AI tomographic instability mechanics model construction and dynamic early warning method and system fusing multi-source data

Abstract

This invention discloses an AI-based fault instability mechanics model construction and dynamic early warning method and system that integrates multi-source data, relating to the fields of coal mine safety monitoring and geological disaster early warning technology. The method includes: collecting and preprocessing multi-source monitoring data to obtain a standardized dataset; constructing a fault instability mechanics model based on mechanical theory, including slip instability criteria and deflection instability criteria; optimizing the fault instability mechanics model using a swarm intelligence optimization algorithm; constructing a hybrid prediction model based on the optimized fault instability mechanics model and a pre-trained neural network model; inputting the standardized dataset into the hybrid prediction model and outputting a comprehensive safety factor reflecting fault stability; and establishing a graded early warning mechanism based on the comprehensive safety factor and its dynamic changing trend, and outputting corresponding early warning information and handling suggestions. This solves the problems of slow convergence and susceptibility to local optima in traditional optimization algorithms.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of coal mine safety monitoring and geological disaster early warning technology, and in particular to the construction of AI fault instability mechanical model and dynamic early warning method and system that integrates multi-source data. Background Technology

[0002] Faults, as common geological structures in underground engineering, are highly susceptible to major disasters such as rock bursts and roof collapses due to their instability, seriously threatening the lives of construction workers and the normal progress of the project. Especially in mining areas with thick, hard roofs, faults are significantly affected by mining activities, and their instability mechanisms involve the coupling of multiple factors such as stress evolution, rock strata movement, and geometric morphology. Traditional monitoring and early warning methods have many limitations.

[0003] Limitations of data acquisition and utilization: Existing technologies mostly rely on single types of monitoring data and fail to effectively integrate real-time status data (such as stress and displacement) with basic mechanical parameters (such as cohesion and internal friction angle). Moreover, the basic mechanical parameters are mostly fixed initial values, which cannot adapt to the dynamic changes during mining. At the same time, the simple removal of outliers in data preprocessing may obliterate the precursor signals of instability and affect the accuracy of early warning.

[0004] Insufficient mechanical models and optimization algorithms: Traditional fault instability mechanical models are mostly based on a single instability criterion (such as only considering slip instability), which fails to fully cover the deflection instability mode unique to thick and hard roof mining areas; and the optimization of model parameters mostly adopts traditional algorithms such as genetic algorithm (GA) and particle swarm optimization (PSO), which have problems such as slow convergence speed, easy to get trapped in local optima, and poor adaptability to complex nonlinear problems, making it difficult to meet the requirements of high-precision parameter optimization.

[0005] Deficiencies in early warning models and threshold settings: Most existing early warning models are purely physical models or purely data-driven models. The former has weak generalization ability, while the latter lacks mechanical theoretical support. Early warning thresholds are mostly empirical values, lacking standardized basis and numerical simulation verification. Moreover, the early warning logic is simple, which easily produces false alarms or missed alarms. Emergency response suggestions lack immediate operability.

[0006] Therefore, there is an urgent need to develop a dynamic early warning system for fault instability that integrates multi-source data, combines mechanical theory with AI intelligent learning capabilities, optimizes parameters efficiently, and provides accurate early warnings. Summary of the Invention

[0007] The core objective of this invention is to address the technical pain points of fault instability early warning in thick and hard roof mining areas by providing an AI-based fault instability mechanical model construction and dynamic early warning method and system that integrates multi-source data. By utilizing the Arctic Fox optimization algorithm through population grouping, multi-stage search, leader incentive and mutation mechanisms, it solves the problems of slow convergence and easy getting trapped in local optima in traditional optimization algorithms, and achieves efficient adaptive optimization of the core parameters of the mechanical model (cohesion, internal friction angle, bulk modulus, shear modulus), thereby improving the adaptability of model parameters to actual geological conditions.

[0008] In a first aspect, embodiments of the present invention provide a method for constructing and dynamically warning of an AI fault instability mechanical model that integrates multi-source data, including:

[0009] Multi-source monitoring data is collected in the fault-affected area and the data is preprocessed to obtain a standardized dataset. The multi-source monitoring data includes at least real-time status data and basic mechanical parameters.

[0010] A fault instability mechanical model is constructed based on mechanical theory. The fault instability mechanical model includes slip instability criteria and deflection instability criteria.

[0011] A swarm intelligence optimization algorithm is used to optimize the parameters of the fault instability mechanical model with the goal of minimizing the deviation between the predicted results of the fault instability mechanical model and the real-time state data.

[0012] Based on the optimized fault instability mechanical model and the pre-trained neural network model, a hybrid prediction model is constructed through adaptive weights;

[0013] The standardized dataset is input into the hybrid prediction model, and a comprehensive safety factor reflecting fault stability is output.

[0014] Based on the comprehensive safety factor and its dynamic change trend, a graded early warning mechanism is established and corresponding early warning information and handling suggestions are output.

[0015] In a preferred embodiment, multi-source monitoring data is collected in the fault-affected area and the data is preprocessed to obtain a standardized dataset, including:

[0016] The collected multi-source monitoring data is indexed by timestamps and spatial coordinates to generate a structured dataset;

[0017] The structured dataset is standardized by performing data standardization, spatiotemporal alignment, and outlier handling to obtain a standardized dataset.

[0018] In a preferred embodiment, the fault instability mechanical model includes a slip instability criterion and a deflection instability criterion, wherein...

[0019] The slip instability criterion is constructed based on the Mohr-Coulomb strength theory and is used to determine the risk of slip instability by comparing the shear stress and shear strength at the fracture surface. The shear strength is determined by the material cohesion, normal stress and internal friction angle.

