Time domain airborne electromagnetic inversion method based on deep learning

Abstract

The present invention belongs to the field of the intersection of geophysical exploration and artificial intelligence technology, and discloses a time-domain airborne electromagnetic inversion method based on deep learning, comprising: obtaining a magnetic field response at each frequency point through a recursive algorithm, received formation thickness, random flight altitude, and acquired measured formation resistivity; transforming the magnetic field response through a Hankel transform kernel function integral to obtain a frequency-domain magnetic induction intensity component; generating a time-domain step response for the frequency-domain magnetic induction intensity component using a preset-order Hankel transform coefficient, and convolving the time-domain step response through a Gaussian integral node to obtain electromagnetic response data in a vertical direction; training a long-short-term memory network to be trained using the electromagnetic response data, formation thickness, and random flight altitude as a training set to obtain a long-short-term memory network; and inputting the measured electromagnetic response data into the long-short-term memory network for inversion to obtain a predicted target flight altitude and target formation resistivity.

Description

Technical Field

[0001] The present invention relates to the field of intersectional geophysical exploration and artificial intelligence, and specifically to, but is not limited to, a time-domain airborne electromagnetic inversion method based on deep learning. Background Art

[0002] Airborne electromagnetics is a geophysical exploration method that uses mobile platforms such as aircraft and airships as the system carrier. Due to its advantages such as high data acquisition efficiency, high detection resolution, and the fact that no ground personnel are required to approach the target area, airborne electromagnetics has been widely used in geological surveys, mineral resource exploration, groundwater resource assessment, and environmental monitoring in areas with complex topography and geology. Airborne electromagnetics is subject to constant turbulence during data acquisition, resulting in errors in the data acquisition system's flight altitude recording. Furthermore, vegetation may cover some of the airborne electromagnetic survey area, potentially leading to the loss of flight altitude data for some survey lines. Given that flight altitude is a primary influencing parameter of airborne electromagnetic data and strongly influences the data, it is essential to recover the system's flight altitude during the inversion and interpretation of airborne electromagnetic data.

[0003] Related technologies use the acquired measured flight altitude to recover meter-level flight altitude recording errors and obtain high-precision electromagnetic data fine-scale modeling results. However, when flight altitude data is missing or the recording errors are large, these methods often cannot obtain accurate airborne electromagnetic data inversion results.

[0004] Therefore, how to quickly and accurately measure the flight altitude and formation resistivity becomes an urgent problem to be solved. Summary of the Invention

[0005] In view of this, the time-domain airborne electromagnetic inversion method based on deep learning provided by the embodiment of the present invention at least solves the problem that related technologies cannot effectively and accurately invert flight altitude and formation resistivity.

[0006] According to a first aspect of an embodiment of the present invention, a time-domain airborne electromagnetic inversion method based on deep learning is provided, comprising:

[0007] The frequency domain forward modeling engine calculates the magnetic field response at each frequency point based on a recursive algorithm, the received formation thickness, random flight altitude, and the acquired measured formation resistivity data;

[0008] The magnetic field response is transformed by integrating the Hankel transform kernel function to obtain a frequency domain magnetic induction intensity component;

[0009] Generating a time-domain step response for the frequency-domain magnetic induction intensity component using a preset-order Hankel transform coefficient, and performing convolution processing on the time-domain step response based on a Gaussian integration node to obtain electromagnetic response data in the vertical direction;

[0010] Using the electromagnetic response data, the formation thickness, and the random flight height as a training set, and using the training set to train a long short-term memory network to obtain a long short-term memory network;

[0011] The measured electromagnetic response data is input into the long short-term memory network for inversion to obtain the predicted target flight altitude and target formation resistivity.

[0012] According to a second aspect of an embodiment of the present invention, an electronic device is provided, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; the memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform an operation corresponding to the method described in the first aspect.

[0013] According to a third aspect of an embodiment of the present invention, a computer storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the method according to the first aspect is implemented.

