A time domain airborne electromagnetic one-dimensional variable number of layers inversion method and system

Abstract

The application discloses a time domain airborne electromagnetic one-dimensional variable layer number inversion method and system, and the system comprises a data processing module, an inversion prediction module, a parameter screening module and a reverse normalization module; the method comprises the following steps: acquiring measured electromagnetic response data by using the data processing module; performing normalization processing on the measured electromagnetic response data; predicting the normalized electromagnetic response data based on a pre-trained variable layer number inversion prediction model by using the inversion prediction module, to obtain a stratum prediction parameter sequence and a corresponding mask prediction sequence; the stratum prediction parameter sequence has a preset maximum layer number, and the stratum prediction parameter sequence comprises a plurality of stratum parameter combinations; the mask prediction sequence is used for identifying effective stratum parameters; screening effective stratum parameters from the stratum prediction parameter sequence based on the mask prediction sequence by using the parameter screening module; and performing reverse normalization processing on the effective stratum parameters by using the reverse normalization module, to obtain reverse normalized effective stratum parameters.

Description

Technical Field

[0001] This application relates to the field of geophysical airborne electromagnetic data exploration and intelligent inversion technology, and relates to, but is not limited to, a time-domain airborne electromagnetic one-dimensional variable layer number inversion method and system. Background Technology

[0002] Airborne Electromagnetic Method (AEM) is a non-contact underground detection method widely used in mineral exploration, groundwater surveys, and environmental geological monitoring. Time-domain AEM inverts the underground electrical structure by receiving the attenuated response of the secondary field induced by the underground medium, offering advantages such as wide coverage, high acquisition efficiency, and strong adaptability to complex terrain.

[0003] In related technologies, forward modeling and parameter updates using iterative optimization methods are computationally complex and time-consuming, failing to meet the requirements for processing large-scale data. While the development of deep learning technology has significantly improved inversion speed through end-to-end mapping methods based on neural networks, it has introduced new drawbacks. Due to the uncertainty of the number of subsurface media layers, the output dimension of neural networks in related technologies cannot adapt to this variability. This leads to the possibility of invalid parameters in the prediction results for models with a large number of actual layers, reducing the accuracy of the inversion results. Furthermore, for models with a small number of actual layers, the information on the subsurface electrical structure cannot be fully represented, resulting in incomplete inversion results.

[0004] In summary, iterative optimization methods suffer from low processing speed, while neural network methods exhibit rigid output structures. Therefore, a time-domain airborne electromagnetic one-dimensional variable-layer inversion method is urgently needed to address the challenge of balancing computational speed and output flexibility in related technologies, thereby achieving efficient and accurate qualitative interpretation of underground electrical structures. Summary of the Invention

[0005] In view of this, embodiments of this application provide a time-domain airborne electromagnetic one-dimensional variable layer number inversion method and system, which at least solves the problem that related technologies cannot simultaneously take into account the computational speed and adapt to the variability of the number of strata.

[0006] The technical solution of this application embodiment is implemented as follows: In a first aspect, embodiments of this application provide a time-domain airborne electromagnetic one-dimensional variable-layer inversion method. The method is applied to a time-domain airborne electromagnetic one-dimensional variable-layer inversion system, which includes: a data processing module, an inversion prediction module, a parameter filtering module, and an inversion normalization module. The method includes: The data processing module is used to acquire the measured electromagnetic response data of the target measurement point; the measured electromagnetic response data is then normalized to obtain normalized electromagnetic response data. Using the aforementioned inversion prediction module, based on a pre-trained variable-layer inversion prediction model, the normalized electromagnetic response data is predicted to obtain a stratigraphic prediction parameter sequence and a corresponding mask prediction sequence; the stratigraphic prediction parameter sequence and the mask prediction sequence are in one-to-one correspondence; the stratigraphic prediction parameter sequence has a preset maximum layer number, and the stratigraphic prediction parameter sequence includes multiple combinations of stratigraphic parameters; the mask prediction sequence is used to identify whether the corresponding stratigraphic parameter combination is a valid layer parameter; Using the parameter filtering module, the effective layer parameters are filtered from the formation prediction parameter sequence based on the mask prediction sequence; The effective layer parameters are denormalized using the denormalization module to obtain the denormalized effective layer parameters; the denormalization process is the inverse transformation process of the normalization process.

[0007] Secondly, embodiments of this application provide a time-domain airborne electromagnetic one-dimensional variable-layer inversion system, the system comprising: a data processing module, an inversion prediction module, a parameter filtering module, and an inverse normalization module; wherein: The data processing module is used to acquire the measured electromagnetic response data of the target measurement point; and to normalize the measured electromagnetic response data to obtain normalized electromagnetic response data. The inversion prediction module is used to predict the normalized electromagnetic response data based on a pre-trained variable-layer inversion prediction model, to obtain a stratigraphic prediction parameter sequence and a corresponding mask prediction sequence; the stratigraphic prediction parameter sequence and the mask prediction sequence are in one-to-one correspondence; the stratigraphic prediction parameter sequence has a preset maximum layer number, and the stratigraphic prediction parameter sequence includes multiple combinations of stratigraphic parameters; the mask prediction sequence is used to identify whether the corresponding stratigraphic parameter combination is a valid layer parameter. The parameter filtering module is used to filter the effective layer parameters from the formation prediction parameter sequence based on the mask prediction sequence. The denormalization module is used to denormalize the effective layer parameters to obtain the denormalized effective layer parameters; the denormalization process is the inverse transformation process of the normalization process.