[0020] The deflection instability criterion is constructed based on the principle of moment balance and is used to determine the risk of deflection instability by comparing the moment that causes the rock mass to deflect with the moment that resists the deflection.

[0021] As a preferred implementation, a swarm intelligence optimization algorithm is employed to optimize the parameters of the fault instability mechanical model with the objective of minimizing the deviation between the predicted results and the real-time state data. This optimization includes:

[0022] An optimization algorithm simulating the hunting behavior of Arctic foxes is used as the swarm intelligence optimization algorithm.

[0023] A fitness function is constructed based on the measured values ​​in the real-time state data and the predicted values ​​output by the fault instability mechanical model.

[0024] With the goal of minimizing the fitness function, the optimization algorithm is driven to optimize the basic parameters of the fault instability mechanical model.

[0025] In a preferred embodiment, the hybrid prediction model includes at least an input module, a feature processing module, a first prediction branch, a second prediction branch, and a fusion module, wherein:

[0026] The input module is used to receive the standardized dataset;

[0027] The first prediction branch outputs a first safety factor prediction value based on the optimized fault instability mechanical model.

[0028] The second prediction branch, based on the pre-trained neural network model, outputs a second safety factor prediction value;

[0029] The fusion module is equipped with an adaptive weighting unit, which is used to dynamically generate weighting coefficients and to weight and fuse the first safety coefficient prediction value and the second safety coefficient prediction value according to the weighting coefficients to generate the final comprehensive safety coefficient.

[0030] In a preferred embodiment, the first safety factor prediction value and the second safety factor prediction value are weighted and fused according to the weighting coefficient to generate the final comprehensive safety factor, specifically using the following formula:

[0031] ;

[0032] in, For the final overall safety factor, The first predicted safety factor value. This is the predicted value of the second safety factor. The weight coefficients output by the adaptive weighting unit.

[0033] In a preferred embodiment, the step of outputting a predicted first safety factor based on the optimized fault instability mechanical model includes:

[0034] Based on the slip instability criterion in the optimized fault instability mechanical model, and combined with the shear stress and normal stress of the fault plane in the standardized dataset, the anti-slip safety factor of the fault plane is calculated.

[0035] Based on the deflection instability criterion in the optimized fault instability mechanical model, and combining the deflection instability moment and the resistance moment, the deflection safety factor of the rock mass is calculated.

[0036] The minimum value between the anti-skid safety factor and the deflection safety factor is taken as the predicted value of the first safety factor.

[0037] As a preferred implementation, a tiered early warning mechanism is established based on the comprehensive safety factor and its dynamic change trend, and corresponding early warning information and handling suggestions are output, including:

[0038] Calculate the rate of change of the comprehensive safety factor at the current moment based on the comprehensive safety factor at two consecutive calculation times;

[0039] Based on the current comprehensive safety factor and its rate of change, the current warning level is determined by matching it with a preset warning threshold system. The preset warning threshold system includes three levels: safety, warning, and emergency warning.

[0040] Based on the warning level, warning information is automatically generated and output, wherein the warning information includes the warning level, risk description and handling suggestions.

[0041] In a preferred embodiment, the emergency warning level is triggered by any of the following conditions:

[0042] The overall safety factor is lower than the first preset threshold; or the growth rate of the shear stress at the fault plane obtained by real-time calculation exceeds the second preset threshold.

[0043] Specifically, when an emergency warning level is triggered solely because the rate of increase of shear stress at the fracture surface exceeds a second preset threshold, a data quality review process is automatically executed, and manual intervention is prompted for confirmation.

[0044] The first preset threshold is determined based on geological stability specifications and historical data.

[0045] Secondly, embodiments of the present invention also provide an AI fault instability mechanical model construction and dynamic early warning system that integrates multi-source data, including:

[0046] The data acquisition and preprocessing module is used to collect multi-source monitoring data in the fault-affected area and preprocess the data to obtain a standardized dataset. The multi-source monitoring data includes at least real-time status data and basic mechanical parameters.

[0047] The fault instability mechanical model construction module is used to construct a fault instability mechanical model based on mechanical theory. The fault instability mechanical model includes slip instability criteria and deflection instability criteria.

[0048] The fault instability mechanical model optimization module is used to optimize the parameters of the fault instability mechanical model by employing a swarm intelligence optimization algorithm with the goal of minimizing the deviation between the predicted results of the fault instability mechanical model and the real-time state data.

[0049] The hybrid prediction model building module is used to construct a hybrid prediction model based on the optimized fault instability mechanical model and the pre-trained neural network model through adaptive weights.

[0050] The comprehensive safety factor calculation module is used to input the standardized dataset into the hybrid prediction model and output a comprehensive safety factor that reflects the fault stability.

[0051] The dynamic early warning module is used to establish a graded early warning mechanism and output corresponding early warning information and handling suggestions based on the comprehensive safety factor and its dynamic change trend.

[0052] Thirdly, embodiments of the present invention also provide an electronic device, the electronic device comprising:

[0053] One or more processors;

[0054] Storage device for storing one or more programs;

[0055] When the one or more programs are executed by the one or more processors, the one or more processors implement the AI ​​fault instability mechanical model construction and dynamic early warning method that integrates multi-source data as described in any embodiment of the present invention.

[0056] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the AI ​​fault instability mechanical model construction and dynamic early warning method for fusing multi-source data as described in any embodiment of the present invention.

