Fully mechanized coal mining face mine pressure real-time monitoring system and method based on multi-sensor fusion

Abstract

The invention belongs to the technical field of safety monitoring of a fully mechanized coal mining face of a coal mine, and discloses a fully mechanized coal mining face mine pressure real-time monitoring system and method based on multi-sensor fusion. According to the system, multi-dimensional physical quantity data are collected in a target stress area through a distributed sensor array, after the data are preprocessed through a data collection module, space-time registration and feature layer fusion are conducted through a multi-sensor fusion processing module by means of a self-adaptive weight dynamic distribution model, and mine pressure comprehensive feature values are output. The real-time monitoring terminal receives and displays the characteristic value, the mine pressure change trend is obtained through intelligent analysis, and when the characteristic value exceeds a preset threshold value or the trend enters a dangerous interval, early warning is triggered. The accuracy and timeliness of mine pressure monitoring are improved, and a reliable guarantee is provided for safe mining of a fully mechanized coal mining face.

Description

Technical Field

[0001] The present invention relates to the technical field of coal mine safety monitoring, and in particular to a real-time monitoring system and method for coal mine pressure in a fully mechanized mining face based on multi-sensor fusion. Background Art

[0002] In coal mining, the safe and stable operation of fully mechanized coal mining faces is directly related to production efficiency and operational safety. Mine pressure monitoring is a key component in preventing roof collapse accidents and ensuring mining continuity. As mine depths continue to increase and geological conditions become more complex, the suddenness and intensity of mine pressure increases significantly, placing higher demands on the reliability and accuracy of real-time monitoring technology.

[0003] Currently, pressure monitoring in fully mechanized mining faces is primarily accomplished through the deployment of pressure sensors, displacement sensors, and other equipment. Monitoring data is transmitted via wired or wireless means to a ground-based monitoring center, where personnel assess the pressure status based on the monitored values. Some systems incorporate a simple data aggregation function, displaying monitoring results from different locations side by side on the terminal, helping managers understand the overall situation at the working face.

[0004] However, existing technologies have many limitations: first, the type of data collected by sensors is single, making it difficult to fully reflect the multi-dimensional characteristics of the mine pressure field; second, data processing is mostly simple forwarding, lacking in-depth integration of information from different sources, resulting in insufficient correlation of monitoring results; third, the early warning mechanism mostly relies on fixed thresholds, which cannot capture the dynamic trends of mine pressure changes in a timely manner, and is prone to early warning lags or false alarms, making it difficult to meet the safety monitoring needs under complex geological conditions. Summary of the Invention

[0005] In view of this, an embodiment of the present invention provides a real-time monitoring system and method for mine pressure in a fully mechanized mining face based on multi-sensor fusion to solve the above technical problems.

[0006] To achieve the above objectives, in a first aspect, a real-time monitoring system for mine pressure in a fully mechanized mining face based on multi-sensor fusion is provided, which comprises: A distributed sensor array is deployed in the target stress area of ​​the fully mechanized mining face, including the structural components of the hydraulic support, the interior and surface of the coal wall, the surrounding rock of the roadway, the boundary of the goaf, and the roof support interface, to collect multi-dimensional physical quantity data related to mine pressure; a data acquisition module, communicatively connected to the distributed sensor array, and configured to preprocess the multi-dimensional physical quantity data to obtain preprocessed multi-dimensional physical quantity data; A multi-sensor fusion processing module is connected to the data acquisition module and is configured with an adaptive weight dynamic allocation model for performing spatiotemporal registration and feature layer fusion on the pre-processed multi-dimensional physical quantity data to output a comprehensive characteristic value of mine pressure; A real-time monitoring terminal is communicatively connected to the multi-sensor fusion processing module, and is used to receive the comprehensive characteristic value of the mine pressure and display it in real time, and to perform intelligent analysis based on the comprehensive characteristic value of the mine pressure to obtain the trend of mine pressure change, and to trigger an early warning when the comprehensive characteristic value of the mine pressure exceeds a preset threshold and / or the trend of mine pressure change enters a dangerous range.

[0007] In a second aspect, a method for real-time monitoring of mine pressure in a fully-mechanized mining face based on multi-sensor fusion is provided. The method is based on the real-time monitoring system for mine pressure in a fully-mechanized mining face based on multi-sensor fusion described in the first aspect, and the method includes: Multi-dimensional physical quantity data related to mine pressure is collected by a distributed sensor array deployed in the target stress area of ​​the fully mechanized mining face. The target stress area includes the structural components of the hydraulic support, the interior and surface of the coal wall, the surrounding rock of the roadway, the boundary of the goaf and the roof support interface. Preprocessing the multi-dimensional physical quantity data by a data acquisition module to obtain preprocessed multi-dimensional physical quantity data; Through a multi-sensor fusion processing module equipped with an adaptive weight dynamic allocation model, the pre-processed multi-dimensional physical quantity data is subjected to spatiotemporal registration and feature layer fusion to output a comprehensive characteristic value of the mine pressure; The comprehensive characteristic value of the mine pressure is received by a real-time monitoring terminal and displayed in real time. The mine pressure change trend is obtained by intelligent analysis based on the comprehensive characteristic value of the mine pressure. When the comprehensive characteristic value of the mine pressure exceeds a preset threshold and / or the mine pressure change trend enters a dangerous range, an early warning is triggered.

[0008] The above technical solution has the following beneficial technical effects: This technical solution realizes the comprehensive collection of multi-dimensional physical quantity data related to mine pressure by deploying distributed sensor arrays in the target stress area of ​​the fully-mechanized mining working face, overcoming the defects of limited monitoring range and insufficient data dimension of a single sensor; the preprocessing process of the data acquisition module effectively improves the reliability of the original data; the multi-sensor fusion processing module uses the adaptive weight dynamic allocation model to perform spatiotemporal alignment and feature layer fusion, and the output comprehensive mine pressure characteristic value can accurately reflect the overall situation of the mine pressure field, solving the problem of poor correlation and difficulty in collaborative analysis of data from different sources; the real-time monitoring terminal can not only display the comprehensive mine pressure characteristic value in real time, but also perform intelligent analysis based on it to obtain the trend of mine pressure changes, and trigger warnings through the dual warning mechanism of threshold exceeding and trend danger, which improves the timeliness and accuracy of the warning, effectively avoids the lag and false alarm risk of single threshold warning, and finally realizes all-round, high-precision and intelligent monitoring of the mine pressure status of the fully-mechanized mining working face, providing strong guarantee for safe and efficient mining of coal mines. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] The accompanying drawings are provided for a better understanding of the present invention and are not intended to limit the present invention. Figure 1 This is a functional block diagram of a real-time monitoring system for mine pressure in a fully mechanized mining face based on multi-sensor fusion according to an embodiment of the present invention; Figure 2 is a schematic structural diagram of a distributed sensor array and a data acquisition module according to an embodiment of the present invention; Figure 3 is a schematic structural diagram of an adaptive weight dynamic allocation model according to an embodiment of the present invention; Figure 4 is a schematic diagram of the deployment location of edge computing nodes in an embodiment of the present invention; Figure 5 2 is a schematic structural diagram of a mine pressure trend prediction model according to an embodiment of the present invention; Figure 6 Schematic diagram of the working principle of the adaptive control interface according to an embodiment of the present invention; Figure 7 This is a flow chart of a method for real-time monitoring of mine pressure in a fully mechanized mining face based on multi-sensor fusion according to an embodiment of the present invention; Figure 8 Schematic diagram of the structure of a computer system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0010] The following description of exemplary embodiments of the present invention is made in conjunction with the accompanying drawings, in which various details of the embodiments of the present invention are included to facilitate understanding. These details should be considered as merely exemplary. Therefore, it should be appreciated by those skilled in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0011] Example 1 like Figure 1 As shown, this embodiment provides a real-time monitoring system for mine pressure in a fully mechanized mining face based on multi-sensor fusion, which includes: A distributed sensor array is deployed in the target stress area of ​​the fully mechanized mining face, including the structural components of the hydraulic support, the interior and surface of the coal wall, the surrounding rock of the roadway, the boundary of the goaf, and the roof support interface, to collect multi-dimensional physical quantity data related to mine pressure; a data acquisition module, communicatively connected to the distributed sensor array, and configured to preprocess the multi-dimensional physical quantity data to obtain preprocessed multi-dimensional physical quantity data; A multi-sensor fusion processing module is connected to the data acquisition module and is configured with an adaptive weight dynamic allocation model for performing spatiotemporal registration and feature layer fusion on the pre-processed multi-dimensional physical quantity data to output a comprehensive characteristic value of mine pressure; A real-time monitoring terminal is communicatively connected to the multi-sensor fusion processing module, and is used to receive the comprehensive characteristic value of the mine pressure and display it in real time, and to perform intelligent analysis based on the comprehensive characteristic value of the mine pressure to obtain the trend of mine pressure change, and to trigger an early warning when the comprehensive characteristic value of the mine pressure exceeds a preset threshold and / or the trend of mine pressure change enters a dangerous range.

