A method and system for monitoring and early warning of rockburst based on multi-source information fusion

Abstract

A method and system for monitoring and early warning of rockbursts based on multi-source information fusion is disclosed. The method includes: collecting multi-dimensional data from the target mining area; integrating and processing the multi-dimensional data using a Kalman filter fusion algorithm; generating a three-dimensional geological structure model of the target mining area using a geological structure inversion method; dividing the three-dimensional geological structure model into regions; predicting the rockburst risk level of each region using a rockburst risk prediction model; and issuing graded early warnings. The system comprises: a data acquisition module for collecting multi-dimensional data from the target mining area; a data fusion module for integrating and processing the data; a model building module for generating a three-dimensional geological structure model; a region division module for dividing the three-dimensional geological structure model into regions; a risk prediction module for risk prediction; a graded early warning module for graded early warning; and an alarm system for executing graded early warning actions. This method and system can achieve accurate monitoring and intelligent early warning of rockbursts.

Description

Technical Field

[0001] This invention belongs to the field of intelligent monitoring technology for rockbursts in coal mines, specifically a rockburst monitoring and early warning method and system based on multi-source information fusion. Background Technology

[0002] Rockbursts are a common and serious geological hazard during mining operations. They are characterized by their suddenness and destructive power, posing a serious threat to the lives of underground workers and the normal operation of mining. As shallow mineral resources become increasingly depleted, mining operations are gradually extending to deeper areas. At the same time, factors such as high ground stress and complex geological structures have significantly increased the frequency and intensity of rockbursts, posing a tremendous challenge to safety management in mining areas.

[0003] Currently, existing methods for monitoring and early warning of rockbursts have many limitations. Traditional monitoring technologies often rely on single types of sensors, such as stress sensors or microseismic sensors, which can only acquire local physical parameter changes in the mining area and cannot comprehensively reflect the geological conditions and stress distribution of the entire mining area. Furthermore, due to the single data source, the monitoring results are often one-sided, easily leading to misjudgments or omissions, thus failing to provide reliable decision-making basis for rockburst prevention and control. To address this problem, there is an urgent need to provide a rockburst monitoring and early warning method and system based on multi-source information fusion. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention provides a method and system for monitoring and early warning of rockbursts based on multi-source information fusion. This method is simple to implement, has low implementation cost, and is highly intelligent. It can achieve efficient and accurate monitoring of rockbursts, effectively reduce the probability of rockburst occurrence, and ensure safe production operations in mining areas. The system has a simple structure, high level of intelligence, and reliable performance. It can achieve accurate monitoring and intelligent early warning of rockbursts, effectively ensuring safe production operations in coal mines.

[0005] To achieve the above objectives, the present invention provides a method for monitoring and early warning of rockbursts based on multi-source information fusion, comprising the following steps:

[0006] Step 1: Collect multi-dimensional data of the target mining area, including topographic parameters, geological structure data, and rock mass physical property information;

[0007] Step 2: Process the multi-dimensional data and use the Kalman filter fusion algorithm to integrate the multi-dimensional data to obtain fused data;

[0008] Step 3: Based on the fused data, use the geological structure inversion method to generate a three-dimensional geological structure model of the target mining area;

[0009] Step 4: Based on the uniformity of rock mass physical property information and the continuity of geological structure data in the 3D geological structure model, the 3D geological structure model is divided into regions to obtain N independent sub-regions; each independent sub-region includes background feature data, which includes terrain parameters, geological structure data and rock mass physical property information;

[0010] Step 5: Collect historical rockburst event data from mining areas, construct a rockburst risk prediction model based on a BP neural network, and optimize the hyperparameters of the rockburst risk prediction model using a particle swarm optimization algorithm. Train the rockburst risk prediction model based on the hyperparameters. Use independent sub-regions as input data, perform risk prediction using the rockburst risk prediction model, and output the rockburst risk level of each region in the target mining area. The rockburst risk level includes low, medium, and high levels.

[0011] Step Six: Classify the early warning level according to the risk level of rockburst and generate early warning reminder suggestions.

[0012] As a preferred embodiment, in step one, the topographic parameters include the surface elevation, slope, aspect, and surface relief of the mining area; the geological structure data include the location, strike, dip angle, fold morphology, and distribution range of faults; and the rock mass physical property information includes rock mass density, elastic modulus, Poisson's ratio, uniaxial compressive strength, and rock mass integrity coefficient.

[0013] As a preferred option, the process of obtaining the fused data in step two is as follows:

[0014] S21: Remove outliers from multidimensional data using the Raida criterion;

[0015] S22: The Kalman filter fusion algorithm is used to integrate the data to obtain fused data;

[0016] S22-1: Establish the system's state equations according to formula (1), and establish the system's observation equations according to formula (2);

[0017] (1);

[0018] (2);

[0019] In the formula, Let k be the system state at time k. Let be the state transition matrix, representing the system transition from ... The dynamic evolution relationship from time k to time k. for The system state at any given moment. for The system noise driving matrix at time t, for System noise at any given moment Let k be the observation vector at time k. Let be the observation matrix at time k. The observation noise at time k;

[0020] S22-2: State prediction; Establish the state prediction equation according to formula (3), so as to... The posterior estimate at time k is used to estimate the state at time k, thus obtaining the prior estimate at time k. ;

[0021] (3);

[0022] In the formula, This represents the optimal estimate of the system state vector at time k-1 after combining the observation information at time k-1;

[0023] S22-3: State Update; Establish the state update equation according to formula (4), use the measurement value at time k to correct the estimated value in the prediction stage, and obtain the posterior estimated value at time k. ;

[0024] (4);

[0025] In the formula, Indicates Kalman gain;

[0026] S22-4: Repeatedly execute S22-2 and S22-3 to continuously fuse the data and finally obtain the fused data.

[0027] In this technical solution, outliers are removed by the Raida criterion and data is integrated by the Kalman filter fusion algorithm, which effectively reduces data noise and errors, ensures the accuracy of the fused data, and lays a solid data foundation for subsequent model building and risk prediction.

[0028] As a preferred option, the process of generating a three-dimensional geological structure model of the target mining area using the geological structure inversion method in step three is as follows:

[0029] S31: Using fused data as constraints, a geological structure inversion method based on Bayesian theory is adopted, and an inversion objective function is constructed according to formula (5). ;

[0030] (5);

[0031] In the formula, m is the model parameter vector, d is the observation data vector, and G(m) is the forward modeling operator. Let C be the initial model parameter vector, and C be the covariance matrix of the model parameters;

[0032] S32: Solving the inversion objective function using the Markov chain Monte Carlo method Obtain the posterior probability distribution of the model parameters;

[0033] S33: Based on the maximum likelihood estimation results of the posterior probability distribution, construct a three-dimensional mesh model of the target mining area, assign the model parameters to the corresponding mesh cells, and generate a three-dimensional geological structure model.