[0014] According to the solution provided by an embodiment of the present invention, the magnetic field response of each frequency point is calculated by a frequency domain forward modeling engine based on a recursive algorithm, the received formation thickness, the random flight height and the acquired measured formation resistivity data; the magnetic field response is transformed by integrating the Hankel transform kernel function to obtain a frequency domain magnetic induction intensity component; a time domain step response is generated for the frequency domain magnetic induction intensity component using a preset order Hankel transform coefficient, and the time domain step response is convolved based on a Gaussian integral node to obtain electromagnetic response data in the vertical direction; the electromagnetic response data, the formation thickness and the random flight height are used as a training set, and the training set is used to train a long short-term memory network to obtain a long short-term memory network; the measured electromagnetic response data is input into the long short-term memory network for inversion to obtain a predicted target flight height and target formation resistivity. In this process, a training set consisting of random flight altitudes, measured formation resistivity, and their calculated airborne electromagnetic responses was established. Finally, a multi-task learning model based on the long short-term memory network was used to synchronously invert altitude and resistivity through a network architecture with shared feature extraction and task branch separation, thereby improving prediction accuracy and efficiency and reducing the risk of overfitting. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive work, among which:

[0016] Figure 1 A schematic flow chart of a time-domain airborne electromagnetic inversion method based on deep learning provided by an embodiment of the present invention;

[0017] Figure 2 A schematic diagram showing the effects of training loss and validation loss during the training process of the model provided in an embodiment of the present invention;

[0018] Figure 3 A schematic diagram illustrating the effect of predicting the loss corresponding to the flight altitude during the training process of the model provided by an embodiment of the present invention;

[0019] Figure 4 A schematic diagram illustrating the effect of predicting the loss corresponding to formation resistivity during the training process of the model provided in an embodiment of the present invention;

[0020] Figure 5 A schematic diagram showing the effect of the inversion result of the target formation resistivity predicted by the first airborne electromagnetic measurement point provided by an embodiment of the present invention;

[0021] Figure 6 A schematic diagram showing the effect of the inversion result of the target formation resistivity predicted by the second airborne electromagnetic measurement point provided by an embodiment of the present invention;

[0022] Figure 7 A schematic diagram showing the effect of the inversion result of the target formation resistivity predicted by the third airborne electromagnetic measurement point provided by an embodiment of the present invention;

[0023] Figure 8 A schematic diagram showing the effect of the inversion result of the target flight altitude predicted by the first airborne electromagnetic measurement point provided by an embodiment of the present invention;

[0024] Figure 9 A schematic diagram showing the effect of the inversion result of the target flight altitude predicted by the second airborne electromagnetic measurement point provided by an embodiment of the present invention;

[0025] Figure 10 A schematic diagram showing the effect of the inversion result of the target flight altitude predicted by the third airborne electromagnetic measurement point provided by an embodiment of the present invention;

[0026] Figure 11 A schematic diagram showing the effect of the inversion result of the target flight altitude predicted on one airborne electromagnetic survey line provided by an embodiment of the present invention;

[0027] Figure 12 A thermal schematic diagram of the inversion result of the target formation resistivity predicted on an airborne electromagnetic survey line provided by an embodiment of the present invention;

[0028] Figure 13 A thermal schematic diagram of the formation resistivity inversion result at a known survey line at a flight altitude provided by an embodiment of the present invention;

[0029] Figure 14 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0031] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be 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.

[0032] It should be pointed out that the terms "first\second\third" involved in the embodiments of the present invention are only used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second\third" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present invention described here can be implemented in an order other than that illustrated or described here.

[0033] Those skilled in the art will understand that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by those skilled in the art in the art to which the embodiments of the present invention pertain. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with those in the context of the prior art and, unless specifically defined as herein, should not be interpreted in an idealized or overly formal sense.

[0034] Figure 1 This is a flow chart of a time-domain airborne electromagnetic inversion method based on deep learning. The time-domain airborne electromagnetic inversion method based on deep learning provided by an embodiment of the present invention can be executed by an electronic device, such as a computer, a server, etc.

[0035] like Figure 1 As shown in Figure 2, the time-domain airborne electromagnetic inversion method based on deep learning includes:

[0036] S101 , calculating the magnetic field response at each frequency point based on a recursive algorithm, received formation thickness, random flight height, and acquired formation resistivity data.

