Transient electromagnetic deep learning inversion method, system, device and medium

Transient electromagnetic data is processed through the Informer model and the multi-head probability sparse self-attention mechanism, the error accumulation problem of traditional RNNs in long-sequence data analysis is solved, and efficient and accurate inversion results are achieved.

CN119045071APending Publication Date: 2024-11-29XINJIANG UNIVERSITY
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202411265107.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-10
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

When traditional RNN neural networks process transient electromagnetic long sequence data, errors accumulate layer by layer, making it difficult to meet the analysis needs.

Method used

The Informer model is used to combine the encoder, decoder and multi-head probability sparse self-attention mechanism, and dynamically adjust the learning rate by forward generation of a layered resistivity model and an optimizer to construct a depth-resistive formation model for inversion.

Benefits of technology

It improves the inversion efficiency and accuracy, can process long-sequence time series data more effectively, maintains the consistency and integrity of the sequence, and significantly improves the calculation speed.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119045071A_ABST
    Figure CN119045071A_ABST
Patent Text Reader

Abstract

The invention provides a transient electromagnetic deep learning inversion method, system and device and a medium, and the method comprises the steps: constructing a layered resistivity model, and generating transient electromagnetic response data corresponding to the layered resistivity model through forward modeling; constructing an Informer model, taking the electromagnetic response data as input data to carry out Informer model training, and meanwhile, dynamically adjusting a learning rate by adopting an optimizer to obtain a trained depth-resistivity stratum model; and using the trained depth-resistivity stratum model to perform inversion on the input and acquired real transient electromagnetic response data to obtain an actual stratum resistivity distribution condition. Compared with a common inversion method at present, the method has the advantages that the inversion efficiency is high, and the result fits the actual geological condition.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of geophysical detection technology, and in particular relates to a transient electromagnetic deep learning inversion method, system, equipment and medium. Background Art

[0002] Transient electromagnetic method is a geophysical detection method based on electromagnetic induction. Due to its low construction cost, large detection depth, and strong detection ability for good metal conductors, it has been widely used in mineral resource exploration, tunnel advance prediction, groundwater hydrological survey, engineering geological survey and other fields. However, in actual application, only by analyzing the secondary field response, it can only qualitatively analyze the underground high (low) resistance anomaly, and cannot accurately describe the physical parameters and burial depth of the underground anomaly. At this time, it is necessary to further quantitatively interpret the distribution of underground physical parameters through transient electromagnetic inversion.

[0003] In the field of transient electromagnetic data interpretation, gradient linear inversion and fully nonlinear inversion are the mainstream inversion methods. Linearized inversion is highly efficient, but it is overly dependent on the initial model, prone to losing formation details, and difficult to accurately depict the formation distribution. And due to the complexity of the underground medium, the three-dimensional inversion method is limited by the huge amount of three-dimensional forward modeling, and the inversion efficiency is very low. The fully nonlinear inversion method is less dependent on the initial model and does not require the calculation of the Jacobian matrix. It has become a hot research direction in recent years. Domestic and foreign scholars have implemented a variety of optimization inversion methods such as genetic algorithms, particle swarm optimization algorithms, simulated annealing algorithms, Bayesian algorithms, etc., but due to the complex electromagnetic theory and data processing process, this type of method is difficult to be put into use on a large scale in actual engineering.

[0004] Since the rise of neural network technology in the 1980s, it has been widely used in image processing, speech recognition, stock prediction and other fields. In recent years, it has also been introduced to solve geophysical inversion problems such as magnetotelluric, seismic, and well logging. Transient electromagnetic neural network inversion is different from nonlinear Monte Carlo-type inversion methods with full-space search solutions. It establishes the matrix mapping relationship between model physical parameters and electromagnetic response through massive theoretical model training. It does not require a large amount of forward calculations and can greatly improve the inversion efficiency. Shallow neural networks such as BP neural network and supervised descent method and deep neural network inversion algorithms such as particle swarm neural network, genetic neural network, and convolutional neural network have been applied to transient electromagnetic inversion. Transient electromagnetic response data decays over time, so transient electromagnetic inversion is a typical time series problem. Based on this, Fan Tao et al. constructed a Seq2Seq inversion model based on long short-term memory network (LSTM), proving the rationality and feasibility of RNN-type inversion methods at the theoretical level. However, LSTM's ability to extract spatial features is slightly weaker than that of temporal features, and the serial processing method of LSTM greatly limits the training speed of the model. Summary of the invention

