Coal mine exploration result visualization and safety monitoring system

Abstract

The invention discloses a coal mine exploration result visualization and safety monitoring system, which is used for carrying out identification, positioning and three-dimensional visualization on geological abnormal areas such as a goaf and a ponding area, and comprises the following steps: collecting underground electromagnetic signals, carrying out preprocessing and feature extraction, and identifying geological abnormal types by combining a deep learning model; and the data are fused with geological drilling data and historical monitoring data to construct a geological spatial information model. The model can be used for generating a three-dimensional image of a geologic structure, a user browses geologic distribution conditions through a visual platform, and mining scheme making and risk assessment are assisted. The system has the function modules of signal acquisition, data processing, abnormity identification, data fusion, three-dimensional modeling, interactive display and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of coal mine safety exploration and geological information processing, and in particular to a coal mine exploration result visualization and safety monitoring system. Background Art

[0002] In coal mining, accurately identifying and locating goaf and waterlogged areas is crucial for ensuring worker safety and improving resource recovery. This is especially true in complex coal seam areas, such as those typically found in coal mines. Due to the widespread presence of damage zones caused by historical small-scale mining, the geological structure exhibits strong heterogeneity, complex interference sources, and unstable signal transmission paths. Traditional geological exploration methods are severely limited in their adaptability and accuracy in such scenarios.

[0003] Existing coal mine exploration technologies mostly rely on traditional nonlinear electromagnetic wave methods, transient electromagnetic methods, or a combination of geophysical exploration and drilling. Although they have some application effects under conventional coal seam conditions, they still have obvious shortcomings in complex geological environments, mainly reflected in the following aspects:

[0004] 1. Single signal reception method: Traditional electromagnetic receiving equipment mostly has fixed sensitivity and frequency bandwidth, which makes it difficult to cope with the underground multi-source interference environment, resulting in the effective signal being masked or the interference being enhanced, affecting subsequent identification and judgment.

[0005] 2. Weak anomaly recognition capability: Existing algorithms usually rely on shallow features or artificial empirical rules, lack deep modeling and spatial coupling analysis of multi-channel nonlinear electromagnetic signals, and cannot achieve high-precision distinction between goaf and waterlogged areas.

[0006] 3. Insufficient data dimension fusion: The current method lacks a mechanism to integrate exploration data with heterogeneous data such as drilling and historical monitoring, resulting in fragmented understanding of geological structures and affecting the accurate judgment of the boundaries and types of abnormal areas.

[0007] 4. Poor 3D modeling and interactive capabilities: Traditional geological models are mostly static or 2D graphical representations, which cannot truly restore the spatial form of underground anomalies and are difficult to support interactive browsing and multiple scheme comparisons.

[0008] Therefore, how to provide a coal mine exploration result visualization and safety monitoring system is a problem that technical personnel in this field urgently need to solve. Summary of the Invention

[0009] One purpose of the present invention is to propose a coal mine exploration results visualization and safety monitoring system. The present invention combines a multi-frequency response adaptive electromagnetic signal receiving array, a deep neural network recognition model, and multi-source data fusion technology to achieve high-precision recognition and three-dimensional spatial modeling of geological anomalies such as goafs and waterlogged areas. It has the advantages of high recognition accuracy, accurate spatial positioning, and strong visualization interactivity.

[0010] A coal mine exploration result visualization and safety monitoring system according to an embodiment of the present invention includes the following steps:

[0011] S1, electromagnetic signal acquisition module, constructs an electromagnetic pulse signal receiving array composed of multiple receiving units with different frequency response characteristics and sensitivities. The receiving units are arranged in a spatially optimized manner. The receiving parameters are adjusted according to the spectrum distribution and signal-to-noise ratio of the received signals of the receiving units to generate the original electromagnetic response data frame;

[0012] S2, signal processing module, pre-processes the original electromagnetic response data frame to form a standardized multi-channel time series signal, and extracts characteristic parameters based on NEMV to generate a structured feature vector set;

[0013] S3, geological identification module, builds a deep neural network model, inputs a structured feature vector set, identifies geological anomaly types in goaf, waterlogged areas, and fracture development areas, and outputs anomaly spatial data including anomaly type, boundary information, and spatial coordinates;

[0014] S4, data fusion module, obtains geological drilling data and historical monitoring data through the configured data access module, fuses them with abnormal spatial data, uses data enhancement methods to expand data dimensions, and generates a geological spatial information model;

[0015] S5, a 3D modeling and visualization module, generates a 3D geological model including geological anomaly areas and their physical parameters based on the geological spatial information model, and converts the 3D geological model into visual graphic data through a graphics rendering engine;

[0016] S6, immersive interaction module, builds a three-dimensional visualization interaction platform, loads visualization graphic data into the three-dimensional visualization interaction platform, and combines virtual reality and augmented reality technologies to achieve multi-angle viewing and interactive operations of geological parameters.

[0017] Optionally, the S1 specifically includes:

[0018] S11. An electromagnetic receiving array construction unit is configured to construct an electromagnetic pulse signal receiving array composed of a plurality of electromagnetic signal receiving units, each electromagnetic signal receiving unit including a coil inductor and a gain adjustment unit, and each electromagnetic signal receiving unit is bound to corresponding three-dimensional spatial coordinate information, the coil inductor having different frequency response characteristics for receiving electromagnetic signals covering low-frequency, medium-frequency, and high-frequency bands, the gain adjustment unit having a parameter setting interface for configuring a gain coefficient and a signal receiving frequency passband parameter range, and the gain coefficient and frequency passband parameters can be updated during an acquisition cycle to adapt to environmental changes;

[0019] S12, a receiving unit arrangement unit, in an initial pre-arrangement state, wherein each electromagnetic signal receiving unit performs a short-cycle sampling to obtain an original electromagnetic waveform signal, analyzes the frequency band amplitude and background noise ratio of the signal at different receiving positions, sets an arrangement spacing of the electromagnetic signal receiving units based on the analysis results, and arranges the electromagnetic signal receiving units according to a predefined geometric structure, wherein the geometric structure arrangement of the receiving units includes a grid arrangement, a spiral arrangement, and a radial arrangement;

[0020] S13, synchronous acquisition unit, each electromagnetic signal receiving unit synchronously acquires the electromagnetic pulse response signal generated by underground excitation according to the set unified time stamp, and the acquired signal forms the original time domain waveform data;

[0021] S14, spectrum analysis unit, performs frequency domain transformation processing on the collected original time domain waveform data, and extracts the amplitude response s of each frequency component f (t), and calculate the frequency band energy value E according to the amplitude response in the target frequency band f , and at the same time extract the background noise amplitude in the non-target frequency band area and form the noise distribution parameters:

[0022]

[0023] Among them, E f is the energy value of frequency band f, t0 and t1 are the start and end time of the analysis time window respectively;

[0024] S15, a parameter adjustment unit, which calculates the signal-to-noise ratio of each frequency band component according to the acquired frequency band energy value and the corresponding noise distribution parameter, and adjusts the gain coefficient and frequency passband parameter of each electromagnetic signal receiving unit in real time;

[0025] S16. The data encapsulation output unit uniformly encodes the original time domain waveform data, gain coefficient and frequency passband parameter to form an original electromagnetic response data frame, wherein the original electromagnetic response data frame includes the acquisition time, receiving unit number, spatial coordinates and current gain configuration status.