[0008] The beneficial effects of the technical solutions provided in this application include at least the following: This application utilizes a data processing module to acquire measured electromagnetic response data of the target measuring point; the measured electromagnetic response data is then normalized to obtain normalized electromagnetic response data. Normalized electromagnetic response data eliminates numerical differences caused by instrument sensitivity or environmental interference, retaining relative variation information reflecting the characteristics of the underground electrical structure, thus enabling high accuracy in subsequent neural network model inversion calculations. Using an inversion prediction module, based on a pre-trained variable-layer inversion prediction model, the normalized electromagnetic response data is predicted to obtain a stratigraphic prediction parameter sequence and a corresponding mask prediction sequence; the stratigraphic prediction parameter sequence and the mask prediction sequence are in one-to-one correspondence; the stratigraphic prediction parameter sequence has a preset maximum layer number and includes multiple combinations of stratigraphic parameters; the mask prediction sequence is used to identify whether the corresponding stratigraphic parameter combination is a valid layer parameter. By utilizing an inversion prediction module, the training and inference processes are unified within a variable-layer inversion prediction model. This model can process measured electromagnetic response data and predict stratigraphic parameter sequences corresponding to the actual subsurface structure. The masked prediction sequence identifies the actual effective layers within the stratigraphic prediction parameter sequence, improving the computational efficiency and accuracy of the inversion results. A parameter filtering module, based on the masked prediction sequence, filters effective layer parameters from the stratigraphic prediction parameter sequence. By converting stratigraphic prediction parameter sequences with a preset maximum layer number into effective layer parameters, invalid stratigraphic parameter combinations are ignored, allowing the final inversion results to adapt to variations in the number of layers in the actual subsurface structure. An inverse normalization module is used to inverse normalize the effective layer parameters, obtaining the inverse-normalized effective layer parameters; inverse normalization is the inverse transformation of normalization. Through inverse normalization, the final inversion result is output, and the final inversion result values ​​can be directly used for subsequent geological interpretation. Therefore, this application, with its variable-layer and effective parameter final inversion results, exhibits high computational speed and adaptability to variations in the number of stratigraphic layers. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein: Figure 1 A flowchart illustrating a time-domain airborne electromagnetic one-dimensional variable layer number inversion method provided in this application embodiment; Figure 2 A schematic diagram of the architecture of the variable layer number inversion prediction model of the method provided in the embodiments of this application; Figure 3 A schematic diagram of the training curve of the mask-weighted loss function of the method provided in the embodiments of this application; Figure 4 A comparative schematic diagram of resistivity inversion results for a first type of 15-layer formation structure provided in this application embodiment; Figure 5 A comparative schematic diagram of resistivity inversion results for a second type of 15-layer formation structure provided in this application embodiment; Figure 6 A comparative schematic diagram of resistivity inversion results for a first type of 20-layer formation structure provided in this application embodiment; Figure 7 A comparative schematic diagram of resistivity inversion results for a second type of 20-layer geological structure using the method provided in this application embodiment; Figure 8 A comparative schematic diagram of resistivity inversion results for a first type of 25-layer formation structure provided in the embodiments of this application; Figure 9 A comparative schematic diagram of resistivity inversion results for a second type of 25-layer formation structure provided in this application embodiment; Figure 10 A comparative schematic diagram of resistivity inversion results for a first type of 30-layer formation structure provided in the embodiments of this application; Figure 11 A comparative schematic diagram of resistivity inversion results for a second type of 30-layer formation structure provided in this application embodiment; Figure 12 This is a schematic diagram of the structure of a time-domain airborne electromagnetic one-dimensional variable layer number inversion system provided in an embodiment of this application. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. The following embodiments are used to illustrate this application, but are not intended to limit the scope of this application. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0011] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0012] It should be noted that the terms "first, second, and third" used in the embodiments of this application are merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first, second, and third" can be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0013] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which the embodiments of this application pertain. It should also be understood that terms such as those defined in general dictionaries should be understood to have a meaning consistent with their meaning in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0014] This application provides a time-domain airborne electromagnetic one-dimensional variable layer number inversion method. The method is applied to a time-domain airborne electromagnetic one-dimensional variable layer number inversion system. The system includes: a data processing module, an inversion prediction module, a parameter filtering module, and an inversion normalization module. Figure 1 A flowchart illustrating a time-domain airborne electromagnetic one-dimensional variable layer number inversion method provided in this application embodiment is shown below. Figure 1 As shown, the method includes at least the following steps: Step S110: Using the data processing module, obtain the measured electromagnetic response data of the target measurement point; normalize the measured electromagnetic response data to obtain normalized electromagnetic response data.

[0015] Electromagnetic measurement systems are installed on aircraft or helicopters to rapidly acquire information on the electrical structure of large underground areas through aerial flight. The target measurement point is the current ground location being measured, and each target measurement point corresponds to a set of data in the generated measured electromagnetic response data.

[0016] The measured electromagnetic response data were recorded by the receiving coil on the aircraft at the target measuring point. The measured electromagnetic response data can reflect the differences in the conductivity of underground rocks.

[0017] The measured electromagnetic response data were normalized to a standard normal distribution with a mean of 0 and a standard deviation of 1, so as to unify the numerical scale.

[0018] Normalized electromagnetic response data eliminates numerical differences caused by instrument sensitivity or environmental interference, and retains relative change information reflecting the characteristics of underground electrical structure, thus enabling high accuracy in subsequent neural network model inversion calculations.

[0019] Step S120: Using the inversion prediction module, based on the pre-trained variable layer number inversion prediction model, the normalized electromagnetic response data is predicted to obtain a stratigraphic prediction parameter sequence and a corresponding mask prediction sequence; the stratigraphic prediction parameter sequence and the mask prediction sequence are in one-to-one correspondence; the stratigraphic prediction parameter sequence has a preset maximum layer number, and the stratigraphic prediction parameter sequence includes multiple sets of stratigraphic parameter combinations; the mask prediction sequence is used to identify whether the corresponding stratigraphic parameter combination is a valid layer parameter.

[0020] The inversion prediction module is equipped with a pre-trained variable layer number inversion prediction model (hereinafter referred to as the variable layer number inversion prediction model). The variable layer number inversion prediction model is a neural network trained based on historical electromagnetic response data and historical stratigraphic label data. By performing forward propagation, it establishes a mapping from electromagnetic response data to stratigraphic parameter combinations.

[0021] The input to the variable layer number inversion prediction model is the normalized electromagnetic response data, and the output is the formation prediction parameter sequence and the corresponding mask prediction sequence.

[0022] The masked prediction sequence assigns semantics to each combination of stratigraphic parameters. For example, the masked prediction sequence consists of 0s and 1s. 1s indicate that the corresponding combination of stratigraphic parameters is a valid stratigraphic parameter set; 0s indicate that the corresponding other combination of stratigraphic parameters is invalid and meaningless. For instance, after inverse prediction of a target measurement point, the output is a stratigraphic prediction parameter sequence of length 39. If the corresponding masked prediction sequence has the first 25 bits as 1s and the last 14 bits as 0s, it indicates that the inversion result for that target measurement point corresponds to a layer thickness of 25, and the parameters at the last 14 positions can be ignored.

