Multi-mining area cluster unmanned mining physical environment dynamic deduction and generation method and system

Abstract

The application relates to the technical field of mining areas, and discloses a multi-mining-area cluster unmanned mining physical environment dynamic deduction and generation method and system. The method comprises the following steps: arranging a multi-type fiber grating sensor integrated sensing network in a target stope; using a three-dimensional laser scanner to scan the target stope and generating a stope three-dimensional space environment model; collecting physical environment parameters of the target stope in real time and data assimilation; constructing a multi-working-condition numerical simulation training database; constructing and training a multi-mining-area physical environment deduction model based on a BP neural network; dynamically deducing a multi-mining-area cluster physical environment and generating a multi-mining-area unmanned mining environment; and performing unmanned equipment operation environment feedback control according to the multi-mining-area unmanned mining environment. The application provides real-time, dynamic and predictable physical environment description for multi-mining-area cluster unmanned mining.

Description

Technical Field

[0001] This invention relates to the field of mining technology, specifically to a method and system for dynamic simulation and generation of the physical environment of unmanned mining in multi-mining area clusters. Background Technology

[0002] One of the core challenges facing unmanned mining in multi-area clusters of deep metal deposits is the complexity and time-varying nature of the stope's physical environment. The stress, displacement, and seepage fields in the surrounding rock continuously evolve during mining, directly impacting the operational safety and production efficiency of unmanned mining equipment. Currently, methods for obtaining the stope's physical environment mainly fall into two categories: The first category is the traditional vibrating wire sensor monitoring method. This method uses devices such as vibrating wire stress gauges, displacement gauges, and piezometers to monitor the surrounding rock of the mining area at specific points. However, vibrating wire sensors have significant technical drawbacks: they are susceptible to electromagnetic interference and humid environments underground, resulting in insufficient monitoring accuracy and long-term stability; they require multi-core cables for signal transmission, leading to complex wiring and high costs; and single-point measurement methods make it difficult to achieve continuous sensing of the physical field of the mining area. In existing solutions, sensor monitoring data and numerical simulation calculations operate independently, with monitoring data and numerical models being disconnected. Monitoring data is only used for threshold alarms, and numerical models are only used for trend prediction. The lack of a data assimilation mechanism leads to a disconnect between monitoring and prediction.

[0003] The second category is single numerical simulation prediction methods. This method establishes a finite element or finite difference model of the mining area to numerically predict stress and displacement changes during the mining process. However, this method has the following problems: it cannot obtain real-time monitoring data for dynamic model correction, leading to systematic deviations between the prediction results and the actual situation; the computational load is enormous, making it difficult to meet the real-time prediction requirements of multi-mining area cluster mining; it ignores the two-way interaction between monitoring data and the numerical model, essentially making it an "offline" prediction; it lacks the ability to extrapolate the physical environment at the multi-mining area cluster scale; existing numerical simulation methods have high computational load and slow response speed, making it difficult to achieve rapid extrapolation under conditions of simultaneous operation in multiple mining areas; changes in the physical environment of a single mining area can affect adjacent mining areas, but existing methods cannot complete multi-mining area collaborative extrapolation within seconds or minutes.

[0004] Fiber Bragg grating (FBG) sensors, as a novel sensing technology, have been applied in the field of geotechnical engineering monitoring. However, the application of existing FBG sensing technology in mining engineering is still limited to the monitoring of single points and single types of parameters. It lacks integrated sensing devices with multiple types of sensors (displacement gauges, strain gauges, piezometers), and even more so, it lacks a complete technical loop from "real-time monitoring of a single mining area" to "multi-mining area cluster extrapolation." In other words, it lacks an environmental extrapolation model from "single mining area" to "multi-mining area." Existing methods can only handle the physical environment of a single mining area and cannot extrapolate the evolution trend of the physical environment of a multi-mining area cluster based on single mining area monitoring data.

[0005] Furthermore, the application of 3D laser scanning technology in mines is mainly concentrated in static scenarios such as goaf boundary detection and shaft deformation monitoring. It has not yet been effectively integrated with physical environment perception data to form a dynamic environment description that supports unmanned equipment operations. There is a lack of dynamic interaction between the physical environment and unmanned equipment. When unmanned equipment operates in the mining area, it needs to obtain the physical environment status of its current location in real time to make safety decisions. However, existing technologies cannot feed the perception data back to the equipment control system in real time to form a closed loop of "environmental perception → equipment control".

[0006] To address these issues, we have developed a method and system for dynamic simulation and generation of the physical environment of unmanned mining in multi-area clusters, which solves the aforementioned technical problems. Summary of the Invention

[0007] This invention provides a method and system for dynamic simulation and generation of the physical environment of unmanned mining in multi-mining area clusters. By integrating real-time sensing of multiple types of fiber optic grating sensors, spatial modeling of three-dimensional laser scanning, and multi-mining area simulation model of BP neural network, it provides a real-time, dynamic, and predictable physical environment description for unmanned mining in multi-mining area clusters.

[0008] Therefore, the present invention provides the following technical solution: A method for dynamically extrapolating and generating the physical environment of unmanned mining in multi-area clusters, the method comprising: Step 1: Deploy an integrated sensing network of multiple types of fiber optic grating sensors within the target mining area; Step 2: Use a 3D laser scanner to scan the target mining area and generate a 3D spatial environment model of the mining area; Step 3: Connect multiple types of fiber optic grating sensors to the physical environment sensing device, and use the physical environment sensing device to collect and assimilate the physical environment parameters of the target mining site in real time. Step 4: Construct a multi-condition numerical simulation training database; Step 5: Construct and train a multi-sampling area physical environment simulation model based on BP neural network; Step 6: Using the trained multi-mining zone physical environment simulation model based on BP neural network, perform dynamic simulation of the physical environment of the multi-mining zone cluster to generate the unmanned mining environment of the multi-mining zone. Step 7: Based on the unmanned mining environment in the multi-mining area, implement feedback control of the unmanned equipment operation environment.

