Multi-modal information acquisition and roof global attitude monitoring method and system

Abstract

The invention provides a multi-modal information acquisition and roof global attitude monitoring method and system, and relates to the technical field of mining safety monitoring and roadway surrounding rock control. The method comprises the following steps of: acquiring acoustic, optical, electrical and mine pressure multi-source data by arranging a fan-shaped drill hole, an annular electrode array and various mine pressure monitoring devices; a wave velocity profile, a fracture parameter set and a resistivity equivalent cloud picture are generated through wavelet noise reduction, fracture recognition and resistivity inversion, space-time registration with a mine pressure data sequence is carried out, and roof partition probability distribution is output through a space-time feature pyramid network. Furthermore, improved XGBoost is adopted for joint inversion of the thickness of the loose circle, the Poisson's ratio and the elastic modulus, probability distribution serves as a space weight, the separation amount and anchoring stress data are fused, and a roof displacement field, a curvature field and a rotation field are reconstructed. And finally, generating an early warning based on a preset safety threshold value, outputting a posture partition map and a monitoring report, and realizing top plate posture transparent deduction and support coordinated regulation and control.

Description

Technical Field

[0001] This invention relates to the field of mine safety monitoring and roadway surrounding rock control technology, and in particular to a method and system for multimodal information acquisition and roof attitude monitoring. Background Technology

[0002] Under current conditions of deep coal mining and ultra-long working faces, the dynamics of roof movement and mine pressure manifestation are becoming increasingly complex. To ensure production safety, various methods are commonly used on-site for observation, including roof separation meters, anchor bolt dynamometers, anchor cable dynamometers, borehole stress gauges, and support resistance monitoring. These methods can reflect the local stability of the roof and the stress conditions of the surrounding rock to a certain extent, but due to the limited information sources, they often fail to fully reveal the overall picture of overburden fracturing and surrounding rock instability.

[0003] With the advancement of smart mines and transparent mining concepts, research trends are gradually shifting towards multimodal monitoring and multi-source information fusion. On the one hand, ultrasonic detection, borehole optical inspection, and resistivity testing are being introduced into the monitoring of surrounding rock and roof in roadways to characterize fracture development, dielectric properties, and structural evolution. On the other hand, traditional mine pressure monitoring data is being combined with acoustic, optical, and electrical acquisition results, and processed in real time through big data platforms and machine learning methods, thereby forming a multi-angle, multi-scale surrounding rock condition perception and prediction system. These methods lay the foundation for achieving transparent extrapolation and comprehensive identification of roof conditions.

[0004] However, existing technologies still have shortcomings. First, the consistency processing of multi-source monitoring data across the spatiotemporal dimensions is insufficient, making it difficult to register and fuse different modal information within a unified framework. Second, existing results typically only provide locally discrete monitoring conclusions, lacking a zonal representation of the continuous field, and failing to intuitively reflect the gradual evolution of the roof from the elastic zone to the loosened zone. Third, parameter inversion mostly relies on numerical simulations or empirical methods, making it difficult to combine real-time observation data for dynamic correction and joint inversion. Therefore, how to achieve integrated processing of multimodal data, intelligent inversion of mechanical parameters, and dynamic reconstruction of the roof's global attitude remain key issues that urgently need to be addressed. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, the purpose of this invention is to provide a multimodal information acquisition and roof global attitude monitoring method and system. Targeting ultra-long working faces and narrow coal pillar side roadways, it integrates multimodal acquisition methods of "sound, light, and electricity" and a multi-parameter mine pressure monitoring device to acquire wave velocity, fracture, resistivity, and mine pressure information of the surrounding rock in real time. Through spatiotemporal feature fusion and inversion calculation, it reconstructs the global displacement, curvature, and rotation state of the roof, thereby realizing transparent roof attitude prediction and coordinated control of support.

[0006] To achieve the above objectives, the present invention provides the following solution: A method for multimodal information acquisition and global attitude monitoring of the roof includes: Fan-shaped boreholes and ring-shaped electrode arrays were arranged on both sides and the roof of the roadway. Ultrasonic detectors, borehole optical sights, and resistivity monitoring devices were installed. At the same time, roof separation instruments, anchor bolt force gauges, anchor cable force gauges, borehole stress gauges, and support working resistance acquisition units were arranged on the cross-section to establish a unified time reference and obtain ultrasonic data sequences, borehole wall image sequences, electrode array potential data sequences, and mine pressure monitoring data sequences. Wavelet denoising and first wave arrival time extraction are performed on the ultrasonic data sequence to form a wave velocity profile; crack intelligent identification is performed on the borehole wall image sequence to obtain a crack parameter set; least squares inversion is used to generate resistivity contour maps on the electrode array potential data sequence; and noise reduction and anomaly removal are performed on the mine pressure monitoring data sequence. The wave velocity profile, the fracture parameter set, the resistivity contour map, and the preprocessed mine pressure monitoring data sequence are spatiotemporally registered, input into a spatiotemporal feature pyramid network, and output the partition probability distribution of elastic zone, plastic zone, fracture zone, and loosening zone. Based on wave velocity profiles, fracture parameter sets, resistivity contour maps, and mine pressure monitoring data sequences, an improved XGBoost joint inversion method was used to retrieve the loosened zone thickness, Poisson's ratio, and elastic modulus. Using the aforementioned partition probability distribution as spatial weights, the delamination amount, anchor bolt force sequence, and anchor cable force sequence in the mine pressure monitoring data sequence are integrated to reconstruct the roof displacement field, curvature field, and rotation field. Based on a preset safety threshold, the displacement field, curvature field, rotation field, and mine pressure monitoring data sequence are judged to generate an early warning, and an attitude zoning map and monitoring report are output for support and coordinated control.

[0007] Preferably, the mine pressure monitoring data sequence includes: delamination amount, anchor bolt force, anchor cable force, borehole stress, and support working resistance.

[0008] Preferably, the set of fracture parameters includes fracture density, width, and a dip rose diagram.

[0009] Preferably, fan-shaped boreholes and annular electrode arrays are arranged on both sides and the roof of the roadway, and ultrasonic detectors, borehole optical sights, and resistivity monitoring devices are installed. Simultaneously, roof separation meters, anchor bolt force gauges, anchor cable force gauges, borehole stress gauges, and support working resistance acquisition units are arranged on the cross-section to establish a unified time reference, obtaining ultrasonic data sequences, borehole wall image sequences, electrode array potential data sequences, and mine pressure monitoring data sequences, including: Fan-shaped boreholes for ultrasonic detection are drilled on both sides and the roof of the tunnel to serve as acquisition channels for ultrasonic data sequences, forming an ultrasonic acquisition station. The acquisition frequency is once per shift to obtain the raw ultrasonic records. An optical borehole sight was installed in the same hole as the ultrasonic borehole to observe the development of cracks in the borehole wall. The image was collected once a day to obtain a sequence of borehole wall images. A ring-shaped electrode array covering the sidewalls and roof is arranged on the surface of the roadway for resistivity monitoring. Real-time continuous monitoring is performed to obtain the potential data sequence of the electrode array. Cross sections are arranged at intervals on the working face, and roof separation instrument, anchor bolt force gauge, anchor cable force gauge, borehole stress gauge and support working resistance acquisition unit are synchronously installed on the cross sections for real-time monitoring of the stress and deformation of the surrounding rock and anchoring system in the roadway, so as to obtain the mine pressure monitoring data sequence.