[0023] By utilizing the inversion prediction module, the training and inference processes are unified in the variable layer number inversion prediction model, which can process measured electromagnetic response data and predict the stratigraphic prediction parameter sequence corresponding to the actual underground structure. The mask prediction sequence identifies the actual effective layers in the stratigraphic prediction parameter sequence, thereby improving the computational efficiency and accuracy of the inversion results of the method in this application.

[0024] Step S130: Using the parameter filtering module, based on the mask prediction sequence, filter the effective layer parameters from the formation prediction parameter sequence.

[0025] The parameter filtering module extracts valid layer parameters from the formation prediction parameter sequence based on the mask values ​​in the mask prediction sequence. The mask prediction sequence includes multiple binary identifiers, each with a value of 1 or 0, indicating whether the formation parameter combination at the corresponding location is a valid layer parameter or an invalid fill.

[0026] By sequentially reading each mask value in the mask prediction sequence, when the mask value is 1, the corresponding resistivity prediction value and layer thickness prediction value in the formation prediction parameter sequence are used as effective layer parameters. When the mask value is 0, the corresponding resistivity prediction value and layer thickness prediction value are ignored.

[0027] By converting the stratigraphic prediction parameter sequence with a preset maximum number of layers into effective layer parameters and ignoring invalid combinations of stratigraphic parameters, the final inversion results can adapt to the actual changes in the number of layers in the underground structure.

[0028] Step S140: Using the inverse normalization module, the effective layer parameters are inversely normalized to obtain the inversely normalized effective layer parameters; the inverse normalization process is the inverse transformation process of the normalization process.

[0029] The effective layer parameters are the normalized predicted values ​​of effective resistivity and effective layer thickness. Inverse normalization is performed on the selected effective layer parameters, restoring them to their original numerical values. Inverse normalization is the inverse operation of normalization. It uses the same normalization parameters as in the training phase of the variable layer number inversion prediction model to perform the inverse operation on the input values, mapping them back to the original physical units. For example, if the resistivity of 100 ohm-meters is normalized to 0.5 during training, then when the effective resistivity prediction value is 0.5, the inverse normalization module will restore the effective resistivity prediction value to 100 ohm-meters.

[0030] After inverse normalization, the output resistivity values ​​are in ohm-meters, and the layer thickness values ​​are in meters. This output is the final inversion result, which can be directly used for subsequent geological interpretation and mapping.

[0031] This application utilizes a data processing module to acquire measured electromagnetic response data of the target measuring point; the measured electromagnetic response data is then normalized to obtain normalized electromagnetic response data. Normalized electromagnetic response data eliminates numerical differences caused by instrument sensitivity or environmental interference, retaining relative variation information reflecting the characteristics of the underground electrical structure, thus enabling high accuracy in subsequent neural network model inversion calculations. Using an inversion prediction module, based on a pre-trained variable-layer inversion prediction model, the normalized electromagnetic response data is predicted to obtain a stratigraphic prediction parameter sequence and a corresponding mask prediction sequence; the stratigraphic prediction parameter sequence and the mask prediction sequence are in one-to-one correspondence; the stratigraphic prediction parameter sequence has a preset maximum layer number and includes multiple combinations of stratigraphic parameters; the mask prediction sequence is used to identify whether the corresponding stratigraphic parameter combination is a valid layer parameter. By utilizing an inversion prediction module, the training and inference processes are unified within a variable-layer inversion prediction model. This model can process measured electromagnetic response data and predict stratigraphic parameter sequences corresponding to the actual subsurface structure. The masked prediction sequence identifies the actual effective layers within the stratigraphic prediction parameter sequence, improving the computational efficiency and accuracy of the inversion results. A parameter filtering module, based on the masked prediction sequence, filters effective layer parameters from the stratigraphic prediction parameter sequence. By converting stratigraphic prediction parameter sequences with a preset maximum layer number into effective layer parameters, invalid stratigraphic parameter combinations are ignored, allowing the final inversion results to adapt to variations in the number of layers in the actual subsurface structure. An inverse normalization module is used to inverse normalize the effective layer parameters, obtaining the inverse-normalized effective layer parameters; inverse normalization is the inverse transformation of normalization. Through inverse normalization, the final inversion result is output, and the final inversion result values ​​can be directly used for subsequent geological interpretation. Therefore, this application, with its variable-layer and effective parameter final inversion results, exhibits high computational speed and adaptability to variations in the number of stratigraphic layers.

[0032] Optionally, the measured electromagnetic response data includes: the center time of the time channel and the corresponding electromagnetic response value; wherein the electromagnetic response value is obtained by observation at multiple preset center times of the time channel.

[0033] In airborne electromagnetic surveying, the center time of a time channel refers to multiple sampling time points arranged at fixed time intervals for recording the electromagnetic response signal. For example, recordings are made at a series of times, from 0.1 milliseconds, 0.2 milliseconds to 10 milliseconds after the pulse transmission. The electromagnetic response value is the voltage signal or magnetic field strength signal measured by the receiving coil at each of these corresponding times. The voltage signal or magnetic field strength signal reflects the strength of the induction of the transmitted pulse by the underground medium.

[0034] Optionally, the normalization process for the measured electromagnetic response data to obtain normalized electromagnetic response data includes: normalizing the center time of the time channel to obtain a normalized center time of the time channel; normalizing the electromagnetic response value to obtain a normalized electromagnetic response value; and obtaining the normalized electromagnetic response data based on the normalized center time of the time channel and the normalized electromagnetic response value; the parameter values ​​used in the normalization process are the same as the parameter values ​​used in the training phase of the variable layer number inversion prediction model.

[0035] During the training phase of the variable layer number inversion prediction model, the following parameter values ​​in formulas (1) to (4) are calculated from the training set of historical electromagnetic response data and historical stratigraphic label data: , , , , In the inversion stage of the variable layer number inversion prediction model, the measured electromagnetic response data are normalized using the above parameter values.

[0036] In formulas (1) to (4), This represents the mean value at the center of the historical timeline. The standard deviation represents the central moment of the historical timeline. This represents the mean of historical electromagnetic response values. This represents the standard deviation of historical electromagnetic response values. This is set to prevent logarithmic operations or extremely small positive numbers with a denominator of zero during the training phase.

