Density profile inversion method, device and equipment for X-mode polarization microwave reflectometer, storage medium and program product

Abstract

The invention relates to a density profile inversion method, device and equipment for an X-mode polarization microwave reflectometer, a medium and a program product, and relates to the technical field of plasma diagnosis. By adopting the method, the efficiency and the precision of density profile inversion can be improved. The method comprises the following steps: acquiring a time-domain signal acquired by an X-mode polarized microwave reflectometer, inputting the time-domain signal into a deep neural network model, and standardizing the time-domain signal by a preprocessing module to obtain a standardized time-domain signal; an input end coding module carries out position information embedding and position coding on the standardized time domain signal to obtain a coded time domain signal; a feature extraction module performs feature extraction and feature enhancement on the coded time domain signal to obtain an enhanced target feature; the feature recognition module recognizes and positions the target features to obtain a recognition and positioning result; and the data output module generates density profile distribution data of the plasma in the X-mode polarized microwave reflectometer according to the identification and positioning result.

Description

Technical Field

[0001] The present application relates to the field of plasma diagnostic technology, and in particular to a density profile inversion method, apparatus, computer equipment, computer-readable storage medium and computer program product for an X-mode polarization microwave reflectometer. Background Art

[0002] Density profile microwave reflectometry is a diagnostic tool widely used in magnetic confinement fusion devices, providing key electron density profile data for fusion research. As a non-contact diagnostic technology, microwave reflectometry has good local measurement characteristics and excellent spatiotemporal resolution, and has a very broad application prospect in the field of fusion.

[0003] Microwave reflectometers are mainly divided into two polarization modes: O-mode and X-mode. For X-mode polarization microwave reflectometers, this traditional profile inversion method based on physical principles not only requires manual time-frequency analysis of the collected time domain data, but also requires a large amount of numerical calculations to invert the density distribution of the plasma, resulting in the problem of low efficiency of the traditional profile inversion method. Summary of the invention

[0004] Based on this, it is necessary to provide a density profile inversion method, device, computer equipment, computer-readable storage medium and computer program product for X-mode polarization microwave reflectometer to address the above technical problems.

[0005] In a first aspect, the present application provides a density profile inversion method for an X-mode polarization microwave reflectometer, comprising:

[0006] Acquire a time domain signal collected by an X-mode polarization microwave reflectometer, input the time domain signal into a trained deep neural network model, and perform standardization processing on the time domain signal by a preprocessing module in the deep neural network model to obtain a standardized time domain signal;

[0007] The input end encoding module in the deep neural network model performs position information embedding and position encoding on the standardized time domain signal to obtain an encoded time domain signal; the feature extraction module in the deep neural network model performs feature extraction and feature enhancement on the encoded time domain signal to obtain an enhanced target feature;

[0008] The feature recognition module in the deep neural network model recognizes and locates the target feature to obtain an identification and positioning result; the data output module in the deep neural network model generates density profile distribution data of the plasma in the X-mode polarization microwave reflectometer based on the identification and positioning result.

[0009] In one embodiment, the method further comprises:

[0010] The density profile inversion data set is split into a training set and a test set; the initial deep neural network model is trained using the training set and the weights of the initial deep neural network model are updated until the error value of the deep neural network model after the weight update on the test set meets a threshold condition, thereby obtaining the trained deep neural network model.

[0011] In one embodiment, the using the training set to train the initial deep neural network model and updating the weights of the initial deep neural network model includes:

[0012] The training set is input into the initial deep neural network model, and the initial deep neural network model performs autoregressive expansion according to the training set; when the initial deep neural network model satisfies any preset condition, the autoregressive process is terminated to obtain a model inversion result; according to the model inversion result, the weights of the initial deep neural network model are updated.

[0013] In one embodiment, before splitting the density profile inversion data set into a training set and a test set, the method further includes:

[0014] Acquire density profile experimental data and time domain experimental data; perform density interpolation on the density profile experimental data to obtain equally spaced radius-density data points, and perform dimension conversion on the time domain experimental data to obtain two-dimensional time domain experimental data; use the two-dimensional time domain experimental data as model input data during training, and use the radius-density data points as target output data during training, to obtain the density profile inversion data set.