[0010] Preferably, the ultrasonic data sequence is subjected to wavelet denoising and first wave arrival time extraction to form a wave velocity profile; the borehole wall image sequence is subjected to intelligent fracture identification to obtain a fracture parameter set; the electrode array potential data sequence is subjected to least squares inversion to generate resistivity contour maps; and the mine pressure monitoring data sequence is subjected to denoising and anomaly removal, including: The ultrasonic data sequence is filtered and denoised, and the arrival time of the first wave is extracted to generate a wave velocity profile. Intelligent crack recognition is performed on the image sequence of the pore wall, and crack density, crack width and crack inclination rose diagram are output to form the crack parameter set used to describe the crack development characteristics; The resistivity inversion of the electrode array potential data sequence is optimized using least squares to generate a resistivity contour map. Based on the monitoring system's processing and statistical analysis procedures for mine pressure data, the mine pressure monitoring data sequence is processed. Combined with the pre-set stress, strain, and displacement safety thresholds, data exceeding the thresholds are marked and removed to obtain a pre-processed mine pressure monitoring data sequence for subsequent registration and fusion.

[0011] Preferably, the wave velocity profile, the fracture parameter set, the resistivity contour map, and the preprocessed mine pressure monitoring data sequence are spatiotemporally registered, input into a spatiotemporal feature pyramid network, and output the probability distribution of the elastic zone, plastic zone, fracture zone, and loosening zone, including: The timestamps of the wave velocity profile, the fracture parameter set, the resistivity contour map, and the mine pressure monitoring data sequence are aligned according to their respective acquisition frequencies to form a time index for fusion. Using the cross-sections arranged at intervals on the working face as spatial indexes, the wave velocity profile, the fracture parameter set, the resistivity contour map, and the mine pressure monitoring data sequence are mapped to a unified cross-section coordinate system according to the cross-section position and the borehole position to complete spatial alignment and obtain the spatiotemporal correspondence for fusion. The spatiotemporally registered wave velocity profile, fracture parameter set, resistivity contour map and mine pressure monitoring data sequence are used as inputs to the four-dimensional data and fused into the spatiotemporal feature pyramid network. Following the process of extracting wave velocity attenuation rate, crack density and resistivity abrupt gradient, using branch network to extract heterogeneous features, and using pyramid structure to compress multi-scale spatiotemporal features, the probability distribution of elastic region, plastic region, fracture region and loosening region is obtained.

[0012] Preferably, following the process of extracting wave velocity attenuation rate, fracture density, and resistivity abrupt gradients, using a branch network to extract heterogeneous features, and utilizing a pyramid structure to compress multi-scale spatiotemporal features, the probability distributions of the elastic region, plastic region, fracture region, and loosening region are obtained, including: Based on the wave velocity profile, the wave velocity attenuation rate, which characterizes the spatial variation of wave velocity, is calculated and used as an acoustic feature input. Based on the fracture parameter set, the fracture density used to describe the degree of fracture development is extracted and used as the image fracture feature input; Based on the resistivity contour map, the resistivity abrupt gradient used to characterize the boundary of electrical anomalies is calculated and used as the input of electrical features; The acoustic feature input, the image gap feature input, and the electrical feature input are respectively fed into the corresponding branch network to extract heterogeneous feature representations; The output of the branch network is aggregated and compressed in a multi-scale pyramid structure in the spatiotemporal feature pyramid network to form a fused spatiotemporal feature representation. Based on the fused spatiotemporal feature representation, the probability distribution of the elastic region, plastic region, fracture region, and loosening region is output.

[0013] Preferably, based on wave velocity profiles, fracture parameter sets, resistivity contour maps, and mine pressure monitoring data sequences, an improved XGBoost joint inversion method is used to retrieve the loosened zone thickness, Poisson's ratio, and elastic modulus, including: Based on the spatiotemporal registration results, the wave velocity profile, fracture parameter set, resistivity contour map and the mine pressure monitoring data sequence are aligned with the cross sections and time indexes arranged at working face intervals to form a set of characteristic parameters for inversion. Using the set of characteristic parameters as input, the improved XGBoost model is invoked, and the target quantities for joint inversion are set as loosened zone thickness, Poisson's ratio and surrounding rock elastic modulus, and joint inversion calculation is performed. The loosened zone thickness sequence, Poisson's ratio sequence, and surrounding rock elastic modulus sequence under the cross-section-time index are obtained as the inversion results of the mechanical property parameters.

[0014] Preferably, using the partitioned probability distribution as spatial weights, the delamination amount, anchor bolt force sequence, and anchor cable force sequence from the mine pressure monitoring data sequence are fused to reconstruct the roof displacement field, curvature field, and rotation field, including: Under each cross section and time index, the probability distribution of the elastic zone, plastic zone, fracture zone and loosening zone is normalized to obtain the spatial weight set corresponding to the cross section position, which is used as the fusion weight. The cross-sections arranged at working face intervals map the delamination amount, anchor bolt force sequence and anchor cable force sequence to a unified cross-section coordinate system, and establish a one-to-one spatiotemporal index relationship with the spatial weight set. The delamination amount is weighted and continuous using the spatial weight set to obtain the cross-sectional displacement distribution; the displacement distribution is constrained and checked at the anchoring position using the anchor bolt force sequence and the anchor cable force sequence to form the top plate displacement field. Under the unified cross-sectional coordinate system, the curvature distribution is calculated based on the spatial variation relationship of the top plate displacement field to obtain the curvature field; Under adjacent time indexes, the rotation amount is calculated based on the directional change relationship of the top plate displacement field to obtain the rotation field.

[0015] A multimodal information acquisition and roof global attitude monitoring system, comprising: The monitoring deployment and synchronization unit is used to deploy fan-shaped boreholes and ring electrode arrays on both sides and the roof of the roadway, install ultrasonic detectors, borehole optical sights, and resistivity monitoring devices, and simultaneously deploy roof separation instruments, anchor bolt force gauges, anchor cable force gauges, borehole stress gauges, and support working resistance acquisition units on the cross-section to establish a unified time reference and obtain ultrasonic data sequences, borehole wall image sequences, electrode array potential data sequences, and mine pressure monitoring data sequences. The acquisition and preprocessing unit is used to perform wavelet denoising and first wave arrival time extraction to form a wave velocity profile on the ultrasonic data sequence, perform intelligent crack identification on the borehole wall image sequence to obtain a crack parameter set, use least squares inversion to generate resistivity contour maps on the electrode array potential data sequence, and perform denoising and anomaly removal on the mine pressure monitoring data sequence. The spatiotemporal registration and partitioning unit is used to perform spatiotemporal registration of the wave velocity profile, the fracture parameter set, the resistivity contour map and the preprocessed mine pressure monitoring data sequence. It inputs a spatiotemporal feature pyramid network and outputs the partition probability distribution of elastic zone, plastic zone, fracture zone and loosening zone. The joint inversion unit is used to jointly invert the loosened zone thickness, Poisson's ratio and elastic modulus using improved XGBoost, based on wave velocity profiles, fracture parameter sets, resistivity contour maps and mine pressure monitoring data sequences. The attitude reconstruction unit is used to reconstruct the roof displacement field, curvature field and rotation field by fusing the delamination amount, anchor bolt force sequence and anchor cable force sequence in the mine pressure monitoring data sequence with the partition probability distribution as spatial weight; The early warning and output unit is used to identify the displacement field, curvature field, rotation field and mine pressure monitoring data sequence based on a preset safety threshold, generate an early warning, and output an attitude zoning map and monitoring report for support and coordinated control.