[0012] The following is a detailed description. This embodiment discloses a real-time monitoring system for mine pressure in a fully-mechanized mining face based on multi-sensor fusion. The system is applicable to fully-mechanized mining faces with gently inclined coal seams of 2.5-5m thickness. The system specifically includes the following components: The deployment of the distributed sensor array must be tailored to the actual working face parameters. On the hydraulic support structures, a fiber Bragg grating pressure sensor is installed at the bottom of each support column and the cylinder of the push jack, forming a dual-point monitoring group. Piezoelectric stress sensors are embedded in 3-meter-deep holes drilled into the coal wall. A monitoring point is placed every five supports (approximately 10 meters) along the strike, with three sensors connected in series at each point to form a stress monitoring chain. Laser displacement sensors are installed on the coal wall surface, corresponding to each stress monitoring chain position. Their rotating pan-tilt heads are fixed to the front of the top beam, providing a ±30° monitoring range. Microseismic sensors are arranged in a 5-meter-by-5-meter triangular grid along the sides of the roadway and along the working face cuts, with two additional rows of sensors added at the goaf boundary. An electromagnetic radiation sensor is fixed to the center of the inner side of the hydraulic support shield beam, maintaining a distance of 1.2-1.5 meters from the coal wall surface.

[0013] The data acquisition module, housed in an intrinsically safe, explosion-proof enclosure, is installed next to the chute transfer machine at the working face and connected to the sensors via flame-retardant mining cables. The module features eight analog inputs (supporting 4-20mA signals) and four network ports (supporting Ethernet / 4G transmission). Preprocessing includes temperature compensation of pressure sensor data (using T-type thermocouples to simultaneously collect ambient temperature), filtering of displacement sensor data (using a fifth-order Butterworth low-pass filter), and noise reduction of microseismic signals (using a wavelet threshold denoising algorithm). All preprocessed data is packaged and stored at 100ms intervals in JSON format, including the sensor ID, timestamp, physical value, and a checksum.

[0014] The multi-sensor fusion processing module uses an industrial-grade server and is deployed in the underground central substation. The adaptive weight dynamic allocation model is built on the TensorFlow framework. The feature extraction layer uses a three-layer convolutional neural network (with kernel sizes of 3×3, 5×5, and 3×3, respectively) to extract features from preprocessed multi-dimensional physical quantity data. The weight allocation layer dynamically generates weight coefficients ranging from 0.1 to 0.9 by calculating the deviation (threshold set at 8%) and fluctuation variance (threshold set at 5%) between the data within the past 10 minutes and the baseline data for the previous 24 hours. Spatiotemporal registration uses linear interpolation to unify data of different sampling frequencies to 50Hz. The feature layer fusion uses an attention mechanism to concatenate weighted feature vectors, ultimately outputting a comprehensive mine pressure characteristic value ranging from 0 to 100 (higher values ​​indicate greater mine pressure risk).

[0015] The real-time monitoring terminal consists of an underground explosion-proof touch screen and a ground-based monitoring center server, both of which synchronize data via the mine's ring network. After receiving comprehensive mine pressure characteristic values, the terminal displays the corresponding values ​​for each support in real-time as a bar graph. It then performs intelligent analysis using an improved Long Short-Term Memory Network (LSTM). The network's input layer includes the characteristic value sequence for the past hour and the current working face advance speed (2-4 m / h), outputting a 30-minute trend curve for mine pressure changes. If the comprehensive mine pressure characteristic value exceeds 80 (a preset threshold), or the slope of the trend curve exceeds 0.5 for five consecutive minutes (entering the danger zone), the terminal immediately triggers a three-level warning: an audible and visual alarm (red warning light and 85dB buzzer) is emitted from the underground terminal, an alert window pops up on the ground terminal, and a text message is sent to the management personnel's mobile phone. The warning area is also highlighted in a 3D simulation map.

[0016] The system of this embodiment forms a distributed array by deploying multiple sensors in the target stress area, which can comprehensively collect multi-dimensional mine pressure data, and improve the data quality through targeted pre-processing by the data acquisition module; the multi-sensor fusion processing module uses the adaptive weight dynamic allocation model to achieve efficient fusion of data spatiotemporal alignment and feature layers, and the output comprehensive mine pressure characteristic value accurately reflects the mine pressure risk; the real-time monitoring terminal generates a trend curve through intelligent analysis, and triggers a graded warning in combination with dual warning conditions, which not only ensures the comprehensiveness and accuracy of monitoring, but also improves the timeliness and pertinence of the warning, effectively reduces the risk of mine pressure accidents in the fully mechanized mining face, and provides strong support for safe and efficient mining.

[0017] Example 2 like Figure 2 As shown, the distributed sensor array includes: Fiber Bragg grating pressure sensors are installed at the bottom of the hydraulic support column cylinder and the push jack cylinder to form a dual-point monitoring group for collecting the working resistance and pushing force of the hydraulic support. The temperature compensation algorithm is used to eliminate the influence of ambient temperature on the measured values. Piezoelectric stress sensors are buried at a preset depth inside the coal wall. The preset depth is determined based on the coal seam thickness and mining process parameters of the fully mechanized mining face. Distributed optical fibers are connected in series to form a stress monitoring chain to collect the three-dimensional stress field distribution inside the coal body. The laser displacement sensor is installed on the rotating platform at the front end of the hydraulic support top beam to dynamically track and collect the amount of roof subsidence and coal wall deformation; The microseismic sensor array is arranged in a triangular grid along the sides of the roadway and the working face cuts, covering the support area of ​​the hydraulic support and the boundary of the goaf. It is used to monitor the energy release rate and source location of microseismic events in the coal rock in real time. The electromagnetic radiation sensor is arranged on the inner side of the hydraulic support shield beam and is used to obtain the electromagnetic radiation signal of the stress concentration area of ​​the coal rock mass. The collection of the electromagnetic radiation signal forms a linkage trigger mechanism with the action of the hydraulic support. The linkage trigger mechanism includes: when the hydraulic support performs a pushing or lifting action, the electromagnetic radiation sensor automatically enters the high-frequency collection mode; when the hydraulic support is in a static state, the electromagnetic radiation sensor automatically switches to the low-frequency duty mode. The sampling frequency of the high-frequency collection mode is higher than the sampling frequency of the low-frequency duty mode.

[0018] In this embodiment, the specific implementation of the distributed sensor array is as follows: Fiber Bragg grating (FBG) pressure sensors are installed at the bottom of the hydraulic support's column cylinder and the push jack cylinder. One sensor is installed at each hydraulic support's column cylinder and push jack cylinder, forming a dual-point monitoring system. The sensors are housed in a fully stainless steel package, with a measurement range of 0-60 MPa and an accuracy of ±0.2% FS. The temperature compensation algorithm, based on dual-wavelength demodulation, monitors the ambient temperature in real time using a reference grating and corrects the measured values ​​using a temperature-strain cross-sensitivity model. The compensation range is -20°C to +80°C. The sensor signals are transmitted to the data acquisition module via a Φ3.0 mm tight-buffered optical cable, with a transmission distance of no less than 2 km.

[0019] The piezoelectric stress sensor used is a YZ-10 piezoelectric triaxial stress sensor, buried at a preset depth within the coal wall. This depth is determined by the coal seam thickness H, calculated as L = 0.3H + 0.5m (H is the coal seam thickness in meters). Three sensors are arranged vertically at each monitoring point, spaced 0.5m apart, to form a stress monitoring chain. The sensors are connected in series using distributed optical fiber, with each group consisting of five monitoring points. These sensors are connected to the data acquisition module via a mining-grade flame-retardant optical cable. With a measurement range of 0-50 MPa and a frequency response of 0.1-10 kHz, they can collect stress components in the X, Y, and Z directions, enabling three-dimensional stress field distribution monitoring within the coal mass.

