Plasma density profile inversion method and device, computer equipment, readable storage medium and program product

Abstract

The application relates to a plasma density profile inversion method, device, equipment, storage medium and program product, and relates to the technical field of plasma diagnosis. The application can improve the efficiency and accuracy of density profile inversion. The method comprises the following steps: acquiring a microwave signal collected by a microwave reflectometer, extracting a medium-frequency signal from the microwave signal, and inputting the medium-frequency signal into an I / Q detector to obtain an I / Q signal through modulation; inputting the I / Q signal into a profile neural network model, performing signal enhancement processing on the I / Q signal by an input coding layer in the profile neural network model to obtain an enhanced target I / Q signal; performing feature extraction and feature enhancement on the target I / Q signal by a backbone network in the profile neural network model through a linear rectification module to obtain reinforced multi-scale features; and performing profile prediction according to the reinforced multi-scale features by an output decoding layer in the profile neural network model to obtain density profile distribution data of plasma in the microwave reflectometer.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of plasma diagnosis, and in particular to a plasma density profile inversion method and device, a computer device, a computer readable storage medium and a computer program product. BACKGROUND

[0002] Microwave reflectometry is an important non-perturbative diagnostic tool, which is widely used in electron density measurement of magnetic confinement fusion devices, and provides key electron density profile data for fusion research. The basic principle of microwave reflectometry is to inject microwave into the plasma, and receive the microwave reflected from the cutoff layer with zero refractive index, and then obtain the electron density information of the cutoff layer from the reflected wave.

[0003] In the traditional technology, the mode of "first acquisition and storage, then offline processing" is generally adopted, and there is a lack of means for real-time density inversion and feedback control by microwave reflectometry. For the reconstruction of the plasma density distribution, the famous inversion methods include the Abel inversion method, the separation of variables method and the slice method using singular value decomposition. However, these traditional density inversion methods generally have problems of complex calculation, high processing delay and limited large-scale computing capacity, and the inversion process usually takes hundreds of milliseconds, which cannot meet the demand of real-time control of future fusion devices. SUMMARY

[0004] Therefore, it is necessary to provide a plasma density profile inversion method, device, computer device, computer readable storage medium and computer program product to solve the above technical problems.

[0005] In a first aspect, the present application provides a plasma density profile inversion method, comprising:

[0006] acquiring a microwave signal collected by a microwave reflectometer, extracting an intermediate frequency signal from the microwave signal, and obtaining an I / Q signal by demodulating the intermediate frequency signal through an I / Q detector;

[0007] inputting the I / Q signal into a profile neural network model, performing signal enhancement processing on the I / Q signal by an input encoding layer in the profile neural network model, and obtaining an enhanced target I / Q signal;

[0008] performing feature extraction and feature enhancement on the target I / Q signal by a backbone network in the profile neural network model through a linear rectification module, and obtaining a reinforced multi-scale feature;

[0009] performing profile prediction according to the reinforced multi-scale feature by an output decoding layer in the profile neural network model, and obtaining density profile distribution data of the plasma in the microwave reflectometer.

[0010] In one of the embodiments, the backbone network in the profile neural network model performs feature extraction and feature enhancement on the target I / Q signal through a linear rectifier module to obtain reinforced multi-scale features, including:

[0011] The linear rectifier module includes a plurality of parallel convolution branches, each of which extracts features of the target I / Q signal through a convolution kernel of different length to obtain a plurality of features of different scales. The linear rectifier module fuses and aggregates the features of different scales and the original features of the target I / Q signal to obtain merged multi-scale features. The linear rectifier module enhances the multi-scale features through a nonlinear activation function to obtain the reinforced multi-scale features.

[0012] In one of the embodiments, the method further includes:

[0013] Extracting an original time-domain signal from the microwave signal, and recording the original time-domain signal in a target information unit in the number matching data storage system for local storage.

[0014] In one of the embodiments, the training of the initial network model and the updating of the model parameters of the initial network model using the training set and the validation set through a deep learning algorithm include:

[0015] Obtaining historical density profile data and extracting the original time-domain signal from the data storage system, and obtaining a training set, a validation set, and a test set according to the historical density profile data and the original time-domain signal. According to the preset initial hyperparameters, an initial network model to be trained is constructed. The initial network model is trained and the model parameters of the initial network model are updated using the training set and the validation set through a deep learning algorithm until the performance indicators of the updated network model on the test set meet the threshold condition, and the profile neural network model is obtained.

[0016] In one of the embodiments, the training of the initial network model and the updating of the model parameters of the initial network model using the training set and the validation set through a deep learning algorithm include:

[0017] The training set is input into the initial network model, and the initial network model performs offline model training according to the training set. If the output result of the initial network model meets all the verification parameters in the validation set, the model training process is terminated and a model inversion result is obtained. Otherwise, the model parameters of the initial network model are adjusted through a self-adaptive optimization algorithm according to the model inversion result.

[0018] In one of the embodiments, the output decoding layer in the profile neural network model performs profile prediction based on the reinforced multi-scale features to obtain the density profile distribution data of the plasma in the microwave reflectometer, including:

[0019] The output decoding layer uses multi-scale convolution to capture the reinforced multi-scale features at different spatial resolutions and uses residual connection to alleviate gradient disappearance in the profile neural network model to obtain a feature capture result; and performs profile prediction based on the feature capture result to obtain the density profile distribution data of the plasma.