[0016] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects: Existing technologies often rely on single-point measurement methods such as delamination meters, anchor bolt stress monitoring, or localized stress monitoring, which cannot fully reflect the overall evolution of the roof and surrounding rock, resulting in one-sided monitoring results and making it difficult to provide a reliable basis for support decisions. This invention, by deploying "sound, light, and electricity" detectors on both sides of the roadway and the roof, and simultaneously collecting multi-parameter mine pressure monitoring data, establishes a unified time reference and forms a multi-dimensional monitoring sequence across modes and locations. This allows for a holistic characterization of roof fracture, loosening, and support stress state, significantly improving the comprehensiveness and reliability of monitoring.

[0017] In the background technology, it is difficult to unify multi-source monitoring data in the spatiotemporal dimension, resulting in results that can only be presented as discrete points or local parameters, lacking the expression of a continuous field. This invention performs spatiotemporal registration of wave velocity profiles, fracture parameter sets, resistivity contour maps, and mine pressure monitoring data sequences, and uses a spatiotemporal feature pyramid network to extract multi-scale heterogeneous features, outputting the partitioned probability distributions of elastic, plastic, fractured, and loosened zones, thereby forming a continuous probability field across the entire cross-section and process, achieving refined partitioning and visual identification of the roof state.

[0018] Existing methods typically rely on numerical simulations or empirical formulas to estimate the thickness and mechanical parameters of the loosened zone, lacking dynamic coupling with real-time monitoring data, resulting in insufficient inversion accuracy. This invention introduces an improved XGBoost algorithm, which combines wave velocity, fracture, and resistivity characteristics with mine pressure data to perform multi-objective inversion of loosened zone thickness, Poisson's ratio, and elastic modulus. Through machine learning for fitting and dynamically correcting nonlinear relationships, the accuracy and real-time performance of mechanical parameter inversion are significantly improved.

[0019] Traditional methods cannot extrapolate the overall displacement, curvature, and rotation information of the roof from discrete monitoring points, making it difficult to identify potential catastrophic processes. This invention, after obtaining the zonal probability distribution, uses it as spatial weights and integrates delamination, anchor bolt and cable force sequences to reconstruct the roof displacement, curvature, and rotation fields, forming a transparent global attitude projection. Further combined with safety threshold discrimination, it can generate early warning signals and output zonal maps and monitoring reports, truly achieving closed-loop support coordination from perception, analysis, to decision-making. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A flowchart of the method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the system structure provided in an embodiment of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] The purpose of this invention is to provide a method and system for multimodal information acquisition and roof attitude monitoring. By integrating multimodal information acquisition, spatiotemporal feature fusion, intelligent inversion and attitude reconstruction, it solves the shortcomings of existing technologies in terms of single monitoring methods, insufficient data fusion, low parameter inversion accuracy and invisible attitude, and significantly improves the scientific nature of roof monitoring and the timeliness of early warning.

[0024] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0025] Figure 1 The method flowchart provided in the embodiments of the present invention is as follows: Figure 1 As shown, this invention provides a method for multimodal information acquisition and top plate global attitude monitoring, including: Step 100: Arrange fan-shaped boreholes and annular electrode arrays on both sides and the roof of the roadway, install ultrasonic detectors, borehole optical sights, resistivity monitoring devices, and simultaneously arrange roof separation instruments, anchor bolt force gauges, anchor cable force gauges, borehole stress gauges, and support working resistance acquisition units on the cross-section to establish a unified time reference and obtain ultrasonic data sequences, borehole wall image sequences, electrode array potential data sequences, and mine pressure monitoring data sequences. Step 200: Perform wavelet denoising and first wave arrival time extraction on the ultrasonic data sequence to form a wave velocity profile; perform intelligent crack identification on the borehole wall image sequence to obtain a crack parameter set; use least squares inversion to generate resistivity contour maps on the electrode array potential data sequence; and perform denoising and anomaly removal on the mine pressure monitoring data sequence. Step 300: Perform spatiotemporal registration of wave velocity profile, fracture parameter set, resistivity contour map and preprocessed mine pressure monitoring data sequence, input spatiotemporal feature pyramid network, and output the partition probability distribution of elastic zone, plastic zone, fracture zone and loosening zone. Step 400: Using wave velocity profiles, fracture parameter sets, resistivity contour maps, and mine pressure monitoring data sequences as features, an improved XGBoost joint inversion method is used to retrieve the loosened zone thickness, Poisson's ratio, and elastic modulus. Step 500: Using the partition probability distribution as spatial weight, merge the delamination amount, anchor bolt force sequence and anchor cable force sequence in the mine pressure monitoring data sequence to reconstruct the roof displacement field, curvature field and rotation field; Step 600: Based on the preset safety threshold, the displacement field, curvature field, rotation field and mine pressure monitoring data sequence are judged, an early warning is generated, and the attitude zoning map and monitoring report are output for support collaborative control.

[0026] In the specific implementation of step 100, fan-shaped boreholes are first drilled on both sides and the roof of the tunnel. These boreholes serve as dedicated channels for ultrasonic detection, used to house ultrasonic detectors and form ultrasonic acquisition stations. By acquiring ultrasonic signals once per shift, the wave velocity changes of the media within the roof and surrounding rock can be obtained, forming an ultrasonic data sequence. To further observe the development of borehole wall fractures, a borehole optical sight is placed in the same borehole as the ultrasonic borehole. This sight can acquire image information of the borehole wall fractures, and a borehole wall image sequence is obtained after daily acquisition. These borehole wall images are used for subsequent intelligent fracture identification. The identification results can be converted into three types of indicators: fracture density, fracture width, and fracture dip rose diagram. These three are collectively referred to as the "fracture parameter set," used to quantitatively describe the development characteristics and spatial distribution patterns of fractures in the surrounding rock.