[0015] In one embodiment, after performing density interpolation on the density profile experimental data to obtain equally spaced radius-density data points, the method further comprises:

[0016] The radius-density data points are tested for validity; when invalid data points are identified in the radius-density data points, the invalid data points are masked.

[0017] In one embodiment, the feature extraction module in the deep neural network model performs feature extraction and feature enhancement on the encoded time domain signal to obtain enhanced target features, including:

[0018] The multiple encoder layers in the feature extraction module respectively extract features from the encoded time domain signal through nonlinear mapping to obtain corresponding multiple features; the feature extraction module merges the multiple features to obtain the enhanced target feature.

[0019] In a second aspect, the present application also provides a density profile inversion device for an X-mode polarization microwave reflectometer, comprising:

[0020] A signal processing module, used to obtain a time domain signal collected by an X-mode polarization microwave reflectometer, input the time domain signal into a trained deep neural network model, and perform standardization processing on the time domain signal by a preprocessing module in the deep neural network model to obtain a standardized time domain signal;

[0021] A feature extraction module is used to perform position information embedding and position encoding on the standardized time domain signal by the input end encoding module in the deep neural network model to obtain an encoded time domain signal; and the feature extraction module in the deep neural network model performs feature extraction and feature enhancement on the encoded time domain signal to obtain an enhanced target feature;

[0022] The result acquisition module is used to identify and locate the target features by the feature recognition module in the deep neural network model to obtain the identification and positioning results; the data output module in the deep neural network model generates the density profile distribution data of the plasma in the X-mode polarization microwave reflectometer according to the identification and positioning results.

[0023] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0024] A time domain signal collected by an X-mode polarization microwave reflectometer is obtained, and the time domain signal is input into a trained deep neural network model. A preprocessing module in the deep neural network model performs standardization processing on the time domain signal to obtain a standardized time domain signal; an input end encoding module in the deep neural network model performs position information embedding and position encoding on the standardized time domain signal to obtain an encoded time domain signal; a feature extraction module in the deep neural network model performs feature extraction and feature enhancement on the encoded time domain signal to obtain an enhanced target feature; a feature recognition module in the deep neural network model recognizes and locates the target feature to obtain an identification and positioning result; a data output module in the deep neural network model generates density profile distribution data of the plasma in the X-mode polarization microwave reflectometer according to the identification and positioning result.

[0025] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:

[0026] A time domain signal collected by an X-mode polarization microwave reflectometer is obtained, and the time domain signal is input into a trained deep neural network model. A preprocessing module in the deep neural network model performs standardization processing on the time domain signal to obtain a standardized time domain signal; an input end encoding module in the deep neural network model performs position information embedding and position encoding on the standardized time domain signal to obtain an encoded time domain signal; a feature extraction module in the deep neural network model performs feature extraction and feature enhancement on the encoded time domain signal to obtain an enhanced target feature; a feature recognition module in the deep neural network model recognizes and locates the target feature to obtain an identification and positioning result; a data output module in the deep neural network model generates density profile distribution data of the plasma in the X-mode polarization microwave reflectometer according to the identification and positioning result.

[0027] In a fifth aspect, the present application further provides a computer program product, including a computer program, which implements the following steps when executed by a processor:

[0028] A time domain signal collected by an X-mode polarization microwave reflectometer is obtained, and the time domain signal is input into a trained deep neural network model. A preprocessing module in the deep neural network model performs standardization processing on the time domain signal to obtain a standardized time domain signal; an input end encoding module in the deep neural network model performs position information embedding and position encoding on the standardized time domain signal to obtain an encoded time domain signal; a feature extraction module in the deep neural network model performs feature extraction and feature enhancement on the encoded time domain signal to obtain an enhanced target feature; a feature recognition module in the deep neural network model recognizes and locates the target feature to obtain an identification and positioning result; a data output module in the deep neural network model generates density profile distribution data of the plasma in the X-mode polarization microwave reflectometer according to the identification and positioning result.

