Method, medium, and apparatus for data error term adaptive estimation for electromagnetic inversion

By employing an adaptive estimation method and utilizing electromagnetic measuring instruments and the cross-dimensional MH acceptance criterion, the uncertainty of noise levels in magnetotelluric inversion was resolved, thereby improving the stability and rationality of the inversion results and enhancing the adaptive adjustment effect of the data error term.

CN122506640BActive Publication Date: 2026-11-10CENT SOUTH UNIV
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202610991611.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-07-06
Publication Date
2026-11-10
Estimated Expiration
2046-07-06

AI Technical Summary

Technical Problem

In magnetotelluric inversion, the uncertainty of the noise level of the observation data leads to the inversion results being affected by subjective factors, making it difficult to improve the stability and rationality of the results.

Method used

An adaptive estimation method is adopted, which acquires data information through electromagnetic measuring instruments, sets adaptive parameters for data error, constructs a tower structure, calculates the joint likelihood function and the cross-dimensional MH acceptance criterion, and performs statistical calculations to estimate the underground conductivity and variance scaling factor, thereby achieving adaptive adjustment of the data error term.

Benefits of technology

This improves the stability and rationality of magnetotelluric inversion results, reduces the adverse effects of artificially set noise levels, and enhances the objectivity and application value of the inversion results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122506640B_ABST
    Figure CN122506640B_ABST
Patent Text Reader

Abstract

The present application relates to the technical field of electromagnetic prospecting and geophysical inversion, and specifically discloses a data error term adaptive estimation method for electromagnetic inversion, a medium and equipment. The data error term adaptive estimation method for electromagnetic inversion comprises the following steps: obtaining the magnetotelluric response data information of a certain research area by using an electromagnetic measuring instrument, setting a data error adaptive parameter, constructing the required tower structure for inversion, calculating the joint likelihood function, the cross-dimensional M-H acceptance criterion under the hyperparameter, and statistically calculating all the proposed models to obtain the underground conductivity and the posterior variance scaling factor in the research area. In the inversion process, the data error term can be adaptively estimated according to the characteristics of the observation data, so as to reduce the adverse effects caused by the artificial setting of the noise level, and improve the stability, rationality and application value of the magnetotelluric inversion result.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of electromagnetic exploration and geophysical inversion technology, specifically to an adaptive estimation method, medium, and device for data error terms in electromagnetic inversion. Background Technology

[0002] Magnetotelluric (MT) is an important geophysical method that uses natural electromagnetic fields as a source to detect the electrical structure of underground materials. It is widely used in deep geological structure research, mineral resource exploration, geothermal exploration, and oil and gas exploration. By inverting measured electromagnetic response data, the spatial distribution of resistivity of underground media can be obtained, providing important evidence for geological interpretation.

[0003] In magnetotelluric inversion methods, the observation data error term is typically a crucial component of the data fitting objective, and its setting directly impacts the determination of the data fit degree and the direction of model parameter updates during the inversion process. The data fit degree, in turn, is closely related to the setting of the observation error noise level. Therefore, the reasonable determination of the observation error noise level is one of the key factors affecting the stability, reliability, and interpretability of the inversion.

[0004] For theoretical models or numerical simulation data, noise levels can usually be directly given based on preset conditions, making the selection of error terms relatively clear. However, under actual observation conditions, the noise level of observed data is often uncertain due to various factors such as instrument accuracy, environmental interference, field source changes, surface conditions, and data preprocessing, making it difficult to accurately characterize using fixed empirical values. If the artificially set noise level is too low, the inversion process may overemphasize data fitting, leading to an overly complex model structure, excessive development of local anomalies, or even false anomalies. If the artificially set noise level is too high, it may weaken the constraint of effective information on model updates, resulting in insufficient resolution and incomplete structure recovery in the inversion results. Therefore, the traditional approach of relying on experience or manually specifying error levels is prone to subjective influence on the inversion results, hindering the improvement of objectivity and adaptability.

[0005] Therefore, there is an urgent need to propose a new method that can adaptively estimate the data error term based on the characteristics of the observed data during the inversion process, so as to reduce the adverse effects of artificially set noise levels and improve the stability, rationality and application value of magnetotelluric inversion results. Summary of the Invention