[0026] Optionally, the S11 specifically includes:

[0027] S111. A receiving unit frequency structure design unit is configured to construct multiple electromagnetic signal receiving units with differentiated frequency responses. Each electromagnetic signal receiving unit is equipped with a coil sensor for sensing underground electromagnetic pulse signals. The frequency response range of each receiving unit is defined as:

[0028] R i =[f i,min ,f i,max ];

[0029] Among them, R irepresents the frequency range that the i-th electromagnetic signal receiving unit can effectively respond to, f i,min Indicates the minimum response frequency of the electromagnetic signal receiving unit, f i,max Indicates the maximum response frequency, multiple R i There is partial overlap between them, which is used to construct the overall continuous spectrum coverage interval;

[0030] S112, a gain parameter setting unit configures an independent gain adjustment path for each electromagnetic signal receiving unit. The gain coefficient of each electromagnetic signal receiving unit is expressed as G i , used to control the amplitude amplification of its output signal, the gain coefficient value meets the following restrictions:

[0031] G i ∈[G min ,G max ];

[0032] Among them, G min The minimum gain value, G max The maximum gain value that can be set is used to adjust the weak signal sensing strength.

[0033] S113, the frequency passband configuration unit, according to the frequency range R that the electromagnetic signal receiving unit can effectively respond to i and the gain coefficient G of the electromagnetic signal receiving unit i , set the signal receiving frequency passband parameters of the corresponding receiving unit. The frequency passband parameters are the frequency window that actually receives the effective signal during sampling, and must be its response interval R i The adjustment mode is dynamically adjusted according to the set gain sensitivity;

[0034] S114. After each acquisition cycle, the set gain coefficient G is updated according to the update strategy set by the system. i The parameter update operation is performed based on the configured frequency passband parameters. The update operation is performed during the acquisition interval.

[0035] Optionally, the S2 specifically includes:

[0036] S21, a time series signal construction unit, which obtains a multi-channel waveform signal in the original electromagnetic response data frame, extracts the electromagnetic signal time series corresponding to each electromagnetic signal receiving unit number, aligns them according to a set unified time reference point, and constructs an initial multi-channel time series matrix with a dimension of C×T, where C is the number of electromagnetic signal receiving units and T is the length of the electromagnetic signal time series;

[0037] S22, the pre-processing execution unit performs normalization, band-pass filtering and noise suppression on the initial multi-channel time series matrix in sequence. The normalization process uses the range normalization method to map the signal value of each channel to the [0, 1] interval to generate a standardized multi-channel time series signal:

[0038]

[0039] Among them, x i,t represents the original multi-channel time series signal of channel i at time t, min(x i )、max(x i ) represent the minimum and maximum values ​​of the original channel i time series signal, x' i,t To standardize multi-channel timing signals;

[0040] S23, NEMV feature extraction unit, performs nonlinear electromagnetic vector modeling on the standardized multi-channel time series signal, extracts the spatiotemporal coupling characteristic parameters at each moment, and constructs a geologically relevant high-dimensional vector set. The spatiotemporal coupling characteristic parameters include instantaneous energy density, vector direction change rate, and vector amplitude extreme point position;

[0041] S24, a structured conversion unit, reconstructs the dimensions and standardizes the format of the geologically relevant high-dimensional vector set, and outputs a structured feature vector set.

[0042] Optionally, the S3 specifically includes:

[0043] S31, a geological identification model construction unit, constructing a deep neural network model for geological anomaly classification, wherein the deep neural network model consists of an input layer, a plurality of hidden layers, and an output layer, wherein the input layer receives a structured feature vector set, the hidden layer adopts a fully connected network structure, and the number of nodes in the output layer is N, where N represents the number of geological anomaly types;

[0044] S32, a model training execution unit, constructs a training sample data set containing anomaly types labeled in goaf, waterlogged areas, and fracture development areas. The training sample data set includes structured feature vectors and corresponding geological anomaly labels. A cross entropy loss function is used to train a deep neural network model to generate a trained deep neural network model L:

[0045]

[0046] Among them, y j is the one-hot encoding value of the actual label, is the predicted probability of the deep neural network model outputting the jth type of anomaly, and N is the number of geological anomaly types;

[0047] S33, an anomaly recognition unit, calling the trained deep neural network model, performs recognition processing on the structured feature vector set, and outputs the geological anomaly type classification label, recognition confidence value and index position corresponding to each input vector;

[0048] S34. The abnormal spatial data generating unit generates abnormal spatial data including abnormal type, boundary information and three-dimensional spatial position based on the geological abnormality classification label and index position and the corresponding three-dimensional spatial coordinate information bound to each electromagnetic signal receiving unit.

[0049] Optionally, the S32 specifically includes:

[0050] S321, a training data preparation unit collects historical geological data of coal mines with known geological anomaly types, obtains structured feature vector samples corresponding to goaf areas, waterlogged areas, and fracture development areas, and forms an input sample set;

[0051] S322, a label encoding unit, adding a manually annotated geological anomaly type label to each input sample set, wherein the geological anomaly type label is represented in a one-hot encoding form. If the total number of geological anomaly types is N, then the label vector is a binary vector of dimension N, where only the jth component is 1 and the rest are 0, indicating the jth type of anomaly;

[0052] S323, a loss function definition unit, defining a cross entropy loss function used in the training process, which is used to measure the difference between the geological anomaly type label predicted by the deep neural network model and the actual geological anomaly type label;

[0053] S324, the model training control unit performs iterative training operations on the deep neural network model based on the input sample set and the geological anomaly type label, optimizes the model using the batch gradient descent method, and outputs the trained deep neural network model when the training loss drops to the maximum training round.

[0054] Optionally, the S4 specifically includes:

[0055] S41. The external data access unit obtains geological drilling data and historical monitoring data through a configured data access module. The geological drilling data includes borehole number, drilling depth interval, lithologic layering information, and measurement parameters. The historical monitoring data includes the occurrence time, spatial coordinates, and corresponding monitoring index values ​​of underground abnormal events.

[0056] S42, an abnormal data matching unit, using the abnormal spatial data as a matching reference, performs registration and matching of the received geological drilling data and the historical monitoring data according to spatial coordinates, extracts drilling samples and monitoring records adjacent to the abnormal spatial position, and establishes a corresponding relationship between the drilling samples and monitoring records adjacent to the abnormal spatial position;

[0057] S43, a data enhancement processing unit, constructing a joint feature space based on the established correspondence between drilling samples and monitoring records adjacent to the abnormal spatial position, merging the abnormal spatial data with the matched geological drilling data and historical monitoring data, using a spatial interpolation method to fill in the sparsely sampled areas, and generating a fused multidimensional feature matrix;

[0058] S44, the spatial information modeling unit, takes the fused multidimensional feature matrix as input, organizes it into structures according to the spatial coordinates of the abnormal spatial data, uses data enhancement methods to expand the data dimension, and generates a geological spatial information model.

