A method, system, and storage medium for multi-parameter collaborative inversion of mine wastewater transport channels based on dynamic step size adjustment.

Abstract

This invention discloses a multi-parameter collaborative inversion method, system, and storage medium for mine wastewater transport channels based on dynamic step-size adjustment. The method includes: collecting geophysical data reflecting multiple physical properties of the underground medium in the detection area of ​​hidden wastewater transport channels in abandoned mines; constructing a standardized multi-parameter feature dataset through location matching; establishing a multi-parameter joint inversion framework; establishing a comprehensive objective function that integrates data fitting accuracy, parameter coordination constraints, and numerical stability; and introducing cross-gradient constraints to achieve spatial structure correlation of the physical property parameter model; using an improved gradient descent method for iterative solution; constructing a dynamic step-size adjustment mechanism that couples first-order iterative information with second-order function characteristics; adjusting the step-size in stages; combining cluster analysis to remove invalid and abnormal data; iterating until convergence conditions are met; optimizing the iterative effect in stages; and finally obtaining a multi-parameter fusion inversion model that fits the actual geological characteristics. This provides technical support for the detection of hidden wastewater transport channels in abandoned mines.

Description

Technical Field

[0001] This invention belongs to the field of mining geological exploration technology, and more specifically, relates to a multi-parameter collaborative inversion method, system and storage medium for mine sewage transport channels based on dynamic step size adjustment. Background Technology

[0002] Detecting hidden pollution transport channels in abandoned mines is a core challenge in mine ecological restoration and environmental protection. After mining operations cease, underground goafs, tunnels, and fault zones easily form hidden pollution transport channels. Pollutants spread and migrate through these channels, not only polluting the soil and groundwater environment but also triggering potential geological disasters, severely hindering the mine's ecological restoration process and posing a potential threat to the surrounding ecosystem and human health. Therefore, locating, spatially characterizing, and analyzing the physical properties of pollution transport channels is a prerequisite for ensuring the effectiveness of mine environmental remediation.

[0003] Current methods for detecting mine wastewater transport channels primarily employ single-parameter inversion techniques, relying on a single conductivity or structural integrity parameter. This approach fails to comprehensively reflect the combined anomalies in the underground media caused by the wastewater transport channel. Single-parameter inversion is susceptible to geological interference, resulting in multiple interpretations and an inability to distinguish between the wastewater transport channel and surrounding normal geological bodies, leading to significant errors in determining the channel's spatial location and connectivity. While some multi-parameter inversion techniques incorporate various physical property parameters, they lack effective parameter coordination and constraint mechanisms, resulting in poor spatial structural consistency among the parameter models and further exacerbating the uncertainty of the inversion results.

[0004] Meanwhile, existing inversion methods mostly employ fixed-step iterative solutions. The step size cannot adapt to the changing characteristics and local morphology of the objective function during the iteration process, resulting in slow convergence speed, a tendency to fall into local optima, and difficulty in balancing inversion accuracy and computational efficiency. Interference from invalid and outlier data also affects the stability of the iterative solution, leading to significant deviations between the final inversion model and actual geological features, thus failing to provide reliable data support for the remediation of pollution transport channels. In summary, there is an urgent need for a technical method that can achieve multi-parameter collaborative constraints, adaptively adjust the iteration step size, and improve inversion accuracy. This would address the shortcomings of existing detection technologies, provide technical support for the detection of hidden pollution transport channels in abandoned mines, and promote high-quality mine ecological restoration. Summary of the Invention

[0005] This invention aims to solve the technical problems of multiple solutions in multi-parameter inversion, rigid iteration step size control, and large deviation between inversion models and actual geological features in the detection of hidden sewage transport channels in abandoned mines. It provides an accurate and efficient multi-parameter collaborative inversion method to improve the inversion accuracy of the spatial morphology and physical properties of sewage transport channels, and to provide reliable technical support for mine ecological restoration and environmental governance.

[0006] To address the aforementioned deficiencies or improvement needs of existing technologies, as a first aspect of this invention, the present invention provides a multi-parameter collaborative inversion method for mine wastewater transport channels based on dynamic step size adjustment, comprising:

[0007] S1. In the detection area of ​​the hidden sewage transport channel in the abandoned mine, multi-dimensional geophysical data reflecting the conductivity, structural integrity and electrochemical characteristics of the underground medium are collected simultaneously. At the same time, the location matching of the multi-dimensional geophysical data is completed, and a standardized multi-parameter feature dataset is constructed.

[0008] S2. Build a multi-parameter joint inversion framework, establish a comprehensive objective function that integrates data fitting accuracy, parameter coordination constraints and numerical stability, introduce cross gradient constraints to associate the spatial distribution characteristics of different physical property parameters, and force the models corresponding to each parameter to maintain spatial structure consistency.

[0009] S3. Taking the objective function as the solution object, the improved gradient descent method is used to start the iterative operation. A dynamic step size adjustment mechanism based on the coupling of first-order iterative information and second-order function characteristics is constructed to complete the adaptive control of the step size. Among them, a mathematical adjustment model for the iterative step size is set. The first-order iterative information represents the iterative trend, and the second-order function characteristics represent the local shape of the objective function. A weight allocation mechanism is introduced to adaptively balance the control effect of the two types of information on the step size.

[0010] S4. Based on the dynamic step size adjustment mechanism of dual-factor coupling, the step size is adaptively adjusted in stages according to the iteration process to carry out the iterative solution of the objective function; at the same time, combined with multi-parameter feature clustering analysis to remove invalid abnormal data, the iteration continues until the preset convergence condition is met, and finally a multi-parameter fusion inversion model that fits the actual geological characteristics is obtained.

[0011] In the iterative solution process, based on the first-order iterative information and the characteristics of the second-order function, the step size and the control priority of the two types of information are dynamically adjusted according to the iterative trend and the changes in the local shape of the objective function. This achieves phased optimization of the convergence speed in the early stage of iteration, the balance between efficiency and accuracy in the middle stage, and the convergence stability in the later stage, thus approximating the global optimal solution.

[0012] Furthermore, the standardized multi-parameter feature dataset in S1 includes: a set of quantitative parameters reflecting the conductivity of the underground medium, a set of physical feature parameters characterizing the structural integrity of the underground geological body, a set of polarization response parameters reflecting the electrochemical properties of the solid-liquid interface of the underground medium, and a set of spatial location coordinate parameters corresponding to each parameter.

[0013] Each parameter set has been standardized to unify the units and data range, and all parameters are matched with spatial location coordinate parameters to achieve normalized correspondence of multi-dimensional geophysical parameters in the spatial dimension.