[0029] The density profile inversion method, device, computer equipment, computer-readable storage medium and computer program product for the X-mode polarization microwave reflectometer directly input the time domain signal collected by the X-mode polarization microwave reflectometer into the trained deep neural network model, and the deep neural network model sequentially performs standardization processing, position information embedding and position encoding, feature extraction and feature enhancement, and feature recognition and positioning on the time domain signal, so as to generate and output the density profile distribution data of the plasma in the X-mode polarization microwave reflectometer according to the recognition and positioning results, thereby avoiding the cumbersome steps of manually performing time-frequency analysis on the collected time domain data and inverting the density distribution of the plasma through a large amount of numerical calculations in the traditional profile inversion method, and can perform real-time and accurate density profile inversion of the X-mode polarization microwave reflectometer, thereby improving the efficiency and accuracy of density profile inversion and meeting the high-precision and full-automation requirements of real-time profile inversion. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0031] Figure 1 FIG. 1 is an application environment diagram of a density profile inversion method for an X-mode polarization microwave reflectometer in one embodiment;

[0032] Figure 2 A schematic flow chart of a density profile inversion method for an X-mode polarization microwave reflectometer in one embodiment;

[0033] Figure 3 is a schematic diagram of the structure of a trained deep neural network model in an embodiment;

[0034] Figure 4 is a schematic diagram of a density profile inversion result in one embodiment;

[0035] Figure 5 is a structural block diagram of a density profile inversion device for an X-mode polarization microwave reflectometer in one embodiment;

[0036] Figure 6 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0038] As a non-contact diagnostic technology, microwave reflectometer has good local measurement characteristics and excellent spatiotemporal resolution, and has a very broad application prospect in the field of fusion. Microwave reflectometers are mainly divided into two polarization modes: O-mode and X-mode. The cutoff frequency of O-mode polarization detection microwaves is only related to the plasma density, and its scope of application is greatly limited; X-mode polarization microwave reflectometer has higher radial resolution and is more suitable for measuring the core area of ​​low density gradients and the boundary area near the density zero point. It is widely used in magnetic confinement fusion devices in various countries around the world. Its basic principle is similar to radar technology. Electromagnetic waves are emitted into non-uniform plasma through the transmitting antenna, and are received by the receiving antenna after being reflected at the corresponding density cutoff layer. By using the cutoff characteristics of microwaves in plasma, the flight time of different frequency detection microwaves propagating in plasma can be measured to determine the position of the cutoff reflection layer and the plasma density distribution.

[0039] The construction of a real-time inversion system for density profile of microwave reflectometer based on deep neural network model is a key research direction in the future. The main disadvantage of the neural network model widely used in fusion devices at home and abroad is that the model uses a simple feedforward neural network with one hidden layer. The simple structural design is not suitable for the profile inversion of X-mode polarized microwave reflectometer, which is specifically reflected in:

[0040] First, the input data of the existing model is the flight time signal (i.e., beat frequency signal) of microwaves propagating in plasma. This means that before each profile inversion, the collected time domain data needs to be subjected to time-frequency analysis, and then the manually extracted beat frequency signal is used as the model input. This process is extremely time-consuming and prone to human errors, thereby reducing the timeliness and reliability of the inversion profile.

[0041] Second, the output data of the existing model is a one-dimensional vector containing plasma position information, while the cutoff frequency of X-mode polarization detection microwave is related to both the position and density of the cutoff layer. The cutoff density information cannot be directly inferred from the cutoff position. Therefore, this model cannot be applied to X-mode polarization profile reflectometers.

[0042] 3. The existing model is only applicable to real-time profile inversion using a single-band microwave reflectometer. However, for fusion devices, the density measurement range of the microwave reflectometer needs to be increased as much as possible, which means that a microwave reflectometer system combining multiple band subsystems is required, resulting in the input and output data of the model needing to simultaneously contain density profile information of multiple band subsystems.

