Multi-magnetic material parameter inversion method and system based on fusion neural network
Through the fusion neural network, combined with the convolutional neural network and the deep BP neural network, the problems of low inversion efficiency and low accuracy of traditional magnetic materials are solved, and fast and efficient multi-permeability prediction is achieved, with an accuracy of 96.3%.
Patent Information
- Application Number
- CN202111173625.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-10-08
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2041-10-08
AI Technical Summary
The traditional nonlinear inversion method of magnetic material parameters based on prior information and initial models is inefficient and low accuracy, making it difficult to accurately obtain the magnetic permeability of multi-magnetic materials.
Using a method based on a fusion neural network, the convolutional neural network and deep BP neural network are combined, and the one-dimensional convolutional layer and multiple hidden layers are used to classify and fine prediction of magnetic field data to obtain parameters of multi-magnetic materials.
Fast and efficient multi-permeability inversion is achieved. The simulation results show that within the error range of ±0.2H/m, the prediction accuracy can reach 96.3%, greatly improving the accuracy and efficiency of the traditional methods.
Smart Images

Figure CN113887137B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of deep learning and magnetic material technology, and specifically relates to a multi-magnetic material parameter inversion method and system based on a fusion neural network. Background Art
[0002] Underwater vehicles are often used to perform a variety of underwater tasks, such as scanning sunken ships, underwater communications, searching for underwater explosives, and performing mine detection and clearance. Underwater vehicles usually perform underwater operations by detecting the magnetic field of the target object. However, the influence of the magnetic field generated by the underwater vehicle itself on the detection cannot be ignored. Magnetic permeability is an important parameter that affects the magnetic field generated by magnetic objects. For underwater vehicles composed of multiple magnetic materials, the magnetic field generated is more complex. Accurately obtaining the magnetic permeability of each magnetic material is particularly important for studying the magnetic field generated by the underwater vehicle itself.
[0003] Magnetic permeability is a physical quantity that characterizes the magnetic properties of magnetic media. The method of determining magnetic permeability by the change of magnetic field around a magnetic object is called the inversion of magnetic permeability. Almost all inverse problems are nonlinear, and most of the methods for solving inverse problems are nonlinear inversion. Nonlinear inversion of magnetic permeability usually establishes the error functional of observed data and forward modeling under the condition of selecting a forward model, and uses nonlinear methods to iteratively minimize the error functional to achieve the purpose of inversion. Summary of the invention
[0004] In order to solve the problems of low efficiency and low accuracy of traditional nonlinear inversion methods of magnetic material parameters based on prior information and initial models, in a first aspect of the present invention, a multi-magnetic material parameter inversion method based on a fusion neural network is provided, comprising: obtaining multi-dimensional magnetic field components of multi-magnetic materials to be inverted under a static current-free background field; inputting the multi-dimensional magnetic field components into a trained convolutional neural network to obtain multiple category labels of the multi-dimensional magnetic field components; inputting each category label and multi-dimensional magnetic field component into multiple trained deep BP neural networks respectively to obtain parameter prediction values of the multi-magnetic materials to be inverted.
[0005] In some embodiments of the present invention, the convolutional neural network includes at least multiple one-dimensional convolutional layers, a maximum pooling layer and a fully connected layer, the multiple one-dimensional convolutional layers are used to receive multi-dimensional magnetic field components of multi-magnetic materials to be inverted or perform convolution operations on multi-dimensional magnetic field components; the fully connected layer is used to output category labels of the multi-dimensional magnetic field components.
[0006] Furthermore, the loss function of the convolutional neural network is a cross entropy loss function.
[0007] In some embodiments of the present invention, each of the deep BP neural networks includes a plurality of hidden layers and an output layer;
[0008] Each of the hidden layers is a fully connected layer, and its activation function is sigmoid or liner;
[0009] The activation function of the output layer is liner.
[0010] Furthermore, the loss function of each deep BP neural network is a mean square error function.