[0057] Compared with existing technologies, the present invention achieves the following beneficial effects:

[0058] (1) This invention integrates real-time status data and basic mechanical parameters to achieve comprehensive coverage of data dimensions and avoid the limitations of single data types; it adopts the interquartile range (IQR) criterion to identify outliers and establishes a dual-channel processing mechanism to ensure data quality and effectively retain instability precursor signals, providing accurate and comprehensive data support for subsequent model construction and early warning, and solving the problem of loss of precursor signals caused by simple removal of outliers in traditional technology.

[0059] (2) This invention adopts the Arctic fox optimization algorithm (PFA) to simulate the natural hunting behavior of Arctic foxes. Through group grouping, multi-stage search, leader incentive and mutation mechanism, it has stronger global exploration ability and local refinement ability compared with traditional algorithms such as genetic algorithm (GA) and particle swarm optimization (PSO), with faster convergence speed and higher optimization accuracy. It can efficiently and adaptively optimize the core parameters of the mechanical model (cohesion, internal friction angle, bulk modulus and shear modulus) and dynamically update the parameters to adapt to the changes in geological conditions during the mining process, thus solving the pain point of weak generalization ability of traditional fixed parameter models.

[0060] (3) Based on the Mohr-Coulomb strength theory and the fault instability characteristics of thick and hard roof mining areas, this invention constructs a sliding-deflection dual-criteria intelligent mechanical model, integrates the sliding instability criterion and the deflection instability criterion, comprehensively describes the multi-mode mechanical mechanism of fault instability, and makes up for the limitation of traditional technology that a single criterion cannot fully cover the unique instability mode of thick and hard roof mining areas.

[0061] (4) The mechanical model optimized by the Arctic Fox Optimization Algorithm (PFA) is combined with the adaptive weighted neural network to dynamically balance the theoretical reliability of the physical model and the generalization ability of the data model. It can not only rely on mechanical theory to ensure the scientific nature of the prediction, but also capture complex nonlinear relationships through data learning. The prediction mean absolute error (MAE≤0.05) and coefficient of determination (R²≥0.92) are significantly better than the traditional single model.

[0062] (5) This invention scientifically calibrates the three-level early warning thresholds, optimizes the early warning logic by combining industry standards, numerical simulation results and historical instability cases, and effectively reduces the false alarm and missed alarm rates. At the same time, it outputs immediate and operable disposal suggestions for different early warning levels, which solves the problems of lack of basis for thresholds, single logic and lack of practicality of disposal suggestions in traditional early warning technology, and provides more reliable decision support for the prevention and control of geological disasters in underground engineering. Attached Figure Description

[0063] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings. The drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0064] Figure 1 This is a flowchart of the AI ​​fault instability mechanical model construction and dynamic early warning method that integrates multi-source data provided in this embodiment of the invention;

[0065] Figure 2 This is a flowchart of the hierarchical early warning decision-making process provided in an embodiment of the present invention;

[0066] Figure 3 This is a schematic diagram illustrating the relationship between fault dip angle and bedrock movement angle provided in an embodiment of the present invention;

[0067] Figure 4 This is a schematic diagram of a rock strata movement model provided in an embodiment of the present invention;

[0068] Figure 5 This is a schematic diagram of the normal stress in a fault zone provided in an embodiment of the present invention;

[0069] Figure 6 This is a schematic diagram of the structure of the AI ​​fault instability mechanical model construction and dynamic early warning system that integrates multi-source data provided in this embodiment of the invention;

[0070] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0071] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0072] Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe operations (or steps) as being processed sequentially, many of these operations (or steps) may be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations may be rearranged. The process may be terminated when its operation is completed, but may also have additional steps not included in the figures. The process may correspond to a method, function, procedure, subroutine, subroutine, etc.

[0073] Example 1

[0074] like Figure 1 The diagram shows a flowchart of the AI ​​fault instability mechanical model construction and dynamic early warning method 100 that integrates multi-source data provided in Embodiment 1 of the present invention. The method 100 specifically includes the following steps:

[0075] S110: Collect multi-source monitoring data in the fault-affected area and preprocess the data to obtain a standardized dataset, wherein the multi-source monitoring data includes at least real-time status data and basic mechanical parameters.

[0076] As a preferred embodiment, combined with Figure 3 As shown, stress sensors, displacement gauges, groundwater monitors, and mining parameter acquisition terminals are strategically deployed at key cross-sections, protective coal pillars, and critical monitoring points in the fault-affected area. This enables synchronous and real-time acquisition of multi-dimensional data, ensuring that the data covers the core influencing factors related to fault instability. The acquired data is divided into two main categories: real-time status data and basic mechanical parameters, as detailed below:

[0077] Real-time status data: shear stress at the fault plane Normal stress at the fracture surface Horizontal displacement of surrounding rock Vertical displacement of surrounding rock Groundwater pressure Humidity of surrounding rock Surrounding rock temperature Working face advance speed Volume of goaf , width of protective coal pillar Overburden thickness Rock strata movement angle Fault dip angle Key layer thickness Direct top height Coal seam thickness ;

[0078] Basic mechanical parameters: cohesion internal friction angle Rock mass compressive strength ,tensile strength bulk modulus shear modulus Rock layer density Friction coefficient between key strata and adjacent rock strata Initial nominal values ​​were obtained through preliminary geological surveys and sampling, laboratory rock sample testing, and large-scale field tests (such as water pressure tests and in-situ deformation tests) and used as variables to be optimized in the model.