[0014] Furthermore, the comprehensive objective function in S2 is specifically as follows:

[0015] The comprehensive objective function is constructed based on the fitting accuracy of multi-dimensional geophysical data, the coordinated constraints of multi-parameter space, and the stability of numerical solutions. It forms a unified solution objective by coupling data fitting terms, cross-gradient constraint terms, and regularization terms, and its expression is:

[0016]

[0017] in, For the comprehensive objective function, The multi-parameter model vector represents the spatial distribution set of parameters characterizing the conductivity, structural integrity, and solid-liquid interface electrochemical properties of the underground medium. It corresponds to the standardized multi-parameter feature dataset collected and processed in the detection area, ensuring the compatibility between the objective function and the input data.

[0018] For the data fitting term, defined as , Define the spatial domain for detecting concealed sewage transport channels in abandoned mines. The inversion model is based on parameter vectors The calculated simulated values ​​of the geophysical response, For measured multi-parameter geophysical data after location matching and standardization, through The integral form minimizes the spatial global deviation between the simulated geophysical response and the measured multi-parameter geophysical data, ensuring that the inversion results fit the actual mining exploration scenario.

[0019] For the cross gradient constraint term, For spatial gradient operators, For tensor product operations, For models with different physical property parameters, the spatial gradient correlation function is used.

[0020] For regularization terms, For the Laplace operator, By adjusting the parameter vector The second-order spatial derivative integral suppresses numerical oscillations caused by abrupt changes in the physical properties of underground geological bodies in the mine during the iterative solution process, thereby improving the stability of the objective function solution.

[0021] Furthermore, the spatial gradient correlation function of the different physical property parameter models is defined as:

[0022]

[0023] in, , These are two different types of physical property parameters, corresponding to two types of parameters in the underground medium's conductivity, structural integrity, and solid-liquid interface electrochemical properties. The spatial structure association of different physical property parameter models is realized through gradient tensor coupling, which forces the boundary morphology and spatial distribution of the models corresponding to each parameter to remain consistent.

[0024] Furthermore, the specific method for representing the iterative trend through first-order iterative information in S3 is as follows:

[0025] Using the iterative residual change of the objective function as the core carrier of first-order iterative information, the iterative trend is quantified by defining the difference in objective function values ​​between two adjacent iterations; let the... The objective function value of the next iteration is , No. The objective function value of the next iteration is ,in For the first The multi-parameter model vector of the next iteration. For the first The multi-parameter model vector of the next iteration, both of which correspond to the spatial distribution characteristics of the physical properties of the underground medium;

[0026] By calculating the change in iterative residual And further transformed into a normalized residual change index. To characterize the completed iterative trend;

[0027] The numerical value directly reflects the rate of change of the objective function during the iteration process. When the value is large, it indicates that the objective function value changes drastically, the iteration is in a rapid descent phase, and the corresponding iteration trend is an efficient convergence trend; when When the value is small, it indicates that the objective function value changes slowly, the iteration approaches the convergence threshold, and the corresponding iteration trend is a slow convergence trend;

[0028] when Abnormal fluctuations can indicate the presence of data interference or local optimum traps during the iteration process.

[0029] Furthermore, the method for characterizing the local morphology of the objective function using second-order function features in S3 is as follows:

[0030] The Hessian matrix of the objective function is used as the carrier of the second-order function characteristics. The local shape of the objective function at the current iteration point is quantitatively characterized by the characteristic parameters of the Hessian matrix, reflecting the function curvature change and extreme value distribution characteristics.

[0031] Regarding the first The multi-parameter model vector of the next iteration Construct the objective function exist Hessian matrix at the location , Let be a square matrix composed of second-order partial derivatives, whose elements are defined as follows: ,in , The first Sub-iteration multi-parameter model vector The first in The, the Each physical property parameter component corresponds to the spatial distribution parameters of the conductivity, structural integrity, and solid-liquid interface electrochemical properties of the underground medium.

[0032] By calculating the Hessian matrix traces Using the eigenvalue spectrum, we complete the representation of the local form of the objective function: the trace of the Hessian matrix. The sum of the elements on the main diagonal of the matrix directly reflects the overall curvature of the objective function at the current iteration point. The larger the value, the steeper the local curvature of the objective function, the more drastic the change in the function shape near the corresponding iteration point, and the higher the probability of the existence of local extrema;

[0033] The eigenvalue spectrum of the Hessian matrix determines the type of local extrema of the objective function by the sign and distribution of the eigenvalues. When all eigenvalues ​​are positive, it indicates that the current iteration point is in the local minimum region of the objective function. When the eigenvalues ​​alternate between positive and negative, it indicates that the current iteration point is in the saddle point region.

[0034] Furthermore, the weight allocation mechanism in S3 is specifically as follows:

[0035] definition The adjustment weights for first-order iterative information, The adjustment weights for the characteristics of the second-order function satisfy... Dynamic allocation is achieved by constructing an adaptive weight mapping function, the specific expression of which is as follows:

[0036]

[0037]

[0038] In the formula, This is an index of normalized residual change. The iterative trend baseline threshold is determined by the overall accuracy level of the multi-parameter feature dataset. Hessian matrix traces ; This is the reference value for the preset Hessian matrix trace;

[0039] In the early stages of iteration, and , The value approaches 0. The value is relatively small. Increased proportion The proportion is reduced, which strengthens the control of the step size by the characteristics of the second-order function, adapts to the state of rapid function descent and gentle curvature, and improves the convergence speed.

[0040] Mid-iteration and The two types of information are balanced and adapted through a weight mapping function. and Synchronous dynamic fine-tuning balances convergence speed and inversion accuracy;

[0041] Later stages of iteration and , The value approaches 1. The value is relatively large. Increased proportion The proportion is reduced, the control of the step size by the first-order iterative information is strengthened, and numerical oscillations are suppressed to approach the global optimal solution.

[0042] Furthermore, the preset convergence condition in S4 is specifically as follows:

[0043] The degree of inversion approximation is characterized by the change magnitude of the objective function in adjacent iterations, the stability of the inversion result is characterized by the spatial distribution difference of the multi-parameter model vector, and the termination boundary of the operation is defined by the preset maximum number of iterations. The iteration terminates when any one of the three judgment dimensions meets the preset requirements.

[0044] As a second aspect of the present invention, a multi-parameter collaborative inversion system for mine wastewater transport channels based on dynamic step size adjustment is also provided, comprising:

[0045] The dataset construction unit is used to simultaneously collect multi-dimensional geophysical data reflecting the conductivity, structural integrity and solid-liquid interface electrochemical properties of underground media in the detection area of ​​hidden sewage channels in abandoned mines. At the same time, it completes the location matching of multi-dimensional geophysical data and constructs a standardized multi-parameter feature dataset.