[0037] The time center time is normalized as shown in formula (1): Formula (1); in, Indicates the first The central moment of the time channel. This is the normalized center time of the time channel. The normalized center time of the time channel follows an approximately standard normal distribution with a mean of 0 and a standard deviation of 1.

[0038] The normalization of the electromagnetic response values ​​is performed according to the sign of the electromagnetic response value, as follows: When the electromagnetic response value is positive, the normalization process for the electromagnetic response value is shown in formula (2): Formula (2); in, Indicates the first The electromagnetic response value corresponding to the center moment of the time channel. This is the normalized electromagnetic response value.

[0039] When the electromagnetic response value is non-positive, the electromagnetic response value is first subjected to a sign-preserving logarithmic transformation, and then normalized to obtain normalized electromagnetic response data.

[0040] The electromagnetic response value is transformed using a sign-preserving logarithmic transformation as shown in formula (3): Formula (3); in, If it is a sign function, then >0, The value is 1; if <0, The value is -1; if =0, Take 0. This is the electromagnetic response value after sign-preserving logarithmic transformation. The sign-preserving logarithmic transformation preserves the sign of the electromagnetic response value while logarithmically compressing its absolute value, thus mapping electromagnetic response values ​​that may span multiple orders of magnitude to a relatively concentrated range.

[0041] The electromagnetic response value after sign-preserving logarithmic transformation is normalized as shown in formula (4): Formula (4); in, This is the normalized electromagnetic response value.

[0042] Optionally, the method further includes: acquiring historical electromagnetic response data and historical stratigraphic label data; the historical stratigraphic label data includes historical stratigraphic parameter sequences and historical mask sequences; the historical electromagnetic response data includes the center time of historical time channels and the corresponding historical electromagnetic response values; the historical stratigraphic parameter sequences include multiple sets of historical stratigraphic parameter combinations, each of the historical stratigraphic parameter combinations including historical resistivity and historical layer thickness; the historical mask sequences are used to identify whether the corresponding historical resistivity and historical layer thickness are valid layer parameters; the historical electromagnetic response data, the historical stratigraphic parameter sequences, and the historical mask sequences are divided into training sets, validation sets, and test sets; the historical electromagnetic response data, historical stratigraphic parameter sequences, and historical mask sequences in the training set, the validation set, and the test set are normalized respectively to obtain... The system comprises a normalized training set, a normalized validation set, and a normalized test set. An initial variable-layer number (VLS) inversion prediction model is constructed based on a Transformer encoder and a neural network model. The historical electromagnetic response data corresponding to the normalized training set is used as input features, and the historical stratigraphic parameter sequence and historical mask sequence corresponding to the normalized training set are used as supervision labels to train the initial VLS inversion prediction model. The performance of the initial VLS inversion prediction model is evaluated using the normalized validation set, and the performance is used as the stopping criterion for the training process to obtain candidate VLS inversion prediction models. The loss function of the neural network model is a mask-weighted loss function. The candidate VLS inversion prediction models are tested based on the normalized test set to obtain the pre-trained VLS inversion prediction model.

[0043] Historical electromagnetic response data and corresponding historical stratigraphic label data are generated through a forward modeling process. Specifically, based on the historical resistivity and historical layer thickness of the historical stratigraphic label data, the frequency domain forward modeling engine uses a recursive algorithm to calculate the magnetic field response at a preset frequency point, and then integrates it into a spatial frequency domain magnetic field component through a Hankel transform. This spatial frequency domain magnetic field component is then transformed into a time domain step response, and convolved with the time derivative of the known emission current waveform to calculate the electromagnetic response value at the set observation time channel, i.e., the historical electromagnetic response data.

[0044] The initial variable-layer inversion prediction model is set with a variable number of layers ranging from 15 to 39. The resistivity parameter is preferably in the range of 1-2000 ohm-meters, and the layer thickness satisfies the constraints of shallow and deep layers with continuous increases.

[0045] In this embodiment of the application, the ratio of data samples in the training set, validation set, and test set is 70%:20%:10%.

[0046] Before training the initial variable-layer inversion prediction model, the mean and standard deviation of the center time of historical time channels and the mean and standard deviation of historical electromagnetic response values ​​are calculated using the training set of historical electromagnetic response data. A very small positive number is set to prevent logarithmic operations or division by zero. Based on these parameter values, the historical electromagnetic response data, historical stratigraphic parameter sequences, and historical mask sequences in the training, validation, and test sets are normalized to obtain normalized training, validation, and test sets, respectively.

[0047] The historical resistivity, historical layer thickness, and historical mask sequence in the historical stratigraphic label data were normalized respectively.

[0048] The normalization of historical resistivity is shown in formula (5): Formula (5); in, Indicates the first The historical resistivity of a finite layer , These represent the logarithmic values ​​of the resistivity in the training set, respectively. The mean and standard deviation; This is the normalized resistivity label value.

[0049] The thickness of the historical layer is normalized as shown in formula (6): Formula (6); in, Indicates the first A rich historical layer; and These represent the minimum and maximum values ​​of the historical layer thickness in the training set, respectively. The normalized layer thickness label value is scaled to the [0,1] range.