[0059] The beneficial effects of the present invention are:

[0060] (1) The present invention constructs an electromagnetic signal receiving array with multi-frequency response and adaptive adjustment capabilities, combines standardized multi-channel time series signal processing and nonlinear electromagnetic vector feature extraction methods, and realizes high-precision signal acquisition and key feature extraction of goaf and waterlogged areas in complex geological environments, effectively improving the anti-interference and accuracy of geological anomaly identification.

[0061] (2) The present invention utilizes a deep neural network model to perform classification training and identification on structured feature vectors, and combines historically labeled samples with a cross-entropy loss function optimization strategy to achieve intelligent identification of multiple types of geological anomalies, significantly improving the generalization ability of the model and geological identification accuracy, thus meeting the classification requirements in complex scenarios.

[0062] (3) The present invention integrates drilling data and historical monitoring data, constructs a joint feature space, and introduces data enhancement methods to generate a high-dimensional geological information model that includes spatial location, physical parameters, and multi-source labels. This improves the comprehensiveness and spatial expression capabilities of geological analysis, and enhances the credibility and decision-making support value of the data.

[0063] (4) The present invention embeds the recognition results into a three-dimensional modeling and visualization interactive platform, and combines virtual reality and augmented reality technologies to achieve three-dimensional expression, real-time browsing and path simulation operations of abnormal areas such as goafs and waterlogged areas, significantly enhancing the intuitive understanding and decision-making efficiency of engineering personnel on underground structures, and is suitable for visualization-assisted evaluation and solution deployment at complex coal mine sites. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0065] Figure 1This is the overall structural framework diagram of a coal mine exploration result visualization and safety monitoring system proposed by the present invention;

[0066] Figure 2 This is a schematic diagram of the structure of the electromagnetic signal acquisition and processing module in the present invention. DETAILED DESCRIPTION

[0067] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0068] refer to Figure 1-2 , a coal mine exploration result visualization and safety monitoring system, comprising the following steps:

[0069] S1, electromagnetic signal acquisition module, constructs an electromagnetic pulse signal receiving array composed of multiple receiving units with different frequency response characteristics and sensitivities. The receiving units are arranged in a spatially optimized manner. The receiving parameters are adjusted according to the spectrum distribution and signal-to-noise ratio of the received signals of the receiving units to generate the original electromagnetic response data frame;

[0070] S2, signal processing module, pre-processes the original electromagnetic response data frame to form a standardized multi-channel time series signal, and extracts characteristic parameters based on NEMV to generate a structured feature vector set;

[0071] S3, geological identification module, builds a deep neural network model, inputs a structured feature vector set, identifies geological anomaly types in goaf, waterlogged areas, and fracture development areas, and outputs anomaly spatial data including anomaly type, boundary information, and spatial coordinates;

[0072] S4, data fusion module, obtains geological drilling data and historical monitoring data through the configured data access module, fuses them with abnormal spatial data, uses data enhancement methods to expand data dimensions, and generates a geological spatial information model;

[0073] S5, a 3D modeling and visualization module, generates a 3D geological model including geological anomaly areas and their physical parameters based on the geological spatial information model, and converts the 3D geological model into visual graphic data through a graphics rendering engine;

[0074] The construction of the three-dimensional geological model is based on the geological spatial information model. First, the entire exploration area is discretized into regular or irregular three-dimensional units through spatial grid division, and the identified geological anomaly boundary data is mapped to the corresponding spatial position. Combined with drilling and monitoring data, multi-source physical attribute parameters such as lithology type, resistivity, porosity, and water content are bound to each spatial unit to form a structured attribute field. The model also integrates the three-dimensional coordinate system, the topological relationship of the geological unit, the attribute field index and rendering tag information, supporting the subsequent graphics engine for visualization generation and spatial query. It is the core data carrier for realizing the digital expression and interactive display of underground geological bodies.

[0075] This implementation method is based on the fused geological spatial information model. The boundary coordinates and physical properties of the abnormal area are hierarchically encoded through a spatial grid division algorithm, and a geological modeling engine is used to construct a three-dimensional geometric structure. Geological anomalies such as goafs and waterlogged areas are expressed in the form of solid units. At the same time, parameters such as lithology, resistivity, and porosity are bound to corresponding spatial units to form a three-dimensional geological model that contains geometric form and physical properties. The model structure has spatial topological relationships and attribute field indexing capabilities, and can be directly read by the rendering engine and converted into visual graphic data, providing a complete data foundation for subsequent visual interaction, path simulation, and risk assessment. It realizes the intuitive expression and digital modeling of complex underground structures, and improves the spatial recognition accuracy and geological information organization efficiency.

[0076] S6, immersive interaction module, builds a three-dimensional visualization interaction platform, loads visualization graphic data into the three-dimensional visualization interaction platform, and combines virtual reality and augmented reality technologies to achieve multi-angle viewing and interactive operations of geological parameters.

[0077] The construction of the three-dimensional visualization interactive platform is based on the geological space information model. First, the data structures such as three-dimensional coordinates, geological anomaly boundaries and physical properties contained therein are converted into a standard graphics rendering format. A visualization model with spatial topological relationships is generated through a three-dimensional modeling engine, and the graphics rendering engine is called to perform scene construction and light and shadow processing. On this basis, virtual reality technology is integrated, and head-mounted display devices and spatial positioning systems are deployed to realize user perspective tracking and mobile interaction in the virtual environment. At the same time, augmented reality technology is used to map the model to the real space, and AR glasses or projection terminals are used to realize virtual and real fusion display, ultimately forming a three-dimensional visualization interactive platform with high-fidelity restoration, attribute interaction and immersive experience.

[0078] This implementation method constructs a three-dimensional visualization interactive platform, loads the three-dimensional geological model generated by the aforementioned geological spatial information model into the platform as graphic data, uses a graphics rendering engine to restore the spatial distribution scenes of coal seam structure, goaf and waterlogged areas, and integrates virtual reality technology to achieve an immersive browsing experience. Users can view geological structure details and zoom in on specific areas from a first-person perspective in a virtual mine environment through head-mounted display devices and spatial positioning controllers, and overlay virtual models onto real operating scenes through augmented reality technology. It supports calling geological attribute information, simulating mining paths and assessing risk areas through AR glasses or projection terminals. The platform significantly improves the intuitive understanding and safety decision-making efficiency of engineering personnel under complex geological conditions, and realizes closed-loop visualization support from data to interaction.