[0037] In the embodiments of the present invention, the formation thickness refers to the thickness of each geological layer, the random flight altitude refers to the height of the measuring instrument relative to the ground, and the formation resistivity data refers to the resistivity value of each formation layer. The magnetic field response at each frequency point can be calculated in a frequency domain forward modeling engine based on a recursive algorithm using the formation thickness, random flight altitude, and acquired formation resistivity data.

[0038] In an embodiment of the present invention, an efficient one-dimensional time-domain airborne electromagnetic forward model is proposed. This model includes an input module that receives input formation parameters (number of layers, resistivity, layer thickness, Cole-Cole parameters), device parameters (transmitter / receiver range, height, waveform type), and time settings. The model's frequency-domain forward modeling engine calculates the magnetic field response at each frequency point based on a recursive algorithm and the received parameters.

[0039] S102. Transform the magnetic field response by integrating the Hankel transform kernel function to obtain a frequency domain magnetic induction intensity component.

[0040] In an embodiment of the present invention, the Hankel transform is a method for achieving such conversion, which completes this process by using the Bessel function as the kernel function, and converts the data describing the magnetic field response from the spatial domain to the frequency domain using the Hankel transform kernel function integral to obtain the frequency domain magnetic induction intensity component.

[0041] S103 , generating a time domain step response for the frequency domain magnetic induction intensity component using a preset order Hankel transform coefficient, and performing convolution processing on the time domain step response based on a Gaussian integration node to obtain electromagnetic response data in the vertical direction.

[0042] In embodiments of the present invention, Gaussian integration nodes refer to the sampling points and weights used during numerical integration, which are selected to accurately approximate the integral result. Frequency-domain magnetic induction intensity components are converted into time-domain step responses using 0.5-order Hankel transform coefficients. This time-domain step response is then precisely convolved using the Gaussian integration node to simulate the propagation of electromagnetic waves in the medium. Ultimately, detailed electromagnetic response data in the vertical direction (i.e., the detection depth direction) is obtained, providing a basis for analyzing underground geological structures. This process integrates frequency-to-time domain conversion, the application of numerical integration methods, and signal processing techniques to accurately reflect the changes in electromagnetic characteristics at different depths. The convolution operation simulates the propagation effects of signals through the system, thereby better understanding the changes in electromagnetic response over time.

[0043] S104: Using the electromagnetic response data, the formation thickness, and the random flight altitude as training set data, and using the training set data to train the long short-term memory network to obtain a long short-term memory network.

[0044] S105: Input the measured electromagnetic response data into the long short-term memory network for inversion to obtain the predicted target flight altitude and target formation resistivity.

[0045] In an embodiment of the present invention, electromagnetic response data, formation thickness, and random flight altitude (10–120 m) are used to form a training set data. A long short-term memory network to be trained is then trained using the training set data to obtain a long short-term memory network. Finally, measured electromagnetic response data are obtained and input into the long short-term memory network for inversion to obtain a predicted target flight altitude and target formation resistivity (the electrical structure of the underground medium).

[0046] Specifically, the measured electromagnetic response data (100×1) can be input into the long short-term memory network for forward propagation to obtain the 29-layer predicted target flight altitude and target formation resistivity.

[0047] In airborne electromagnetic data inversion, electromagnetic response data are essentially time series signals with the following characteristics: 1. Temporal dependence: Electromagnetic field variations at different time points are closely related to the physical evolution of formation parameters. 2. Long-range correlation: Shallow formation parameters may affect early electromagnetic responses, while deeper parameters affect later responses, necessitating the capture of long-term dependencies across time steps. 3. Noise sensitivity: Measured data often contain random noise, requiring the model to possess robust feature extraction capabilities.