[0011] In the above-mentioned embodiment, the method of obtaining the multi-dimensional magnetic field components of the multi-magnetic material to be inverted under a static current-free background field includes: performing comsol modeling on the multi-magnetic material to be inverted; determining multiple magnetic field measurement lines of the multi-magnetic material to be inverted; and calculating the magnetic field components of each magnetic field measurement line using finite element and interpolation methods.
[0012] The second aspect of the present invention provides a multi-magnetic material parameter inversion system based on a fusion neural network, including: an acquisition module for acquiring multi-dimensional magnetic field components of the multi-magnetic material to be inverted under a static current-free background field; a first prediction module for inputting the multi-dimensional magnetic field components into a trained convolutional neural network to obtain multiple category labels of the multi-dimensional magnetic field components; a second prediction module for inputting each category label and multi-dimensional magnetic field component into multiple trained deep BP neural networks, respectively, to obtain parameter prediction values of the multi-magnetic material to be inverted.
[0013] Furthermore, the acquisition module includes a modeling unit, a determination unit, and a calculation unit. The modeling unit is used to perform comsol modeling on the multi-magnetic material to be inverted; the determination unit is used to determine multiple magnetic field measurement lines of the multi-magnetic material to be inverted; and the calculation unit is used to calculate the magnetic field component of each magnetic field measurement line using finite element and interpolation methods.
[0014] The third aspect of the present invention provides an electronic device, comprising: one or more processors; a storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the multi-magnetic material parameter inversion method based on a fusion neural network provided in the first aspect of the present invention.
[0015] A fourth aspect of the present invention provides a computer-readable medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the multi-magnetic material parameter inversion method based on a fusion neural network provided in the first aspect of the present invention.
[0016] The beneficial effects of the present invention are:
[0017] 1. The fusion neural network proposed in the present invention uses a one-dimensional convolutional network. The convolution kernel in the network is used to identify local features, which greatly reduces the number of parameters. At the same time, the classification function of the one-dimensional convolutional network can quickly determine the range of multiple parameters; the multi-layer BP deep neural network contains multiple hidden layers, which can accurately obtain the corresponding multiple parameters.
[0018] 2. The fusion neural network proposed in the present invention can quickly and efficiently invert the multi-permeability while reducing the number of neural network parameters. It can be seen from the simulation results that within the error range of ±0.2H / m, the prediction accuracy of the fusion neural network proposed in the present invention for the multi-permeability parameters of multi-magnetic materials can be as high as 96.3%. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] Figure 1 It is a basic flow chart of a multi-magnetic material parameter inversion method based on a fusion neural network in some embodiments of the present invention;
[0020] Figure 2 It is a schematic diagram of a specific process of a multi-magnetic material parameter inversion method based on a fusion neural network in some embodiments of the present invention;
[0021] Figure 3 is a schematic diagram of the structure of a convolutional neural network in some embodiments of the present invention;
[0022] Figure 4 It is a schematic diagram of the structure of a deep BP neural network in some embodiments of the present invention;
[0023] Figure 5 A schematic diagram of multi-magnetic material simulation modeling in some embodiments of the present invention;
[0024] Figure 6 Detailed dimension diagram for multi-magnetic material simulation modeling in some embodiments of the present invention;
[0025] Figure 7 A schematic diagram of selecting measurement lines of multiple magnetic materials in some embodiments of the present invention;
[0026] Figure 8 is a trend comparison diagram of the accuracy of the traditional BP neural network and the fusion neural network based on the fluctuation error in some embodiments of the present invention;
[0027] Fig. 9 It is a schematic structural diagram of a multi-magnetic material parameter inversion system based on a fusion neural network in some embodiments of the present invention;
[0028] Fig.10 It is a schematic diagram of the structure of an electronic device in some embodiments of the present invention. DETAILED DESCRIPTION
[0029] The principles and features of the present invention are described below in conjunction with the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.
[0030] refer to Figure 1 and Figure 2 In the first aspect of the present invention, a method for parameter inversion of multi-magnetic materials based on a fusion neural network is provided, comprising: S100. Obtaining multi-dimensional magnetic field components of the multi-magnetic materials to be inverted under a static current-free background field; S200. Inputting the multi-dimensional magnetic field components into a trained convolutional neural network to obtain multiple category labels of the multi-dimensional magnetic field components; S300. Inputting each category label and multi-dimensional magnetic field component into multiple trained deep BP neural networks respectively to obtain parameter prediction values of the multi-magnetic materials to be inverted.