[0079] In this embodiment, S1 specifically includes:

[0080] S11. An electromagnetic receiving array construction unit is configured to construct an electromagnetic pulse signal receiving array composed of a plurality of electromagnetic signal receiving units, each electromagnetic signal receiving unit including a coil inductor and a gain adjustment unit, and each electromagnetic signal receiving unit is bound to corresponding three-dimensional spatial coordinate information, the coil inductor having different frequency response characteristics for receiving electromagnetic signals covering low-frequency, medium-frequency, and high-frequency bands, the gain adjustment unit having a parameter setting interface for configuring a gain coefficient and a signal receiving frequency passband parameter range, and the gain coefficient and frequency passband parameters can be updated during an acquisition cycle to adapt to environmental changes;

[0081] S12, a receiving unit arrangement unit, in an initial pre-arrangement state, wherein each electromagnetic signal receiving unit performs a short-cycle sampling to obtain an original electromagnetic waveform signal, analyzes the frequency band amplitude and background noise ratio of the signal at different receiving positions, sets an arrangement spacing of the electromagnetic signal receiving units based on the analysis results, and arranges the electromagnetic signal receiving units according to a predefined geometric structure, wherein the geometric structure arrangement of the receiving units includes a grid arrangement, a spiral arrangement, and a radial arrangement;

[0082] In this embodiment, during the initial deployment of the electromagnetic signal receiving array, the system first controls all receiving units to perform a short-cycle synchronous sampling to obtain the original electromagnetic waveform signal at each receiving position, and analyzes its frequency band amplitude and the corresponding background noise ratio. According to the differences in signal-to-noise ratios in different areas, the deployment spacing between receiving units is dynamically adjusted, with the deployment spacing in high signal-to-noise ratio areas increased and the deployment in low signal-to-noise ratio or high-interference areas more dense, in order to enhance the ability to capture local abnormal signals. The receiving units are arranged in a preset geometric structure. Grid, spiral or radial structures can be selected according to the detection task to achieve orderly coverage of the target area and enhanced directionality. This deployment method improves the spatial resolution and overall detection accuracy of signal reception, providing high-quality raw data support for subsequent extraction and identification of geological anomaly features.

[0083] S13, synchronous acquisition unit, each electromagnetic signal receiving unit synchronously acquires the electromagnetic pulse response signal generated by underground excitation according to the set unified time stamp, and the acquired signal forms the original time domain waveform data;

[0084] S14, spectrum analysis unit, performs frequency domain transformation processing on the collected original time domain waveform data, and extracts the amplitude response s of each frequency component f (t), and calculate the frequency band energy value E according to the amplitude response in the target frequency band f , and at the same time extract the background noise amplitude in the non-target frequency band area and form the noise distribution parameters:

[0085]

[0086] Among them, E f is the energy value of frequency band f, t0 and t1 are the start and end time of the analysis time window respectively;

[0087] This formula is used to calculate the energy of electromagnetic signals within a target frequency band. It effectively measures the energy intensity of a frequency component within a specific time window, useful for distinguishing valid signals from background noise. By integrating the square of the frequency domain signal's amplitude, it reflects the cumulative power of the electromagnetic response within that frequency band. In principle, energy calculation is based on integrating the square of the modulus of the frequency response function over a time interval. This effectively characterizes the energy distribution characteristics of electromagnetic pulse signals at different frequencies, providing a quantitative basis for subsequent signal-to-noise ratio calculations and receiver parameter adjustments.

[0088] S15, a parameter adjustment unit, which calculates the signal-to-noise ratio of each frequency band component according to the acquired frequency band energy value and the corresponding noise distribution parameter, and adjusts the gain coefficient and frequency passband parameter of each electromagnetic signal receiving unit in real time;

[0089] This implementation performs spectrum analysis on the collected electromagnetic signals, extracting the energy value of each frequency component and combining it with the background noise distribution of its corresponding frequency band. The signal-to-noise ratio is then calculated to determine the signal validity. The system dynamically adjusts the gain coefficient and frequency passband parameters of each electromagnetic signal receiving unit based on the signal-to-noise ratio results for each frequency band: when the signal-to-noise ratio in a certain frequency band is high, the system automatically increases the gain of that band to enhance the signal response; otherwise, it reduces the gain or narrows the passband to shield the noise, thereby achieving real-time optimal configuration of the gain and passband. This method improves the quality of effective signal acquisition, enhances the system's adaptability in multi-source interference environments, and enhances the accuracy of geological anomaly identification. It is particularly suitable for the stable extraction of weak abnormal signals in complex coal mine scenarios.

[0090] S16. The data encapsulation output unit uniformly encodes the original time domain waveform data, gain parameters and frequency passband parameters to form an original electromagnetic response data frame, wherein the original electromagnetic response data frame includes the acquisition time, receiving unit number, spatial coordinates and current gain configuration status.

[0091] This embodiment realizes dynamic adaptive adjustment of signal acquisition parameters by constructing a receiving array composed of multiple electromagnetic signal receiving units with different frequency response characteristics, and configuring a gain adjustment unit and a frequency passband parameter setting interface for each receiving unit. In the initial sampling stage, the spectrum distribution and background noise conditions of each receiving position are obtained through short-period electromagnetic acquisition, and the energy value and noise amplitude of the target frequency band are extracted using frequency domain analysis. The signal-to-noise ratio is calculated and the gain coefficient and frequency passband parameters are dynamically adjusted. Between acquisition cycles, the gain value and passband range are updated in real time through the system control module, so that each receiving unit can adaptively optimize the signal response capability according to the current geological electromagnetic environment, avoid strong interference frequency bands, and enhance the sensitivity to effective weak signals, thereby improving the overall signal quality and the recognition accuracy of anomalies. It is particularly suitable for stable acquisition in electromagnetically complex environments such as small kiln damage areas.

[0092] In this embodiment, the S11 specifically includes:

[0093] S111. A receiving unit frequency structure design unit is configured to construct multiple electromagnetic signal receiving units with differentiated frequency responses. Each electromagnetic signal receiving unit is equipped with a coil sensor for sensing underground electromagnetic pulse signals. The frequency response range of each receiving unit is defined as:

[0094] R i =[f i,min ,f i,max ];

[0095] Among them, R i represents the frequency range that the i-th electromagnetic signal receiving unit can effectively respond to, f i,minIndicates the minimum response frequency of the electromagnetic signal receiving unit, f i,max Indicates the maximum response frequency, multiple R i There is partial overlap between them, which is used to construct the overall continuous spectrum coverage interval;

[0096] S112, a gain parameter setting unit configures an independent gain adjustment path for each electromagnetic signal receiving unit. The gain coefficient of each electromagnetic signal receiving unit is expressed as G i , used to control the amplitude amplification of its output signal, the gain coefficient value meets the following restrictions:

[0097] G i ∈[G min ,G max ];

[0098] Among them, G min The minimum gain value, G max The maximum gain value that can be set is used to adjust the weak signal sensing strength.

[0099] S113, the frequency passband configuration unit, according to the frequency range R that the electromagnetic signal receiving unit can effectively respond to i and the gain coefficient G of the electromagnetic signal receiving unit i , set the signal receiving frequency passband parameters of the corresponding receiving unit. The frequency passband parameters are the frequency window that actually receives the effective signal during sampling, and must be its response interval R i The adjustment mode is dynamically adjusted according to the set gain sensitivity;

[0100] The dynamic sensitivity function adjusts the frequency passband in real time. During the initialization phase, the system sets a gain sensitivity mapping rule. When the receiving unit's gain coefficient increases to enhance weak signals, the system automatically narrows the frequency passband to suppress broadband noise interference. When the gain coefficient decreases to avoid saturation of strong signals, the frequency passband is correspondingly widened to improve coverage, thus achieving a dynamic balance between signal enhancement and interference rejection. This mechanism ensures that the frequency passband is always optimally configured under different gain conditions, effectively improving signal acquisition quality and the system's environmental adaptability in complex electromagnetic environments.