[0048] It can be understood that in an embodiment of the present invention, forward modeling is performed based on a recursive algorithm, the received formation thickness, the random flight height and the obtained measured formation resistivity to obtain the magnetic field response of each frequency point; the magnetic field response is transformed by integrating the Hankel transform kernel function to obtain the frequency domain magnetic induction intensity component; the frequency domain magnetic induction intensity component is subjected to a time domain step response using a preset order Hankel transform coefficient, and the time domain step response is convolved based on a Gaussian integral node to obtain electromagnetic response data in the vertical direction; the electromagnetic response data, the formation thickness and the random flight height are used as a training set, and the long short-term memory network to be trained is trained using the training set to obtain a long short-term memory network; the measured electromagnetic response data is input into the long short-term memory network for inversion to obtain the predicted target flight height and target formation resistivity. In this process, a training set consisting of random flight altitudes, measured formation resistivity, and their calculated airborne electromagnetic responses was established. Finally, a multi-task learning model based on the long short-term memory network was used to synchronously invert altitude and resistivity through a network architecture with shared feature extraction and task branch separation, thereby improving prediction accuracy and efficiency and reducing the risk of overfitting.

[0049] In some embodiments of the present invention, S104 may be implemented through S1041 to S1042, which is described in the following steps.

[0050] S1041. Preprocess the training data set to obtain a preprocessed data set, and obtain a predicted flight altitude and a predicted formation resistivity based on the preprocessed data set, the first long short-term memory layer, the second long short-term memory layer, the third long short-term memory layer, the fourth long short-term memory layer, the fifth long short-term memory layer, the discard layer, and the fully connected layer.

[0051] In some embodiments of the present invention, a long short-term memory network (LSTM) consists of five LSTM layers, a dropout layer, and a fully connected layer. The first LSTM layer has 32 hidden units, the second LSTM layer has 64 hidden units, the third LSTM layer has 128 hidden units, the fourth LSTM layer has 256 hidden units, and the fifth LSTM layer has 512 hidden units. The training dataset is preprocessed and then fed into a five-layer LSTM, a dropout layer, and a fully connected layer for processing. Finally, the fully connected layer outputs the predicted flight altitude and formation resistivity. The dropout layer is a random dropout layer.

[0052] S1042. Train the long short-term memory network to be trained based on the predicted flight altitude and the predicted formation resistivity to obtain a long short-term memory network.

[0053] In some embodiments of the present invention, a loss value is calculated based on the predicted flight altitude and the predicted formation resistivity in combination with a loss function, and the loss value is used to adjust the parameters of the long short-term memory network to be trained to achieve the purpose of training until the training conditions are met, and finally a long short-term memory network is obtained.

[0054] In some embodiments of the present invention, preprocessing the training set in S1041 to obtain the preprocessed training set can be implemented through S201 to S202, which is explained in the following steps.

[0055] S201 , normalizing the electromagnetic response data, formation thickness, and random flight height in a training set to obtain first electromagnetic response data, first formation thickness, and first random flight height.

[0056] In some embodiments of the present invention, the electromagnetic response data, formation thickness and random flight height in the training set are standardized (mean-variance normalization) to obtain processed first electromagnetic response data, first formation thickness and first random flight height.

[0057] S202 : Normalize the tag height using a normalized scaler to obtain a first height, and normalize the tag resistivity using a minimum-maximum scaler to obtain a first formation resistivity.

[0058] In some embodiments of the present invention, the training set also includes label data, which includes 29 layers of label heights and 29 layers of label resistivity. The 29 layers of label heights are standardized using a standard scaler, that is, they are standardized using StandardScaler to obtain a first height, and the 29 layers of label resistivity are normalized using a minimum and maximum scaler, that is, they are normalized using MinMaxScaler to obtain a first formation resistivity.

[0059] In some embodiments of the present invention, the predicted flight altitude and predicted formation resistivity obtained based on the preprocessed training set, the first long short-term memory layer, the second long short-term memory layer, the third long short-term memory layer, the fourth long short-term memory layer, the fifth long short-term memory layer, the discard layer and the fully connected layer in S1041 can be achieved through S301 to S303, which is explained in the following steps.

[0060] S301. Input the first electromagnetic response data, the first formation thickness and the first random flight height into the first long short-term memory layer to obtain the first time series feature, and input the first time series feature into the batch normalization layer to obtain the second time series feature.

[0061] In some embodiments of the present invention, the first electromagnetic response data (shape = (100, 1)), the first formation thickness, and the first random flight altitude are input into the first long short-term memory layer, which outputs a 100×32 time series feature, retaining all time steps. The feature dimensions (32 dimensions) are then normalized using a BatchNorm layer to accelerate convergence and reduce the risk of vanishing gradients, resulting in the second time series feature.