[0046] The objective function construction unit is used to build a multi-parameter joint inversion framework, establish a comprehensive objective function that integrates data fitting accuracy, parameter coordination constraints and numerical stability, introduce cross gradient constraints to associate the spatial distribution characteristics of different physical property parameters, and force the models corresponding to each parameter to maintain spatial structure consistency.

[0047] The dynamic step size adjustment unit is used to initiate iterative calculations using an improved gradient descent method with the objective function as the solution object. It constructs a dynamic step size adjustment mechanism based on the coupling of first-order iterative information and second-order function characteristics to complete the adaptive control of the step size. Specifically, a mathematical adjustment model for the iterative step size is set, which uses first-order iterative information to represent the iterative trend and second-order function characteristics to represent the local shape of the objective function. A weight allocation mechanism is introduced to adaptively balance the control effect of the two types of information on the step size.

[0048] The inversion model generation unit is used to adaptively adjust the step size in stages according to the iteration process based on the dynamic step size adjustment mechanism of dual-factor coupling, and to carry out the iterative solution of the objective function. At the same time, it combines multi-parameter feature clustering analysis to remove invalid and abnormal data, and continues to iterate until the preset convergence condition is met, and finally obtains a multi-parameter fusion inversion model that fits the actual geological characteristics.

[0049] In the iterative solution process, based on the first-order iterative information and the characteristics of the second-order function, the step size and the control priority of the two types of information are dynamically adjusted according to the iterative trend and the changes in the local shape of the objective function. This achieves phased optimization of the convergence speed in the early stage of iteration, the balance between efficiency and accuracy in the middle stage, and the convergence stability in the later stage, thus approximating the global optimal solution.

[0050] As a third aspect of the invention, a computer-readable storage medium is also provided, on which a computer program is stored, which is executed by a processor as described in any one of the claims, a multi-parameter collaborative inversion method for mine sewage conveying channels based on dynamic step size adjustment.

[0051] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects:

[0052] 1. The present invention provides a multi-parameter collaborative inversion method for mine wastewater transport channels based on dynamic step size adjustment. This method involves simultaneously collecting multi-dimensional geophysical data reflecting the conductivity, structural integrity, and electrochemical properties of the underground medium at the detection area of ​​concealed wastewater transport channels in abandoned mines. After data location matching, a standardized multi-parameter feature dataset is constructed. A multi-parameter joint inversion framework is then built, establishing a comprehensive objective function that integrates data fitting accuracy, parameter coordination constraints, and numerical stability. Cross-gradient constraints are introduced to correlate the spatial distribution characteristics of different physical property parameters, forcing the corresponding models of each parameter to maintain spatial structural consistency. This technical feature achieves standardized integration of multi-source geophysical data, avoiding the limitations of single-parameter inversion. Cross-gradient constraints effectively correlate the spatial characteristics of different physical property parameters, significantly reducing the ambiguity of multi-parameter inversion. This allows the inversion framework to adapt to the detection needs of wastewater transport channels in complex geological environments of mines, laying a foundation for subsequent iterative solutions that align with actual geological characteristics.

[0053] 2. The multi-parameter collaborative inversion method for mine wastewater transport channels based on dynamic step size adjustment of this invention uses a comprehensive objective function as the solution object, employs an improved gradient descent method to initiate iterative calculations, and constructs a dynamic step size adjustment mechanism based on the dual-factor coupling of first-order iterative information and second-order function characteristics. An iterative step size mathematical adjustment model is set, using first-order iterative information to represent the iterative trend and second-order function characteristics to represent the local shape of the objective function. A weight allocation mechanism is introduced to adaptively balance the effects of these two types of information on step size regulation. This technique overcomes the drawbacks of the fixed step size in traditional gradient descent methods, achieving adaptive control of the iterative step size. Through dual-factor coupling, it accurately captures the iterative process and local features of the objective function, ensuring a high degree of adaptation between step size regulation and iterative state. The weight allocation mechanism guarantees the synergy of the two types of information on step size regulation, effectively improving the convergence speed of iterative calculations while avoiding the problem of falling into local optima during iteration.

[0054] 3. The multi-parameter collaborative inversion method for mine pollution transport channels based on dynamic step size adjustment of this invention adaptively adjusts the step size in stages according to the iteration process through a dynamic step size adjustment mechanism based on two-factor coupling, and iteratively solves the objective function. Simultaneously, it combines multi-parameter feature clustering analysis to eliminate invalid and abnormal data, continuously iterating until a preset convergence condition consisting of the objective function change amplitude, model vector stability, and maximum number of iterations is met, ultimately obtaining a multi-parameter fusion inversion model that closely matches the actual geological characteristics. This technical feature achieves dynamic and accurate iterative solution, clustering analysis eliminates invalid and abnormal data to ensure the validity of the iterative data, and the multi-dimensional preset convergence condition balances inversion accuracy and computational efficiency, avoiding over-iteration or under-iteration. This allows the final multi-parameter fusion inversion model to characterize the spatial morphology, connectivity, and physical properties of hidden pollution transport channels in abandoned mines, providing reliable geological model support for the detection of pollution transport channels. Attached Figure Description

[0055] Figure 1 This is a flowchart of the multi-parameter collaborative inversion method for mine sewage conveying channels based on dynamic step size adjustment, according to an embodiment of the present invention.

[0056] Figure 2 This is a schematic diagram illustrating the correlation characteristic curve between the normalized residual and the trace value of the Hessian matrix during the iterative process in an embodiment of the present invention.

[0057] Figure 3 This is a schematic diagram of the dynamic change curves of the first-order / second-order feature weight allocation at each iteration stage in an embodiment of the present invention.

[0058] Figure 4 This is a schematic diagram of the system units in an embodiment of the present invention. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0060] Example 1

[0061] Please refer to Figure 1 This embodiment 1 provides a multi-parameter collaborative inversion method for mine sewage transport channels based on dynamic step size adjustment, including:

[0062] S1. In the detection area of ​​the hidden sewage transport channel in the abandoned mine, multi-dimensional geophysical data reflecting the conductivity, structural integrity and electrochemical characteristics of the underground medium are collected simultaneously. At the same time, the location matching of the multi-dimensional geophysical data is completed, and a standardized multi-parameter feature dataset is constructed.

[0063] S2. Build a multi-parameter joint inversion framework, establish a comprehensive objective function that integrates data fitting accuracy, parameter coordination constraints and numerical stability, introduce cross gradient constraints to associate the spatial distribution characteristics of different physical property parameters, and force the models corresponding to each parameter to maintain spatial structure consistency.

[0064] S3. Taking the objective function as the solution object, the improved gradient descent method is used to start the iterative operation. A dynamic step size adjustment mechanism based on the coupling of first-order iterative information and second-order function characteristics is constructed to complete the adaptive control of the step size. Among them, a mathematical adjustment model for the iterative step size is set. The first-order iterative information represents the iterative trend, and the second-order function characteristics represent the local shape of the objective function. A weight allocation mechanism is introduced to adaptively balance the control effect of the two types of information on the step size.