[0020] The laser displacement sensor is mounted on a motorized rotating pan / tilt platform at the front end of the hydraulic support's top beam. The platform features an explosion-proof design, a rotation range of ±45°, a pitch range of ±15°, and a positioning accuracy of ±0.1°. The sensor has a measurement range of 50-500mm, an accuracy of ±0.01%FS, and a sampling frequency of 10kHz. Through a PLC control system, the platform automatically adjusts its monitoring angle based on the hydraulic support's position, dynamically tracking roof subsidence and coal wall deformation. Measurement data is transmitted to the data acquisition module via Industrial Ethernet with a transmission cycle of ≤10ms.

[0021] The microseismic sensor array is arranged in a triangular grid pattern along the sides of the tunnel and the working face cuts. The MS6100 microseismic sensors have a grid spacing of 5m×5m. Each sensor adopts a three-component design with a sensitivity of ≥2V / m / s, a frequency response of 1-1000Hz, and a dynamic range of ≥120dB. The sensor is connected to the underground microseismic data acquisition substation through a mine intrinsically safe cable. The substation adopts a DSP (Digital Signal Processor) and FPGA (Field-Programmable Gate Array) architecture and can process 32 channels of signals simultaneously. The system uses a time difference positioning algorithm to calculate the three-dimensional coordinates and energy release rate of the earthquake source through the microseismic signals received by at least 4 sensors. The positioning accuracy is ≤3m and the energy resolution is ≤ J.

[0022] The KBD5 intrinsically safe electromagnetic radiation sensor for mining is installed inside the hydraulic support shield beam, 1.2 meters from the coal face. The sensor has a measurement range of 0-1000mV / m, a frequency response of 1-100MHz, and a sampling frequency that can be dynamically adjusted between 1Hz and 100Hz. A linkage trigger mechanism is triggered by the action signal from the hydraulic support's electro-hydraulic control system: when a push-slide or lift action is detected, the sensor sampling frequency automatically switches to 50Hz (high-frequency acquisition mode). If the hydraulic support remains stationary for more than 5 minutes, the sampling frequency automatically drops to 1Hz (low-frequency monitoring mode). Collected data is transmitted to the data acquisition module via a 485 bus at a transmission rate of 9600bps.

[0023] The distributed sensor array achieves precise calibration of stress data through temperature compensation and dual-point monitoring of fiber Bragg grating pressure sensors, and combines the distributed series connection of piezoelectric stress sensors to form a three-dimensional perception network of the internal stress field of the coal body. The dynamic tracking of the laser displacement sensor is used to construct a correlation map between surface deformation and internal stress. Then, through the grid layout of the microseismic sensor array and the linkage trigger mechanism of the electromagnetic radiation sensor (intelligent switching of the acquisition frequency with the action of the hydraulic support), a full-dimensional mine pressure monitoring closed loop is formed from static to dynamic, from point measurement to surface monitoring, and from immediate status to evolution trend. It not only avoids the information island limitations of traditional single sensor monitoring, but also can capture the weak precursor signals of multi-physical field coupling during the manifestation of mine pressure, realizing the transformation of mine pressure risk from passive warning to early prediction.

[0024] Example 3 like Figure 3 As shown, the adaptive weight dynamic allocation model includes: The feature extraction layer is used to extract features from the output data of each sensor using a convolutional neural network to obtain the local feature vector corresponding to each sensor; The weight allocation layer is used to dynamically adjust the weight coefficient based on the data credibility of each sensor. The data credibility is calculated by the deviation between the current data and the historical benchmark data and the data fluctuation variance; The fusion decision layer is used to multiply the weight coefficient output by the weight distribution layer by the local feature vector output by the feature extraction layer element by element to obtain the weighted local feature vector, and fuse all the weighted local feature vectors to output the comprehensive feature value of the mine pressure.

[0025] In specific implementation, the adaptive weight dynamic allocation model described in this embodiment is built based on the TensorFlow 2.0 framework. The model input is the preprocessed sensor data matrix of five types (dimension is 100×5, where 100 is the time series length and 5 is the number of sensor types), and the output is the comprehensive characteristic value of mine pressure ranging from 0 to 100. The specific implementation of each layer is as follows: The feature extraction layer employs a three-layer convolutional neural network (CNN) architecture, with differentiated convolution kernels designed for different sensor data types. For numerical data from fiber Bragg grating pressure sensors and piezoelectric stress sensors, 1D convolution kernels (3×1 kernel size, 32 kernels) are used for feature extraction. For dynamic curve data from laser displacement sensors, 2D convolution kernels (3×3 kernel size, 64 kernels) are used to capture spatiotemporal variations. For waveform data from microseismic sensors and electromagnetic radiation sensors, 1D dilated convolutions (with a dilation rate of 2, kernel size 5×1, and 48 kernels) are used to enhance long-range dependencies. Each convolution layer is followed by a batch normalization (BatchNorm) layer and a leaky rectified linear unit (LeakyReLU) activation function (α=0.01). After global average pooling, the output is a 64-dimensional local feature vector (five sets of local feature vectors for five sensor types).

[0026] The weight allocation layer calculates data credibility using a two-factor evaluation model. First, 1000 data sets from the system's stable operation period are selected as a historical baseline dataset. The deviation D (D = |μ_current - μ_base| / σ_base, where μ is the mean and σ is the standard deviation) between the current sliding window (window size 10s) and the baseline data is calculated. The variance V (V = Var (current_data)) of the current window data is also calculated. Data credibility C is calculated using a normalized weighted formula: C = 0.6 × exp (-2D) + 0.4 × exp (-V / σ_base²), mapped to the range [0, 1]. The weight coefficients W_i are dynamically adjusted using a normalized exponential function (softmax): W_i = exp (C_i) / Σexp (C_j) (i, j = 1...5), ensuring that the sum of the weights is 1. If a sensor's data is abnormal (e.g., out of range), its weight coefficient is forcibly set to 0.01.

[0027] The fusion decision layer performs a three-level fusion operation: the first level multiplies the Wi output by the weight distribution layer by the local feature vector V_i output by the feature extraction layer element by element. , get the weighted feature vector; the second level uses the attention mechanism to splice the 5 groups of weighted feature vectors (using the learnable attention matrix A∈R^64×64, the attention matrix A is a 64-row 64-column real number matrix), and calculates the fusion feature matrix At the third level, M is reduced in dimension using a two-layer fully connected network (hidden layer dimension 128, output layer dimension 1). Sigmoid function is used to map the model to the interval [0, 100], and the output is the comprehensive characteristic value of the mine pressure. The model is trained using the Adam optimizer (initial learning rate 0.001, decaying by 10% every 50 rounds). The loss function is the mean squared error between the predicted value and the manually annotated risk value. The training batch size is 32, and 1000 rounds are repeated until the loss converges (validation set loss < 0.05).

[0028] The advantage of the above technical solution is that the convolutional neural network of the feature extraction layer captures the physical essence of different sensor data (such as the step characteristics of the pressure signal and the spectral characteristics of the microseismic signal) through differentiated convolution kernels, and cooperates with the weight distribution layer to perform a two-factor credibility assessment based on deviation and fluctuation variance, thereby realizing dynamic suppression of interference data (for example, the weight of the sensor temporarily impacted is automatically reduced to below 0.01) and enhanced amplification of the effective signal (for example, the weight of the sensor in the stress concentration area can be increased to above 0.3). This synergistic mechanism enables the comprehensive characteristic value of the mine pressure output by the fusion decision layer to not only remove the influence of environmental noise, but also capture the implicit correlation between multiple sensor data (for example, the leading 10s correlation between the sudden change of hydraulic support pressure and the release of microseismic energy), avoiding the limitations of traditional fixed weight fusion.

[0029] Example 4 like Figure 4 As shown, the real-time monitoring system for mine pressure in a comprehensive mining face based on multi-sensor fusion also includes an edge computing node, which is deployed between the data acquisition module and the multi-sensor fusion processing module, and is used to perform real-time filtering and feature dimensionality reduction processing on the pre-processed data to obtain a low-dimensional feature data set that meets spatiotemporal consistency. The low-dimensional feature data set contains the target feature parameters of each sensor data and the data volume is reduced to within a preset proportion range of the pre-processed multi-dimensional physical quantity data. The preset proportion range is dynamically adjusted according to the sensor type and data redundancy.