[0009] Optionally, in step 1, a multi-type fiber optic grating sensor integrated sensing network is deployed at key locations such as the retreat roadways, cross-cut roadways, and stope roof of the target stope. The multi-type fiber optic grating sensors include: fiber optic grating displacement gauges, fiber optic grating embedded strain gauges, and fiber optic grating piezometers. The fiber optic grating displacement gauge is used to measure the displacement and settlement of multiple parts in the deep surrounding rock of the stope. It senses external displacement through a measuring probe, converts it into the deflection of an equal-strength beam through a wedge, and two fiber optic gratings attached to the upper and lower surfaces of the equal-strength beam sense the strain change. The fiber optic grating embedded strain gauge adopts a thin-necked tube protective encapsulation structure, consisting of a protective steel tube, transmission optical fiber, fiber optic grating, sleeve, and bonding material, and is used to monitor the strain field changes inside the surrounding rock of the stope. The fiber optic grating piezometer adopts a diaphragm structure encapsulation, which transmits external water seepage pressure to the diaphragm structure through a permeable stone, causing the fiber optic grating to stretch or contract, and is used to monitor the seepage field changes in the surrounding rock of the stope.

[0010] Optionally, step 2 includes: Step 21: Set up a main station at the junction of the cross-cut roadway and the segmented horizontal roadway in the target mining area, set up an auxiliary station at the junction of the cross-cut roadway and the return roadway, and set up supplementary stations at regular intervals to cover the area blocked by protruding rock blocks. Step 22: Using a target-based stitching method, at least three reflective targets are ensured between adjacent scanning stations. The targets are evenly distributed to cover the scanning object. A 3D laser scanner is used for fine scanning to obtain the original point cloud data. Step 23: Take pictures of the scanning environment using the built-in camera or an external digital camera to obtain color information, which is used to match the original point cloud data and give the original point cloud data realistic color and texture information. Step 24: The original point cloud data is denoised using the MCMD-Z automatic denoising algorithm to obtain high-density three-dimensional point cloud data. The high-density three-dimensional point cloud data is then stitched together, coordinate transformed, and texture mapped to generate the three-dimensional spatial environment model of the mining site.

[0011] Optionally, in step 3, based on the geological conditions and mining plan of the target mining area, a FLAC3D three-dimensional numerical calculation model is established to calculate the stress and displacement field variations of the target mining area under different mining progresses, obtaining numerical simulation results; the numerical simulation results are compared with the actual monitoring data collected by the integrated sensing network of multiple types of fiber optic grating sensors to calculate the systematic error E of the numerical simulation. ; in, For the first The stress values ​​obtained from numerical simulation at each monitoring point For the first The actual stress values ​​obtained from monitoring at each monitoring point The total number of monitoring points is used; the numerical simulation results are systematically reduced and corrected according to the calculated systematic error, so that the corrected numerical simulation results are consistent with the actual monitoring data in a statistical sense, thus completing the initial data assimilation of the actual monitoring data with the FLAC3D three-dimensional numerical calculation model.

[0012] Optionally, in step 4, based on the corrected FLAC3D three-dimensional numerical calculation model, a multi-condition extended simulation is performed on the target mining area to generate the multi-condition numerical simulation training database. The condition variables include: cross-sectional dimensions of the mining roadway, mining depth, mining progress, distance to the working face, measuring point height, and roadway location. Numerical calculations are performed for each condition, and stress and displacement data at each monitoring point in the multi-type fiber optic grating sensor integrated sensing network are extracted. The obtained error coefficients are systematically reduced and corrected to form the multi-condition numerical simulation training database containing multiple sets of sample data.

[0013] Optionally, in step 5, the constructed multi-mining area physical environment simulation model based on BP neural network includes an input layer, a hidden layer, and an output layer. The input layer includes 6 nodes, which correspond to 6 working condition variables in the multi-working condition numerical simulation training database: the cross-sectional size of the mining roadway, the mining middle section, the mining progress, the distance to the working face, the height of the measuring point, and the roadway location. The distance to the working face and the mining progress are dynamically updated by real-time monitoring data, while the cross-sectional size of the mining roadway, the height of the measuring point, the roadway location, and the mining middle section are determined by the working condition variables in step 4. The hidden layer adopts a two-layer hidden layer structure, with 12 nodes in the first hidden layer and 8 nodes in the second hidden layer. The interlayer transfer function adopts an S-shaped function. The output layer includes two nodes, namely the surrounding rock stress and the surrounding rock displacement. When training the multi-sampling area physical environment simulation model based on BP neural network, the learning efficiency, target accuracy and maximum number of training iterations are set. Multiple sets of data in the multi-condition numerical simulation training database are used as the training set and multiple sets of data are used as the test set to train the model. During the training process, the error function is backpropagated, and the error function is made to reach the expected value by continuously adjusting the threshold and network weights.

[0014] Optionally, step 6 includes: Step 61: For the currently being mined single mining area, update the input parameters of the multi-mining area physical environment inference model based on BP neural network according to real-time monitoring data. The multi-mining area physical environment inference model based on BP neural network outputs the stress and displacement prediction values ​​of the current location in real time. Step 62: For adjacent mining areas that have not yet been mined, according to the planned mining sequence and mining area structure parameters, input the corresponding network parameters into the multi-mining area physical environment inference model based on BP neural network. The multi-mining area physical environment inference model based on BP neural network infers the stress and displacement evolution trend of the mining area at different mining stages. Step 63: For scenarios where multiple mining areas operate simultaneously, the physical environment simulation model based on the BP neural network simulates the physical environment changes of each mining area in parallel, analyzes the superposition effect of mining stress, and generates physical environment data. Step 64: The physical environment data generated by the simulation is fused with the three-dimensional spatial environment model of the mining area to generate a complete unmanned mining environment for multiple mining areas, including four dimensions: spatial geometric information, stress field distribution, displacement field distribution, and seepage field distribution.

[0015] Optionally, step 7 includes: Step 71: Based on the current physical environment of the mining area, determine the safety level of each work area and divide it into safe work areas, warning work areas, and prohibited work areas; Step 72: When monitoring or simulation data shows that the stress value in a certain area is close to or exceeds the preset threshold, a warning signal is sent to the unmanned equipment control system to adjust the equipment operation path or suspend operation. Step 73: Feedback the trend of physical environment changes to the unmanned equipment scheduling system, dynamically adjust the operation sequence between multiple mining areas, and realize closed-loop control from environmental perception to risk warning and then to equipment regulation.