[0031] It can be understood that the fusion neural network in the present invention is composed of a one-dimensional convolutional network and multiple deep BP neural networks. The one-dimensional convolutional network is used for coarse classification and roughly determines the range of multiple combination parameters; multiple deep BP neural networks are used for refined prediction. That is: 1. Multiple groups of multi-dimensional magnetic field data are used as the input of the fusion neural network, first pass through the one-dimensional convolutional network, and output the category label corresponding to each group of data (the range of multiple combination parameters); 2. The category labels correspond to the corresponding deep BP neural network, and then the multiple groups of multi-dimensional magnetic field data are input into the corresponding deep BP neural network for refined prediction, and the parameters of each magnetic material are output. Among them, the parameters of the magnetic material include but are not limited to initial magnetic permeability, maximum magnetic permeability, remanence, coercive force and saturation magnetic induction intensity, excitation magnetic field intensity, etc.
[0032] refer to Figure 3 In step S200 of some embodiments of the present invention, the convolutional neural network includes at least a plurality of one-dimensional convolutional layers, a maximum pooling layer and a fully connected layer, the plurality of one-dimensional convolutional layers are used to receive the multi-dimensional magnetic field components of the multi-magnetic materials to be inverted or to perform convolution operations on the multi-dimensional magnetic field components; the fully connected layer is used to output the category labels of the multi-dimensional magnetic field components.
[0033] Preferably, the convolutional neural network is a one-dimensional convolutional network, and its detailed parameters include:
[0034] 0. The one-dimensional convolutional network has four layers: two convolutional layers, one maximum pooling layer, and one fully connected layer.
[0035] 1. The one-dimensional convolution layer is used as the input layer. The dimension of the input magnetic field data is nx1, the number of one-dimensional convolution kernels is 16, and the size of the convolution kernel is 3x1;
[0036] 2. The second layer is a one-dimensional convolution layer with 16 one-dimensional convolution kernels, a convolution kernel size of 3x1, and an activation function of the hyperbolic tangent function tanh;
[0037] 3. The third layer is the maximum pooling layer, the number of pooling cores is 1, and the pooling core size is 3x1;
[0038] 4. The fourth layer is the fully connected output layer, with 9 neuron nodes, the activation function is softmax, and the output is 9 categories.
[0039] 5. The loss function of the one-dimensional convolutional network is the cross entropy loss function;
[0040] 6. The training algorithm for one-dimensional convolutional network is Adam;
[0041] 7. The training set accounts for 80% of the total data set, and the test set accounts for 20% of the total data;
[0042] 8. The number of training iterations epochs = 400.
[0043] refer to Figure 4 In some embodiments S300 of the present invention, each deep BP neural network includes multiple hidden layers and an output layer; each hidden layer is a fully connected layer, and its activation function is sigmoid or liner; the activation function of the output layer is liner.
[0044] Specifically, the detailed parameters of the deep BP neural network include:
[0045] 0. The deep BP neural network consists of 4 hidden layers and one output layer;
[0046] 1. The dimension of the input magnetic field data is nx1;
[0047] 2. The first layer is a fully connected layer with 16 neuron nodes and sigmoid activation function;
[0048] 3. The second layer is a fully connected layer with 8 neuron nodes and sigmoid activation function;
[0049] 4. The third layer is a fully connected layer with 4 neuron nodes and sigmoid activation function;
[0050] 5. The fourth layer is a fully connected layer with 2 neuron nodes and the activation function is liner;
[0051] 6. The fifth layer is the output layer, which outputs 3 parameters and the activation function is liner;
[0052] 7. The loss function of the deep BP neural network is MSE (mean square error);
[0053] 8. The training algorithm is Bayesian regularization algorithm;
[0054] 9. The training set accounts for 70% of the total data set, the test set accounts for 15% of the total data, and the validation set accounts for 15% of the total data;
[0055] 10. The number of training iterations epochs = 700.