[0065] S4. Based on the dynamic step size adjustment mechanism of dual-factor coupling, the step size is adaptively adjusted in stages according to the iteration process to carry out the iterative solution of the objective function; at the same time, combined with multi-parameter feature clustering analysis to remove invalid abnormal data, the iteration continues until the preset convergence condition is met, and finally a multi-parameter fusion inversion model that fits the actual geological characteristics is obtained.

[0066] In the iterative solution process, based on the first-order iterative information and the characteristics of the second-order function, the step size and the control priority of the two types of information are dynamically adjusted according to the iterative trend and the changes in the local shape of the objective function. This achieves phased optimization of the convergence speed in the early stage of iteration, the balance between efficiency and accuracy in the middle stage, and the convergence stability in the later stage, thus approximating the global optimal solution.

[0067] This embodiment 1 further elaborates on the above process.

[0068] (1) Dataset construction

[0069] In the detection of hidden pollution transport channels in abandoned mines, in order to capture the comprehensive physical property anomalies of the underground media to support subsequent inversion analysis, it is necessary to first carry out systematic collection and standardized processing of multi-dimensional geophysical data. Based on the underground geological conditions of the detection area, three types of core geophysical data are collected simultaneously to specifically capture the physical property characteristics of the underground media's conductivity, the structural integrity of the underground geological body, and the electrochemical properties of the underground media's solid-liquid interface. Each of the three types of data has its own detection focus, forming a comprehensive coverage of the physical property changes in the pollution transport channels.

[0070] Geophysical data on the conductivity of underground media can reflect the differences in conductivity caused by changes in porosity and composition of underground media due to the existence of sewage transport channels; geophysical data on the integrity of geological structures can characterize the changes in physical characteristics caused by geological structural damage such as cracks and cavities formed by sewage transport channels; geophysical data on the electrochemical properties of solid-liquid interfaces can capture changes in electrochemical signals such as polarization response generated by interfacial reactions between pollutants and surrounding media within the channel. These three types of data complement and corroborate each other, fully restoring the comprehensive physical property anomalies of underground media caused by sewage transport channels.

[0071] Considering the diverse sources of multi-dimensional data and the differences in spatial benchmarks, it is necessary to simultaneously match the locations of various data. Using a uniformly established three-dimensional spatial coordinate system for the exploration area as a benchmark, the acquisition points and exploration profiles of various geophysical data are associated with the three-dimensional spatial coordinates within this coordinate system, thus unifying the spatial benchmark for all geophysical data. For point-based geophysical data, the three-dimensional coordinates of the acquisition points are directly assigned to the corresponding data. For profile-based and area-based geophysical data, corresponding three-dimensional spatial coordinates are matched to each sampling unit according to the exploration trajectory and sampling interval. Simultaneously, system positioning deviations of different exploration devices are corrected to ensure the consistency and accuracy of various data in spatial location. Binding various types of geophysical data to their corresponding spatial locations allows geophysical data from different dimensions to undergo correlation analysis under the same spatial benchmark.

[0072] Meanwhile, in order to eliminate the impact of differences in the dimensions and ranges of different parameters on the accuracy of subsequent inversion, all parameter sets are standardized to unify the data dimensions and ranges. At the same time, it is ensured that all physical property parameters and spatial location coordinate parameters are matched, realizing the normalized correspondence of multi-dimensional geophysical parameters in the spatial dimension, and providing a unified, accurate and collaborative data foundation for subsequent multi-parameter joint inversion.

[0073] (2) Construction of the objective function

[0074] To address the challenges of traditional single-parameter inversion methods that fail to comprehensively characterize the anomalies in the physical properties of mine wastewater transport channels, and the lack of a systematic collaborative mechanism in multi-parameter inversion, a multi-parameter joint inversion framework needs to be established. The core of this framework is to integrate the inversion processes of three types of physical property parameters—conductivity, structural integrity, and electrochemical characteristics of the solid-liquid interface—into a single system. This will construct an integrated inversion architecture adapted to multi-dimensional geophysical data, providing a unified technical carrier and logical support for multi-parameter collaborative analysis and synchronous modeling, thus achieving an upgrade from dispersed parameter inversion to integrated collaborative inversion.

[0075] During the construction process, the framework's unified spatial domain was first established based on the spatial benchmark and parameter dimensions of the previously standardized multi-parameter feature dataset. This domain was then matched with the three-dimensional spatial coordinate system of the detection area to ensure spatial compatibility between the framework and the input data. Next, the basic architecture for multi-parameter parallel inversion was built, and modular units compatible with inversion logic for different physical property parameters were designed. This allows various parameters to be inverted synchronously within the framework, while also reserving interfaces for embedding constraints to provide technical adaptability for the subsequent introduction of mechanisms such as cross-gradient constraints and numerical stability constraints. Finally, parameter collaboration and interaction rules within the framework were established, clarifying the data transfer and result feedback logic during the inversion process of different physical property parameters. This ensures that the inversion results of various parameters can be correlated, complementary, and mutually verified, breaking down the isolation barriers of single-parameter inversion.

[0076] The core objective of this framework is to break down the barriers between isolated inversions of different physical properties, incorporating multiple parameters reflecting the conductivity, structural integrity, and electrochemical characteristics of the solid-liquid interface of the underground medium into a single inversion system. This ensures that various parameters are interconnected and mutually reinforcing during the inversion process. By establishing this framework, a unified inversion process and spatial benchmark can be created, adapting to the characteristics of data after pre-standardized processing. This provides structural support for the subsequent establishment of a comprehensive objective function and the introduction of constraints, allowing the inversion process to align with the synergistic physical property laws of the underground geological bodies in the mine.

[0077] After the framework is built, based on the standardized multi-parameter feature dataset processed in the early stage, a comprehensive objective function is established that integrates data fitting accuracy, parameter co-constraints, and numerical stability. This function forms a unified solution objective by coupling data fitting terms, cross-gradient constraint terms, and regularization terms, taking into account the fit between the inversion results and the measured data, the spatial correlation of different parameter models, and the numerical stability of the solution process. Specifically, the comprehensive objective function is constructed based on the multi-dimensional geophysical data fitting accuracy, multi-parameter spatial co-constraints, and numerical solution stability. The expression is as follows:

[0078]

[0079] in, For the comprehensive objective function, The multi-parameter model vector represents the spatial distribution set of parameters characterizing the conductivity, structural integrity, and solid-liquid interface electrochemical properties of the underground medium. It corresponds to the standardized multi-parameter feature dataset collected and processed in the detection area, ensuring the compatibility between the objective function and the input data.

