Mine water disaster intelligent alarm system responding to multidimensional physical field parameter abnormity

Abstract

The invention relates to the technical field of mine safety monitoring and geophysical exploration, in particular to a mine water disaster intelligent alarm system responding to multi-dimensional physical field parameter abnormity, which comprises a full-space data acquisition module for acquiring monitoring area observation data and calling a background geological physical model; the ideal reference reconstruction module is used for constructing an ideal physical field reference in a non-abnormal state; the double difference extraction module is used for calculating an observation residual error of observation data relative to a reference and a theoretical residual error of simulation response relative to the reference; the inversion coupling verification module is used for driving the theoretical residual error to approach to the observation residual error and extracting parameter sensitivity characteristics; the intelligent alarm judgment module is used for analyzing sensitivity characteristics and convergence states; generating an alarm instruction if the feature is determined to be the entity fluid abnormal feature; if it is determined that the target abnormal interference feature is not the target abnormal interference feature, signal suppression processing is executed; according to the invention, the false alarm rate of mine water disaster detection is greatly reduced.

Description

Technical Field

[0001] This invention relates to the fields of mine safety monitoring and geophysical exploration technology, specifically to an intelligent alarm system for mine water hazards that responds to anomalies in multidimensional physical field parameters. Background Technology

[0002] In the scenario of advanced water hazard detection in coal mine roadway excavation or mining face, intelligent alarm system relies on accurate multi-dimensional physical field observation data to ensure operational safety and avoid geological risks. Underground detection equipment usually needs to combine data collected by sensor array with background geological model to perceive potential water inrush structures ahead in real time.

[0003] For intelligent identification of mine water hazards, existing solutions generally adopt a direct inversion imaging architecture based on raw data. This involves acquiring electromagnetic or DC electric field signals through sensors placed on the tunnel walls, and using geophysical inversion algorithms to directly reconstruct the electrical distribution of the underground medium. Highly conductive areas are directly identified as water hazard targets to correct the safety warning model. Although this solution is feasible in an ideal environment with a uniform background field, it directly processes raw data containing a large amount of environmental noise and lacks differential constraints based on an ideal physical field. When encountering complex mine environments, the inversion algorithm is prone to misidentifying the high resistance effect of tunnel cavities, the low resistance interference of metal supports, or random geological noise as abnormal features of solid fluids, leading to erroneous alarm commands. Furthermore, direct inversion without sensitivity feature verification struggles to eliminate false artifacts caused by the mathematical properties of the algorithm, leading to false alarms or invalid calculations due to pathological oscillations during the alarm decision process. This makes it difficult to support high-confidence, interference-resistant real-time risk assessment for mine water hazard monitoring. Therefore, establishing an intelligent alarm mechanism with ideal benchmark reconstruction and dual differential verification capabilities, effectively eliminating static environmental interference and algorithm artifacts while improving the ability to capture weak water signals and alarm accuracy, has become an urgent technical problem to be solved. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides an intelligent alarm system for mine water hazards that responds to anomalies in multidimensional physical field parameters. Specifically, the technical solution of this invention includes: The full-space data acquisition module is used to acquire multi-dimensional physical field observation data of the monitoring area and retrieve the preset background geophysical model; The ideal benchmark reconstruction module is used to perform forward modeling based on the background geophysical model and construct an ideal physical field benchmark for the monitoring area under anomaly-free conditions. The dual difference extraction module is used to initialize the forward model based on preset fluid anomaly body parameters, generate simulation response containing anomalies, and calculate the observation residuals of multidimensional physical field observation data relative to the ideal physical field benchmark, as well as the theoretical residuals of simulation response containing anomalies relative to the ideal physical field benchmark. The inversion coupling verification module is used to drive the theoretical residual to approximate the observation residual based on the Levenberg-Marquardt algorithm. During the iteration process, the Jacobian matrix is ​​generated, and the singular value decomposition of the Jacobian matrix is ​​performed to extract parameter sensitivity features. The intelligent alarm decision module is used to analyze parameter sensitivity characteristics and iterative convergence status. If it is determined to be an abnormal feature of the physical fluid, an alarm command is generated; if it is determined to be a non-target abnormal interference feature, signal suppression processing is performed.

[0005] Preferably, the method for constructing an ideal physical field benchmark for a monitoring area under anomaly-free conditions includes: extracting channel or cavity geometric parameters and media average physical property parameters from the background geophysical model; substituting the channel or cavity geometric parameters and media average physical property parameters into the physical field control equations, performing forward numerical simulation under anomaly-free conditions in the full-space coordinate system, and generating full-space benchmark field data; and using the full-space benchmark field data as the ideal physical field benchmark.

[0006] Preferably, the method for generating simulation responses containing anomalies includes: defining fluid anomaly parameters, which include spatial location vectors, geometric feature parameters, and electrical parameters; constructing a full-space spherical conductive fluid model based on the fluid anomaly parameters; performing coupled calculations based on the full-space spherical conductive fluid model and the background geophysical model to obtain the simulation physical field response under the current parameter state, and using the simulation physical field response as the simulation response containing anomalies.

