Transient electromagnetic inversion model construction method based on target data set driving

Abstract

The invention relates to a transient electromagnetic inversion model construction method based on target data set driving, and the method comprises the steps: constructing an initial training set, training a first convolutional neural network through the initial training set, and obtaining an initial transient electromagnetic inversion network model; inputting to-be-measured electromagnetic response data into the initial transient electromagnetic inversion model for prediction to obtain a resistivity value; performing forward modeling on the resistivity value to obtain electromagnetic response data of the forward modeling, and constructing prediction data based on the electromagnetic response data of the forward modeling and the resistivity value; extracting data similar to the to-be-tested electromagnetic response data in the initial training set, and constructing a target data set based on the similar data and the prediction data; and parameters in the initial transient electromagnetic inversion model are migrated to the second convolutional neural network to carry out iterative learning on the target data set pair, and the transient electromagnetic inversion model is obtained. The invention aims to improve the inversion precision of the to-be-tested data by the network when deep learning inversion is carried out on the to-be-tested data through the iterative training strategy.

Description

Technical Field

[0001] The present invention relates to the field of transient electromagnetic inversion technology, and in particular to a method for constructing a transient electromagnetic inversion model driven by a target data set. Background Art

[0002] At present, transient electromagnetic inversion is mainly divided into two types: linear inversion and nonlinear inversion. Linear inversion methods, such as Occam inversion and Marquardt inversion, are highly dependent on the initial model, and the quality of the inversion results depends largely on whether a suitable initial model can be found; nonlinear inversion methods, such as simulated annealing algorithm, particle swarm algorithm and Bayesian inversion, usually require huge amounts of computation due to the complexity of geophysical models. With the improvement of hardware computing power, electromagnetic inversion based on deep learning has become a research hotspot. Vladimir Puzyrev (2019) first explored the potential of deep learning methods in electromagnetic inversion. Subsequently, in recent years, deep learning-based methods have also been widely studied in the fields of CSEM inversion, MT inversion, AEM inversion, etc.

[0003] However, the above-mentioned deep learning-based inversion methods are completely data-driven, and their performance depends largely on the richness of the training sample set. When processing data with different distributions from the training sample set, the effect is often unsatisfactory. In response to this, many scholars have made improvements by adding physical constraints to the loss function of the deep learning network to control the training process of the network. For example, Jin et al. (2019) introduced the Jacobi differential operator in the forward model of the electromagnetic response of logging while drilling (LWD) predicted by convolutional neural networks to construct a composite loss function of the model and data mismatch function. Sun et al. (2020) constructed a forward operator Γ based on a recurrent neural network (RNN) to simulate wave propagation and realized unsupervised deep learning seismic inversion. Its training process is equivalent to the optimization of conventional deterministic inversion methods. W.Liu et al. (2022) proposed to incorporate the physical laws of magnetotelluric wave propagation into a purely data-driven deep learning method (PlainDNN), adding physically based unfitted data to the loss function to guide network training, and verified it in magnetotelluric one-dimensional inversion. In addition, some scholars have combined deep learning with traditional methods to give full play to the advantages of both methods. For example, Asifetal. (2022) predicted the Jacobian matrix in the least squares inversion through a neural network, integrated the neural network into the traditional least squares inversion, and verified it in transient electromagnetic one-dimensional inversion. On this basis, Asifetal. (2022) obtained the forward operator through neural network training, integrated it into the least squares inversion, and finally achieved excellent performance in aviation transient electromagnetic one-dimensional inversion.

[0004] Due to the complexity, non-uniqueness, and high requirements for time and computing resources of geophysical models, it is quite challenging to construct large and detailed geophysical datasets and train networks. In addition, purely data-driven deep learning methods mainly learn the inversion operator L hidden between the input and output inside the training set, so when testing on data outside the training set, the effect is often unsatisfactory. In other words, data-driven deep learning inversion methods have difficulty dealing with the fine details between data and models when faced with unseen data, unless large-scale training is performed when the data to be tested has the same statistical distribution. This leads to certain limitations in the practical application of data-driven machine learning methods in geophysical inversion.

[0005] Although the improvements of the above two methods have reduced the dependence on large-scale data to a certain extent and improved the generalization ability of the network, they have also complicated the network training process to a certain extent and increased the consumption of computing resources in the training stage. Summary of the invention

[0006] The purpose of the present invention is to provide a transient electromagnetic inversion model construction method based on target data set driving, which gradually approaches the real inversion result by repeatedly updating the model and adjusting parameters, thereby overcoming the limitations of traditional data-driven methods and improving the inversion accuracy and reliability.