[0016] A system for dynamic simulation and generation of the physical environment of unmanned mining in multi-area clusters, the system comprising: An integrated sensing network deployment unit is used to deploy a multi-type fiber optic grating sensor integrated sensing network within the target mining area. The three-dimensional spatial environment model generation unit uses a three-dimensional laser scanner to scan the target mining area and generate a three-dimensional spatial environment model of the mining area. The parameter acquisition and assimilation unit connects multiple types of fiber optic grating sensors to the physical environment sensing device, and uses the physical environment sensing device to acquire and assimilate the physical environment parameters of the target sampling site in real time. Database construction unit, constructing a multi-condition numerical simulation training database; The environmental simulation model construction and training unit is used to construct and train a multi-sampling area physical environment simulation model based on a BP neural network. The environmental dynamic simulation unit uses a trained multi-mining zone physical environment simulation model based on BP neural network to dynamically simulate the physical environment of the multi-mining zone cluster and generate an unmanned mining environment for the multi-mining zone. The feedback control unit performs feedback control of the unmanned equipment operation environment based on the unmanned mining environment in multiple mining areas.

[0017] A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, performs the steps of the method for dynamic simulation and generation of the physical environment of unmanned mining in multi-mining area clusters.

[0018] The present invention provides a method and system for dynamic simulation and generation of the physical environment of unmanned mining in multi-mining areas. It proposes and implements for the first time a four-level data fusion and intelligent simulation architecture: "integrated sensing of multiple types of fiber optic grating sensors → numerical simulation error correction → BP neural network multi-condition training → multi-mining area cluster simulation." Unlike existing technologies that only focus on single-mining area point monitoring or offline numerical simulation, this application achieves a complete technical closed loop from real-time sensing to multi-mining area simulation. It constructs a BP neural network simulation model that integrates multiple working condition variables (cross-sectional dimensions, mining depth, mining progress, measuring point locations, etc.), with a training sample size of 360 groups. The model's prediction accuracy R-value reaches 0.997, and the average prediction error is only... The accuracy is 4.5%, far exceeding existing numerical simulation methods, and the calculation time is reduced from tens of hours in numerical simulation to milliseconds. A point cloud processing method for the mining area spatial environment based on the MCMD-Z automatic noise reduction algorithm is proposed. Combined with high-precision fiber optic grating physical sensing data, it achieves the fusion generation of a two-dimensional environment of "spatial geometry + physical field," providing a complete description of the operating environment for unmanned equipment. A real-time feedback closed loop from environmental perception to equipment control is established, automatically adjusting the unmanned equipment operation strategy based on the physical environment simulation results, enabling the system to have proactive safety protection capabilities. Unlike the existing architecture where environmental monitoring and equipment control are independent, this application achieves deep integration of the two. Compared with existing technologies, this application has the following beneficial effects: The inference accuracy is significantly improved: the average error between the prediction results of the BP neural network inference model and the measured data is only 4.5%, and the model regression coefficient R value reaches 0.997. Compared with the prediction error of more than 30% by traditional numerical simulation methods, the accuracy is improved by nearly an order of magnitude.

[0019] The simulation speed has been greatly improved: the numerical simulation calculation time for a single mining area is about tens of hours, while the BP neural network model can complete a simulation in just milliseconds, which is more than six orders of magnitude faster, making it possible to perform real-time simulations for multi-mining area clusters.

[0020] Multi-mining area collaborative simulation capability: The model can simulate the physical environment evolution of multiple mining areas in parallel, achieving for the first time a leap from single mining area monitoring to multi-mining area cluster simulation. It can analyze the superposition effect of mining-induced stress and ensure the safety of multi-mining area collaborative mining.

[0021] Proactive safety control closed loop: By feeding the simulation results back to the unmanned equipment control system in real time, a proactive safety control closed loop of "environmental perception → risk warning → equipment control" is realized, transforming passive alarm into active control and greatly improving the operational safety level of unmanned mining.

[0022] High system integration: The field physical environment sensing device integrates data acquisition, storage and transmission functions of multi-channel and multi-type fiber optic grating sensors. It has a compact size (280mm×185mm×52mm) and is easy to deploy in the narrow space underground. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0024] Figure 1 This is a flowchart of a method for dynamic simulation and generation of the physical environment of unmanned mining in multi-mining area clusters, according to a specific embodiment of the present invention. Figure 2 This is a schematic diagram of the deployment of an integrated sensing network for multiple types of fiber Bragg grating sensors in a specific embodiment of the present invention; Figure 3 This is a structural diagram of a mining site physical environment sensing device in a specific embodiment of the present invention; Figure 4 This is a flowchart of three-dimensional laser scanning point cloud data acquisition and processing in a specific embodiment of the present invention; Figure 5 This is a structural diagram of a BP neural network inference model in a specific embodiment of the present invention; Figure 6 This is a schematic diagram illustrating the physical environment simulation of a multi-sampling area cluster in a specific embodiment of the present invention; Figure 7 This is a curve comparing the predicted values ​​and measured values ​​of the inference model in a specific embodiment of the present invention; Figure 8 This is a schematic diagram of the structure of the dynamic simulation and generation system of physical environment for unmanned mining in a multi-mining area cluster, according to a specific embodiment of the present invention. Detailed Implementation

[0025] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0026] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0027] like Figure 1 As shown, Figure 1 This is a flowchart of a method for dynamically extrapolating and generating the physical environment of unmanned mining in multi-mining area clusters, according to a specific embodiment of the present invention. The method includes: Step 1: Construct an integrated sensing network of multiple types of fiber Bragg grating sensors.

[0028] In key locations such as the longwall roadways, crossroads, and roof of the target mining area, a multi-type fiber optic grating sensor integrated sensing network is deployed, including: Fiber Bragg grating displacement gauge: Used to measure displacement and settlement in multiple locations within deep rock layers of a mining area. Its measurement principle is as follows: An external displacement is sensed by a measuring probe, converted into the deflection of a beam of equal strength by a wedge, and then two fiber gratings attached to the upper and lower surfaces of the beam sense the strain change. The relationship between displacement D and wavelength change is: D=[(λs-λ0)-Kt(Ts-T0)] / Kp; Where λs is the measurement wavelength, λ0 is the initial wavelength, Kt is the temperature coefficient, Kp is the displacement coefficient, and Ts and T0 are the measurement temperature and the initial temperature, respectively. Fiber Bragg grating embedded strain gauge: Employing a narrow-necked tube protective enclosure structure, it consists of a protective steel tube, transmission optical fiber, fiber Bragg grating, sleeve, and bonding material. It is used to monitor strain field changes within the surrounding rock of a mining area. The relationship between strain S and wavelength change is: S=[(λs-λ0)-Kt(Ts-T0)] / Kp; Fiber Bragg grating piezometer: This type of gauge uses a diaphragm structure for encapsulation. External water seepage pressure is transmitted to the diaphragm through a permeable rock, causing the fiber grating to stretch or contract. It is used to monitor changes in the seepage field within the surrounding rock of a mining area. The relationship between seepage pressure P and wavelength change is as follows: P=[(λs-λ0)-Kt(Ts-T0)] / Kp; The aforementioned sensors are connected in series via optical fibers to the field physical environment sensing device.