[0006] The purpose of this invention is to provide an adaptive estimation method for data error terms in electromagnetic inversion. This method includes steps such as obtaining magnetotelluric response data of a study area using electromagnetic measuring instruments, setting adaptive data error parameters, constructing the necessary tower structure for inversion, calculating the joint likelihood function, calculating the cross-dimensional MH acceptance criterion under hyperparameters, and statistically calculating the subsurface conductivity and its posterior variance scaling factor of all proposed models within the study area. This method enables adaptive estimation of data error terms based on the characteristics of observed data during the inversion process, reducing the adverse effects of artificially set noise levels and improving the stability, rationality, and application value of the magnetotelluric inversion results. The specific technical solution is as follows: An adaptive estimation method for data error terms in electromagnetic inversion includes the following steps: Step S1: Obtain magnetotelluric response data of a certain study area using electromagnetic measuring instruments; Step S2: Set the adaptive parameters for data error, specifically: use a global scaling factor to uniformly correct the initial covariance matrix; Step S3: Construct the required tower structure for inversion to obtain the existing tower structure model; Step S4: Calculate the number of tower permutations and combinations at different depth nodes, and construct a Markov chain starting from the first sampling. Step S5: Obtain a new candidate model for the tower structure at the t-th sampling time based on the existing tower structure model. ; Step S6: Calculate the joint likelihood function; Step S7: Calculation of the cross-dimensional MH acceptance criterion under hyperparameters, specifically including: proposing a distribution strategy; using a parallel tempering algorithm under cross-dimensional Bayesian inversion with hyperparameters to calculate the acceptance rate; for temperatures respectively Given two chains, obtain the inter-chain swap acceptance probability under the condition of introducing hyperparameters. ; Step S8: Perform a judgment, specifically: if the inter-chain swap acceptance probability of the proposed model satisfies 0 < If the value is less than 1, then the new candidate model is accepted, which means the existing tower structure at the (t+1)th sampling time is obtained. for: Otherwise, reject the new candidate model, thus obtaining the existing tower structure at the (t+1)th sampling time. for: , Given the previous existing tower structure model; take t=t+1, and perform a second judgment: if t is less than or equal to the maximum number of samplings designed, return to S5; otherwise, proceed to the next step. Step S9: Perform statistical calculations on all proposed models to obtain the underground electrical conductivity and its posterior variance scaling factor within the study area.

[0007] Preferably, in S3: constructing the required tower structure for the inversion includes designing the maximum depth of the tower structure. The maximum number of nodes in the tower structure was calculated. : .

[0008] Preferably, in S5: the nodes of the existing tower structure model are classified as existing nodes. Nodes can be generated and demiseable nodes , Represented as empty nodes; select any type of node to perform a tower-like structural transformation to form a new candidate model. ; If a node that can be generated is selected, it represents a new candidate tower structure node. The new candidate model is;

[0009] ; If a destructible node is selected, it represents a new candidate tower structure node. The new candidate model is:

[0010] ; Selecting one of the existing nodes means only the value of the selected node will be changed. The new candidate model is:

[0011] ; in: Indicates in Depth and The specific tower-like structural arrangement and combination of new candidate models under each node; Indicates the number of nodes is Wavelet coefficient values ​​at time Indicates the selected first The value after the node changes.

[0012] Preferably, in S5, the new candidate model under the tower structure is inversely transformed to obtain the proposed model of the conductivity structure; the inverse transformation uses wavelet basis functions.

[0013] Preferably, the magnetotelluric response data collected in step S1 exists in four different datasets: apparent resistivity and phase in TE and TM modes. Step S6 specifically includes: Fitting functions for different data types As shown in the following formula:

[0014] ; in: Indicates the first Forward response data vectors corresponding to each dataset; This represents the corresponding observation data vector. This represents the diagonal covariance matrix constructed from the standard deviations of the initial observation errors of this type of data; Indicates the variance scaling factor; superscript Represents transposition; Overall fitting function As shown in the following formula:

[0015] ; Joint likelihood function Calculate using the following formula:

[0016] ; in: Indicates the number of data points. Representation matrix The determinant of the matrix, specifically the variance scaling factor, is used in the inversion study. .

[0017] Preferably, step S7 includes the following steps: Step S7.1: Recommended distribution strategy: In each iteration, randomly select one of the four types of update operations with equal probability to generate candidate models. Step S7.2, Parallel refractory algorithm under cross-dimensional Bayesian inversion with hyperparameters, specifically includes: Let the model parameters be The observation data is Then the temperature The posterior distribution under Written as:

[0018] ; in: Let be the likelihood function. This represents the model's prior distribution; Let the first Chain and the The states of the chains before the exchange are as follows: and Then a single exchange proposal is represented as:

[0019] ; We construct the Metropolis-Hasting acceptance criterion with a wavelet domain tower structure based on the hyperparameter and parallel tempering method, calculate the acceptance rate of the proposed model, and select to accept or reject new candidate models based on the acceptance rate. The recommended model acceptance rate for selecting existing nodes The expression is:

[0020] ; Among them: Among them: For model parameters and variance scaling factor The likelihood function of the observed data under the given conditions; For model parameters and variance scaling factor difference The likelihood function of the observed data under the given conditions; The recommended model acceptance rate for selecting productive nodes The expression is:

[0021] ; in: for The probability distribution of the number of permutations under each node; for The probability distribution of the number of permutations under each node; This represents the number of deprecated nodes in the candidate model. This represents the number of nodes that can be generated in the current model. The acceptance rate of the suggested model that selects demiseable nodes The expression is:

[0022] ; in: This represents the number of degradable nodes in the current model. This represents the number of nodes that can be generated by the candidate model. For the newly generated suggested model By simply changing the value of the variance scaling factor, and according to its acceptance criteria, determining whether to accept the model by calculating the ratio of its likelihood function to the proposed distribution using the following formula:

[0023] ; in: For the construction of MH acceptance criteria; This represents the variance scaling factor in the current model. Generate candidate model variance scaling factor under the condition The probability density; Indicates the variance scaling factor in the candidate model Under the condition of reverse process to generate the variance scaling factor of the current model The probability density; Step S7.3: Exchange different chains, specifically: For temperatures respectively and The two chains, their exchange acceptance probability as follows:

[0024] ; in: Representation Model At temperature The probability of it happening; Representation Model At temperature The probability of it happening; Representation Model At temperature The probability of it happening; Representation Model At temperature The probability of it happening; Covariance of actual data Represented as:

[0025] ; in: This is the variance scaling factor; Indicates the initial relative covariance structure; The following exchange acceptance probabilities between chains are obtained under the condition of introducing hyperparameters;

[0026] ; in: Indicates the first Model parameters of the chain and the corresponding variance scaling factor Likelihood function value under the given conditions; Indicates the first Model parameters of the chain and the corresponding variance scaling factor The likelihood function value under the given conditions.

[0027] Preferably, the four types of update operations include: Model parameter perturbation, in the candidate perturbation node set A node is randomly selected in the middle for perturbation; Hyperparameter perturbation involves perturbing the variance scaling factor component and generating new candidate values ​​centered on the current value using a Gaussian proposal distribution. The generation strategy, in the candidate generation node set A node is randomly selected from the data to generate the data; The extinction strategy, in the set of candidate extinction nodes A node is randomly selected and eliminated.

[0028] In addition, the present invention also provides an electromagnetic probability inversion device, comprising: a processor and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the adaptive estimation method for data error terms in electromagnetic inversion as described above.

[0029] In addition, the present invention also includes a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the adaptive estimation method for data error terms for electromagnetic inversion as described above.

[0030] In addition to the objectives, features, and advantages described above, the present invention has other objectives, features, and advantages. The invention will now be described in further detail with reference to the figures. Attached Figure Description

[0031] To more clearly illustrate the technical solutions of the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a schematic diagram of the tower structure with a maximum depth of 3 in a preferred embodiment of the present invention; Figure 2 This is a schematic diagram of a tower structure with perturbed, generated, and destroyed nodes in a preferred embodiment of the present invention; Figure 3 This is a schematic diagram of the dual-box model and station location in a preferred embodiment of the present invention; Figure 4 This is a schematic diagram of the posterior mean model in a preferred embodiment of the present invention; Figure 5 This is a schematic diagram of the posterior distribution of the variance scaling factor in a preferred embodiment of the present invention. Detailed Implementation

[0033] To enable those skilled in the art to better understand the present invention, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Example:

[0034] This embodiment discloses an adaptive estimation method for data error terms in magnetotelluric inversion, specifically for the subsurface conductivity model. With observation data The response relationship between them can be mapped by forward operators. However, in actual exploration, due to the limited number of stations and frequencies used, coupled with unavoidable human interference noise and the influence of other natural environmental factors, researchers often cannot obtain comprehensive and accurate magnetotelluric data. Therefore, the data will contain error terms, expressed as: 1); in: This represents the error term in the observed data; Given underground model parameters Under the given conditions, the forward modeling operator is computed.

[0035] We assume that the observation error follows a zero-mean, independent, and identically distributed Gaussian model. This assumption is based on the fact that observation errors are typically composed of the superposition of multiple independent small perturbations, and it also facilitates the construction of the likelihood function and posterior inference. Therefore, the observation error follows the following distribution: 2); in: This represents the standard deviation of the error for different observation data; Indicates the first One error value; The standard deviation is expressed as The normal distribution; Therefore, the overall error vector can be written as: 3); In Bayesian inversion, the use of cross-dimensional Bayesian methods to invert the magnetotelluric response mainly involves obtaining the posterior probability density distribution of the relevant model based on known data. Through prior distribution Data likelihood distribution and data constants The distribution is as follows: 4); because It can be considered as a normalization constant, therefore: 5); For the specific likelihood function, it is assumed that the error follows a Gaussian distribution, that is, given the model parameters... and error level standard deviation Under the condition of, the likelihood function of the observed data Written as: 6); in: Indicates that given model parameters and data error standard deviation Observational data under the conditions The probability density of occurrence.