[0050] Figure 2 A schematic diagram of the architecture of the variable-layer-number inversion prediction model of the method provided in the embodiments of this application. For example... Figure 2 As shown, the input features of the model are normalized historical electromagnetic response data with a shape of (seq_len, 2), representing the center time of seq_len historical time channels and their corresponding historical electromagnetic response values. This input feature is processed by an input mapping layer (Dense(96)) through feature mapping and linear transformation to obtain a 96-dimensional temporal feature vector. The 96-dimensional temporal feature vector (input x) then enters the core encoder, which consists of three identical Transformer coding blocks concatenated. Figure 2The code is labeled "Transformer encoding block × 3". The output of the first Transformer encoding block serves as the input to the second Transformer encoding block, and the output of the second Transformer encoding block serves as the input to the third Transformer encoding block. The operation of the first Transformer encoding block is as follows: "Input x" enters the multi-head self-attention (4-head) module, used to capture global dependencies between temporal channels. The output of the multi-head self-attention module passes through a random dropout layer with a dropout rate of 0.1. Its output is added to the input x through a residual branch. The result of the addition operation is processed by layer normalization to obtain the intermediate feature x'. The intermediate feature x' enters the feedforward network (Dense(192, ReLU)-Dense(96)), which consists of two fully connected layers connected in series. The output of the feedforward network passes through a dropout layer with a dropout rate of 0.1. The output and x' are then added together through another residual branch, and normalized by a second layer to complete the computation of the first Transformer coding block. After the depth temporal features are extracted by three cascaded Transformer coding blocks, the feature sequence output by the third Transformer coding block is flattened and passed sequentially through a high-dimensional fully connected layer (Dense(256)) and a nonlinear activation function (ReLU), a dropout layer with a dropout rate of 0.2, and a fully connected output layer (Dense(out_len×2)). The data shape is adjusted to (out_len, 2) by Reshape((out_len,2)). The output is a sequence of formation prediction parameters with a shape of (out_len, 2). Here, out_len is the preset maximum number of layers, and each row of this sequence corresponds to a combination of formation parameters containing predicted resistivity and predicted layer thickness. The output of this combination of formation parameters, together with the simultaneously generated mask prediction sequence, provides a unified framework for the system's parameter filtering module to select effective formation parameters from the formation prediction parameter sequence.

[0051] Optionally, the method further includes: the dimension of the historical electromagnetic response data is (N, 33, 2); where N represents the number of samples of the historical electromagnetic response data, 33 represents the number of center moments of the historical time channel, and 2 represents the center moment of each historical time channel and the corresponding historical electromagnetic response value; the dimension of the historical mask sequence is (N, 39), where 39 represents the preset maximum number of layers.

[0052] In some embodiments, 33 may be the number of historical time-center moments that strike a balance between signal resolution, computational efficiency, and neural network processing power, characterizing the main features of historical electromagnetic response data without making the data too large.

[0053] The dimension (N, 39) of the history mask sequence consists of multiple binary identifiers corresponding to the output history resistivity and history layer thickness, used to identify whether the output history resistivity and history layer thickness are valid layer parameters. 1 indicates a valid layer parameter, and 0 indicates an invalid padding bit. Therefore, the number of valid layer parameters is usually less than 39.

[0054] In this embodiment, during the training phase of the variable layer number inversion prediction model, the effective layer parameters correspond to the effective historical resistivity and effective historical layer thickness, enabling the variable layer number inversion prediction model to effectively fit the historical resistivity and effective historical layer thickness. During the application phase of the variable layer number inversion prediction model, the effective layer parameters are identified as the predicted values ​​of effective resistivity and effective layer thickness.

[0055] For historical electromagnetic response data, the total number of layers is denoted as... ,and The value range is 15 to 40; since the last layer is a half-space, only the previous layer is considered. The layer performs supervision and prediction. Historical electromagnetic response data can be represented as: Formula (7); Where X refers to historical electromagnetic response data, R is a real number field, and 33 represents the number of central moments of historical time channels, which is the fixed time channel length.

[0056] Before inputting historical electromagnetic response data into the initial variable-layer inversion prediction model based on the Transformer encoder, a linear transformation is first performed on the historical electromagnetic response data through the mapping layer (i.e., a fully connected layer) at the input of the model. This linear transformation projects the historical electromagnetic response data X of dimension (33,2) onto a high-dimensional feature space, so that the subsequent Transformer encoder can effectively extract temporal correlation features. The linear transformation is shown in Equation (8): Formula (8); in, It is the weight matrix of the mapping layer, with dimensions (2,D). Let be the bias vector, with dimensions (1, D). After this linear transformation, the output initial features are... The shape is (33,D), where D is the preset dimension of the hidden layer of the model.

[0057] Using historical electromagnetic response data X as the independent variable, a prediction matrix of 39 rows and 2 columns is output after nonlinear transformation of the initial variable-layer inversion prediction model. Prediction matrix Each row i corresponds to the predicted resistivity and layer thickness of the i-th layer. This is the prediction matrix of the initial variable layer number inversion prediction model. As shown in formula (9): Formula (9); in, This indicates the preset maximum number of layers. In this embodiment, since the total number of layers n in the historical stratigraphic model ranges from 15 to 40, and the last layer is a half-space, only the first n-1 layers are predicted. =39. This represents the nonlinear mapping function implemented by the initial variable-layer inversion prediction model. Output prediction matrix. Each row corresponds to a set of predicted resistivity and layer thickness values.

[0058] Because the actual number of underground electrical structure layers is uncertain, while the output dimension of the neural network needs to remain fixed, this embodiment uses a combination of zero-padding and single masking for representation. Let the total number of layers in the historical stratigraphic model be n, where the nth layer is an infinitely extending half-space layer without a finite thickness. Let K = n-1, where K represents the number of finite-thickness layers participating in the network prediction. For historical stratigraphic label data with a total of n layers, this embodiment only uses the resistivity and thickness of the first K finite-thickness layers as network output labels; the nth half-space layer is not output as a finite-thickness layer parameter and is not included in subsequent zero-padding terms. To ensure a unified output dimension for historical stratigraphic label data with different layers, from row K+1 to row... The rows are zero-filled using (0,0). The corresponding historical stratigraphic parameter sequence is shown in formula (10): Formula (10); in, This represents the historical resistivity of the i-th finite thickness layer. This represents the historical layer thickness of the i-th finite thickness layer. From row K+1 to row Zero-filled items are used to supplement fixed output dimensions and do not represent the actual strata. This represents the maximum number of parameter pairs for the finite thickness layer output by the neural network. In this embodiment, the total number of layers in the historical stratigraphic model ranges from 15 to 40, with the last layer being a half-space layer. Therefore, the maximum number of finite thickness layers is 39. .

[0059] Meanwhile, to distinguish between the real effective layer and the padding value, the corresponding historical mask sequence M is constructed as shown in formula (11): Formula (11); in, .

[0060] By concatenating the historical stratigraphic parameter sequence Y with the historical mask sequence M according to their positions, a prediction matrix of masked historical stratigraphic label data is constructed. As shown in formula (12): Formula (12); in, The first two components represent the historical resistivity and historical layer thickness, respectively, and the third component represents the corresponding historical mask value, which is used to identify whether the position belongs to the effective layer parameters.