[0020] In a second aspect, the present application further provides a plasma density profile inversion device, including:

[0021] A signal extraction module is configured to acquire a microwave signal collected by a microwave reflectometer, extract an intermediate frequency signal from the microwave signal, and demodulate the intermediate frequency signal by an I / Q detector to obtain an I / Q signal.

[0022] A signal enhancement module is configured to input the I / Q signal into a profile neural network model, perform signal enhancement processing on the I / Q signal by an input encoding layer in the profile neural network model, and obtain an enhanced target I / Q signal.

[0023] A feature extraction module is configured to perform feature extraction and feature enhancement on the target I / Q signal by a linear rectifier module in a backbone network of the profile neural network model to obtain reinforced multi-scale features.

[0024] A profile prediction module is configured to perform profile prediction based on the reinforced multi-scale features by an output decoding layer in the profile neural network model to obtain the density profile distribution data of the plasma in the microwave reflectometer.

[0025] In a third aspect, the present application further provides a computer device including a memory and a processor, the memory storing a computer program, and the processor implementing the following steps when executing the computer program:

[0026] The microwave signal collected by the microwave reflectometer is acquired, an intermediate frequency signal is extracted from the microwave signal, and the intermediate frequency signal is demodulated by an I / Q detector to obtain an I / Q signal; the I / Q signal is input into a profile neural network model, the I / Q signal is subjected to signal enhancement processing by an input encoding layer in the profile neural network model, and an enhanced target I / Q signal is obtained; a backbone network in the profile neural network model passes through a linear rectification module to perform feature extraction and feature enhancement on the target I / Q signal, and a reinforced multi-scale feature is obtained; and a profile prediction is performed according to the reinforced multi-scale feature by an output decoding layer in the profile neural network model, and density profile distribution data of the plasma in the microwave reflectometer is obtained.

[0027] In a fourth aspect, the present application also provides a computer readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the following steps:

[0028] The microwave signal collected by the microwave reflectometer is acquired, an intermediate frequency signal is extracted from the microwave signal, and the intermediate frequency signal is demodulated by an I / Q detector to obtain an I / Q signal; the I / Q signal is input into a profile neural network model, the I / Q signal is subjected to signal enhancement processing by an input encoding layer in the profile neural network model, and an enhanced target I / Q signal is obtained; a backbone network in the profile neural network model passes through a linear rectification module to perform feature extraction and feature enhancement on the target I / Q signal, and a reinforced multi-scale feature is obtained; and a profile prediction is performed according to the reinforced multi-scale feature by an output decoding layer in the profile neural network model, and density profile distribution data of the plasma in the microwave reflectometer is obtained.

[0029] In a fifth aspect, the present application also provides a computer program product comprising a computer program, wherein the computer program, when executed by a processor, implements the following steps:

[0030] The microwave signal collected by the microwave reflectometer is acquired, an intermediate frequency signal is extracted from the microwave signal, and the intermediate frequency signal is demodulated by an I / Q detector to obtain an I / Q signal; the I / Q signal is input into a profile neural network model, the I / Q signal is subjected to signal enhancement processing by an input encoding layer in the profile neural network model, and an enhanced target I / Q signal is obtained; a backbone network in the profile neural network model passes through a linear rectification module to perform feature extraction and feature enhancement on the target I / Q signal, and a reinforced multi-scale feature is obtained; and a profile prediction is performed according to the reinforced multi-scale feature by an output decoding layer in the profile neural network model, and density profile distribution data of the plasma in the microwave reflectometer is obtained.

[0031] The plasma density profile inversion method, device, computer device, computer readable storage medium and computer program product can obtain the I / Q signal by extracting and mediating the microwave signal collected by the microwave reflectometer, input the I / Q signal into the profile neural network model, sequentially perform signal enhancement, feature extraction and feature enhancement on the I / Q signal by the profile neural network model, and perform profile prediction according to the reinforced multi-scale features, so that the density profile distribution data of the plasma in the microwave reflectometer can be obtained in real time, thereby avoiding the mode of “first collection and storage, then offline processing” in the traditional profile inversion method and omitting the cumbersome steps of inverting the plasma density distribution through a large number of complex numerical calculations. The microwave reflectometer can be accurately and in real time density profile inversion, the efficiency and accuracy of the density profile inversion are improved, and the real-time control requirement of the plasma density profile inversion of the future fusion device can be met. In addition, the end-to-end profile neural network model designed according to the characteristics of the microwave reflectometer is used, the model can learn the mapping relationship of the density profile distribution directly from the core original data (I / Q signal) of the microwave reflectometer, does not need to rely on other diagnostic intermediate results or assumptions, and provides an efficient technical path for the real-time application of the microwave reflectometer diagnosis. BRIEF DESCRIPTION OF DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the drawings needed to be used in the description of the embodiments of the present application or the related art will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other related drawings can be obtained by those skilled in the art without creative labor.