[0007] To achieve the above object, the present invention provides the following solutions:

[0008] The transient electromagnetic inversion model construction method based on the target data set drive includes:

[0009] Constructing an initial training set, and using the initial training set to train a first convolutional neural network to obtain an initial transient electromagnetic inversion network model;

[0010] Inputting the electromagnetic response data to be measured into the initial transient electromagnetic inversion model for prediction to obtain the resistivity value;

[0011] Performing forward simulation on the resistivity value to obtain electromagnetic response data of the forward simulation, and constructing prediction data based on the electromagnetic response data of the forward simulation and the resistivity value;

[0012] Extracting data similar to the electromagnetic response data to be tested in the initial training set, and constructing a target data set based on the similar data and the predicted data;

[0013] The parameters in the initial transient electromagnetic inversion model are transferred to the second convolutional neural network to iteratively learn the target data set pair to obtain a transient electromagnetic inversion model.

[0014] Optionally, obtaining the initial training set includes:

[0015] The thickness of the top layer and the depth of the bottom boundary of the underground geological structure are preset, and divided into several layers according to the method of cumulative increasing thickness of each layer;

[0016] Determine a plurality of control points between the top layer and the bottom layer, interpolate the control points, and obtain resistivity-thickness sample data between the top layer and the bottom layer;

[0017] Perform transient electromagnetic one-dimensional numerical simulation according to the resistivity-thickness sample data to obtain corresponding electromagnetic response;

[0018] Based on the resistivity value and the corresponding electromagnetic response, a sample data set is obtained, and the initial training set is divided from the sample data set.

[0019] Optionally, interpolating the control points includes: interpolating using a B-spline interpolation method.

[0020] Optionally, the convolutional neural network includes: an input layer, an output layer, a convolutional layer and a fully connected layer, wherein the input layer is used to input the electromagnetic response, the output layer is used to output the resistivity value, and the data features extracted by the convolutional layer are input into the fully connected layer.

[0021] Optionally, extracting data in the initial training set that is similar to the electromagnetic response data to be tested includes:

[0022] Obtain a relative average error between the electromagnetic response data in the initial training set and the electromagnetic response data to be measured; obtain the electromagnetic response data in the initial training set corresponding to the relative average error in the preset range as similar electromagnetic response data, and extract the resistivity value corresponding to the similar electromagnetic response data in the initial training set to obtain the similar data.

[0023] The present invention also provides a transient electromagnetic inversion method driven by a target data set, and uses a transient electromagnetic inversion model construction method driven by a target data set to construct a transient electromagnetic inversion model, input the electromagnetic response data to be predicted into the transient electromagnetic inversion model, and obtain the predicted resistivity value.

[0024] The beneficial effects of the present invention are as follows: the present invention aims to improve the accuracy of the network inversion of the test data when the test data is subjected to deep learning inversion through an iterative training strategy. In this method, by repeatedly updating the model and adjusting the parameters, the real inversion result is gradually approached, thereby overcoming the limitations of the traditional data-driven method and improving the inversion accuracy and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] 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 will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0026] Figure 1 A framework diagram of a method for constructing a transient electromagnetic inversion model driven by a target data set according to an embodiment of the present invention;

[0027] Figure 2 A schematic diagram of sample set data generation according to an embodiment of the present invention;

[0028] Figure 3 This is a network framework diagram of iterative inversion driven by a target data set according to an embodiment of the present invention;

[0029] Figure 4 This is a network model training loss graph of an embodiment of the present invention;

[0030] Figure 5 This is a performance evaluation diagram of an initial network model according to an embodiment of the present invention;

[0031] Figure 6 A comparison diagram of network response errors of iterative inversion driven by target data set according to an embodiment of the present invention;

[0032] Figure 7 1 is a comparison diagram of a certain sample point to be tested in an embodiment of the present invention using different methods, wherein Figure (a) is a response data fitting diagram, and Figure (b) is a resistivity value fitting diagram. DETAILED DESCRIPTION

[0033] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0034] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0035] This embodiment provides a method for constructing a transient electromagnetic inversion model driven by a target data set, including:

[0036] Constructing an initial training set, using the initial training set to train the first convolutional neural network, and obtaining an initial transient electromagnetic inversion network model;

[0037] Input the electromagnetic response data to be measured into the initial transient electromagnetic inversion model for prediction to obtain the resistivity value;

[0038] Perform forward simulation on the resistivity value, obtain electromagnetic response data of the forward simulation, and construct prediction data based on the electromagnetic response data of the forward simulation and the resistivity value;

[0039] Extracting data similar to the electromagnetic response data to be tested from the initial training set, and constructing a target data set based on the similar data and the predicted data;

[0040] Furthermore, extracting data similar to the electromagnetic response data to be tested in the initial training set includes:

[0041] Obtain the relative average error of the response between the electromagnetic response data in the initial training set and the electromagnetic response data to be tested; obtain the electromagnetic response data in the initial training set corresponding to the relative average error of the response within a preset range as similar electromagnetic response data, and extract the resistivity value corresponding to the similar electromagnetic response data in the initial training set to obtain similar data.

[0042] The parameters in the initial transient electromagnetic inversion model are transferred to the second convolutional neural network to iteratively learn the target data set pair to obtain the transient electromagnetic inversion model.

[0043] Furthermore, obtaining the initial training set includes: presetting the top thickness and bottom boundary depth of the underground geological structure, and dividing it into several layers according to the method of cumulative increasing thickness of each layer; determining several control points between the top and bottom layers, interpolating the control points, and obtaining resistivity-thickness sample data between the top and bottom layers; performing transient electromagnetic one-dimensional numerical simulation based on the resistivity-thickness sample data to obtain corresponding electromagnetic responses; obtaining a sample data set based on the resistivity value and the corresponding electromagnetic response, and dividing the initial training set from the sample data set.

[0044] Specifically, considering that the resistivity values ​​of actual underground strata are usually distributed continuously rather than segmented, and the continuous distribution characteristics can well reflect the complexity of underground geological structures. Therefore, in order to generate relatively smooth and longitudinally continuous underground resistivity values, this embodiment uses the following method to generate sample set data: first, the thickness of the top layer is set to 1m, the boundary depth of the last layer is set to 750m, and each layer is divided into 15 layers according to the method of cumulative increasing thickness, so as to ensure that the random characteristics of underground resistivity distribution are fully reflected; 5 control points are determined between 1 and 750m, as shown in the attached figure. Figure 2As shown in the dark dots, the depths of two control points are fixed at 1m and 750m, and the other three control points are randomly determined, but the depth between the control points is required to be within 50-200m. The resistivity of the control points is randomly generated within 10-1000Ω·m, and the B-spline interpolation method is used to interpolate based on the five control points, thereby making the resistivity curve smoother, and finally obtaining a continuously distributed resistivity value from 1m to 750m. Figure 2 As shown. Using a rectangular loop source device, the generated resistivity-thickness sample data is substituted into the transient electromagnetic one-dimensional numerical simulation method to calculate the corresponding electromagnetic response. The coil length and width are 500m, the number of receiving time channels is 61, and the sampling time is 1e -4 ~1e -1 s, the emission current is 10A.

[0045] Furthermore, the convolutional neural network includes: an input layer, an output layer, a convolutional layer and a fully connected layer, wherein the input layer is used to input electromagnetic response, the output layer is used to output resistivity value, and the data features extracted by the convolutional layer are input into the fully connected layer.

[0046] Specifically, the convolutional neural network model framework is used as Figure 3 , the input of the network is electromagnetic response data, and the output is resistivity value. The network structure mainly consists of input layer and output layer, 4 convolutional layers and 2 fully connected layers. The input layer has 61 neurons, representing 61 electromagnetic response values; the output layer has 15 neurons, corresponding to 15 layers of resistivity values. The convolution kernel sizes of the 4 convolutional layers are 2×1, 3×1, 3×1 and 3×1, respectively, and the number of convolution kernels is 64, 128, 256 and 512, respectively, with a step size of 1; the pooling layer uses Max-pooling, with a size of 2×1 and a step size of 2. The data features extracted by the convolution layer are input into the fully connected layer through the flattening layer. The number of neurons in the two fully connected layers are 1024 and 512, respectively. The activation function uses tanh. In order to quantitatively evaluate the convergence of the network training process, the following loss function is defined:

[0047]

[0048] Where N represents the number of layers of resistivity value, represents the resistivity value of the jth layer of the network prediction model, y j Indicates the resistivity value corresponding to the real geoelectric model.