[0061] By using the above method, samples with different number of layers can be incorporated into the same training framework for learning without changing the network output dimension, thereby enabling the expression of a stratum model with a variable number of layers from 15 to 39.

[0062] In some embodiments, the total number of layers ranging from 15 to 40 represents the inversion prediction range designed and optimized by the variable layer number inversion prediction model. This is a parameter pre-set when constructing historical stratigraphic label data in this application embodiment to cover the typical complexity in the target exploration scenario. In actual geological exploration, there may be simple layer structures corresponding to fewer than 14 effective layer parameters, i.e., the total number of layers in the historical stratigraphic label data is less than 15.

[0063] Optionally, the mask-weighted loss function is configured to: calculate the error between the predicted output value of the historical stratigraphic parameter combination and the historical stratigraphic parameter combination for the historical stratigraphic parameter combination that is identified as a valid stratigraphic parameter in the historical mask sequence; and ignore the prediction error for the historical stratigraphic parameter combination that is identified as invalid padding in the historical mask sequence.

[0064] Optionally, the predicted output value corresponding to the combination of historical stratigraphic parameters includes the predicted output value of historical resistivity and the predicted output value of historical layer thickness; the mask weighted loss function is shown in formula (13): Formula (13); in, For mask-weighted loss function, To preset the maximum number of layers, The i-th historical mask value in the historical mask sequence. For the i-th historical resistivity, For the i-th historical layer, The predicted output value for the i-th historical resistivity is... The predicted output value is the thickness of the i-th historical layer. It is a very small positive number.

[0065] The mask-weighted loss function calculates the prediction error only when the mask value is 1, ignoring the filling positions when the mask value is 0.

[0066] To ensure that training only applies to valid layer parameters, this application employs a mask-weighted loss function. The mask-weighted loss function uses the prediction matrix of masked historical stratigraphic label data as the supervision label. During calculation, the first two components of each row in the supervision label matrix are extracted as historical resistivity and historical layer thickness labels, and the third component is extracted as the historical mask sequence value. Prediction errors are accumulated only at valid layer locations identified by the historical mask sequence, avoiding invalid fill locations from participating in parameter updates during model training.

[0067] Figure 3 A schematic diagram of the training curve of the mask-weighted loss function of the method provided in the embodiments of this application. Figure 3 As shown, the "Loss Curve" represents the curve of the mask-weighted loss function, showing how the loss value of the loss function on the training and validation sets (Val) changes with the number of iterations during training. The vertical axis represents the loss value, and the horizontal axis represents the number of training iterations (200 in total). The blue curve represents the loss value on the training set, and the orange curve represents the loss value on the validation set. In the early stages of training (approximately the first 50 iterations), the loss value decreases rapidly; subsequently, both curves decrease slowly and tend to stabilize, eventually fluctuating around a low loss value (approximately 0.008). This indicates that the mask-weighted loss function effectively guides the optimization of model parameters, resulting in stable convergence during training and no overfitting.

[0068] This loss function calculates the error only at the location of the true effective layer, thereby reducing the interference of invalid padding values ​​on model training and improving the network's ability to learn and predict the structure with varying number of layers.

[0069] Optionally, the mask prediction sequence includes multiple binary identifiers; the binary identifiers include a first value and a second value; wherein: the first value is used to indicate that the corresponding combination of formation parameters is a valid formation parameter; the second value is used to indicate that the corresponding combination of formation parameters is an invalid fill.

[0070] Optionally, the formation prediction parameter sequence includes resistivity prediction values ​​and layer thickness prediction values; the effective layer parameters include effective resistivity prediction values ​​and effective layer thickness prediction values; the step of filtering the effective layer parameters from the formation prediction parameter sequence based on the mask prediction sequence includes: traversing each of the binary identifiers in the mask prediction sequence; when the binary identifier is a first value, determining the corresponding resistivity prediction value as an effective resistivity prediction value; determining the corresponding layer thickness prediction value as an effective layer thickness prediction value; when the binary identifier is a second value, ignoring the resistivity prediction value and the layer thickness prediction value in the corresponding formation parameter combination.

[0071] By filtering the stratigraphic prediction parameter sequence using a masked prediction sequence, valid prediction results corresponding to the actual number of subsurface layers can be accurately separated. This avoids the interference of invalid parameters introduced by a fixed output dimension on the reliability of the inversion results, thus ensuring the purity and geological rationality of the final output results of the variable layer number inversion prediction method.

[0072] To verify the generalization performance of the variable layer number inversion prediction model in this application, Figures 4-11 The resistivity inversion results of the pre-trained variable layer number inversion prediction model on the test set are shown. Figures 4-11 In the diagram, the horizontal axis represents resistivity (ρ), measured in ohm-meters (Ω·m), using a logarithmic scale, while the vertical axis represents depth, measured in meters (m), increasing downwards from the Earth's surface. Figures 4-11 In the diagram, the solid blue line (True) represents the known true stratigraphic structure curve of the test sample, and the inflection points of the solid blue line correspond to the resistivity and depth of different strata. The dashed orange line (Pred) represents the predicted curve obtained by the method of this application. A direct comparison demonstrates that for stratigraphic structures with different total number of layers (15, 20, 25, 30, etc.), the predicted curves of this application highly match the true stratigraphic structure curves in terms of resistivity and depth, proving that the variable-layer-number inversion prediction model of this application has good generalization ability when handling variable-layer-number inversion tasks.

[0073] Figure 4 A schematic diagram showing the comparison of resistivity inversion results for a first type of 15-layer geological structure using the method provided in this application embodiment. Figure 5 A schematic diagram comparing the resistivity inversion results of the method provided in this application for a second type of 15-layer geological structure. (See attached diagram.) Figure 4 and Figure 5 As shown, Figure 4 The comparison between the predicted curve of the method and the actual model curve is shown for a 15-layer test sample 1. Figure 5This shows the comparison between the predicted curve and the actual model curve for another 15-layer test sample 2. Figure 4 and Figure 5 Both sets of comparisons show that the method in this application can achieve a high degree of agreement between the predicted curve and the real model for different samples with a 15-layer structure.