[0079] The above collected data is timestamped to the second. and three-dimensional spatial coordinates It is indexed twice and stored as a structured dataset, forming a multi-source fused original dataset. The format of a single data entry is:

[0080]

[0081] The meanings of each parameter are consistent with the aforementioned data details. The subscript "0" indicates the initial nominal value corresponding to the parameter. It fully covers all core parameters related to real-time operating conditions and basic mechanical properties, providing comprehensive input for subsequent model calculations.

[0082] As a preferred embodiment, to ensure data quality and model input adaptability, preprocessing is performed according to the following steps to output a standardized dataset. .

[0083] Data standardization: The Z-score standardization method is used to uniformly process all real-time state data and initial values ​​of basic mechanical parameters, completely eliminating the dimensional differences between different parameters and ensuring the consistency of model training and calculation.

[0084] Spatiotemporal alignment: To address the issue of missing data, missing data is supplemented by linear interpolation based on timestamps to ensure the continuity of the time dimension; spatial coordinate matching is used to match the location consistency of data collected by different terminals to ensure that multi-source data at the same monitoring point correspond one-to-one, thereby improving the spatial consistency of data.

[0085] Outlier Handling: Outlier data is identified using the interquartile range (IQR) criterion, and the definition is... ( It is the first quartile. (second quartile), exceeding Data within the specified range are marked as "suspected outliers"; a dual-channel processing mechanism is established, where normal data is directly input into the model, while suspected outliers are marked separately and their original characteristics are preserved. After being input into the model, targeted stability analysis is performed, without direct smooth replacement, to avoid losing the precursor signals of fault instability.

[0086] S120: Construct a fault instability mechanical model based on mechanical theory. The fault instability mechanical model includes slip instability criteria and deflection instability criteria.

[0087] As a preferred embodiment, combined with Figure 4 Based on the Mohr-Coulomb strength theory and combined with the fault instability characteristics of thick and hard roof mining areas, a fault instability mechanical model based on slip instability criteria and deflection instability criteria is constructed.

[0088] The slip instability criterion is as follows:

[0089] ;

[0090] in, Shear stress at the fault plane, The normal stress of the fault plane reflects the anti-slip stability of the fault plane. : Internal friction angle (basic mechanical parameter, variable to be optimized).

[0091] Among them, the deflection instability criterion is based on the moment balance equation, defining the deflection instability moment. With resistance moment ,when At that time, the rock mass underwent deflection and instability, among which, the deflection and instability moment The calculation formula and parameter mapping relationship are as follows:

[0092]

[0093] in, : Uniformly distributed load on the overlying strata (mechanical parameter, calculated from the unit weight of the overlying strata and the thickness of the overlying strata). : BF side length of rock mass at the end of key layer (geometric parameter, derived from the horizontal projection distance of the A end of key layer, the thickness of key layer, and the rock layer movement angle). : Length of the rock mass FA at the end of the key layer : The weight of the rock mass at the end of the key layer (calculated from the rock mass density and the volume of the key layer). The equivalent length of the rock mass moment at the end of the key layer (derived from the horizontal span, average height and rock layer movement angle of the key layer). : Horizontal tensile force component of key rock strata (calculated by the tensile strength of key strata and the angle of rock strata movement). : Length of the rock mass BA at the end of the key layer (equal to the horizontal projection distance of the A end of the key layer). Vertical pressure at the fault plane : Length of the BC side of the rock mass at the end of the key layer : The horizontal distance from the point of application of vertical pressure on the fault plane to the center of rotation.

[0094] Among them, the mechanical parameters are: ( The density of the overlying rock strata. Density of rock strata (Acceleration due to gravity) ( (for the tensile strength of the key layer) (Normal stress at the fault plane) (Heavy weight of rock at the end of the key layer); among which , These are the horizontal projection distances from the movement lines of the key strata at ends A and P to the fault plane, calculated based on the relative position of the working face and the fault, and the angle of strata movement. Calculated;

[0095] Geometric parameters: ; ; ; , , , .

[0096] in, Average height of rock mass at the end of key strata. The horizontal span of the key layer is derived from the width of the goaf.

[0097] Resistance moment A simplified calculation formula based on the deflection mechanics model of the key layer of a thick, hard roof is adopted. It is suitable for scenarios where one end of the key layer is cut by a fault and the other end is constrained by adjacent rock layers.

[0098] As a preferred embodiment, the basic mechanical parameters As variables to be optimized, a nonlinear mapping relationship between parameters and multi-source data is constructed: The parameter values ​​are dynamically updated by employing a swarm intelligence optimization algorithm.

[0099] in, Cohesion : Angle of internal friction Bulk modulus Shear modulus : Nonlinear mapping function : Preprocessed dataset.

[0100] S130: Using a swarm intelligence optimization algorithm, the parameters of the fault instability mechanical model are optimized with the goal of minimizing the deviation between the predicted results of the fault instability mechanical model and the real-time state data.

[0101] As a preferred embodiment, the Polar Fox Optimization Algorithm (PFA) is used as a swarm intelligence optimization algorithm. Its core objective is to minimize the deviation between the predicted results of the fault instability mechanical model and the measured real-time state data, so as to achieve efficient adaptive optimization of the core parameters of the model and ensure that the model is accurately adapted to the dynamically changing geological conditions.

[0102] The Polar Fox Optimization Algorithm (PFA) is a swarm intelligence optimization algorithm that simulates the hunting behavior of Arctic foxes. Its core logic originates from the efficient foraging strategies of Arctic foxes in harsh environments. By simulating the natural behavior of "random exploration - target location - precise approximation - leaping hunt," it achieves global optimization of the problem. The algorithm allocates the probability proportions of the exploration and development phases using random variables and balances global search and local refinement capabilities using dynamically decaying variables to avoid getting trapped in local optima. Its core mechanisms include: expanding the search range through random movement in the random exploration phase; calculating the difference between the target location phase and the optimal solution based on the signal propagation distance in the target location phase; gradually narrowing the search radius by locking the direction in the precise approximation phase; and obtaining the optimal solution through directional leaps in the leaping hunt phase. It features both fast convergence speed and high optimization accuracy, making it suitable for parameter optimization of complex nonlinear systems.