[0080] For the data fitting term, defined as , Define the spatial domain for detecting concealed sewage transport channels in abandoned mines. The inversion model is based on parameter vectors The calculated simulated values ​​of the geophysical response, For measured multi-parameter geophysical data after location matching and standardization, through The integral form minimizes the spatial global deviation between the simulated geophysical response and the measured multi-parameter geophysical data, ensuring that the inversion results fit the actual mining exploration scenario.

[0081] For the cross gradient constraint term, For spatial gradient operators, For tensor product operations, For models with different physical property parameters, the spatial gradient correlation function is used.

[0082] For regularization terms, For the Laplace operator, By adjusting the parameter vector The second-order spatial derivative integral suppresses numerical oscillations caused by abrupt changes in the physical properties of underground geological bodies in the mine during the iterative solution process, thereby improving the stability of the objective function solution.

[0083] Furthermore, the spatial gradient correlation function of different physical property parameter models was further investigated. To explain, its specific definition is as follows:

[0084]

[0085] The construction process of this formula first defines two types of physical property parameter components based on the vector dot product principle and the synergistic characteristics of underground mineral media. , The spatial gradient vectors are respectively... and ,in For spatial gradient operators, , , , , These are the three-dimensional spatial coordinates of the detection area of ​​the hidden sewage transport channel in the abandoned mine. The gradient vector directly reflects the changing trend and direction of the corresponding physical parameters in the underground space, adapting to the spatial location matching requirements of multi-dimensional parameters.

[0086] According to the rules of vector dot product, the dot product expression for two types of gradient vectors is: ,in for and The included angle between them has a range of values. This formula establishes a mathematical relationship between the gradient vector dot product and the vector magnitude and the included angle, providing a theoretical basis for quantifying the consistency of gradient direction.

[0087] To address the ambiguity of multi-parameter inversion in mine wastewater transport channels, the core requirement is to enforce spatial structural consistency among models with different physical property parameters. Because the physical property anomalies caused by wastewater transport channels are synergistic, the same geological anomaly area corresponds to… and The directions should converge, that is... Approaching 0, Approaching 1; based on this requirement, the dot product formula is transformed to... Separately, obtain ;

[0088] Based on the above deformation results, a spatial gradient correlation function is constructed. This function is essentially the cosine of the angle between the gradient vectors of two types of physical property parameters, and its value ranges from... Between 1 and 1, the spatial structural consistency of models with different physical property parameters can be directly quantified;

[0089] when When the value approaches 1, it indicates that the spatial gradient directions of the two types of parameter models are consistent, and the structural consistency is good; when When the deviation is 1, the parameter model is forcibly corrected through the deviation value fed back by the function, realizing the spatial structure constraint under the gradient tensor coupling, which echoes the spatial position matching feature of multi-dimensional parameters and effectively reduces the ambiguity of multi-parameter inversion.

[0090] , These are two different types of physical property parameters, corresponding to two types of parameters in the underground medium's conductivity, structural integrity, and solid-liquid interface electrochemical properties. The spatial structure association of different physical property parameter models is realized through gradient tensor coupling, which forces the boundary morphology and spatial distribution of the models corresponding to each parameter to remain consistent.

[0091] (3) Dynamic step size adjustment

[0092] Traditional gradient descent methods employ fixed step sizes for iterative calculations, which are ill-suited to the complex variations in objective functions during multi-parameter inversion of mine wastewater transport channels. This often results in slow convergence, getting trapped in local optima, or numerical oscillations in later stages. Therefore, this paper proposes an improved gradient descent method to initiate iterative calculations, using a constructed comprehensive objective function as the core solution object. The key is to establish a dynamic step-size adjustment mechanism based on the coupling of first-order iterative information and second-order function characteristics. This mechanism enables adaptive control of the iteration step size, balancing iterative efficiency and inversion accuracy, and adapting to the multi-parameter inversion requirements under complex geological conditions in mines.

[0093] After the iterative operation begins, the core state of the iterative process is captured through first-order iterative information and second-order function characteristics, providing data support for step size control. First-order iterative information uses the change in the iterative residual of the objective function as its core carrier, and the iterative trend is quantified by defining the difference in objective function values ​​between two adjacent iterations; let the... The objective function value of the next iteration is , No. The objective function value of the next iteration is ,in For the first The multi-parameter model vector of the next iteration. For the first The multi-parameter model vector of the next iteration, both of which correspond to the spatial distribution characteristics of the physical properties of the underground medium;

[0094] By calculating the change in iterative residual And further transformed into a normalized residual change index. To characterize the completed iterative trend;

[0095] The numerical value directly reflects the rate of change of the objective function during the iteration process. When the value is large, it indicates that the objective function value changes drastically, the iteration is in a rapid descent phase, and the corresponding iteration trend is an efficient convergence trend; when When the value is small, it indicates that the objective function value changes slowly, the iteration approaches the convergence threshold, and the corresponding iteration trend is a slow convergence trend;

[0096] when Abnormal fluctuations can indicate the presence of data interference or local optimum traps during the iteration process.

[0097] The second-order function features use the Hessian matrix of the objective function as the carrier of the second-order function features. The feature parameters of the Hessian matrix are used to quantitatively characterize the local shape of the objective function at the current iteration point, reflecting the function curvature change and extreme value distribution characteristics.

[0098] Regarding the first The multi-parameter model vector of the next iteration Construct the objective function exist Hessian matrix at the location , Let be a square matrix composed of second-order partial derivatives, whose elements are defined as follows: ,in , The first Sub-iteration multi-parameter model vector The first in The, the Each physical property parameter component corresponds to the spatial distribution parameters of the conductivity, structural integrity, and solid-liquid interface electrochemical properties of the underground medium.

[0099] By calculating the Hessian matrix traces Using the eigenvalue spectrum, we complete the representation of the local form of the objective function: the trace of the Hessian matrix. The sum of the elements on the main diagonal of the matrix directly reflects the overall curvature of the objective function at the current iteration point. The larger the value, the steeper the local curvature of the objective function, the more drastic the change in the function shape near the corresponding iteration point, and the higher the probability of the existence of local extrema;

[0100] The eigenvalue spectrum of the Hessian matrix determines the type of local extrema of the objective function by the sign and distribution of the eigenvalues. When all eigenvalues ​​are positive, it indicates that the current iteration point is in the local minimum region of the objective function. When the eigenvalues ​​alternate between positive and negative, it indicates that the current iteration point is in the saddle point region.