[0027] In practice, a ring-shaped electrode array covering the roof and sidewalls is also deployed on the roadway surface. This array, composed of a uniformly distributed multi-electrode unit, monitors the resistivity of the roof and sidewall rock mass. During monitoring, the array performs real-time continuous data acquisition, obtaining potential difference data between the electrodes and forming an electrode array potential data sequence. This data sequence, after subsequent inversion calculations, generates a resistivity contour map. This map visually reflects the spatial distribution and anomalous areas of electrical parameters within the roof and surrounding rock, providing a basis for identifying loosened zones and fracture extent.

[0028] In practical implementation, roof separation meters, anchor bolt force gauges, anchor cable force gauges, borehole stress gauges, and support working resistance acquisition units are simultaneously installed on cross-sections spaced at intervals in the working face. The roof separation meter monitors the separation between the roof and the surrounding rock layer; the anchor bolt and anchor cable force gauges monitor the stress state of the anchoring system; the borehole stress gauge records stress changes within the rock mass; and the support working resistance acquisition unit collects real-time data on the load-bearing capacity of the hydraulic support. These devices are linked together using a unified time reference to obtain data on separation amount, anchor bolt force, anchor cable force, borehole stress, and support working resistance, forming a mine pressure monitoring data sequence. This data sequence dynamically reflects the stress and deformation process of the roadway surrounding rock and support system, corroborating the aforementioned ultrasonic, optical, and resistivity monitoring results, providing a reliable data foundation for subsequent multimodal fusion and mechanical parameter inversion.

[0029] In the specific implementation of step 200 of this embodiment, the acquired ultrasonic data sequence is first filtered and denoised, preferably using wavelet transform to remove high-frequency noise from the signal, thereby retaining the effective components related to the rock mass structure. Subsequently, first-arrival time extraction is performed on the denoised signal to identify the time point when the ultrasonic wave first arrives in the rock mass. This first-arrival time information, combined with borehole geometry parameters and propagation path, can be used to calculate the wave velocity distribution of the rock mass and ultimately generate a wave velocity profile. The wave velocity profile can intuitively reflect the integrity of the roof and surrounding rock; low-velocity anomaly zones typically correspond to fracture development or loosened zones, providing basic data for subsequent mechanical parameter inversion.

[0030] In practical implementation, the borehole wall image sequence is input into the fracture intelligent identification module. Fracture features are extracted using image processing and pattern recognition methods to obtain fracture density, fracture width, and fracture dip rose diagrams, forming a fracture parameter set. This "fracture parameter set" is a dataset used to quantitatively describe the degree of fracture development and spatial distribution characteristics of the rock mass, reflecting the fracture status of the surrounding rock structure. Simultaneously, a least-squares optimized resistivity inversion method is applied to the electrode array potential data sequence to generate resistivity contour maps. These resistivity contour maps reveal differences in conductivity within the rock mass; resistivity anomalies are often highly correlated with water-bearing, fractured, or loosened areas, and can corroborate the results of ultrasonic and fracture identification.

[0031] In practice, the mine pressure monitoring data sequence undergoes preprocessing. First, the raw acquired signals are denoised to eliminate electrical interference and environmental noise. Then, according to the data processing and statistical analysis workflow of the monitoring system, the data undergoes consistency verification. Based on pre-set stress, strain, and displacement safety thresholds, data exceeding these thresholds are marked and outliers are removed to avoid misjudgments caused by sensor errors or accidental disturbances. After this processing, the resulting preprocessed mine pressure monitoring data sequence can be spatiotemporally fused with wave velocity profiles, fracture parameter sets, and resistivity contour maps, providing high-quality input data for roof attitude reconstruction and early warning.

[0032] In the specific implementation of step 300 of this embodiment, the wave velocity profile, fracture parameter set, resistivity contour map, and mine pressure monitoring data sequence are first timestamped according to their respective acquisition frequencies to establish a unified time indexing system. Ultrasonic data is acquired once per shift, fracture image data is acquired once daily, resistivity data is acquired continuously in real time, and mine pressure monitoring data is acquired continuously in real time. Through the unified processing of the time index, the comparability of data from different sources in the same time dimension can be guaranteed, laying the foundation for subsequent spatiotemporal registration. In practical implementation, cross-sections spaced at intervals on the working face are used as spatial indexes to map various monitoring results to a unified cross-sectional coordinate system. Specifically, wave velocity profiles are mapped to borehole locations, fracture parameter sets are mapped to borehole wall image locations, resistivity contour maps are mapped to the annular electrode arrangement area, and mine pressure monitoring data are mapped to the locations of delamination meters, anchor bolts, and anchor cable force gauges arranged on the cross-sections. Through the unified processing of spatial locations, a spatiotemporal registration relationship corresponding to the cross-sectional coordinate system is formed, ensuring the collaborative fusion of multimodal data in the spatial dimension. In practical implementation, the spatiotemporally registered wave velocity profile, fracture parameter set, resistivity contour map, and mine pressure monitoring data are fed into the spatiotemporal feature pyramid network as four-dimensional inputs. The spatiotemporal feature pyramid network is a hierarchical feature fusion structure used to extract features at multiple scales in time and space. First, key indicators reflecting rock mass damage and failure are extracted, including wave velocity attenuation rate, fracture density, and resistivity gradient abrupt change. These indicators are then fed into the branch networks as input features. Acoustic features, image features, and electrical features are extracted through the branch networks, and multi-scale compression and fusion are performed within the pyramid structure. In practical implementation, the multi-scale spatiotemporal features obtained through pyramid structure fusion are used as the discrimination input, and the probability calculation module outputs the partitioned probability distribution of different regions of the roof. This probability distribution includes four categories: elastic zone, plastic zone, fracture zone, and loosening zone, each corresponding to the mechanical state of the rock mass at different evolution stages. The elastic zone indicates that the rock mass is in a recoverable deformation stage, the plastic zone indicates irreversible deformation, the fracture zone indicates the presence of obvious fractures, and the loosening zone indicates overall loosening and instability of the rock mass. Through these partitioned probability distributions, a quantitative basis can be provided for subsequent mechanical parameter inversion and roof attitude reconstruction. In the specific implementation of step 400 of this embodiment, the wave velocity profile, fracture parameter set, resistivity contour map, and preprocessed mine pressure monitoring data sequence obtained in the preceding steps are used as feature inputs. The wave velocity profile reflects the spatial variation of wave velocity in the rock mass medium; the fracture parameter set (including fracture density, fracture width, and fracture dip rose diagram) is used to quantitatively characterize the degree of fracture development in the borehole wall; the resistivity contour map reveals the differences in conductivity within the rock mass; and the mine pressure monitoring data sequence includes parameters such as delamination amount, anchor bolt force, anchor cable force, borehole stress, and support working resistance. These input features together constitute a multimodal feature vector describing the mechanical state of the roof rock mass, providing a unified data foundation for subsequent mechanical parameter inversion. In practical implementation, an improved XGBoost model is used to train and infer the aforementioned multimodal features. XGBoost is a machine learning method based on gradient boosting trees, which has the advantages of handling high-dimensional sparse features and strong generalization ability. To adapt to the scenario of this invention, the model is improved, including introducing partitioned probability distributions as spatial weights for sample weighting and optimizing the tree structure depth and splitting criteria to improve the fitting ability to nonlinear features. In this way, the model can more accurately capture the complex relationship between acoustic, electrical, and stress responses, achieving a fine inversion of rock mass parameters. In practice, an improved XGBoost model is used as the core to perform joint inversion of loosened zone thickness, Poisson's ratio, and elastic modulus. "Joint inversion" means solving for multiple objective parameters simultaneously within a unified computational framework, rather than calculating them independently. Loosened zone thickness characterizes the extent of roof failure, Poisson's ratio reflects the ratio of the rock mass's lateral to longitudinal deformation capacity, and elastic modulus measures the rock mass's resistance to deformation. By outputting these three types of parameters as parallel parameters during training and inference, the mutual constraints between parameters are preserved, thus making the results more consistent with actual geomechanical characteristics.

