Transient electromagnetic method data denoising method, related device and computer storage medium

Abstract

The invention provides a transient electromagnetic method data denoising method, a related device and a computer storage medium, and the method comprises the steps: carrying out the preprocessing of transient electromagnetic method forward modeling data of a measurement point for the transient electromagnetic method forward modeling data of each measurement point after obtaining a transient electromagnetic method data set, and obtaining the target transient electromagnetic method forward modeling data of the measurement point; and then, inputting the target transient electromagnetic method forward modeling data of the measuring point into the denoising model, and outputting the denoised transient electromagnetic method forward modeling data of the measuring point. The denoising model uses early-stage data (target transient electromagnetic method forward modeling data) with a high signal-to-noise ratio as input, and predicts and outputs complete denoised transient electromagnetic method forward modeling data, so that the transient electromagnetic method data can be denoised quickly and effectively.

Description

Technical Field

[0001] The present application relates to the field of electromagnetic detection technology, and in particular to a method for denoising transient electromagnetic data, a related device and a computer storage medium. Background Art

[0002] At present, transient electromagnetic method is widely used in mineral exploration, hydrogeological survey and other fields. However, in actual data collection, transient electromagnetic method data are often interfered by various electromagnetic noises, which seriously affect the subsequent data processing and interpretation. The noise sources are mainly divided into natural noise and cultural noise, which is more obvious in the later data. Natural noise includes geomagnetic field, plasma emitted by the sun, and lightning discharge. Cultural noise mainly comes from transmission line network, low-frequency transmitter and radio transmitter. In addition, the interaction between underground metal pipelines, cables and conductors and cultural noise can also cause coupled electromagnetic noise, which further increases the complexity of the data. Although many methods have been proposed to remove noise from transient electromagnetic method data, including denoising methods based on statistics, transform domain, filtering and prediction, traditional methods still have some limitations. Summary of the invention

[0003] In view of this, the present application provides a transient electromagnetic method data denoising method, related devices and computer storage medium, which can quickly and effectively denoise transient electromagnetic method data.

[0004] The first aspect of the present application provides a method for denoising transient electromagnetic method data, comprising:

[0005] Acquire a transient electromagnetic method data set; wherein the transient electromagnetic method data set includes transient electromagnetic method forward modeling data of at least one measuring point;

[0006] For each of the transient electromagnetic method forward modeling data of the measuring point, preprocessing the transient electromagnetic method forward modeling data of the measuring point is performed to obtain target transient electromagnetic method forward modeling data of the measuring point;

[0007] The target transient electromagnetic method forward modeling data of the measuring point is input into a denoising model, and the denoised transient electromagnetic method forward modeling data of the measuring point is output; wherein the denoising model is obtained by training a deep learning model with a training sample set; and the training sample set includes training sample transient electromagnetic method forward modeling data of at least one measuring point.

[0008] Optionally, the transient electromagnetic method forward modeling data of each measuring point is preprocessed to obtain target transient electromagnetic method forward modeling data of the measuring point, including:

[0009] For each of the transient electromagnetic method forward modeling data of the measuring point, normalizing the transient electromagnetic method forward modeling data of the measuring point to obtain normalized transient electromagnetic method forward modeling data of the measuring point;

[0010] The target length of the normalized transient electromagnetic method forward modeling data of the measuring point is intercepted to obtain the target transient electromagnetic method forward modeling data.

[0011] Optionally, the method for constructing the denoising model includes:

[0012] Constructing a training sample set; wherein the training sample set includes training sample transient electromagnetic method forward modeling data of at least one measuring point;

[0013] For each of the training sample transient electromagnetic method forward modeling data of the measuring point, normalizing the training sample transient electromagnetic method forward modeling data of the measuring point to obtain target training sample transient electromagnetic method forward modeling data of the measuring point;

[0014] For each target training sample transient electromagnetic method forward modeling data of a measuring point, masking is performed on the target training sample transient electromagnetic method forward modeling data of the measuring point to obtain masked target training sample transient electromagnetic method forward modeling data of the measuring point;

[0015] Input the masked target training sample transient electromagnetic method forward modeling data of the measuring point into the deep learning model, and output the predicted data of the training sample transient electromagnetic method of the measuring point;

[0016] The parameters in the deep learning model are adjusted by using the error between the predicted data of the training sample transient electromagnetic method of the measuring point and the forward modeling data of the training sample transient electromagnetic method of the measuring point, until the error between the predicted data of the training sample transient electromagnetic method of the measuring point and the forward modeling data of the training sample transient electromagnetic method of the measuring point meets the preset convergence condition, and the deep learning model is used as the denoising model.

[0017] Optionally, the deep learning model is based on an encoder, a decoder and a multi-head attention mechanism, and the encoder and the decoder are both stacked by multi-layer gated recurrent unit networks.