[0056] refer to Figures 5 to 7 In the above embodiment, the method of obtaining the multi-dimensional magnetic field components of the multi-magnetic material to be inverted under a static current-free background field includes: performing comsol modeling on the multi-magnetic material to be inverted; determining multiple magnetic field measurement lines of the multi-magnetic material to be inverted; and calculating the magnetic field components of each magnetic field measurement line using finite element and interpolation methods.
[0057] Specifically, the above steps are explained by taking an underwater vehicle as a carrier for inverting multiple magnetic materials and magnetic permeability as a parameter of a magnetic material as an example:
[0058] S101. Under static and current-free background field, COMSO modeling of underwater vehicles is performed, such as Figure 5 As shown in the figure, the underwater vehicle is mainly composed of three parts: the cylindrical shell is the main body, which is 1.3 meters long; the two small cylinders inside represent the second and third parts respectively, the height of the small cylinder is 0.08 meters, the radius of the circle is 0.03 meters, and the detailed dimensions are as follows Figure 6 As shown;
[0059] S102, the three parts correspond to three magnetic materials respectively, the shell magnetic permeability u1: 200-400 (including: cast iron, nickel-zinc ferrite);
[0060] The two small cylindrical parts correspond to magnetic permeability u2: 1000-3000 (including: nickel-iron alloy, manganese-zinc ferrite); u3: 7000-9000 (including: silicon steel);
[0061] S103, when the reduced background fields x, y, and z are all 23 A / m, 5 measuring lines with a length of 1.55 meters are selected at 0.2 meters below the underwater vehicle, and the corresponding magnetic field y component values are obtained by using the finite element method every 0.05 meters on each measuring line, and finally the interpolation method is used to obtain all the magnetic field y component values of the 5 measuring lines: By(160x1); the measuring lines are selected as follows: Figure 7 shown.
[0062] It can be understood that the above process can be used for prediction, and can also be used to obtain the multi-dimensional magnetic field components of the multi-magnetic material to be inverted under the static current-free background field, and can also be used for the samples required for the fusion neural network training. Therefore, the above steps also include:
[0063] S104, u1 is divided into 200, 300, 400, u2 is divided into 1000, 2000, 3000, u3 is divided into 7000, 8000, 9000, then u1, u2, u3 are combined into nine categories in the following way: 200 1000 [7000 8000 9000], 200 2000 [7000 8000 9000], 200 3000 [7000 8000 9000], 300 1000 [7000 8000 9000], 300 2000 [7000 8000 9000], 300 3000 [70008000 9000], 400 1000 [7000 8000 9000], 400 2000[7000 8000 9000], 400 3000[7000 8000 9000].
[0064] S105. After the 9 combinations are determined, each data set is generated with intervals of u1-16: 0.25: u1+20, u2-160: 2.5: u2+200, and u3-160: 2.5: u3+200. At this time, each class has a total of 435 sets of u1, u2, and u3 magnetic permeability combinations, and the 9 classes have a total of 9x435 sets of data sets.
[0065] Different combinations of S106, u1, u2, and u3 can produce corresponding different magnetic field components By, so there are 435 groups of By in each category, and there are 9x435 groups of By in the nine categories.
[0066] S107. After By is normalized, it is used as the input of the one-dimensional convolutional neural network in the fusion neural network for training, and the category label is also one-hot encoded as the output.
[0067] S108. The above is the source of data for training convolutional networks in fusion neural networks.
[0068] Further, based on the above samples, taking 9 deep BP neural networks as an example, the 9 deep BP neural networks correspond to the deep BP neural networks under the categories in S104, but at this time, the magnetic permeability under each category is divided into intervals of: u1-50: 0.5: u1+50, u2-500: 5: u2+500, 6500: 15: 9500. At this time, the data set of each deep BP neural network is 201 groups, and the 9 deep BP neural networks have a total of 9x201 groups;
[0069] Different combinations of S1010, u1, u2, and u3 can produce corresponding different magnetic field components By, so each deep BP neural network will have 201 groups of By, and nine BPs have a total of 9x435 groups of By;
[0070] S1011, normalize By in each deep BP neural network and use it as the input of the deep BP neural network in the fusion neural network, and use the corresponding values of u1, u2, and u3 as outputs to train each deep BP neural network;
[0071] S1012. The above is the source of data for training deep BP neural network in fusion neural network (training sample set);
[0072] After S1013, one-dimensional convolutional network and 9 deep BP training are completed, the final fusion neural network is obtained.