[0033] In practical implementation, the loosened zone thickness distribution, Poisson's ratio field, and elastic modulus field obtained through the improved XGBoost joint inversion are organized according to the cross-sectional location and time index to generate corresponding spatiotemporal inversion results. These inversion results can not only be used as independent indicators for roof stability analysis, but also be coupled with the aforementioned zonal probability distribution to form a two-dimensional identification framework combining zonal probability and mechanical parameters. This approach allows for the acquisition of both the spatial probability distribution of rock mass failure zones and the quantitative parameters of the rock mass's mechanical properties, achieving a complete chain from qualitative identification to quantitative assessment, providing solid support for subsequent roof attitude reconstruction and safety early warning. As an example, the improved XGBoost of this invention uses multimodal features consisting of "wave velocity profile, fracture parameter set, resistivity contour map, and mine pressure monitoring data sequence" as a unified input, maintaining consistency with the data organization after spatiotemporal registration, and achieving unified representation across modes, locations, and times. On the output side, the model uses loosened zone thickness, Poisson's ratio, and surrounding rock elastic modulus as parallel objectives, forming a "single input - multiple objectives" joint inversion structure, so that different mechanical parameters are consistently estimated within the same framework, avoiding coupling distortion caused by separate estimation. This approach of "using sound velocity, fracture parameters, resistivity, etc. as feature parameters → inverting and calculating loosened zone thickness, Poisson's ratio, and surrounding rock elastic modulus" is consistent with the disclosed information, only changing the objectives to parallel outputs in order to maintain the inherent constraint relationship between parameters. The model aligns with the upstream "spatiotemporal feature pyramid network" results at the data interface: first, the pyramid process completes the spatiotemporal registration and feature extraction of multi-source data (including the acquisition of wave velocity attenuation rate, fracture density, and resistivity abrupt gradients), and then the features consistent with the cross-section location and time index are input into the improved XGBoost. In this way, the upstream is responsible for "extraction and compression of multi-scale, heterogeneous features", and the downstream is responsible for "joint inversion of mechanical parameters based on these features". The two are seamlessly connected under the same cross-section coordinate system and time index, ensuring the consistency and traceability of training and inference data.

[0034] During runtime, the input features (wave velocity, fracture, resistivity, and mine pressure) are first organized according to the established unified time index and cross-sectional coordinate system, and then fed into the improved XGBoost with cross-section × time as the sample unit. In the training phase, multi-objective joint optimization is adopted to directly obtain the regressors of the three mechanical parameters. In the inference phase, the newly arrived data is processed in the same index and then entered into the model, outputting the loosening zone thickness, Poisson's ratio, and elastic modulus sequence on the cross-section-time grid for subsequent attitude reconstruction and early warning stages.

[0035] Furthermore, in this specific implementation, the embodiment first calculates the wave velocity attenuation rate from the wave velocity profile to characterize the spatial variation of wave velocity in the rock mass. The wave velocity attenuation rate refers to the degree to which the speed of ultrasound waves decreases with spatial distance during propagation in the surrounding rock, reflecting the internal density and integrity of the medium. Subsequently, the fracture density is extracted from the fracture parameter set. Fracture density is the number of fractures per unit length or unit area, used to measure the degree of rock fragmentation. Then, the resistivity abrupt change gradient is calculated from the resistivity contour map. This gradient characterizes the abrupt boundary of the rock mass's electrical parameters in space, revealing areas of electrical anomalies and potential damage zones. These three features represent the acoustic, image fracture, and electrical dimensions, respectively, and are the core inputs for subsequent partitioning and discrimination.

[0036] In practice, acoustic features, crack features, and electrical features are fed into corresponding branch networks for processing. A branch network is a sub-network structure built for different types of data within a unified neural network framework. Each branch extracts the most discriminative feature representation based on the characteristics of the input features. For example, the acoustic branch network uses convolution to extract wave velocity attenuation rate sequences to identify velocity change patterns; the image crack branch performs feature compression on indicators such as crack density to identify structural development trends; and the electrical branch performs nonlinear transformation on resistivity gradients to capture abrupt boundary information. This branching processing method ensures that heterogeneous data can extract effective features independently without interfering with each other, maintaining fusionability within the unified framework.

[0037] In practical implementation, the heterogeneous features extracted from each branch network are fed into a spatiotemporal feature pyramid network for multi-scale convergence and compression. The pyramid structure refers to the expression of features at different scales through hierarchical convolution and downsampling mechanisms, thereby capturing spatiotemporal features that possess both global trends and local details. In this invention, the pyramid structure can simultaneously preserve the large-scale evolution patterns of the surrounding rock and local fracture anomalies. Finally, through the fused spatiotemporal feature representation, a probabilistic discriminant model is established, outputting the partitioned probability distribution of the roof region, including four categories: elastic zone, plastic zone, fracture zone, and loosened zone. The probability value of each category represents the likelihood of the surrounding rock being in that state, thus achieving quantification and visualization of the roof state across the entire cross-section and the entire time domain.

[0038] In the specific implementation of step 500 of this embodiment, under each cross-section and time index, the probability distribution of the elastic zone, plastic zone, fracture zone, and loosening zone is first normalized to obtain a spatial weight set that corresponds one-to-one with the cross-section coordinates. The spatial weight set is used to express the relative probability of the same location in the four states and is used as the weighting coefficient during subsequent fusion. Subsequently, using the cross-sections arranged at work face intervals as spatial indices, the delamination amount, anchor bolt force sequence, and anchor cable force sequence are mapped to a unified cross-section coordinate system according to their observation point positions, and a one-to-one index relationship is established with the corresponding spatial weight set according to position and time, forming a spatiotemporal correspondence for fusion and reconstruction.

[0039] In practical implementation, the delamination amount is weighted and continuous based on the spatial weight set: on the cross-sectional coordinates, according to the grid or measuring point topology, the delamination amount at each location is weighted and summarized according to the probability weights of four types of partitions to obtain the displacement estimate at that location, and the continuity and monotonicity requirements are maintained between adjacent locations, thus forming the cross-sectional displacement distribution. To improve the consistency between the displacement estimate and the anchoring force response, force constraints and checks are introduced at the anchoring locations of the anchor bolts and anchor cables. The observed anchor bolt force sequences and anchor cable force sequences are used to check the consistency of the displacement at this location and make necessary amplitude adjustments to ensure that the displacement change trend and the force change trend remain compatible, and finally obtain the top plate displacement field that meets the monitoring constraint conditions.