[0018] Optionally, the masked target training sample transient electromagnetic method forward modeling data of the measuring point is input into a deep learning model, and outputting the training sample transient electromagnetic method prediction data of the measuring point, including:

[0019] Inputting the masked target training sample transient electromagnetic method forward modeling data of the measuring point into the deep learning model, wherein the encoder of the deep learning model processes the masked target training sample transient electromagnetic method forward modeling data of the measuring point to obtain a hidden state vector;

[0020] In the multi-head attention mechanism of the deep learning model, determining a context vector according to the hidden state vector;

[0021] In a decoder of the deep learning model, updating a current hidden state based on the context vector;

[0022] According to the updated hidden state, the weight matrix of the linear transformation and the bias term, the prediction data of the transient electromagnetic method of the training sample of the measuring point is determined.

[0023] Optionally, in the multi-head attention mechanism of the deep learning model, determining the context vector according to the hidden state vector includes:

[0024] In the multi-head attention mechanism of the deep learning model, the hidden state vector is linearly transformed to obtain the query vector, key vector and value vector of all attention heads;

[0025] For each attention head, the query vector, key vector, and value vector of the attention head are calculated using the scaled dot product attention mechanism to obtain the output of the attention head;

[0026] A context vector is determined based on the output of the attention head and the weight matrix.

[0027] Optionally, the deep learning model uses mean square error as a loss function during model training.

[0028] The second aspect of the present application provides a denoising device for transient electromagnetic method data, comprising:

[0029] An acquisition unit, used for acquiring a transient electromagnetic method data set; wherein the transient electromagnetic method data set includes transient electromagnetic method forward modeling data of at least one measuring point;

[0030] A preprocessing unit, for preprocessing the transient electromagnetic method forward modeling data of each measuring point, to obtain target transient electromagnetic method forward modeling data of the measuring point;

[0031] A denoising unit is used to input the target transient electromagnetic method forward modeling data of the measuring point into a denoising model, and output the denoised transient electromagnetic method forward modeling data of the measuring point; wherein the denoising model is obtained by training a deep learning model with a training sample set; and the training sample set includes training sample transient electromagnetic method forward modeling data of at least one measuring point.

[0032] Optionally, the preprocessing unit includes:

[0033] A first normalization unit is used to perform normalization processing on the transient electromagnetic method forward modeling data of each measuring point to obtain normalized transient electromagnetic method forward modeling data of the measuring point;

[0034] The interception unit is used to intercept the target length of the normalized transient electromagnetic method forward modeling data of the measuring point to obtain the target transient electromagnetic method forward modeling data.

[0035] Optionally, the construction unit of the denoising model includes:

[0036] A training sample set construction unit is used to construct a training sample set; wherein the training sample set includes training sample transient electromagnetic method forward modeling data of at least one measuring point;

[0037] A second normalization unit is used to perform normalization processing on the training sample transient electromagnetic method forward modeling data of each measuring point to obtain target training sample transient electromagnetic method forward modeling data of the measuring point;

[0038] A masking unit is used for performing masking processing on the target training sample transient electromagnetic method forward modeling data of each measuring point to obtain the masked target training sample transient electromagnetic method forward modeling data of the measuring point;

[0039] A training unit, used for inputting the masked target training sample transient electromagnetic method forward modeling data of the measuring point into a deep learning model, and outputting the predicted data of the training sample transient electromagnetic method of the measuring point;

[0040] An adjustment unit is used to adjust the parameters in the deep learning model by using the error between the predicted data of the training sample transient electromagnetic method of the measuring point and the forward modeling data of the training sample transient electromagnetic method of the measuring point, until the error between the predicted data of the training sample transient electromagnetic method of the measuring point and the forward modeling data of the training sample transient electromagnetic method of the measuring point meets a preset convergence condition, and the deep learning model is used as a denoising model.

[0041] Optionally, the deep learning model is based on an encoder, a decoder and a multi-head attention mechanism, and the encoder and the decoder are both stacked by multi-layer gated recurrent unit networks.

[0042] Optionally, the training unit includes:

[0043] A processing unit, configured to input the masked target training sample transient electromagnetic method forward modeling data of the measuring point into a deep learning model, wherein an encoder of the deep learning model processes the masked target training sample transient electromagnetic method forward modeling data of the measuring point to obtain a hidden state vector;

[0044] A first determining unit is used to determine a context vector according to the hidden state vector in a multi-head attention mechanism of the deep learning model;

[0045] An updating unit, configured to update a current hidden state in a decoder of the deep learning model based on the context vector;

[0046] The second determination unit is used to determine the prediction data of the training sample transient electromagnetic method of the measuring point according to the updated hidden state, the weight matrix of the linear transformation and the bias term.

[0047] Optionally, the first determining unit includes:

[0048] A linear transformation unit, used to perform a linear transformation on the hidden state vector in the multi-head attention mechanism of the deep learning model to obtain query vectors, key vectors and value vectors of all attention heads;

[0049] An attention unit, for calculating the query vector, the key vector, and the value vector of each attention head using a scaled dot product attention mechanism to obtain an output of the attention head;

[0050] The first determination subunit is used to determine a context vector according to the output of the attention head and the weight matrix.

[0051] Optionally, the deep learning model uses mean square error as a loss function during model training.

[0052] A third aspect of the present application provides an electronic device, including:

[0053] one or more processors;

[0054] a storage device having one or more programs stored thereon;

[0055] When the one or more programs are executed by the one or more processors, the one or more processors implement the transient electromagnetic method data denoising method as described in any one of the first aspects.