[0062] S302. Input the second time series feature into the second long short-term memory layer to obtain a third time series feature, and input the third time series feature into the batch normalization layer to obtain a fourth time series feature.

[0063] In some embodiments of the present invention, the second time series feature is input into the second long short-term memory layer, and a 100×64 feature is output. The feature dimension (64 dimensions) is normalized again through the BatchNorm layer to obtain the fourth time series feature.

[0064] S303 , performing task-independent branch prediction based on the fourth time series feature, the third long short-term memory layer, the fourth long short-term memory layer, and the fifth long short-term memory layer to obtain a predicted flight altitude and a predicted formation resistivity.

[0065] In some embodiments of the present invention, the fourth time series feature is input into the third long short-term memory layer to predict height, and the fourth time series feature is input into the fourth long short-term memory layer and the fifth long short-term memory layer to predict resistivity. By assigning independent branches to different tasks, shared information between tasks is preserved while reducing interference between tasks, thereby improving the overall performance of the model.

[0066] In some embodiments of the present invention, S303 may be implemented through S3031 to S3032, which is described in the following steps.

[0067] S3031. Obtain the predicted flight altitude based on the fourth temporal feature, the third long short-term memory layer, the batch normalization layer, and the dropout layer.

[0068] In some embodiments of the present invention, the fourth time series feature is input into the third long short-term memory layer, and a 100×128 feature is output. The feature dimension (128 dimensions) is normalized by the BatchNorm layer to obtain the fifth time series feature. The fifth time series feature is randomly shielded by 20% of the neurons in the discard layer to prevent overfitting, and the predicted flight height is obtained.

[0069] S3032. Obtain the predicted formation resistivity based on the fourth time series feature, the fourth long short-term memory layer, the fifth long short-term memory layer, the batch normalization layer, and the discard layer.

[0070] In some embodiments of the present invention, the fourth time series feature is input into the fourth long short-term memory layer, outputting 100×256 features, and the feature dimension (256 dimensions) is normalized through the BatchNorm layer to obtain the sixth time series feature. The sixth time series feature is input into the fifth long short-term memory layer for deep processing to obtain 100×512 features, and the feature dimension (512 dimensions) is normalized through the BatchNorm layer to obtain the seventh time series feature. The seventh time series feature is randomly shielded by 20% of neurons through the discard layer to prevent overfitting, and the predicted formation resistivity is obtained.

[0071] In some embodiments of the present invention, S1042 can be implemented through S401 to S402, which is explained through the following steps.

[0072] S401 : Obtain a first loss by predicting a flight altitude and a first altitude, and obtain a second loss by predicting a formation resistivity and a first formation resistivity.

[0073] S402. Obtain a total loss through the first loss and the second loss, and train the long short-term memory network to be trained through the total loss to obtain a long short-term memory network.

[0074] In some embodiments of the present invention, the mean square error between the predicted flight altitude and the first altitude is calculated as the first loss, the mean absolute error between the predicted formation resistivity and the first formation resistivity is calculated as the second loss, the first loss and the second loss are summed to obtain the total loss, and the long short-term memory network to be trained is trained using the total loss to obtain a long short-term memory network.

[0075] Among them, the total loss loss calculation formula is as follows:

[0076] ;

[0077] In the above formula, h_pred is the predicted flight altitude, h_true is the first altitude, r_pred is the predicted formation resistivity, and r_true is the first formation resistivity.

[0078] α = 1, β = 0.5 (initial values). L2 regularization (λ = 0.0001) was added to constrain the weights of the fully connected layers. The Adam optimizer was used with an initial learning rate of 0.001 and dynamic decay using ReduceLROnPlateau. An early stopping mechanism was implemented, terminating training if the validation set loss did not decrease after 30 epochs.

[0079] in, , ;

[0080] In the above formula, MSE represents mean square error, MAE represents mean absolute error, m and n are the number of samples, and yi For the The true value of the sample, For the The predicted value of samples, m and n are the same.