[0040] In practical implementation, the curvature distribution is calculated based on the spatial variation relationship of the roof displacement field under a unified cross-sectional coordinate system, resulting in a curvature field that characterizes the degree of curvature of the roof on a spatial scale. Under adjacent time indices, the rotation is calculated based on the directional variation relationship of the roof displacement field, resulting in a rotation field that characterizes the rotational features of the roof attitude on a time scale. The displacement field, curvature field, and rotation field are organized and stored according to cross-sectional coordinates and time indices, and serve as input data for subsequent early warning judgment and visualization output, supporting the formation of attitude zoning maps and monitoring reports.

[0041] In the specific implementation of step 600 of this embodiment, a preset safety threshold library is first established based on the site design specifications, historical monitoring baselines, and trial operation data. The threshold library includes displacement limits for key locations in the displacement field, curvature limits for the curvature field, rotation limits for the rotation field, and control limits for delamination, anchor bolt force, anchor cable force, borehole stress, and support working resistance in the mine pressure monitoring data sequence, while also providing the corresponding units and value ranges. To ensure consistency in the judgment, the displacement field, curvature field, and rotation field are organized using cross-sectional coordinates and time indices, and aligned with the mine pressure monitoring data sequence under the same cross-section and time index, forming an input set for threshold judgment. For indicators with different dimensions, units are unified according to the dimensional descriptions and conversion relationships in the threshold library to ensure comparability and accuracy in subsequent comparisons.

[0042] In practice, a combination of single-index exceedance judgment, combined relationship judgment, and duration judgment is used to determine the safety status. Single-index exceedance judgment compares the corresponding values ​​of displacement, curvature, rotation, and mine pressure monitoring data sequences under the cross-section and time index with a threshold library; any index exceeding its control limit triggers an exceedance flag. Combined relationship judgment marks a location as a high-concern area when simultaneous increases in displacement, curvature, or rotational abrupt changes are accompanied by abnormal changes in delamination or anchorage force. Duration judgment continuously checks adjacent time indices; if the exceedance status does not recover within a preset minimum duration, a warning is confirmed. To reduce the impact of occasional noise, a moving average or median filter can be applied to the input set before judgment to smooth abnormal peaks. Simultaneously, by combining the spatial weights of the zonal probability distribution, sensitivity is increased for locations with a higher probability of being in fractured or loosened zones, and the false alarm probability is reduced for locations with a higher probability of being in elastic zones, thus making the warning more closely reflect the actual state of the surrounding rock.

[0043] In practice, once an early warning is confirmed, the system generates an attitude zoning map and a monitoring report. The attitude zoning map overlays the probability distributions of four types of zones and the contour lines of the displacement field, curvature field, and rotation field under a unified cross-sectional coordinate system, and prominently annotates the location of the exceedance, the name of the exceedance index, and the exceedance magnitude. The monitoring report records the time index, cross-sectional location, relevant sensor number, threshold type, measured value, threshold value, duration, and processing status, and can include trend curves of the mine pressure monitoring data sequence for source tracing analysis. These charts serve as the basis for coordinated control of support systems and can be used to prompt targeted inspections and adjustments to the anchoring system and support parameters on-site, and to arrange reinforcement or retesting procedures when necessary. All early warning records and processing results are archived according to cross-sectional coordinates and time indexes, forming a traceable monitoring and handling closed loop, which facilitates subsequent review and long-term safety assessment.

[0044] Corresponding to the above methods, such as Figure 2 As shown, this embodiment also provides a multimodal information acquisition and top plate global attitude monitoring system, including: The monitoring deployment and synchronization unit is used to deploy fan-shaped boreholes and ring electrode arrays on both sides and the roof of the roadway, install ultrasonic detectors, borehole optical sights, and resistivity monitoring devices, and simultaneously deploy roof separation instruments, anchor bolt force gauges, anchor cable force gauges, borehole stress gauges, and support working resistance acquisition units on the cross-section to establish a unified time reference and obtain ultrasonic data sequences, borehole wall image sequences, electrode array potential data sequences, and mine pressure monitoring data sequences. The acquisition and preprocessing unit is used to perform wavelet denoising and first wave arrival time extraction to form a wave velocity profile on the ultrasonic data sequence, perform intelligent crack identification on the borehole wall image sequence to obtain a crack parameter set, use least squares inversion to generate resistivity contour maps on the electrode array potential data sequence, and perform denoising and anomaly removal on the mine pressure monitoring data sequence. The spatiotemporal registration and partitioning unit is used to perform spatiotemporal registration of the wave velocity profile, the fracture parameter set, the resistivity contour map and the preprocessed mine pressure monitoring data sequence. It inputs a spatiotemporal feature pyramid network and outputs the partition probability distribution of elastic zone, plastic zone, fracture zone and loosening zone. The joint inversion unit is used to jointly invert the loosened zone thickness, Poisson's ratio and elastic modulus using improved XGBoost, based on wave velocity profiles, fracture parameter sets, resistivity contour maps and mine pressure monitoring data sequences. The attitude reconstruction unit is used to reconstruct the roof displacement field, curvature field and rotation field by fusing the delamination amount, anchor bolt force sequence and anchor cable force sequence in the mine pressure monitoring data sequence with the partition probability distribution as spatial weight; The early warning and output unit is used to identify the displacement field, curvature field, rotation field and mine pressure monitoring data sequence based on a preset safety threshold, generate an early warning, and output an attitude zoning map and monitoring report for support and coordinated control.

[0045] The beneficial effects of this invention are as follows: 1) This invention first achieves unified acquisition of "sound, light, electricity" and mine pressure data on both sides and roof of the roadway, and uses a unified time reference and cross-sectional coordinate organization as the sequential input, directly solving the problems of limited perspective of single monitoring methods and fragmented data from multiple sources. By synchronously arranging fan-shaped boreholes, ring electrode arrays, and cross-sections, an observation network covering the roof and sides is formed, which can simultaneously obtain wave velocity profiles, fracture parameter sets, resistivity contour maps, and mine pressure monitoring data sequences, significantly improving spatial coverage and information completeness, and reducing monitoring blind spots.

[0046] 2) The spatiotemporal registration and spatiotemporal feature pyramid network proposed in this invention maps multimodal results to a unified time index and cross-sectional coordinate system, and extracts and compresses features at multiple scales, outputting the probability distribution of elastic, plastic, fractured, and loosened zones. Compared with traditional methods that only provide discrete point conclusions, this scheme expresses the evolution of surrounding rock with a continuous probability field, making the zoning boundaries clearer and the transition zones more identifiable, thereby reducing the uncertainty in judgment caused by zoning bias and providing stable input for subsequent inversion and early warning.