[0056] A fourth aspect of the present application provides a computer storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the method for denoising transient electromagnetic method data as described in any one of the first aspects is implemented.

[0057] It can be seen from the above scheme that the present application provides a denoising method, related device and computer storage medium for transient electromagnetic method data. After acquiring the transient electromagnetic method data set, the transient electromagnetic method forward modeling data of each measuring point is preprocessed to obtain the target transient electromagnetic method forward modeling data of the measuring point; finally, the target transient electromagnetic method forward modeling data of the measuring point is input into the denoising model, and the denoised transient electromagnetic method forward modeling data of the measuring point is output. The denoising model uses early high signal-to-noise ratio data (target transient electromagnetic method forward modeling data) as input, predicts and outputs complete denoised transient electromagnetic method forward modeling data, thereby realizing fast and effective denoising of transient electromagnetic method data. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0059] Figure 1 A specific flow chart of a method for denoising transient electromagnetic method data provided in an embodiment of the present application;

[0060] Figure 2 A flowchart of a method for preprocessing transient electromagnetic method forward modeling data of a measuring point provided in another embodiment of the present application;

[0061] Figure 3 A flowchart of a method for constructing a denoising model provided in another embodiment of the present application;

[0062] Figure 4 A schematic diagram of a masking method provided in another embodiment of the present application;

[0063] Figure 5 A flowchart of data processing by a deep learning model provided in another embodiment of the present application;

[0064] Figure 6 A flowchart of a method for determining a context vector provided by another embodiment of the present application;

[0065] Figure 7 A schematic diagram of a deep learning network structure based on self-supervised learning provided in another embodiment of the present application;

[0066] Figure 8 A schematic diagram of another embodiment of the present application using a logarithmic scale to show the change of the loss function of a self-supervised deep learning method;

[0067] Fig. 9 A schematic diagram of unsupervised denoising comparison provided by another embodiment of the present application;

[0068] Fig.10 A schematic diagram of a device for implementing denoising of transient electromagnetic method data provided in another embodiment of the present application.

[0069] Fig.11 A schematic diagram of an electronic device for implementing a method for denoising transient electromagnetic method data provided by another embodiment of the present application. DETAILED DESCRIPTION

[0070] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0071] The term "including" and its variations used herein are open inclusions, i.e., "including but not limited to". The term "based on" means "based at least in part on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". The relevant definitions of other terms will be given in the following description.

[0072] It should be noted that the information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0073] It should be noted that the concepts such as "first" and "second" mentioned in this application are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0074] It should be noted that the modifications of "one" and "plurality" mentioned in the present application are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0075] The present application embodiment provides a method for denoising transient electromagnetic method data, such as Figure 1 As shown, the specific steps include:

[0076] S101. Acquire a transient electromagnetic method data set.

[0077] The transient electromagnetic method data set includes transient electromagnetic method forward modeling data of at least one measuring point.

[0078] Specifically, according to the information of the transient electromagnetic method transmitting source, receiving device and measuring point, the frequency domain magnetic field intensity H(ω) is obtained by forward calculation, and then the sine and cosine transforms are used to obtain the magnetic field impulse response and step response respectively. The expressions are as follows:

[0079]

[0080] in, represents the magnetic field pulse response, h(t) represents the magnetic field step response, Im[H(ω)] is the imaginary part of the electromagnetic field in the frequency domain, d represents the differential operation, ω represents the angular frequency, and t represents time.

[0081] In the specific implementation process of the present application, technicians can select a magnetic field pulse response or a step response according to actual application conditions to generate a transient electromagnetic method data set.

[0082] S102 , preprocessing the transient electromagnetic method forward modeling data of each measuring point to obtain target transient electromagnetic method forward modeling data of the measuring point.

[0083] Optionally, in another embodiment of the present application, an implementation of step S102 is as follows: Figure 2 As shown, including:

[0084] S201 , for each measuring point's transient electromagnetic method forward modeling data, normalize the transient electromagnetic method forward modeling data to obtain normalized transient electromagnetic method forward modeling data of the measuring point.

[0085] It can be understood that through data normalization, the data is scaled to between 0 and 1 to improve the prediction accuracy of the model.

[0086] S202, intercepting the target length of the normalized transient electromagnetic method forward modeling data of the measuring point to obtain target transient electromagnetic method forward modeling data.

[0087] The target length may be pre-set and modified by technicians, experts, etc., such as 15 time steps, 20 time steps, 47 time steps, etc., and is not limited here.

[0088] S103, inputting the target transient electromagnetic method forward modeling data of the measuring point into the denoising model, and outputting the denoised transient electromagnetic method forward modeling data of the measuring point.

[0089] The denoising model is obtained by training a deep learning model with a training sample set; the training sample set includes training sample transient electromagnetic method forward modeling data of at least one measuring point.

[0090] Optionally, in another embodiment of the present application, an implementation method of a method for constructing a denoising model is as follows: Figure 3 As shown, including:

[0091] S301: Construct a training sample set.

[0092] The training sample set includes training sample transient electromagnetic method forward modeling data of at least one measuring point.