[0033] Figure 1 The application environment diagram of the plasma density profile inversion method in an embodiment;

[0034] Figure 2 The flowchart of the plasma density profile inversion method in an embodiment;

[0035] Figure 3 The structural diagram of the linear rectification module in an embodiment;

[0036] Figure 4 The flowchart of the feature extraction and feature enhancement step in an embodiment;

[0037] Figure 5 The flowchart of the plasma density profile inversion method in a specific embodiment;

[0038] Figure 6 The flowchart of the plasma density profile inversion method in an application embodiment;

[0039] Figure 7 Figure 1 shows a schematic diagram of a data dimensionality reduction module according to an application embodiment;

[0040] Figure 8 Figure 2 shows a block diagram of a plasma density profile inversion device according to an application embodiment;

[0041] Figure 9 Figure 3 shows an internal structure diagram of a computer device according to an application embodiment. DETAILED DESCRIPTION

[0042] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and should not be used to limit the present application.

[0043] With the rapid development of artificial intelligence, more and more neural network models are used in fusion research. Future fusion devices need intelligent control systems to maintain optimal performance, and these control systems rely on accurate real-time monitoring of the plasma. For experimental equipment used in nuclear fusion research (such as a tokamak device), whether it is the boundary density in the low-confinement mode or the edge-local-mode and density pedestal structure in the high-confinement mode, all of them play a crucial role in plasma confinement, so fast and accurate calculation and identification of plasma boundary parameters and modes are one of the key factors for future fusion parameter control. The data-driven electron density inversion method establishes a complex nonlinear relationship between the original input signal and the physical parameter, so that the density change can be directly reflected according to the original signal, omitting the time-consuming calculation step in the middle, reaching the level of real-time processing, and being applied to the feedback control of future fusion devices.

[0044] The plasma density profile inversion method provided by the embodiments of the present application can be applied in the application environment as shown in Figure 1 . Among them, the terminal can communicate with the server through the network. The data storage system can store the data required to be processed by the server. The data storage system can be integrated on the server, or placed on the cloud or other network servers. In the application environment as shown in Figure 1 , the terminal can be but is not limited to various personal computers, notebook computers, smart phones and tablet computers. The server can be implemented by an independent server or a server cluster composed of multiple servers.

[0045] In one embodiment, as shown in Figure 2 , a plasma density profile inversion method is provided, which can be applied to the terminal in Figure 1 . The method can include the following steps:

[0046] Step S201: Acquire the microwave signal collected by the microwave reflectometer, extract the intermediate frequency signal from the microwave signal, and adjust the intermediate frequency signal through the I / Q detector to obtain the I / Q signal.

[0047] Intermediate frequency (IF) signals are typically obtained by frequency conversion of high-frequency signals (such as microwave signals), which simplifies the processes of signal amplification, filtering, and demodulation, thereby improving the performance and efficiency of the system.

[0048] The I / Q signal is a combination of an in-phase signal (I) and a quadrature signal (Q). The I signal is the in-phase component, representing the real part of the signal or the phase component of the original signal; the Q signal is the quadrature component, representing the imaginary part of the signal, and it is 90 degrees out of phase with the I signal.

[0049] Specifically, in response to the received plasma density profile inversion command, the terminal acquires the microwave signal collected by the microwave reflector, extracts a portion of the signal from the microwave signal and mixes it with the local oscillator signal to obtain the intermediate frequency signal, and then modulates the intermediate frequency signal through I / Q detectors (in-phase and quadrature detectors) to obtain the I / Q signal.

[0050] Step S202: Input the I / Q signal into the profile neural network model, and perform signal enhancement processing on the I / Q signal by the input encoding layer in the profile neural network model to obtain the enhanced target I / Q signal.

[0051] The input coding layer increases the channel dimension of the data, enhances its representational power, and prepares for effective feature extraction.

[0052] Specifically, the terminal inputs the I / Q signal into the profile neural network model, and performs signal enhancement processing on the I / Q signal through the input coding layer in the profile neural network model to obtain the enhanced target I / Q signal.

[0053] Step S203: The backbone network in the profile neural network model extracts and enhances the target I / Q signal through a linear rectification module to obtain enhanced multi-scale features.

[0054] Among them, the linear rectifier module (i.e., the Inception module) is as follows: Figure 3 As shown, its core idea is to fuse features at different scales through a multi-branch parallel structure, while introducing sparse connections to reduce computational costs. Specifically, each linear rectified module includes four branches. After the initial 1×1 convolution, three branches apply 3×1, 5×1, and 7×1 convolutions to extract multi-scale features. The outputs are then concatenated, fused through another 1×1 convolution, and connected to the input through residual connections to enhance features and preserve gradient flow.

[0055] Specifically, the terminal extracts features of the target I / Q signal by the backbone network in the profile neural network model through the linear rectifier module to obtain multi-scale features, and then enhances the multi-scale features to obtain the enhanced multi-scale features.

[0056] In step S204, the output decoding layer in the profile neural network model performs profile prediction according to the enhanced multi-scale features to obtain the density profile distribution data of the plasma in the microwave reflectometer.

[0057] Specifically, the terminal performs profile prediction according to the enhanced multi-scale features by the output decoding layer in the profile neural network model to generate the predicted density profile distribution data of the plasma in the microwave reflectometer.