[0101] S114. After each acquisition cycle, the set gain coefficient G is updated according to the update strategy set by the system. i The parameter update operation is performed based on the configured frequency passband parameters. The update operation is performed during the acquisition interval.

[0102] The gain coefficient and frequency passband parameters of the electromagnetic signal receiving unit are adjusted in real time. According to the energy distribution and signal-to-noise ratio statistics of each channel signal in the current cycle, the system automatically evaluates the optimal receiving sensitivity and frequency window of each receiving unit in the next cycle, and completes parameter writing during the acquisition interval. This method can effectively enhance the response capability to weak abnormal signals in the target frequency band by periodically and adaptively optimizing the receiving parameters, while suppressing background noise interference, improving signal quality and detection accuracy, and enhancing the stability and adaptability of the system in complex electromagnetic environments.

[0103] This embodiment achieves high-sensitivity sensing and adaptive parameter control of electromagnetic pulse signals in different frequency bands by constructing an array of electromagnetic signal receiving units with differentiated frequency responses and configuring an independent gain adjustment path and frequency passband control mechanism for each unit. During the initial configuration phase, the system sets the frequency response range and gain adjustment parameters of each receiving unit, and dynamically adjusts the receiving parameters based on environmental changes after each acquisition cycle, ensuring that the system always maintains optimal signal reception under complex geological interference conditions. This approach effectively improves the system's ability to capture weak underground electromagnetic signals and the continuity of spectrum coverage, enhances the basic signal quality for subsequent geological anomaly identification, and significantly improves the detection accuracy and spatial positioning accuracy of abnormal areas such as goafs and waterlogged areas.

[0104] In this embodiment, S2 specifically includes:

[0105] S21, a time series signal construction unit, which obtains a multi-channel waveform signal in the original electromagnetic response data frame, extracts the electromagnetic signal time series corresponding to each electromagnetic signal receiving unit number, aligns them according to a set unified time reference point, and constructs an initial multi-channel time series matrix with a dimension of C×T, where C is the number of electromagnetic signal receiving units and T is the length of the electromagnetic signal time series;

[0106] S22, the pre-processing execution unit performs normalization, band-pass filtering and noise suppression on the initial multi-channel time series matrix in sequence. The normalization process uses the range normalization method to map the signal value of each channel to the [0, 1] interval to generate a standardized multi-channel time series signal:

[0107]

[0108] Among them, x i,t represents the original multi-channel time series signal of channel i at time t, min(x i )、max(x i ) represent the minimum and maximum values ​​of the original channel i time series signal, x' i,t To standardize multi-channel timing signals;

[0109] This formula uses range normalization to linearly scale the electromagnetic signal values ​​of each channel according to the difference between its minimum and maximum values, mapping them to a standardized interval between 0 and 1. This method maintains the relative amplitude relationships of the original signals, eliminating dimensional differences between different receiving channels while preserving the signal's changing trends and dynamic characteristics. This facilitates the unified processing and comparison of information from different channels during subsequent feature extraction, improving the stability and adaptability of the model's input data.

[0110] S23, NEMV feature extraction unit, performs nonlinear electromagnetic vector modeling on the standardized multi-channel time series signal, extracts the spatiotemporal coupling characteristic parameters at each moment, and constructs a geologically relevant high-dimensional vector set. The spatiotemporal coupling characteristic parameters include instantaneous energy density, vector direction change rate, and vector amplitude extreme point position;

[0111] The nonlinear electromagnetic vector modeling maps the time series data from multiple electromagnetic signal receiving channels into a vector set corresponding to each moment, constructing a coupled expression of the electromagnetic response in time and space. The system extracts the instantaneous energy density of the signal at each sampling moment to measure the local signal strength; calculates the vector direction change rate to reflect the relative phase and direction changes between different channels; identifies the local extreme points of the vector amplitude to characterize the response position and morphological characteristics of the abnormal area. Through the joint modeling of the above characteristic parameters, the system generates a high-dimensional geological feature vector set, which not only retains the key differences in the electromagnetic signal's time domain and spatial distribution, but also enhances the model's ability to identify geological anomaly structures, effectively improving the accuracy and robustness of subsequent deep neural networks in the classification and identification of goaf and waterlogged areas.

[0112] S24, a structured conversion unit, reconstructs the dimensions and standardizes the format of the geologically relevant high-dimensional vector set, and outputs a structured feature vector set.

[0113] This implementation method constructs a structured feature vector set for use by the geological identification module by performing multi-channel time series construction, normalization preprocessing, NEMV feature extraction, and structured conversion on the raw electromagnetic response data frames. The system first extracts the time series of each receiving unit and performs time alignment to construct a standardized multi-channel time series matrix. Signal quality is improved through range normalization, bandpass filtering, and noise suppression. Nonlinear electromagnetic vector modeling is applied to extract spatiotemporal coupling features such as instantaneous energy, directional change rate, and extreme point location, ultimately reconstructing them into a unified high-dimensional feature vector. This method significantly improves the information density and discriminative power of electromagnetic response data in geological anomaly identification, provides high-quality input for subsequent model identification, and enhances the system's recognition accuracy and stability under complex geological conditions.

[0114] In this embodiment, S3 specifically includes:

[0115] S31, a geological identification model construction unit, constructing a deep neural network model for geological anomaly classification, wherein the deep neural network model consists of an input layer, a plurality of hidden layers, and an output layer, wherein the input layer receives a structured feature vector set, the hidden layer adopts a fully connected network structure, and the number of nodes in the output layer is N, where N represents the number of geological anomaly types;

[0116] S32, a model training execution unit, constructs a training sample data set containing anomaly types labeled in goaf, waterlogged areas, and fracture development areas. The training sample data set includes structured feature vectors and corresponding geological anomaly labels. A cross entropy loss function is used to train a deep neural network model to generate a trained deep neural network model L:

[0117]

[0118] Among them, y j is the one-hot encoding value of the actual label, is the predicted probability of the deep neural network model outputting the jth type of anomaly, and N is the number of geological anomaly types;

[0119] The loss function used in this formula is the cross-entropy function, which measures the discrepancy between the model's predictions and the actual labels in geological anomaly classification tasks. This function performs a logarithmic comparison of the predicted probability of each geological anomaly class with the actual label. The closer the model's predictions are to the actual label, the smaller the loss value, and vice versa. This effectively guides the model to continuously optimize parameters during training to improve classification accuracy. This function, based on the principle of information entropy, quantifies the degree of inconsistency between the predicted and actual distributions and is a commonly used optimization objective in deep learning multi-classification tasks.

[0120] S33, an anomaly recognition unit, calling the trained deep neural network model, performs recognition processing on the structured feature vector set, and outputs the geological anomaly type classification label, recognition confidence value and index position corresponding to each input vector;

[0121] S34. The abnormal spatial data generating unit generates abnormal spatial data including abnormal type, boundary information and three-dimensional spatial position based on the geological abnormality classification label and index position and the corresponding three-dimensional spatial coordinate information bound to each electromagnetic signal receiving unit.

