Single-valued transient electromagnetic apparent resistivity calculation method, system and device based on deep learning and storage medium

Abstract

The invention provides a single-valued transient electromagnetic apparent resistivity calculation method based on deep learning, and the method comprises the following steps: a), carrying out the normalization processing of an actually-measured induced voltage, and obtaining a normalized induced magnetic field; b) a deep neural network model comprising an input layer, a plurality of hidden layers and an output layer is constructed, the input layer receives the normalized induced magnetic field, and the output layer outputs transient field parameters; c) training the deep neural network model by using a Nadam optimization algorithm, wherein a training sample set comprises multiple groups of actually measured induced voltages and transient field parameters corresponding to the actually measured induced voltages; and d) inputting the actually measured induced voltage to be measured into the trained deep neural network model, obtaining the output transient field parameters, and calculating the apparent resistivity according to the theoretical formula of the transient electromagnetic field. Compared with a traditional method, the method has the advantages that the fitting capacity of nonlinear data is higher, and the transient electromagnetic apparent resistivity can be solved more accurately.

Description

Technical Field

[0001] The present invention belongs to the field of electromagnetic underground surveying, and specifically relates to a single-value transient electromagnetic apparent resistivity calculation method, system, equipment and storage medium based on deep learning. Background Art

[0002] Transient electromagnetic method (TEM) is a time-domain electromagnetic detection method, which is widely used in hydrology, engineering, environment and other fields because it can detect the electrical changes of underground strata. The apparent resistivity of the whole area is an important parameter in transient electromagnetic method, which can directly reflect the electrical characteristics of underground strata. However, due to the complex implicit function relationship between resistivity parameters and transient response, it is quite difficult to directly solve the apparent resistivity by analytical method.

[0003] Although some direct calculation methods have been studied, these methods still face many challenges in practical applications. With the continuous development of artificial intelligence technology, the advantages of using machine learning methods to analyze and solve apparent resistivity are gradually emerging. In particular, the deep neural network (DNN) model has better representation capabilities for nonlinear functions than the traditional artificial neural network (ANN), and has high fault tolerance, robustness and self-organization. Summary of the invention

[0004] The purpose of the present invention is to provide a single-value transient electromagnetic apparent resistivity calculation method, system, device and storage medium based on deep learning.

[0005] The technical solution of the present invention is as follows:

[0006] A single-value transient electromagnetic apparent resistivity calculation method based on deep learning includes the following steps:

[0007] a) normalizing the measured induced voltage to obtain a normalized induced magnetic field;

[0008] b) constructing a deep neural network model including an input layer, multiple hidden layers and an output layer, wherein the input layer receives the normalized induced magnetic field, and the output layer outputs the transient field parameters;

[0009] c) using the Nadam optimization algorithm to train the deep neural network model, the training sample set includes multiple sets of measured induced voltages and their corresponding transient field parameters;

[0010] d) The measured induced voltage to be measured is input into the trained deep neural network model to obtain the output transient field parameters, and the apparent resistivity is calculated according to the theoretical formula of the transient electromagnetic field.

[0011] Furthermore, the deep neural network model in step b) has a 6-layer structure, including 1 input layer, 4 hidden layers and 1 output layer.

[0012] Furthermore, the theoretical formula of the transient electromagnetic field is specifically:

[0013]

[0014] Among them, a is the radius of the launch loop; u is the transient field parameter; erf(u) is the probability integral; μ0 is the magnetic permeability of the uniform half space; π is pi; I0 is the launch current amplitude; e is a natural constant.

[0015] Furthermore, the number of hidden layer neurons in the 6-layer deep neural network model in step b) is determined by error calculation.

[0016] Furthermore, the normalization process of the measured induced voltage is specifically as follows:

[0017] The measured induced voltage value is normalized to the interval [0, 1] using the maximum value method. The formula is:

[0018]

[0019] Among them, x i and are the values ​​before and after normalization, x max 、x min are the maximum and minimum values ​​before normalization.

[0020] Furthermore, the input of the input layer is transmitted through the activation function, processed by each hidden layer, and finally reaches the output layer.

[0021] The present invention also provides a system based on the above method, the system comprising:

[0022] A data normalization module is used to normalize the measured transient electromagnetic induction voltage data;

[0023] Deep neural network model building module, used to build and train deep neural network models;

[0024] A transient field parameter prediction module is used to predict transient field parameters using a trained deep neural network model;

[0025] The apparent resistivity calculation module is used to calculate the full-range apparent resistivity based on the predicted transient field parameters.