[0081] like Figure 2 As shown, Figure 2 Schematic diagram of the effects of training loss and validation loss during the training process of the model provided in an embodiment of the present invention. Figure 2 The horizontal axis represents the number of training rounds, and the vertical axis represents the loss (mean squared error). This model, as well as the following models, is a long short-term memory network. Training loss refers to the model's performance on the training set, that is, how well the model fits the training data. It is measured by calculating the difference between the model's predicted values ​​and the actual labels. Validation loss refers to the model's performance on the validation set, a subset of data from the training set that is not included in the model parameter update process. Validation loss is used to assess the model's ability to predict unseen data, that is, its generalization ability.

[0082] like Figure 3 As shown, Figure 3 This example illustrates the effect of the model's predicted flight altitude loss during training. After each training iteration, the model makes predictions for each sample in the training set and calculates the loss between these predictions and the actual flight altitude. This loss reflects the model's performance in its current state.

[0083] like Figure 4 As shown, Figure 4 This example illustrates the effect of model training on the loss of formation resistivity predictions. After each training iteration, the model makes predictions for each sample in the training set and calculates the loss between these predictions and the actual formation resistivity. This loss value reflects the model's performance in its current state.

[0084] like Figure 5 、 Figure 6 and Figure 7 As shown, Figure 5 A schematic diagram of the effect of the inversion result of the target formation resistivity predicted by the first airborne electromagnetic measurement point provided by an embodiment of the present invention, Figure 6 A schematic diagram of the effect of the inversion result of the target formation resistivity predicted by the second airborne electromagnetic measurement point provided by an embodiment of the present invention, Figure 7 This is a schematic diagram of the effect of the inversion result of the target formation resistivity predicted by the third airborne electromagnetic measurement point provided by an embodiment of the present invention. Figure 5 、 Figure 6 and Figure 7The difference is that the three measuring points are different. After the model is trained, the model is used to start the inversion. The input is the electromagnetic response (dBz / dt) of 100 time points. The inversion effect of the target formation resistivity predicted by a single measuring point is as follows: Figure 5 , where the measured formation resistivity can be obtained by itself. And Figure 8 、 Figure 9 and Figure 10 As shown, Figure 8 This is a schematic diagram of the effect of the inversion result of the target flight altitude predicted by the first airborne electromagnetic measurement point provided by an embodiment of the present invention. Figure 9 A schematic diagram showing the effect of the inversion result of the target flight altitude predicted by the second airborne electromagnetic measurement point provided by an embodiment of the present invention. Figure 10 A schematic diagram of the effect of the inversion result of the target flight altitude predicted by the third airborne electromagnetic measurement point provided by an embodiment of the present invention. Figure 8 、 Figure 9 and Figure 10 The difference is that the three measurement points are different. After the model is trained, the inversion is started using the model. The input is the electromagnetic response (dBz / dt) at 100 time points. The inversion effect of the target flight altitude predicted by a single measurement point is as follows Figure 8 、 Figure 9 and Figure 10 , where the actual height can be set randomly, such as 10m to 120m.

[0085] like Figure 11 As shown, Figure 11 A schematic diagram of the effect of the inversion result of the target flight altitude predicted on an airborne electromagnetic survey line provided by an embodiment of the present invention. A survey line can be a series of measurements along a specific path, wherein the actual altitude can be randomly set, such as 10m to 120m. Figure 11 The height difference can be calculated from the actual altitude and the predicted target flight altitude. Figure 12 A thermal schematic diagram of the inversion results of the target formation resistivity predicted on an airborne electromagnetic survey line provided by an embodiment of the present invention. Figure 12 for Figure 11 A thermal diagram of the predicted target formation resistivity inversion results on the profile where the survey line is located.

[0086] like Figure 13 As shown, Figure 13 This is a thermal diagram of the inversion results of formation resistivity on a known survey line at a flight height provided by an embodiment of the present invention. The flight height can be set randomly, such as from 10m to 120m.

[0087] In summary, this deep learning method, based on long-short-term memory networks, aims to address the problem of incorrect or missing flight altitude records during the inversion of airborne electromagnetic data, often caused by terrain undulations or ground obstructions. Traditional methods typically rely on accurate records of flight altitude during the inversion process. However, due to irregularities in flight paths or environmental factors, flight altitude is often inaccurately recorded or not recorded at all, thus affecting the accuracy of data inversion and subsequent 3D modeling.