[0093] It should be noted that the specific implementation process of obtaining the training sample transient electromagnetic method forward modeling data of the measuring point can refer to the method for obtaining the transient electromagnetic method forward modeling data of the measuring point in step S101, which will not be repeated here.

[0094] S302 , for each measuring point's training sample transient electromagnetic method forward modeling data, normalize the training sample transient electromagnetic method forward modeling data to obtain target training sample transient electromagnetic method forward modeling data of the measuring point.

[0095] S303 , performing mask processing on the target training sample transient electromagnetic method forward modeling data of each measuring point to obtain the masked target training sample transient electromagnetic method forward modeling data of the measuring point.

[0096] For example, there are M target training sample transient electromagnetic method forward modeling data of the mth target training sample transient electromagnetic method forward modeling data. Where N is the number of time gates.

[0097] Perform L masking operations on the target training sample transient electromagnetic method forward modeling data of each measuring point, and the length of each mask is different. The specific operation is to retain the data of the previous time gate and set the data of the remaining time gates to zero to form the input data (the target training sample transient electromagnetic method forward modeling data of the measuring point after masking) Without mask processing, the complete forward modeling data x m The target output of the network (training sample transient electromagnetic method forward modeling data of the measurement point) is denoted as y m ,like Figure 4 In this way, the masked input data and its corresponding complete data form an input-output pair for subsequent training of the deep learning model.

[0098] S304, inputting the masked target training sample transient electromagnetic method forward modeling data of the measuring point into the deep learning model, and outputting the predicted data of the training sample transient electromagnetic method of the measuring point.

[0099] In the specific implementation process of the present application, the deep learning model shown can be but is not limited to being based on an encoder, a decoder and a multi-head attention mechanism, and the encoder and the decoder are both stacked by a multi-layer gated recurrent unit network.

[0100] The Seq2Seq-GRU model can be used, but is not limited to, based on the encoder and decoder structure, which can effectively extract the characteristics of the data itself. The model uses a multi-layer gated recurrent unit (GRU) network layer to achieve more efficient calculation. At the same time, a multi-head attention mechanism is introduced to enable the model to focus on different parts of the data.

[0101] Specifically, the input sequence Simplified to x I =(x i |1≤i≤I), where I represents the length of valid data, and the target output sequence ym is simplified to y=(y n |1≤n≤N).

[0102] Optionally, in another embodiment of the present application, an implementation of step S304 is as follows: Figure 5 As shown, including:

[0103] S501. Input the masked target training sample transient electromagnetic method forward modeling data of the measuring point into the deep learning model. The encoder of the deep learning model processes the masked target training sample transient electromagnetic method forward modeling data of the measuring point to obtain a hidden state vector.

[0104] Specifically, when the encoder processes the input sequence (the masked target training sample transient electromagnetic method forward data of the measurement point), it processes the input data x at the current time step i i and the hidden state vector s at the previous time step i-1 To obtain the hidden state vector s of the current time step i , the process can be expressed as:

[0105] s i =f encoder (s i-1 ,x i ), where f encoder Represents a multi-layer gated recurrent unit (GRU) structure. The encoder processes the entire input sequence step by step and finally generates a hidden state vector s = {s1, s2, ..., s I}.

[0106] S502. In the multi-head attention mechanism of the deep learning model, determine the context vector according to the hidden state vector.

[0107] Optionally, in another embodiment of the present application, an implementation of step S502 is as follows: Figure 6 As shown, including:

[0108] S601. In the multi-head attention mechanism of the deep learning model, a linear transformation is performed on the hidden state vector to obtain the query vector, key vector and value vector of all attention heads.

[0109] Specifically, in the multi-head attention mechanism, the hidden state s is first linearly transformed, that is, the data is mapped into different vector representations through matrix multiplication, so as to obtain the query vector Q, key vector K and value vector V of all attention heads for subsequent attention calculations. Q , K = sw V , V = sw K

[0110] S602: For each attention head, the query vector, key vector and value vector of the attention head are calculated using the scaled dot product attention mechanism to obtain the output of the attention head.

[0111] Specifically, for each attention head, the scaled dot product attention mechanism is used to calculate its output. The scaled dot product attention mechanism is one of the core technologies for calculating attention weights. Its basic idea is to use the dot product of the query vector and the key vector to measure their similarity, thereby determining the attention weights between different inputs. The calculation formula is:

[0112] The softmax function converts each element of the input vector into a probability value between 0 and 1. K Indicates the dimension of the key vector, used to scale the dot product.

[0113] S603: Determine a context vector according to the output of the attention head and the weight matrix.

[0114] Specifically, the outputs of all attention heads are concatenated ("concat") together and multiplied by the weight matrix W of the output layer O , get the context vector c:

[0115] c=concat(head1,head2,…head H )W O ; where h ranges from 1 to H (where H is the total number of attention heads). In the above neural network layer, the weight matrices of all linear transformation operations are trainable parameters of the model and are optimized through network training.

[0116] S503. In the decoder of the deep learning model, update the current hidden state based on the context vector.