[0043] The density profile inversion method for X-mode polarization microwave reflectometry provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown in FIG. 1 , the terminal communicates with the server through the network. The data storage system can store the data that the server needs to process. The data storage system can be integrated on the server or placed on the cloud or other network servers. Figure 1 In the application environment shown, the terminal can be, but is not limited to, various personal computers, notebook computers, smart phones, and tablet computers. The server can be implemented as an independent server or a server cluster consisting of multiple servers.

[0044] In one embodiment, Figure 2 As shown in FIG. 1 , a density profile inversion method for X-mode polarization microwave reflectometry is provided. The method is applied to Figure 1 The terminal in is used as an example to illustrate, including the following steps:

[0045] Step S201, obtain the time domain signal collected by the X-mode polarization microwave reflectometer, input the time domain signal into the trained deep neural network model, and use the preprocessing module in the deep neural network model to standardize the time domain signal to obtain a standardized time domain signal.

[0046] The time domain signal may be an in-phase and orthogonal I / Q signal, wherein the I component represents the real part of the signal and the Q component represents the imaginary part of the signal.

[0047] Among them, Figure 3 As shown, the deep neural network model is composed of a preprocessing module, an input encoding module, a feature extraction module, a feature recognition module and a data output module connected in sequence. Since the deep neural network model has a reasonable structural design, good feature extraction performance and can accurately recognize and extract features of multi-dimensional I / Q time domain data, it overcomes the deficiency that the existing profile inversion model can only process the beat frequency signal manually extracted after time-frequency analysis of the I / Q time domain data. The deep neural network model is particularly suitable for multi-period, high-dimensional I / Q time domain data generated simultaneously by multi-band subsystems.

[0048] Specifically, the preprocessing module is used to standardize the input time domain signal and then provide the standardized time domain signal to the next module. In this embodiment, the input data of the model are two orthogonal original time domain I / Q signals output by the X-mode polarization microwave reflectometer system, and the data dimension depends on the voltage scanning period and system sampling rate of the fast-sweep microwave reflectometer, wherein the system sampling rate is 62.5 MHz, and the voltage scanning period of the microwave source is 50 microseconds (including 40 microseconds of sweep time and 10 microseconds of pause time), which enables the microwave reflectometer to obtain a density profile every 50 microseconds. The number of detection frequency points within the 40 microsecond sweep time is: 40 (microseconds) × 62.5 (MHz) = 2500 (points), and each data point contains a real part and an imaginary part (I / Q signal). In addition, in order to reduce the influence of density turbulence and plasma disturbance on profile inversion, an average profile is calculated every 5 cycles. Therefore, the final total input data dimension is 2 (I / Q orthogonal signal) × 2500 (pieces) × 5 (average cycle) = 25000, and then the data is further processed into two-dimensional data, that is, the shape is (12500, 3); the target data is organized in the form of density points, and the radius R ranges from 1.95 to 2.37. By performing density interpolation on the points in this range, equally spaced radius-density data points are obtained. The number of data points is 115, and the data shape is (115, 2). If there are invalid numerical points (NaN values), they are masked in the model operation.

[0049] In step S202, the input-end encoding module in the deep neural network model embeds position information and performs position encoding on the standardized time domain signal to obtain an encoded time domain signal; and the feature extraction module in the deep neural network model performs feature extraction and feature enhancement on the encoded time domain signal to obtain enhanced target features.

[0050] Specifically, in the input encoding module of this embodiment, the real part and the imaginary part of the standardized time domain signal are first connected, so as to realize seamless information exchange between these components in the process of position encoding and embedding. The standardized time domain signal is first embedded with position information through the position encoder, and then encoded by a multilayer perceptron containing two hidden layers. The multilayer perceptron acting as an encoder contains two linear operation layers and a nonlinear operation layer using a GELU nonlinear activation function, so that it can not only retain the position information in the sequence, but also this encoding method enables the neural network model to more sensitively capture the sequential relationship in the data.

[0051] Step S203, the feature recognition module in the deep neural network model recognizes and locates the target features to obtain recognition and positioning results; the data output module in the deep neural network model generates density profile distribution data of the plasma in the X-mode polarization microwave reflectometer based on the recognition and positioning results.