[0005] The purpose of the present invention is to provide a transient electromagnetic deep learning inversion method, system, device and medium to solve the problem that traditional RNN neural network cannot meet the needs of long sequence data analysis due to error accumulation layer by layer.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] In a first aspect, the present invention provides a transient electromagnetic deep learning inversion method, comprising:

[0008] Construct a layered resistivity model and generate transient electromagnetic response data corresponding to the layered resistivity model through forward modeling;

[0009] Construct an Informer model, use the electromagnetic response data as input data to train the Informer model, and use the optimizer to dynamically adjust the learning rate to obtain the trained depth-resistivity formation model;

[0010] The trained depth-resistivity formation model is used to invert the real transient electro-magnetic response data collected by the input to obtain the actual formation resistivity distribution.

[0011] Furthermore, a layered resistivity model is constructed, including:

[0012] According to the natural rock resistivity range and stratum distribution depth characteristics in actual geological data, a corresponding constrained layered resistivity model is designed.

[0013] Furthermore, the transient electromagnetic response data corresponding to the layered resistivity model is generated through forward modeling, including:

[0014] The transient electromagnetic response data corresponding to the layered resistivity model are obtained by a one-dimensional transient electromagnetic forward modeling program based on Hankel transform and cosine transform, sampling time-apparent resistivity;

[0015] The synthetic sample set includes: t Transient electromagnetic data D measured by time channels at different times; number of layers N in the layered resistivity model p The depth-resistivity geoelectric model M synthesized by the corresponding resistivity values ​​divides the sample set into training set, validation set and test set.

[0016] Furthermore, the Informer model is constructed, including:

[0017] The Informer model includes an encoder, a decoder, and a multi-head probabilistic sparse self-attention. The input data of the model is transient electromagnetic response data. The dimension of the input layer is composed of time channels N tThe output data of the model is the depth-resistivity geoelectric model, and the dimension of the output layer is determined by the number of layers N of the layered resistivity model. p Determine; input the feature vector obtained by the encoder into the decoder for decoding, and output the predicted formation resistivity model parameters.

[0018] Furthermore, the electromagnetic response data is used as input data for Informer model training, and the optimizer is used to dynamically adjust the learning rate to obtain the trained depth-resistivity formation model, including:

[0019] The training set is used to train the Informer model. During the training process, the parameters of the model are adjusted by minimizing the error between the predicted results and the true model. The adaptive moment estimation algorithm Adam is used as the optimization algorithm, and the ReduceLROnPlateau optimizer is used to dynamically adjust the learning rate.

[0020] Furthermore, the Adam algorithm is used to optimize the weights and biases, and the ReduceLROnPlateau optimizer sets the initial learning rate lr = 0.01, the period adjustment parameter patience = 10, and the learning rate adjustment factor parameter factor = 0.1.

[0021] Furthermore, the trained depth-resistivity formation model is used to invert the real transient electromagnetic response data collected by the input to obtain the actual formation resistivity distribution, including:

[0022] The preprocessed real transient electromagnetic response data is input into the trained depth-resistivity formation model. The model extracts and maps the features of the input data through a deep learning algorithm and outputs a preliminary distribution image of the formation resistivity.

[0023] In a second aspect, the present invention provides a transient electromagnetic deep learning inversion system, comprising:

[0024] A data acquisition module is used to construct a layered resistivity model and generate transient electromagnetic response data corresponding to the layered resistivity model through forward modeling;

[0025] The model building module is used to build the Informer model, use the electromagnetic response data as input data to train the Informer model, and use the optimizer to dynamically adjust the learning rate to obtain the trained depth-resistivity formation model;

[0026] The inversion output module is used to use the trained depth-resistivity formation model to invert the real transient electromagnetic response data collected by the input to obtain the actual formation resistivity distribution.

[0027] In a third aspect, the present invention provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of a transient electromagnetic deep learning inversion method when executing the computer program.

[0028] In a fourth aspect, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of a transient electromagnetic deep learning inversion method are implemented.