[0036] Within the hierarchical Bayesian framework, the inverted model parameters can be defined as follows: Therefore, the joint posterior probability density of the model parameters and the noise level can be expressed as: 7); in: Representation Model The prior distribution; The prior distribution representing the standard deviation of the error level; Therefore, the hierarchical Bayesian method essentially extends the fixed error parameters in traditional Bayesian inversion into random variables that can participate in inference, enabling joint estimation of model parameters, model dimension, and noise level within a unified probabilistic framework. Its constructed MH acceptance criterion is specifically written in the following form: 8); in: This indicates the MH acceptance criteria that have been constructed; This represents the prior distribution of the candidate model; The standard deviation of the error level of the candidate model; Represents model parameters and error level standard deviation The likelihood function of the observed data under the given conditions; Represents model parameters and error level standard deviation The likelihood function of the observed data under the given conditions; This represents the probability of generating the current model given the candidate models; This represents the probability of generating a candidate model given the current model. This represents the standard deviation at the current model error level. Generate candidate model error level standard deviation under the condition of The probability density; Indicates the standard deviation of the error level of the candidate model. Under the condition of reverse process to generate the standard deviation of the current model error level The probability density; Let be a Jacobian matrix, take .

[0037] The adaptive estimation method for data error terms in electromagnetic inversion in this embodiment specifically includes the following steps: Step S1: Obtain magnetotelluric response data of a certain study area using electromagnetic measuring instruments.

[0038] Step S2: Set adaptive parameters for data error, specifically by uniformly correcting the initial covariance matrix using a global scaling factor. Details are as follows: In cross-dimensional Bayesian inversion, this embodiment introduces a hierarchical Bayesian inversion method, a Bayesian method that establishes a probabilistic model hierarchically. It not only treats the model parameters as unknowns but also estimates higher-level parameters such as error levels or prior distribution parameters; these parameters are called hyperparameters. This allows for simultaneous evaluation of model parameters and uncertainties during the inversion process, making the results more objective and reliable. When the error level standard deviation... When considered as an unknown and included in the sampling, the standard deviation vector of the observation error for MT data is... .

[0039] If the standard deviation of each data point is estimated as an independent hyperparameter, the dimension of the hyperparameters will expand rapidly, thus reducing sampling efficiency. Therefore, this invention does not treat the standard deviation of each data point as an independent hyperparameter, but instead uses a global scaling factor to uniformly correct the initial covariance matrix, that is, the actual data covariance is expressed as: 9); in: This represents the initial relative covariance structure. This represents the variance scaling factor, used to characterize the scaling of the overall noise level. (The remaining text appears to be incomplete and possibly contains errors.) Using hyperparameters for inversion, this approach helps reduce the dimensionality of hyperparameters while preserving the original relative structure of the error, thus avoiding increased inversion complexity and decreased sampling stability due to an excessive number of noise-related hyperparameters.

[0040] Step S3: Construct the required tower structure for inversion, obtaining an existing tower structure model. The existing tower structure model includes the maximum depth of the designed tower structure. Thus, the maximum number of nodes in the tower structure can be calculated. ,as follows:

[0041] .

[0042] Therefore, the designed tower structure can be represented as follows (taking a maximum depth of 3 as an example), such as Figure 1 As shown.

[0043] Step S4: Calculate the number of tower permutations and combinations at different depth nodes, and construct a Markov chain starting from the first sampling.

[0044] Step S5: Obtain a new candidate model for the tower structure at the t-th sampling time based on the existing tower structure model. .

[0045] For a Markov chain, considering the existing tower structure model at the t-th sampling time, the nodes of the tower structure can be classified as existing nodes. Nodes can be generated and demiseable nodes , This is represented as an empty node. Select any type of node and perform a tower-like structural transformation to form a new candidate model. The number of samples t ranges from 1 to one million; If a node that can be generated is selected, it represents a new candidate tower structure node. The new candidate model is;

[0046] ; If a destructible node is selected, it represents a new candidate tower structure node. The new candidate model is:

[0047] ; Selecting one of the existing nodes means only the value of the selected node will be changed. The new candidate model is:

[0048] ; in: Indicates in Depth and The specific tower-like structural arrangement and combination of new candidate models under each node; Indicates the number of nodes is Wavelet coefficient values ​​at time Indicates the selected first The value after the node changes.

[0049] For a detailed schematic diagram of the perturbable, generated, and destroyed nodes of the tower structure, please refer to [link / reference]. Figure 2 .

[0050] Step S5: Perform an inverse transformation based on the existing tower structure model to obtain a new candidate model for the tower structure at the t-th sampling time. The inverse transform uses wavelet basis functions.

[0051] Step S6: Calculate the joint likelihood function. Details are as follows: During the sampling process, data fitting is required. Because the collected magnetotelluric data includes four different datasets—apparent resistivity and phase—in TE and TM modes, The fitting function for different data types takes the following form: 10); in: Indicates the first Fitting functions for each dataset; Indicates the first Forward response data vectors corresponding to each dataset; This represents the corresponding observation data vector. This represents the diagonal covariance matrix constructed from the standard deviations of the initial observation errors of this type of data; superscript This indicates transpose.