[0058] In this embodiment, the I / Q signal is obtained by extracting and adjusting the microwave signal collected by the microwave reflectometer, and then the I / Q signal is input into the profile neural network model. The profile neural network model sequentially performs signal enhancement, feature extraction and feature enhancement on the I / Q signal, and performs profile prediction according to the enhanced multi-scale features, so that the density profile distribution data of the plasma in the microwave reflectometer can be obtained in real time. Thus, the mode of "first collection and storage, then offline processing" in the traditional profile inversion method is avoided, and the cumbersome step of inverting the plasma density distribution through a large number of complex numerical calculations is omitted. The microwave reflectometer can be accurately inverted in real time, improving the efficiency and accuracy of the density profile inversion, and meeting the real-time control requirements of the future fusion device for plasma density profile inversion. In addition, the present application utilizes an end-to-end profile neural network model specially designed for the characteristics of the microwave reflectometer. The model can directly learn the mapping relationship of the density profile distribution from the core raw data (I / Q signal) of the microwave reflectometer, without relying on other intermediate results or assumptions, providing an efficient technical path for real-time application of microwave reflectometer diagnosis.

[0059] In one embodiment, as shown in FIG. 4, Figure 4 The step S203 of extracting and enhancing features of the target I / Q signal by the backbone network in the profile neural network model through the linear rectifier module can include the following steps:

[0060] In step S401, a plurality of parallel convolution branches in the linear rectifier module extract features of the target I / Q signal through convolution kernels of different lengths to obtain a plurality of features of different scales.

[0061] In step S402, the linear rectifier module fuses and summarizes the features of different scales and the original features of the target I / Q signal to obtain the merged multi-scale features.

[0062] Step S403, the multi-scale features are enhanced by the linear rectification module through a nonlinear activation function to obtain enhanced multi-scale features.

[0063] The nonlinear activation function can be a Relu activation function. The introduction of the nonlinear activation function enables each layer of the network to perform more complex mapping and learning, avoids overly simple linear relationships, and thus improves the expression ability of the model.

[0064] Specifically, the terminal extracts features of the target I / Q signal by multiple parallel convolution branches in the linear rectification module through convolution kernels of different lengths to obtain corresponding features of multiple different scales; the linear rectification module fuses and summarizes the features of multiple different scales and the original features of the target I / Q signal to obtain merged multi-scale features; and the linear rectification module enhances the multi-scale features through a Relu activation function to obtain enhanced multi-scale features.

[0065] In this embodiment, the profile neural network model captures features at different spatial resolutions using multi-scale convolution and uses residual connections to mitigate gradient vanishing in deep networks, thereby significantly improving the efficiency and stability of feature extraction.

[0066] In one of the embodiments, the method further comprises the following steps:

[0067] The original time-domain signal is extracted from the microwave signal, and the original time-domain signal is recorded in the target information unit for local storage according to the matching of the number of the microwave reflectometer and the target information unit in the data storage system.

[0068] The information unit can be a basic unit of signal storage or processing, such as a byte, and the target information unit refers to an information unit corresponding to the number of the microwave reflectometer in the data storage system.

[0069] Specifically, the terminal extracts a part of the signal in the microwave signal as the original time-domain signal, matches the target information unit in the data storage system according to the preset number of the microwave reflectometer, and then records the original time-domain signal in the target information unit for local storage, which is beneficial to efficient storage and subsequent use (such as offline model training and manual profile drawing) of the original time-domain signal.

[0070] In one of the embodiments, the method further comprises the following steps:

[0071] The historical density profile data is acquired and the original time domain signal is extracted from the data storage system, the training set, the verification set and the test set are obtained according to the historical density profile data and the original time domain signal, the initial network model to be trained is constructed according to the preset initial hyperparameter, the initial network model is trained and the model parameters of the initial network model are updated by using the training set and the verification set through the deep learning algorithm until the performance index of the updated network model on the test set meets the threshold condition, and the profile neural network model is obtained.

[0072] In the profile neural network model, the model parameters (such as the weights and biases of the neural network) are important components of the model, which determine the connection strength between neurons in the network. During the training process, the model parameters are constantly updated by optimization algorithms (such as adaptive optimization algorithms) to enable the model to better fit the data and complete the task.

[0073] The initial hyperparameter refers to a parameter that needs to be manually set outside the training process of the machine learning or deep learning model. The setting of these parameters has an important influence on the performance of the model, but unlike the parameters of the model itself, the hyperparameters are set before training and are usually not learned through training data.

[0074] The performance index can be the accuracy and efficiency of the density profile inversion.

[0075] Specifically, the terminal splits the historical density profile data and the original time domain signal extracted from the data storage system into a training set, a verification set and a test set according to a certain proportion, and constructs an initial network model to be trained according to the preset initial hyperparameter, and then trains the initial network model by using the training set and the verification set through the deep learning algorithm, so that the model can learn the correlation between different types of density profile data and original time domain signals, help the model adjust the model parameters during the training process, and ensure the accuracy of the plasma density profile inversion. Finally, the test set is used to test the trained network model to ensure that it has a certain density profile inversion accuracy on different types of I / Q signals. The accuracy is evaluated by F1 value and recall rate to ensure that it has strong generalization ability in actual application and can cope with various types of microwave signals. The test can help confirm the actual effect of the model and timely find potential problems.

[0076] In one of the embodiments, the step of training the initial network model by using the training set and the verification set through the deep learning algorithm and updating the model parameters of the initial network model can include the following steps:

[0077] The training set is input into the initial network model, and the initial network model performs offline model training according to the training set; if the output result of the initial network model meets all the verification parameters in the verification set, the model training process is terminated and the model inversion result is obtained; otherwise, the model parameters of the initial network model are adjusted by the adaptive optimization algorithm according to the model inversion result.