[0030] In this embodiment, the edge computing node uses an industrial-grade edge gateway and is deployed in an intrinsically safe control cabinet in the working face chute. It establishes communication with the data acquisition module (transmission rate 1Gbps) and the multi-sensor fusion processing module (transmission delay <20ms) through mining Gigabit Ethernet. It runs a real-time operating system based on Ubuntu 20.04 LTS, supports 24 parallel data processing channels, and has a maximum single-channel processing rate of 800Mbps.

[0031] Real-time filtering uses a hierarchical cascade architecture. To address the high-frequency noise (100-500 Hz) of fiber Bragg grating pressure sensors and piezoelectric stress sensors, a 32nd-order FIR (Finite Impulse Response Low-Pass Filter) low-pass filter (cutoff frequency 80 Hz, passband ripple 0.1 dB) is deployed in the first stage. An adaptive LMS filter algorithm (32nd order, step size 0.01) is used in the second stage to eliminate periodic electromagnetic interference. To address the random jitter noise of the laser displacement sensor, a Kalman filter is used (state equation x(k)=x(k-1)+v(k), observation equation z(k)=x(k)+w(k), process noise covariance Q=diag([0.01, 0.001]), measurement noise covariance R=0.1). To address the pulse interference of microseismic and electromagnetic radiation sensors, wavelet threshold denoising is used (using the db8 wavelet, 5 decomposition layers, and a soft threshold function λ=0.6745×median). (|w|)), all filtering algorithms are executed by real-time threads written in C++, and the processing delay is controlled within 50ms.

[0032] Feature dimensionality reduction processing consists of two steps: spatiotemporal registration and dimensionality compression. Spatiotemporal registration synchronizes the timestamps of each sensor data to UTC (accuracy ±1ms) through the NTP service (Network Time Protocol). The monitoring area is divided into 5m×5m spatial grids, and the multi-source data within each grid are bound to the same spatial coordinates. Dimensionality compression extracts target feature parameters for different sensor data: peak value, mean, variance, and kurtosis (4-dimensional features) are extracted for pressure / stress data; maximum subsidence, deformation rate, and cumulative displacement (3-dimensional features) are extracted for displacement data; energy, frequency, and peak amplitude are extracted for microseismic data (3-dimensional features); and intensity mean and spectral peak frequency are extracted for electromagnetic radiation data (2-dimensional features). Principal component analysis (PCA) is then used to compress the original 512-dimensional data into 32-dimensional feature vectors, retaining 95% of the variance contribution rate. The t-SNE algorithm is used during the compression process to maintain the local topological structure of the feature space, ensuring that the physical meaning of the reduced data is interpretable.

[0033] The dynamic data reduction mechanism is based on the redundancy assessment model. The system presets a basic ratio range of 15%-40%. During initialization, the initial ratio is allocated according to the sensor type (20% for microseismic / electromagnetic radiation sensors, 30% for pressure / stress sensors, and 35% for displacement sensors). The data redundancy index (R=1 - mutual information entropy / joint entropy) within each 5-minute sliding window is calculated in real time. When R>0.7 (high redundancy), the retention ratio of the corresponding sensor data is automatically reduced (by 5% each time, with a minimum of 15%). When R<0.3 (low redundancy), the retention ratio is increased (by 5% each time, with a maximum of 40%). For example, the microseismic sensor automatically reduces the data volume to 15% during the period without obvious mine pressure activity (R=0.82), retaining only event features with energy >10^4J. During the stress concentration period (R=0.25), the ratio is automatically increased to 40%, retaining complete waveform feature parameters. The final output low-dimensional feature dataset uses Protocol Buffer (Protocol Buffer). Buffers) format, including sensor ID, spatiotemporal coordinates, 32-dimensional feature vectors, and data reduction ratio labels. The size of a single frame of data is stable at 256 bytes (22.3% of the original data).

[0034] The edge computing node plays a key role in middle-layer processing by being deployed between the data acquisition module and the multi-sensor fusion processing module. Its real-time filtering processing designed for different sensor data characteristics can effectively eliminate environmental noise and interference signals, and improve data purity; feature dimensionality reduction processing accurately extracts the core feature parameters of each sensor data, greatly reduces the data dimension while retaining key information, reduces the load of data transmission and storage, and ensures the consistency of multi-source data in time and space dimensions through spatiotemporal alignment, avoiding analysis deviations caused by data dislocation; the preset ratio range mechanism based on dynamic adjustment of sensor type and data redundancy can automatically compress the data volume to improve transmission efficiency when data redundancy is high, and retain richer data details when key information is dense, achieving a dynamic balance between data processing efficiency and information integrity, so that the real-time response capability and resource utilization efficiency of the entire monitoring system are synergistically improved.

[0035] Example 5 The arrangement density of the distributed sensor array satisfies: In the hydraulic support area, each hydraulic support is equipped with at least two fiber grating pressure sensors, and the two sensors of the dual-point monitoring group are respectively arranged at the bottom of the column cylinder and the cylinder of the push jack; In the coal wall area, piezoelectric stress sensors are arranged along the strike at a first preset spacing, where the first preset spacing is an integer multiple of the hydraulic support spacing, and each stress monitoring chain includes at least three piezoelectric stress sensors connected in series; In the top plate area, laser displacement sensors are arranged along the inclination at a second preset spacing, wherein the second preset spacing is 0.5-1 times the first preset spacing, and the laser displacement sensor installed on each rotating pan / tilt head covers an angle range greater than or equal to ±30°; On both sides of the tunnel and in the working face cut area, a microseismic sensor array is arranged in a triangular grid, with the side length of each grid unit being 2-3 times the first preset spacing; An electromagnetic radiation sensor is configured on the inner side of each hydraulic support shield beam, and the arrangement height of the electromagnetic radiation sensor matches the height of the area with the strongest electromagnetic radiation signal of the coal wall; The layout positions of all sensors are located by three-dimensional coordinates and stored in the real-time monitoring terminal. The three-dimensional coordinate data includes sensor type, installation angle, measurement direction and relative position relationship with the nearest hydraulic support.

[0036] In a specific embodiment, each hydraulic support is equipped with two fiber Bragg grating (FBG) pressure sensors (with a measurement range of 0-60 MPa) in the hydraulic support area, forming a dual-point monitoring system. One sensor is vertically mounted on the central axis of the column cylinder bottom via a threaded interface, with the sensor's sensing surface tightly aligned with the inner wall of the cylinder bottom. This sensor is used to collect the column's axial working resistance. The other sensor is horizontally mounted on the inside of the rodless cavity end cap of the push jack cylinder. It is rigidly connected to the cylinder via a flange to collect push force data. For example, in a working face configuration with 150 hydraulic supports, a total of 300 fiber Bragg grating (FBG) pressure sensors are deployed in this area. The spacing between sensors on adjacent supports is consistent with the support center distance (1.5 m). All sensor cables are centrally stored through threading holes on the inside of the support shield beams and connected to the data collection substation in the chute.

[0037] In the coal wall area, piezoelectric stress sensors (for three-axis stress measurement) are arranged along the working face at a predetermined spacing of 3 meters, twice the spacing between hydraulic supports (1.5 meters). Each stress monitoring chain consists of three sensors connected in series and embedded within the coal wall using a 50 mm diameter borehole. The borehole depth is calculated according to the formula L = 0.3H + 0.5m (where H is the coal seam thickness, averaging 3.5 meters), with a predetermined depth of 1.55 meters. The three sensors are distributed axially within the borehole, with adjacent sensors spaced 0.5 meters apart. The top sensor is 0.3 meters from the coal wall surface, and the bottom sensor is 0.25 meters from the hole bottom. Epoxy resin grouting is used to secure the sensors to the borehole wall, ensuring a stress transfer efficiency of ≥90%.

[0038] In the roof area, laser displacement sensors (measurement accuracy ±0.01% FS) are placed at a second preset spacing along the inclination of the working surface. This second preset spacing is 0.5 times the first preset spacing (3m), or 1.5m. The sensor is mounted on the trunnion at the front end of the hydraulic support's top beam via an explosion-proof rotating pan-tilt platform. The platform has a horizontal rotation range of ±45° and a vertical pitch range of ±20°, ensuring an effective coverage angle of ≥±30° (horizontally). This allows for simultaneous monitoring of roof subsidence within a 1.5m range on both sides of the centerline. Every two hydraulic supports share one laser displacement sensor, and the sensor's laser emission port is maintained at a vertical distance of 0.8-1.0m from the roof surface to prevent mechanical obstruction during support movement.