[0103] To adapt to the parameter optimization requirements of fault mechanics models, the Arctic Fox Optimization Algorithm (PFA) is configured as follows:

[0104] Grouping Design: Four functionally differentiated group groups (free search, leader-dependent, experience-dependent, and diligent hunter) are used to balance exploration and development capabilities through division of labor and cooperation. The initial behavior matrix of each group is as follows:

[0105] , ;

[0106] , ;

[0107] in, : Initial behavior matrix of the free search group (containing 5 parameters, corresponding to the maximum experimental power factor, leader power factor, experimental power increment coefficient, leader power increment coefficient, and maximum power correction coefficient in sequence). Initial behavior matrix of leader-dependent groups (containing 5 parameters, with the same meaning as above) ), Initial behavior matrix of experience-dependent groups (containing 5 parameters, with the same meaning as above) ), Initial behavior matrix of the diligent hunting group (containing 5 parameters, with the same meaning as above) ).

[0108] Key parameter settings: Population size Maximum number of iterations , variable factor Leader motivation threshold .

[0109] The multi-stage search mechanism includes the following stages:

[0110] Experience based on phases: by updating jump power Jump direction Optimize search paths using historical experience;

[0111] Leadership based on stage: through the leader power factor Control the intensity of individual following of the leader to quickly approach the optimal solution;

[0112] Leader incentive phase: When the leader incentive threshold is met, the positions of some individuals are randomly updated to break through the local optimum;

[0113] Mutation phase: Replace inferior solutions in the population, maintain population vitality, and ensure global optimization capabilities.

[0114] In a preferred embodiment, the PFA algorithm aims to minimize the deviation between the model prediction results and the measured values, constructs a fitness function, and dynamically adjusts the weight coefficients according to the sensor accuracy. The fitness function is specifically expressed as follows:

[0115] ;

[0116] in, Number of data samples Shear stress error weighting coefficient : No. Predicted shear stress values ​​at the fracture surface of each sample. : No. Measured values ​​of cross-sectional shear stress for each sample. Normal stress error weighting coefficient : No. Predicted normal stress values ​​at the fault plane for each sample. : No. Measured values ​​of normal stress at the fracture surface of each sample. Displacement error weighting coefficient : No. Predicted surrounding rock displacement values ​​for each sample. : No. Measured values ​​of surrounding rock displacement for each sample.

[0117] The weighting coefficient adjustment rules are as follows: The initial values ​​are all If the historical measurement error (RMSE) of the stress sensor is twice that of the displacement sensor, then adjust accordingly. Alternatively, the weight combination can be optimized through cross-validation.

[0118] As a preferred embodiment, optimizing the parameters of the above-mentioned fault instability mechanical model includes the following steps:

[0119] Initialize the PFA population, with each individual corresponding to a set of parameter solutions. The value range is expanded based on the initial nominal value. ;

[0120] For each set of parameters, solve the fault instability mechanical model constructed in step S120 above, calculate the predicted stress / displacement corresponding to each set of parameters, and combine the real-time measured data to calculate the Fit value of each solution through the fitness function above.

[0121] The population is iteratively updated through a multi-stage search mechanism of the PFA algorithm (experience search → leader following → incentive mutation → fatigue simulation);

[0122] Iterate to or Stop at time (adapted to geological data) (Inherent errors of left and right), output the optimal parameter solution. This completes the adaptive optimization of the model.

[0123] S140: Based on the optimized fault instability mechanical model and the pre-trained neural network model, a hybrid prediction model is constructed through adaptive weights.

[0124] As a preferred embodiment, this embodiment is based on an optimized fault instability mechanical model and a pre-trained neural network model. A hybrid prediction model is constructed through adaptive weights. The core objective is to integrate the theoretical reliability of the physical model with the nonlinear fitting capability of the neural network to accurately output a comprehensive safety factor that reflects the overall stability of the fault. The hybrid prediction model specifically includes the following modules or layers connected in sequence:

[0125] Input layer: Used to receive standardized datasets This dataset contains multidimensional real-time state data and fundamental mechanical parameters, including the number of neurons. ;

[0126] Hidden layers: Connected after the input layer, these layers perform non-linear transformations and deep feature extraction on the input data. In a specific configuration, this module consists of three fully connected layers with 64, 32, and 16 neurons respectively, using ReLU as the activation function to alleviate the vanishing gradient problem and improve the model's representational ability.

[0127] Parallel prediction module: Receives the output features from the hidden layer and divides them into two independent prediction branches:

[0128] First prediction branch: Based on the optimized fault instability mechanical model, calculates and outputs the first safety factor prediction value according to the input features. ;

[0129] The second prediction branch: Based on the pre-trained neural network model, it calculates and outputs a predicted value for the second security factor according to the input features. ;

[0130] Adaptive fusion layer: Equipped with a single-neuron weight learner for dynamically generating weight coefficients. (scope ), and based on the weighting coefficient, predict the first safety factor value. Compared with the second safety factor prediction value Weighted fusion is performed to generate the final comprehensive safety coefficient. The input data feature vector is then used as the output dynamic weights. The fusion formula is as follows:

[0131] ;

[0132] Output layer: Connected after the adaptive fusion layer, in a specific configuration: 1 neuron, using... (Scaling factor 4) maps the output to the [0,4] interval, adapting to a higher range of safety factors; in the training data The scaling process is performed synchronously on the labels to avoid errors in the loss function calculation; for Even in an extremely stable state, the original values ​​are retained for training, allowing the model to learn its feature differences.