[0101] To achieve coordinated control of step size by the two types of information, a weight allocation mechanism is introduced, defining... The adjustment weights for first-order iterative information, The adjustment weights for the characteristics of the second-order function satisfy... Dynamic allocation is achieved by constructing an adaptive weight mapping function, the specific expression of which is as follows:

[0102]

[0103]

[0104] In the formula, This is an index for the change in normalized residuals; The iterative trend baseline threshold is determined by the overall accuracy level of the multi-parameter feature dataset. Hessian matrix traces ; This is the reference value for the preset Hessian matrix trace;

[0105] In the early stages of iteration, and , The value approaches 0. The value is relatively small. Increased proportion The proportion is reduced, which strengthens the control of the step size by the characteristics of the second-order function, adapts to the state of rapid function descent and gentle curvature, and improves the convergence speed.

[0106] Mid-iteration and The two types of information are balanced and adapted through a weight mapping function. and Synchronous dynamic fine-tuning balances convergence speed and inversion accuracy;

[0107] Later stages of iteration and , The value approaches 1. The value is relatively large. Increased proportion The proportion is reduced, the control of the step size by the first-order iterative information is strengthened, and numerical oscillations are suppressed to approach the global optimal solution.

[0108] like Figure 2 and Figure 3 As shown, based on the dynamic weight allocation mechanism and the quantitative feedback of the two types of features in the preceding process, the step size achieves adaptive adaptation through a two-step adjustment of "basic setting + secondary correction". The basic step size is determined based on the trace value of the Hessian matrix. The smaller the trace value, the gentler the local curvature of the objective function. The basic step size is increased by a preset ratio to adapt to the requirement of fast convergence. The larger the trace value, the steeper the curvature. The basic step size is reduced by a corresponding ratio to reduce the risk of numerical fluctuations caused by local extrema. At the same time, the step size direction is corrected by combining the eigenvalue spectrum of the Hessian matrix. The direction is finely adjusted for saddle point regions to break through local constraints, and the direction is kept stable for local minimum regions to avoid local optimum traps.

[0109] Subsequently, a second correction is completed by combining the weight ratio of each iteration stage. In the early stage of iteration, the step size is increased by relying on the weight allocation dominated by the characteristics of the second-order function to improve the convergence efficiency. In the middle stage of iteration, the step size is finely adjusted by balancing the weights of the two types of information to balance efficiency and accuracy. In the later stage of iteration, the step size is reduced by relying on the weight allocation dominated by the first-order iteration information to suppress numerical oscillation. Finally, the appropriate step size for each iteration is determined, and the matching of step size with iteration process and objective function shape is completed to ensure the stability and accuracy of multi-parameter inversion process.

[0110] (4) Generation of inversion model

[0111] In the iterative solution stage of multi-parameter collaborative inversion, the complex underground geological environment of the mine is prone to problems such as data interference and fluctuations in the iteration state. Relying solely on a single control logic can easily lead to insufficient inversion accuracy or low computational efficiency. To address this, based on the previously constructed dual-factor coupled dynamic step-size adjustment mechanism, the step-size is adaptively adjusted in stages according to the iteration process. Combining the dynamic changes in the iteration trend and the local shape of the objective function, the step-size and control priority are adjusted to achieve adaptation throughout the entire iteration process.

[0112] During the iterative solution process, multi-parameter feature clustering analysis is introduced simultaneously to remove potentially invalid outliers from the previously standardized data. Clustering algorithms are used to classify and analyze multi-dimensional physical property parameter data. For example, density clustering (DBSCAN) is employed to identify core data clusters and isolated outliers based on the spatial density distribution of data points, filtering out discrete data points that deviate from the normal data distribution range and fail to reflect true geological characteristics. Alternatively, K-means clustering can be used to divide the data into optimal clusters, identifying data that deviates from all cluster centers and exceeds a preset threshold as outliers, preventing such data from interfering with iterative calculations, ensuring the validity of the input data, and providing reliable support for iteration. Density clustering is more suitable for the discrete distribution characteristics of outliers in mine geological data, while K-means clustering is suitable for scenarios with relatively clear data cluster boundaries; the choice can be flexibly made based on the data distribution characteristics of the exploration area.

[0113] Meanwhile, multi-dimensional preset convergence conditions are set, constructing a judgment system from three core dimensions: inversion approximation degree, result stability, and computational boundaries, to ensure the rationality of the iteration termination time. Specifically, the inversion approximation degree is judged by the change amplitude of the objective function between adjacent iterations, using the difference in objective function values ​​between two adjacent iterations as the core quantitative indicator, combined with a preset approximation threshold to achieve accurate judgment. Specifically, the difference between the objective function value of the current iteration and the previous iteration is calculated and converted into a normalized change amplitude. When this amplitude is consistently less than the preset threshold, it indicates that the objective function has approached the minimum region, and the inversion result gradually approaches the global optimum, preliminarily determining that the convergence condition is met. If the change amplitude is consistently greater than the threshold, it indicates that the inversion is still in the rapid iteration stage and further computation is required.

[0114] The stability of the results is judged by the spatial distribution differences of the multi-parameter model vectors. Focusing on the spatial distribution characteristics of each physical property parameter in the multi-parameter model vectors, the distribution deviation of the corresponding parameters in the three-dimensional space of the detection area between two adjacent iterations is calculated, including differences in parameter values ​​and consistency of spatial boundary morphology. When the distribution deviation tends to stabilize and remains within the preset allowable range, it indicates that the inverted physical property parameter model has conformed to the actual underground geological characteristics and will not fluctuate significantly with the increase in the number of iterations, thus ensuring the reliability of the inversion results. If the distribution deviation fluctuates significantly, further iterations are needed to optimize the model stability.

[0115] The calculation termination boundary is defined by setting a maximum number of iterations. This number is reasonably set in combination with the scale of the mine exploration area, multiple parameter dimensions and computing equipment performance. It not only reserves sufficient space for iteration optimization to ensure the accuracy of the inversion results, but also effectively avoids infinite iteration caused by the objective function getting stuck in local minima or data interference, reduces computing resource consumption and ensures the efficiency of the inversion process.

[0116] The three judgment dimensions complement each other. The iteration terminates when any one of them meets the preset requirements. The iteration continues until the convergence condition is met, and finally outputs a multi-parameter fusion inversion model that fits the actual geological characteristics of the abandoned mine, accurately representing the spatial morphology and physical properties of the hidden pollution transport channels.

[0117] Example 2

[0118] Please refer to Figure 4 This embodiment 2 provides a multi-parameter collaborative inversion system for mine wastewater transport channels based on dynamic step size adjustment, including:

[0119] The dataset construction unit is used to simultaneously collect multi-dimensional geophysical data reflecting the conductivity, structural integrity and solid-liquid interface electrochemical properties of underground media in the detection area of ​​hidden sewage channels in abandoned mines. At the same time, it completes the location matching of multi-dimensional geophysical data and constructs a standardized multi-parameter feature dataset.