[0039] A triangular grid of microseismic sensors (three components) was arranged along the sides of the tunnel and in the working face cutout area. The side length of each grid cell was set to 2.5 times the first preset spacing (3m), or 7.5m. The specific arrangement involved drilling holes (Ø50mm diameter, 1.5m depth) at the vertices of an equilateral triangle in the left and right sides of the tunnel and on the roof. The sensors were fixed in place with expansion bolts, with the sensor axes at a 45° angle to the horizontal. This ensured that the x-axis of the three components ran along the tunnel, the y-axis was perpendicular to the sides, and the z-axis was vertical. The triangular grid in the cutout area covered the hydraulic support support area and a 5m radius outside the goaf boundary. Sensors at the grid nodes were connected in series via intrinsically safe mining cables. Every 10 sensors were connected to a microseismic data acquisition substation, located in an explosion-proof junction box in the drift.

[0040] Regarding the placement of electromagnetic radiation sensors, a mining-grade intrinsically safe electromagnetic radiation sensor is installed on the inner side of each hydraulic support's shield beam. The sensor's installation height was determined through preliminary field testing: before mining resumed at the working face, temporary monitoring points were set up every 10 meters along the coal wall. The electromagnetic radiation signal strength at various heights (1.2-2.2 meters) was measured, and the average height of the strongest signal area (1.8 meters) was used as the fixed installation height. The sensor is connected to the steel plate on the inner side of the shield beam via a magnetic base and secured with a U-shaped clamp. Its detection surface faces the coal wall surface, maintaining a horizontal distance of 1.2 meters from the coal wall. This ensures that the relative position of the sensor and the coal wall deviates by no more than ±100 mm during the hydraulic support's raising and lowering process.

[0041] All sensors are positioned using a total station (with an angular measurement accuracy of 0.5") for three-dimensional coordinates. A local coordinate system (with the x-axis along the working face, the y-axis along the inclination, and the z-axis perpendicular) is established with the intersection of the working face cut and the transport chute as the coordinate origin (0,0,0). Positioning data includes the sensor's unique ID, device type, three-dimensional coordinates (x, y, z, accurate to 0.01m), installation angle (angle with the coordinate system axis, accurate to 0.1°), measurement direction vector, relative distance to the nearest hydraulic support (accurate to 0.1m), and azimuth (accurate to 1°). This data is uploaded to the real-time monitoring terminal via a wireless transmission module and stored in a spatial data table in the database. It can be accessed through the terminal's 3D visualization interface, enabling spatial correlation and display of sensor locations and monitoring data.

[0042] Example 6 The real-time monitoring terminal is also equipped with a mine pressure trend prediction model, which adopts a long short-term memory network LSTM to predict the mine pressure change trend within a preset time period in the future based on the historical mine pressure comprehensive characteristic value sequence and the real-time mine pressure comprehensive characteristic value.

[0043] like Figure 5 As shown, in this embodiment, the mine pressure trend prediction model is built based on the PyTorch framework and deployed on the GPU server of the real-time monitoring terminal. The specific implementation method is as follows: The model architecture utilizes a three-layer stacked LSTM structure. The input layer receives a sequence of historical pressure-comprehensive feature values ​​(time step T = 1440, corresponding to 24 hours of data) and real-time feature values ​​(dimension 1). Data concatenation involves concatenating the historical pressure feature sequence (time step T = 1440) with the real-time feature values ​​(dimension 1) along the feature dimension (e.g., using torch.cat ) to form a unified input tensor for the LSTM model's preprocessing. The hidden layer contains three LSTM units (with 128, 64, and 32 neurons, respectively). The output layer is mapped to a sequence of predicted pressure values ​​(dimension 720) for the next 12 hours (720 minutes) via a fully connected layer. A dropout layer (dropout rate 0.2) is added after each LSTM layer to prevent overfitting. The output layer uses a ReLU activation function to ensure non-negative predicted values. The model has approximately 200,000 parameters and is trained using the Adam (Adaptive Moment Estimation) optimizer with an initial learning rate of 0.001 that decays by 50% every 10 epochs.

[0044] The data preprocessing module performs Z-score normalization on the historical series of comprehensive characteristic values ​​of rock pressure. To maintain temporal characteristics, a sliding window approach is used to construct training samples (window size 1440, step size 60). Each sample contains 1440 historical values ​​and the corresponding 720 future real values. Missing values ​​are filled using cubic spline interpolation, with an interpolation error within ±0.5%.

[0045] The model was trained using historical data from the past six months (approximately 1.8 million records), split into training, validation, and test sets in an 8:1:1 ratio. The training batch size was 64, with each epoch consisting of 28,125 iterations, for a total of 100 epochs. Mean Squared Error (MSE) was used as the loss function, and Dynamic Time Warping (DTW) distance was introduced as an auxiliary metric to ensure the shape similarity between the predicted and true curves. An early stopping mechanism was used during training, terminating training if the validation set loss did not decrease for five consecutive epochs. The final mean absolute error (MAE) on the test set was kept below 3.2%.

[0046] During the real-time prediction process, the system collects the latest comprehensive characteristic values ​​of mine pressure every 10 minutes, updates the historical sequence (removing the earliest data and adding the latest data), and after preprocessing, inputs the data into a trained LSTM model. The model outputs a 12-hour trend curve for mine pressure changes, which is then denormalized to restore the actual values. The prediction results are displayed in three formats: a time series curve (the horizontal axis represents the next 12 hours, and the vertical axis represents the comprehensive characteristic values ​​of mine pressure); a heat map (using a color gradient to indicate the mine pressure risk level); and a digital warning (comparing the predicted value with the preset threshold to generate a risk level report).

[0047] The model update mechanism utilizes an online learning strategy, with incremental training performed daily at 2:00 AM using new data from the previous day. During incremental training, the parameters of the first two LSTM layers are frozen, and only the output layer weights are updated. The learning rate is reduced to 0.0001, and training is repeated for five epochs. A model performance monitoring system is also maintained. If the prediction error exceeds 5% for three consecutive days, a full retraining process is triggered to ensure the model's adaptability to dynamic changes in the work surface.

[0048] The warning trigger logic compares the predicted value for the next 12 hours with preset thresholds (safety threshold 50, warning threshold 70, and danger threshold 85). An alert is triggered when the following conditions are met: the predicted value exceeds the warning threshold; the slope of the prediction curve continuously increases by more than 0.1 / hour; or the rate of change of the predicted value exceeds 15% within the next two hours. Alert levels are divided into three levels: yellow (warning threshold), orange (danger threshold), and red (predicted value exceeds the danger threshold and continues to rise). Alert information is simultaneously notified to relevant personnel via audio and visual alarms, SMS push notifications, and system pop-up notifications.

[0049] The mine pressure trend prediction model is also equipped with an adaptive learning mechanism. When the deviation between the actual mine pressure change and the predicted mine pressure change trend exceeds a preset deviation threshold, the parameter update of the mine pressure trend prediction model is automatically triggered. The parameter update is based on incremental training of multiple sets of mine pressure data samples collected recently, and the incremental training adopts an online gradient descent algorithm.

[0050] Specifically, in this embodiment, the adaptive learning mechanism of the mine pressure trend prediction model is implemented in the following manner: A dedicated deviation calculation program is deployed on the real-time monitoring terminal. This program compares the actual monitored comprehensive characteristic values ​​of the mine pressure with the model predictions every 10 minutes. During this comparison, the system simultaneously evaluates two metrics: first, the similarity between the actual and predicted value sequences using a dynamic time warping algorithm; second, the absolute error percentage at each time point. If the sequence similarity falls below a preset threshold (corresponding to an original DTW distance exceeding 0.8) and the average error percentage over three consecutive time points exceeds 15%, the system determines that the current prediction deviation is excessive and automatically triggers a model parameter update process.

[0051] The system maintains a sliding window sample pool that holds the most recent 1,000 data sets. Each data set contains a historical feature value sequence from the past 24 hours and the corresponding actual observations for the next 12 hours. When a parameter update is triggered, the system randomly selects the 50 most recent data sets from the sample pool as incremental training sets, ensuring that the data used for model updates reflects the latest status of the current working area.

[0052] Online gradient descent uses the Adagrad (Adaptive Gradient Algorithm) optimization algorithm for incremental training, setting a small learning rate to prevent large fluctuations in model parameters. During parameter updates, the system freezes the parameters of the first two layers of the LSTM model and only updates the weights of the output layer and the final LSTM layer. This ensures that the model retains its memory of historical patterns while quickly adapting to new data features. Each incremental training cycle involves five rounds of iteration, processing 10 data sets per round. By accumulating gradient information over multiple rounds, the system simulates the effects of processing larger batches of data, ensuring the stability of parameter updates.