[0078] Specifically, the terminal adjusts the model parameters of the initial network model using the adaptive optimization algorithm to optimize the inversion accuracy of the model. The adaptive optimization algorithm is the Adam optimization algorithm. The Adam optimization algorithm can dynamically adjust the learning rate by combining momentum and adaptive learning rate, which helps to accelerate the training process and avoid overfitting or slow training problems. Adam performs well in processing sparse data and a large number of parameters, can efficiently optimize the weights and biases of the model, and quickly converges to a better solution, thereby improving the efficiency of the plasma density profile inversion. The adaptive optimization algorithm can improve the stability of the training process, especially when the data set is large or the features are complex, avoiding the problem of too high or too low learning rate that may occur in traditional optimization algorithms.

[0079] In one of the embodiments, in step S204, the output decoding layer in the profile neural network model performs profile prediction based on the reinforced multi-scale features to obtain the density profile distribution data of the plasma in the microwave reflectometer, which can include the following steps:

[0080] The output decoding layer uses multi-scale convolution to capture the reinforced multi-scale features at different spatial resolutions and uses residual connection to alleviate the gradient vanishing in the profile neural network model to obtain a feature capture result. Based on the feature capture result, the density profile distribution data of the plasma is obtained.

[0081] Specifically, the output decoding layer uses multi-scale convolution to capture the reinforced multi-scale features at different spatial resolutions, which enhances the feature extraction effect through parameter sharing and lightweight convolution, reduces the computational complexity, and uses residual connection to alleviate the gradient vanishing in the deep network, thereby significantly improving the training efficiency and stability, and quickly and accurately predicting the density profile distribution data of the plasma from the original microwave signal.

[0082] In one embodiment, as shown in Figure 5 A specific embodiment of a plasma density profile inversion method is provided, which specifically includes the following steps:

[0083] Step S501, obtaining historical density profile data and extracting original time domain signals from a data storage system, obtaining a training set, a verification set and a test set from the historical density profile data and the original time domain signals; constructing an initial network model to be trained according to preset initial hyperparameters.

[0084] Step S502, input the training set into the initial network model, and perform offline model training according to the training set by the initial network model; if the output result of the initial network model meets all the verification parameters in the verification set, terminate the model training process and obtain the model inversion result; otherwise, adjust the model parameters of the initial network model according to the model inversion result through the self-adaptive optimization algorithm until the performance index of the updated network model on the test set meets the threshold condition, and obtain the profile neural network model.

[0085] Step S503, extract the intermediate frequency signal from the microwave signal collected by the microwave reflectometer, and demodulate the intermediate frequency signal through the I / Q detector to obtain the I / Q signal; input the I / Q signal into the profile neural network model, and perform signal enhancement processing on the I / Q signal by the input coding layer in the profile neural network model to obtain the enhanced target I / Q signal.

[0086] Step S504, the backbone network in the profile neural network model performs feature extraction and feature enhancement on the target I / Q signal through the linear rectification module to obtain the reinforced multi-scale feature.

[0087] Step S505, the output decoding layer in the profile neural network model uses multi-scale convolution to capture the reinforced multi-scale feature at different spatial resolutions, and uses residual connection to alleviate the gradient disappearance in the profile neural network model to obtain the feature capture result; based on the feature capture result, the profile is predicted to obtain the density profile distribution data of the plasma.

[0088] The beneficial effects brought by the above embodiments are as follows:

[0089] 1. The problem of complex process and poor real-time performance of traditional methods is solved:

[0090] At present, the method based on CNN is usually limited to beat frequency signal extraction or local feature calculation, and still needs to rely on complex post-processing algorithm to finally obtain the density distribution, and the real-time performance is difficult to guarantee. The present scheme realizes the direct mapping from the original I / Q signal to the density distribution through the end-to-end network architecture, omits the intermediate complex calculation steps, compresses the traditional serial processing process of hundreds of milliseconds to 4 milliseconds, and improves the real-time performance.

[0091] 2. The problem of high inference delay and error accumulation of advanced models is solved:

[0092] Although the precision of the autoregressive model based on Transformer is high, the inference time and the length of the predicted sequence are linearly or even quadratically related, and the point-by-point prediction from the edge to the core of the plasma will cause error accumulation, resulting in decreased accuracy and increased calculation delay in the core, which cannot meet the needs of high-resolution real-time inversion. The linear rectifier used in this scheme can quickly calculate the density distribution of the entire profile, and its inference time is only related to the length of the input sequence. The prediction accuracy of the ELM (Edge Localized Modes) event is 97%, and the average error in the core region is less than 3%, ensuring the reliability of the data.

[0093] 3. The problem of poor universality and flexibility of special hardware solutions is solved:

[0094] Although the data processing scheme based on special hardware such as FPGA has low delay, it is usually tailored for specific algorithms and diagnostic quantities, and the hardware logic is fixed, making it difficult to adapt to different devices or algorithm iterations. This scheme can synchronize and transfer data with high real-time requirements to, for example, a general LabVIEW platform, while complex inversion algorithms are executed on a general server. This architecture can take advantage of the synchronization of LabVIEW, and also combine the powerful computing power and flexibility of general computing platforms, so that algorithm updates and model optimization do not need to change the hardware design, enhancing the universality and maintainability of the system.