[0052] in, Figure 4 The density profile inversion results of the X-mode polarized microwave reflectometer under different magnetic field parameters in the magnetic confinement fusion device, where the square points represent the density profile distribution results manually inverted by physical methods, and the dotted line represents the density profile distribution results predicted by the deep neural network model. It can be seen that under the condition of different discharge parameters of the magnetic confinement fusion device, the method in this embodiment can quickly and accurately invert the density profile distribution data of the plasma.

[0053] Specifically, the feature recognition module in this embodiment adopts an architecture design based on an attention decoder to identify and locate enhanced target features. The structure of the attention decoder is similar to that of the attention encoder, consisting of 6 decoder layers, and the number of heads of the multi-head attention (Self-Attention) mechanism in each layer is 8. In the self-attention layer, the query (Query), key (Key) and value (Value) are divided into 8 parts, each part performs attention calculation independently, and finally the results of these 8 parts are merged. Each layer includes a self-attention layer and a feed-forward neural network layer (Feed-Forward Network, FFN), which are connected by residual connection (Residual Connection) and layer normalization (Layer Normalization). The dimension of the feed-forward neural network is 1024, which is twice the input dimension of 512. After the data features are deeply processed by the attention encoder, they are passed to the decoder. This process also benefits from the self-attention mechanism, which ensures that the model can take into account the information of the entire sequence when generating prediction results. This processing flow of the attention model not only improves the accuracy of the prediction, but also enhances the interpretability of the model, allowing us to intuitively understand how the model allocates attention to different parts of the input sequence. The data output module in this embodiment is used to output the density profile distribution data of the plasma in the X-mode polarization microwave reflectometer. The data is analyzed by a multi-layer perceptron to obtain a prediction result.

[0054] In the density profile inversion method for the X-mode polarization microwave reflectometer, the time domain signal collected by the X-mode polarization microwave reflectometer is directly input into the trained deep neural network model, and the deep neural network model sequentially performs standardization processing, position information embedding and position encoding, feature extraction and feature enhancement, and feature recognition and positioning on the time domain signal, and then the density profile distribution data of the plasma in the X-mode polarization microwave reflectometer can be generated and output according to the recognition and positioning results, thereby avoiding the cumbersome steps of manually performing time-frequency analysis on the collected time domain data and inverting the density distribution of the plasma through a large amount of numerical calculations in the traditional profile inversion method, and can perform real-time and accurate density profile inversion of the X-mode polarization microwave reflectometer, thereby improving the efficiency and accuracy of the density profile inversion and meeting the high-precision and full-automation requirements of real-time profile inversion. In addition, the present application can obtain the density information and position information of the plasma by directly inputting the time domain signal collected by the X-mode polarization microwave reflectometer. The data output module of the model can efficiently and accurately output the density profile distribution data of the plasma, which solves the defect that the existing profile inversion method can only obtain the plasma position information by inputting the artificially extracted beat frequency signal, and overcomes the problems of the existing profile inversion method such as low timeliness, insufficient accuracy and only applicable to O-mode polarization microwave reflectometer.

[0055] In one embodiment, the method of the present application further includes the following steps:

[0056] The density profile inversion data set is split into a training set and a test set. The initial deep neural network model is trained using the training set and the weights of the initial deep neural network model are updated until the error value of the deep neural network model after the weight update on the test set meets the threshold condition, thereby obtaining a trained deep neural network model.

[0057] Among them, in the deep neural network model, weights are an important part of the model, which determines the connection strength between each neuron in the network. During the training process, the weights are continuously updated through optimization algorithms (such as gradient descent) to enable the network to better fit the data and complete the task.

[0058] Specifically, the terminal splits the density profile inversion data set into a training set and a test set; uses the training set to train the initial deep neural network model and uses the stochastic gradient descent method to update the weights of the initial deep neural network model until the error value of the deep neural network model after the weight update on the test set drops below 3.0%, thereby obtaining a trained deep neural network model.