[0117] Specifically, the decoder is based on the context vector c generated by the encoder and the hidden state h at the previous moment n-1 And the predicted output of the previous moment, update the current hidden state h n :

[0118] Among them, f decoder It is also a multi-layer gated recurrent unit (GRU) structure.

[0119] S504: Determine prediction data of the training sample transient electromagnetic method of the measuring point according to the updated hidden state, the weight matrix of the linear transformation and the bias term.

[0120] It is understandable that the output of the decoder is usually a high-dimensional vector. In order to map it to the dimension of the target sequence, a linear transformation layer is usually added after the GRU layer. It is expressed as:

[0121] Among them, W linear and b linear is the weight matrix and bias term of the linear transformation. The network model uses a self-supervised learning strategy to restore the complete sequence from the masked partial sequence.

[0122] S305 , judging whether the error between the predicted data of the training sample transient electromagnetic method of the measuring point and the forward modeling data of the training sample transient electromagnetic method of the measuring point meets a preset convergence condition.

[0123] The preset convergence condition is preset and changed by a technician or authorized staff and is not limited here.

[0124] It should be noted that, in the specific implementation process of the present application, it is not limited to using the preset convergence conditions to train the model, and the model can also be trained using a preset number of iterations, which is also not limited here.

[0125] Specifically, if it is determined that the error between the predicted data of the training sample transient electromagnetic method of the measuring point and the forward modeling data of the training sample transient electromagnetic method of the measuring point meets the preset convergence condition, step S306 is executed; if it is determined that the error between the predicted data of the training sample transient electromagnetic method of the measuring point and the forward modeling data of the training sample transient electromagnetic method of the measuring point does not meet the preset convergence condition, step S307 is executed.

[0126] S306: Use the deep learning model as a denoising model.

[0127] S307, adjusting the parameters in the deep learning model by using the error between the predicted data of the training sample transient electromagnetic method of the measuring point and the forward modeling data of the training sample transient electromagnetic method of the measuring point.

[0128] Optionally, in another embodiment of the present application, mean square error is used as a loss function during model training of the deep learning model.

[0129] It is understandable that the mean square error (MSE) as a loss function is very suitable for the sensitivity of prediction errors in regression problems and the ability to intuitively reflect the difference between predicted values ​​and actual values. The mathematical expression of MSE is:

[0130] Where N is the number of samples and y is n The true value of the nth sample, is the predicted value of the model for the nth sample. In the specific implementation process of this application, the Adam optimizer can be used to train the network but is not limited to it.

[0131] Now, this application takes the ground loop source transient electromagnetic method as an example to illustrate the implementation process and effect of the method proposed in this application.

[0132] Step 1: Construction of training sample dataset

[0133] The example of the embodiment of the present application generates transient electromagnetic method data based on a one-dimensional forward model to construct a training and test data set for a deep learning model. The geoelectric model used contains 30 layers of layered media, of which the total thickness of the first 29 layers is 1500 meters, the thickness of the shallowest first layer is 20 meters, the thickness of the deepest 29th layer is 106.7 meters, and the 30th layer is half space. In order to reflect the characteristics of different underground structures, the resistivity is set in the range of 0.1 to 10000Ωm, and 20 logarithmically uniformly distributed resistivity values ​​are selected within this range. Using the above parameters, 500,000 sets of geoelectric models were randomly generated, and the formation conductivity and corresponding induced electromotive force data of each set of models were calculated, recorded as a one-dimensional data set. Subsequently, 1000 three-dimensional geoelectric models were constructed, and three-dimensional forward calculations were performed, and finally about 5000 sets of data were retained as three-dimensional data sets. The one-dimensional and three-dimensional data sets were merged for network training.

[0134] Step 2: Generation of mask data

[0135] Based on step 1, a masked data generation strategy is introduced. Specifically, for the data sequence of each measurement point, different degrees of masking (replacing with zero values) are performed for the later time periods. In each masking operation, the first several time gate data of the sequence are retained, and the remaining time gate data are masked to generate multiple training samples with different mask lengths. Finally, the masked data sequence is used as the input of the model, and the complete induced electromotive force sequence is used as the corresponding output label, thereby constructing a self-supervised learning task without relying on paired labeled data.

[0136] In the data preprocessing stage, the simulated induced electromotive force data is logarithmically transformed to reduce the fluctuations in different numerical ranges, and further standardized to make it suitable for the input requirements of the neural network. The processed data set is divided into training set, validation set and test set in a ratio of 8:1:1 to ensure the training effect and generalization ability of the model.

[0137] Step 3: Deep learning model construction

[0138] This application adopts a deep learning network structure based on self-supervised learning (see Figure 7 ), mainly composed of an encoder, a decoder, and a multi-head attention mechanism. The overall architecture adopts the classic Seq2Seq model design. Both the encoder and the decoder are stacked with multiple layers of GRU (Gated Recurrent Unit) networks to effectively extract and represent the deep features of time series. The multi-head attention mechanism is introduced between the encoder and the decoder, which can capture key information in the data and thus improve the processing accuracy of the model. Through the self-supervised learning strategy, the model learns the ability to reconstruct the complete signal from partially masked data, effectively realizing the prediction and denoising of transient electromagnetic data.