[0007] Preferably, the method for calculating the observation residual and the theoretical residual includes: performing vector difference operation between the multidimensional physical field observation data and the ideal physical field reference to extract the observation residual containing potential anomalous signals and environmental noise; performing vector difference operation between the simulation response containing anomalous signals and the ideal physical field reference to extract the theoretical anomalous signals generated only by the full-space spherical conductive fluid model as the theoretical residual.

[0008] Preferably, the method for driving the theoretical residuals to approximate the observed residuals includes: constructing an inversion objective function with the goal of minimizing the sum of squared residuals; introducing a damping factor and constructing a system of linear equations containing the Jacobian matrix, the damping identity matrix, and parameter corrections; solving the system of linear equations to obtain the parameter corrections and using the parameter corrections to update the fluid anomaly parameters; regenerating the theoretical residuals based on the updated fluid anomaly parameters and determining whether the inversion objective function satisfies the preset convergence conditions; if not, adjusting the damping factor and proceeding to the next iteration.

[0009] Preferably, the method for extracting parameter sensitivity features by performing singular value decomposition on the Jacobian matrix includes: extracting the current Jacobian matrix in real time during the iteration process; performing singular value decomposition on the Jacobian matrix to obtain a singular value matrix and eigenvectors; constructing a sensitivity analysis matrix based on the singular values, and mapping the principal singular values ​​to the control strength index of the fluid anomaly parameters on the observed data.

[0010] Preferably, the method for determining anomalies in physical fluids includes: detecting whether the inversion objective function converges to a preset residual threshold range; if converged, checking whether the principal singular values ​​in the sensitivity analysis matrix exhibit stable high sensitivity characteristics, and determining whether the finally converged fluid anomaly parameters are within a preset reasonable physical space range; if convergence, high sensitivity characteristics, and a reasonable physical space range are simultaneously satisfied, then the anomaly in the multidimensional physical field observation data is determined to be caused by a real conductive fluid, and is confirmed as anomaly in physical fluids.

[0011] Preferably, the method for determining non-target abnormal interference features includes: if the inversion objective function cannot converge within a preset maximum number of iterations; or if the sensitivity analysis matrix shows that the Jacobian matrix is ​​ill-conditioned and the principal singular values ​​exhibit low sensitivity or irregular oscillation characteristics; then the anomaly in the multidimensional physical field observation data is determined to be a false signal caused by environmental electromagnetic interference or algorithm artifacts, confirmed as a non-target abnormal interference feature, and alarm commands are prohibited from being generated.

[0012] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention effectively solves the problem that the original data is easily affected by environmental noise when directly inverted by existing technologies by constructing an ideal benchmark reconstruction and a dual difference extraction mechanism. Unlike directly processing the original signal containing a lot of noise, this solution constructs an ideal physical field benchmark under anomaly-free conditions based on the background geological and physical model. It uses vector difference operations to remove the cavity effect of the tunnel and the static background interference of metal facilities, so that the system only processes the residual signal containing potential anomalies, which significantly improves the signal-to-noise ratio and the ability to capture weak water signals. 2. This invention introduces an inversion coupling verification and Jacobi matrix sensitivity analysis strategy, which significantly reduces the false alarm rate caused by algorithm artifacts. By performing singular value decomposition on the Jacobi matrix during the iteration process and combining it with a dual weighting system to eliminate the dimensional differences between multidimensional data and model parameters, this method can distinguish between real anomalies caused by real conductive fluids and false anomalies caused by ill-conditioned oscillations of the algorithm from mathematical characteristics, providing a rigorous basis for the true and false judgment of mine water hazard detection. 3. This invention employs a parameterized spherical conductive fluid model and the Levenberg-Marquardt adaptive iterative technique to achieve efficient convergence and stability in inversion calculations. By reducing the complex voxel inversion to the optimization of a few geometric and physical parameters, and using a set of normal equations containing damping factors to dynamically adjust the parameter update step size, this method ensures that the model parameters are continuously differentiable with respect to grid properties while taking into account the stability of the gradient descent method and the speed of the Gauss-Newton method, thus meeting the stringent timeliness requirements of downhole real-time alarms. 4. This invention establishes an intelligent alarm judgment and signal suppression mechanism for multidimensional physical field parameter anomalies, giving the system extremely high alarm confidence and anti-interference capability. The system not only makes judgments based on residual convergence and sensitivity characteristics, but also verifies the spatial location rationality of the anomaly through coordinate projection transformation. It can automatically identify and suppress non-target interference characteristics caused by cable leakage or metal mesh support, ensuring that only physical fluid anomalies that meet physical space constraints and have continuous signal-to-noise ratio support trigger alarms, effectively avoiding uncertainties in geological risk assessment. Attached Figure Description