[0059] The error in this embodiment is defined as:

[0060]

[0061] Among them, m is the number of training times, is the model prediction density, is the target density.

[0062] In one embodiment, in the above embodiment, using the training set to train the initial deep neural network model and updating the weights of the initial deep neural network model specifically includes the following steps:

[0063] The training set is input into the initial deep neural network model, and the initial deep neural network model performs autoregressive expansion according to the training set; when the initial deep neural network model meets any preset condition, the autoregressive process is terminated to obtain the model inversion result; according to the model inversion result, the weights of the initial deep neural network model are updated.

[0064] Specifically, the terminal inputs the training set into the deep neural network model, assuming the initial conditions R = 2.37 m, N e = 0×10 19 m -3 , the model is gradually expanded by autoregression, and the calculation formula is as follows:

[0065]

[0066] By going from O0 to O n The final density inversion profile is obtained by connecting the sequences in series. In the above formula, represents the I / Q time domain signal input to the X-mode polarization microwave reflectometer system, O n Indicates that step length n contains R and N e The autoregressive process terminates when one of the following conditions is met:

[0067] 1. R < 1.95 m;

[0068] 2. N e > N e,max In the embodiment, N e,max =5×10 19 m -3 is the maximum measurement density of the X-mode polarization microwave reflectometer system;

[0069] 3. The total inversion step exceeds the total number of data points, which is 115.

[0070] Then, according to the model inversion results, the stochastic gradient descent method is used to update the weights of the initial deep neural network model.

[0071] In one embodiment, before splitting the density profile inversion data set into a training set and a test set, the following steps are also included:

[0072] Obtain density profile experimental data and time domain experimental data; perform density interpolation on the density profile experimental data to obtain equally spaced radius-density data points, and perform dimension conversion on the time domain experimental data to obtain two-dimensional time domain experimental data; use the two-dimensional time domain experimental data as model input data during training, and use the radius-density data points as target output data during training to obtain a density profile inversion data set.

[0073] Specifically, a density profile inversion data set can be generated by the following method: first, the target output data of the density profile inversion data set in this embodiment consists of 30,000 plasma density profiles generated by more than 500 discharge experiments of the fusion device. The target output data is organized in the form of density points, and the radius R ranges from 1.95 to 2.37. By performing density interpolation on the points in this range, equally spaced radius-density data points are obtained. The number of data points is 115, and the data shape is (115, 2). If there are invalid numerical points (NaN values), mask processing is performed in the model operation; second, the model input data of the density profile inversion data set consists of two orthogonal time domain I / Q signals collected by an X-mode polarized microwave reflectometer, and its data dimension is two-dimensional data of (12500, 3).

[0074] In one embodiment, after density interpolation is performed on the density profile experimental data to obtain equally spaced radius-density data points, the following steps are also included:

[0075] The radius-density data points are tested for validity; when invalid data points are identified in the radius-density data points, the invalid data points are masked.

[0076] Specifically, the terminal performs a validity check on the radius-density data points to determine whether there are invalid data points in the radius-density data points; when invalid data points are identified in the radius-density data points, the invalid data points are masked to quickly and accurately remove the invalid data points in the radius-density data points.

[0077] In one embodiment, in the above step S302, the feature extraction module in the deep neural network model performs feature extraction and feature enhancement on the encoded time domain signal to obtain enhanced target features, which specifically includes the following steps:

[0078] The multiple encoder layers in the feature extraction module respectively extract features of the encoded time domain signal through nonlinear mapping to obtain corresponding multiple features; the feature extraction module merges the multiple features to obtain enhanced target features.