[0052] Overall fitting function As shown in the following formula: 11); Furthermore, assuming that the errors of different data types are independent and follow a Gaussian distribution, the joint likelihood function... It can be represented as: 12); in: Indicates the number of data points. Representation matrix The determinant of the matrix, specifically the variance scaling factor, is used in the inversion study. .

[0053] Step S7, calculation of the cross-dimensional MH acceptance criterion under hyperparameters, is as follows: Based on image compression and wavelet basis function Bayesian inversion, wavelet coefficients are represented by tree-structured nodes. This invention adds a variance scaling factor to this structure. Therefore, the MH acceptance criterion is established and calculated as follows: Step S7.1 suggests the following distribution strategy: In each iteration, one of the four types of update operations listed in Table 1 is randomly selected with equal probability to generate candidate models. These four types of proposal operations together constitute the candidate model update mechanism in the hierarchical Bayesian generation-death algorithm. Specifically, model parameter perturbation, generation, and death operations are mainly used to adjust the model structure, while hyperparameter perturbation is used to update the noise level.

[0054] Table 1. Suggested update strategies for lower-level Bayesian inversion in the generation-death algorithm.

[0055]

[0056] This embodiment focuses on the variance scaling factor. The disturbance. Each time to When performing a perturbation, use the current value. Gaussian distribution centered on Randomly select a new value The specific derivation is as follows: 13); in: Indicates from the current value A new candidate value is generated centered at the Gaussian proposal distribution. The distribution of is expressed as: 14); in: This represents the standard deviation of the proposed distribution; This indicates the update step size of the variance scaling factor.

[0057] Step S7.2, Parallel refractory algorithm for cross-dimensional Bayesian inversion after introducing hyperparameters, details are as follows: The basic idea of ​​parallel tempering is to introduce a temperature parameter to perform temperature-based processing on the posterior distribution. Let the model parameters be... The observation data is Then the temperature The posterior distribution under It can be written as: 15); in: Let be the likelihood function. For the model The prior distribution of . When At that time, the above formula degenerates into the original target posterior distribution; with temperature... As temperature increases, the distribution gradually flattens out, which is beneficial for the parameter chain to perform a global search over a larger range; with increasing temperature... As the probability decreases, the distribution gradually concentrates towards the local posterior distribution, which is beneficial for finely characterizing regions with high posterior probabilities. Furthermore, parallel tempering sets a fixed temperature sequence throughout the sampling process. A Markov chain is constructed at each temperature level. Each chain undergoes intra-chain updates at its corresponding temperature using the standard Markov chain Monte Carlo method, while periodically allowing state exchanges between adjacent temperature chains. Let the th... Chain and the The states of the chains before the exchange are as follows: and Then a single exchange proposal can be represented as: 16); Through this inter-chain exchange mechanism, the information obtained by the high-temperature chain during the global search process can be transferred to the low-temperature chain, thereby improving the ability of the low-temperature chain to escape local high-probability regions and enhancing the overall sampling process's efficiency in exploring complex posterior distributions.

[0058] We construct the Metropolis-Hasting acceptance criterion with a wavelet domain tower structure based on the hyperparameter and parallel tempering method, calculate the acceptance rate of the proposed model, and select to accept or reject new candidate models based on the acceptance rate. The recommended model acceptance rate for selecting existing nodes The expression is:

[0059] 17); in: For model parameters and variance scaling factor The likelihood function of the observed data under the given conditions; For model parameters and variance scaling factor difference The likelihood function of the observed data under the given conditions; The recommended model acceptance rate for selecting productive nodes The expression is:

[0060] 18); in: for The probability distribution of the number of permutations under each node; for The probability distribution of the number of permutations under each node; This represents the number of deprecated nodes in the candidate model. This represents the number of nodes that can be generated in the current model. The acceptance rate of the suggested model that selects demiseable nodes The expression is:

[0061] 19); in: This represents the number of degradable nodes in the current model. This represents the number of nodes that can be generated by the candidate model. For the newly generated suggested model By simply changing the value of the variance scaling factor, and according to its acceptance criteria, determining whether to accept the model by calculating the ratio of its likelihood function to the proposed distribution using the following formula:

[0062] 20); in: For the construction of MH acceptance criteria; This represents the variance scaling factor in the current model. Generate candidate model variance scaling factor under the condition The probability density; Indicates the variance scaling factor in the candidate model Under the condition of reverse process to generate the variance scaling factor of the current model The probability density.

[0063] S6.3: Acceptance rate calculated based on S6.2 Make a judgment; if 0 < If the value is less than 1, then the proposed model is accepted. Otherwise, the suggested model is rejected. .

[0064] S6.4: Exchange between different chains To ensure detail balance during multi-temperature chain joint sampling, a sampling interval of 20 times should be used, and the Metropolis–Hastings criterion should be applied. For temperatures of... The two chains, their exchange acceptance probability It can be written as: twenty one); in: Representation Model At temperature The probability of it happening; Representation Model At temperature The probability of it happening; Representation Model At temperature The probability of it happening; Representation Model At temperature The probability of it happening.