[0013] The present invention will be further explained below with reference to the accompanying drawings and embodiments: Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0015] Example 1: Please see Figure 1 A mine water hazard intelligent alarm system responding to anomalies in multidimensional physical field parameters, comprising: The full-space data acquisition module is used to acquire multi-dimensional physical field observation data of the monitoring area and retrieve the preset background geophysical model; The ideal benchmark reconstruction module is used to perform forward modeling based on the background geophysical model and construct an ideal physical field benchmark for the monitoring area under anomaly-free conditions. The dual difference extraction module is used to initialize the forward model based on preset fluid anomaly body parameters, generate simulation response containing anomalies, and calculate the observation residuals of multidimensional physical field observation data relative to the ideal physical field benchmark, as well as the theoretical residuals of simulation response containing anomalies relative to the ideal physical field benchmark. The inversion coupling verification module is used to drive the theoretical residual to approximate the observation residual based on the Levenberg-Marquardt algorithm. During the iteration process, the Jacobian matrix is ​​generated, and the singular value decomposition of the Jacobian matrix is ​​performed to extract parameter sensitivity features. The intelligent alarm decision module is used to analyze parameter sensitivity characteristics and iterative convergence status. If it is determined to be an abnormal feature of the physical fluid, an alarm command is generated; if it is determined to be a non-target abnormal interference feature, signal suppression processing is performed.

[0016] This embodiment details the overall architecture and operational logic of the system, which is primarily applied to advanced water hazard detection scenarios in coal mine roadway excavation or mining faces. The full-space data acquisition module uses a high-sensitivity sensor array, such as transient electromagnetic probes or DC electrode arrays, deployed on the roadway walls or face to acquire real-world physical field data and load prior geological knowledge to generate multi-dimensional physical field observation data. Simultaneously, it retrieves a background geophysical model from a database, containing the three-dimensional geometric structure of the roadway and the basic physical properties of the surrounding rock. The ideal benchmark reconstruction module establishes an absolutely secure reference system. Based on the background geophysical model, it performs forward modeling using the finite element method or finite difference method to construct an ideal physical field benchmark for the monitoring area under anomaly-free conditions. This benchmark represents the theoretical signal that the sensors should receive without water hazard. The dual-difference extraction module removes background interference, initializes the forward model based on preset fluid anomaly parameters, generates a simulation response containing anomalies, and calculates the observation residuals of the observation data relative to the benchmark and the theoretical residuals of the simulation response relative to the benchmark, respectively. Based on this, the inversion coupling verification module verifies the authenticity of the anomaly through iterative optimization. It drives the theoretical residual to approximate the observed residual using the Levenberg-Marquardt algorithm. The specific implementation of this driving process is as follows: An objective function representing the Euclidean distance between the theoretical and observed residuals is constructed. An iterative strategy combining gradient descent and Gauss-Newton methods is used to calculate the parameter update step size by solving a set of normal equations containing damping factors, thereby continuously reducing the theoretical residual numerically and converging it to the observed residual. During the iteration process, the Jacobian matrix is ​​generated in real time and subjected to singular value decomposition to extract parameter sensitivity features. The intelligent alarm decision module outputs a decision based on the mathematical characteristics of the inversion. In response to the analysis results indicating a solid fluid anomaly (i.e., convergence and high sensitivity), the system generates an alarm command. In response to the determination of a non-target anomaly interference characteristic (i.e., divergence or pathological oscillation), the system performs signal suppression processing. This embodiment constructs a digital twin verification environment by introducing ideal benchmark reconstruction and a dual-difference mechanism. The system no longer directly processes the raw data containing a large amount of environmental noise, but processes the residuals. This design significantly improves the signal-to-noise ratio. Combined with subsequent Jacobian matrix sensitivity analysis, it can fundamentally distinguish between anomalies caused by real water bodies and false anomalies caused by algorithm artifacts or environmental interference, greatly reducing the false alarm rate of mine water hazard detection in complex geological noise backgrounds.

[0017] Example 2: Methods for constructing an ideal physical field benchmark for a monitoring area under anomaly-free conditions include: Extract the channel or cavity geometric parameters and the average physical property parameters of the medium from the background geophysical model; Substitute the channel or cavity geometric parameters and the average physical property parameters of the medium into the physical field control equations, and perform forward numerical simulation without anomalies in the full space coordinate system to generate full space reference field data. Use the full-space reference field data as the ideal physical field reference.

[0018] This embodiment details the specific steps for constructing an ideal physical field benchmark, aiming to eliminate the cavity effect in roadways and the static interference of metal facilities on detection signals. The system extracts key parameters of the background geophysical model from the mine geological database, including the geometric parameters of the channel or cavity that define the cross-sectional shape, size, and excavation depth of the roadway, as well as the average physical property parameters of the medium that define the electrical parameters of the surrounding rock under dry, water-free conditions. For electric field strength, Permeability; The background spatial conductivity distribution function; For Hamiltonian operators, The variable is time; this embodiment uses the all-space transient electromagnetic diffusion equation as the core operator:

[0019] To ensure its feasibility, this embodiment specifies its mathematical mapping rules: assuming the monitoring area is divided into a roadway spatial domain. With the surrounding rock medium ,but: in Assign the value to air conductivity, and take... To avoid matrix singularities, Obtained by interpolation of three-dimensional physical property data from the background geophysical model; To excite the source current density, its spatial distribution is defined as along the predetermined trajectory of the transmitting coil. The Dirac distribution; to achieve numerical solution by computer, this embodiment explicitly adopts the backward Eulerian difference scheme to discretize the time term: in, Using time-step indexing, continuous partial differential equations are transformed into discrete time-stepped linear equation systems:

[0020] in, Here is the curl stiffness matrix, and its elements are calculated using the following formula: , The generalized mass matrix contains conductivity weights, and its elements are... This definition ensures that the parameters of non-uniform media are correctly represented in the grid integral, avoiding the calculation error of non-uniform fields caused by the traditional method of extracting conductivity as a scalar. As the source term vector, in order to accurately simulate the actual coil excitation source in the mine and solve the problem of source term distribution in the finite element mesh, this embodiment uses its first term vector. Each component is strictly defined as following the trajectory of the transmitting coil. Line integral:

[0021] in, The number of coil turns. Explicitly defined as having linear shut-off edges The trapezoidal wave during the turn-off period This ensures the numerical boundedness and physical reality of the source energy input. The time step is defined as follows: In the full spatial coordinate system, which simultaneously includes the tunnel space and the surrounding rock space, the model without anomalies is meshed and numerically simulated. To address the characteristics of electromagnetic field vector fluctuations, this embodiment explicitly selects first-order edge elements as basis functions. Its mathematical definition is: Let the centroid coordinates of the vertices of the tetrahedral element be... For the i-th edge connecting vertices a and b, ,in, To eliminate spurious solutions, the electric field degrees of freedom are defined on the mesh edges to represent the length of the edge. In this numerical simulation, to accurately handle the infinite underground half-space and eliminate spurious reflections caused by artificially truncated boundaries, the system sets perfectly matched layer PML or infinite element boundary conditions at the truncated boundaries of the model. Simultaneously, adaptive unstructured tetrahedral meshing technology is used to discretize the computational domain, and local mesh refinement is performed on the tunnel surface where the sensor array is located and the expected front-line detection area. For example, the mesh cell size is set to 1 / 10 of the skin depth to ensure the accuracy of calculations for weak reference field signals. The physical field response at the preset sensor locations is calculated to generate full-space reference field data. This data is then solidified into an ideal physical field reference, serving as a zero-point reference for subsequent differential calculations. This embodiment constructs a full-space reference field to accurately quantify the inherent background response generated by the high-resistivity characteristics typically exhibited by roadway cavities and the low-resistivity characteristics typically exhibited by known metal facilities. In subsequent steps, by subtracting this reference from the observation data, the interference of static environmental factors can be completely eliminated, allowing weak water body anomaly signals to stand out in the complex mine background field.

[0022] Example 3: Methods for generating simulation responses containing anomalies include: Define fluid anomaly parameters, which include spatial position vector, geometric feature parameters, and electrical parameters; construct a full-space spherical conductive fluid model based on the fluid anomaly parameters; The simulation physical field response under the current parameter state is obtained by coupling the full-space spherical conductive fluid model with the background geophysical model, and the simulation physical field response is used as the simulation response with anomalies.

[0023] This embodiment details the specific method for generating simulation responses containing anomalies. To mathematically simulate the water hazard source, this embodiment employs a parameterized spherical model to represent the potential water inrush structure. Fluid anomaly parameters are defined, and a parameter vector is constructed that includes a spatial position vector containing the centroid coordinates of the anomaly, geometric characteristic parameters representing the equivalent radius of the spherical anomaly, and electrical parameters representing the conductivity of the fluid inside the anomaly. Based on the above parameters, a full-space spherical conductive fluid model is embedded into the background geological model; in the coupled calculation stage, the system performs explicit parameter-mesh mapping operations: Given a finite element mesh... The geometric center of each unit is The volume is Then the coupling conductivity of this unit Determined by the following step smoothing function:

[0024] in, This is the spatial position vector in the fluid anomaly body parameters. The specific method for obtaining this information is as follows: based on the center coordinates of the grid cells. Spatial interpolation indexing is performed in the background geophysical model retrieved by the full-space data acquisition module to extract the background electrical conductivity value of the surrounding rock at that location; if If located inside the cavity of the tunnel, then Take the air conductivity, such as ; It is a smoothing factor, and its physical dimension is the reciprocal of length. To address the issue that relying solely on the minimum grid size in the original definition might lead to the function degenerating into a sharp step function, this embodiment clarifies... The principle for selecting the value of is to balance the convergence of the inversion algorithm and the accuracy of the boundary representation: Definition ,in, The dimensionless constant used to control the steepness of the S-curve is taken in the range of 4 to 6, with a typical value of 5, corresponding to 99% of the conductivity variation range. For a smooth transition bandwidth in physical space; the key is, It is not simply determined by the minimum grid size, but is defined as:

[0025] in, To ensure the minimum number of coverage grids required for effective numerical gradient calculation, a value of 3 to 5 is chosen. This refers to the local mesh size around the anomaly. A preset physical transition lower limit, such as 0.5 meters, is used. This constraint ensures that even with local ultra-fine mesh refinement, the smoothing function will not degenerate into a non-differentiable step function, thus guaranteeing the continuous differentiability of the inversion objective function with respect to the model's geometric parameters, which is beneficial for iterative convergence. This mapping ensures that the model parameters are continuously differentiable with respect to mesh properties. Based on finite element discretization theory, stiffness matrices containing conductivity information are constructed respectively. With the mass matrix In the coupled computation phase, the system performs time-step iterative solutions, for each time step... Solve the following system of linear equations: ; This enables the coupled calculation of the full-space spherical conductive fluid model and the background geophysical model; the forward modeling algorithm is used to calculate the physical field response under the background plus spherical coupled model to obtain the simulated physical field response under the current parameter state; the response is output as a simulation response with anomalies for subsequent residual approximation calculation; This embodiment uses a parameterized spherical model instead of an arbitrary-shaped voxel model, which greatly reduces the number of unknowns in the inversion process, reducing thousands of grid parameters to a few geometric and physical parameters. This dimensionality reduction not only reduces computational complexity but also makes the subsequent Levenberg-Marquardt inversion algorithm more likely to converge, thus meeting the stringent timeliness requirements of real-time mine alarms.

[0026] Example 4: Methods for calculating observed residuals and theoretical residuals include: Vector difference operations are performed between multidimensional physical field observation data and ideal physical field benchmark to extract observation residuals containing potential anomalous signals and environmental noise; By performing vector difference operations between the simulation response containing the anomalies and the ideal physical field reference, the theoretical anomaly signal generated only by the full-space spherical conductive fluid model is extracted as the theoretical residual.

[0027] This embodiment details the specific method for calculating the observation residual and the theoretical residual. To construct the target difference quantity for inversion, this embodiment performs vector operations; the observation residual is extracted by performing vector difference operations between the multidimensional physical field observation data actually collected by the sensor and the ideal physical field reference, as shown in the following formula:

[0028] in, : Derived from the observation residual vector calculated in real time, its physical meaning is the signal in the real world that exceeds the background expectation; : The vector of observation data acquired by the sensor; : Derived from the baseline vector calculated in the previous step; Extract the theoretical residuals and perform vector difference between the simulation response with anomalies under the current parameters and the ideal physical field reference. The calculation formula is as follows:

[0029] in, : Derived from the theoretical residual vector of simulation calculation, its physical meaning is the pure anomalous signal that should be generated if the current water body model exists; : Derived from the simulation response containing anomalies calculated in the preceding steps; The dual-difference mechanism in this embodiment is a key preprocessing step. The observation residuals eliminate the background influence of the surrounding rock and tunnel, retaining only potential anomalies and noise, while the theoretical residuals represent pure theoretical anomalies. By driving the theoretical residuals to approximate the observation residuals, the system is actually looking for the spherical model that best explains the current observation residuals, rather than directly fitting the original data. This greatly improves the inversion's ability to capture weak signals and its accuracy.

[0030] Example 5: Methods that drive the approximation of theoretical residuals to observed residuals include: Construct an inversion objective function with the goal of minimizing the sum of squared residuals; By introducing a damping factor, a system of linear equations is constructed that includes the Jacobian matrix, the damping identity matrix, and parameter corrections. Solve the linear equation system to obtain the parameter correction amount, and use the parameter correction amount to update the fluid anomaly parameters; regenerate the theoretical residual based on the updated fluid anomaly parameters, and determine whether the inversion objective function satisfies the preset convergence condition; If the condition is not met, the damping factor is adjusted and the process proceeds to the next iteration.

[0031] This embodiment details the specific steps for handling nonlinear and ill-conditioned problems in inversion using the Levenberg-Marquardt algorithm; it constructs the inversion objective function, defining the objective function as the sum of squared residuals:

[0032] in, : Derived from the calculated objective function value, its physical meaning is the degree of fit between the current model and the observed data; : Fluid anomaly body parameter vector; To find the parameter correction, a system of linear equations including the damping factor is constructed during the iteration process:

[0033] in, The Jacobian matrix, derived from numerical differential calculations, physically represents the partial derivatives of the theoretical residuals with respect to each parameter. Here, we clarify the dimensional definitions of matrices and vectors: Let the number of effective observation data points acquired by the full-space data acquisition module be... That is, the dimension of the observed data, the number of fluid anomaly parameters to be inverted is That is, the dimension of the model parameters; at this point, the observed residual vector With the theoretical residual vector All Column vector, parameter adjustment for Column vectors; Jacobian matrix Then it is clearly defined as Matrix; specifically, the system constructs this matrix using the finite difference method, for parameters The Each component, its Jacobian element The calculation formula is:

[0034] in, To address the numerical differentiation distortion caused by the large range between spatial coordinates (meter-level) and conductivity parameters, which is a preset small perturbation, this embodiment employs an adaptive step size strategy:

[0035] Set relative disturbance coefficient Absolute Lower World , For the corresponding unit vector; The damping factor originates from the adaptive adjustment strategy, and is distinguished from the permeability in the physical model. Symbols are used here. Its initial value Usually set to The largest diagonal element This is multiplied by a factor of 1 to ensure the stability of the initial iteration; The identity matrix, whose dimensions are explicitly defined as follows: , and matrix product Consistency in dimensions ensures the formula Dimension matching for addition operations; : Parameter correction vector; Solving the above system of equations yields the parameter corrections, and the parameters are updated. Based on this, the theoretical residuals are regenerated using the updated parameters, and the new objective function value is calculated. It is then determined whether the inverted objective function satisfies the preset convergence conditions. If the conditions are met, the iteration stops. If the conditions are not met, the damping factor is adjusted according to the degree of decrease in the objective function, and the next iteration begins. The specific strategy for adjusting the damping factor is based on the gain ratio. The determination and calculation formula is as follows:

[0036] Wherein, the denominator term represents the decrease in the objective function predicted by the linear approximation model. Clearly defined as the theoretical residual function The first-order Taylor expansion is a linear approximation; if This indicates that the update is valid, and the parameter correction amount is accepted, while the damping factor is reduced. ,in, It is a constant greater than 1, for example, 2; if This indicates that the update is invalid, the parameter correction amount is rejected, and the damping factor is increased. This causes the algorithm to degenerate into gradient descent to ensure stability; This embodiment cleverly solves the singularity problem that may exist in the Jacobian matrix by introducing a damping factor, namely the computational collapse caused by the non-invertibility of the matrix. When the matrix is ​​close to ill-conditioned, the damping factor increases and the algorithm degenerates into gradient descent to ensure stability. When it is close to the extreme point, the damping factor decreases and the algorithm approximates the Gauss-Newton method to accelerate convergence. This mechanism ensures that the inversion process is both stable and fast in complex mining environments. Example 6: Methods for extracting parameter sensitivity features by performing singular value decomposition on the Jacobian matrix include: During the iteration process, the current Jacobian matrix is ​​extracted in real time; Singular value decomposition is performed on the Jacobian matrix to obtain the singular value matrix and eigenvectors; A sensitivity analysis matrix is ​​constructed based on singular values, and the principal singular values ​​are mapped as indicators of the control strength of fluid anomaly parameters on the observed data.

[0037] This embodiment details the method for extracting parameter sensitivity features. The Jacobian matrix implies the control capability of model parameters over observed data. During the convergence phase of the LM iteration, the current Jacobian matrix is ​​extracted in real time. To simultaneously eliminate dimensional differences in multidimensional observed data, such as mixed data of electric field V / m and magnetic field T, and differences in physical dimensions of model parameters, such as the difference between position m and conductivity S / m, and to ensure the objectivity and numerical stability of sensitivity analysis, the system constructs a dual weighting system. The first step is to construct the data weighting matrix:

[0038] in, For the first The measured noise standard deviation of each observation channel is derived from the background noise statistics of the full-space data acquisition module. The specific statistical method is as follows: during the silent time window before the system transmitter is turned on, continuous noise measurements are taken for each channel. For example, with 1024 samples, calculate the variance of the sampled sequence and take the square root. This is used to quantify the current level of electromagnetic interference. To prevent extremely small quantities with a denominator of zero, a parameter weighting matrix is ​​constructed. The weighting coefficient is defined as follows: , For the first The initial values ​​of the parameters, For example, the lower limit of the preset parameter benchmark. To avoid matrix singularities caused by initial parameters being zero, such as the origin; calculate the fully weighted dimensionless Jacobian matrix. For the matrix Perform singular value decomposition:

[0039] in, : The fully weighted dimensionless Jacobian matrix has the following dimensions. ,in For the dimensions of the observation data, For model parameter dimensions; : an orthogonal matrix, where for A left singular vector matrix of dimension for A right singular vector matrix of dimension; The singular value matrix derived from the decomposition calculation is a diagonal matrix with dimensions of . diagonal elements These are the weighted singular values, which characterize the sensitivity in the sense of signal-to-noise ratio; A sensitivity analysis matrix is ​​constructed based on singular values, and this matrix is ​​rigorously defined as a normalized resolution density matrix:

[0040] in, Principal singular values, Let be the numerical stability constant, and take . Simultaneously, the principal singular values ​​are mapped as an index of the control strength of fluid anomaly parameters on the observed data. ,because The noise normalization information is already included, and its calculation formula is simplified to: This indicator directly quantifies the intensity multiple of the abnormal signal relative to the background noise in the most sensitive direction, in decibels. This embodiment utilizes dual-weighted SVD technology, which not only extracts mathematical features but also achieves physical normalization, enabling sensitivity comparisons of different physical field components and parameters with different dimensions on the same scale. By extracting these features, the system can determine whether the currently inverted water model is an entity actually supported by the data or a false solution cobbled together by mathematical calculations, thus providing a mathematical basis for subsequent authenticity determination.

[0041] Example 7: Methods for identifying anomalous characteristics of solid fluids include: Check whether the inversion objective function converges to the preset residual threshold range; If convergence is achieved, check whether the principal singular values ​​in the sensitivity analysis matrix exhibit stable high sensitivity characteristics, and determine whether the finally converged fluid anomaly parameters are within the preset reasonable physical space range. If the convergence and high sensitivity characteristics are met simultaneously and the data is located within a reasonable physical space, then the anomaly in the multidimensional physical field observation data is determined to be caused by a real conductive fluid, and is confirmed as an anomaly characteristic of a physical fluid.