[0095] 4. A special real-time inversion program suitable for reflectometer diagnosis is realized:

[0096] Due to the fundamental differences in diagnostic principles, the density inversion methods developed for other diagnostic devices such as polarimeters and interferometers cannot be directly applied to the processing of reflectometer signals. This scheme is a end-to-end deep learning model designed specifically for the characteristics of reflectometer IQ signals. It directly learns the mapping relationship of density distribution from the core raw data (IQ signal) of the reflectometer, without relying on other diagnostic intermediate results or assumptions, providing an efficient technical path for the real-time application of reflectometer diagnosis.

[0097] To more clearly illustrate the plasma density profile inversion method provided by the embodiments of the present application, the following will specifically describe the plasma density profile inversion method with an application embodiment. In one embodiment, as shown in Figure 6 The present application also provides a plasma density profile inversion method, which specifically includes the following steps:

[0098] The microwave signal to be processed is collected by a microwave reflectometer acquisition system, and then the microwave signal is divided into two parts. One part of the signal is amplified and transmitted to the plasma, while the other part of the signal is coupled to the receiver as a reference for storage. In the plasma, the microwave is reflected at the cutoff layer, and then the reflected signal is collected by the receiving antenna, which converts the received signal into an intermediate frequency signal by mixing the local oscillator signal, and the intermediate frequency signal is demodulated by the in-phase and quadrature detector (I / Q detector). The data acquisition and control system (DACS) acquires the I / Q signal, which can realize the synchronization of signal acquisition, local storage and TCP / IP streaming through Labview (a graphical programming language) in this process. Then the I / Q signal is uploaded to the server. Here, the TCP protocol is used instead of the UDP protocol to ensure that the data packets arrive in order and are not lost. Then the profile neural network model built by the Inception neural network architecture on the server is used to process the I / Q signal in real time, and the predicted plasma density distribution is given. Finally, the plasma control system (PCS) controls the charging and aeration system according to the electron density profile data to complete the feedback control.

[0099] The overall architecture of the profile neural network model is composed of an input encoding layer, a backbone network and an output decoding layer, which cooperatively perform feature extraction and profile prediction. The input encoding layer increases the channel dimension of the data and enhances the representation ability, preparing for effective feature extraction. The backbone network includes two linear rectifier functions Inception_1 and Inception_2, each of which includes four branches. After the initial 1x1 convolution, three branches apply 3x1, 5x1 and 7x1 convolutions to extract multi-scale features. Then the outputs are connected, fused by another 1x1 convolution, and connected with the input through a residual connection to enhance the features and maintain the gradient flow. The data reduction module (i.e. Reduction module) is as follows Figure 7As shown, its main role is to perform specific operations on the input data to achieve feature compression and information extraction. Through the data dimension reduction module, the network can better focus on important feature information, thereby improving the performance of the model. Specifically, the data dimension reduction module receives the output result of the previous layer (i.e., the output result of Inception_1), and then processes the received output result through the four parallel branches in the module. Each branch first adjusts the channel size through 1x1 convolution, and then uses maximum pooling or 3x1, 5x1 and 7x1 convolution for down-sampling. This process reduces the sequence length by half while doubling the channel count and preserving multi-scale feature information. Then, the reduced data is subjected to secondary feature extraction by Inception_2 to obtain high-dimensional features. Finally, the output decoding layer processes the high-dimensional features extracted by the backbone network to produce the predicted plasma density distribution. This model uses multi-scale convolution to capture features at different spatial resolutions and uses residual connections to mitigate gradient vanishing in deep networks, thereby significantly improving training efficiency and stability.

[0100] The beneficial effects brought by the above embodiments are as follows:

[0101] The present embodiment proposes a plasma density distribution real-time inversion system based on streaming and deep learning, which is a lightweight density real-time inversion mechanism. The system achieves a millisecond-level response from raw signal acquisition to density distribution output. The system works collaboratively by three modules: synchronous acquisition and transmission, end-to-end inversion, and feedback control. The streaming module based on Labview mainly undertakes the functions of signal synchronous acquisition and real-time transmission. The density real-time inversion module based on deep learning is mainly deployed on a graphics processing server to achieve end-to-end density mapping. The feedback control function is completed by the plasma control system (PCS) according to the density data calculated by the inversion module.

[0102] It should be understood that although each step in the flowchart involved in each embodiment as described above is displayed in sequence according to the arrow, these steps are not necessarily executed in the order indicated by the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, at least part of the steps in the flowchart involved in each embodiment as described above can include multiple steps or stages, which are not necessarily executed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be alternately or alternately executed with at least part of other steps or steps or stages in other steps.

[0103] Based on the same inventive concept, the application also provides a plasma density profile inversion device for implementing the above-mentioned plasma density profile inversion method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more plasma density profile inversion device embodiments provided below can refer to the limitations of the plasma density profile inversion method described above, which will not be repeated here.

[0104] In one exemplary embodiment, as shown in Figure 8 a plasma density profile inversion device is provided, which can include:

[0105] The signal extraction module 801 is configured to acquire a microwave signal collected by a microwave reflectometer, extract an intermediate frequency signal from the microwave signal, and obtain an I / Q signal by demodulating the intermediate frequency signal through an I / Q detector.