[0065] If the standard deviation of each data point is estimated as an independent hyperparameter, the hyperparameter dimension will rapidly expand, thus reducing sampling efficiency. Therefore, this invention does not treat the standard deviation of each data point as an independent hyperparameter, but instead uses a global scaling factor to uniformly correct the initial covariance matrix, i.e., the actual data covariance... Represented as:

[0066] ; in: This is the variance scaling factor; The standard deviation for each data point.

[0067] Because this invention employs hierarchical Bayesian methods, hyperparameters are further introduced into its inversion process, altering the joint posterior distribution and its corresponding likelihood function form. Therefore, the inter-chain exchange acceptance probability under the introduced hyperparameter condition can be derived, and its specific expression is as follows: twenty two); in: Indicates the first Model parameters of the chain and the corresponding variance scaling factor Likelihood function value under the given conditions; Indicates the first Model parameters of the chain and the corresponding variance scaling factor The likelihood function value under the given conditions.

[0068] Step S8: Repeat S4 to S6, and save the obtained sampling model every Z samplings, until the number of samplings t = Q, where Q is the maximum number of samplings designed, which is generally one million. Step S9: Perform statistical calculations on all the sampling models saved in S8 to obtain the underground conductivity and its posterior variance scaling factor within the study area. .

[0069] To examine the impact of noise levels on magnetotelluric inversion results, model parameterization and variance scaling factors are discussed here. Simultaneously, they are treated as unknowns. This embodiment designs a synthetic model that simultaneously includes both high-conductivity and low-conductivity anomalies, referred to as a "two-box model." For example... Figure 3 As shown, the model consists of two rectangular anomalies with electrical conductivities of 0.1 S / m and 0.001 S / m, respectively, embedded in a uniform background with a conductivity of 0.01 S / m. The dimensions of both anomalies are... The model space extent is set to... The inversion mesh is uniformly partitioned horizontally and logarithmically partitioned vertically, ultimately discretizing the model into... Sub-regions. This paper sets the maximum depth of the wavelet tree structure to [number]. Therefore, the maximum number of nodes is 1024. In the observation system design, a total of 17 measuring points are deployed, such as... Figure 3 The red inverted triangle in the middle shows the apparent resistivity and phase response in TE and TM modes, respectively. Eight frequencies were used, ranging from 10 Hz to 0.01 Hz, uniformly distributed on a logarithmic scale, and 3% Gaussian noise was added. A total of 544 apparent resistivity / phase data points for TE and TM modes were generated. Correspondingly, the initial relative covariance matrix... Based on this noise level, the actual covariance is consistent with the initial covariance. The true value of the corresponding variance scaling factor is set to .

[0070] A total of 1 million rj-MCMC samples were performed. The initial model given was a uniform half-space model with a conductivity of 0.01 (S / m), and its initial variance scaling factor was given. The burn-in phase involves the first 100,000 samples, followed by saving the model every 20 samples, for a total of 45,000 models for posterior inference. Figure 4 The posterior mean conductivity model obtained from the inversion is presented, with the red rectangle indicating the extent of the true anomalies. Overall, the inversion results recover the spatial locations and relative electrical characteristics of the two anomalies relatively well, but there is a certain degree of smoothing diffusion at their boundaries. The background region maintains a relatively stable conductivity level overall. In general, compared with the true model, the inversion results still exhibit a certain degree of smoothing transition near the anomaly boundaries, but overall they can reflect the main electrical distribution characteristics of the model.

[0071] Furthermore, the variance scaling factor obtained from the inversion is... An analysis was conducted. Figure 5 Given The posterior probability density distribution, where the red dashed line represents the true value. Overall, the variance scaling factor The posterior distributions are all concentrated within a finite range, and the probability peak is 1.28, which is close to the true value. This indicates that the introduced noise hyperparameters can be effectively constrained during the inversion process and can recover the true noise level well.

[0072] In addition, this embodiment also provides a magnetotelluric inversion device, including: a processor and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the adaptive estimation method for data error terms in electromagnetic inversion as described above.

[0073] In addition, this embodiment also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the adaptive estimation method for data error terms in electromagnetic inversion as described above.

[0074] It should be noted that the division of the various modules in the above system is merely a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. These modules can be implemented entirely in software via processing element calls; they can be fully implemented in hardware; or some modules can be implemented by processing element calls to software, while others are implemented in hardware. Each module can be a separate processing element, or it can be integrated into a chip in the aforementioned device. Alternatively, it can be stored as program code in the memory of the aforementioned device, and its functions can be called and executed by a processing element of the device. Furthermore, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each of the above modules can be completed through the integrated logic circuits in the hardware of the processor element or through software instructions.