[0088] To address this issue, the present invention employs a long short-term memory (LSTM) network, which can dynamically recover missing flight altitude information and effectively integrate it into the inversion process. By introducing deep learning techniques, the LSTM network can automatically estimate and recover flight altitude during the inversion process based on the long-term dependencies of time series data. This capability is particularly important for areas with complex terrain or numerous obstructions, ensuring accurate inversion even with incomplete or inaccurate flight altitude data.

[0089] In addition, through the gating mechanism and memory units of the long short-term memory network, the model can effectively handle the temporal dependencies in the time series, reduce the impact caused by incomplete altitude records, and thus improve the accuracy of the inversion results. This not only improves the recovery effect of the flight altitude, but also enhances the stability of the overall inversion. The present invention uses the long short-term memory network learning method to not only effectively restore the flight altitude information, but also improve the inversion accuracy, stability and processing efficiency, especially under complex geological and environmental conditions, which has significant advantages. Figure 14 , shows a schematic structural diagram of an electronic device according to an embodiment of the present invention. The specific embodiment of the present invention does not limit the specific implementation of the electronic device.

[0090] like Figure 14 As shown, the electronic device may include: a processor (processor) 502 , a communication interface (Communications Interface) 504 , a memory (memory) 506 , and a communication bus 508 .

[0091] in:

[0092] The processor 502 , the communication interface 504 , and the memory 506 communicate with each other via a communication bus 508 .

[0093] The communication interface 504 is used to communicate with other electronic devices or servers.

[0094] The processor 502 is configured to execute the program 510 , and specifically may execute the relevant steps in the above method embodiment.

[0095] Specifically, the program 510 may include program codes, which include computer operation instructions.

[0096] Processor 502 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The one or more processors included in a smart device may be of the same type, such as one or more CPUs, or different types, such as one or more CPUs and one or more ASICs.

[0097] The memory 506 is used to store the program 510. The memory 506 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk storage.

[0098] The program 510 may be specifically configured to enable the processor 502 to execute operations corresponding to the methods described in the above method embodiments.

[0099] The specific implementation of each step in program 510 can be found in the corresponding descriptions of the corresponding steps and units in the above-mentioned method embodiments, and will not be repeated here. Those skilled in the art will clearly understand that for the convenience and brevity of description, the specific working processes of the above-mentioned devices and modules can refer to the corresponding process descriptions in the above-mentioned method embodiments, and will not be repeated here.

[0100] It should be pointed out that, according to the needs of implementation, the various components / steps described in the embodiments of the present invention can be split into more components / steps, or two or more components / steps or partial operations of components / steps can be combined into new components / steps to achieve the purpose of the embodiments of the present invention.

[0101] The methods according to the embodiments of the present invention described above can be implemented in hardware, firmware, or as software or computer code that can be stored on a recording medium (such as a CD ROM, RAM, floppy disk, hard disk, or magneto-optical disk), or as computer code originally stored on a remote recording medium or non-transitory machine-readable medium downloaded over a network and then stored on a local recording medium. Thus, the methods described herein can be processed by such software stored on a recording medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware (such as an ASIC or FPGA). It will be understood that a computer, processor, microprocessor controller, or programmable hardware includes a storage component (e.g., RAM, ROM, flash memory, etc.) that can store or receive software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the methods described herein are implemented. Furthermore, when a general-purpose computer accesses the code for implementing the methods described herein, the execution of the code transforms the general-purpose computer into a dedicated computer for performing the methods described herein.

[0102] Those skilled in the art will appreciate that the units and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the embodiments of the present invention.

[0103] The above implementation methods are only used to illustrate the embodiments of the present invention, and are not intended to limit the embodiments of the present invention. Ordinary technicians in the relevant technical field may make various changes and modifications without departing from the spirit and scope of the embodiments of the present invention. Therefore, all equivalent technical solutions also fall within the scope of the embodiments of the present invention, and the scope of patent protection of the embodiments of the present invention should be defined by the claims.