[0029] like Figure 2 The diagram shows the deployment of a multi-type fiber Bragg grating sensor integrated sensing network. Two sensor data acquisition sections, DM1 and DM2, are arranged in the -645m mid-section of the mining roadway. Each section has one set of fiber Bragg grating displacement gauges, one set of fiber Bragg grating embedded strain gauges, and one set of fiber Bragg grating piezometers installed on the roof. These sensors are connected in series via optical fiber to a physical environment sensing device. This device has eight measurement channels, a measurement frequency of 100Hz, and a measurement wavelength range of 1528nm~1568nm. The acquired data is transmitted in real-time to the control center via an Ethernet (TCP / IP) interface. This device is located in a chamber in the roadway outside the mining area.

[0030] like Figure 3 The diagram shows the structure of the physical environment sensing device used in the mining area. The device measures 280mm × 185mm × 52mm and integrates a fiber Bragg grating demodulator, a data acquisition module, an embedded processor, a network communication module, and a large-capacity data storage module. The fiber Bragg grating demodulator has eight independent fiber optic measurement channels, each capable of supporting up to 17 sensors. The measurement wavelength range is 1528nm~1568nm, with an accuracy of 2~5pm, repeatability better than ±2pm, a dynamic range of 18dB, and a wavelength resolution of 1pm. The data acquisition module uses a 16-bit ADC. The embedded processor is an ARM Cortex-A9 dual-core processor equipped with 512MB of DDR3 memory. The large-capacity data storage module has a capacity of 256GB SSD. The device transmits the acquired data to the control center in real time via an Ethernet (TCP / IP) interface. The operating temperature is 0~50℃, the power consumption is 10W, and the power supply is AC220V.

[0031] Step 2: Real-time modeling of the 3D laser scanning mining site environment.

[0032] A 3D laser scanner was used to scan the roadways and space within the mining area to acquire high-density 3D point cloud data. The specific parameters of the 3D laser scanner were: distance accuracy 1.2mm±10ppm, point accuracy 3mm@50m / 6mm@100m, wavelength 1550nm (invisible), and scanning rate up to 1,000,000 points / second.

[0033] like Figure 4 The diagram shown is a flowchart of 3D laser scanning point cloud data acquisition and processing. The process includes the following steps: (21) Site layout: The main site is set at the junction of the cross-vein roadway and the segmented horizontal roadway, and the auxiliary site is set at the junction of the cross-vein roadway and the mining roadway. Supplementary sites are set every 5m to cover the area blocked by protruding rock blocks.

[0034] (22) Point cloud data acquisition: The target-based stitching method is adopted, and there are at least 3 reflective targets between adjacent scanning stations. The targets are evenly distributed to cover the scanning object for fine scanning.

[0035] (23) Image data acquisition: Use the built-in camera or an external digital camera to take pictures of the scanning environment and obtain color information for matching with point cloud data.

[0036] (24) Point cloud data processing: The MCMD-Z automatic denoising algorithm is used to denoise the original point cloud data. The algorithm extracts the maximum coordination set fitting plane from the neighborhood point set based on the h-MCS method, calculates the orthogonal distance OD from each point to the fitting plane, and calculates the Rz-score value: Rz_j=|OD_j-median(OD_m)| / MAD(OD); In the formula, OD_j represents the orthogonal distance from the j-th point to be judged in the point cloud data to the reference plane fitted by the h-MCS method. This orthogonal distance refers to the perpendicular distance from the point to be judged along the normal direction of the reference plane to the reference plane, and the unit is the length unit in the scanner coordinate system (e.g., mm). The larger the absolute value of OD_j, the farther the point deviates from the local reference plane, and the more likely it is to be a noise point or an anomaly.

[0037] `median(OD_m)`: Represents the median of the orthogonal distances from all `m` points in the neighborhood of point `p_i` to the fitted plane. Specifically, `m` is the total number of points in the neighborhood of `p_i` (i.e., `k` neighborhood points), and the index `m` in `OD_m` iterates through all points in the neighborhood (m = 1, 2, ..., k). `median(OD_m)` takes the middle value after arranging these `k` orthogonal distance values ​​in ascending order (the exact middle value when `k` is odd, and the average of the two middle values ​​when `k` is even). The median is a robust location parameter in statistics. Compared to the arithmetic mean, the median is not affected by extreme values ​​and more accurately reflects the "normal" distance level of most points in the neighborhood to the reference plane.

[0038] MAD(OD): Represents the median absolute deviation (MAD) of the orthogonal distances from all points in the neighborhood set to the fitted plane, a robust measure of dispersion. The formula is: MAD(OD) = 1.483 × median(|OD_j - median(OD_m)|), where 1.483 is a consistency scaling factor, ensuring that MAD(OD) aligns with the standard deviation σ when the data follows a normal distribution. The specific calculation process is as follows: First, calculate the absolute value of the difference between the OD value of each point in the neighborhood and the median of the neighborhood. Then, take the median of these absolute differences and multiply it by the scaling factor 1.483. A larger MAD(OD) value indicates greater fluctuation in the distance from each point in the neighborhood to the reference plane, suggesting higher surface roughness or more noise in the area.

[0039] Rz_j: Represents the robust Z-score of the j-th point to be judged in the point cloud data. Its calculation formula is Rz_j = |OD_j - median(OD_m)| / MAD(OD). The numerator |OD_j - median(OD_m)| represents the absolute value of the deviation between the orthogonal distance of the point to be judged and the median of its neighborhood, while the denominator MAD(OD) represents the dispersion of the point set within the neighborhood. Rz_j is a dimensionless statistic used to measure the degree of deviation of the point to be judged relative to the overall level of its neighborhood. When Rz_j < 2.5, it indicates that the deviation of the point is within the normal statistical fluctuation range and is judged as a normal point and retained; when Rz_j ≥ 2.5, it indicates that the deviation of the point exceeds 2.5 times the normal statistical fluctuation range and is judged as an outlier (noise point) and removed. The selection of the threshold 2.5 is based on the normal distribution assumption in statistics—under a normal distribution, approximately 98.76% of the data points fall within ±2.5 standard deviations (MAD).