[0074] Figure 6 A schematic diagram showing the comparison of resistivity inversion results for a first type of 20-layer geological structure using the method provided in this application embodiment. Figure 7 A schematic diagram comparing the resistivity inversion results of the method provided in this application for a second type of 20-layer geological structure. (See attached diagram.) Figure 6 and Figure 7 As shown, Figure 6 and Figure 7 The predicted curves and actual model curves of the described method for two different 20-layer geological structures are compared. The comparison results show that for the 20-layer geoelectric structure, the described method is stable on different samples and can effectively recover the resistivity of each layer.

[0075] Figure 8 A schematic diagram showing the comparison of resistivity inversion results for a first type of 25-layer geological structure using the method provided in this application embodiment. Figure 9 A schematic diagram comparing the resistivity inversion results of the method provided in this application for a second type of 25-layer geological structure. (See attached diagram.) Figure 8 and Figure 9 As shown, Figure 8 and Figure 9 The predicted curves and actual model curves for two test samples with 25 stratigraphic layers are presented respectively. It can be seen that when the number of layers increases to 25, the prediction results of the method still maintain good consistency with the actual model, verifying its ability to handle more complex models.

[0076] Figure 10 A schematic diagram showing the comparison of resistivity inversion results for a first type of 30-layer geological structure using the method provided in this application embodiment. Figure 11 A schematic diagram comparing the resistivity inversion results of the method provided in this application for a second type of 30-layer geological structure. (See attached diagram.) Figure 10 and Figure 11 As shown, Figure 10 and Figure 11 The predicted curves and actual model curves for two test samples of 30-layer stratigraphic structures are shown. Even near the upper limit of the preset layer number range, the predicted curves of the method for different samples still closely match the actual curves, demonstrating the effectiveness and robustness of the variable layer number inversion prediction model in handling highly complex stratigraphic structures.

[0077] This application employs a parameter representation method combining zero-padding and single-mask, enabling historical stratigraphic label data from different layers to be trained within the same initial variable-layer inversion prediction model framework. Simultaneously, a mask-weighted loss function avoids interference from invalid padding values ​​during training. Compared to traditional point-by-point iterative inversion, this application eliminates the need for repeated calls to the forward solver during the inference phase, directly outputting effective layer parameters from measured electromagnetic response data, thereby improving the efficiency of large-scale data processing. The temporal feature extraction network based on the Transformer encoder can more fully utilize the global correlation information between early and late time traces, enhancing the joint representation ability of shallow and deep structures.

[0078] Figure 12 A schematic diagram of a time-domain airborne electromagnetic one-dimensional variable layer number inversion system provided in this application embodiment is shown below. Figure 12 As shown, this application proposes a time-domain airborne electromagnetic one-dimensional variable-layer inversion system. The system 1200 includes: a data processing module 1210, an inversion prediction module 1220, a parameter filtering module 1230, and an inverse normalization module 1240; wherein: The data processing module 1210 is used to acquire the measured electromagnetic response data of the target measurement point; and to normalize the measured electromagnetic response data to obtain normalized electromagnetic response data. The inversion prediction module 1220 is used to predict the normalized electromagnetic response data based on a pre-trained variable-layer inversion prediction model to obtain a stratigraphic prediction parameter sequence and a corresponding mask prediction sequence; the stratigraphic prediction parameter sequence and the mask prediction sequence are in one-to-one correspondence; the stratigraphic prediction parameter sequence has a preset maximum layer number, and the stratigraphic prediction parameter sequence includes multiple sets of stratigraphic parameter combinations; the mask prediction sequence is used to identify whether the corresponding stratigraphic parameter combination is a valid layer parameter. The parameter filtering module 1230 is used to filter the effective layer parameters from the formation prediction parameter sequence based on the mask prediction sequence; The denormalization module 1240 is used to denormalize the effective layer parameters to obtain the denormalized effective layer parameters; the denormalization process is the inverse transformation process of the normalization process.

[0079] It should be noted that the description of the above system embodiments is similar to the description of the above method embodiments, and has similar beneficial effects. For technical details not disclosed in the system embodiments of this application, please refer to the description of the method embodiments of this application for understanding.

[0080] It should be noted that, in the embodiments of this application, if the above-described time-domain airborne electromagnetic one-dimensional variable-layer inversion method is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of this application, or the part that contributes to the related technology, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause an electronic device to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, mobile hard drives, read-only memory (ROM), magnetic disks, or optical disks. Thus, the embodiments of this application are not limited to any specific hardware and software combination.

[0081] Correspondingly, embodiments of this application provide a computer-readable storage medium storing a computer program thereon. When executed by a processor, this computer program implements the steps in any of the time-domain airborne electromagnetic one-dimensional variable layer number inversion methods described in the above embodiments. Correspondingly, embodiments of this application also provide a computer program product, which, when executed by a processor of an electronic device, is used to implement the steps in any of the time-domain airborne electromagnetic one-dimensional variable layer number inversion methods described in the above embodiments.

[0082] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the above-described embodiments are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0083] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0084] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0085] The units described above as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units; some or all of the units may be selected to achieve the purpose of the embodiments of this application according to actual needs. In addition, each functional unit in the embodiments of this application may be fully integrated into one processing unit, or each unit may be a separate unit, or two or more units may be integrated into one unit; the integrated unit may be implemented in hardware or in the form of hardware plus software functional units.

[0086] Alternatively, if the integrated units described above are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to related technologies, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause the device automatic test line to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0087] The methods disclosed in the several method embodiments provided in this application can be arbitrarily combined to obtain new method embodiments without conflict. The features disclosed in the several method or device embodiments provided in this application can be arbitrarily combined to obtain new method embodiments or device embodiments without conflict.

[0088] The above description is merely an embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A time-domain airborne electromagnetic one-dimensional variable-layer inversion method, characterized in that, The method is applied to a time-domain airborne electromagnetic one-dimensional variable-layer inversion system, the system comprising: a data processing module, an inversion prediction module, a parameter filtering module, and an inverse normalization module; the method comprises: The data processing module is used to acquire the measured electromagnetic response data of the target measurement point; the measured electromagnetic response data is then normalized to obtain normalized electromagnetic response data. Using the aforementioned inversion prediction module, based on a pre-trained variable-layer inversion prediction model, the normalized electromagnetic response data is predicted to obtain a stratigraphic prediction parameter sequence and a corresponding mask prediction sequence; the stratigraphic prediction parameter sequence and the mask prediction sequence are in one-to-one correspondence; the stratigraphic prediction parameter sequence has a preset maximum layer number, and the stratigraphic prediction parameter sequence includes multiple combinations of stratigraphic parameters; the mask prediction sequence is used to identify whether the corresponding stratigraphic parameter combination is a valid layer parameter; Using the parameter filtering module, the effective layer parameters are filtered from the formation prediction parameter sequence based on the mask prediction sequence; The effective layer parameters are denormalized using the denormalization module to obtain the denormalized effective layer parameters; the denormalization process is the inverse transformation process of the normalization process.