[0079] Specifically, the feature extraction module in this embodiment adopts an architecture design based on the attention encoder. The encoded time domain signal is sent to the attention encoder, and the attention encoder extracts features from the encoded time domain signal through nonlinear mapping. The attention encoder consists of 6 encoder layers, and the number of heads of the multi-head attention (Self-Attention) mechanism of each layer is 8. In the self-attention layer, the query (Query), key (Key) and value (Value) are divided into 8 parts, each part performs attention calculation independently, and finally the results of these 8 parts are merged. Each layer includes a self-attention layer and a feed-forward neural network layer (Feed-Forward Network, FFN), which are connected by residual connection (ResidualConnection) and layer normalization (Layer Normalization). The dimension of the feed-forward neural network is 1024, which is twice the input dimension of 512. The attention encoder can deeply analyze the long-range dependencies in the data, which is particularly important for predicting data such as plasma density profiles that may have long-distance correlations. The self-attention mechanism of the attention model allows the model to perform parallel calculations when processing sequence data, greatly improving training efficiency while avoiding the gradient vanishing problem common in traditional sequence models. This mechanism enables the model to flexibly adjust its structural parameters to adapt to tasks of different complexity, demonstrating strong generalization capabilities.

[0080] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0081] Based on the same inventive concept, the embodiment of the present application also provides a density profile inversion device for X-mode polarization microwave reflectometer for realizing the density profile inversion method for X-mode polarization microwave reflectometer mentioned above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific definition of one or more density profile inversion device embodiments for X-mode polarization microwave reflectometer provided below can refer to the definition of density profile inversion method for X-mode polarization microwave reflectometer above, and will not be repeated here.

[0082] In an exemplary embodiment, Figure 5 As shown, a density profile inversion device for an X-mode polarization microwave reflectometer is provided, comprising:

[0083] The signal processing module 501 is used to obtain a time domain signal collected by an X-mode polarization microwave reflectometer, input the time domain signal into a trained deep neural network model, and perform standardization processing on the time domain signal by a preprocessing module in the deep neural network model to obtain a standardized time domain signal;

[0084] The feature extraction module 502 is used to perform position information embedding and position encoding on the standardized time domain signal by the input end encoding module in the deep neural network model to obtain an encoded time domain signal; and perform feature extraction and feature enhancement on the encoded time domain signal by the feature extraction module in the deep neural network model to obtain an enhanced target feature;

[0085] The result acquisition module 503 is used to identify and locate the target features by the feature recognition module in the deep neural network model to obtain the identification and positioning results; the data output module in the deep neural network model generates the density profile distribution data of the plasma in the X-mode polarization microwave reflectometer according to the identification and positioning results.

[0086] In one embodiment, the density profile inversion device for an X-mode polarization microwave reflectometer also includes a model training module, which is used to split the density profile inversion data set into a training set and a test set; use the training set to train the initial deep neural network model and update the weights of the initial deep neural network model until the error value of the deep neural network model after the weight update on the test set meets a threshold condition, thereby obtaining the trained deep neural network model.

[0087] In one embodiment, the model training module is also used to input the training set into the initial deep neural network model, and the initial deep neural network model performs autoregressive expansion according to the training set; when the initial deep neural network model meets any preset condition, the autoregressive process is terminated to obtain a model inversion result; and according to the model inversion result, the weights of the initial deep neural network model are updated.

[0088] In one embodiment, the density profile inversion device for an X-mode polarization microwave reflectometer also includes a data acquisition module, which is used to acquire density profile experimental data and time domain experimental data; density interpolation is performed on the density profile experimental data to obtain equally spaced radius-density data points, and dimension conversion is performed on the time domain experimental data to obtain two-dimensional time domain experimental data; the two-dimensional time domain experimental data is used as model input data during training, and the radius-density data points are used as target output data during training to obtain the density profile inversion data set.

[0089] In one embodiment, the density profile inversion device for an X-mode polarization microwave reflectometer further includes an invalid detection module for performing validity detection on the radius-density data points; when invalid data points are identified to exist in the radius-density data points, masking is performed on the invalid data points.

[0090] In one embodiment, the feature extraction module 502 is also used to perform feature extraction on the encoded time domain signal through nonlinear mapping by multiple encoder layers in the feature extraction module to obtain corresponding multiple features; the feature extraction module merges the multiple features to obtain the enhanced target feature.

[0091] Each module in the density profile inversion device for X-mode polarization microwave reflectometer can be implemented in whole or in part by software, hardware and their combination. Each module can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to each module.