[0040] After noise reduction, point cloud stitching, coordinate transformation, and texture mapping are performed to generate a three-dimensional spatial environment model of the mining site.

[0041] Step 3: Physical environment data collection and data assimilation in the single sampling area.

[0042] After the sensors are installed, the physical environment sensing device is used to collect the physical environment parameters (displacement, strain, and seepage pressure) of the stope in real time. At the same time, based on the geological conditions of the stope and the mining plan, a FLAC3D three-dimensional numerical calculation model is established to calculate the stress field and displacement field variation laws of the stope under different mining progress.

[0043] The numerical simulation results are compared with the actual monitoring data to calculate the systematic error E of the numerical simulation: ; in, For the first The stress values ​​obtained from numerical simulation at each monitoring point For the first The actual stress values ​​obtained from monitoring at each monitoring point The total number of monitoring points is given. Error analysis revealed that the numerical simulation results were on average 36.7% (stress) and 37.8% (displacement) higher than expected. Based on this, a systematic reduction correction was performed on the numerical simulation results to achieve initial data assimilation between the monitoring data and the numerical model.

[0044] Step 4: Construct a multi-condition numerical simulation training database.

[0045] Based on the corrected numerical model, a multi-condition extended simulation of the mining area is performed to generate a training database. The condition variables include: The cross-sectional dimensions of the mining roadway are available in two sizes: 4m×4m and 3m×3m. Mining depths: three intermediate sections: -555m, -645m, and -735m; Mining progress: three steps: 4m, 8m, and 12m; Distance to the working surface: 1m~10m; Measurement point height: 2m (side wall) and 4m (top plate); Roadway locations: Upper mining roadway (code 0) and lower mining roadway (code 1).

[0046] Numerical calculations were performed for each working condition to extract stress and displacement data at each monitoring point. The data were then reduced according to the error coefficient obtained in step 3 to form a training database containing 360 sets of sample data.

[0047] Before training, the data is preprocessed by normalization to eliminate dimensional differences between different physical quantities. Pn=(P-Pmin) / (Pmax-Pmin), Pn∈[0,1]; In the formula, P represents the original value of a certain working condition variable in the multi-working-condition numerical simulation training database. P can be any physical quantity or geometric parameter to be normalized in the database, such as: distance to the working face (in meters), roadway cross-sectional area (in m²), mining progress (in meters), height of the measuring point (in meters), mining section (in meters), surrounding rock stress (in MPa), or surrounding rock displacement (in mm). The value of P varies with different working conditions and different monitoring points.

[0048] Pmin: Represents the minimum value of a specific variable under all working conditions in the multi-condition numerical simulation training database. For example, when P represents the distance to the working face, Pmin is the minimum value (1m) of that variable in all 360 sets of sample data; when P represents the surrounding rock stress, Pmin is the minimum stress value of that variable in all 360 sets of sample data. Pmin is obtained by statistical calculation through traversing the corresponding variable columns of the entire database and taking the minimum value.

[0049] Pmax: Represents the maximum value of a specific variable under all working conditions in the multi-condition numerical simulation training database. For example, when P represents the distance to the working face, Pmax is the maximum value (10m) of that variable in all 360 sets of sample data; when P represents the surrounding rock stress, Pmax is the maximum stress value of that variable in all 360 sets of sample data. Pmax is obtained by statistically calculating the maximum value by traversing the corresponding variable columns throughout the entire database.

[0050] Pn: Represents the dimensionless standardized value after min-max normalization. The value of Pn is in the closed interval [0, 1]. When P is Pmin, Pn = 0; when P is Pmax, Pn = 1; when P is between Pmin and Pmax, Pn is linearly mapped to the corresponding value in the interval (0, 1).

[0051] The purpose of normalization preprocessing is as follows: In the multi-condition numerical simulation training database, the dimensions and numerical ranges of variables for different conditions vary greatly. For example, the distance to the working face ranges from 1 to 10 m, the cross-sectional area of ​​the roadway ranges from 9 to 16 m², and the stress of the surrounding rock can range from several MPa to tens of MPa. If normalization is not performed and the raw data is directly input into the BP neural network for training, variables with large numerical ranges will dominate during error backpropagation, severely weakening the influence of variables with small numerical ranges on model training. By mapping all variables to a unified [0, 1] interval through min-max normalization, the adverse effects caused by differences in dimensions and numerical ranges among variables can be eliminated, allowing the BP neural network to give equal attention to each input variable during training, thereby accelerating model convergence and improving prediction accuracy.

[0052] Step 5: Construction and training of a multi-sampling area physical environment inference model based on BP neural network.

[0053] like Figure 5 The diagram shown is a structural diagram of the constructed BP neural network derivation model. The network structure includes: Input layer (6 nodes): distance to working face (d), roadway cross-sectional area (S), mining progress (j), height of measuring point (h), roadway location (0 for top, 1 for bottom), and mining section (m). The input value of each node is the normalized value of the corresponding working condition variable, with a value range of [0, 1].

[0054] Hidden Layers: A two-layer hidden layer structure is adopted, with 12 nodes in the first hidden layer and 8 nodes in the second hidden layer. The inter-layer transfer function adopts a sigmoid function. ; x represents the weighted sum of the outputs of all neurons in the previous layer received by the current neuron, and its mathematical expression is: ; Where m is the number of neurons in the previous layer; This represents the connection weight between the k-th neuron in the previous layer and the current neuron. The initial value is randomly assigned and continuously adjusted during training using the backpropagation algorithm. is the output value of the k-th neuron in the previous layer; b is the bias term of the current neuron, initially assigned a random value, and continuously adjusted during training through the backpropagation algorithm.

[0055] Output layer (2 nodes): surrounding rock stress (σ) and surrounding rock displacement (c).