[0029] Compared with the prior art, the present invention has the following technical effects:

[0030] The present invention uses the transient electromagnetic response data (converted into sampling time-apparent resistivity in this paper) corresponding to the layered resistivity model generated by forward modeling, uses the electromagnetic response data as the input data feature to train the Informer model, and uses the ReduceLROnPlateau optimizer to dynamically adjust the learning rate, and finally obtains the depth-resistivity formation model as the output result; in view of the nonlinear and complex relationship between transient electromagnetic data and the actual geological model, the present invention uses the Informer model in transient electromagnetic inversion and verifies its effectiveness. Compared with the current common inversion methods, the present invention has the advantages of high inversion efficiency and results that fit the actual geological conditions. The Informer model can accurately capture the long-term dependencies in the input sequence based on the integrity of the sequence information. This feature is very important for processing long sequence time series problems and helps to maintain the coherence and integrity of the sequence. Compared with traditional models such as LSTM and GRU, when processing long sequences, the Informer model can more effectively retain important information in the sequence, thereby improving the accuracy of inversion.

[0031] The Informer model introduced in the present invention has higher generalization ability and operation speed. This means that the Informer model can show better performance on different data sets and can complete the calculation task faster. In transient electromagnetic inversion, this feature will greatly improve the efficiency of inversion and also help to process large-scale data.

[0032] In transient electromagnetic inversion, the present invention needs to process a large amount of time series data. Neural network structures such as CNN and GNN lack the ability to process time series problems, while RNN and the improved LSTM, Transformer and other model architectures are inefficient due to the serial processing mode. The original design of the Informer model is to deal with long sequence time series problems, so it shows excellent performance in processing such tasks. It can effectively process long sequence data and improve the calculation speed while maintaining the inversion accuracy, providing a new solution for transient electromagnetic deep neural network inversion. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a logic block diagram of the inversion algorithm of an example of the present invention.

[0034] Figure 2 Schematic diagram of the composition of the training sample set of the example of the present invention.

[0035] Figure 3 Schematic diagram of the network structure of the deep learning Informer model of the example of the present invention.

[0036] Figure 4 1 and 2. The inversion results of the three-layer geoelectric model of the present invention are exemplified, wherein (a) is model 1; (b) is model 2; (c) is model 7; (d) is model 27; (e) is model 35; (f) is model 46; (g) is model 61; and (g) is model 93.

[0037] Figure 5 This is an example of the inversion results of a five-layer geoelectric model according to an example of the present invention, where (a) is model 9; (b) is model 23; (c) is model 28; (d) is model 39; (e) is model 58; (f) is model 82; (g) is model 93; and (g) is model 99.

[0038] Figure 6 Comparison of the inversion methods provided for the examples of the present invention, where (a) is the inversion results of the three methods of Occam, LSTM and Informer for the K-type three-layer geoelectric model, and (b) is the inversion results of the three methods of Occam, LSTM and Informer for the HKH-type five-layer geoelectric model.

[0039] Figure 7 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0040] The present invention is further described below in conjunction with the accompanying drawings:

[0041] Example 1, please refer to Figure 7 The present invention provides a transient electromagnetic deep learning inversion method, comprising:

[0042] Construct a layered resistivity model and generate transient electromagnetic response data corresponding to the layered resistivity model through forward modeling;

[0043] Construct an Informer model, use the electromagnetic response data as input data to train the Informer model, and use the optimizer to dynamically adjust the learning rate to obtain the trained depth-resistivity formation model;

[0044] The trained depth-resistivity formation model is used to invert the real transient electro-magnetic response data collected by the input to obtain the actual formation resistivity distribution.

[0045] Example 2, please refer to Figure 1 The present invention provides a transient electromagnetic deep learning inversion method, which specifically includes:

[0046] ⑴Sample set generation:

[0047] ①First, design a corresponding constrained layered resistivity model based on the natural rock resistivity range and stratum distribution depth characteristics in actual geological data;

[0048] ② Obtain the transient electromagnetic response data (sampling time-apparent resistivity) corresponding to the layered resistivity model generated in the previous step through a transient electromagnetic one-dimensional forward modeling program based on Hankel transform and cosine transform;

[0049] The synthetic sample set includes: t Transient electromagnetic data D measured by time channels at different times; number of layers N in the layered resistivity model p The corresponding resistivity values ​​are used to synthesize the depth-resistivity geoelectric model M, and the sample set is divided into a training set, a validation set, and a test set. 90% of the data is used for training, 5% for validation, and 5% for testing the performance of the model.