[0073] refer to Figure 8 , the above fusion neural network is evaluated by the validation data set:
[0074] S1015. For each type of data in step S104, generate magnetic permeability combinations at intervals of u1-50: 0.1: u1+50, u2-500: 1: u2+500, 6500: 3: 9500. There are 1001 types in each group. 101-901 of each group are intercepted as verification data. Therefore, after the 9 groups are merged, the verification data set has a total of 801x9=7209 groups, wherein the magnetic permeability range corresponding to the intercepted data is: u1: 160-440; u2: 600-3400; u3: 6800-9200; and includes the material magnetic permeability parameter range specified in step S102;
[0075] S1016, there are also 7209 groups of By corresponding to 7209 groups of magnetic permeabilities. The normalized By is used as the input of the fusion neural network, and the output is the corresponding predicted value of magnetic permeability u1, u2, u3. The predicted value is subtracted from the standard validation set in step S1015 for error analysis, so as to evaluate the fusion neural network.
[0076] S1017, when the absolute fluctuation error is 0.2, the accuracy of the fusion neural network is as high as 96.3%;
[0077] S1018. When the absolute fluctuation error has different values, the accuracy comparison between the fusion neural network and the traditional BP neural network is shown in Figure 8.
[0078] It can be understood that although the above-mentioned changes and measurements of magnetic permeability are based on the y-axis component By in the diagram, they cannot be used as an absolute limitation of the magnetic field data in the present invention, that is: when the reference coordinate system changes, the magnetic field components of other axes can be regarded as equivalent to the y-axis component By in the present invention through relevant coordinate transformations or physical laws (Ampere's law, Biot-Savart law, Lorentz law, etc.).
[0079] Example 2
[0080] refer to Fig. 9 According to the second aspect of the present invention, a multi-magnetic material parameter inversion system 1 based on a fusion neural network is provided, comprising: an acquisition module 11, used to obtain the multi-dimensional magnetic field components of the multi-magnetic material to be inverted under a static current-free background field; a first prediction module 12, used to input the multi-dimensional magnetic field components into a trained convolutional neural network to obtain multiple category labels of the multi-dimensional magnetic field components; a second prediction module 13, used to input each category label and multi-dimensional magnetic field component into multiple trained deep BP neural networks respectively to obtain parameter prediction values of the multi-magnetic material to be inverted.
[0081] Furthermore, the acquisition module 11 includes a modeling unit, a determination unit, and a calculation unit. The modeling unit is used to perform comsol modeling on the multi-magnetic material to be inverted; the determination unit is used to determine multiple magnetic field measurement lines of the multi-magnetic material to be inverted; and the calculation unit is used to calculate the magnetic field components of each magnetic field measurement line using finite element and interpolation methods.
[0082] Example 3
[0083] refer to Fig.10 According to the third aspect of the present invention, an electronic device is provided, comprising: one or more processors; a storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the method provided by the second aspect of the present invention.
[0084] The electronic device 500 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 501, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 502 or a program loaded from a storage device 508 into a random access memory (RAM) 503. In the RAM 503, various programs and data required for the operation of the electronic device 500 are also stored. The processing device 501, the ROM 502, and the RAM 503 are connected to each other via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.
[0085] Typically, the following devices may be connected to the I / O interface 505: an input device 506 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 507 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 508 including, for example, a hard disk, etc.; and a communication device 509. The communication device 509 may allow the electronic device 500 to communicate with other devices wirelessly or by wire to exchange data. Although Fig.10The electronic device 500 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead. Fig.10 Each block shown in the figure may represent one device, or may represent multiple devices as required.