[0120] The objective function construction unit is used to build a multi-parameter joint inversion framework, establish a comprehensive objective function that integrates data fitting accuracy, parameter coordination constraints and numerical stability, introduce cross gradient constraints to associate the spatial distribution characteristics of different physical property parameters, and force the models corresponding to each parameter to maintain spatial structure consistency.

[0121] The dynamic step size adjustment unit is used to initiate iterative calculations using an improved gradient descent method with the objective function as the solution object. It constructs a dynamic step size adjustment mechanism based on the coupling of first-order iterative information and second-order function characteristics to complete the adaptive control of the step size. Specifically, a mathematical adjustment model for the iterative step size is set, which uses first-order iterative information to represent the iterative trend and second-order function characteristics to represent the local shape of the objective function. A weight allocation mechanism is introduced to adaptively balance the control effect of the two types of information on the step size.

[0122] The inversion model generation unit is used to adaptively adjust the step size in stages according to the iteration process based on the dynamic step size adjustment mechanism of dual-factor coupling, and to carry out the iterative solution of the objective function. At the same time, it combines multi-parameter feature clustering analysis to remove invalid and abnormal data, and continues to iterate until the preset convergence condition is met, and finally obtains a multi-parameter fusion inversion model that fits the actual geological characteristics.

[0123] In the iterative solution process, based on the first-order iterative information and the characteristics of the second-order function, the step size and the control priority of the two types of information are dynamically adjusted according to the iterative trend and the changes in the local shape of the objective function. This achieves phased optimization of the convergence speed in the early stage of iteration, the balance between efficiency and accuracy in the middle stage, and the convergence stability in the later stage, thus approximating the global optimal solution.

[0124] Example 3

[0125] This embodiment 3 also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it can implement any step of a multi-parameter collaborative inversion method for mine sewage conveying channels based on dynamic step size adjustment.

[0126] The computer-readable storage medium may include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0127] For a description of the computer-readable storage medium provided in this application, please refer to the above method embodiments; further details will not be repeated here.

[0128] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-parameter collaborative inversion method for mine wastewater transport channels based on dynamic step size adjustment, characterized in that, include: S1. In the detection area of ​​the hidden sewage transport channel in the abandoned mine, multi-dimensional geophysical data reflecting the conductivity, structural integrity and electrochemical characteristics of the underground medium are collected simultaneously. At the same time, the location matching of the multi-dimensional geophysical data is completed, and a standardized multi-parameter feature dataset is constructed. S2. Build a multi-parameter joint inversion framework, establish a comprehensive objective function that integrates data fitting accuracy, parameter coordination constraints and numerical stability, introduce cross gradient constraints to associate the spatial distribution characteristics of different physical property parameters, and force the models corresponding to each parameter to maintain spatial structure consistency. S3. Taking the objective function as the solution object, the improved gradient descent method is used to start the iterative operation. A dynamic step size adjustment mechanism based on the coupling of first-order iterative information and second-order function characteristics is constructed to complete the adaptive control of the step size. Among them, a mathematical adjustment model for the iterative step size is set. The first-order iterative information represents the iterative trend, and the second-order function characteristics represent the local shape of the objective function. A weight allocation mechanism is introduced to adaptively balance the control effect of the two types of information on the step size. S4. Based on the dynamic step size adjustment mechanism of dual-factor coupling, the step size is adaptively adjusted in stages according to the iteration process to carry out the iterative solution of the objective function; at the same time, combined with multi-parameter feature clustering analysis to remove invalid abnormal data, the iteration continues until the preset convergence condition is met, and finally a multi-parameter fusion inversion model that fits the actual geological characteristics is obtained. In the iterative solution process, based on the first-order iterative information and the characteristics of the second-order function, the step size and the control priority of the two types of information are dynamically adjusted according to the iterative trend and the changes in the local shape of the objective function. This achieves phased optimization of the convergence speed in the early stage of iteration, the balance between efficiency and accuracy in the middle stage, and the convergence stability in the later stage, thus approximating the global optimal solution.

2. The multi-parameter collaborative inversion method for mine wastewater conveyance channels based on dynamic step size adjustment according to claim 1, characterized in that, The standardized multi-parameter feature dataset in S1 includes: a set of quantitative parameters reflecting the conductivity of the underground medium, a set of physical feature parameters characterizing the structural integrity of the underground geological body, a set of polarization response parameters reflecting the electrochemical properties of the solid-liquid interface of the underground medium, and a set of spatial location coordinate parameters corresponding to each parameter. Each parameter set has been standardized to unify the units and data range, and all parameters are matched with spatial location coordinate parameters to achieve normalized correspondence of multi-dimensional geophysical parameters in the spatial dimension.

3. The multi-parameter collaborative inversion method for mine wastewater conveyance channels based on dynamic step size adjustment according to claim 1, characterized in that, The specific objective function in S2 is as follows: The comprehensive objective function is constructed based on the fitting accuracy of multi-dimensional geophysical data, the coordinated constraints of multi-parameter space, and the stability of numerical solutions. It forms a unified solution objective by coupling data fitting terms, cross-gradient constraint terms, and regularization terms, and its expression is: in, For the comprehensive objective function, The multi-parameter model vector represents the spatial distribution set of parameters characterizing the conductivity, structural integrity, and solid-liquid interface electrochemical properties of the underground medium. It corresponds to the standardized multi-parameter feature dataset collected and processed in the detection area, ensuring the compatibility between the objective function and the input data. For the data fitting term, defined as , Define the spatial domain for detecting concealed sewage transport channels in abandoned mines. The inversion model is based on parameter vectors The calculated simulated values ​​of the geophysical response, For measured multi-parameter geophysical data after location matching and standardization, through The integral form minimizes the spatial global deviation between the simulated geophysical response and the measured multi-parameter geophysical data, ensuring that the inversion results fit the actual mining exploration scenario; For the cross gradient constraint term, For spatial gradient operators, For tensor product operations, For models with different physical property parameters, the spatial gradient correlation function is used. For regularization terms, For the Laplace operator, By adjusting the parameter vector The second-order spatial derivative integral suppresses numerical oscillations caused by abrupt changes in the physical properties of underground geological bodies in the mine during the iterative solution process, thereby improving the stability of the objective function solution.

4. The multi-parameter collaborative inversion method for mine wastewater conveyance channels based on dynamic step size adjustment according to claim 3, characterized in that, The spatial gradient correlation function of the different physical property parameter models is defined as: in, , These are two different types of physical property parameters, corresponding to two types of parameters in the underground medium's conductivity, structural integrity, and solid-liquid interface electrochemical properties. The spatial structure association of different physical property parameter models is realized through gradient tensor coupling, which forces the boundary morphology and spatial distribution of the models corresponding to each parameter to remain consistent.