[0053] Before and after each parameter update, the system evaluates model performance using an independent validation dataset. Validation metrics include root mean square error (RMSE) and mean absolute percentage error (MAPE). If the RMSE of the updated model on the validation set decreases by more than 5% and the MAPE is less than 10%, the parameter update is considered valid. Otherwise, the system automatically rolls back to the pre-update parameter state and marks the sample causing the deviation as an outlier to prevent it from interfering with subsequent model training.

[0054] To balance the model's adaptability to new data with its stability to historical patterns, the system incorporates an exponential decay mechanism into the gradient update process. This mechanism weights historical gradient information, making recent gradients have a greater impact on parameter updates while the influence of older gradients gradually decreases. This design enables the model to rapidly respond to changes in working face geological conditions while retaining a memory of long-term patterns of rock pressure fluctuations.

[0055] During the exception handling process, if five consecutive parameter updates fail to effectively improve the model's prediction accuracy, the system automatically triggers a full retraining process. The system then retrains the entire LSTM model using all data from the last 30 days. After retraining, the new model is compared with the old model. By evaluating their performance on the validation set, the superior model is selected for production operation, ensuring that the system maintains good predictive capabilities even in extreme situations.

[0056] The system keeps a detailed log of each parameter update, including the trigger time, model performance metrics before and after the update, the number of samples trained, and the magnitude of the parameter gradient change. These logs are stored in a distributed file system in standard JSON format, enabling subsequent auditing and retrospective analysis, ensuring traceability and transparency of the model update process.

[0057] Example 7 The warning levels of the real-time monitoring terminal are divided into three levels, corresponding to abnormal mine pressure, serious abnormal mine pressure and dangerous mine pressure. Different levels of warning correspond to different processing plans, and the processing plans are stored in the plan database of the real-time monitoring terminal.

[0058] like Figure 6 As shown, the real-time monitoring terminal specifically includes: an adaptive control interface for pushing control suggestions based on the comprehensive characteristic value of mine pressure to the control system of the hydraulic support, the control suggestions including the support resistance adjustment amplitude and action timing, and receiving the execution feedback results returned by the control system of the hydraulic support.

[0059] Specifically, the real-time monitoring terminal's warning levels are based on a dual indicator: the comprehensive characteristic value of the mine pressure and its changing trend. Abnormal mine pressure (Level 1 warning) occurs when the comprehensive characteristic value is between 50 and 70, or when it increases by more than 10% within an hour but does not reach the warning level. Upon triggering, the system automatically sends a text message warning to the intrinsically safe mobile phones of the team leader and safety officer at the work face. A yellow warning box is displayed on the terminal interface. The contingency plan includes increasing monitoring frequency every 30 minutes, assigning inspection personnel to focus on checking the column travel of hydraulic supports and the integrity of the coal wall in the corresponding area, and recording abnormal data without affecting normal mining. When the characteristic value corresponding to a major abnormal mine pressure (secondary warning) is in the range of 70-85, or increases by more than 15% within 20 minutes, the terminal will emit a buzzing alarm (volume ≥85dB) for 10 seconds after being triggered, and the red warning light will flash. The emergency plan will be activated: the working face personnel will be notified through broadcasting to suspend coal mining operations and start the local support reinforcement process (such as controlling the pressure of the hydraulic support column to 1.1 times the rated value). The dispatching room will remotely retrieve the microseismic and stress change curves of the area within 1 hour, and the technical personnel will conduct remote analysis and judgment through the video conferencing module of the terminal. Mine pressure danger (Level 3 warning) corresponds to a characteristic value ≥ 85, or an increase of more than 20% within 5 minutes. In addition to audible and visual alarms, the terminal automatically sends an interlock signal to the mine safety monitoring system, forcibly shutting off the power to the shearer and face conveyor at the working face. The emergency plan requires all personnel to evacuate to the safe refuge chamber within 10 minutes, and the emergency support plan is activated (three adjacent hydraulic supports are simultaneously raised to their maximum travel, and the sliding jacks are locked). Simultaneously, an encrypted message containing real-time data, location coordinates, and the emergency plan number is sent via the 5G private network to the mine dispatch center and the Group Safety Supervision Department. The response plan is stored in the terminal's MySQL relational database, categorized by three dimensions. For example, the Level 3 warning plan for the fault zone is stored separately. The technical department updates the plan monthly based on recent mining data and accident cases. Update records are electronically signed and timestamped to ensure traceability.

[0060] The adaptive control interface of the real-time monitoring terminal uses an intrinsically safe Ethernet switch (supporting IEEE 802.3af) to communicate with the hydraulic support's electro-hydraulic control system. The physical interface is a circular connector with a key for preventing mis-insertion. The communication protocol uses Modbus TCP / IP, with a data rate of 100 Mbps and a transmission delay of ≤50 ms. The logic for generating control recommendations is as follows: the terminal calculates the comprehensive characteristic value of the mine pressure every five minutes. When the characteristic value is between 40 and 50, it generates a recommendation to maintain the current support; when it is between 50 and 65, it generates a recommendation to fine-tune the support resistance (for example, increasing the column pressure by 5%-8%); and when it exceeds 65, it generates a recommendation to strengthen the support (increasing the pressure by 10%-15%). The timing of action is determined by combining the shearer's position signal (obtained via UWB positioning with an accuracy of ±30 cm). For example, when the shearer is more than 5 meters away from the target support and is not cutting coal, the push jack is triggered; if the shearer has been stopped for more than three minutes, the column pressure is adjusted. The control suggestion is encapsulated in JSON format, including the target support number, support resistance adjustment range, action execution time window and safety verification code. After the hydraulic support control system is executed, the feedback result is returned through the same interface, including the actual adjusted pressure value, action completion time, whether there is an execution abnormality (such as column seal leakage) and other information. The terminal verifies the feedback result. If the actual adjustment value deviates from the recommended value by more than 3%, a secondary adjustment suggestion is automatically generated; if there are three consecutive abnormal feedbacks, a secondary warning is triggered and maintenance personnel are notified. The interface has a built-in watchdog timer, which automatically records breakpoint data when communication is interrupted. After the connection is restored, unfinished control instructions and feedback information are transmitted first to ensure the reliability of the control link.

[0061] Example 8 like Figure 7 As shown, this embodiment provides a method for real-time monitoring of mine pressure in a fully-mechanized mining working face based on multi-sensor fusion. The method is based on the real-time monitoring system for mine pressure in a fully-mechanized mining working face based on multi-sensor fusion, and the method includes: S10: Collecting multi-dimensional physical quantity data related to mine pressure through a distributed sensor array deployed in the target stress area of ​​the fully mechanized mining face, wherein the target stress area includes the structural components of the hydraulic support, the interior and surface of the coal wall, the surrounding rock of the roadway, the boundary of the goaf, and the roof support interface; S20: Preprocessing the multi-dimensional physical quantity data through a data acquisition module to obtain preprocessed multi-dimensional physical quantity data; S30: performing spatiotemporal registration and feature layer fusion on the pre-processed multi-dimensional physical quantity data through a multi-sensor fusion processing module equipped with an adaptive weight dynamic allocation model, and outputting a comprehensive characteristic value of the mine pressure; S40: The comprehensive characteristic value of the mine pressure is received through a real-time monitoring terminal and displayed in real time, and an intelligent analysis is performed based on the comprehensive characteristic value of the mine pressure to obtain a trend of mine pressure change. When the comprehensive characteristic value of the mine pressure exceeds a preset threshold and / or the trend of mine pressure change enters a dangerous range, an early warning is triggered.

[0062] This real-time monitoring method for mine pressure in fully mechanized mining working faces based on multi-sensor fusion realizes the comprehensive collection of multi-dimensional physical quantity data related to mine pressure by deploying distributed sensor arrays in all directions in the target stress area, overcoming the limitations of traditional single-area monitoring. After preprocessing by the data acquisition module, the data is subjected to spatiotemporal alignment and feature layer fusion with the help of a multi-sensor fusion processing module configured with an adaptive weight dynamic allocation model. The weight can be dynamically adjusted according to the credibility of different sensor data, so that the output comprehensive characteristic value of mine pressure can more accurately reflect the overall state of mine pressure, avoiding the deviation caused by simple data superposition. The real-time monitoring terminal realizes closed-loop management of the entire process from data collection to risk warning through real-time display, intelligent analysis of the comprehensive characteristic value of mine pressure and triggering of early warning, which not only ensures real-time control of mine pressure changes, but also can identify dangers in advance through trend analysis. Different levels of warnings correspond to differentiated processing plans, further improving the pertinence and effectiveness of responding to mine pressure anomalies, and overall improving the comprehensiveness, accuracy and timeliness of mine pressure monitoring in fully mechanized mining working faces.