[0034] This technical solution employs a Bayesian-based geological structure inversion method, using fused data as constraints. The objective function is solved using the Markov chain Monte Carlo method, which quantifies the uncertainty of model parameters and yields a posterior probability distribution of model parameters that more closely reflects real geological conditions. This technique significantly enhances the accuracy of the 3D geological model, enabling the constructed 3D geological structure model to clearly represent the geological structure and rock mass physical properties of the mining area, providing a precise spatial framework for subsequent regional delineation.

[0035] As a preferred option, the process of dividing the three-dimensional geological structure model into regions in step four is as follows:

[0036] S41: Using the elastic modulus and uniaxial compressive strength in the rock mass physical property information as the dividing index, calculate the rock mass physical property difference coefficient between any two points in the three-dimensional geological structure model according to formula (6). ;

[0037] (6);

[0038] In the formula, and The elastic modulus at two points are respectively. and The uniaxial compressive strengths at two points are respectively. and These are the maximum values ​​of elastic modulus and uniaxial compressive strength within the mining area, respectively.

[0039] S42: Set the difference coefficient threshold When the difference coefficient of rock mass physical properties between two points When two points are located in the same region, they are assigned to the same sub-region.

[0040] S43: Combining the distribution of faults and folds in geological structural data, areas separated by faults or with fold morphological differences exceeding a preset difference threshold are divided into different sub-regions.

[0041] S44: After the initial partitioning is completed using the region growing algorithm, the boundaries are smoothed to obtain N independent sub-regions.

[0042] In this technical solution, the regions are divided based on the uniformity of the physical properties of the rock mass and the continuity of the geological structure. By combining the calculation of the difference coefficient and the distribution of the geological structure, and through the regional growth algorithm and boundary smoothing, the resulting independent sub-regions are highly representative, ensuring the consistency of geological conditions in each region and providing clear targets for hierarchical monitoring.

[0043] Furthermore, in order to obtain a high-accuracy rockburst risk prediction model, in step five, the historical rockburst event data in the mining area includes background feature data of the event occurrence area and specific attributes of the rockburst event; the background feature data of the event occurrence area includes the topographic parameters, geological structure data and rock mass physical property information of the area; the specific attributes of the rockburst event include the spatiotemporal information of the event, the impact intensity and the rockburst risk probability value corresponding to the event, and the spatiotemporal information of the event includes the time and location coordinates of the rockburst occurrence.

[0044] As a preferred option, the process of constructing the rockburst risk prediction model based on a BP neural network in step five is as follows:

[0045] S151: Construct the basic BP neural network model, which includes an input layer, multiple hidden layers, and an output layer. The input layer is used to receive background feature data of independent sub-regions. Multiple hidden layers are used to perform nonlinear transformations and feature mappings on the data received by the input layer. The output layer is used to output the probability value of rockburst risk for independent sub-regions.

[0046] S152: Take the regional feature data corresponding to the event location in the historical mining area rockburst event data as input, take the rockburst risk probability value corresponding to the event as output, train the BP neural network basic model, and save the model parameters that meet the preset accuracy. After training, a rockburst risk prediction model based on BP neural network is obtained.

[0047] In this technical solution, a risk prediction model based on a BP neural network is constructed and trained using historical rockburst event data. This enables the model to effectively capture the complex nonlinear relationships of rockburst occurrence and ensure prediction accuracy.

[0048] As a preferred option, in step five, the process of optimizing the hyperparameters of the rockburst risk prediction model using the particle swarm optimization algorithm is as follows:

[0049] S251: The hyperparameter combination consists of the learning rate and the number of hidden layer neurons in the rockburst risk prediction model;

[0050] S252: Randomly generate a set of particles, where the position of each particle represents a set of hyperparameters;

[0051] S253: Use validation set data to evaluate model prediction accuracy as a fitness value. As shown in formula (7);

[0052] (7);

[0053] In the formula, To predict the correct number of samples, This represents the total number of samples;

[0054] S254: Based on fitness value Update the individual optimal position of each particle and the global optimal position of the entire particle swarm, and obtain the updated particle velocity according to formula (8). The updated particle position is obtained according to formula (9). ;

[0055] (8);

[0056] (9);

[0057] In the formula, is the velocity of particle i in the k-th iteration, and w is the inertia weight. and These are two different learning factors. and Two random numbers within the interval [0, 1] It is the optimal position of particle i up to the k-th iteration. It is the globally optimal position of the entire particle swarm up to the k-th iteration. It is the position of particle i in the k-th iteration;

[0058] S255: Repeatedly generate a set of particles and update the individual optimal position of each particle and the global optimal position of the entire particle swarm until the preset number of iterations is reached to obtain the optimal hyperparameter combination.

[0059] In this technical solution, the particle swarm optimization algorithm is used to optimize the model hyperparameters, avoiding the subjectivity of human experience in selection. This enables the model to maintain high prediction accuracy under different geological conditions and accurately output the risk level of each region, providing a scientific basis for risk prevention and control.

[0060] As a preferred option, the process of classifying early warnings based on the risk level of rockburst and generating early warning reminders in step six is ​​as follows:

[0061] For situations where the risk level of rockburst is low, a Level 1 warning is issued, the alarm is controlled to perform the Level 1 warning action, and at the same time, a reminder message is generated to implement periodic sampling monitoring of the low-risk area, which is displayed in real time on the monitor.

[0062] For cases where the risk level of rockburst is medium, a level two warning is issued, the alarm is controlled to perform the level two warning action, and at the same time, a reminder message is generated to carry out fixed-point continuous monitoring of the medium-risk area and dynamically adjust the monitoring frequency, which is displayed in real time on the display.

[0063] For situations where the risk level of rockburst is high, a level 3 early warning is issued, the alarm is controlled to perform the level 3 early warning action, and at the same time, a reminder message is generated to deploy a comprehensive three-dimensional monitoring network for high-risk areas and link the early warning response mechanism, which is displayed in real time on the monitor.

[0064] This technical solution, through the automatic generation of early warning classifications and alert suggestions, not only can timely and effective reminders be provided to relevant personnel when anomalies occur, but it can also intelligently generate decision-making information for reference, significantly shortening response time. Furthermore, implementing a tiered monitoring strategy based on different risk levels ensures effective monitoring through differentiated approaches, avoids wasting monitoring resources, and allows for the timely detection of potential risks and the implementation of corresponding countermeasures, greatly reducing the probability of rockbursts and ensuring safe production in the mining area.