[0026] The present invention also provides an electronic device, comprising a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the above-mentioned single-value transient electromagnetic apparent resistivity calculation method based on deep learning.

[0027] The present invention also provides a computer-readable storage medium, characterized in that it stores a computer program, and when the computer program is executed by a processor, it implements the single-value transient electromagnetic apparent resistivity calculation method based on deep learning as described above.

[0028] Compared with the prior art, the present invention has the following advantages:

[0029] Compared with traditional methods, the present invention has stronger fitting ability for abstract data and can solve transient electromagnetic apparent resistivity more accurately.

[0030] The present invention proposes a single DNN model to solve the apparent resistivity, which avoids the emergence of multiple solutions in traditional methods and improves the calculation accuracy.

[0031] The present invention uses a trained DNN model to calculate the apparent resistivity, which does not require a large number of iterative calculations and improves the calculation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The accompanying drawings generally illustrate various embodiments by way of example and not limitation, and together with the description and claims, serve to illustrate the embodiments of the invention. Where appropriate, the same reference numerals are used throughout the drawings to refer to the same or similar parts. Such embodiments are illustrative and are not intended to be exhaustive or exclusive embodiments of the present apparatus or method.

[0033] Figure 1 The invention process schematic diagram of the present invention is shown;

[0034] Figure 2 The relationship diagram of the transient field parameter u obtained by the transient induced magnetic field of the present invention is shown;

[0035] Figure 3 A schematic diagram of a deep neural network model of the present invention is shown. DETAILED DESCRIPTION

[0036] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0037] Embodiment 1:

[0038] Data preprocessing: The measured induced voltage is normalized to obtain the normalized induced magnetic field F(u). The normalization process uses the maximum value method to normalize the value to the [0,1] interval.

[0039] Constructing DNN model: A 6-layer DNN structure is adopted, the input layer receives the normalized induced magnetic field F(u), and the output layer outputs the transient field parameter u. The number of neurons in the hidden layer is determined by formula (4), where n and m are the number of input units and the number of output units, respectively, and a is a constant between [1,10]. In this embodiment, n=1, m=1, a is 5, and the number of neurons in the hidden layer is calculated to be 7.

[0040] Deep learning is to find the corresponding rules between known data and answers through training samples. These rules can then be applied to new data and enable the computer to generate answers autonomously. The learning process consists of two processes: forward propagation of signals and back propagation of errors. Its structure generally includes three parts: input layer, hidden layer, and output layer. The core of deep learning lies in the existence of multiple hidden layers. The present invention generally uses a 6-layer DNN structure, as shown in Figure (3), and the hollow circles in each layer represent neurons.

[0041] The specific learning process is: during forward propagation, the sample is passed from the input layer, and the value of the input layer is transmitted by the activation function, processed by each hidden layer, and finally reaches the output layer. The Sigmoid activation function expression is:

[0042]

[0043] Where: p is the value of the previous layer of neurons, and f(p) is the value passed to the next layer of neurons. This paper chooses the Sigmoid activation function, that is, the values ​​passed to each other in the neural network are all passed according to formula (3). The last step of forward propagation is to calculate the error between the value passed to the output layer and the expected output value:

[0044]

[0045] Where: t k is the expected value, p k is the actual output value, k is the number of samples; for the DNN constructed by the present invention, t k is the value of the transient field parameter u in the training sample, p k It is the value of F(u) in the training sample transmitted to the output layer through the neurons in each layer of the neural network.

[0046] According to the solution formula of transient electromagnetic field, different physical parameters have different dimensions. Different physical parameters directly input into the network may become abnormally large after weighted accumulation, making it difficult for the network to converge. Therefore, it is necessary to normalize the input vector of the network. The input and output values ​​in the DNN of the present invention use the maximum value method:

[0047]

[0048] Normalize the value to the interval [0, 1], xi and are the values ​​before and after normalization, x max 、x min are the maximum and minimum values ​​before normalization. In order to enhance the generalization ability of DNN, the normalized samples are randomly sorted before training.

[0049] The number of neurons in the hidden layer also has a great impact on the constructed DNN. Choosing an appropriate number of neurons in the hidden layer can accelerate the convergence of the DNN.