[0050] ⑵Build the Informer model: The Informer model consists of an encoder, a decoder, and a multi-head probable sparse self-attention, which can capture long-term dependencies in the sequence.

[0051] ① The input data of the model is transient electromagnetic response data. The dimension of the input layer is composed of time channel N t Decide;

[0052] ② The output data of the model is the depth-resistivity geoelectric model. The dimension of the output layer is determined by the number of layers N of the layered resistivity model. p Decide;

[0053] ③ Input the feature vector obtained by the encoder into the decoder for decoding, and output the predicted formation resistivity model parameters.

[0054] ⑶ Model training: Use the training set to train the Informer model. During the training process, the parameters of the model are adjusted by minimizing the error between the predicted result and the true model (this paper uses the mean square error between the two). This paper uses the adaptive moment estimation algorithm (Adam) as the optimization algorithm and the ReduceLROnPlateau optimizer to dynamically adjust the learning rate.

[0055] (4) Model validation: Use the validation set to evaluate the performance of the trained Informer model. Calculate the mean square error between the predicted results and the true value to verify the accuracy of the model.

[0056] ⑸ Model inversion: Use the trained Informer model to quickly invert the real transient electromagnetic response data collected by the input to obtain the actual formation resistivity distribution.

[0057] The above steps are used to implement the transient electromagnetic deep learning inversion method based on the Informer model.

[0058] In step ① of process (1), the number of layers of the geoelectric model is N p =3, the formation resistivity of the generated geoelectric model is between 20Ω·m and 500Ω·m, and the resistivity difference between adjacent formations is greater than 1%, the formation thickness is between 30m and 200m, and the inter-layer difference is greater than 1%; the number of layers N in the geoelectric model p =5, the formation resistivity of the generated geoelectric model is between 20Ω·m and 600Ω·m, and the resistivity difference between adjacent formations is greater than 1%, the formation thickness is between 40m and 300m, and the inter-layer difference is greater than 1%;

[0059] In process (1) step ②, the forward modeling program sets the sampling time start and end interval to [0.1, 10] ms, the emission current to 1 A, the number of time channels to 60, and the extraction is performed at equal logarithmic intervals.

[0060] In process (2), the number of LSTM layers is set to 2, the number of nodes in the hidden layer is set to 128, the number of attention heads of the multi-head probabilistic sparse self-attention mechanism is set to 4, and the batch size is set to 64.

[0061] In process (3), the Adam algorithm is used to optimize the weights and biases, and the ReduceLROnPlateau optimizer sets the initial learning rate lr = 0.01, the period adjustment parameter patience = 10, and the learning rate adjustment factor parameter factor = 0.1.

[0062] Example 3: Sample set generation

[0063] The present invention uses a large amount of sampling time-apparent resistivity obtained by transient electromagnetic forward simulation as input data, and based on the deep learning Informer model, adjusts the hyperparameter settings and the network structure according to the problem characteristics of transient electromagnetic inversion, and obtains the final depth-resistivity geoelectric model output data (i.e., the formation resistivity distribution) after training.

[0064] The basic idea of ​​transient electromagnetic data inversion is to establish the mapping relationship between model parameters and transient electromagnetic response through a large number of theoretical model calculations. Transient electromagnetic forward modeling obtains the electromagnetic data response of the observation point by establishing a geoelectric model. In an ideal state, the forward modeling process of transient electromagnetic response is represented by a function, which is:

[0065] d=f(m)

[0066] Among them, d is the electromagnetic response data observed at the receiving point (converted into sampling time-apparent resistivity in this paper); m represents the distribution of the geoelectric model (depth-resistivity). Then the inversion of transient electromagnetic method can be expressed as its inverse function:

[0067] m=y(d)

[0068] Traditional inversion methods require a lot of complex mathematical calculations to establish the physical connection between underground resistivity distribution and electromagnetic field. The introduction of deep learning models can transform the above complex inverse function solution process into a matrix mapping relationship:

[0069] T θ :D→M

[0070] The vector spaces D and M have bases d1=f(m1), d2=f(m2)...d i =f(m i ) and m1=f(d1), m2=f(d2)...m i =f(d i ), θ is the hyperparameter setting of the Informer model after training and the grid structure data.