[0056] The learning efficiency (lr) was set to 0.01, the target accuracy (goal) to 0.001, and the maximum number of training iterations to 2000. 300 datasets from the database were used as the training set, and 60 datasets as the test set to train the model. Backpropagation of the error function was used during training. E1=Σ(ti-Oi)² / 2; E1 - Backpropagation error function value, representing the total error between the network's current output and the expected output; ti - the expected output value of the i-th node of the output layer, that is, the measured surrounding rock stress value or measured surrounding rock displacement value corresponding to the multi-condition numerical simulation training database described in step 4; Oi - The actual calculated output value of the network at the i-th node of the output layer, i.e., the surrounding rock stress or displacement value predicted by the multi-mining area physical environment inference model based on BP neural network under the current network weights and thresholds. The ∑ symbol sums up all nodes (i=1,2) in the output layer, where i=1 corresponds to the surrounding rock stress node and i=2 corresponds to the surrounding rock displacement node. During the training process, the connection weights and thresholds between nodes in each layer are continuously adjusted through the backpropagation algorithm, so that the value of the backpropagation error function E1 gradually decreases until the preset target accuracy is reached.

[0057] Step 6: Dynamic simulation and generation of the physical environment of the multi-sampling area cluster.

[0058] like Figure 6 The diagram shows a simulation of the physical environment of a multi-mining zone cluster. It illustrates the spatial arrangement of four mining blocks (blocks 1, 2, 3, and 4) and the data flow process for the physical environment simulation: real-time monitoring data (including displacement, strain, and seepage pressure) collected at the monitoring point in block 1 is used as input and transmitted via fiber optic cable to the physical environment sensing device, then through the mine's existing network to the control center. The control center inputs the real-time monitoring data into a trained BP neural network simulation model. The model then performs parallel simulations of the stress and displacement evolution trends of each mining area at different mining stages, analyzes the superposition effect of mining-induced stress, and provides a decision-making basis for the safe and coordinated mining of the multi-mining zone cluster.

[0059] The trained BP neural network model is used for dynamic simulation of the physical environment of a multi-sampling cluster: (61) For the currently mining single mining area, the network input parameters (mining progress j, distance to working face d, etc.) are updated according to real-time monitoring data. The model outputs the predicted values ​​of stress and displacement at the current location in real time. The average error between the predicted value and the measured result is controlled within 4.5%. The regression coefficient R value of the model output data reaches 0.997.

[0060] (62) For adjacent mining areas that have not yet been mined, the corresponding network parameters are input according to the planned mining sequence and mining area structure parameters, and the model can quickly deduce the stress and displacement evolution trend of the mining area at different mining stages.

[0061] (63) For scenarios where multiple mining areas operate simultaneously, the model can extrapolate the changes in the physical environment of each mining area and analyze the superposition effect of mining stress, providing a basis for decision-making for safe and coordinated mining of multiple mining areas.

[0062] (64) The physical environment data generated by the simulation is integrated with the three-dimensional spatial environment model of the mining area generated in step 2 to generate a complete unmanned mining environment for multiple mining areas, including four dimensions: spatial geometric information, stress field distribution, displacement field distribution and seepage field distribution.

[0063] like Figure 7The figure shows a curve comparing the predicted values ​​of the model with the measured values. The X-axis represents the measurement point number, and the Y-axis represents the displacement value (unit: mm). The solid line represents the measured value curve, which is actually collected from each monitoring point in the integrated sensing network of the multi-type fiber optic grating sensors; the dashed line represents the predicted value curve, which is derived from the trained BP neural network model for each monitoring point. The figure indicates that the average error between the predicted and measured values ​​is 4.5%, and the regression coefficient R-value of the model output data is 0.997, indicating a high degree of consistency between the model prediction results and the measured data, verifying the accuracy and reliability of the method in the simulation of the physical environment of the mining area.

[0064] Step 7, feedback control of unmanned equipment operating environment.

[0065] The multi-area unmanned mining environment data generated in step 6 is transmitted to the unmanned equipment centralized control system in real time. (71) Based on the current physical environment of the mining area (stress level, displacement rate, etc.), automatically determine the safety level of each working area and divide it into "safe working area", "early warning working area" and "prohibited working area".

[0066] (72) When monitoring or simulation data shows that the stress value in a certain area is close to or exceeds the preset threshold, an early warning signal is automatically sent to the unmanned equipment control system to adjust the equipment operation path or suspend operation.

[0067] (73) Feedback the trend of physical environment changes to the unmanned equipment scheduling system, dynamically adjust the operation sequence between multiple mining areas, and realize closed-loop control of "environmental perception → risk warning → equipment regulation".

[0068] This invention integrates real-time sensing of multiple types of fiber optic grating sensors, spatial modeling of three-dimensional laser scanning, and multi-mining area simulation model of BP neural network to achieve a closed-loop chain of "real-time sensing → data assimilation → multi-mining area simulation → environment generation → equipment feedback", providing real-time, dynamic, and predictable physical environment description for unmanned mining in multi-mining area clusters. In a specific embodiment of the present invention, the physical environment of the multi-mining zone in the -645m section of the Sanshandao gold mine is simulated and generated, specifically including: (1) Sensor network deployment.

[0069] Two fiber Bragg grating (FBG) sensor data acquisition sections, DM1 and DM2, were installed in the -645m section of the mining roadway at the Sanshandao Gold Mine. Each section has one set of FBG displacement gauges, one FBG embedded strain gauge, and one FBG piezometer installed at the roof position. The sensors are connected to a physical environment sensing device via cables inside steel protective conduits. The device is placed in a chamber in the roadway outside the mining area, and the data is transmitted in real-time to the control center via the mine's existing network.

[0070] (2) Three-dimensional laser scanning space modeling.

[0071] The -645m mid-section cross-cut roadway, segmented horizontal roadway, and mining roadway were scanned using a Leica P40 3D laser scanner, with a total of 12 scanning stations set up. After processing with the MCMD-Z automatic noise reduction algorithm, high-density point cloud data was obtained. A 3D spatial environment model of the mining area was generated by point cloud stitching and texture mapping.

[0072] (3) Training of BP neural network inference model.

[0073] Based on the FLAC3D numerical model, extended simulations were performed for 18 working conditions, extracting 360 sets of stress and displacement data. These data were then reduced by a 37% error coefficient to construct a training database. The BP neural network achieved a target accuracy of 0.001 after 52 iterations of training, with convergence of the root mean square error. The trained model was used to predict the stress and displacement of the roof of the -645m mid-section mining roadway. Comparison with measured data showed that the maximum stress prediction error was 0.02 MPa, the maximum displacement prediction error was 0.09 mm, and the average error was 4.5%.

[0074] (4) Multi-sampling area simulation and verification.