[0106] The signal enhancement module 802 is configured to input the I / Q signal into a profile neural network model, perform signal enhancement processing on the I / Q signal by an input encoding layer in the profile neural network model, and obtain an enhanced target I / Q signal.

[0107] The feature extraction module 803 is configured to perform feature extraction and feature enhancement on the target I / Q signal by a backbone network in the profile neural network model through a linear rectification module, and obtain a reinforced multi-scale feature.

[0108] The profile prediction module 804 is configured to perform profile prediction on the reinforced multi-scale feature according to an output decoding layer in the profile neural network model, and obtain density profile distribution data of the plasma in the microwave reflectometer.

[0109] In one embodiment, the feature extraction module 803 is further configured to perform feature extraction on the target I / Q signal by a plurality of parallel convolution branches in the linear rectification module through convolution kernels of different lengths, respectively, to obtain a plurality of different scale features corresponding to the target I / Q signal; fuse and aggregate the plurality of different scale features and the original features of the target I / Q signal by the linear rectification module, to obtain merged multi-scale features; and perform feature enhancement on the multi-scale features by the linear rectification module through a nonlinear activation function, to obtain the reinforced multi-scale features.

[0110] In one embodiment, the device can further include a signal storage module configured to extract an original time domain signal from the microwave signal; and record the original time domain signal in a target information unit in a data storage system according to the number matching data of the microwave reflectometer, for local storage.

[0111] In an embodiment, the device can further include a model training module configured to obtain historical density profile data and extract original time domain signals from a data storage system, obtain a training set, a validation set and a test set according to the historical density profile data and the original time domain signals, construct an initial network model to be trained according to preset initial hyperparameters, and train the initial network model by a deep learning algorithm and update model parameters of the initial network model using the training set and the validation set until a performance index of the updated network model on the test set meets a threshold condition, and obtain a profile neural network model.

[0112] In an embodiment, the model training module is further configured to input the training set into the initial network model, perform offline model training on the initial network model according to the training set, terminate the model training process and obtain a model inversion result if an output result of the initial network model meets all validation parameters in the validation set, and adjust the model parameters of the initial network model according to the model inversion result by an adaptive optimization algorithm.

[0113] In an embodiment, the profile prediction module 804 is configured to capture reinforced multi-scale features at different spatial resolutions by multi-scale convolution using an output decoding layer, and use a residual connection to alleviate gradient vanishing in the profile neural network model, to obtain a feature capture result, and perform profile prediction based on the feature capture result to obtain density profile distribution data of the plasma.

[0114] The above-described various modules in the plasma density profile inversion device can be realized by software, hardware and combinations thereof in whole or in part. The above-described various modules can be embedded in or independent of a processor in a computer device in hardware form, or can be stored in a memory in a computer device in software form, so as to be called and executed by a processor to perform operations corresponding to the above-described various modules.

[0115] In an exemplary embodiment, a computer device is provided, which can be a terminal, and an internal structure diagram of the computer device can be as shown in FIG. 8. Figure 9The computer device shown in the figure includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. Among them, 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. Among them, the processor of the computer device is used to provide computing and control capability. 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 operating system and the computer program in the non-volatile storage medium to run. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be realized through WIFI, mobile cellular network, near field communication (NFC) or other technologies. The computer program is executed by the processor to realize a plasma density profile inversion method. 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. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0116] Those skilled in the art can understand that, Figure 9 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0117] In one embodiment, a computer device is also provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to realize the steps in each of the above method embodiments.

[0118] In one embodiment, a computer readable storage medium is provided, which stores a computer program, and the computer program is executed by a processor to realize the steps in each of the above method embodiments.

[0119] In one embodiment, a computer program product is provided, including a computer program, and the computer program is executed by a processor to realize the steps in each of the above method embodiments.

[0120] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.

[0121] It can be understood by those skilled in the art that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing related hardware through a computer program. 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 above-mentioned embodiments of each method. In the embodiments provided in the present application, any reference to memory, database or other medium 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 storage, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive 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 but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., without being limited thereto.

[0122] Any technical features in the above embodiments can be combined, and for the sake of brevity, not all possible combinations are described above, however, any combination of these technical features is deemed to be within the scope of the present application.

[0123] The above embodiments only express several implementation manners of the present application, and the description is relatively specific and detailed, but it should not be understood as a limitation on the patent scope of the present application. It should be pointed out that, for ordinary skilled persons in the art, several modifications 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 should be subject to the appended claims.