[0071] 2. Building the Informer Model

[0072] The Informer model is an improvement on the Transformer model, and its overall structure still retains the encoder-decoder architecture. At the encoder layer, the multi-head probable sparse self-attention mechanism is used to filter out the most important queries, reducing the computational complexity of the traditional self-attention mechanism; self-attention distillation is used to gradually reduce the sequence length to reduce the network dimension, reduce the number of network parameters, and improve the efficiency and performance of the model; a generative decoding mechanism is proposed at the decoding layer, which directly outputs the results when predicting the sequence, greatly improving the inversion speed of the model.

[0073] The calculation formula of the probabilistic sparse self-attention mechanism is as follows: The input sequence is the transient electromagnetic response D = (d1, d2...d i ), d i is the i-th element of the input sequence. First, three different representations are calculated through linear transformation: Q = DWQ, K = DWK, V = DWV, where WQ, WK, WV are learnable weight matrices. Then, the scaled dot product operation is performed through query-key-value, and after optimization, the expression is obtained as follows:

[0074]

[0075] Where Q is the query matrix; K is the key matrix; V is the value matrix; d k is the input dimension of the key. is a sparse matrix of the same size as Q, and Only included in the sparse evaluation M(q i ,K) of the Top-u Query under. According to the above theory, the computational complexity is reduced from O(L 2 ) is reduced to O(LlnL), thereby reducing the computational cost of each layer of the sparse self-attention mechanism.

[0076] The distillation mechanism gives higher weight to the dominant features of the main information and generates feature maps in the next layer, so that the length of the input sequence is halved layer by layer to reduce the high-dimensional space complexity and computational memory consumption caused by long sequence input. At the n+1 layer, an n-layer focused self-attention feature map is generated. The expression of the distillation operation is:

[0077]

[0078] Where: is the n+1th layer matrix; It represents a one-dimensional convolution operation on the attention module. It uses the exponential linear unit (ELU) as the activation function and passes through the maximum pooling layer (Max-pooling) to halve the amount of computation in each layer.

[0079] For the decoder, the input sequence is divided into a known sequence and a predicted sequence that needs to mask the future feature factor data, expressed as:

[0080]

[0081] Where: is the input transient electromagnetic response time series; Concat(·) represents the concatenation operation; For historical sequence; The forecast sequence masked by the future characteristic factor data; L token is the known sequence length; L y is the length of the prediction sequence; dm is the dimension of the prediction model.

[0082] In order to reduce the interference of the autoregressive process on the training results, a masked sparse multi-head attention mechanism is introduced to avoid the influence of unpredicted information on the current prediction point. The generative decoder can generate the entire prediction result at one time during the training process, which significantly improves the decoding efficiency.

[0083] 3 Model Training

[0084] Transient electromagnetic inversion is a complex nonlinear process that requires a lot of training to well characterize the resistivity distribution of the formation. Therefore, the ratio of training, validation and test samples is set to 0.9:0.05:0.05. Other hyperparameter settings are shown in Table 1.

[0085] Learning rate plays a vital role in deep learning. It directly affects the speed, stability and generalization ability of model training. A suitable learning rate can help the model converge to the optimal solution faster while avoiding falling into the local optimal solution or oscillation. Too low a learning rate will cause the model to converge slowly or even stagnate during training and fail to achieve ideal performance. On the contrary, too high a learning rate may cause the loss function to fluctuate around the minimum value, making it impossible for the model to converge stably.

[0086] The present invention uses the ReduceLROnPlateau optimizer to dynamically adjust the learning rate. This optimizer can automatically adjust the learning rate according to the changes in the monitoring indicators, thereby better guiding the model training process. When the performance indicators on the validation set stop improving, ReduceLROnPlateau will gradually reduce the learning rate so that the model can better converge to the optimal solution.

[0087] In this way, the model can automatically adjust the learning rate during training to adapt to the characteristics of the data and the performance of the model. This adaptive learning rate adjustment method can help the model converge faster, improve generalization ability, and avoid overfitting or underfitting problems.