[0139] Step 4: Train and tune the model

[0140] First, the data is logarithmized and preprocessed using mean-variance normalization. Next, the data is masked, and the effective length of the masked data is set to 15-20, which is divided according to the area where the signal-to-noise ratio in the data decreases. The masked data is used as the input of the model, and the original data is used as the target output. This time, the Seq2Seq model is used, which uses a three-layer GRU encoder and decoder, and combines a multi-head attention mechanism to enhance the decoding effect.

[0141] During training, the learning rate is set to 0.001, the optimizer is Adam, and the training is performed for 500 epochs. During the training process, the loss function of the self-supervised deep learning method is displayed using a logarithmic scale (see Figure 8), where the vertical axis is the logarithmic scale of the loss value. The mean square error loss decreases rapidly in the early stage of training and gradually converges after about 400 epochs, indicating that the model converges well and has high stability during training.

[0142] Step 5: Use the trained model to denoise the noisy induced electromotive force data

[0143] After the model training is completed, the Seq2Seq-GRU network is used to perform denoising tests on noisy transient electromagnetic method data. The test data is generated by three-dimensional forward modeling. The model is a plate-like model. Two burial depths of 50m and 1000m are selected for testing. The input data length of the test phase is set to 20, that is, the data of the first 20 moments of the signal are used as input. Fig. 9 As shown in the figure, in these two test cases, the model's predicted data is basically consistent with the noise-free forward data, showing a high degree of consistency. This shows that the Seq2Seq-GRU network can effectively predict the complete time series from early high signal-to-noise ratio data, verifying the effectiveness of the model in denoising tasks.

[0144] By introducing a self-supervised learning strategy, the reliance on labeled data sets is significantly reduced, and the tedious noise simulation process is avoided. At the same time, the model can fully explore the characteristic laws of the data itself and has the flexibility to process input data of different lengths.

[0145] It can be seen from the above scheme that the present application provides a method for denoising transient electromagnetic method data. After acquiring the transient electromagnetic method data set, the transient electromagnetic method forward modeling data of each measuring point is preprocessed to obtain the target transient electromagnetic method forward modeling data of the measuring point; finally, the target transient electromagnetic method forward modeling data of the measuring point is input into the denoising model, and the denoised transient electromagnetic method forward modeling data of the measuring point is output. The denoising model uses early high signal-to-noise ratio data (target transient electromagnetic method forward modeling data) as input, predicts and outputs complete denoised transient electromagnetic method forward modeling data, thereby realizing fast and effective denoising of transient electromagnetic method data.

[0146] The present application embodiment provides a denoising device for transient electromagnetic method data, such as Fig.10 As shown, specifically including:

[0147] The acquisition unit 1001 is used to acquire a transient electromagnetic method data set.

[0148] The transient electromagnetic method data set includes transient electromagnetic method forward modeling data of at least one measuring point.

[0149] The preprocessing unit 1002 is used to preprocess the transient electromagnetic method forward modeling data of each measuring point to obtain target transient electromagnetic method forward modeling data of the measuring point.

[0150] Optionally, in another embodiment of the present application, an implementation of the preprocessing unit 1002 includes:

[0151] The first normalization unit is used for normalizing the transient electromagnetic method forward modeling data of each measuring point to obtain normalized transient electromagnetic method forward modeling data of the measuring point.

[0152] The interception unit is used to intercept the target length of the normalized transient electromagnetic method forward modeling data of the measuring point to obtain the target transient electromagnetic method forward modeling data.

[0153] For the specific working process of the units disclosed in the above embodiments of the present application, please refer to the corresponding method embodiments, such as Figure 2 As shown, no further description is given here.

[0154] The denoising unit 1003 is used to input the target transient electromagnetic method forward modeling data of the measuring point into the denoising model, and output the denoised transient electromagnetic method forward modeling data of the measuring point; wherein the denoising model is obtained by training the deep learning model with a training sample set; the training sample set includes training sample transient electromagnetic method forward modeling data of at least one measuring point.

[0155] For the specific working process of the units disclosed in the above embodiments of the present application, please refer to the corresponding method embodiments, such as Figure 1 As shown, no further description is given here.

[0156] Optionally, in another embodiment of the present application, an implementation of a construction unit of a denoising model includes:

[0157] The training sample set construction unit is used to construct the training sample set.

[0158] The training sample set includes training sample transient electromagnetic method forward modeling data of at least one measuring point.

[0159] The second normalization unit is used to perform normalization processing on the training sample transient electromagnetic method forward modeling data of each measuring point to obtain the target training sample transient electromagnetic method forward modeling data of the measuring point.

[0160] The mask unit is used for performing mask processing on the target training sample transient electromagnetic method forward modeling data of each measuring point to obtain the masked target training sample transient electromagnetic method forward modeling data of the measuring point.

[0161] The training unit is used to input the masked target training sample transient electromagnetic method forward modeling data of the measuring point into the deep learning model, and output the predicted data of the training sample transient electromagnetic method of the measuring point.

[0162] An adjustment unit is used to adjust the parameters in the deep learning model by using the error between the predicted data of the training sample transient electromagnetic method of the measuring point and the forward modeling data of the training sample transient electromagnetic method of the measuring point, until the error between the predicted data of the training sample transient electromagnetic method of the measuring point and the forward modeling data of the training sample transient electromagnetic method of the measuring point meets the preset convergence condition, and the deep learning model is used as a denoising model.