[0133] Through the above design, the hybrid prediction model can dynamically balance physical mechanisms and data patterns to achieve accurate and robust assessment of fault stability.

[0134] As a preferred embodiment, the training steps of the hybrid prediction model are as follows:

[0135] Training data: Historical fault monitoring data from more than three working faces were selected, covering three scenarios: stable, early warning, and emergency early warning. The training set and test set were split in an 8:2 ratio. The training set was used for model parameter learning, and the test set was used for generalization capability verification. The above data sources include geological survey reports of thick and hard roof mining areas, laboratory rock sample test data, and field stress / displacement monitoring records.

[0136] Core training parameters: The loss function uses mean squared error (MSE).

[0137] ;

[0138] Using the Adam optimizer, the learning rate batch size Training epochs = 200; L2 regularization strategy is used (weight decay coefficient). ).

[0139] Model validation: Calculate the mean absolute error (MAE) using the test set. ), coefficient of determination ( This ensures the accuracy of model predictions; additional verification of abnormal data processing effects ensures that no precursory signals of instability are missed.

[0140] S150: Input the standardized dataset into the hybrid prediction model and output a comprehensive safety factor that reflects fault stability.

[0141] As a preferred embodiment, based on the description of each module of the hybrid prediction model in step S140 above, after the hybrid prediction model completes the deep feature extraction of the data in the hidden layer, it generates a comprehensive safety coefficient through dual-branch parallel computation and dynamic weighted fusion. The steps are as follows:

[0142] First prediction branch: Based on the optimized fault instability mechanical model, calculates and outputs the predicted value of the first safety factor according to the input characteristics. ;

[0143] The second prediction branch: Based on a pre-trained neural network model, it calculates and outputs a predicted value for the second safety factor according to the input features. ;

[0144] Next, the first safety factor prediction value is calculated in the adaptive fusion layer. Second safety factor prediction value Weighted fusion is performed to generate the final comprehensive safety coefficient. The fusion formula is as follows:

[0145] .

[0146] Among them, the predicted value of the first safety factor mentioned above Obtain it through the following steps:

[0147] Based on the slip instability criterion in the optimized fault instability mechanical model, and combined with the shear stress and normal stress of the fault plane in the standardized dataset, the anti-slip safety factor of the fault plane is calculated: ;

[0148] Based on the deflection instability criterion in the optimized fault instability mechanical model, and combining the deflection instability moment and the resisting moment, the deflection safety factor of the rock mass is calculated:

[0149] ;

[0150] The minimum value between the anti-skid safety factor and the deflection safety factor is taken as the predicted value of the first safety factor: .

[0151] As a preferred embodiment, the above-mentioned hybrid prediction model outputs a comprehensive safety factor once per second. And the individual safety factors, and simultaneously calculate the rate of change of the overall safety factor. and the rate of increase of shear stress at the fault plane Among them, the rate of change of the comprehensive safety factor The formula is as follows:

[0152] ;

[0153] in, The change in the safety factor over three consecutive calculation periods. : Calculation cycle interval (in this embodiment, it is fixed at 3s, that is, the time difference of 3 consecutive calculation cycles).

[0154] S160: Based on the comprehensive safety factor and its dynamic change trend, establish a graded early warning mechanism and output corresponding early warning information and handling suggestions.

[0155] As a preferred embodiment, combined with Figure 2 As shown, this embodiment constructs a three-level early warning mechanism with static thresholds and dynamic trends to achieve accurate early warning of fault instability risks and output actionable handling suggestions. The specific design is as follows:

[0156] The preset threshold system calibration references relevant specifications for fault stability in the "Coal Mine Safety Regulations," combined with statistical analysis of historical instability cases in thick and hard roof mining areas. Numerical simulation results (safety factor range under different steady-state conditions) calibration, specifically: critical value of the rate of change of the safety factor: Critical value for shear stress growth rate: =0.1MPa / h; Cumulative reduction threshold of safety factor: ≥0.05.

[0157] Level 3 Early Warning Judgment Rules:

[0158] Security level: ,and (If the safety factor does not decrease significantly), output "safe" information and simultaneously push a mechanical stability report;

[0159] Warning level: and (Safety factor continues to decline), or But three consecutive cycles And the cumulative decline (Dimensionless safety factor change), outputs "early warning" information, along with stress / displacement evolution curves and trend analysis;

[0160] Emergency Alert Level: ,or (Based on statistics of critical rates of instability in thick, hard-topped faults); when only due to When a single indicator triggers an emergency warning, the system will automatically initiate a data quality review process and prompt for manual intervention for confirmation; finally, it will output an "emergency warning" message, trigger an audible and visual alarm, and push immediate handling suggestions (immediately stop the work face advancement, evacuate personnel from dangerous areas, and increase the frequency of real-time monitoring of the fault zone).

[0161] The early warning information is output in a standardized structured format, including data acquisition time, spatial coordinates of the acquisition point, comprehensive safety factor, individual safety factors, and mechanical evolution trend. Structured information including change curves, warning levels, and corresponding handling recommendations.