[0075] It should be understood that the above-described device embodiments are merely illustrative, and the device of this application can also be implemented in other ways. For example, the division of units / modules in the above embodiments is only a logical functional division, and there may be other division methods in actual implementation. For example, multiple units, modules, or components may be combined, or integrated into another system, or some features may be ignored or not executed.

[0076] Furthermore, unless otherwise specified, the functional units / modules in the various embodiments of this application can be integrated into one unit / module, or each unit / module can exist physically separately, or two or more units / modules can be integrated together. The integrated units / modules described above can be implemented in hardware or as software program modules.

[0077] When an integrated unit / module is implemented in hardware, the hardware can be digital circuits, analog circuits, etc. The physical implementation of the hardware structure includes, but is not limited to, transistors, memristors, etc. Unless otherwise specified, the processor can be any suitable hardware processor, such as a Central Processing Unit (CPU), Graphics Processing Unit (GPU), Application Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), controller, microcontroller, microprocessor, or other electronic component. Unless otherwise specified, memory can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as a USB flash drive, Random-Access Memory (RAM), Static Random-Access Memory (SRAM), Dynamic Random-Access Memory (DRAM), Enhanced Dynamic Random-Access Memory (EDRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), Resistive Random Access Memory (RRAM), High-Bandwidth Memory (HBM), and Hybrid Memory Cube (HMC). Cube, magnetic storage, flash memory, disk, optical disk, portable hard drive or magnetic disk, and other media that can store program code.

[0078] If the integrated unit / module is implemented as a software program module and sold or used as an independent product, it can be stored in a computer-readable storage device. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application.

[0079] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An adaptive estimation method for data error terms in electromagnetic inversion, characterized in that, Includes the following steps: Step S1: Obtain magnetotelluric response data of a certain study area using electromagnetic measuring instruments; Step S2: Set the data error adaptive parameter, specifically: adjust the variance scaling factor s 2 As an unknown random variable to be estimated and included in the sampling, the variance scaling factor is used as a global scaling factor to uniformly correct the initial covariance matrix, so that the variance scaling factor and the model parameters participate together in the posterior probability calculation. Step S3: Construct the required tower structure for inversion to obtain the existing tower structure model; Step S4: Calculate the number of tower permutations and combinations at different depth nodes, and construct a Markov chain starting from the first sampling. Step S5: Obtain a new candidate model for the tower structure at the t-th sampling time based on the existing tower structure model. ; Step S6: Calculate the joint likelihood function, wherein the joint likelihood function is the model parameters and the variance scaling factor s. 2 The likelihood function of the observed data under the given conditions; Step S7, calculation of the cross-dimensional MH acceptance criterion under hyperparameters, specifically includes: S7.1, Recommended distribution strategy: Specifically, in each iteration, one of the four types of update operations is randomly selected with equal probability to generate candidate models. The four types of update operations include model parameter perturbation, generation strategy, elimination strategy and hyperparameter perturbation. The hyperparameter perturbation is to perturb the variance scaling factor and generate new candidate values ​​with the current value as the center using a Gaussian recommendation distribution. S7.2 Parallel refrigeration algorithm under cross-dimensional Bayesian inversion after introducing hyperparameters, calculate the acceptance rate; wherein, for the case where the value of the variance scaling factor is only changed in the newly generated proposal model, the ratio of its likelihood function to the proposal distribution is used to determine whether to accept the proposal model; S7.3, for temperatures respectively Given two chains, obtain the inter-chain swap acceptance probability under the condition of introducing hyperparameters. The inter-chain exchange acceptance probability is calculated based on the ratio of the likelihood function value of the j-th chain under the model parameters and corresponding variance scaling factor to the likelihood function value of the i-th chain under the model parameters and corresponding variance scaling factor, as well as the temperatures of the two chains. Step S8: Perform a judgment, specifically: if the inter-chain swap acceptance probability of the proposed model satisfies 0 < If the value is less than 1, then the new candidate model is accepted, which means the existing tower structure at the (t+1)th sampling time is obtained. for: Otherwise, reject the new candidate model, thus obtaining the existing tower structure at the (t+1)th sampling time. for: , Given the previous existing tower structure model; take t=t+1, and perform a second judgment: if t is less than or equal to the maximum number of samplings designed, return to S5; otherwise, proceed to the next step. Step S9: Perform statistical calculations on all proposed models to obtain the underground electrical conductivity and its posterior variance scaling factor within the study area.

2. The adaptive estimation method for data error terms in electromagnetic inversion according to claim 1, characterized in that, In S3: Constructing the required tower structure for the inversion includes designing the maximum depth of the tower structure. ; The maximum number of nodes in the tower structure was calculated. : .

