps are weighted averaged to obtain the fused saliency map D at each segmentation scale. smap , the original crack zone extension range saliency image D is weighted by the fusion parameter f, and the expression is: D vis =(1-f)×D smap +f×D The obtained crack zone extension range saliency image D vis , used as input to the end-to-end crack band extension range saliency image description model for subsequent training and testing.

5. The automatic monitoring method for geological disasters according to claim 4, characterized in that: Step S203 specifically includes: (1) All the descriptions of the fracture zone expansion displacement distances in the fracture zone expansion range database training set were counted, and multiple data with abnormal distances were selected to establish a fracture zone expansion displacement attribute database; 92% of the data in the fracture zone expansion range data training set appeared in the fracture zone expansion displacement attribute database, including the current fracture zone expansion position and the next fracture zone expansion displacement position; based on multiple attributes in the fracture zone expansion displacement attribute database, the fracture zone expansion range significance image D was constructed. vis Make attribute predictions; (2) Expand the crack band to the saliency image D vis The image is resized to a 128×128 pixel square and input into the crack zone extension displacement attribute detection network to generate a 6×6 pixel multi-dimensional rough spatial response map L8. Each point in the spatial response map L8 is directly in D vis Perform convolution operation on it.

6. The automatic monitoring method for geological disasters according to claim 5, characterized in that: Step S205 specifically includes: According to the threshold b, the top N attributes with high probability ranking are selected {g 1 ,g 2 …g N }, find their respective corresponding positions on the spatial response map L8; from the spatial response map L7 to the spatial response map L8 layer, perform dimension transformation mapping on the significant image features of the crack zone extension range, and select {g 1 ,g 2 …g N The mapping weights of the spatial response graphs connected to the} are cumulatively added in each dimension of the spatial response graph L7 to obtain an importance vector h of the same dimension; after taking the average of each position of the significant image features of the crack zone extension range, the importance of each dimension is weighted, and the expression is: D vis-attr =h⊙L7 Where D vis-attr is the displacement characteristic of the crack zone expansion, L7 is the spatial response diagram, h is the vector, and ⊙ is the Hadamard product; The weighted fracture zone expansion displacement characteristic D vis-attr The data is input into the subsequent long-short-term memory neural network to generate a description of the extension range of the fracture zone under the rock stratum fracture.

7. The automatic monitoring method for geological disasters according to claim 1, characterized in that: In step S3, the collaborative monitoring of the deformation process of the curved subsidence zone, fracture zone and caving zone in the goaf is completed by combining the delineation results of the underground abnormal body with the collaborative monitoring, including: Step 1: Obtain measurement data of the curved subsidence zone and the goaf boundary. The distributed optical fiber sensor illuminates the curved subsidence zone and the goaf boundary target to obtain measurement data and obtain the light flux density γ of the rock formation stress redistribution. m ; Step 2: Obtain the cross-sectional structural information of the curved subsidence zone and the goaf boundary of the reconstructed object, and obtain the optical fiber optical characteristic parameters; Step 3: Use imaging software to discretize the imaging target and obtain the grid structure information of the curved subsidence zone and the goaf boundary; Step 4: Based on the optical transmission model and finite element theory, the cross-sectional structural information of the reconstructed curved subsidence zone and the goaf boundary and the optical fiber characteristic parameters are used as prior information to establish the curved subsidence zone and goaf boundary data with a limited angle on the surface, as well as the linear relationship between the rock stress redistribution of the curved subsidence zone and the goaf boundary target distribution inside the reconstructed target; Step 5: Convert the linear relationship of rock stress redistribution into a 1 / 2 norm minimization problem, expressed as: Where η is the regularization parameter, G is the system matrix, E is the target three-dimensional stress distribution of the curved subsidence zone and the goaf boundary to be solved, γ m is the light flux density of rock stress redistribution, ||||2 is the absolute value of the 2-norm, || || 1 / 2 is the absolute value of the 1 / 2 norm; Step 6: The linear relationship of rock stress redistribution is solved by using the semi-threshold iteration technique and introducing the semi-threshold operator. The expression is: The semi-threshold iterative technique is expressed as: F(E)E+G T (γ-GE) Where, O η,1 / 2 (·) is the semi-threshold iterative function, C is the rock formation stress value, η is the regularization parameter, For downward recursion, F(E) is the linear value of the target three-dimensional stress distribution of the curved subsidence zone and the goaf boundary to be solved, [F(E)] i is the set of linear values ​​of the three-dimensional stress distribution of the i-th curved subsidence zone and the boundary of the goaf to be solved, γ is the light flux density, and T is the transposed matrix; Step 7: Introduce a dynamic tracking method for the goaf boundary to reduce the number of iterations. Let F(E)≈E. Each iteration has the following process: From n+1 =sup(p n+1,1 ),p n+1,1 =O(F(E n )) Where Z n+1 is the boundary value of the goaf for n+1 iterations, sup() is the dynamic tracking function, and p n+1,1 is the node value for dynamic tracking of the goaf boundary between the n+1th and 1st iterations, O() is the iteration result function, F() is the linear function of the target three-dimensional stress distribution of the curved subsidence zone and the goaf boundary to be solved, E n The current distribution of the target three-dimensional stress distribution of the curved subsidence zone and the goaf boundary after n iterations is E n+1 The current distribution of the target three-dimensional stress distribution of the curved subsidence zone and the goaf boundary after n+1 iterations is p n+1,k+1 is the node trajectory for dynamic tracking of the goaf boundary of the n+1th and k+1th iterations, p n+1,k is the node trajectory for dynamic tracking of the goaf boundary of the n+1th and kth iterations, t n+1,k The time taken for the +1 and kth iterations; Step 8: Through the iterative process, the reconstruction result E of step n is obtained n , when ||γ-GE n || / ||γ||≤1e -5 or ||E n -GE n || / ||γ||≤1e -8 When , the iteration stops; where e is the Reynolds index; Step nine, using imaging software to discretize the imaging target to obtain a grid with a finer grid size; Step 10: Perform image fusion of the reconstruction result, the curved subsidence zone of the imaging target, and the boundary profile structure of the goaf, and display the result.