[0122] This implementation method constructs a geological anomaly recognition model based on a deep neural network. The structured feature vector set is input into a model composed of a multi-layer fully connected network for training. The training samples cover labeled geological types such as goaf, waterlogged areas, and fracture development areas. The model parameters are optimized through the cross-entropy loss function to generate a geological recognition model with classification capabilities. In actual applications, the system calls the trained model to identify the newly collected structured feature vectors, outputs the anomaly type label, recognition confidence, and its index position in the data matrix, and combines the spatial coordinates of the receiving unit to map and generate three-dimensional geological anomaly data containing boundary information. This implementation method realizes an intelligent closed-loop recognition process from electromagnetic data to the spatial positioning of geological anomalies, significantly improving the recognition accuracy and spatial positioning accuracy, and providing a high-quality data foundation for subsequent visualization and risk assessment.

[0123] In this embodiment, the S32 specifically includes:

[0124] S321, a training data preparation unit collects historical geological data of coal mines with known geological anomaly types, obtains structured feature vector samples corresponding to goaf areas, waterlogged areas, and fracture development areas, and forms an input sample set;

[0125] S322, a label encoding unit, adding a manually annotated geological anomaly type label to each input sample set, wherein the geological anomaly type label is represented in a one-hot encoding form. If the total number of geological anomaly types is N, then the label vector is a binary vector of dimension N, where only the jth component is 1 and the rest are 0, indicating the jth type of anomaly;

[0126] S323, a loss function definition unit, defining a cross entropy loss function used in the training process, which is used to measure the difference between the geological anomaly type label predicted by the deep neural network model and the actual geological anomaly type label;

[0127] S324, the model training control unit performs iterative training operations on the deep neural network model based on the input sample set and the geological anomaly type label, optimizes the model using the batch gradient descent method, and outputs the trained deep neural network model when the training loss drops to the maximum training round.

[0128] This implementation method collects historical coal mine data with known geological anomaly types, extracts structured feature vectors corresponding to goaf areas, waterlogged areas, and fissure development areas, and uses one-hot encoding to annotate geological anomaly labels to construct a training sample set. During the training process, a cross-entropy loss function is introduced to measure the error between the model output and the actual label. The training process is controlled by a batch gradient descent algorithm, and the parameters of the deep neural network model are optimized within a set number of rounds. This method realizes structured modeling and label supervised learning of coal mine geological anomaly data, effectively improving the model's ability to classify anomaly types and recognition accuracy, and providing a reliable data foundation and intelligent recognition capabilities for subsequent spatial positioning and risk assessment of geological anomalies.

[0129] In this embodiment, the S4 specifically includes:

[0130] S41. The external data access unit obtains geological drilling data and historical monitoring data through a configured data access module. The geological drilling data includes borehole number, drilling depth interval, lithologic layering information, and measurement parameters. The historical monitoring data includes the occurrence time, spatial coordinates, and corresponding monitoring index values ​​of underground abnormal events.

[0131] S42, an abnormal data matching unit, using the abnormal spatial data as a matching reference, performs registration and matching of the received geological drilling data and the historical monitoring data according to spatial coordinates, extracts drilling samples and monitoring records adjacent to the abnormal spatial position, and establishes a corresponding relationship between the drilling samples and monitoring records adjacent to the abnormal spatial position;

[0132] Using the aforementioned anomaly spatial data as a matching reference, the team extracts its three-dimensional spatial coordinate information. Within a unified spatial reference system, the borehole spatial locations in the accessed geological drilling data and the monitoring point coordinates in the historical monitoring data are aligned. A spatial proximity search algorithm is used to select the drilling samples and monitoring records closest to each anomaly spatial location within a given radius. Spatial index relationships are then established to achieve one-to-many or many-to-one data mapping. This method not only accurately connects geological anomaly identification results with actual drilling and monitoring data, but also establishes a spatial fusion channel between cross-source data, providing a highly consistent data foundation for subsequent feature enhancement and spatial modeling, improving the integrity of geological modeling and the accuracy of spatial analysis.

[0133] S43, a data enhancement processing unit, constructing a joint feature space based on the established correspondence between drilling samples and monitoring records adjacent to the abnormal spatial position, merging the abnormal spatial data with the matched geological drilling data and historical monitoring data, using a spatial interpolation method to fill in the sparsely sampled areas, and generating a fused multidimensional feature matrix;

[0134] Based on the established correspondence between abnormal spatial positions and drilling samples and monitoring records, the abnormal spatial data, drilling attribute parameters and historical monitoring indicators are first normalized according to the spatial position, and the corresponding data dimensions are aligned and merged in the same spatial grid to construct a joint feature space.

[0135] To address the sparse distribution of drilling samples, the inverse distance weighted method is used for spatial interpolation. That is, the data values ​​of the missing points are estimated by weighted average based on their distances from the surrounding known data points. The interpolation weight is inversely proportional to the distance, so as to fill the missing areas of geological attributes in three-dimensional space. The final output fused multidimensional feature matrix can fully express the anomaly type, physical parameters and monitoring characteristics within each spatial unit, providing a complete geological data foundation for subsequent three-dimensional modeling and risk analysis, and improving the spatial continuity and parameter integrity of the model.

[0136] S44, the spatial information modeling unit, takes the fused multidimensional feature matrix as input, organizes it into structures according to the spatial coordinates of the abnormal spatial data, uses data enhancement methods to expand the data dimension, and generates a geological spatial information model.

[0137] The data enhancement method expands upon the original features by introducing redundant attribute dimensions and simulating positional perturbations. For attribute enhancement, redundant feature dimensions reflecting regional consistency are constructed by introducing the statistical attribute means and variances of adjacent geological units. For spatial enhancement, simulated samples are generated by perturbing a known spatial point cloud at small scales, simulating various spatial distribution variations and improving the robustness of the model. The resulting geological spatial information model not only includes the anomaly type and attribute parameter set for each spatial location but also exhibits spatial continuity and multi-scale representation capabilities. This provides comprehensive, high-precision data support for 3D modeling and geological analysis, significantly improving the accuracy of anomaly boundary fitting and the integrity of the geological model.

[0138] This implementation introduces geological drilling data and historical monitoring data through a configured data access module. Using the anomalous spatial data as a reference, the multi-source data is aligned and matched according to spatial coordinates, establishing a correlation between the anomalous location and drilling samples and monitoring records. On this basis, spatial location, lithologic attributes, and monitoring indicators are integrated, and spatial interpolation methods are used to complete sparse data areas, generating a fused multidimensional feature matrix. A structured geological spatial information model is further constructed, achieving the binding of spatial location with multi-source geological information. This method significantly improves the data integrity and expression dimensionality of geological anomaly areas, providing more comprehensive and accurate basic data support for subsequent three-dimensional modeling and safety decision-making.

[0139] Example 1:

[0140] To verify the feasibility of this invention, we applied it to a coal mine in City A, a typical destructive coal seam affected by historical mining operations at small kilns. The mine has complex geological conditions, with numerous goafs, waterlogged areas, and fault fissures. Because the remaining areas of the small kilns have not been systematically explored, numerous "geological blind spots" remain. Traditional geophysical exploration and drilling methods suffer from low resolution, weak response, and inaccurate coordinates, making them unable to provide accurate geological support for safe underground construction.