Claims

1. A time-domain airborne electromagnetic inversion method based on deep learning, characterized in that: include: Based on the recursive algorithm, the received formation thickness, random flight altitude and the measured formation resistivity, forward modeling is performed to obtain the magnetic field response at each frequency point; The magnetic field response is transformed by integrating the Hankel transform kernel function to obtain a frequency domain magnetic induction intensity component; Generating a time-domain step response for the frequency-domain magnetic induction intensity component using a preset-order Hankel transform coefficient, and performing convolution processing on the time-domain step response based on a Gaussian integration node to obtain electromagnetic response data in the vertical direction; Using the electromagnetic response data, the formation thickness, and the random flight height as a training set, and preprocessing the training set to obtain a preprocessed training set; Obtaining a predicted flight altitude and a predicted formation resistivity based on the preprocessed training set, the first long short-term memory layer, the second long short-term memory layer, the third long short-term memory layer, the fourth long short-term memory layer, the fifth long short-term memory layer, the dropout layer, and the fully connected layer; Training a long short-term memory network to be trained based on the predicted flight altitude and the predicted formation resistivity to obtain a long short-term memory network; the long short-term memory network is composed of five long short-term memory layers, a dropout layer, and a fully connected layer, wherein the number of hidden units in the first long short-term memory layer is 32, the number of hidden units in the second long short-term memory layer is 64, the number of hidden units in the third long short-term memory layer is 128, the number of hidden units in the fourth long short-term memory layer is 256, and the number of hidden units in the fifth long short-term memory layer is 512; The measured electromagnetic response data is input into the long short-term memory network for inversion to obtain the predicted target flight altitude and target formation resistivity.

2. The method according to claim 1, characterized in that The training set also includes label data, wherein the label data includes a label flight height and a label formation resistivity; The preprocessing of the training set to obtain a preprocessed training set includes: Normalizing the electromagnetic response data, the formation thickness, and the random flight height in the training set to obtain first electromagnetic response data, first formation thickness, and first random flight height; The tag flying height is normalized using a normalized scaler to obtain a first height, and the tag formation resistivity is normalized using a minimum-maximum scaler to obtain a first formation resistivity.

3. The method according to claim 2, characterized in that The step of obtaining a predicted flight altitude and a predicted formation resistivity based on the preprocessed training set, the first long short-term memory layer, the second long short-term memory layer, the third long short-term memory layer, the fourth long short-term memory layer, the fifth long short-term memory layer, the discard layer, and the fully connected layer includes: Inputting the first electromagnetic response data, the first formation thickness, and the first random flight height into the first long short-term memory layer to obtain a first time series feature, and inputting the first time series feature into a batch normalization layer to obtain a second time series feature; Inputting the second time series feature into the second long short-term memory layer to obtain a third time series feature, and inputting the third time series feature into the batch normalization layer to obtain a fourth time series feature; Task-independent branch prediction is performed based on the fourth timing feature, the third long short-term memory layer, the fourth long short-term memory layer, and the fifth long short-term memory layer to obtain the predicted flight altitude and the predicted formation resistivity.

4. The method according to claim 3, characterized in that The performing task-independent branch prediction based on the fourth time series feature, the third long short-term memory layer, the fourth long short-term memory layer, and the fifth long short-term memory layer to obtain the predicted flight altitude and the predicted formation resistivity includes: Obtaining the predicted flight altitude based on the fourth temporal feature, the third long short-term memory layer, the batch normalization layer, and the dropout layer; The predicted formation resistivity is obtained based on the fourth time series feature, the fourth long short-term memory layer, the fifth long short-term memory layer, the batch normalization layer, and the discard layer.

5. The method according to claim 2, characterized in that The step of training the long short-term memory network to be trained based on the predicted flight altitude and the predicted formation resistivity to obtain the long short-term memory network comprises: Obtain a first loss by using the predicted flight altitude and the first altitude, and obtain a second loss by using the predicted formation resistivity and the first formation resistivity; A total loss is obtained by using the first loss and the second loss, and the long short-term memory network to be trained is trained by using the total loss to obtain the long short-term memory network.

Images