[0042] This embodiment details the triple verification logic for determining abnormal characteristics of a physical fluid; it performs a convergence verification to check whether the inversion objective function converges to a preset residual threshold range to confirm the success of the mathematical fitting; the preset residual threshold... Based on the standard deviation of ambient background noise Dynamic setting, usually taken The number of observed data points is the expected weighted sum of squared residuals, which is close to the degrees of freedom of the data. Sensitivity feature verification is performed to check whether the principal singular values ​​in the sensitivity analysis matrix exhibit stable high sensitivity characteristics. The specific judgment logic is as follows: the system introduces a length of... The iterative sliding window requires control strength indices for all iteration steps within the window. All are greater than the preset threshold ,For example And the relative standard deviation of the principal singular value within the window is less than This logic is used to exclude false positives caused by transient sensitivity spikes due to numerical differentiation crossing grid boundaries, ensuring that only anomalies supported by a sustained signal-to-noise ratio are recognized. Perform a physical space rationality check to determine whether the parameters of the finally converged fluid anomaly are within the preset reasonable physical space range; based on the fluid anomaly position vector output by the inversion. Defined in the full-space absolute coordinate system of the background geological model, while a reasonable range is defined in the local coordinate system with the tunneling face as the reference, the system needs to perform an explicit coordinate projection transformation: obtaining the absolute coordinates of the center of the current tunneling face. and the unit vector of the tunneling axis direction ; Calculate the displacement vector of the anomaly relative to the working surface. ; Calculate the longitudinal projection distance and lateral deviation distance ; The pre-defined reasonable physical space range is parameterized and constrained as follows: in, To effectively advance the detection range, such as 100m, For the effective field of view half angle, such as Furthermore, physical constraints are imposed on the radius of the anomalous body, for example... In response to simultaneously satisfying convergence and high sensitivity characteristics and being located within a reasonable physical space range, the system determines that the observed anomaly is caused by a real conductive fluid, confirms it as a physical fluid anomaly, and triggers an alarm. The judgment logic established in this embodiment constitutes the core risk control mechanism. By introducing coordinate projection transformation, it rigorously solves the mapping problem between the absolute coordinates output by the algorithm and the relative position of the engineering site. It effectively eliminates situations where the fit is good but physically impossible, such as when the abnormal body is located in the mined-out area behind. It also eliminates situations where there is an anomaly but the signal is extremely weak and unreliable. This stringent judgment standard ensures that the alarm signal has a very high degree of confidence, avoiding panic and unnecessary work stoppages caused by false alarms.

[0043] Example 8: Methods for identifying non-target anomalous interference features include: If the inversion objective function fails to converge within the preset maximum number of iterations; or if the sensitivity analysis matrix shows that the Jacobian matrix is ​​ill-conditioned and the principal singular values ​​exhibit low sensitivity or irregular oscillation characteristics; If the anomaly in the multidimensional physical field observation data is determined to be a false signal caused by environmental electromagnetic interference or algorithm artifacts, it is confirmed as a non-target anomaly interference feature, and alarm commands are prohibited from being generated.

[0044] This embodiment details the suppression logic for common false anomalies in mines; it performs divergence suppression judgment, responding to the fact that the inversion objective function cannot be reduced below the threshold within the preset maximum number of iterations, indicating that the existing single-sphere model cannot explain the observation data and may be complex distributed noise, which is judged as interference; it performs ill-conditioned suppression judgment, responding to the sensitivity analysis matrix showing that the Jacobian matrix is ​​ill-conditioned, i.e., the calculated condition number If the principal singular value exhibits low sensitivity, or if the parameters show irregular large oscillations at the end of the iteration (i.e., the variance of the parameter update in the last 5 iterations exceeds 100 times the preset convergence tolerance), the system determines that the anomaly in the observed data is caused by environmental electromagnetic interference, such as false signals caused by cable leakage, metal mesh support coupling, or algorithm artifacts. Once a non-target abnormal interference feature is confirmed, the system will forcibly prohibit the generation of alarm commands and mark it as an interference signal on the human-machine interface for technical personnel to refer to. This embodiment effectively filters out non-water-related interference by identifying ill-conditioned features during the inversion process. For example, metal supports often generate strong secondary fields with spatial shapes that do not conform to the spherical pattern, which can lead to non-convergence or abnormal singular values ​​in the inversion. This system utilizes this characteristic to automatically classify such interference as non-target features, thereby achieving intelligent signal purification in the complex downhole electromagnetic environment and ensuring the continuity of production.