[0086] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through a communication device 509, or installed from a storage device 508, or installed from a ROM 502. When the computer program is executed by the processing device 501, the above functions defined in the method of the embodiment of the present disclosure are executed. It should be noted that the computer-readable medium described in the embodiment of the present disclosure can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection with one or more conductors, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In an embodiment of the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In an embodiment of the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to: wire, optical cable, RF (radio frequency), etc., or any suitable combination of the foregoing.
[0087] The computer-readable medium may be included in the electronic device, or may exist independently without being installed in the electronic device. The computer-readable medium carries one or more computer programs. When the one or more programs are executed by the electronic device, the electronic device:
[0088] Computer program code for performing the operations of embodiments of the present disclosure may be written in one or more programming languages or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, Python, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on a user's computer, partially on a user's computer, as a separate software package, partially on a user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).
[0089] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.
[0090] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims
1. A multi-magnetic material parameter inversion method based on a fusion neural network, characterized in that: include: Obtaining multi-dimensional magnetic field components of the multi-magnetic material to be inverted under a static current-free background field: performing comsol modeling on the multi-magnetic material to be inverted; Determining a plurality of magnetic field measurement lines of multiple magnetic materials to be inverted; Calculate the magnetic field components of each magnetic field measurement line using finite element and interpolation methods; The multidimensional magnetic field component is input into the trained convolutional neural network to obtain multiple category labels of the multidimensional magnetic field component; the convolutional neural network includes at least multiple one-dimensional convolutional layers, a maximum pooling layer and a fully connected layer, the multiple one-dimensional convolutional layers are used to receive the multidimensional magnetic field components of the multi-magnetic material to be inverted or to perform convolution operations on the multidimensional magnetic field components; the fully connected layer is used to output the category labels of the multidimensional magnetic field components; the loss function of the convolutional neural network is a cross entropy loss function; Each category label and multi-dimensional magnetic field component are respectively input into multiple trained deep BP neural networks to obtain parameter prediction values of multi-magnetic materials to be inverted; each of the deep BP neural networks includes multiple hidden layers and an output layer; each of the hidden layers is a fully connected layer, and its activation function is sigmoid or liner; the activation function of the output layer is liner; the loss function of each deep BP neural network is a mean square error function.
2. A multi-magnetic material parameter inversion system based on a fusion neural network, characterized in that: include: An acquisition module is used to acquire multi-dimensional magnetic field components of the multi-magnetic material to be inverted under a static current-free background field: Comsol modeling is performed on the multi-magnetic material to be inverted; Determining a plurality of magnetic field measurement lines of multiple magnetic materials to be inverted; Calculate the magnetic field components of each magnetic field measurement line using finite element and interpolation methods; The first prediction module is used to input the multidimensional magnetic field component into the trained convolutional neural network to obtain multiple category labels of the multidimensional magnetic field component; the convolutional neural network includes at least multiple one-dimensional convolutional layers, a maximum pooling layer and a fully connected layer, the multiple one-dimensional convolutional layers are used to receive the multidimensional magnetic field components of the multi-magnetic material to be inverted or perform convolution operations on the multidimensional magnetic field components; the fully connected layer is used to output the category labels of the multidimensional magnetic field components; the loss function of the convolutional neural network is a cross entropy loss function; The second prediction module is used to input each category label and multi-dimensional magnetic field component into multiple trained deep BP neural networks respectively to obtain parameter prediction values of multi-magnetic materials to be inverted; each of the deep BP neural networks includes multiple hidden layers and an output layer; each of the hidden layers is a fully connected layer, and its activation function is sigmoid or liner; the activation function of the output layer is liner; the loss function of each deep BP neural network is a mean square error function.
3. An electronic device, comprising: one or more processors; A storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the multi-magnetic material parameter inversion method based on a fusion neural network as described in claim 1.
4. A computer readable medium having a computer program stored thereon, wherein: When the computer program is executed by a processor, the multi-magnetic material parameter inversion method based on a fusion neural network as described in claim 1 is implemented.
Citation Information
Patent Citations
Magnetotelluric nonlinear inversion method based on convolutional neural network
CN111812732A
Target classification method, based on sound wave propagation equation, for two-way coupling deep learning
WO2020258615A1