[0049] The Adam optimizer was used for network training, and the hyperparameters in the network were set as follows: learning rate lr = 0.0001, batch size batch_size = 1024, training epochs = 5000, and early stopping mechanism was used for training strategy. A total of 40,000 sample data sets were generated, and the data sets were divided according to the ratio of training set: validation set: test set = 6.3:2.7:1.0.

[0050] like Figure 4 The figure is the loss function image of the model network training. The dotted line is the validation set error, and the solid line is the training set error. The early stopping mechanism is used, and it can be seen that there is no overfitting phenomenon in the model training process. It can be seen that the loss value on the training set (Train loss) and the loss value on the validation set (Validation loss) continue to decrease during the training process and eventually converge. The decrease in the loss value on the training set indicates that the model can gradually learn the patterns and rules in the training data and continuously optimize the parameters of the model. This shows that the model has no obvious overfitting phenomenon during the training process and has a strong fitting ability.

[0051] This embodiment selects three geoelectric model data corresponding to the minimum value, average value and maximum value of the relative average error from the test set. Figure 5 The resistivity-depth curve comparison between the original model and the inversion model and the electromagnetic response comparison between the original model and the inversion model of the three models are shown. Figure 5 Three data with the minimum, average and maximum relative average errors are selected to draw the resistivity-depth map of the original geoelectric model and the inverted geoelectric model (first row) and the electromagnetic response signal map of the original geoelectric model and the inverted geoelectric model (second row). It can be observed from the figure that the inversion model fits the original model well. Although there is a certain gap in some layers, it is still within an acceptable range overall. In particular, for the model with the largest relative average error, although there are large errors in the resistivity prediction of some layers, the overall trend of resistivity with depth is consistent with the original model, and its corresponding electromagnetic response error is only 9.4%. This further illustrates the high accuracy of the inversion results. It meets the assumption that the deep learning inversion method is efficient and the accuracy can be guaranteed to a certain extent.

[0052] Specifically, Figure 1The upper left module is the initial network model training part, the upper right part is the data set required for iterative inversion, and the lower part is the iterative training part. First, the initial training set is generated by numerical simulation, that is, the Train data in the upper left module, and then the convolutional neural network model is trained to obtain the initial transient electromagnetic inversion network model, and finally the electromagnetic response data to be measured, that is, the Objective data in the upper right module, is predicted for the first time to obtain the corresponding resistivity value DL Predict. Then, the electromagnetic response data in the Train data and the electromagnetic response data to be measured (Objective data) are extracted. The data with a relative average error (such as formula 2) within 5% is extracted, that is, Objective dataend, including the corresponding resistivity value. Afterwards, the Objective data end is merged with the DL Predict (resistivity value) and the electromagnetic response obtained by forward simulation of the DLPredict to form the data set New Data (in the lower module) required for iterative inversion. Finally, a transfer learning strategy is adopted for the initial network Model-1, i.e., the initial transient electromagnetic inversion network model. That is, the network framework is kept unchanged, the model parameters in Model-1 are migrated to Model (the second convolutional neural network) to initialize the model parameters, and then the New Data is learned. This process is iterated multiple times to achieve iterative inversion driven by the target data set.

[0053]

[0054] Where M is the number of time channels, d j is the real electromagnetic response of the jth time channel, Predict the electromagnetic response of the network for the jth time channel.

[0055] Figure 6 The solid line in the middle is the model response error predicted by the initial network model for 50 test data, and the dotted line is the model response error predicted by the iterative inversion network driven by the target data set. Figure 6 It can be clearly observed that after iterative inversion driven by the target data set, except for individual data, the overall response error of the data to be measured is reduced to a low level, indicating that the iterative inversion strategy driven by the target data set has a certain effect. In addition, by comparing the results of a certain sample point under different methods, such as Figure 7 , where the dotted line is the true value, the dashed line is the result obtained by the initial network inversion, and the dotted line is the result after 5 iterations of the iterative inversion network driven by the target data set. Figure 7 It can be observed from the sub-graph (a) that after 5 iterations, the response error of the data approaches the true response value, and the response accuracy is improved to a certain extent. Figure 7(b) It can be clearly observed that the result after 5 iterations (dashed line) is closer to the true resistivity value, which is more obvious in the shallow layer with a depth of 100-200m. In summary, this embodiment improves the inversion accuracy of the network for the measured data through simple iterative inversion.