[0088] Table 1 Hyperparameter settings

[0089]

[0090]

[0091] 4 Model Validation

[0092] In order to verify the inversion results of the trained model for the three-layer model, the forward model data generated by the parameter settings in Table 2 were used to test the model adaptability. Eight groups of data were randomly selected from the test set. The inversion results of the model for the formation resistivity distribution and the target results are shown in Figure 2. Figure 5 As shown in the figure, all eight groups of samples accurately reflect the formation resistivity and thickness information of the real model. Most of the curve shapes are completely consistent, and a few have certain fitting errors, but they are all less than the maximum allowable error. It is proved that the transient electromagnetic deep learning inversion method based on the Informer model has a good learning ability for the three-layer geoelectric model.

[0093] Table 2 Parameter settings of the three-layer model

[0094]

[0095] Furthermore, the forward model data generated by the parameter settings in Table 3 are used to verify the inversion effect of the trained model on the five-layer model. Eight groups of data are randomly selected from the test set, and the results are as follows: Figure 6 As shown. The transient electromagnetic deep learning inversion method based on the Informer model shows good learning ability in the five-layer geoelectric model. Through the analysis of 8 groups of samples, we observed that although the inversion results of the five-layer model are less accurate than those of the three-layer model, these samples accurately reflect the formation resistivity and thickness information, and most of the curve shapes are completely consistent. Only a few samples have slight fitting errors, but these errors are within the maximum allowable error range. This result not only confirms the effectiveness of the deep learning method, but also highlights its potential in solving geological exploration problems, especially compared with traditional methods, it has a higher degree of automation and data processing efficiency.

[0096] Table 3 Parameter settings of the five-layer model

[0097]

[0098] 5 Model inversion

[0099] The Occam and LSTM inversion methods were introduced for comparative analysis, and the measured transient electromagnetic response data were randomly selected for inversion (ensuring that the training volume of the LSTM and Informer models is consistent). The results of the three inversion methods are shown in Figure 2. Figure 6 As shown in Table 4 and Table 5, the relative errors of the inversion results of the Informer model for the three-layer geoelectric model and the four-layer geoelectric model are smaller than those of the LSTM inversion results, indicating that the Informer model is a method with too high precision.

[0100] Table 4 Parameter settings and inversion results of the three-layer geoelectric model

[0101]

[0102] Table 5 Parameter settings and inversion results of the five-layer geoelectric model

[0103]

[0104]

[0105] In yet another embodiment of the present invention, a transient electromagnetic deep learning inversion system is provided, which can be used to implement the above-mentioned transient electromagnetic deep learning inversion method. Specifically, the system includes:

[0106] A data acquisition module is used to construct a layered resistivity model and generate transient electromagnetic response data corresponding to the layered resistivity model through forward modeling;

[0107] The model building module is used to build the Informer model, use the electromagnetic response data as input data to train the Informer model, and use the optimizer to dynamically adjust the learning rate to obtain the trained depth-resistivity formation model;

[0108] The inversion output module is used to use the trained depth-resistivity formation model to invert the real transient electromagnetic response data collected by the input to obtain the actual formation resistivity distribution.

[0109] The division of modules in the embodiments of the present invention is schematic and is only a logical function division. There may be other division methods in actual implementation. In addition, each functional module in each embodiment of the present invention may be integrated into one processor, or may exist physically separately, or two or more modules may be integrated into one module. The above-mentioned integrated modules may be implemented in the form of hardware or in the form of software functional modules.

[0110] In another embodiment of the present invention, a computer device is provided, the computer device comprising a processor and a memory, the memory being used to store a computer program, the computer program comprising program instructions, and the processor being used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc., which are the computing core and control core of the terminal, which are suitable for implementing one or more instructions, and are specifically suitable for loading and executing one or more instructions in a computer storage medium to implement a corresponding method flow or corresponding function; the processor described in the embodiment of the present invention can be used for the operation of a transient electromagnetic deep learning inversion method.

[0111] In another embodiment of the present invention, the present invention also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device for storing programs and data. It is understandable that the computer-readable storage medium here can include both built-in storage media in a computer device and, of course, extended storage media supported by the computer device. The computer-readable storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by a processor are also stored in the storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of a transient electromagnetic deep learning inversion method in the above embodiment.