[0065] This invention provides a method for monitoring and early warning of rockburst based on multi-source information fusion. First, by collecting multi-dimensional information such as topographic parameters, geological structure data, and rock mass physical properties, it overcomes the limitations of traditional single-source data and can comprehensively reflect the geological conditions of the mining area. Second, by preprocessing the data and using a Kalman filter fusion algorithm, data noise and errors can be effectively reduced, and the reliability of the data is improved through multi-dimensional data fusion. Simultaneously, the accuracy of the fused data is ensured, laying a solid data foundation for the accurate construction of subsequent models and accurate prediction of risk levels. Next, using the accurate fused data as constraints, a three-dimensional geological structure model of the target mining area is generated using a geological structure inversion method. This model clearly and accurately presents the geological structure and rock mass physical property distribution of the mining area, comprehensively reflecting the geological conditions and stress distribution of the entire mining area, and providing a precise spatial framework for regional division. Furthermore, based on the uniformity of rock mass physical properties and the continuity of geological structures... The regional division results in highly representative independent sub-regions, ensuring consistency in geological conditions within each region and providing clear targets for tiered monitoring. This helps reduce the complexity of subsequent rockburst risk prediction models and conserves computational resources during the risk prediction process. Then, by constructing a rockburst risk prediction model based on a backpropagation (BP) neural network and optimizing its hyperparameters using a particle swarm optimization algorithm, the subjectivity of manual experience-based selection is avoided, ensuring high prediction accuracy under various geological conditions. Using the optimized BP neural network to predict risks in the divided independent sub-regions not only significantly improves prediction accuracy but also simultaneously presents the risk status of each sub-region. This allows relevant personnel to understand the situation in each sub-region simultaneously, thus simultaneously grasping the overall and local conditions of the target mining area. This facilitates comprehensive consideration when taking action, leading to more scientific, rational, and effective measures, and better optimizing resource allocation. Finally, tiered early warning systems based on different risk levels can promptly and effectively alert relevant personnel to take tiered measures when anomalies occur, providing a scientific basis for risk prevention and control.

[0066] This method is simple to implement, low in cost, and highly intelligent. It improves data reliability and enhances the accuracy of the three-dimensional geological model by integrating multi-dimensional data. At the same time, it uses an optimized BP neural network for prediction, which significantly improves the ability to predict risk levels. It can efficiently and accurately output the risk levels of each region. Combined with graded early warning, it can provide a reliable decision-making basis for rockburst risk prevention and control. Using this method can achieve efficient and accurate monitoring of rockbursts, effectively reduce the probability of rockbursts, and ensure safe production operations in mining areas.

[0067] The present invention also provides a rockburst monitoring and early warning method system based on multi-source information fusion, which is used to implement the rockburst monitoring and early warning method based on multi-source information fusion, including a processor, an alarm, a display and a memory;

[0068] The processor includes a data acquisition module, a data fusion module, a model building module, a region division module, a risk prediction module, and a graded early warning module;

[0069] The data acquisition module is used to collect multi-dimensional data of the target mining area, including topographic parameters, geological structure data and rock mass physical property information;

[0070] The data fusion module is used to remove outliers from multi-dimensional data and to integrate the multi-dimensional data using a Kalman filter fusion algorithm to obtain fused data.

[0071] The model building module is used to generate a three-dimensional geological structure model of the target mining area based on the fused data and the geological structure inversion method.

[0072] The region division module is used to divide the three-dimensional geological structure model into N independent sub-regions based on the uniformity of rock mass physical property information and the continuity of geological structure data in the three-dimensional geological structure model.

[0073] The risk prediction module is used to take the background feature data of independent sub-regions as input data for the rockburst risk prediction model based on BP neural network, predict the risk, and output the rockburst risk level of each region in the target mining area.

[0074] The graded early warning module is used to classify early warnings according to the risk level of rockburst and generate early warning reminders.

[0075] The alarm is used to perform graded warning actions based on the warning graded results;

[0076] The display is used to show early warning reminders and suggestions in real time;

[0077] The memory is used by the processor to store and retrieve data.

[0078] In this invention, the data acquisition module provides a data receiving channel for the data fusion module, allowing direct input of received multi-dimensional data. The data fusion module automatically performs noise reduction and fusion processing on the data, obtaining accurate fused data and providing a reliable data foundation for the precise construction of subsequent 3D geological structure models. The model construction module automatically constructs a 3D geological structure model that clearly and accurately presents the geological structure and rock mass physical properties of the mining area. The region division module efficiently divides the target mining area into multiple highly representative independent sub-regions, providing clear targets for subsequent risk prediction, reducing the complexity of the risk prediction module, and saving computational power. The risk prediction module significantly improves the accuracy and efficiency of risk prediction. The tiered early warning module alerts relevant personnel to take targeted measures, shortening response time. The alarm provides timely and effective alerts to relevant personnel in case of anomalies. The display allows relevant personnel to easily view early warning information. By configuring the memory, historical data can be easily stored and retrieved in real time.

[0079] The system is simple in structure, highly intelligent, and reliable in performance. It can accurately monitor and provide intelligent early warning of rockbursts, effectively ensuring safe production operations in coal mines. Attached Figure Description

[0080] Figure 1 This is a flowchart of the monitoring method in this invention;

[0081] Figure 2 This is a flowchart of the process for generating a three-dimensional geological structure model in this invention;

[0082] Figure 3 This is a block diagram of the monitoring system in this invention. Detailed Implementation

[0083] The invention will now be further described with reference to the accompanying drawings.

[0084] like Figure 1 As shown, this invention provides a method for monitoring and early warning of rockbursts based on multi-source information fusion, comprising the following steps:

[0085] Step 1: Collect multi-dimensional data of the target mining area, including topographic parameters, geological structure data, and rock mass physical property information;

[0086] As a preferred embodiment, the topographic parameters include the surface elevation, slope, aspect, and surface relief of the mining area; the geological structure data include the location, strike, dip angle, fold morphology, and distribution range of faults; and the rock mass physical property information includes rock mass density, elastic modulus, Poisson's ratio, uniaxial compressive strength, and rock mass integrity coefficient.

[0087] As an example, terrain parameters are collected using a combination of UAV aerial surveying and ground total station. The UAV is equipped with a high-precision optical camera and lidar, and conducts multi-angle, full-coverage aerial photography of the mining area along a preset flight path. Dense point cloud data is generated through photogrammetry, and the surface elevation, slope, aspect, and surface relief of the mining area are extracted after processing. For complex terrain areas that are difficult for UAVs to cover, manual supplementary surveying is conducted using a total station to ensure the completeness of terrain parameters.

[0088] Topographic parameters are the foundation for constructing a three-dimensional topographic framework of a mining area. They can provide a reference for the spatial positioning of subsequent geological structure models, and at the same time affect the stress distribution and rock mass stability of the mining area. They are an important basis for analyzing the environment in which rockbursts occur.