[0050] In the present invention, a hidden layer, namely ANN model, is first used to determine, and then Determine the number of neurons in the hidden layer, where: n and m are the number of input units and output units respectively. In the constructed neural network model, the values ​​of n and m are both 1; a is a constant between [1, 10]; n1 is the number of neurons in the hidden layer, the initial value can be 3, and the number of neurons is gradually increased to 13, and its error value is calculated by formula (4). During the training process, small batch gradient descent can be used, that is, the samples are divided into 10 groups of data as a batch for training. The 10 groups of data jointly determine the gradient descent direction of this training to reduce the randomness in the training process. The training algorithm uses the commonly used Nadam algorithm, selects different numbers of neurons in the hidden layer by arithmetically, and uses 100 and 200 groups of samples for 5000 trainings.

[0051] Model training: The Nadam algorithm is used to train the DNN model. The training samples are 200,000 sets of measured induced voltages and their corresponding transient field parameters u. The emission current is 1A, and the emission wireframe radius a=300m. During the training process, the DNN model with an error of 5.5311×10^-7 is saved.

[0052] Apparent resistivity calculation: Input the measured induced voltage to be measured into the trained DNN model to obtain the transient field parameter u. Then calculate the overall apparent resistivity according to the transient electromagnetic field solution formula.

[0053]

[0054] Among them, a is the radius of the launch loop; u is the transient field parameter; erf(u) is the probability integral; μ0 is the magnetic permeability of the uniform half space; π is pi; I0 is the launch current amplitude; e is a natural constant.

[0055] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A single-value transient electromagnetic apparent resistivity calculation method based on deep learning, characterized in that: The following steps are involved: a) normalizing the measured induced voltage to obtain a normalized induced magnetic field; b) constructing a deep neural network model including an input layer, multiple hidden layers and an output layer, wherein the input layer receives the normalized induced magnetic field, and the output layer outputs the transient field parameters; c) using the Nadam optimization algorithm to train the deep neural network model, the training sample set includes multiple sets of measured induced voltages and their corresponding transient field parameters; d) The measured induced voltage to be measured is input into the trained deep neural network model to obtain the output transient field parameters, and the apparent resistivity is calculated according to the theoretical formula of the transient electromagnetic field.

2. The single-value transient electromagnetic apparent resistivity calculation method based on deep learning according to claim 1 is characterized in that: The deep neural network model in step b) has a 6-layer structure, including 1 input layer, 4 hidden layers and 1 output layer.

3. The single-value transient electromagnetic apparent resistivity calculation method based on deep learning according to claim 1 is characterized in that: The theoretical formula of the transient electromagnetic field is specifically: Among them, a is the radius of the launch loop; u is the transient field parameter; erf(u) is the probability integral; μ0 is the magnetic permeability of the uniform half space; π is pi; I0 is the launch current amplitude; e is a natural constant.

4. The single-value transient electromagnetic apparent resistivity calculation method based on deep learning according to claim 1 is characterized in that: In step b), the number of hidden layer neurons of the 6-layer deep neural network model is determined by error calculation.

5. The single-value transient electromagnetic apparent resistivity calculation method based on deep learning according to claim 1 is characterized in that: The normalization process of the measured induced voltage is specifically as follows: The measured induced voltage value is normalized to the interval [0, 1] using the maximum value method. The formula is: Among them, x i and are the values ​​before and after normalization, x max 、x min are the maximum and minimum values ​​before normalization.

6. The single-value transient electromagnetic apparent resistivity calculation method based on deep learning according to claim 1 is characterized in that: The input of the input layer is transmitted through the activation function, processed by each hidden layer, and finally reaches the output layer.

7. A system for implementing the method according to any one of claims 1 to 6, the system comprising: A data normalization module is used to normalize the measured transient electromagnetic induction voltage data; Deep neural network model building module, used to build and train deep neural network models; A transient field parameter prediction module is used to predict transient field parameters using a trained deep neural network model; The apparent resistivity calculation module is used to calculate the full-range apparent resistivity based on the predicted transient field parameters.

8. An electronic device, characterized in that: It comprises a memory and a processor, the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to perform the single-value transient electromagnetic apparent resistivity calculation method based on deep learning according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: It stores a computer program, which, when executed by a processor, implements the single-value transient electromagnetic apparent resistivity calculation method based on deep learning as described in any one of claims 1 to 6.

Images