[0141] During the implementation of the present invention, 64 electromagnetic signal receiving units were first deployed in the 17# working face area of ​​the coal mine. These units were grouped according to their multi-frequency response characteristics to form a spatially optimized receiving array structure. The system automatically adjusted the gain of the receiving units based on the initial electromagnetic environment and noise response, and collected raw electromagnetic response data. After acquisition, the collected multi-channel data was subjected to feature extraction using a nonlinear electromagnetic vector method to generate a standardized feature vector set. The deep neural network recognition model constructed by the present invention was used to identify and classify goaf and waterlogged areas. The recognition results were spatially encoded and output as an abnormal spatial dataset.

[0142] In the data fusion stage, the system imports the drilling data and regional monitoring records of the coal mine since 2007 through the data access module, covering 88 boreholes and 116 abnormal monitoring points. In the S43 data enhancement processing unit of the present invention, the system matches the abnormal spatial data with the drilling parameters (such as lithology, resistivity, porosity), and uses the inverse distance weighted interpolation method to fill the data in the sparse spatial distribution area. For example, the original sampling points of the goaf area numbered between K17-18 and K17-20 are insufficient. After interpolation processing, a complete abnormal boundary surface is generated. The processing unit outputs a fusion feature matrix containing abnormal type, physical parameters and spatial coordinates to provide basic data for three-dimensional modeling.

[0143] The system's resulting 3D geological model successfully captured the spatial configuration of the main goaf beneath the 17# working face, clearly identifying an irregular waterlogged area approximately 57 meters long, 24 meters wide, and 4.2 meters high. The model also visualized its boundary variations and surrounding rock properties. Engineers performed path simulations on the VR interactive platform and, based on the visualization results, adjusted the planned tunnel layout to avoid high-risk areas, reducing the risk of water inrush during excavation.

[0144] To further verify the effectiveness of the system in practical applications, the project team selected two adjacent working face areas for comparative testing. One area used the system for exploration and design, while the other used traditional geophysical methods. The comparison period was three months, covering three drilling verifications and one local water inrush emergency treatment. The main technical indicators are shown in the following table:

[0145] Table 1 Comparison of application effects of different exploration systems in coal mines

[0146]

[0147]

[0148] The above data demonstrates that this method significantly outperforms traditional exploration methods in terms of recognition accuracy, modeling capabilities, and risk control. Exploration time was shortened by approximately 40.9%, recognition accuracy increased by over 17%, and boundary error was controlled within ±5 meters. This effectively supported dynamic design adjustments, successfully avoided crossing high-risk areas, and reduced construction safety hazards.

[0149] In a specific construction application, for example, a borehole numbered D17-03 was not identified as an anomaly using traditional methods, resulting in a water inrush during tunneling. However, the system, after identifying it as an abnormally high water pressure area, was preemptively sealed and prevented from causing any water inrush during the entire construction process. Engineering team members reported that the system's 3D visualization and real-time interactive display significantly improved their understanding and assessment of geological structures, significantly enhancing its practicality in safety risk assessment and plan development.

[0150] In summary, the present invention achieves high-precision exploration and expression of goaf and waterlogged areas under complex geological conditions by constructing a complete electromagnetic data acquisition, feature extraction, depth recognition, multi-source fusion and three-dimensional visualization process. It has higher recognition accuracy, stronger visualization capabilities and better engineering adaptability, and can significantly improve coal mine safety monitoring and risk management capabilities, providing a replicable solution for coal mines under similar geological conditions.

[0151] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A coal mine exploration result visualization and safety monitoring system, characterized in that: The steps include: S1, electromagnetic signal acquisition module, constructs an electromagnetic pulse signal receiving array composed of multiple receiving units with different frequency response characteristics and sensitivities. The receiving units are arranged in a spatially optimized manner. The receiving parameters are adjusted according to the spectrum distribution and signal-to-noise ratio of the received signals of the receiving units to generate the original electromagnetic response data frame; S2, signal processing module, pre-processes the original electromagnetic response data frame to form a standardized multi-channel time series signal, and extracts characteristic parameters based on NEMV to generate a structured feature vector set; S3, geological identification module, builds a deep neural network model, inputs a structured feature vector set, identifies geological anomaly types in goaf, waterlogged areas, and fracture development areas, and outputs abnormal spatial data; S4, data fusion module, obtains geological drilling data and historical monitoring data through the configured data access module, fuses them with abnormal spatial data, uses data enhancement methods to expand data dimensions, and generates a geological spatial information model; S5, a 3D modeling and visualization module, generates a 3D geological model including geological anomaly areas and their physical parameters based on the geological spatial information model, and converts the 3D geological model into visual graphic data through a graphics rendering engine; S6, immersive interaction module, builds a three-dimensional visualization interaction platform, loads visualization graphic data into the three-dimensional visualization interaction platform, and combines virtual reality and augmented reality technologies to achieve multi-angle viewing and interactive operations of geological parameters.

2. A coal mine exploration result visualization and safety monitoring system according to claim 1, characterized in that: Said S1 specifically includes: S11. An electromagnetic receiving array construction unit is configured to construct an electromagnetic pulse signal receiving array composed of a plurality of electromagnetic signal receiving units, each electromagnetic signal receiving unit including a coil inductor and a gain adjustment unit, and each electromagnetic signal receiving unit is bound to corresponding three-dimensional spatial coordinate information, the coil inductor having different frequency response characteristics for receiving electromagnetic signals covering low-frequency, medium-frequency, and high-frequency bands, the gain adjustment unit having a parameter setting interface for configuring a gain coefficient and a signal receiving frequency passband parameter range, and the gain coefficient and frequency passband parameters can be updated during an acquisition cycle to adapt to environmental changes; S12, a receiving unit arrangement unit, in an initial pre-arrangement state, wherein each electromagnetic signal receiving unit performs a short-cycle sampling to obtain an original electromagnetic waveform signal, analyzes the frequency band amplitude and background noise ratio of the signal at different receiving positions, sets an arrangement spacing of the electromagnetic signal receiving units based on the analysis results, and arranges the electromagnetic signal receiving units according to a predefined geometric structure, wherein the geometric structure arrangement of the receiving units includes a grid arrangement, a spiral arrangement, and a radial arrangement; S13, synchronous acquisition unit, each electromagnetic signal receiving unit synchronously acquires the electromagnetic pulse response signal generated by underground excitation according to the set unified time stamp, and the acquired signal forms the original time domain waveform data; S14, spectrum analysis unit, performs frequency domain transformation processing on the collected original time domain waveform data, and extracts the amplitude response s of each frequency component f (t), and calculate the frequency band energy value E according to the amplitude response in the target frequency band f , and at the same time extract the background noise amplitude in the non-target frequency band area and form the noise distribution parameters: Among them, E f is the energy value of frequency band f, t0 and t1 are the start and end time of the analysis time window respectively; S15, a parameter adjustment unit, which calculates the signal-to-noise ratio of each frequency band component according to the acquired frequency band energy value and the corresponding noise distribution parameter, and adjusts the gain coefficient and frequency passband parameter of each electromagnetic signal receiving unit in real time; S16. The data encapsulation output unit uniformly encodes the original time domain waveform data, gain parameters and frequency passband parameters to form an original electromagnetic response data frame, wherein the original electromagnetic response data frame includes the acquisition time, receiving unit number, spatial coordinates and current gain configuration status.