[0162] Based on the above embodiments, the present invention achieves the following beneficial effects:

[0163] (1) This invention integrates real-time status data and basic mechanical parameters to achieve comprehensive coverage of data dimensions and avoid the limitations of single data types; it adopts the interquartile range (IQR) criterion to identify outliers and establishes a dual-channel processing mechanism to ensure data quality and effectively retain instability precursor signals, providing accurate and comprehensive data support for subsequent model construction and early warning, and solving the problem of loss of precursor signals caused by simple removal of outliers in traditional technology.

[0164] (2) This invention adopts the Arctic fox optimization algorithm (PFA) to simulate the natural hunting behavior of Arctic foxes. Through group grouping, multi-stage search, leader incentive and mutation mechanism, it has stronger global exploration ability and local refinement ability compared with traditional algorithms such as genetic algorithm (GA) and particle swarm optimization (PSO), with faster convergence speed and higher optimization accuracy. It can efficiently and adaptively optimize the core parameters of the mechanical model (cohesion, internal friction angle, bulk modulus and shear modulus) and dynamically update the parameters to adapt to the changes in geological conditions during the mining process, thus solving the pain point of weak generalization ability of traditional fixed parameter models.

[0165] (3) Based on the Mohr-Coulomb strength theory and the fault instability characteristics of thick and hard roof mining areas, this invention constructs a sliding-deflection dual-criteria intelligent mechanical model, integrates the sliding instability criterion and the deflection instability criterion, comprehensively describes the multi-mode mechanical mechanism of fault instability, and makes up for the limitation of traditional technology that a single criterion cannot fully cover the unique instability mode of thick and hard roof mining areas.

[0166] (4) This invention combines the mechanical model optimized by the Arctic Fox Optimization Algorithm (PFA) with an adaptive weighted neural network to dynamically balance the theoretical reliability of the physical model and the generalization ability of the data model. It can not only rely on mechanical theory to ensure the scientific nature of the prediction, but also capture complex nonlinear relationships through data learning. The prediction mean absolute error (MAE≤0.05) and coefficient of determination (R²≥0.92) are significantly better than the traditional single model.

[0167] (5) This invention scientifically calibrates the three-level early warning thresholds, optimizes the early warning logic by combining industry standards, numerical simulation results and historical instability cases, and effectively reduces the false alarm and missed alarm rates. At the same time, it outputs immediate and operable disposal suggestions for different early warning levels, which solves the problems of lack of basis for thresholds, single logic and lack of practicality of disposal suggestions in traditional early warning technology, and provides more reliable decision support for the prevention and control of geological disasters in underground engineering.

[0168] Example 2

[0169] Figure 6 This is a schematic diagram of the structure of an AI fault instability mechanical model construction and dynamic early warning system that integrates multi-source data, as provided in Embodiment 2 of the present invention. Figure 6 As shown, the system includes:

[0170] The data acquisition and preprocessing module 610 is used to collect multi-source monitoring data in the fault-affected area and preprocess the data to obtain a standardized dataset. The multi-source monitoring data includes at least real-time status data and basic mechanical parameters.

[0171] The fault instability mechanical model construction module 620 is used to construct a fault instability mechanical model based on mechanical theory. The fault instability mechanical model includes slip instability criteria and deflection instability criteria.

[0172] The fault instability mechanical model optimization module 630 is used to optimize the parameters of the fault instability mechanical model by employing a swarm intelligence optimization algorithm with the goal of minimizing the deviation between the predicted results of the fault instability mechanical model and the real-time state data.

[0173] The hybrid prediction model building module 640 is used to build a hybrid prediction model based on the optimized fault instability mechanical model and the pre-trained neural network model through adaptive weights.

[0174] The comprehensive safety factor calculation module 650 is used to input the standardized dataset into the hybrid prediction model and output a comprehensive safety factor that reflects the fault stability.

[0175] The dynamic early warning module 660 is used to establish a graded early warning mechanism and output corresponding early warning information and handling suggestions based on the comprehensive safety coefficient and its dynamic change trend.

[0176] The AI ​​fault instability mechanical model construction and dynamic early warning system that integrates multi-source data provided in this embodiment of the invention can execute the AI ​​fault instability mechanical model construction and dynamic early warning method that integrates multi-source data provided in any of the above embodiments of the invention. It has the corresponding functions and beneficial effects of executing the AI ​​fault instability mechanical model construction and dynamic early warning method that integrates multi-source data. For detailed process, please refer to the relevant operations of the AI ​​fault instability mechanical model construction and dynamic early warning method that integrates multi-source data in the foregoing embodiments.

[0177] Example 3

[0178] Figure 7 This is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. The electronic device 10 is intended to represent various forms of digital computers, and may also represent various forms of mobile devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the invention described and / or claimed herein.

[0179] like Figure 7 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded into the RAM 13 from storage unit 18. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0180] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0181] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, central processing unit (CPU), graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 11 executes the AI ​​fault instability mechanics model construction and dynamic early warning method fused with multi-source data described above.

[0182] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0183] The above embodiments are merely illustrative examples and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. An AI fault instability mechanics model construction and dynamic early warning method for fusing multi-source data, characterized in that, include: Multi-source monitoring data is collected in the fault-affected area and the data is preprocessed to obtain a standardized dataset. The multi-source monitoring data includes at least real-time status data and basic mechanical parameters. A fault instability mechanical model is constructed based on mechanical theory. The fault instability mechanical model includes slip instability criteria and deflection instability criteria. A swarm intelligence optimization algorithm is used to optimize the parameters of the fault instability mechanical model with the goal of minimizing the deviation between the predicted results of the fault instability mechanical model and the real-time state data. Based on the optimized fault instability mechanical model and the pre-trained neural network model, a hybrid prediction model is constructed through adaptive weights; The standardized dataset is input into the hybrid prediction model, and a comprehensive safety factor reflecting fault stability is output. Based on the comprehensive safety factor and its dynamic change trend, a graded early warning mechanism is established and corresponding early warning information and handling suggestions are output.