[0056] Based on the transient electromagnetic inversion network of deep learning, without taking any measures, the relative average error of the model and the relative average error of the response of the inversion of 50 samples different from the training set (the number of samples of the measured data of a single survey line is relatively close) are 0.0504 and 0.0310 respectively. The accuracy is high, which meets the assumption that the deep learning inversion method is efficient and the accuracy can be guaranteed to a certain extent.

[0057] The target dataset proposed in this embodiment consists of two parts:

[0058] (1) Data similar to the data to be tested in the training set, that is, data sets with response errors less than the specified threshold (default 5%); (2) Data sets consisting of resistivity values ​​predicted by the network and corresponding responses, and the transfer learning strategy is used for the initial training model to learn the target data set. After 5 iterations of inversion, the proposed method has a model relative average error and a response relative average error of 0.0379 and 0.0105, respectively. Compared with the initial prediction, both the model error and the response error are reduced by 3 to 5 times.

[0059] In summary, this embodiment improves the inversion accuracy of the network for the data to be measured through simple iterative inversion.

[0060] This embodiment also provides a transient electromagnetic inversion method driven by a target data set. A transient electromagnetic inversion model is constructed by using a transient electromagnetic inversion model construction method driven by a target data set. The electromagnetic response data to be predicted is input into the transient electromagnetic inversion model to obtain a predicted resistivity value.

[0061] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for constructing a transient electromagnetic inversion model based on a target data set drive, characterized in that: include: Constructing an initial training set, and using the initial training set to train a first convolutional neural network to obtain an initial transient electromagnetic inversion network model; Inputting the electromagnetic response data to be measured into the initial transient electromagnetic inversion model for prediction to obtain the resistivity value; Performing forward simulation on the resistivity value to obtain electromagnetic response data of the forward simulation, and constructing prediction data based on the electromagnetic response data of the forward simulation and the resistivity value; Extracting data similar to the electromagnetic response data to be tested in the initial training set, and constructing a target data set based on the similar data and the predicted data; The parameters in the initial transient electromagnetic inversion model are transferred to the second convolutional neural network to iteratively learn the target data set pair to obtain a transient electromagnetic inversion model.

2. The method for constructing a transient electromagnetic inversion model based on target data set drive according to claim 1, characterized in that: Acquiring the initial training set includes: The thickness of the top layer and the depth of the bottom boundary of the underground geological structure are preset, and divided into several layers according to the method of cumulative increasing thickness of each layer; Determine a plurality of control points between the top layer and the bottom layer, interpolate the control points, and obtain resistivity-thickness sample data between the top layer and the bottom layer; Perform transient electromagnetic one-dimensional numerical simulation according to the resistivity-thickness sample data to obtain corresponding electromagnetic response; Based on the resistivity value and the corresponding electromagnetic response, a sample data set is obtained, and the initial training set is divided from the sample data set.

3. The method for constructing a transient electromagnetic inversion model based on target data set drive according to claim 2, characterized in that: Interpolating the control points includes: interpolating using a B-spline interpolation method.

4. The method for constructing a transient electromagnetic inversion model based on target data set drive according to claim 2, characterized in that: The convolutional neural network includes: an input layer, an output layer, a convolutional layer and a fully connected layer, wherein the input layer is used to input the electromagnetic response, the output layer is used to output the resistivity value, and the data features extracted by the convolutional layer are input to the fully connected layer.

5. The method for constructing a transient electromagnetic inversion model based on target data set drive according to claim 1, characterized in that: Extracting data in the initial training set that is similar to the electromagnetic response data to be tested includes: Obtain a relative average error between the electromagnetic response data in the initial training set and the electromagnetic response data to be measured; obtain the electromagnetic response data in the initial training set corresponding to the relative average error in the preset range as similar electromagnetic response data, and extract the resistivity value corresponding to the similar electromagnetic response data in the initial training set to obtain the similar data.

6. A transient electromagnetic inversion method driven by a target data set, characterized in that: A transient electromagnetic inversion model is constructed using the target data set driven transient electromagnetic inversion model construction method described in any one of claims 1 to 5, and the electromagnetic response data to be predicted is input into the transient electromagnetic inversion model to obtain a predicted resistivity value.

Images