[0089] Geological structural data is collected through a comprehensive approach utilizing geological mapping, geophysical exploration, and drilling. Geologists conduct field geological mapping to record the location of faults (marked with latitude and longitude coordinates), fault strike (the direction of the intersection of the fault line and the horizontal plane), fault dip angle (the angle at which the fault plane dips downwards), fold morphology (such as anticlines and synclines), and their distribution range. Geophysical exploration methods, such as seismic exploration and ground-penetrating radar, are used to detect hidden underground geological structures and determine their extension. Drilling and sampling are conducted in key structural areas to obtain core samples for verification and refinement of geological structural characteristics. Geological structures are a key factor influencing the occurrence of rockbursts; faults and folds lead to stress concentration, and their data can provide direct clues for identifying high-risk areas, forming the core content of the structural framework in constructing a three-dimensional geological structural model.

[0090] Information on the physical properties of the rock mass was collected through a combination of laboratory tests and field tests. Rock samples were collected from different locations in the mining area and sent to the laboratory for testing to determine the rock mass density, elastic modulus, Poisson's ratio, and uniaxial compressive strength. An acoustic wave tester was used to test the rock mass in the field, and the rock mass integrity coefficient was calculated based on the sound wave propagation speed.

[0091] Step 2: Process the multi-dimensional data and use the Kalman filter fusion algorithm to integrate the multi-dimensional data to obtain fused data;

[0092] As a preferred option, the process of obtaining fused data is as follows:

[0093] S21: Remove outliers from multidimensional data using the Raida criterion;

[0094] S22: The Kalman filter fusion algorithm is used to integrate the data to obtain fused data;

[0095] S22-1: Establish the system's state equations according to formula (1), and establish the system's observation equations according to formula (2);

[0096] (1);

[0097] (2);

[0098] In the formula, Let k be the system state at time k. Let be the state transition matrix, representing the system transition from ... The dynamic evolution relationship from time k to time k. for The system state at any given moment. for The system noise driving matrix at time t, for System noise at any given moment Let k be the observation vector at time k. Let be the observation matrix at time k. The observation noise at time k;

[0099] S22-2: State prediction; Establish the state prediction equation according to formula (3), so as to... The posterior estimate at time k is used to estimate the state at time k, thus obtaining the prior estimate at time k. ;

[0100] (3);

[0101] In the formula, This represents the predicted value of the system state vector at time k, based on all observations up to time k-1. This represents the optimal estimate of the system state vector at time k-1 after combining the observation information at time k-1;

[0102] S22-3: State update; Establish the state update equation according to formula (4) to use the measurement value at time k to correct the estimated value in the prediction stage and obtain the posterior estimated value at time k.

[0103] (4);

[0104] In the formula, Indicates Kalman gain, Indicates the integration of the observation vector at time k. Then, the optimal estimate of the system state vector at time k;

[0105] S22-4: Repeatedly execute S22-2 and S22-3 to continuously fuse the data and finally obtain the fused data.

[0106] By using the Raida criterion to remove outliers and the Kalman filter fusion algorithm to integrate the data, data noise and errors were effectively reduced, ensuring the accuracy of the fused data and laying a solid data foundation for subsequent model building and risk prediction.

[0107] Step 3: Based on the fused data, use the geological structure inversion method to generate a three-dimensional geological structure model of the target mining area;

[0108] like Figure 2 As shown, as a preferred method, the process of generating a three-dimensional geological structure model of the target mining area using the geological structure inversion method is as follows:

[0109] S31: Using fused data as constraints, a geological structure inversion method based on Bayesian theory is adopted, and an inversion objective function is constructed according to formula (5). ;

[0110] (5);

[0111] In the formula, m is the model parameter vector, d is the observation data vector, and G(m) is the forward modeling operator. Let C be the initial model parameter vector, and C be the covariance matrix of the model parameters;

[0112] S32: Solving the inversion objective function using the Markov chain Monte Carlo method Obtain the posterior probability distribution of the model parameters;

[0113] An initial set of model parameters is determined, based on existing geological knowledge, regional geological data, and previous exploration data, serving as the starting point for the Markov chain. New candidate values ​​for model parameters are generated according to certain probability rules, considering the current distribution of model parameters and the characteristics of the inversion objective function to ensure that the generated candidate parameters are within a reasonable range.

[0114] Calculate the inversion objective function value corresponding to the new candidate parameter and compare it with the objective function value of the current model parameter. If the objective function value of the new candidate parameter is smaller, it indicates that it better meets the constraints of the fused data, and the candidate parameter is directly accepted as the current model parameter. If the objective function value of the new candidate parameter is larger, it is not directly discarded, but rather its acceptance is determined by a probability. This probability is related to the difference between the objective function values ​​of the two parameters; the smaller the difference, the higher the probability of acceptance, thus avoiding getting trapped in local optima and ensuring the comprehensiveness of the search.

[0115] The process of repeatedly generating candidate parameters, calculating the objective function value, comparing and deciding whether to accept them forms a Markov chain. During this iterative process, the distribution of model parameters is continuously monitored. When the parameter distribution stabilizes—that is, when the statistical characteristics of the parameters (such as mean and variance) no longer change significantly after multiple iterations—the Markov chain is considered to have converged. At this point, all accepted model parameters from the iteration process are collected, forming the posterior probability distribution of the model parameters, reflecting the possible values ​​of each model parameter and their probabilities under the constraints of the fused data.

[0116] S33: Based on the maximum likelihood estimation results of the posterior probability distribution, construct a three-dimensional mesh model of the target mining area, assign the model parameters to the corresponding mesh cells, and generate a three-dimensional geological structure model.

[0117] From the posterior probability distribution of the model parameters, find the combination of model parameters with the highest probability of occurrence, which is the maximum likelihood estimation result. This represents the model parameters that are most likely to approximate the actual geological conditions under the current constraints of the fused data.

[0118] The size and density of the 3D mesh are determined based on the topographical extent, geological complexity, and accuracy requirements of the target mining area for subsequent analysis. For areas with complex geological structures and high interest, the mesh is made denser to ensure detailed representation of the model; for areas with relatively simple geological conditions, the mesh can be made sparser to improve computational efficiency.

[0119] A three-dimensional mesh is applied to the entire target mining area, with each mesh cell corresponding to a specific spatial location within the mining area. The model parameters obtained from the maximum likelihood estimation are then assigned according to the spatial location of the mesh cells. Each mesh cell is assigned geological parameters corresponding to its location, such as lithology, thickness, porosity, and permeability of the rock strata.

[0120] Using professional 3D modeling software, parameterized mesh cells are integrated and visualized. Based on the spatial coordinates and parameter information of the mesh cells, a three-dimensional model with a sense of depth is constructed, showcasing the distribution of different rock layers, the morphology of geological structures, and the spatial variations of rock mass physical properties within the mining area, thus obtaining a three-dimensional geological structure model of the target mining area.