[0092] In an exemplary embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as shown in FIG. Figure 6As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, near field communication (NFC) or other technologies. When the computer program is executed by the processor, a density profile inversion method for an X-mode polarization microwave reflectometer is implemented. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0093] Those skilled in the art will understand that Figure 6 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0094] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

[0095] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0096] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0097] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0098] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0099] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0100] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A density profile inversion method for X-mode polarization microwave reflectometry, characterized in that: The method comprises: Acquire a time domain signal collected by an X-mode polarization microwave reflectometer, input the time domain signal into a trained deep neural network model, and perform standardization processing on the time domain signal by a preprocessing module in the deep neural network model to obtain a standardized time domain signal; The input end encoding module in the deep neural network model performs position information embedding and position encoding on the standardized time domain signal to obtain an encoded time domain signal; the feature extraction module in the deep neural network model performs feature extraction and feature enhancement on the encoded time domain signal to obtain an enhanced target feature; The feature recognition module in the deep neural network model recognizes and locates the target feature to obtain an identification and positioning result; the data output module in the deep neural network model generates density profile distribution data of the plasma in the X-mode polarization microwave reflectometer based on the identification and positioning result.

2. The method according to claim 1, characterized in that: The method further comprises: Split the density profile inversion dataset into training and testing sets; The training set is used to train the initial deep neural network model and update the weights of the initial deep neural network model until the error value of the deep neural network model after the weight update on the test set meets the threshold condition, thereby obtaining the trained deep neural network model.

3. The method according to claim 2, characterized in that The using the training set to train the initial deep neural network model and updating the weights of the initial deep neural network model includes: Inputting the training set into the initial deep neural network model, and allowing the initial deep neural network model to perform autoregressive expansion according to the training set; When the initial deep neural network model satisfies any preset condition, the autoregressive process is terminated to obtain a model inversion result; According to the model inversion result, the weights of the initial deep neural network model are updated.

4. The method according to claim 2, characterized in that: Before splitting the density profile inversion data set into a training set and a test set, the method further includes: Obtain density profile experimental data and time domain experimental data; Performing density interpolation on the density profile experimental data to obtain equally spaced radius-density data points, and performing dimension conversion on the time domain experimental data to obtain two-dimensional time domain experimental data; The two-dimensional time-domain experimental data is used as model input data during training, and the radius-density data points are used as target output data during training to obtain the density profile inversion data set.

5. The method according to claim 4, characterized in that After performing density interpolation on the density profile experimental data to obtain equally spaced radius-density data points, the method further includes: Performing validity detection on the radius-density data points; When invalid data points are identified in the radius-density data points, mask processing is performed on the invalid data points.

6. The method according to any one of claims 1 to 5, characterized in that: The feature extraction module in the deep neural network model performs feature extraction and feature enhancement on the encoded time domain signal to obtain enhanced target features, including: The plurality of encoder layers in the feature extraction module respectively extract features from the encoded time domain signal through nonlinear mapping to obtain a plurality of corresponding features; The feature extraction module combines a plurality of the features to obtain the enhanced target feature.

7. A density profile inversion device for X-mode polarization microwave reflectometer, characterized in that: The device comprises: A signal processing module, used to obtain a time domain signal collected by an X-mode polarization microwave reflectometer, input the time domain signal into a trained deep neural network model, and perform standardization processing on the time domain signal by a preprocessing module in the deep neural network model to obtain a standardized time domain signal; A feature extraction module is used to perform position information embedding and position encoding on the standardized time domain signal by the input end encoding module in the deep neural network model to obtain an encoded time domain signal; and the feature extraction module in the deep neural network model performs feature extraction and feature enhancement on the encoded time domain signal to obtain an enhanced target feature; The result acquisition module is used to identify and locate the target features by the feature recognition module in the deep neural network model to obtain the identification and positioning results; the data output module in the deep neural network model generates the density profile distribution data of the plasma in the X-mode polarization microwave reflectometer according to the identification and positioning results.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images