[0063] In some embodiments, the step of collecting multi-dimensional physical quantity data related to the mine pressure in the target stress area of ​​the fully mechanized mining working face through a distributed sensor array includes: A dual-point monitoring system is formed by fiber Bragg grating pressure sensors placed at the bottom of the hydraulic support column cylinder and the push jack cylinder to collect the working resistance and pushing force of the hydraulic support. The temperature compensation algorithm is used to eliminate the influence of ambient temperature on the measured values. The three-dimensional stress field distribution inside the coal body is collected by piezoelectric stress sensors buried at a preset depth inside the coal wall. The preset depth is determined according to the coal seam thickness and mining process parameters of the fully mechanized mining face. The piezoelectric stress sensors are connected in series using distributed optical fibers to form a stress monitoring chain. The laser displacement sensor installed on the rotating platform at the front end of the hydraulic support top beam is used to dynamically track and collect the roof subsidence and coal wall deformation. The energy release rate and source location of microseismic events in the coal and rock mass are monitored in real time by using a microseismic sensor array arranged in a triangular grid along the sides of the roadway and the working face cuts. The microseismic sensor array covers the support area of ​​the hydraulic support and the boundary of the goaf. With the help of an electromagnetic radiation sensor arranged on the inner side of the hydraulic support shield beam, the electromagnetic radiation signal of the stress concentration area of ​​the coal rock mass is obtained. The collection of the electromagnetic radiation signal and the action of the hydraulic support form a linkage trigger mechanism. The linkage trigger mechanism includes: when the hydraulic support performs a pushing or lifting action, the electromagnetic radiation sensor automatically enters a high-frequency collection mode; when the hydraulic support is in a static state, the electromagnetic radiation sensor automatically switches to a low-frequency duty mode. The sampling frequency of the high-frequency collection mode is higher than the sampling frequency of the low-frequency duty mode.

[0064] In some embodiments, the step of performing data processing by the adaptive weight dynamic allocation model includes: The feature extraction layer uses a convolutional neural network to extract features from the output data of each sensor and obtain the local feature vector corresponding to each sensor; The weight distribution layer dynamically adjusts the weight coefficient based on the data credibility of each sensor. The data credibility is calculated by the deviation between the current data and the historical benchmark data and the data fluctuation variance. The weight coefficient output by the weight distribution layer is multiplied element by element by the local feature vector output by the feature extraction layer through the fusion decision layer to obtain the weighted local feature vector, and all the weighted local feature vectors are fused to output the comprehensive feature value of the mine pressure.

[0065] In some embodiments, the method further includes: deploying an edge computing node between the data acquisition module and the multi-sensor fusion processing module, and using the edge computing node to perform real-time filtering and feature dimensionality reduction processing on the preprocessed data to obtain a low-dimensional feature data set that satisfies spatiotemporal consistency. The low-dimensional feature data set includes target feature parameters of each sensor data and the amount of data is reduced to a preset proportion range of the preprocessed multi-dimensional physical quantity data, and the preset proportion range is dynamically adjusted according to the sensor type and data redundancy.

[0066] In some embodiments, the distributed sensor array arrangement step satisfies: In the hydraulic support area, each hydraulic support is equipped with at least two fiber grating pressure sensors, and the two sensors of the dual-point monitoring group are respectively arranged at the bottom of the column cylinder and the cylinder of the push jack; In the coal wall area, piezoelectric stress sensors are arranged along the strike at a first preset spacing, where the first preset spacing is an integer multiple of the hydraulic support spacing, and each stress monitoring chain includes at least three piezoelectric stress sensors connected in series; In the top plate area, laser displacement sensors are arranged along the inclination at a second preset spacing, wherein the second preset spacing is 0.5-1 times the first preset spacing, and the laser displacement sensor installed on each rotating pan / tilt head covers an angle range greater than or equal to ±30°; On both sides of the tunnel and in the working face cut area, a microseismic sensor array is arranged in a triangular grid, with the side length of each grid unit being 2-3 times the first preset spacing; An electromagnetic radiation sensor is configured on the inner side of each hydraulic support shield beam, and the arrangement height of the electromagnetic radiation sensor matches the height of the area with the strongest electromagnetic radiation signal of the coal wall; The layout positions of all sensors are located by three-dimensional coordinates and stored in the real-time monitoring terminal. The three-dimensional coordinate data includes sensor type, installation angle, measurement direction and relative position relationship with the nearest hydraulic support.

[0067] In some embodiments, the step of the real-time monitoring terminal performing intelligent analysis to obtain the mine pressure change trend also includes: configuring a mine pressure trend prediction model, the mine pressure trend prediction model adopts a long short-term memory network LSTM, based on the historical mine pressure comprehensive characteristic value sequence and the real-time mine pressure comprehensive characteristic value, to predict the mine pressure change trend within a preset time period in the future.

[0068] In some embodiments, the method further includes: configuring an adaptive learning mechanism for the mine pressure trend prediction model, and automatically triggering parameter updates of the mine pressure trend prediction model when the deviation between the actual mine pressure change and the predicted mine pressure change trend exceeds a preset deviation threshold, and the parameter updates are incrementally trained based on the latest collected multiple groups of mine pressure data samples, and the incremental training adopts an online gradient descent algorithm.

[0069] In some embodiments, the step of triggering an early warning by the real-time monitoring terminal includes: dividing the early warning level into three levels, corresponding to abnormal mine pressure, major abnormal mine pressure and dangerous mine pressure, respectively. Different levels of early warning correspond to different processing plans. The processing plans are stored in the plan database of the real-time monitoring terminal, and the processing plans of the corresponding levels are called when the early warning is triggered.

[0070] In some embodiments, the operating steps of the real-time monitoring terminal also include: pushing control suggestions based on the comprehensive characteristic value of mine pressure to the control system of the hydraulic support through the adaptive control interface, the control suggestions including the support resistance adjustment amplitude and action timing, and receiving the execution feedback results returned by the control system of the hydraulic support.

[0071] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0072] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein the program implements any one of the above methods when executed by a processor.

[0073] If the integrated modules / units are implemented as software functional units and sold or used as standalone products, they can be stored in a computer-readable storage medium. Based on this understanding, the present invention can also implement all or part of the process steps in the above-mentioned method embodiments by using a computer program to instruct the relevant hardware. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium. Of course, there are other types of readable storage media, such as quantum memory and graphene memory. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practices in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practices, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0074] The present invention also provides an electronic device. The electronic device according to an embodiment of the present invention includes: one or more processors; and a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method provided by the present invention.

[0075] Reference below Figure 8 , which shows a schematic structural diagram of a computer system 800 of an electronic device suitable for implementing an embodiment of the present invention. Figure 8 The electronic device shown is only an example and should not limit the functions and scope of use of the embodiments of the present invention.

[0076] like Figure 8 As shown, computer system 800 includes a central processing unit (CPU) 801, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 802 or a program loaded from a storage unit 808 into a random access memory (RAM) 803. Various programs and data required for the operation of computer system 800 are also stored in RAM 803. CPU 801, ROM 802, and RAM 803 are connected to each other via a bus 804. An input / output (I / O) interface 805 is also connected to bus 804.

[0077] The following components are connected to the I / O interface 805: an input section 806 including a keyboard, mouse, and the like; an output section 807 including devices such as a cathode ray tube (CRT), a liquid crystal display (LCD), and speakers; a storage section 808 including a hard disk; and a communication section 809 including a network interface card such as a LAN card or a modem. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the I / O interface 805 as needed. Removable media 811, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, is installed in the drive 810 as needed, so that computer programs read from the removable media can be installed in the storage section 808 as needed.

[0078] In particular, according to embodiments disclosed herein, the processes described in the main step diagrams above can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for executing the methods shown in the main step diagrams. In the above embodiments, the computer program can be downloaded and installed from a network via the communication section 809 and / or installed from removable media 811. When the computer program is executed by the central processing unit 801, the above-described functions defined in the system of the present invention are performed.