[0121] A geological structure inversion method based on Bayesian theory, combined with fused data as constraints, is employed. The objective function is solved using the Markov chain Monte Carlo method, which quantifies the uncertainty of model parameters and yields a posterior probability distribution of model parameters that more closely reflects real geological conditions. This technique significantly enhances the accuracy of the 3D geological model, enabling the constructed 3D geological structure model to clearly represent the geological structure and rock mass physical properties of the mining area, providing a precise spatial framework for subsequent regional delineation.

[0122] Step 4: Based on the uniformity of rock mass physical property information and the continuity of geological structure data in the 3D geological structure model, the 3D geological structure model is divided into regions to obtain N independent sub-regions; each independent sub-region includes background feature data, which includes terrain parameters, geological structure data and rock mass physical property information;

[0123] As a preferred method, the process of dividing the three-dimensional geological structure model into regions is as follows:

[0124] S41: Using the elastic modulus and uniaxial compressive strength in the rock mass physical property information as the dividing index, calculate the rock mass physical property difference coefficient between any two points in the three-dimensional geological structure model according to formula (6). ;

[0125] (6);

[0126] In the formula, and The elastic modulus at two points are respectively. and The uniaxial compressive strengths at two points are respectively. and These are the maximum values ​​of elastic modulus and uniaxial compressive strength within the mining area, respectively.

[0127] S42: Set the difference coefficient threshold When the difference coefficient of rock mass physical properties between two points When two points are located in the same region, they are assigned to the same sub-region.

[0128] S43: Combining the distribution of faults and folds in geological structural data, areas separated by faults or with fold morphological differences exceeding a preset difference threshold are divided into different sub-regions.

[0129] Spatial distribution information of faults is extracted from a 3D geological structure model, including the fault's strike, dip, dip angle, and extent. The faults are then visualized using 3D modeling software, clearly defining their specific locations and morphologies within the model. For areas completely separated by faults, the presence of faults disrupts the continuity of the rock mass, resulting in significant differences in stress states and geological evolution on either side. Therefore, these fault-separated areas are directly divided into different preliminary sub-regions.

[0130] To analyze the differences in fold morphology, first determine the indicators to measure these differences, such as the dip angle of the fold's axial plane, the strike of the hinge, and the dip angle of its limbs. Then, calculate the indicator values ​​for folds in different regions and compare them with a preset difference threshold. When the difference in indicator values ​​between two adjacent regions exceeds the preset threshold, it indicates that their fold morphologies are significantly different, reflecting poor continuity in their geological structure. These two regions are then divided into different preliminary sub-regions.

[0131] S44: After the initial partitioning is completed using the region growing algorithm, the boundaries are smoothed to obtain N independent sub-regions.

[0132] In the initial partitioning of the region growing algorithm, points that satisfy the rock mass physical property difference coefficient condition and are not separated by geological structures are used as seed points. Starting from the seed point, neighboring points that meet the condition are continuously added to the current region until no more points can be added, forming multiple initial regions. However, the boundaries of the regions obtained at this time may be relatively rough, with jagged or irregular protrusions and depressions.

[0133] A moving average method is used for boundary smoothing. This involves selecting a series of points on the boundary, taking a certain number of neighboring boundary points as the center of each point, and calculating the average coordinates of these neighboring points. This average is then used as the smoothed coordinates of the original point. The boundary is adjusted point by point to smooth out any abrupt changes. Simultaneously, for overly sharp corners on the boundary, curve fitting is used to fit a smooth curve based on the boundary points near the corner, replacing the original corner segment and making the boundary transition more natural.

[0134] During the smoothing process, the smoothed boundaries are continuously checked to ensure they accurately reflect the actual geological structure, avoiding the obscuring of important geological boundary features due to over-smoothing. After multiple iterative adjustments, the boundaries are smoothed and conform to geological reality, ultimately resulting in N independent sub-regions with clear boundaries and regular shapes.

[0135] Based on the uniformity of rock mass physical properties and the continuity of geological structure, the regions are divided. By combining the difference coefficient calculation and the distribution of geological structures, and through regional growth algorithm and boundary smoothing processing, the resulting independent sub-regions are highly representative, ensuring the consistency of geological conditions within each region and providing clear targets for hierarchical monitoring.

[0136] Step 5: Collect historical rockburst event data from mining areas, construct a rockburst risk prediction model based on a BP neural network, and optimize the hyperparameters of the rockburst risk prediction model using a particle swarm optimization algorithm. Train the rockburst risk prediction model based on the hyperparameters. Use independent sub-regions as input data, perform risk prediction using the rockburst risk prediction model, and output the rockburst risk level of each region in the target mining area. The rockburst risk level includes low, medium, and high levels.

[0137] In order to obtain a high-accuracy prediction model for rockburst risk, the historical rockburst event data in the mining area includes background feature data of the event occurrence area and specific attributes of the rockburst event; the background feature data of the event occurrence area includes the topographic parameters, geological structure data and rock mass physical property information of the area; the specific attributes of the rockburst event include the spatiotemporal information of the event, the impact intensity and the corresponding rockburst risk probability value, and the spatiotemporal information of the event includes the time and location coordinates of the rockburst occurrence.

[0138] As a preferred option, the process of constructing a rockburst risk prediction model based on a BP neural network is as follows:

[0139] S151: Construct the basic BP neural network model, which includes an input layer, multiple hidden layers, and an output layer. The input layer receives background feature data from independent sub-regions. Multiple hidden layers perform nonlinear transformations and feature mappings on the data received from the input layer. The hidden layers use the Sigmoid activation function to enhance the model's ability to fit nonlinear relationships. The output layer outputs the probability value of rockburst risk for independent sub-regions. The output layer has three nodes, corresponding to low, medium, and high rockburst risk levels, respectively.

[0140] S152: Take the regional feature data corresponding to the event location in the historical mining area rockburst event data as input, take the rockburst risk probability value corresponding to the event as output, train the BP neural network basic model, and save the model parameters that meet the preset accuracy. After training, a rockburst risk prediction model based on BP neural network is obtained.

[0141] By constructing a risk prediction model based on a BP neural network and training it with historical rockburst event data, the model can effectively capture the complex nonlinear relationships of rockburst occurrence and ensure prediction accuracy.