3. The adaptive estimation method for data error terms in electromagnetic inversion according to claim 2, characterized in that, In S5: Classify the nodes of the existing tower structure model as existing nodes. Nodes can be generated and demiseable nodes , Represented as empty nodes; select any type of node to perform a tower-like structural transformation to form a new candidate model. ; If a node that can be generated is selected, it represents a new candidate tower structure node. The new candidate model is; ; If a destructible node is selected, it represents a new candidate tower structure node. The new candidate model is: ; Selecting one of the existing nodes means only the value of the selected node will be changed. The new candidate model is: ; in: Indicates in Depth and The specific tower-like structural arrangement and combination of new candidate models under each node; Indicates the number of nodes is Wavelet coefficient values ​​at time Indicates the selected first The value after the node changes.

4. The adaptive estimation method for data error terms in electromagnetic inversion according to claim 3, characterized in that, In S5, the new candidate model under the tower structure is inversely transformed to obtain the proposed model of the conductivity structure; the inverse transformation uses wavelet basis functions.

5. The adaptive estimation method for data error terms in electromagnetic inversion according to claim 3, characterized in that, The magnetotelluric response data collected in step S1 exists in four different datasets: apparent resistivity and phase in TE and TM modes. Step S6 specifically includes: Fitting functions for different data types As shown in the following formula: ; in: Indicates the first Forward response data vectors corresponding to each dataset; This represents the corresponding observation data vector. This represents the diagonal covariance matrix constructed from the standard deviations of the initial observation errors of this type of data; Indicates the variance scaling factor; superscript Represents transposition; Overall fitting function As shown in the following formula: ; Joint likelihood function Calculate using the following formula: ; in: Indicates the number of data points. Representation matrix The determinant of the matrix, specifically the variance scaling factor, is used in the inversion study. .

6. The adaptive estimation method for data error terms in electromagnetic inversion according to claim 5, characterized in that, Step S7 includes the following steps: Step S7.2, Parallel refractory algorithm under cross-dimensional Bayesian inversion with hyperparameters, specifically includes: Let the model parameters be The observation data is Then the temperature The posterior distribution under Written as: ; in: Let be the likelihood function. This represents the model's prior distribution; Let the first Chain and the The states of the chains before the exchange are as follows: and Then a single exchange proposal is represented as: ; We construct the Metropolis-Hasting acceptance criterion with a wavelet domain tower structure based on the hyperparameter and parallel tempering method, calculate the acceptance rate of the proposed model, and select to accept or reject new candidate models based on the acceptance rate. The recommended model acceptance rate for selecting existing nodes The expression is: ; in: For model parameters and variance scaling factor The likelihood function of the observed data under the given conditions; For model parameters and variance scaling factor difference The likelihood function of the observed data under the given conditions; Acceptance rate of recommended models that can generate nodes The expression is: ; in: for The probability distribution of the number of permutations under each node; for The probability distribution of the number of permutations under each node; This represents the number of deprecated nodes in the candidate model. This represents the number of nodes that can be generated in the current model. The acceptance rate of the suggested model that selects demiseable nodes The expression is: ; in: This represents the number of degradable nodes in the current model. This represents the number of nodes that can be generated by the candidate model. For the newly generated suggested model By simply changing the value of the variance scaling factor, and according to its acceptance criteria, determining whether to accept the model by calculating the ratio of its likelihood function to the proposed distribution using the following formula: ; in: For the construction of MH acceptance criteria; This represents the variance scaling factor in the current model. Generate candidate model variance scaling factor under the condition The probability density; Indicates the variance scaling factor in the candidate model Under the condition of reverse process to generate the variance scaling factor of the current model The probability density; Step S7.3: Exchange different chains, specifically: For temperatures respectively and The two chains, their exchange acceptance probability as follows: ; in: Representation Model At temperature The probability of it happening; Representation Model At temperature The probability of it happening; Representation Model At temperature The probability of it happening; Representation Model At temperature The probability of it happening; Covariance of actual data Represented as: ; in: This is the variance scaling factor; Indicates the initial relative covariance structure; The following exchange acceptance probabilities between chains are obtained under the condition of introducing hyperparameters; ; in: Indicates the first Model parameters of the chain and the corresponding variance scaling factor Likelihood function value under the given conditions; Indicates the first Model parameters of the chain and the corresponding variance scaling factor The likelihood function value under the given conditions.

7. The adaptive estimation method for data error terms in electromagnetic inversion according to claim 6, characterized in that, The four types of update operations include: Model parameter perturbation, in the candidate perturbation node set A node is randomly selected in the middle for perturbation; Hyperparameter perturbation involves perturbing the variance scaling factor component and generating new candidate values ​​centered on the current value using a Gaussian proposal distribution. The generation strategy, in the candidate generation node set A node is randomly selected from the data to generate the data; The extinction strategy, in the set of candidate extinction nodes A node is randomly selected and eliminated.

8. An electromagnetic probability inversion device, characterized in that, include: A processor and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the adaptive estimation method for data error terms for electromagnetic inversion as described in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the adaptive estimation method for data error terms for electromagnetic inversion as described in any one of claims 1 to 7.

Citation Information

Patent Citations

  • Resource electromagnetic exploration efficient probability inversion method, medium and electronic equipment

    CN121069507A