[0075] The trained model was used to simulate the mining process of four blocks in adjacent mining areas (line 1460 to line 1520), predicting the stress evolution trajectory of each block under different mining steps. The simulation results accurately identified the stress concentration area (maximum stress up to 7.1 MPa) at the acute angle of the block floor, guiding the targeted optimization of the support scheme.

[0076] Accordingly, embodiments of the present invention also provide a system for dynamic simulation and generation of the physical environment of unmanned mining in multi-mining area clusters, such as... Figure 8 The diagram shown is a structural schematic of the system. This multi-area cluster unmanned mining physical environment dynamic simulation and generation system includes the following modules: The integrated sensing network deployment unit 801 deploys an integrated sensing network of multiple types of fiber optic grating sensors within the target mining area. The three-dimensional spatial environment model generation unit 802 uses a three-dimensional laser scanner to scan the target mining area and generate a three-dimensional spatial environment model of the mining area. The parameter acquisition and assimilation unit 803 connects multiple types of fiber optic grating sensors to the physical environment sensing device, and uses the physical environment sensing device to acquire and assimilate the physical environment parameters of the target sampling site in real time. Database construction unit 804, constructs a multi-condition numerical simulation training database; The environmental simulation model construction and training unit 805 is used to construct and train a multi-sampling area physical environment simulation model based on a BP neural network. The environmental dynamic simulation unit 806 uses a trained multi-mining area physical environment simulation model based on BP neural network to perform dynamic simulation of the physical environment of the multi-mining area cluster and generate the multi-mining area unmanned mining environment. The feedback control unit 807 performs feedback control of the unmanned equipment operation environment based on the unmanned mining environment in the multi-mining area.

[0077] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0078] The present invention also provides a storage medium, which is a computer-readable storage medium storing a computer program thereon, the computer program being executable when it is run. Figure 1 The method shown may include some or all of the steps. The storage medium may include read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, etc. The storage medium may also include non-volatile memory or non-transitory memory, etc.

[0079] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data provider to another website, computer, server, or data provider via wired or wireless means.

Claims

1. A method for dynamic simulation and generation of the physical environment of unmanned mining in multi-area clusters, characterized in that, The method includes: Step 1: Deploy an integrated sensing network of multiple types of fiber Bragg grating sensors within the target mining area; Step 2: Use a 3D laser scanner to scan the target mining area and generate a 3D spatial environment model of the mining area; Step 3: Connect multiple types of fiber Bragg grating sensors to the physical environment sensing device, and use the physical environment sensing device to collect and assimilate the physical environment parameters of the target mining site in real time. Step 4: Construct a multi-condition numerical simulation training database; Step 5: Construct and train a multi-sampling area physical environment simulation model based on a BP neural network; Step 6: Using the trained multi-mining zone physical environment simulation model based on BP neural network, perform dynamic simulation of the physical environment of the multi-mining zone cluster to generate the multi-mining zone unmanned mining environment. Step 7: Based on the unmanned mining environment in the multi-mining area, implement feedback control of the unmanned equipment operation environment; In step 5, the constructed multi-mining area physical environment simulation model based on BP neural network includes an input layer, a hidden layer, and an output layer. The input layer includes 6 nodes, which correspond to 6 working condition variables in the multi-working condition numerical simulation training database: the cross-sectional size of the mining roadway, the mining middle section, the mining progress, the distance to the working face, the height of the measuring point, and the roadway location. The distance to the working face and the mining progress are dynamically updated by real-time monitoring data, while the cross-sectional size of the mining roadway, the height of the measuring point, the roadway location, and the mining middle section are determined by the working condition variables in step 4. The hidden layer adopts a two-layer hidden layer structure, with 12 nodes in the first hidden layer and 8 nodes in the second hidden layer. The interlayer transfer function adopts an S-shaped function. The output layer includes two nodes, namely the surrounding rock stress and the surrounding rock displacement. When training the multi-sampling area physical environment simulation model based on BP neural network, the learning efficiency, target accuracy and maximum number of training times are set. Multiple sets of data in the multi-condition numerical simulation training database are used as the training set and multiple sets of data are used as the test set to train the model. During the training process, the error function is backpropagated and the error function is made to reach the expected value by continuously adjusting the threshold and network weights. Step 6 includes: Step 61: For the currently being mined single mining area, update the input parameters of the multi-mining area physical environment inference model based on BP neural network according to real-time monitoring data. The multi-mining area physical environment inference model based on BP neural network outputs the stress and displacement prediction values ​​of the current location in real time. Step 62: For adjacent mining areas that have not yet been mined, according to the planned mining sequence and mining area structure parameters, input the corresponding network parameters into the multi-mining area physical environment inference model based on BP neural network. The multi-mining area physical environment inference model based on BP neural network infers the stress and displacement evolution trend of the mining area at different mining stages. Step 63: For scenarios where multiple mining areas operate simultaneously, the physical environment simulation model based on the BP neural network simulates the physical environment changes of each mining area in parallel, analyzes the superposition effect of mining stress, and generates physical environment data. Step 64: The physical environment data generated by the simulation is fused with the three-dimensional spatial environment model of the mining area to generate a complete unmanned mining environment for multiple mining areas, including four dimensions: spatial geometric information, stress field distribution, displacement field distribution, and seepage field distribution.

2. The method for dynamic simulation and generation of the physical environment of multi-mining area cluster unmanned mining according to claim 1, characterized in that, In step 1, a multi-type fiber optic grating sensor integrated sensing network is deployed at key locations such as the mining roadways, cross-cutting roadways, and mining roof of the target mining area. The multi-type fiber optic grating sensors include: fiber optic grating displacement gauges, fiber optic grating embedded strain gauges, and fiber optic grating piezometers.

3. The method for dynamic simulation and generation of the physical environment of multi-mining area cluster unmanned mining according to claim 1, characterized in that, Step 2 includes: Step 21: Set up a main station at the junction of the cross-cut roadway and the segmented horizontal roadway in the target mining area, set up an auxiliary station at the junction of the cross-cut roadway and the return roadway, and set up supplementary stations at regular intervals to cover the area blocked by protruding rock blocks. Step 22: Using a target-based stitching method, at least three reflective targets are ensured between adjacent scanning stations. The targets are evenly distributed to cover the scanning object. A 3D laser scanner is used for fine scanning to obtain the original point cloud data. Step 23: Take pictures of the scanning environment using the built-in camera or an external digital camera to obtain color information, which is used to match the original point cloud data and give the original point cloud data realistic color and texture information. Step 24: The original point cloud data is denoised using the MCMD-Z automatic denoising algorithm to obtain high-density three-dimensional point cloud data. The high-density three-dimensional point cloud data is then stitched together, coordinate transformed, and texture mapped to generate the three-dimensional spatial environment model of the mining site.