[0142] As a preferred approach, the process of optimizing the hyperparameters of the rockburst risk prediction model using the particle swarm optimization algorithm is as follows:

[0143] S251: The hyperparameter combination consists of the learning rate and the number of hidden layer neurons in the rockburst risk prediction model;

[0144] S252: Randomly generate a set of particles, where the position of each particle represents a set of hyperparameters;

[0145] S253: Use validation set data to evaluate model prediction accuracy as a fitness value. As shown in formula (7);

[0146] (7);

[0147] In the formula, To predict the correct number of samples, This represents the total number of samples;

[0148] S254: Based on fitness value Update the individual optimal position of each particle and the global optimal position of the entire particle swarm, and obtain the updated particle velocity according to formula (8). The updated particle position is obtained according to formula (9). ;

[0149] (8);

[0150] (9);

[0151] In the formula, is the velocity of particle i in the k-th iteration, and w is the inertia weight. and These are two different learning factors. and Two random numbers within the interval [0, 1] It is the optimal position of particle i up to the k-th iteration. It is the globally optimal position of the entire particle swarm up to the k-th iteration. It is the position of particle i in the k-th iteration;

[0152] S255: Repeatedly generate a set of particles and update the individual optimal position of each particle and the global optimal position of the entire particle swarm until the preset number of iterations is reached to obtain the optimal hyperparameter combination.

[0153] The particle swarm optimization algorithm was used to optimize the model hyperparameters, avoiding the subjectivity of human experience in selection. This enabled the model to maintain high prediction accuracy under different geological conditions and accurately output the risk level of each region, providing a scientific basis for risk prevention and control.

[0154] Step Six: Classify the early warning level according to the risk level of rockburst and generate early warning reminder suggestions.

[0155] As a preferred approach, the process of classifying early warnings based on the risk level of rockburst and generating early warning reminders is as follows: For low-risk rockburst situations, a Level 1 warning is issued, controlling the alarm to execute the Level 1 warning action. Simultaneously, a reminder message is generated to implement periodic sampling monitoring of low-risk areas, and this is displayed in real-time on a monitor. For medium-risk rockburst situations, a Level 2 warning is issued, controlling the alarm to execute the Level 2 warning action. Simultaneously, a reminder message is generated to implement fixed-point continuous monitoring and dynamically adjust the monitoring frequency of medium-risk areas, and this is displayed in real-time on a monitor. For high-risk rockburst situations, a Level 3 warning is issued, controlling the alarm to execute the Level 3 warning action. Simultaneously, a reminder message is generated to deploy a comprehensive three-dimensional monitoring network and link it with the early warning response mechanism in high-risk areas, and this is displayed in real-time on a monitor.

[0156] A low-risk area indicates that the area has never experienced a rockburst event in its historical records, and according to the rockburst risk prediction model, the probability of a future rockburst event is extremely low. Furthermore, the rock mass in the area has stable physical properties, a simple geological structure, no obvious stress concentration, and minimal disturbance to the rock mass from mining activities.

[0157] A medium-risk area indicates that no rockburst events have occurred in the area historically, but risk prediction models show a certain probability of future occurrence. The area may have localized differences in rock physical properties or complex geological structures, such as small folds without significant stress concentration zones. Changes in mining operations may trigger potential rockburst risks during the mining process.

[0158] High-risk areas indicate that the area has a history of rockburst events, or that risk prediction models assess a high probability of future rockburst events. These areas exhibit significant differences in rock mass physical properties, complex geological structures, and the presence of obvious faults and large folds, making them prone to stress concentration. Mining activities cause significant disturbance to the rock mass, greatly increasing the likelihood of triggering rockbursts.

[0159] Periodic sampling monitoring of low-risk areas involves developing a distribution plan for sampling monitoring points based on the scope and terrain characteristics of the low-risk areas to ensure that the monitoring points can evenly cover the entire area.

[0160] In medium-risk areas, fixed monitoring points are set up at key locations based on geological structure and mining plans. Each monitoring point is equipped with stress sensors, microseismic sensors and other equipment to achieve continuous monitoring of parameters such as rock mass stress and microseismic activity.

[0161] A comprehensive, three-dimensional monitoring network and coordinated early warning and response mechanism are established for high-risk areas. This network involves constructing a comprehensive, three-dimensional monitoring system. On the surface, multiple monitoring base stations are set up, equipped with high-precision GPS displacement monitoring systems and meteorological monitoring equipment to monitor real-time changes in surface displacement and the impact of meteorological conditions on the rock mass. In the underground mining face and surrounding rock mass, stress sensors, displacement sensors, microseismic sensors, and infrared thermometers are densely deployed to achieve comprehensive monitoring of parameters such as rock mass stress, displacement, microseismic activity, and temperature.

[0162] By automatically generating early warning grading and alert suggestions, timely and effective reminders can be provided to relevant personnel when anomalies occur. Furthermore, intelligent decision-making information can be generated for reference, significantly shortening response time. In addition, implementing tiered monitoring strategies based on different risk levels ensures effective monitoring through differentiated approaches, avoids wasting monitoring resources, and allows for the timely detection of potential risks and the implementation of corresponding measures, greatly reducing the probability of rockbursts and ensuring safe production in the mining area.

[0163] This invention provides a method for monitoring and early warning of rockburst based on multi-source information fusion. First, by collecting multi-dimensional information such as topographic parameters, geological structure data, and rock mass physical properties, it overcomes the limitations of traditional single-source data and can comprehensively reflect the geological conditions of the mining area. Second, by preprocessing the data and using a Kalman filter fusion algorithm, data noise and errors can be effectively reduced, and the reliability of the data is improved through multi-dimensional data fusion. Simultaneously, the accuracy of the fused data is ensured, laying a solid data foundation for the accurate construction of subsequent models and accurate prediction of risk levels. Next, using the accurate fused data as constraints, a three-dimensional geological structure model of the target mining area is generated using a geological structure inversion method. This model clearly and accurately presents the geological structure and rock mass physical property distribution of the mining area, comprehensively reflecting the geological conditions and stress distribution of the entire mining area, and providing a precise spatial framework for regional division. Furthermore, based on the uniformity of rock mass physical properties and the continuity of geological structures... The regional division results in highly representative independent sub-regions, ensuring consistency in geological conditions within each region and providing clear targets for tiered monitoring. This helps reduce the complexity of subsequent rockburst risk prediction models and conserves computational resources during the risk prediction process. Then, by constructing a rockburst risk prediction model based on a backpropagation (BP) neural network and optimizing its hyperparameters using a particle swarm optimization algorithm, the subjectivity of manual experience-based selection is avoided, ensuring high prediction accuracy under various geological conditions. Using the optimized BP neural network to predict risks in the divided independent sub-regions not only significantly improves prediction accuracy but also simultaneously presents the risk status of each sub-region. This allows relevant personnel to understand the situation in each sub-region simultaneously, thus simultaneously grasping the overall and local conditions of the target mining area. This facilitates comprehensive consideration when taking action, leading to more scientific, rational, and effective measures, and better optimizing resource allocation. Finally, tiered early warning systems based on different risk levels can promptly and effectively alert relevant personnel to take tiered measures when anomalies occur, providing a scientific basis for risk prevention and control.