[0079] It should be noted that the computer-readable medium described in the present invention may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. Computer-readable storage media may include, but are not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or components, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection having one or more conductors, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device, or component. In the present invention, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. This propagated data signal may take a variety of forms, including, but not limited to, electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device. Program code embodied on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wireline, optical cable, RF, or any suitable combination thereof.

[0080] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the above-mentioned module, program segment, or a part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0081] The above specific embodiments do not limit the scope of protection of the present invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may occur depending on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A real-time monitoring system for mine pressure in fully mechanized mining working face based on multi-sensor fusion, characterized in that: include: A distributed sensor array is deployed in the target stress area of ​​the fully mechanized mining face, including the structural components of the hydraulic support, the interior and surface of the coal wall, the surrounding rock of the roadway, the boundary of the goaf, and the roof support interface, to collect multi-dimensional physical quantity data related to mine pressure; a data acquisition module, communicatively connected to the distributed sensor array, and configured to preprocess the multi-dimensional physical quantity data to obtain preprocessed multi-dimensional physical quantity data; A multi-sensor fusion processing module is connected to the data acquisition module and is configured with an adaptive weight dynamic allocation model for performing spatiotemporal registration and feature layer fusion on the pre-processed multi-dimensional physical quantity data to output a comprehensive characteristic value of mine pressure; A real-time monitoring terminal is communicatively connected to the multi-sensor fusion processing module, and is used to receive the comprehensive characteristic value of the mine pressure and display it in real time, and to perform intelligent analysis based on the comprehensive characteristic value of the mine pressure to obtain the trend of mine pressure change. When the comprehensive characteristic value of the mine pressure exceeds a preset threshold and / or the trend of mine pressure change enters a dangerous range, an early warning is triggered.

2. The system according to claim 1, wherein: The distributed sensor array comprises: Fiber Bragg grating pressure sensors are installed at the bottom of the hydraulic support column cylinder and the push jack cylinder to form a dual-point monitoring group for collecting the working resistance and pushing force of the hydraulic support. The temperature compensation algorithm is used to eliminate the influence of ambient temperature on the measured values. Piezoelectric stress sensors are buried at a preset depth inside the coal wall. The preset depth is determined based on the coal seam thickness and mining process parameters of the fully mechanized mining face. Distributed optical fibers are connected in series to form a stress monitoring chain to collect the three-dimensional stress field distribution inside the coal body. The laser displacement sensor is installed on the rotating platform at the front end of the hydraulic support top beam to dynamically track and collect the amount of roof subsidence and coal wall deformation; The microseismic sensor array is arranged in a triangular grid along the sides of the roadway and the working face cuts, covering the support area of ​​the hydraulic support and the boundary of the goaf. It is used to monitor the energy release rate and source location of microseismic events in the coal rock in real time. The electromagnetic radiation sensor is arranged on the inner side of the hydraulic support shield beam and is used to obtain the electromagnetic radiation signal of the stress concentration area of ​​the coal rock mass. The collection of the electromagnetic radiation signal forms a linkage trigger mechanism with the action of the hydraulic support. The linkage trigger mechanism includes: when the hydraulic support performs a pushing or lifting action, the electromagnetic radiation sensor automatically enters the high-frequency collection mode; when the hydraulic support is in a static state, the electromagnetic radiation sensor automatically switches to the low-frequency duty mode. The sampling frequency of the high-frequency collection mode is higher than the sampling frequency of the low-frequency duty mode.

3. The system according to claim 1, wherein: The adaptive weight dynamic allocation model includes: The feature extraction layer is used to extract features from the output data of each sensor using a convolutional neural network to obtain the local feature vector corresponding to each sensor; The weight allocation layer is used to dynamically adjust the weight coefficient based on the data credibility of each sensor. The data credibility is calculated by the deviation between the current data and the historical benchmark data and the data fluctuation variance; The fusion decision layer is used to multiply the weight coefficient output by the weight distribution layer by the local feature vector output by the feature extraction layer element by element to obtain the weighted local feature vector, and fuse all the weighted local feature vectors to output the comprehensive feature value of the mine pressure.

4. The system according to claim 1, wherein: It also includes an edge computing node, which is deployed between the data acquisition module and the multi-sensor fusion processing module, and is used to perform real-time filtering and feature dimensionality reduction processing on the preprocessed data to obtain a low-dimensional feature data set that meets spatiotemporal consistency. The low-dimensional feature data set contains the target feature parameters of each sensor data and the data volume is reduced to a preset proportion range of the preprocessed multi-dimensional physical quantity data. The preset proportion range is dynamically adjusted according to the sensor type and data redundancy.

5. The system according to claim 2, wherein: The arrangement density of the distributed sensor array satisfies: In the hydraulic support area, each hydraulic support is equipped with at least two fiber grating pressure sensors, and the two sensors of the dual-point monitoring group are respectively arranged at the bottom of the column cylinder and the cylinder of the push jack; In the coal wall area, piezoelectric stress sensors are arranged along the strike at a first preset spacing, where the first preset spacing is an integer multiple of the hydraulic support spacing, and each stress monitoring chain includes at least three piezoelectric stress sensors connected in series; In the top plate area, laser displacement sensors are arranged along the inclination at a second preset spacing, wherein the second preset spacing is 0.5-1 times the first preset spacing, and the laser displacement sensor installed on each rotating pan / tilt head covers an angle range greater than or equal to ±30°; On both sides of the tunnel and in the working face cut area, a microseismic sensor array is arranged in a triangular grid, with the side length of each grid unit being 2-3 times the first preset spacing; An electromagnetic radiation sensor is configured on the inner side of each hydraulic support shield beam, and the arrangement height of the electromagnetic radiation sensor matches the height of the area with the strongest electromagnetic radiation signal of the coal wall; The layout positions of all sensors are located and stored in the real-time monitoring terminal through three-dimensional coordinate data. The three-dimensional coordinate data includes sensor type, installation angle, measurement direction and relative position relationship with the nearest hydraulic support.

6. The system according to claim 1, wherein: The real-time monitoring terminal is also equipped with a mine pressure trend prediction model, which adopts a long short-term memory network LSTM to predict the mine pressure change trend within a preset time period in the future based on the historical mine pressure comprehensive characteristic value sequence and the real-time mine pressure comprehensive characteristic value.

7. The system according to claim 6, characterized in that The mine pressure trend prediction model is also equipped with an adaptive learning mechanism. When the deviation between the actual mine pressure change and the predicted mine pressure change trend exceeds a preset deviation threshold, the parameter update of the mine pressure trend prediction model is automatically triggered. The parameter update is based on incremental training of multiple sets of mine pressure data samples collected recently, and the incremental training adopts an online gradient descent algorithm.

8. The system according to claim 1, wherein: The warning levels of the real-time monitoring terminal are divided into three levels, corresponding to abnormal mine pressure, serious abnormal mine pressure and dangerous mine pressure. Different levels of warning correspond to different processing plans, and the processing plans are stored in the plan database of the real-time monitoring terminal.

9. The system according to claim 1, wherein: The real-time monitoring terminal specifically includes: The adaptive control interface is used to push control suggestions based on the comprehensive characteristic value of mine pressure to the control system of the hydraulic support. The control suggestions include the support resistance adjustment range and action timing, and receive the execution feedback results returned by the control system of the hydraulic support.

10. A real-time monitoring method for mine pressure in fully mechanized mining working face based on multi-sensor fusion, characterized in that: The method is based on the real-time monitoring system for mine pressure in a fully-mechanized mining working face based on multi-sensor fusion according to any one of claims 1 to 9, and the method comprises: Multi-dimensional physical quantity data related to mine pressure is collected by a distributed sensor array deployed in the target stress area of ​​the fully mechanized mining face. The target stress area includes the structural components of the hydraulic support, the interior and surface of the coal wall, the surrounding rock of the roadway, the boundary of the goaf and the roof support interface. Preprocessing the multi-dimensional physical quantity data by a data acquisition module to obtain preprocessed multi-dimensional physical quantity data; Through a multi-sensor fusion processing module equipped with an adaptive weight dynamic allocation model, the pre-processed multi-dimensional physical quantity data is subjected to spatiotemporal registration and feature layer fusion to output a comprehensive characteristic value of the mine pressure; The comprehensive characteristic value of the mine pressure is received by a real-time monitoring terminal and displayed in real time. The mine pressure change trend is obtained by intelligent analysis based on the comprehensive characteristic value of the mine pressure. When the comprehensive characteristic value of the mine pressure exceeds a preset threshold and / or the mine pressure change trend enters a dangerous range, an early warning is triggered.

Images