2. The method of claim 1, wherein, Multi-source monitoring data were collected in the fault-affected area, and the data were preprocessed to obtain a standardized dataset, including: The collected multi-source monitoring data is indexed by timestamps and spatial coordinates to generate a structured dataset; The structured dataset is subjected to data standardization, spatiotemporal alignment, and outlier handling to obtain a standardized dataset.

3. The method of claim 1, wherein, The fault instability mechanical model includes slip instability criteria and deflection instability criteria, wherein, The slip instability criterion is constructed based on the Mohr-Coulomb strength theory and is used to determine the risk of slip instability by comparing the shear stress and shear strength at the fracture surface. The shear strength is determined by the material cohesion, normal stress and internal friction angle. The deflection instability criterion is constructed based on the principle of moment balance and is used to determine the risk of deflection instability by comparing the moment that causes the rock mass to deflect with the moment that resists the deflection.

4. The method of claim 3, wherein, A swarm intelligence optimization algorithm is employed to optimize the parameters of the fault instability mechanical model, with the objective of minimizing the deviation between the predicted results and real-time state data. This optimization includes: An optimization algorithm simulating the hunting behavior of Arctic foxes is used as the swarm intelligence optimization algorithm. A fitness function is constructed based on the measured values ​​in the real-time state data and the predicted values ​​output by the fault instability mechanical model. With the goal of minimizing the fitness function, the optimization algorithm is driven to optimize the basic parameters of the fault instability mechanical model.

5. The method of claim 1, wherein, The hybrid prediction model includes at least an input module, a first prediction branch, a second prediction branch, and a fusion module, wherein: The input module is used to receive the standardized dataset; The first prediction branch outputs a first safety factor prediction value based on the optimized fault instability mechanical model. The second prediction branch, based on the pre-trained neural network model, outputs a second safety factor prediction value; The fusion module is equipped with an adaptive weighting unit, which is used to dynamically generate weighting coefficients and to weight and fuse the first safety coefficient prediction value and the second safety coefficient prediction value according to the weighting coefficients to generate the final comprehensive safety coefficient.

6. The method according to claim 5, characterized in that, The first safety factor prediction value and the second safety factor prediction value are weighted and fused according to the weighting coefficient to generate the final comprehensive safety factor, specifically using the following formula: ； in, For the final overall safety factor, The first predicted safety factor value. This is the predicted value of the second safety factor. The weight coefficients output by the adaptive weighting unit.

7. The method according to claim 5, characterized in that, The first safety factor prediction value is output based on the optimized fault instability mechanical model, including: Based on the slip instability criterion in the optimized fault instability mechanical model, and combined with the shear stress and normal stress of the fault plane in the standardized dataset, the anti-slip safety factor of the fault plane is calculated. Based on the deflection instability criterion in the optimized fault instability mechanical model, and combining the deflection instability moment and the resistance moment, the deflection safety factor of the rock mass is calculated. The minimum value between the anti-skid safety factor and the deflection safety factor is taken as the predicted value of the first safety factor.

8. The method according to claim 1, characterized in that, Based on the comprehensive safety factor and its dynamic trend, a tiered early warning mechanism is established and corresponding early warning information and handling suggestions are output, including: Calculate the rate of change of the comprehensive safety factor at the current moment based on the comprehensive safety factor at two consecutive calculation times; Based on the current comprehensive safety factor and its rate of change, the current warning level is determined by matching it with a preset warning threshold system. The preset warning threshold system includes three levels: safety, warning, and emergency warning. Based on the warning level, warning information is automatically generated and output, wherein the warning information includes the warning level, risk description and handling suggestions.

9. The method according to claim 8, characterized in that, The emergency warning level is triggered by any of the following conditions: The overall safety factor is lower than the first preset threshold; or the growth rate of the shear stress at the fault plane obtained by real-time calculation exceeds the second preset threshold. Specifically, when an emergency warning level is triggered solely because the rate of increase of shear stress at the fracture surface exceeds a second preset threshold, a data quality review process is automatically executed, and manual intervention is prompted for confirmation. The first preset threshold is determined based on geological stability specifications and historical data.

10. A system for constructing and dynamically warning of an AI-based fault instability mechanical model that integrates multi-source data, characterized in that: include: The data acquisition and preprocessing module is used to collect multi-source monitoring data in the fault-affected area and preprocess the data to obtain a standardized dataset. The multi-source monitoring data includes at least real-time status data and basic mechanical parameters. The fault instability mechanical model construction module is used to construct a fault instability mechanical model based on mechanical theory. The fault instability mechanical model includes slip instability criteria and deflection instability criteria. The fault instability mechanical model optimization module is used to optimize the parameters of the fault instability mechanical model by employing a swarm intelligence optimization algorithm with the goal of minimizing the deviation between the predicted results of the fault instability mechanical model and the real-time state data. The hybrid prediction model building module is used to construct a hybrid prediction model based on the optimized fault instability mechanical model and the pre-trained neural network model through adaptive weights. The comprehensive safety factor calculation module is used to input the standardized dataset into the hybrid prediction model and output a comprehensive safety factor that reflects the fault stability. The dynamic early warning module is used to establish a graded early warning mechanism and output corresponding early warning information and handling suggestions based on the comprehensive safety factor and its dynamic change trend.

Images