[0047] 3) Regarding the acquisition of mechanical parameters, this invention employs an improved XGBoost for joint inversion, using wave velocity profiles, fracture parameter sets, resistivity contour maps, and mine pressure monitoring data sequences as features to simultaneously solve for the loosened zone thickness, Poisson's ratio, and elastic modulus. Joint inversion maintains the inherent coupling relationship between parameters, which is superior to the consistency loss caused by separate estimations. It can better reflect the actual nonlinear coupled response of the surrounding rock, improving the accuracy and stability of mechanical parameter estimation and providing a reliable basis for quantitative assessment.

[0048] 4) In terms of attitude quantification, this invention uses the partitioned probability distribution as spatial weights, integrating the delamination amount, anchor bolt force sequence, and anchor cable force sequence to construct the roof displacement field, curvature field, and rotation field. This reconstruction path of "probability weight constraint plus monitoring quantity verification" effectively connects the observation data with the meaning of structural mechanics, preserving the objectivity of multi-source monitoring while introducing the spatial prior of partitioned results, forming a continuous field expression from displacement to curvature to rotation, and realizing a transparent representation of the roof's attitude across the entire domain.

[0049] 5) In terms of risk management, this invention discriminates displacement field, curvature field, rotation field, and mine pressure monitoring data sequences based on preset safety thresholds, and adjusts sensitivity by combining spatial weights of zoning probabilities, effectively suppressing false alarms and missed alarms caused by occasional noise. The output attitude zoning map and monitoring reports clearly visualize the over-limit locations, indicators, and durations, directly serving support coordination and on-site handling decisions, improving the interpretability and enforceability of early warnings, and shortening the decision-making chain from problem discovery to taking measures.

[0050] 6) In terms of engineering implementation, this invention adopts the sensor system and layout commonly used on-site. The steps are modularized, the terminology is standardized, and the data is organized traceably using cross-sectional and time indexes, facilitating phased deployment and maintenance according to the working face's progress. This system is compatible with existing monitoring platforms and reporting processes, making it easy to scale up. Under the same resource conditions, it obtains higher-value evaluation results through information fusion and joint inversion, reducing the cost of repeated experiments and blind reinforcement, and overall improving the efficiency and quality of mine roof monitoring and support management.

[0051] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0052] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for multimodal information acquisition and global attitude monitoring of the roof, characterized in that, include: Fan-shaped boreholes and ring-shaped electrode arrays were arranged on both sides and the roof of the roadway. Ultrasonic detectors, borehole optical sights, and resistivity monitoring devices were installed. At the same time, roof separation instruments, anchor bolt force gauges, anchor cable force gauges, borehole stress gauges, and support working resistance acquisition units were arranged on the cross-section to establish a unified time reference and obtain ultrasonic data sequences, borehole wall image sequences, electrode array potential data sequences, and mine pressure monitoring data sequences. Wavelet denoising and first wave arrival time extraction are performed on the ultrasonic data sequence to form a wave velocity profile; crack intelligent identification is performed on the borehole wall image sequence to obtain a crack parameter set; least squares inversion is used to generate resistivity contour maps on the electrode array potential data sequence; and noise reduction and anomaly removal are performed on the mine pressure monitoring data sequence. The wave velocity profile, the fracture parameter set, the resistivity contour map, and the preprocessed mine pressure monitoring data sequence are spatiotemporally registered, input into a spatiotemporal feature pyramid network, and output the partition probability distribution of elastic zone, plastic zone, fracture zone, and loosening zone. Based on wave velocity profiles, fracture parameter sets, resistivity contour maps, and mine pressure monitoring data sequences, an improved XGBoost joint inversion method was used to retrieve the loosened zone thickness, Poisson's ratio, and elastic modulus. Using the aforementioned partition probability distribution as spatial weights, the delamination amount, anchor bolt force sequence, and anchor cable force sequence in the mine pressure monitoring data sequence are integrated to reconstruct the roof displacement field, curvature field, and rotation field. Based on a preset safety threshold, the displacement field, curvature field, rotation field, and mine pressure monitoring data sequence are judged to generate an early warning, and an attitude zoning map and monitoring report are output for support and coordinated control.

2. The method for multimodal information acquisition and top plate global attitude monitoring according to claim 1, characterized in that, The mine pressure monitoring data sequence includes: delamination amount, anchor bolt force, anchor cable force, borehole stress, and support working resistance.

3. The method for multimodal information acquisition and top plate global attitude monitoring according to claim 1, characterized in that, The set of fracture parameters includes fracture density, width, and dip rose diagram.

4. The method for multimodal information acquisition and top plate global attitude monitoring according to claim 1, characterized in that, Fan-shaped boreholes and annular electrode arrays were arranged on both sides of the roadway and in the roof. Ultrasonic detectors, borehole optical sights, and resistivity monitoring devices were installed. Simultaneously, roof separation meters, anchor bolt force gauges, anchor cable force gauges, borehole stress gauges, and support working resistance acquisition units were deployed at the cross-sections to establish a unified time reference. This yielded ultrasonic data sequences, borehole wall image sequences, electrode array potential data sequences, and mine pressure monitoring data sequences, including: Fan-shaped boreholes for ultrasonic detection are drilled on both sides and the roof of the tunnel to serve as acquisition channels for ultrasonic data sequences, forming an ultrasonic acquisition station. The acquisition frequency is once per shift to obtain the raw ultrasonic records. An optical borehole sight was installed in the same hole as the ultrasonic borehole to observe the development of cracks in the borehole wall. The image was collected once a day to obtain a sequence of borehole wall images. A ring-shaped electrode array covering the sidewalls and roof is arranged on the surface of the roadway for resistivity monitoring. Real-time continuous monitoring is performed to obtain the potential data sequence of the electrode array. Cross sections are arranged at intervals on the working face, and roof separation instrument, anchor bolt force gauge, anchor cable force gauge, borehole stress gauge and support working resistance acquisition unit are synchronously installed on the cross sections for real-time monitoring of the stress and deformation of the surrounding rock and anchoring system in the roadway, so as to obtain the mine pressure monitoring data sequence.

5. The method for multimodal information acquisition and top plate global attitude monitoring according to claim 1, characterized in that, The ultrasonic data sequence is subjected to wavelet denoising and first wave arrival time extraction to form a wave velocity profile; the borehole wall image sequence is subjected to intelligent fracture identification to obtain a fracture parameter set; the electrode array potential data sequence is subjected to least squares inversion to generate resistivity contour maps; and the mine pressure monitoring data sequence is subjected to denoising and anomaly removal, including: The ultrasonic data sequence is filtered and denoised, and the arrival time of the first wave is extracted to generate a wave velocity profile. Intelligent crack recognition is performed on the image sequence of the pore wall, and crack density, crack width and crack inclination rose diagram are output to form the crack parameter set used to describe the crack development characteristics; The resistivity inversion of the electrode array potential data sequence is optimized using least squares to generate a resistivity contour map. Based on the monitoring system's processing and statistical analysis procedures for mine pressure data, the mine pressure monitoring data sequence is processed. Combined with the pre-set stress, strain, and displacement safety thresholds, data exceeding the thresholds are marked and removed to obtain a pre-processed mine pressure monitoring data sequence for subsequent registration and fusion.