3. A coal mine exploration result visualization and safety monitoring system according to claim 2, characterized in that: The S11 specifically includes: S111. A receiving unit frequency structure design unit is configured to construct multiple electromagnetic signal receiving units with differentiated frequency responses. Each electromagnetic signal receiving unit is equipped with a coil sensor for sensing underground electromagnetic pulse signals. The frequency response range of each receiving unit is defined as: R i =[f i,min ,f i,max ]; Among them, R i represents the frequency range that the i-th electromagnetic signal receiving unit can effectively respond to, f i,min Indicates the minimum response frequency of the electromagnetic signal receiving unit, f i,max Indicates the maximum response frequency, multiple R i There is partial overlap between them, which is used to construct the overall continuous spectrum coverage interval; S112, a gain parameter setting unit configures an independent gain adjustment path for each electromagnetic signal receiving unit. The gain coefficient of each electromagnetic signal receiving unit is expressed as G i , used to control the amplitude amplification of its output signal, the gain coefficient value meets the following restrictions: G i ∈[G min ,G max ]; Among them, G min The minimum gain value, G max The maximum gain value that can be set is used to adjust the weak signal sensing strength. S113, the frequency passband configuration unit, according to the frequency range R that the electromagnetic signal receiving unit can effectively respond to i and the gain coefficient G of the electromagnetic signal receiving unit i , set the signal receiving frequency passband parameters of the corresponding receiving unit. The frequency passband parameters are the frequency window that actually receives the effective signal during sampling, and must be its response interval R i The adjustment mode is dynamically adjusted according to the set gain sensitivity; S114. After each acquisition cycle, the set gain coefficient G is updated according to the update strategy set by the system. i The parameter update operation is performed based on the configured frequency passband parameters. The update operation is performed during the acquisition interval.

4. A coal mine exploration result visualization and safety monitoring system according to claim 2, characterized in that: The S2 specifically includes: S21, a time series signal construction unit, which obtains a multi-channel waveform signal in the original electromagnetic response data frame, extracts the electromagnetic signal time series corresponding to each electromagnetic signal receiving unit number, aligns them according to a set unified time reference point, and constructs an initial multi-channel time series matrix with a dimension of C×T, where C is the number of electromagnetic signal receiving units and T is the length of the electromagnetic signal time series; S22, the pre-processing execution unit performs normalization, band-pass filtering and noise suppression on the initial multi-channel time series matrix in sequence. The normalization process uses the range normalization method to map the signal value of each channel to the [0, 1] interval to generate a standardized multi-channel time series signal: Among them, x i,t represents the original multi-channel time series signal of channel i at time t, min(x i )、max(x i ) represent the minimum and maximum values ​​of the original channel i time series signal, x ' i,t To standardize multi-channel timing signals; S23, NEMV feature extraction unit, performs nonlinear electromagnetic vector modeling on the standardized multi-channel time series signal, extracts the spatiotemporal coupling characteristic parameters at each moment, and constructs a geologically relevant high-dimensional vector set. The spatiotemporal coupling characteristic parameters include instantaneous energy density, vector direction change rate, and vector amplitude extreme point position; S24, a structured conversion unit, reconstructs the dimensions and standardizes the format of the geologically relevant high-dimensional vector set, and outputs a structured feature vector set.

5. The coal mine exploration result visualization and safety monitoring system according to claim 1, characterized in that: The S3 specifically includes: S31, a geological identification model construction unit, constructing a deep neural network model for geological anomaly classification, wherein the deep neural network model consists of an input layer, a plurality of hidden layers, and an output layer, wherein the input layer receives a structured feature vector set, the hidden layer adopts a fully connected network structure, and the number of nodes in the output layer is N, where N represents the number of geological anomaly types; S32, a model training execution unit, constructs a training sample data set containing anomaly types labeled in goaf, waterlogged areas, and fracture development areas. The training sample data set includes structured feature vectors and corresponding geological anomaly labels. A cross entropy loss function is used to train a deep neural network model to generate a trained deep neural network model L: Among them, y j is the one-hot encoding value of the actual label, is the predicted probability of the deep neural network model outputting the jth type of anomaly, and N is the number of geological anomaly types; S33, an anomaly recognition unit, calling the trained deep neural network model, performs recognition processing on the structured feature vector set, and outputs the geological anomaly type classification label, recognition confidence value and index position corresponding to each input vector; S34. The abnormal spatial data generating unit generates abnormal spatial data including abnormal type, boundary information and three-dimensional spatial position based on the geological abnormality classification label and index position and the corresponding three-dimensional spatial coordinate information bound to each electromagnetic signal receiving unit.

6. A coal mine exploration result visualization and safety monitoring system according to claim 5, characterized in that: The S32 specifically includes: S321, a training data preparation unit collects historical geological data of coal mines with known geological anomaly types, obtains structured feature vector samples corresponding to goaf areas, waterlogged areas, and fracture development areas, and forms an input sample set; S322, a label encoding unit, adding a manually annotated geological anomaly type label to each input sample set, wherein the geological anomaly type label is represented in a one-hot encoding form. If the total number of geological anomaly types is N, then the label vector is a binary vector of dimension N, where only the jth component is 1 and the rest are 0, indicating the jth type of anomaly; S323, a loss function definition unit, defining a cross entropy loss function used in the training process, which is used to measure the difference between the geological anomaly type label predicted by the deep neural network model and the actual geological anomaly type label; S324, the model training control unit performs iterative training operations on the deep neural network model based on the input sample set and the geological anomaly type label, optimizes the model using the batch gradient descent method, and outputs the trained deep neural network model when the training loss drops to the maximum training round.

7. A coal mine exploration result visualization and safety monitoring system according to claim 1, characterized in that: The S4 specifically includes: S41. The external data access unit obtains geological drilling data and historical monitoring data through a configured data access module. The geological drilling data includes borehole number, drilling depth interval, lithologic layering information, and measurement parameters. The historical monitoring data includes the occurrence time, spatial coordinates, and corresponding monitoring index values ​​of underground abnormal events. S42, an abnormal data matching unit, using the abnormal spatial data as a matching reference, performs registration and matching of the received geological drilling data and the historical monitoring data according to spatial coordinates, extracts drilling samples and monitoring records adjacent to the abnormal spatial position, and establishes a corresponding relationship between the drilling samples and monitoring records adjacent to the abnormal spatial position; S43, a data enhancement processing unit, constructing a joint feature space based on the established correspondence between drilling samples and monitoring records adjacent to the abnormal spatial position, merging the abnormal spatial data with the matched geological drilling data and historical monitoring data, using a spatial interpolation method to fill in the sparsely sampled areas, and generating a fused multidimensional feature matrix; S44, the spatial information modeling unit, takes the fused multidimensional feature matrix as input, organizes it into structures according to the spatial coordinates of the abnormal spatial data, uses data enhancement methods to expand the data dimension, and generates a geological spatial information model.

Images