5. The multi-parameter collaborative inversion method for mine wastewater conveyance channels based on dynamic step size adjustment according to claim 1, characterized in that, The specific method for representing the iterative trend using first-order iterative information in S3 is as follows: Using the iterative residual change of the objective function as the core carrier of first-order iterative information, the iterative trend is quantified by defining the difference in objective function values ​​between two adjacent iterations; let the... The objective function value of the next iteration is , No. The objective function value of the next iteration is ,in For the first The multi-parameter model vector of the next iteration. For the first The multi-parameter model vector of the next iteration, both of which correspond to the spatial distribution characteristics of the physical properties of the underground medium; By calculating the change in iterative residual And further transformed into a normalized residual change index. To characterize the completed iterative trend; The numerical value directly reflects the rate of change of the objective function during the iteration process. When the value is large, it indicates that the objective function value changes drastically, the iteration is in a rapid descent phase, and the corresponding iteration trend is an efficient convergence trend; when When the value is small, it indicates that the objective function value changes slowly, the iteration approaches the convergence threshold, and the corresponding iteration trend is a slow convergence trend; when Abnormal fluctuations can indicate the presence of data interference or local optimum traps during the iteration process.

6. The multi-parameter collaborative inversion method for mine wastewater conveyance channels based on dynamic step size adjustment according to claim 1, characterized in that, The method for characterizing the local shape of the objective function using the second-order function features in S3 is as follows: The Hessian matrix of the objective function is used as the carrier of the second-order function characteristics. The local shape of the objective function at the current iteration point is quantitatively characterized by the characteristic parameters of the Hessian matrix, reflecting the function curvature change and extreme value distribution characteristics. Regarding the first The multi-parameter model vector of the next iteration Construct the objective function exist Hessian matrix at the location , Let be a square matrix composed of second-order partial derivatives, whose elements are defined as follows: ,in , The first Sub-iteration multi-parameter model vector The first in The, the Each physical property parameter component corresponds to the spatial distribution parameters of the conductivity, structural integrity, and solid-liquid interface electrochemical properties of the underground medium. By calculating the Hessian matrix traces Using the eigenvalue spectrum, we complete the representation of the local form of the objective function: the trace of the Hessian matrix. The sum of the elements on the main diagonal of the matrix directly reflects the overall curvature of the objective function at the current iteration point. The larger the value, the steeper the local curvature of the objective function, the more drastic the change in the function shape near the corresponding iteration point, and the higher the probability of the existence of local extrema; The eigenvalue spectrum of the Hessian matrix determines the type of local extrema of the objective function by the sign and distribution of the eigenvalues. When all eigenvalues ​​are positive, it indicates that the current iteration point is in the local minimum region of the objective function. When the eigenvalues ​​alternate between positive and negative, it indicates that the current iteration point is in the saddle point region.

7. The multi-parameter collaborative inversion method for mine wastewater conveyance channels based on dynamic step size adjustment according to claim 1, characterized in that, The weight allocation mechanism in S3 is as follows: definition The adjustment weights for first-order iterative information, The adjustment weights for the characteristics of the second-order function satisfy... Dynamic allocation is achieved by constructing an adaptive weight mapping function, the specific expression of which is as follows: In the formula, This is an index for the change in normalized residuals; The iterative trend baseline threshold is determined by the overall accuracy level of the multi-parameter feature dataset; Hessian matrix traces ; This is the reference value for the preset Hessian matrix trace; In the early stages of iteration, and , The value approaches 0. The value is relatively small. Increased proportion The proportion is reduced, which strengthens the control of the step size by the characteristics of the second-order function, adapts to the state of rapid function descent and gentle curvature, and improves the convergence speed. Mid-iteration and The two types of information are balanced and adapted through a weight mapping function. and Synchronous dynamic fine-tuning balances convergence speed and inversion accuracy; Later stages of iteration and , The value approaches 1. The value is relatively large. Increased proportion The proportion is reduced, the control of the step size by the first-order iterative information is strengthened, and numerical oscillations are suppressed to approach the global optimal solution.

8. The multi-parameter collaborative inversion method for mine wastewater conveyance channels based on dynamic step size adjustment according to claim 1, characterized in that, The specific convergence condition preset in S4 is as follows: The degree of inversion approximation is characterized by the change magnitude of the objective function in adjacent iterations, the stability of the inversion result is characterized by the spatial distribution difference of the multi-parameter model vector, and the termination boundary of the operation is defined by the preset maximum number of iterations. The iteration terminates when any one of the three judgment dimensions meets the preset requirements.

9. A multi-parameter collaborative inversion system for mine wastewater conveyance channels based on dynamic step size adjustment, characterized in that, include: The dataset construction unit is used to simultaneously collect multi-dimensional geophysical data reflecting the conductivity, structural integrity and solid-liquid interface electrochemical properties of underground media in the detection area of ​​hidden sewage channels in abandoned mines. At the same time, it completes the location matching of multi-dimensional geophysical data and constructs a standardized multi-parameter feature dataset. The objective function construction unit is used to build a multi-parameter joint inversion framework, establish a comprehensive objective function that integrates data fitting accuracy, parameter coordination constraints and numerical stability, introduce cross gradient constraints to associate the spatial distribution characteristics of different physical property parameters, and force the models corresponding to each parameter to maintain spatial structure consistency. The dynamic step size adjustment unit is used to initiate iterative calculations using an improved gradient descent method with the objective function as the solution object. It constructs a dynamic step size adjustment mechanism based on the coupling of first-order iterative information and second-order function characteristics to complete the adaptive control of the step size. Specifically, a mathematical adjustment model for the iterative step size is set, which uses first-order iterative information to represent the iterative trend and second-order function characteristics to represent the local shape of the objective function. A weight allocation mechanism is introduced to adaptively balance the control effect of the two types of information on the step size. The inversion model generation unit is used to adaptively adjust the step size in stages according to the iteration process based on the dynamic step size adjustment mechanism of dual-factor coupling, and to carry out the iterative solution of the objective function. At the same time, it combines multi-parameter feature clustering analysis to remove invalid and abnormal data, and continues to iterate until the preset convergence condition is met, and finally obtains a multi-parameter fusion inversion model that fits the actual geological characteristics. In the iterative solution process, based on the first-order iterative information and the characteristics of the second-order function, the step size and the control priority of the two types of information are dynamically adjusted according to the iterative trend and the changes in the local shape of the objective function. This achieves phased optimization of the convergence speed in the early stage of iteration, the balance between efficiency and accuracy in the middle stage, and the convergence stability in the later stage, thus approximating the global optimal solution.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer program is executed by a processor as described in any one of claims 1-8: a multi-parameter collaborative inversion method for mine sewage transport channels based on dynamic step size adjustment.

Images