4. The method for dynamic simulation and generation of the physical environment of multi-mining area cluster unmanned mining according to claim 1, characterized in that, In step 3, based on the geological conditions and mining plan of the target mining area, a FLAC3D three-dimensional numerical calculation model is established to calculate the stress and displacement field variations of the target mining area under different mining progresses, obtaining numerical simulation results. The numerical simulation results are then compared with actual monitoring data collected by a multi-type fiber optic grating sensor integrated sensing network to calculate the systematic error E of the numerical simulation. ； in, For the first The stress values ​​obtained from numerical simulation at each monitoring point For the first The actual stress values ​​obtained from monitoring at each monitoring point The total number of monitoring points is used; the numerical simulation results are systematically reduced and corrected according to the calculated systematic error, so that the corrected numerical simulation results are consistent with the actual monitoring data in a statistical sense, thus completing the initial data assimilation of the actual monitoring data with the FLAC3D three-dimensional numerical calculation model.

5. The method for dynamic simulation and generation of the physical environment of multi-mining area cluster unmanned mining according to claim 4, characterized in that, In step 4, based on the corrected FLAC3D three-dimensional numerical calculation model, a multi-condition extended simulation is performed on the target mining area to generate the multi-condition numerical simulation training database. The condition variables include: cross-sectional dimensions of the mining roadway, mining depth, mining progress, distance to the working face, measuring point height, and roadway location. Numerical calculations are performed for each condition, and stress and displacement data at each monitoring point in the multi-type fiber optic grating sensor integrated sensing network are extracted. The obtained error coefficients are systematically reduced and corrected to form the multi-condition numerical simulation training database containing multiple sets of sample data.

6. The method for dynamic simulation and generation of the physical environment of multi-mining area cluster unmanned mining according to claim 1, characterized in that, Step 7 includes: Step 71: Based on the current physical environment of the mining area, determine the safety level of each work area and divide it into safe work areas, warning work areas, and prohibited work areas; Step 72: When monitoring or simulation data shows that the stress value in a certain area is close to or exceeds the preset threshold, a warning signal is sent to the unmanned equipment control system to adjust the equipment operation path or suspend operation. Step 73: Feedback the trend of physical environment changes to the unmanned equipment scheduling system, dynamically adjust the operation sequence between multiple mining areas, and realize closed-loop control from environmental perception to risk warning and then to equipment regulation.

7. A system for dynamic simulation and generation of the physical environment of unmanned mining in multi-area clusters, characterized in that, The system includes: An integrated sensing network deployment unit is used to deploy a multi-type fiber optic grating sensor integrated sensing network within the target mining area. The three-dimensional spatial environment model generation unit uses a three-dimensional laser scanner to scan the target mining area and generate a three-dimensional spatial environment model of the mining area. The parameter acquisition and assimilation unit connects multiple types of fiber optic grating sensors to the physical environment sensing device, and uses the physical environment sensing device to acquire and assimilate the physical environment parameters of the target sampling site in real time. Database construction unit, constructing a multi-condition numerical simulation training database; The environmental simulation model construction and training unit is used to construct and train a multi-sampling area physical environment simulation model based on a BP neural network. The environmental dynamic simulation unit uses a trained multi-mining zone physical environment simulation model based on BP neural network to dynamically simulate the physical environment of the multi-mining zone cluster and generate an unmanned mining environment for the multi-mining zone. The feedback control unit performs feedback control of the unmanned equipment operation environment based on the unmanned mining environment in multiple mining areas; The environmental simulation model construction and training unit constructs a multi-mining area physical environment simulation model based on a BP neural network, which includes an input layer, a hidden layer, and an output layer. The input layer includes six nodes, each corresponding to one of the six working condition variables in the multi-working condition numerical simulation training database: the cross-sectional size of the mining roadway, the mining middle section, the mining progress, the distance to the working face, the height of the measuring point, and the roadway location. The distance to the working face and the mining progress are dynamically updated by real-time monitoring data, while the cross-sectional size of the mining roadway, the height of the measuring point, the roadway location, and the mining middle section are determined by the working condition variables in step 4. The hidden layer adopts a two-layer hidden layer structure, with 12 nodes in the first hidden layer and 8 nodes in the second hidden layer. The interlayer transfer function adopts an S-shaped function. The output layer includes two nodes, namely the surrounding rock stress and the surrounding rock displacement. When training the multi-sampling area physical environment simulation model based on BP neural network, the learning efficiency, target accuracy and maximum number of training times are set. Multiple sets of data in the multi-condition numerical simulation training database are used as the training set and multiple sets of data are used as the test set to train the model. During the training process, the error function is backpropagated and the error function is made to reach the expected value by continuously adjusting the threshold and network weights. The environmental dynamic simulation unit, when using a trained multi-mining area physical environment simulation model based on a BP neural network to dynamically simulate the physical environment of a multi-mining area cluster and generate an unmanned mining environment for the multi-mining area, specifically includes: For the currently operating single mining area, the input parameters of the multi-mining area physical environment inference model based on BP neural network are updated according to real-time monitoring data. The multi-mining area physical environment inference model based on BP neural network outputs the predicted stress and displacement values ​​at the current location in real time. For adjacent mining areas that have not yet been mined, according to the planned mining sequence and mining area structure parameters, the corresponding network parameters are input into the multi-mining area physical environment inference model based on BP neural network. The multi-mining area physical environment inference model based on BP neural network infers the stress and displacement evolution trend of the mining area at different mining stages. For scenarios involving simultaneous operation in multiple mining areas, a multi-mining area physical environment simulation model based on BP neural network is used to simulate the changes in the physical environment of each mining area in parallel, analyze the superposition effect of mining stress, and generate physical environment data. The physical environment data generated by the simulation is integrated with the three-dimensional spatial environment model of the mining area to generate a complete unmanned mining environment for multiple mining areas, including four dimensions: spatial geometric information, stress field distribution, displacement field distribution, and seepage field distribution.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by the processor, it executes the steps of the method for dynamic simulation and generation of the physical environment of unmanned mining in multi-mining area clusters as described in any one of claims 1 to 6.

Images