[0163] For the specific working process of the units disclosed in the above embodiments of the present application, please refer to the corresponding method embodiments, such as Figure 3 As shown, no further description is given here.

[0164] Optionally, in another embodiment of the present application, the deep learning model is based on an encoder, a decoder and a multi-head attention mechanism, and the encoder and the decoder are both stacked by a multi-layer gated recurrent unit network.

[0165] The specific working process of the units disclosed in the above embodiments of the present application can be found in the corresponding method embodiments, which will not be repeated here.

[0166] Optionally, in another embodiment of the present application, an implementation of the training unit includes:

[0167] The processing unit is used to input the masked target training sample transient electromagnetic method forward modeling data of the measuring point into the deep learning model. The encoder of the deep learning model processes the masked target training sample transient electromagnetic method forward modeling data of the measuring point to obtain a hidden state vector.

[0168] The first determination unit is used to determine a context vector according to a hidden state vector in a multi-head attention mechanism of a deep learning model.

[0169] An update unit is used to update the current hidden state based on the context vector in the decoder of the deep learning model.

[0170] The second determination unit is used to determine the prediction data of the training sample transient electromagnetic method of the measuring point according to the updated hidden state, the weight matrix of the linear transformation and the bias term.

[0171] For the specific working process of the units disclosed in the above embodiments of the present application, please refer to the corresponding method embodiments, such as Figure 5 As shown, no further description is given here.

[0172] Optionally, in another embodiment of the present application, an implementation of the first determining unit includes:

[0173] The linear transformation unit is used to perform linear transformation on the hidden state vector in the multi-head attention mechanism of the deep learning model to obtain the query vector, key vector and value vector of all attention heads.

[0174] The attention unit is used to calculate the query vector, key vector and value vector of each attention head using the scaled dot product attention mechanism to obtain the output of the attention head.

[0175] The first determination subunit is used to determine the context vector according to the output of the attention head and the weight matrix.

[0176] For the specific working process of the units disclosed in the above embodiments of the present application, please refer to the corresponding method embodiments, such as Figure 6 As shown, no further description is given here.

[0177] Optionally, in another embodiment of the present application, mean square error is used as a loss function during model training of the deep learning model.

[0178] The specific working process of the units disclosed in the above embodiments of the present application can be found in the corresponding method embodiments, which will not be repeated here.

[0179] It can be seen from the above scheme that the present application provides a denoising device for transient electromagnetic method data. After the acquisition unit 1001 acquires the transient electromagnetic method data set, the preprocessing unit 1002 preprocesses the transient electromagnetic method forward modeling data of each measuring point to obtain the target transient electromagnetic method forward modeling data of the measuring point; finally, the denoising unit 1003 inputs the target transient electromagnetic method forward modeling data of the measuring point into the denoising model, and outputs the denoised transient electromagnetic method forward modeling data of the measuring point. The denoising model uses early high signal-to-noise ratio data (target transient electromagnetic method forward modeling data) as input, predicts and outputs complete denoised transient electromagnetic method forward modeling data, thereby realizing fast and effective denoising of transient electromagnetic method data.

[0180] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0181] Another embodiment of the present application provides an electronic device, such as Fig.11 As shown, including:

[0182] One or more processors 1101.

[0183] The storage device 1102 stores one or more programs.

[0184] When the one or more programs are executed by the one or more processors 1101 , the one or more processors 1101 implement the method for denoising transient electromagnetic method data as described in any one of the above embodiments.

[0185] Another embodiment of the present application provides a computer storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the method for denoising transient electromagnetic method data as described in any one of the above embodiments is implemented.

[0186] In the context of the present application, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0187] It should be noted that the computer-readable medium mentioned above in the present application may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0188] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device.

[0189] Another embodiment of the present application provides a computer program product. When the computer program product is executed, it is used to perform the above-mentioned transient electromagnetic method data denoising method.

[0190] In particular, according to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via a communication device, or installed from a storage device, or installed from a ROM. When the computer program is executed by a processing device, the above-mentioned functions defined in the method of the embodiment of the present application are executed.

[0191] Although the subject matter has been described in language specific to structural features and / or method logic actions, it should be understood that the subject matter defined in this application is not necessarily limited to the specific features or actions described above. On the contrary, the specific features and actions described above are merely example forms of implementing this application.

[0192] Although several specific implementation details are included in the above discussion, these should not be interpreted as limiting the scope of the present application. Certain features described in the context of a separate embodiment can also be implemented in a single embodiment in combination. On the contrary, the various features described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable sub-combination.

[0193] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of application involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above application concept. For example, the above features are replaced with (but not limited to) technical features with similar functions applied in the present application.