6. The method for multimodal information acquisition and top plate global attitude monitoring according to claim 1, characterized in that, The wave velocity profile, the fracture parameter set, the resistivity contour map, and the preprocessed mine pressure monitoring data sequence are spatiotemporally registered. The spatiotemporal feature pyramid network is input, and the probability distributions of the elastic zone, plastic zone, fracture zone, and loosened zone are output, including: The timestamps of the wave velocity profile, the fracture parameter set, the resistivity contour map, and the mine pressure monitoring data sequence are aligned according to their respective acquisition frequencies to form a time index for fusion. Using the cross-sections arranged at intervals on the working face as spatial indexes, the wave velocity profile, the fracture parameter set, the resistivity contour map, and the mine pressure monitoring data sequence are mapped to a unified cross-section coordinate system according to the cross-section position and the borehole position to complete spatial alignment and obtain the spatiotemporal correspondence for fusion. The spatiotemporally registered wave velocity profile, fracture parameter set, resistivity contour map and mine pressure monitoring data sequence are used as inputs to the four-dimensional data and fused into the spatiotemporal feature pyramid network. Following the process of extracting wave velocity attenuation rate, crack density and resistivity abrupt gradient, using branch network to extract heterogeneous features, and using pyramid structure to compress multi-scale spatiotemporal features, the probability distribution of elastic region, plastic region, fracture region and loosening region is obtained.

7. The method for multimodal information acquisition and top plate global attitude monitoring according to claim 6, characterized in that, Following the process of extracting wave velocity attenuation rate, fracture density, and resistivity abrupt gradients, using a branch network to extract heterogeneous features, and employing a pyramid structure to compress multi-scale spatiotemporal features, the probability distributions of the elastic, plastic, fracture, and loosened regions are obtained, including: Based on the wave velocity profile, the wave velocity attenuation rate, which characterizes the spatial variation of wave velocity, is calculated and used as an acoustic feature input. Based on the fracture parameter set, the fracture density used to describe the degree of fracture development is extracted and used as the image fracture feature input; Based on the resistivity contour map, the resistivity abrupt gradient used to characterize the boundary of electrical anomalies is calculated and used as the input of electrical features; The acoustic feature input, the image gap feature input, and the electrical feature input are respectively fed into the corresponding branch network to extract heterogeneous feature representations; The output of the branch network is aggregated and compressed in a multi-scale pyramid structure in the spatiotemporal feature pyramid network to form a fused spatiotemporal feature representation. Based on the fused spatiotemporal feature representation, the probability distribution of the elastic region, plastic region, fracture region, and loosening region is output.

8. The method for multimodal information acquisition and top plate global attitude monitoring according to claim 6, characterized in that, Characterized by wave velocity profiles, fracture parameter sets, resistivity contour maps, and mine pressure monitoring data sequences, an improved XGBoost joint inversion method was used to retrieve the loosened zone thickness, Poisson's ratio, and elastic modulus, including: Based on the spatiotemporal registration results, the wave velocity profile, fracture parameter set, resistivity contour map and the mine pressure monitoring data sequence are aligned with the cross sections and time indexes arranged at working face intervals to form a set of characteristic parameters for inversion. Using the set of characteristic parameters as input, the improved XGBoost model is invoked, and the target quantities for joint inversion are set as loosened zone thickness, Poisson's ratio and surrounding rock elastic modulus, and joint inversion calculation is performed. The loosened zone thickness sequence, Poisson's ratio sequence, and surrounding rock elastic modulus sequence under the cross-section-time index are obtained as the inversion results of the mechanical property parameters.

9. The method for multimodal information acquisition and top plate global attitude monitoring according to claim 1, characterized in that, Using the aforementioned partition probability distribution as spatial weights, the delamination amount, anchor bolt force sequence, and anchor cable force sequence from the mine pressure monitoring data sequence are fused to reconstruct the roof displacement field, curvature field, and rotation field, including: Under each cross section and time index, the probability distribution of the elastic zone, plastic zone, fracture zone and loosening zone is normalized to obtain the spatial weight set corresponding to the cross section position, which is used as the fusion weight. The cross-sections arranged at working face intervals map the delamination amount, anchor bolt force sequence and anchor cable force sequence to a unified cross-section coordinate system, and establish a one-to-one spatiotemporal index relationship with the spatial weight set. The delamination amount is weighted and continuous using the spatial weight set to obtain the cross-sectional displacement distribution; the displacement distribution is constrained and checked at the anchoring position using the anchor bolt force sequence and the anchor cable force sequence to form the top plate displacement field. Under the unified cross-sectional coordinate system, the curvature distribution is calculated based on the spatial variation relationship of the top plate displacement field to obtain the curvature field; Under adjacent time indexes, the rotation amount is calculated based on the directional change relationship of the top plate displacement field to obtain the rotation field.

10. A multimodal information acquisition and roof full-domain attitude monitoring system, characterized in that, include: The monitoring deployment and synchronization unit is used to deploy fan-shaped boreholes and ring electrode arrays on both sides and the roof of the roadway, install ultrasonic detectors, borehole optical sights, and resistivity monitoring devices, and simultaneously deploy roof separation instruments, anchor bolt force gauges, anchor cable force gauges, borehole stress gauges, and support working resistance acquisition units on the cross-section to establish a unified time reference and obtain ultrasonic data sequences, borehole wall image sequences, electrode array potential data sequences, and mine pressure monitoring data sequences. The acquisition and preprocessing unit is used to perform wavelet denoising and first wave arrival time extraction to form a wave velocity profile on the ultrasonic data sequence, perform intelligent crack identification on the borehole wall image sequence to obtain a crack parameter set, use least squares inversion to generate resistivity contour maps on the electrode array potential data sequence, and perform denoising and anomaly removal on the mine pressure monitoring data sequence. The spatiotemporal registration and partitioning unit is used to perform spatiotemporal registration of the wave velocity profile, the fracture parameter set, the resistivity contour map and the preprocessed mine pressure monitoring data sequence. It inputs a spatiotemporal feature pyramid network and outputs the partition probability distribution of elastic zone, plastic zone, fracture zone and loosening zone. The joint inversion unit is used to jointly invert the loosened zone thickness, Poisson's ratio and elastic modulus using improved XGBoost, based on wave velocity profiles, fracture parameter sets, resistivity contour maps and mine pressure monitoring data sequences. The attitude reconstruction unit is used to reconstruct the roof displacement field, curvature field and rotation field by fusing the delamination amount, anchor bolt force sequence and anchor cable force sequence in the mine pressure monitoring data sequence with the partition probability distribution as spatial weight; The early warning and output unit is used to identify the displacement field, curvature field, rotation field and mine pressure monitoring data sequence based on a preset safety threshold, generate an early warning, and output an attitude zoning map and monitoring report for support and coordinated control.

Images