2. The method according to claim 1, characterized in that, The method further includes: Historical electromagnetic response data and historical stratigraphic label data are acquired; the historical stratigraphic label data includes historical stratigraphic parameter sequences and historical mask sequences; the historical electromagnetic response data includes the center time of historical time channels and the corresponding historical electromagnetic response values; the historical stratigraphic parameter sequences include multiple sets of historical stratigraphic parameter combinations, each of which includes historical resistivity and historical layer thickness; the historical mask sequences are used to identify whether the corresponding historical resistivity and historical layer thickness are valid layer parameters. The historical electromagnetic response data, the historical stratigraphic parameter sequence, and the historical mask sequence are divided into training set, validation set, and test set; The historical electromagnetic response data, historical stratigraphic parameter sequences, and historical mask sequences in the training set, the validation set, and the test set are normalized respectively to obtain normalized training set, normalized validation set, and normalized test set. An initial variable layer number inversion prediction model is constructed based on a Transformer encoder and a neural network model; The historical electromagnetic response data corresponding to the normalized training set is used as input features, and the historical stratigraphic parameter sequence and historical mask sequence corresponding to the normalized training set are used as supervision labels to train the initial variable layer number inversion prediction model. The performance of the initial variable layer number inversion prediction model is evaluated using the normalized validation set, and the performance is used as the stopping criterion for the training process to obtain candidate variable layer number inversion prediction models; wherein, the loss function of the neural network model is a mask-weighted loss function; The candidate variable layer number inversion prediction model is tested based on the normalized test set to obtain the pre-trained variable layer number inversion prediction model.

3. The method according to claim 2, characterized in that, The method further includes: The dimension of the historical electromagnetic response data is (N, 33, 2); where N represents the number of samples of the historical electromagnetic response data, 33 represents the number of center moments of the historical time channel, and 2 represents the historical electromagnetic response value of each center moment of the historical time channel. The dimension of the historical mask sequence is (N, 39), where 39 represents the preset maximum number of layers.

4. The method according to claim 2, characterized in that, The mask-weighted loss function is configured as follows: For the historical stratigraphic parameter combinations that are identified as effective stratigraphic parameters in the historical mask sequence, calculate the error between the predicted output value of the historical stratigraphic parameter combination and the historical stratigraphic parameter combination; For the historical stratigraphic parameter combinations that are identified as invalid padding in the historical mask sequence, the prediction error is ignored.

5. The method according to claim 4, characterized in that, The predicted output value corresponding to the combination of historical stratigraphic parameters includes the predicted output value of historical resistivity and the predicted output value of historical layer thickness; the formula for the mask-weighted loss function is: ； in, For mask-weighted loss function, To preset the maximum number of layers, The i-th historical mask value in the historical mask sequence. For the i-th historical resistivity, For the i-th historical layer, The predicted output value for the i-th historical resistivity is... The predicted output value is the thickness of the i-th historical layer. It is a very small positive number.

6. The method according to claim 1, characterized in that, The mask prediction sequence includes multiple binary identifiers; each binary identifier includes a first value and a second value; wherein: The first value is used to indicate that the corresponding combination of formation parameters is an effective formation parameter; The second value is used to indicate that the corresponding combination of formation parameters is an invalid fill.

7. The method according to claim 6, characterized in that, The formation prediction parameter sequence includes predicted resistivity and predicted layer thickness; the effective layer parameters include predicted effective resistivity and predicted effective layer thickness; the step of filtering the effective layer parameters from the formation prediction parameter sequence based on the mask prediction sequence includes: Iterate through each of the binary identifiers in the mask prediction sequence; When the binary identifier is the first value, the corresponding resistivity prediction value is determined as the effective resistivity prediction value; the corresponding layer thickness prediction value is determined as the effective layer thickness prediction value. When the binary identifier is the second value, the predicted resistivity value and the predicted layer thickness value in the corresponding formation parameter combination are ignored.

8. The method according to claim 1, characterized in that, The measured electromagnetic response data includes: the center time of the time channel and the corresponding electromagnetic response value; wherein, the electromagnetic response value is obtained by observation at multiple pre-set center times of the time channel.

9. The method according to claim 8, characterized in that, The normalization process for the measured electromagnetic response data to obtain normalized electromagnetic response data includes: The time of the time channel center is normalized to obtain the normalized time channel center time. The electromagnetic response value is normalized to obtain the normalized electromagnetic response value. Based on the normalized time channel center time and the normalized electromagnetic response value, the normalized electromagnetic response data is obtained. The parameter values ​​used in the normalization process are the same as those used in the training phase of the variable layer number inversion prediction model.

10. A time-domain airborne electromagnetic one-dimensional variable-layer inversion system, characterized in that, The system includes: a data processing module, an inversion prediction module, a parameter filtering module, and an inversion normalization module; wherein: The data processing module is used to acquire the measured electromagnetic response data of the target measurement point; and to normalize the measured electromagnetic response data to obtain normalized electromagnetic response data. The inversion prediction module is used to predict the normalized electromagnetic response data based on a pre-trained variable-layer inversion prediction model, to obtain a stratigraphic prediction parameter sequence and a corresponding mask prediction sequence; the stratigraphic prediction parameter sequence and the mask prediction sequence are in one-to-one correspondence; the stratigraphic prediction parameter sequence has a preset maximum layer number, and the stratigraphic prediction parameter sequence includes multiple combinations of stratigraphic parameters; the mask prediction sequence is used to identify whether the corresponding stratigraphic parameter combination is a valid layer parameter. The parameter filtering module is used to filter the effective layer parameters from the formation prediction parameter sequence based on the mask prediction sequence. The denormalization module is used to denormalize the effective layer parameters to obtain the denormalized effective layer parameters; the denormalization process is the inverse transformation process of the normalization process.

Images