8. The automatic monitoring method for geological disasters according to claim 7, characterized in that: Step 2 specifically includes: (2.1) Reconstruct the cross-sectional structural information of the curved subsidence zone and goaf boundary of the object; use the human-computer interactive semi-automatic segmentation method in 3DMED software to perform tissue segmentation on the curved subsidence zone and goaf to obtain the cross-sectional structure of the curved subsidence zone and goaf boundary of the imaging target; (2.2) Obtaining optical characteristic parameters: Based on regional diffuse optical tomography, the optical characteristic parameters of each rock stress node at the curved subsidence zone and the boundary of the goaf within the imaging target are obtained.

9. The automatic monitoring method for geological disasters according to claim 7, characterized in that: Step 4 specifically includes: (4.1) Light transmission model, which uses the diffusion approximation equation to describe the transmission process of light within the imaging target; (4.2) Based on the finite element theory, the cross-sectional structural information and optical characteristic parameters of the reconstructed curved subsidence zone and the goaf boundary are integrated, the diffusion approximation equation is discretized, and a linear relationship between the surface measurement data and the rock stress redistribution distribution of the reconstructed curved subsidence zone and the goaf boundary target is constructed. The expression is: γ m =GE。 10. An automatic monitoring device for geological disasters, characterized in that: The device is implemented by the automatic geological disaster monitoring method according to any one of claims 1 to 9, and the device includes: The joint monitoring module for deformation of coal mine goafs above ground and on the surface uses InSAR radar satellites to perform InSAR surface deformation inversion processing for several consecutive months. Combined with the deployment of BeiDou GNSS above ground at the on-site location, this module eliminates the influence of atmospheric delay errors and completes joint monitoring of deformation of coal mine goafs above ground and on the surface. The underground anomaly delineation module is used to locate the curved and subsided zones, fracture zones, and caving zones in the goaf. An electrical method system, a microseismic monitoring system, and distributed fiber optic sensors are installed underground to monitor underground resistivity, fracture zone expansion, and rock fragmentation changes in the caving zone. This allows for dynamic tracking and monitoring of goaf boundaries, thereby completing the delineation of underground anomalies. The dynamic coupling monitoring module is used to carry out collaborative monitoring based on the joint monitoring results of the surface goaf deformation and the delineation results of the underground abnormal bodies, so as to complete the collaborative dynamic coupling monitoring of the deformation process of the curved subsidence zone, fracture zone and caving zone in the goaf.

Images