[0112] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0113] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0114] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0115] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0116] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A transient electromagnetic deep learning inversion method, characterized in that: include: Construct a layered resistivity model and generate transient electromagnetic response data corresponding to the layered resistivity model through forward modeling; Construct an Informer model, use the electromagnetic response data as input data to train the Informer model, and use the optimizer to dynamically adjust the learning rate to obtain the trained depth-resistivity formation model; The trained depth-resistivity formation model is used to invert the real transient electro-magnetic response data collected by the input to obtain the actual formation resistivity distribution.

2. A transient electromagnetic deep learning inversion method according to claim 1, characterized in that: Construct a layered resistivity model, including: According to the natural rock resistivity range and stratum distribution depth characteristics in actual geological data, a corresponding constrained layered resistivity model is designed.

3. A transient electromagnetic deep learning inversion method according to claim 1, characterized in that: The transient electromagnetic response data corresponding to the layered resistivity model is generated through forward modeling, including: The transient electromagnetic response data corresponding to the layered resistivity model are obtained by a one-dimensional transient electromagnetic forward modeling program based on Hankel transform and cosine transform, sampling time-apparent resistivity; The synthetic sample set includes: t Transient electromagnetic data D measured by time channels at different times; number of layers N in the layered resistivity model p The depth-resistivity geoelectric model M synthesized by the corresponding resistivity values ​​divides the sample set into training set, validation set and test set.

4. A transient electromagnetic deep learning inversion method according to claim 3, characterized in that: Build the Informer model, including: The Informer model includes an encoder, a decoder, and a multi-head probabilistic sparse self-attention. The input data of the model is transient electromagnetic response data. The dimension of the input layer is composed of time channels N t The output data of the model is the depth-resistivity geoelectric model, and the dimension of the output layer is determined by the number of layers N of the layered resistivity model. p Determine; input the feature vector obtained by the encoder into the decoder for decoding, and output the predicted formation resistivity model parameters.

5. The transient electromagnetic deep learning inversion method according to claim 1, characterized in that: The electromagnetic response data is used as input data to train the Informer model, and the optimizer is used to dynamically adjust the learning rate to obtain the trained depth-resistivity formation model, including: The training set is used to train the Informer model. During the training process, the parameters of the model are adjusted by minimizing the error between the predicted results and the true model. The adaptive moment estimation algorithm Adam is used as the optimization algorithm, and the ReduceLROnPlateau optimizer is used to dynamically adjust the learning rate.

6. A transient electromagnetic deep learning inversion method according to claim 5, characterized in that: The Adam algorithm is used to optimize weights and biases. The ReduceLROnPlateau optimizer sets the initial learning rate lr = 0.01, the period adjustment parameter patience = 10, and the learning rate adjustment factor parameter factor = 0.

1.

7. A transient electromagnetic deep learning inversion method according to claim 1, characterized in that: The trained depth-resistivity formation model is used to invert the real transient electromagnetic response data collected by the input to obtain the actual formation resistivity distribution, including: The preprocessed real transient electromagnetic response data is input into the trained depth-resistivity formation model. The model extracts and maps the features of the input data through a deep learning algorithm and outputs a preliminary distribution image of the formation resistivity.

8. A transient electromagnetic deep learning inversion system, characterized in that: include: A data acquisition module is used to construct a layered resistivity model and generate transient electromagnetic response data corresponding to the layered resistivity model through forward modeling; The model building module is used to build the Informer model, use the electromagnetic response data as input data to train the Informer model, and use the optimizer to dynamically adjust the learning rate to obtain the trained depth-resistivity formation model; The inversion output module is used to use the trained depth-resistivity formation model to invert the real transient electromagnetic response data collected by the input to obtain the actual formation resistivity distribution.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the transient electromagnetic deep learning inversion method as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of a transient electromagnetic deep learning inversion method as described in any one of claims 1 to 7 are implemented.

Citation Information

Cited By

  • Transient electromagnetic rapid inversion method based on automatic differentiation

    CN120195757A

  • Transient electromagnetic vertical component detection data one-dimensional real-time accurate inversion method and system

    CN120762119A