Claims

1. A method for plasma density profile inversion, characterized in that, The method includes: The microwave signal collected by the microwave reflectometer is acquired, the intermediate frequency signal is extracted from the microwave signal, and the intermediate frequency signal is regulated by the I / Q detector to obtain the I / Q signal; The I / Q signal is input into a profile neural network model, and the input encoding layer in the profile neural network model performs signal enhancement processing on the I / Q signal to obtain the enhanced target I / Q signal; The backbone network of the profile neural network model extracts and enhances features from the target I / Q signal through a linear rectification module. Multiple parallel convolutional branches in the linear rectification module extract features from the target I / Q signal using convolutional kernels of different lengths, resulting in multiple features at different scales. The linear rectification module then fuses and summarizes these multiple features at different scales with the original features of the target I / Q signal to obtain merged multi-scale features. Finally, the linear rectification module enhances these multi-scale features using a non-linear activation function to obtain enhanced multi-scale features. The output decoding layer of the profile neural network model performs profile prediction based on the enhanced multi-scale features. The output decoding layer captures the enhanced multi-scale features at different spatial resolutions using multi-scale convolutions and uses residual connections to mitigate gradient vanishing in the profile neural network model, obtaining feature capture results. Based on these feature capture results, profile prediction is performed to obtain the density profile distribution data of the plasma in the microwave reflectometer. The overall architecture of the profile neural network model consists of the input encoding layer, the backbone network, and the output decoding layer. The backbone network includes two linear rectified functions, Inception_1 and Inception_2, each with four branches. After the initial 1×1 convolution, three branches apply 3×1 and 5×1 convolutions. Multi-scale features are extracted using 7×1 convolutions, and the outputs are then concatenated and fused using another 1×1 convolution. The outputs are then connected to the input via residual connections to enhance features and preserve gradient flow. The data dimensionality reduction module receives the output from Inception_1 and processes it through four parallel branches. Each branch first adjusts the channel size using a 1×1 convolution, then downsamples using max pooling or 3×1, 5×1, and 7×1 convolutions. This process halves the sequence length while doubling the channel count and preserving multi-scale feature information. Inception_2 then performs secondary feature extraction on the dimensionality-reduced data to obtain high-dimensional features. Finally, the output decoding layer processes the high-dimensional features extracted by the backbone network to generate the predicted plasma density distribution.

2. The method according to claim 1, characterized in that, The method further includes: Extract the original time-domain signal from the microwave signal; According to the number of the microwave reflectometer, the target information unit in the data storage system is matched, and the original time domain signal is recorded into the target information unit for local storage; the target information unit refers to the information unit in the data storage system that corresponds to the number.

3. The method according to claim 2, characterized in that, The method further includes: Historical density profile data is acquired and the original time-domain signal is extracted from the data storage system. Training set, validation set and test set are obtained based on the historical density profile data and the original time-domain signal. Based on the preset initial hyperparameters, construct the initial network model to be trained; The initial network model is trained and updated using a deep learning algorithm using the training set and the validation set until the performance index of the updated network model on the test set meets the threshold condition, thus obtaining the profile neural network model.

4. The method according to claim 3, characterized in that, The step of training the initial network model using the training set and the validation set through a deep learning algorithm and updating the model parameters of the initial network model includes: The training set is input into the initial network model, and the initial network model performs offline model training based on the training set; If the output of the initial network model satisfies all the validation parameters in the validation set, then the model training process is terminated and the model inversion result is obtained; Otherwise, the model parameters of the initial network model are adjusted using an adaptive optimization algorithm based on the model inversion results.

5. A plasma density profile inversion device, characterized in that, The device includes: The signal extraction module is used to acquire the microwave signal collected by the microwave reflector, extract the intermediate frequency signal from the microwave signal, and adjust the intermediate frequency signal through the I / Q detector to obtain the I / Q signal; The signal enhancement module is used to input the I / Q signal into the profile neural network model, and the input encoding layer in the profile neural network model performs signal enhancement processing on the I / Q signal to obtain the enhanced target I / Q signal; The feature extraction module is used to extract and enhance features from the target I / Q signal through the backbone network of the profile neural network model via a linear rectification module. Multiple parallel convolutional branches in the linear rectification module extract features from the target I / Q signal using convolutional kernels of different lengths, obtaining multiple features at different scales. The linear rectification module then fuses and summarizes the multiple features at different scales and the original features of the target I / Q signal to obtain merged multi-scale features. Finally, the linear rectification module enhances the multi-scale features using a non-linear activation function to obtain enhanced multi-scale features. The profile prediction module is used to perform profile prediction based on the enhanced multi-scale features by the output decoding layer in the profile neural network model. The output decoding layer captures the enhanced multi-scale features at different spatial resolutions using multi-scale convolutions and uses residual connections to mitigate gradient vanishing in the profile neural network model, obtaining feature capture results. Based on the feature capture results, profile prediction is performed to obtain the density profile distribution data of the plasma in the microwave reflectometer. The overall architecture of the profile neural network model consists of the input encoding layer, the backbone network, and the output decoding layer. The backbone network includes two linear rectified functions, Inception_1 and Inception_2, each with four branches. After an initial 1×1 convolution, three branches apply a 3×1 convolution. Multi-scale features are extracted using 5×1 and 7×1 convolutions, and the outputs are then concatenated and fused using another 1×1 convolution. The outputs are then connected to the input via residual connections to enhance features and preserve gradient flow. The data dimensionality reduction module receives the output from Inception_1 and processes it through four parallel branches. Each branch first adjusts the channel size using a 1×1 convolution, then downsamples using max pooling or 3×1, 5×1, and 7×1 convolutions. This process halves the sequence length while doubling the channel count and preserving multi-scale feature information. Inception_2 then performs secondary feature extraction on the dimensionality-reduced data to obtain high-dimensional features. Finally, the output decoding layer processes the high-dimensional features extracted by the backbone network to generate the predicted plasma density distribution.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 4.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 4.

Images