[0164] This method is simple to implement, low in cost, and highly intelligent. It improves data reliability and enhances the accuracy of the three-dimensional geological model by integrating multi-dimensional data. At the same time, it uses an optimized BP neural network for prediction, which significantly improves the ability to predict risk levels. It can efficiently and accurately output the risk levels of each region. Combined with graded early warning, it can provide a reliable decision-making basis for rockburst risk prevention and control. Using this method can achieve efficient and accurate monitoring of rockbursts, effectively reduce the probability of rockbursts, and ensure safe production operations in mining areas.

[0165] like Figure 3As shown, based on the same inventive concept, the present invention also provides a rockburst monitoring and early warning method system based on multi-source information fusion, which is used to implement the rockburst monitoring and early warning method based on multi-source information fusion, including a processor, an alarm, a display and a memory;

[0166] The processor includes a data acquisition module, a data fusion module, a model building module, a region division module, a risk prediction module, and a graded early warning module;

[0167] The data acquisition module is used to collect multi-dimensional data of the target mining area, including topographic parameters, geological structure data and rock mass physical property information;

[0168] The data fusion module is used to remove outliers from multi-dimensional data and to integrate the multi-dimensional data using a Kalman filter fusion algorithm to obtain fused data.

[0169] The model building module is used to generate a three-dimensional geological structure model of the target mining area based on the fused data and the geological structure inversion method.

[0170] The region division module is used to divide the three-dimensional geological structure model into N independent sub-regions based on the uniformity of rock mass physical property information and the continuity of geological structure data in the three-dimensional geological structure model.

[0171] The risk prediction module is used to take the background feature data of independent sub-regions as input data for the rockburst risk prediction model based on BP neural network, predict the risk, and output the rockburst risk level of each region in the target mining area.

[0172] The graded early warning module is used to classify early warnings according to the risk level of rockburst and generate early warning reminders.

[0173] The alarm is used to perform graded warning actions based on the warning graded results;

[0174] The display is used to show early warning reminders and suggestions in real time;

[0175] The memory is used by the processor to store and retrieve data.

[0176] As a preferred embodiment, the processor is an industrial computer. Preferably, the alarm is an audible and visual alarm capable of broadcasting preset voice messages. Specifically, the preset voice message corresponding to a Level 1 warning action is: periodic sampling monitoring of low-risk areas; the preset voice message corresponding to a Level 2 warning action is: continuous fixed-point monitoring of medium-risk areas with dynamic adjustment of monitoring frequency; and the preset voice message corresponding to a Level 3 warning action is: deployment of a comprehensive three-dimensional monitoring network and linkage with the warning response mechanism for high-risk areas.

[0177] In this invention, the data acquisition module provides a data receiving channel for the data fusion module, allowing direct input of received multi-dimensional data. The data fusion module automatically performs noise reduction and fusion processing on the data, obtaining accurate fused data and providing a reliable data foundation for the precise construction of subsequent 3D geological structure models. The model construction module automatically constructs a 3D geological structure model that clearly and accurately presents the geological structure and rock mass physical properties of the mining area. The region division module efficiently divides the target mining area into multiple highly representative independent sub-regions, providing clear targets for subsequent risk prediction, reducing the complexity of the risk prediction module, and saving computational power. The risk prediction module significantly improves the accuracy and efficiency of risk prediction. The tiered early warning module alerts relevant personnel to take targeted measures, shortening response time. The alarm provides timely and effective alerts to relevant personnel in case of anomalies. The display allows relevant personnel to easily view early warning information. By configuring the memory, historical data can be easily stored and retrieved in real time.

[0178] The system is simple in structure, highly intelligent, and reliable in performance. It can accurately monitor and provide intelligent early warning of rockbursts, effectively ensuring safe production operations in coal mines.

Claims

1. A method for monitoring and early warning of rockbursts based on multi-source information fusion, characterized in that, Includes the following steps: Step 1: Collect multi-dimensional data of the target mining area, including topographic parameters, geological structure data, and rock mass physical property information; Step 2: Process the multi-dimensional data and use the Kalman filter fusion algorithm to integrate the multi-dimensional data to obtain fused data; Step 3: Based on the fused data, use the geological structure inversion method to generate a three-dimensional geological structure model of the target mining area; Step 4: Based on the uniformity of rock mass physical property information and the continuity of geological structure data in the 3D geological structure model, the 3D geological structure model is divided into regions to obtain N independent sub-regions; each independent sub-region includes background feature data, which includes terrain parameters, geological structure data and rock mass physical property information; The process of dividing the three-dimensional geological structure model into regions is as follows: S41: Using the elastic modulus and uniaxial compressive strength in the rock mass physical property information as the dividing index, calculate the rock mass physical property difference coefficient between any two points in the three-dimensional geological structure model according to formula (6). ; （6）； In the formula, and The elastic modulus at two points are respectively. and The uniaxial compressive strengths at two points are respectively. and These are the maximum values ​​of elastic modulus and uniaxial compressive strength within the mining area, respectively. S42: Set the difference coefficient threshold When the difference coefficient of rock mass physical properties between two points When two points are located in the same region, they are assigned to the same sub-region. S43: Combining the distribution of faults and folds in geological structural data, areas separated by faults or with fold morphological differences exceeding a preset difference threshold are divided into different sub-regions. S44: After the initial partitioning is completed using the region growing algorithm, the boundaries are smoothed to obtain N independent sub-regions; Step 5: Collect historical rockburst event data from mining areas, construct a rockburst risk prediction model based on a BP neural network, and optimize the hyperparameters of the rockburst risk prediction model using a particle swarm optimization algorithm. Train the rockburst risk prediction model based on the hyperparameters. Use independent sub-regions as input data, perform risk prediction using the rockburst risk prediction model, and output the rockburst risk level of each region in the target mining area. The rockburst risk level includes low, medium, and high levels. Step Six: Classify the early warning level according to the risk level of rockburst and generate early warning reminder suggestions.

2. The method for monitoring and early warning of rockbursts based on multi-source information fusion according to claim 1, characterized in that, In step one, the topographic parameters include the surface elevation, slope, aspect, and surface relief of the mining area; the geological structure data include the location, strike, dip angle, fold morphology, and distribution range of faults; and the rock mass physical property information includes rock mass density, elastic modulus, Poisson's ratio, uniaxial compressive strength, and rock mass integrity coefficient.