[0045] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A mine water disaster intelligent alarm system responding to multi-dimensional physical field parameter anomaly, characterized in that, The method comprises the following steps: a full-space data acquisition module is used to acquire multi-dimensional physical field observation data of a monitoring area and to call a preset background geophysical model; an ideal reference reconstruction module is used to perform forward calculation based on the background geophysical model and to construct an ideal physical field reference of the monitoring area in an abnormal state; a double-difference extraction module is used to initialize a forward model based on preset fluid anomaly body parameters, to generate an abnormal simulation response, and to calculate observation residuals of the multi-dimensional physical field observation data relative to the ideal physical field reference and theoretical residuals of the abnormal simulation response relative to the ideal physical field reference; an inversion coupling verification module is used to drive the theoretical residuals to approximate the observation residuals based on the Levenberg-Marquardt algorithm, to generate a Jacobian matrix in an iterative process, and to perform singular value decomposition on the Jacobian matrix to extract parameter sensitivity characteristics; an intelligent alarm decision module is used to analyze the parameter sensitivity characteristics and the iterative convergence state, to generate an alarm instruction if it is determined that the entity fluid anomaly characteristics are present, and to perform signal suppression processing if it is determined that non-target abnormal interference characteristics are present.

2. The intelligent mine water disaster alarm system responding to multi-dimensional physical field parameter anomaly according to claim 1, characterized in that, The method for constructing the ideal physical field reference of the monitoring area in the abnormal state comprises the following steps: channel or cavity geometric parameters and medium average physical property parameters in the background geophysical model are extracted; the channel or cavity geometric parameters and the medium average physical property parameters are substituted into a physical field control equation, forward numerical simulation in an abnormal state is performed in a full-space coordinate system, and full-space reference field data are generated; the full-space reference field data are used as the ideal physical field reference.

3. The intelligent mine water disaster alarm system responding to multi-dimensional physical field parameter anomaly according to claim 2, characterized in that, The method for generating the abnormal simulation response comprises the following steps: fluid anomaly body parameters are defined, the fluid anomaly body parameters include a spatial position vector, geometric characteristic parameters and electrical parameters; a full-space spherical conductive fluid model is constructed based on the fluid anomaly body parameters; coupling calculation is performed based on the full-space spherical conductive fluid model and the background geophysical model, simulation physical field responses in a current parameter state are acquired, and the simulation physical field responses are used as the abnormal simulation response.

4. The intelligent mine water disaster alarm system responding to multi-dimensional physical field parameter anomaly according to claim 3, characterized in that, The calculation of the observation residuals and the theoretical residuals comprises the following steps: vector difference operation is performed on the multi-dimensional physical field observation data and the ideal physical field reference, and observation residuals containing potential abnormal signals and environmental noise are extracted; vector difference operation is performed on the abnormal simulation response and the ideal physical field reference, and theoretical abnormal signals generated only by the full-space spherical conductive fluid model are extracted as the theoretical residuals.

5. The intelligent mine water disaster alarm system responding to multi-dimensional physical field parameter anomaly according to claim 4, characterized in that, The method for driving the theoretical residuals to approximate the observation residuals comprises the following steps: an inversion objective function is constructed to minimize the residual square sum; a damping factor is introduced, a linear equation set containing the Jacobian matrix, a damping unit matrix and a parameter correction amount is constructed; the linear equation set is solved to acquire the parameter correction amount, the fluid anomaly body parameters are updated by using the parameter correction amount; the theoretical residuals are regenerated based on the updated fluid anomaly body parameters, and it is determined whether the inversion objective function meets a preset convergence condition; if not, the damping factor is adjusted and the next round of iteration is entered.

6. The intelligent mine water disaster alarm system responding to multi-dimensional physical field parameter anomaly according to claim 5, characterized in that, The method for singular value decomposition of the Jacobian matrix to extract parameter sensitivity characteristics comprises: extracting the current Jacobian matrix in real time during the iteration process; singular value decomposition of the Jacobian matrix to obtain a singular value matrix and an eigenvector; constructing a sensitivity analysis matrix based on the singular values, and mapping the dominant singular value to an index of the control strength of the fluid anomaly body parameters on the observation data.

7. The intelligent mine water disaster alarm system responding to multi-dimensional physical field parameter anomaly according to claim 6, characterized in that, The method for determining the entity fluid anomaly characteristics comprises: detecting whether the inversion objective function converges to a preset residual threshold range; if the convergence is detected, checking whether the dominant singular value in the sensitivity analysis matrix shows a stable high sensitivity characteristic, and determining whether the finally converged fluid anomaly body parameters are located within a preset reasonable physical space range; if the convergence, high sensitivity characteristic and reasonable physical space range are simultaneously satisfied, determining that the anomaly in the multi-dimensional physical field observation data is caused by a conductive fluid actually existing, and confirming the entity fluid anomaly characteristics.

8. The intelligent mine water disaster alarm system responding to multi-dimensional physical field parameter anomaly according to claim 7, characterized in that, The method for determining the non-target anomaly interference characteristics comprises: if the inversion objective function cannot converge within a preset maximum number of iterations; or if the sensitivity analysis matrix shows that the Jacobian matrix is in a sick state, and the dominant singular value shows a low sensitivity or irregular oscillation characteristic; determining that the anomaly in the multi-dimensional physical field observation data is a false signal caused by environmental electromagnetic interference or algorithm artifacts, confirming the non-target anomaly interference characteristics, and prohibiting the generation of an alarm instruction.

Images