Claims

1. A method for denoising transient electromagnetic method data, characterized in that: include: Acquire a transient electromagnetic method data set; wherein the transient electromagnetic method data set includes transient electromagnetic method forward modeling data of at least one measuring point; For each of the transient electromagnetic method forward modeling data of the measuring point, preprocessing the transient electromagnetic method forward modeling data of the measuring point is performed to obtain target transient electromagnetic method forward modeling data of the measuring point; The target transient electromagnetic method forward modeling data of the measuring point is input into a denoising model, and the denoised transient electromagnetic method forward modeling data of the measuring point is output; wherein the denoising model is obtained by training a deep learning model with a training sample set; and the training sample set includes training sample transient electromagnetic method forward modeling data of at least one measuring point.

2. The method for denoising transient electromagnetic method data according to claim 1, characterized in that: The method of preprocessing the transient electromagnetic method forward modeling data of each measuring point to obtain target transient electromagnetic method forward modeling data of the measuring point includes: For each of the transient electromagnetic method forward modeling data of the measuring point, normalizing the transient electromagnetic method forward modeling data of the measuring point to obtain normalized transient electromagnetic method forward modeling data of the measuring point; The target length of the normalized transient electromagnetic method forward modeling data of the measuring point is intercepted to obtain the target transient electromagnetic method forward modeling data.

3. The method for denoising transient electromagnetic method data according to claim 1, characterized in that: The method for constructing the denoising model comprises: Constructing a training sample set; wherein the training sample set includes training sample transient electromagnetic method forward modeling data of at least one measuring point; For each of the training sample transient electromagnetic method forward modeling data of the measuring point, normalizing the training sample transient electromagnetic method forward modeling data of the measuring point to obtain target training sample transient electromagnetic method forward modeling data of the measuring point; For each target training sample transient electromagnetic method forward modeling data of a measuring point, masking is performed on the target training sample transient electromagnetic method forward modeling data of the measuring point to obtain masked target training sample transient electromagnetic method forward modeling data of the measuring point; Input the masked target training sample transient electromagnetic method forward modeling data of the measuring point into the deep learning model, and output the predicted data of the training sample transient electromagnetic method of the measuring point; The parameters in the deep learning model are adjusted by using the error between the predicted data of the training sample transient electromagnetic method of the measuring point and the forward modeling data of the training sample transient electromagnetic method of the measuring point, until the error between the predicted data of the training sample transient electromagnetic method of the measuring point and the forward modeling data of the training sample transient electromagnetic method of the measuring point meets the preset convergence condition, and the deep learning model is used as the denoising model.

4. The method for denoising transient electromagnetic method data according to claim 3, characterized in that: The deep learning model is based on an encoder, a decoder and a multi-head attention mechanism, and both the encoder and the decoder are stacked by multi-layer gated recurrent unit networks.

5. The method for denoising transient electromagnetic method data according to claim 4, characterized in that: The masked target training sample transient electromagnetic method forward modeling data of the measuring point is input into the deep learning model, and the training sample transient electromagnetic method prediction data of the measuring point is output, including: Inputting the masked target training sample transient electromagnetic method forward modeling data of the measuring point into the deep learning model, wherein the encoder of the deep learning model processes the masked target training sample transient electromagnetic method forward modeling data of the measuring point to obtain a hidden state vector; In the multi-head attention mechanism of the deep learning model, determining a context vector according to the hidden state vector; In a decoder of the deep learning model, updating a current hidden state based on the context vector; According to the updated hidden state, the weight matrix of the linear transformation and the bias term, the prediction data of the transient electromagnetic method of the training sample of the measuring point is determined.

6. The method for denoising transient electromagnetic method data according to claim 5, characterized in that: In the multi-head attention mechanism of the deep learning model, determining the context vector according to the hidden state vector includes: In the multi-head attention mechanism of the deep learning model, the hidden state vector is linearly transformed to obtain the query vector, key vector and value vector of all attention heads; For each attention head, the query vector, key vector, and value vector of the attention head are calculated using the scaled dot product attention mechanism to obtain the output of the attention head; A context vector is determined based on the output of the attention head and the weight matrix.

7. The method for denoising transient electromagnetic method data according to claim 1, characterized in that: The deep learning model uses mean square error as the loss function during model training.

8. A denoising device for transient electromagnetic method data, characterized in that: include: An acquisition unit, used for acquiring a transient electromagnetic method data set; wherein the transient electromagnetic method data set includes transient electromagnetic method forward modeling data of at least one measuring point; A preprocessing unit, for preprocessing the transient electromagnetic method forward modeling data of each measuring point, to obtain target transient electromagnetic method forward modeling data of the measuring point; A denoising unit is used to input the target transient electromagnetic method forward modeling data of the measuring point into a denoising model, and output the denoised transient electromagnetic method forward modeling data of the measuring point; wherein the denoising model is obtained by training a deep learning model with a training sample set; and the training sample set includes training sample transient electromagnetic method forward modeling data of at least one measuring point.

9. An electronic device, characterized in that: include: one or more processors; a storage device having one or more programs stored thereon; When the one or more programs are executed by the one or more processors, the one or more processors are enabled to implement the transient electromagnetic method data denoising method according to any one of claims 1 to 7.

10. A computer storage medium, characterized in that: A computer program is stored thereon, wherein when the computer program is executed by a processor, the method for denoising transient electromagnetic method data as claimed in any one of claims 1 to 7 is implemented.

Images