3. The method for monitoring and early warning of rockbursts based on multi-source information fusion according to claim 2, characterized in that, In step two, the process of obtaining the fused data is as follows: S21: Remove outliers from multidimensional data using the Raida criterion; S22: The Kalman filter fusion algorithm is used to integrate the data to obtain fused data; S22-1: Establish the system's state equations according to formula (1), and establish the system's observation equations according to formula (2); （1）； （2）； In the formula, Let k be the system state at time k. Let be the state transition matrix, representing the system transition from ... The dynamic evolution relationship from time k to time k. for The system state at any given moment. for The system noise driving matrix at time t, for System noise at any given moment Let k be the observation vector at time k. Let be the observation matrix at time k. The observation noise at time k; S22-2: State prediction; Establish the state prediction equation according to formula (3), so as to... The posterior estimate at time k is used to estimate the state at time k, thus obtaining the prior estimate at time k. ; （3）； In the formula, This represents the optimal estimate of the system state vector at time k-1 after combining the observation information at time k-1; S22-3: State Update; Establish the state update equation according to formula (4), use the measurement value at time k to correct the estimated value in the prediction stage, and obtain the posterior estimated value at time k. ; （4）； In the formula, Indicates Kalman gain; S22-4: Repeatedly execute S22-2 and S22-3 to continuously fuse the data and finally obtain the fused data.

4. The method for monitoring and early warning of rockbursts based on multi-source information fusion according to claim 3, characterized in that, In step three, the process of generating a three-dimensional geological structure model of the target mining area using the geological structure inversion method is as follows: S31: Using fused data as constraints, a geological structure inversion method based on Bayesian theory is adopted, and an inversion objective function is constructed according to formula (5). ; （5）； In the formula, m is the model parameter vector, d is the observation data vector, and G(m) is the forward modeling operator. Let C be the initial model parameter vector, and C be the covariance matrix of the model parameters; S32: Solving the inversion objective function using the Markov chain Monte Carlo method Obtain the posterior probability distribution of the model parameters; S33: Based on the maximum likelihood estimation results of the posterior probability distribution, construct a three-dimensional mesh model of the target mining area, assign the model parameters to the corresponding mesh cells, and generate a three-dimensional geological structure model.

5. The method for monitoring and early warning of rockbursts based on multi-source information fusion according to claim 4, characterized in that, In step five, the historical rockburst event data in the mining area includes background feature data of the event location and specific attributes of the rockburst event; the background feature data of the event location includes the topographic parameters, geological structure data and rock mass physical property information of the area; the specific attributes of the rockburst event include the spatiotemporal information of the event, the impact intensity and the rockburst risk probability value corresponding to the event, and the spatiotemporal information of the event includes the time and location coordinates of the rockburst occurrence.

6. The method for monitoring and early warning of rockbursts based on multi-source information fusion according to claim 5, characterized in that, In step five, the process of constructing a rockburst risk prediction model based on a BP neural network is as follows: S151: Construct the basic BP neural network model, which includes an input layer, multiple hidden layers, and an output layer. The input layer is used to receive background feature data of independent sub-regions. Multiple hidden layers are used to perform nonlinear transformations and feature mappings on the data received by the input layer. The output layer is used to output the probability value of rockburst risk for independent sub-regions. S152: Take the regional feature data corresponding to the event location in the historical mining area rockburst event data as input, take the rockburst risk probability value corresponding to the event as output, train the BP neural network basic model, and save the model parameters that meet the preset accuracy. After training, a rockburst risk prediction model based on BP neural network is obtained.

7. The method for monitoring and early warning of rockbursts based on multi-source information fusion according to claim 6, characterized in that, In step five, the process of optimizing the hyperparameters of the rockburst risk prediction model using the particle swarm optimization algorithm is as follows: S251: The hyperparameter combination consists of the learning rate and the number of hidden layer neurons in the rockburst risk prediction model; S252: Randomly generate a set of particles, where the position of each particle represents a set of hyperparameters; S253: Use validation set data to evaluate model prediction accuracy as a fitness value. As shown in formula (7); （7）； In the formula, To predict the correct number of samples, This represents the total number of samples; S254: Based on fitness value Update the individual optimal position of each particle and the global optimal position of the entire particle swarm, and obtain the updated particle velocity according to formula (8). The updated particle position is obtained according to formula (9). ; （8）； （9）； In the formula, is the velocity of particle i in the k-th iteration, and w is the inertia weight. and These are two different learning factors. and Two random numbers within the interval [0, 1] It is the optimal position of particle i up to the k-th iteration. It is the globally optimal position of the entire particle swarm up to the k-th iteration. It is the position of particle i in the k-th iteration; S255: Repeatedly generate a set of particles and update the individual optimal position of each particle and the global optimal position of the entire particle swarm until the preset number of iterations is reached to obtain the optimal hyperparameter combination.

8. The method for monitoring and early warning of rockburst based on multi-source information fusion according to claim 7, characterized in that, In step six, the process of classifying early warnings based on the risk level of rockburst and generating early warning reminders is as follows: For situations where the risk level of rockburst is low, a Level 1 warning is issued, the alarm is controlled to perform the Level 1 warning action, and at the same time, a reminder message is generated to implement periodic sampling monitoring of the low-risk area, which is displayed in real time on the monitor. For cases where the risk level of rockburst is medium, a level two warning is issued, the alarm is controlled to perform the level two warning action, and at the same time, a reminder message is generated to carry out fixed-point continuous monitoring of the medium-risk area and dynamically adjust the monitoring frequency, which is displayed in real time on the display. For situations where the risk level of rockburst is high, a level 3 early warning is issued, the alarm is controlled to perform the level 3 early warning action, and at the same time, a reminder message is generated to deploy a comprehensive three-dimensional monitoring network for high-risk areas and link the early warning response mechanism, which is displayed in real time on the monitor.

9. A method and system for monitoring and early warning of rockbursts based on multi-source information fusion, used to implement the method for monitoring and early warning of rockbursts based on multi-source information fusion as described in any one of claims 1 to 8, characterized in that, Includes processor, alarm, display, and memory; The processor includes a data acquisition module, a data fusion module, a model building module, a region division module, a risk prediction module, and a graded early warning module; The data acquisition module is used to collect multi-dimensional data of the target mining area, including topographic parameters, geological structure data and rock mass physical property information; The data fusion module is used to remove outliers from multi-dimensional data and to integrate the multi-dimensional data using a Kalman filter fusion algorithm to obtain fused data. The model building module is used to generate a three-dimensional geological structure model of the target mining area based on the fused data and the geological structure inversion method. The region division module is used to divide the three-dimensional geological structure model into N independent sub-regions based on the uniformity of rock mass physical property information and the continuity of geological structure data in the three-dimensional geological structure model. The risk prediction module is used to take the background feature data of independent sub-regions as input data for the rockburst risk prediction model based on BP neural network, predict the risk, and output the rockburst risk level of each region in the target mining area. The graded early warning module is used to classify early warnings according to the risk level of rockburst.

Images