Multi-channel ion temperature inference method and system for magnetic confinement nuclear fusion device

Abstract

The invention discloses a multichannel ion temperature inference method and system for a magnetic confinement nuclear fusion device. The method comprises the following steps: acquiring a charge exchange composite spectrum curve as input data; performing data analysis on the input data to obtain ion temperature data as an ion temperature label; constructing a historical experiment data set according to the input data and the ion temperature label; constructing a neural network model, and carrying out model training on the neural network model by adopting a historical experiment data set to obtain a trained neural network model; and arranging the trained neural network model under a TFLite reasoning framework, inputting multi-channel real-time acquired spectral data into the trained neural network model, and carrying out multi-channel real-time ion temperature parallel deduction to obtain the real-time ion temperature. According to the invention, multichannel ultrafast ion temperature deduction can be obtained, and real-time ion temperature measurement requirements in a feedback control scene are met.

Description

Technical Field

[0001] The present invention relates to the field of ion temperature diagnosis for magnetic confinement fusion devices, and particularly to a multi-channel ion temperature inference method and system for magnetic confinement fusion devices. Background Art

[0002] For magnetic confinement fusion devices represented by tokamak devices, high-power and long-pulse operation is an important operating state and a key operating goal. Under high-power and long-pulse operating conditions, it is necessary to continuously understand the state of the plasma and perform feedback control on the plasma at the necessary time to keep it within the target operating range. This requires obtaining real-time measurement results of key plasma parameters. Plasma ion temperature is a very critical physical quantity, which is related to the core heat transport of the plasma and directly affects the judgment of whether the triple product of the fusion plasma reaches the ignition condition.

[0003] Charge exchange recombination spectroscopy is a common and important ion temperature diagnostic method for magnetic confinement fusion devices. This diagnosis collects and analyzes the de-excitation radiation after the charge exchange recombination reaction between ions in the plasma and neutral beam ions, and obtains ion temperature information through the Doppler broadening of this radiation spectrum. Due to the existence of multiple different reactions in the plasma, there are multiple spectral lines with different wavelengths and different broadenings covering each other near this radiation. Therefore, it is necessary to de-spectrum the obtained radiation spectrum to infer the ion temperature result. The existing mature de-spectroscopic method is to assume the composition structure of the spectrum, write the spectral intensity as a form of superposition of multiple Gaussian peaks, and iteratively solve to obtain the ion temperature through the least squares fitting method. This method can obtain relatively accurate plasma ion temperature information, but this method needs to perform iterative solution on the diagnostic data, cannot perform parallel calculation, has a slow calculation speed, and is difficult to meet the requirements of real-time measurement in the feedback control scenario.

[0004] In view of this, the present application is specifically proposed. Summary of the Invention

[0005] The technical problem to be solved by the present invention is that the de-spectroscopic method in the existing temperature inference for magnetic confinement fusion devices cannot meet the problem of real-time inference of ion temperature and cannot efficiently and quickly calculate the plasma ion temperature. The purpose of the present invention is to provide a multi-channel ion temperature inference method and system for magnetic confinement fusion devices, which fit the mapping relationship from the spectral data obtained from historical experiments to the plasma ion temperature based on a neural network model, and use multi-channel parallel technology to meet the requirement of obtaining real-time ion temperature in control feedback, and can more efficiently and quickly infer the ion temperature.

[0006] The present invention is realized through the following technical solutions:

[0007] In a first aspect, the present invention provides a multi-channel ion temperature inference method for a magnetic confinement fusion device, the method comprising:

[0008] Obtain a charge exchange recombination spectroscopy curve as input data X; and perform data analysis on the input data X to obtain ion temperature data as an ion temperature label Y;

[0009] Construct a historical experimental data set based on the input data X and the ion temperature label Y;

[0010] Construct a neural network model, and use the historical experimental data set to train the neural network model to obtain a trained neural network model;

[0011] Deploy the trained neural network model under the TFLite inference framework, and input the multi-channel real-time acquired spectral data into the trained neural network model for multi-channel real-time parallel ion temperature inference to obtain the real-time ion temperature.

[0012] Further, obtain a charge exchange recombination spectroscopy curve as input data X; and perform data analysis on the input data X to obtain ion temperature data as an ion temperature label Y, including:

[0013] Obtain a charge exchange recombination spectroscopy curve as input data X according to the historical experimental data of the magnetic confinement fusion device; the input data X refers to a spectrum with a dimension of 1×512;

[0014] Perform data analysis on the input data X using the least squares fitting method and perform abnormal data cleaning to obtain ion temperature data as an ion temperature label Y; the ion temperature label Y refers to a scalar number of the ion temperature at a set position in the plasma.

[0015] Further, each set of data in the historical experimental data set is represented as a spectrum with a dimension of 1×512 and an ion temperature label with a dimension of 1×1 inferred therefrom.

[0016] Further, constructing a neural network model includes:

[0017] Input layer: The input layer size is 1×512;

[0018] First convolutional neural network layer: Construct a first convolutional neural network layer with 4 layers and 32 convolutional kernels, and perform preliminary feature extraction on the input data X based on the first convolutional neural network layer; at this time, the output matrix size of the neural network model is converted to 256×32;

[0019] The second convolutional neural network layer and the third convolutional neural network layer: Construct a second convolutional neural network layer with 3 layers and 64 convolutional kernels, and a third convolutional neural network layer with 3 layers and 128 convolutional kernels; The first convolutional neural network layer, the second convolutional neural network layer, and the third convolutional neural network layer are combined in the form of cross-layer residual connections to obtain the final neural network feature map; At this time, the output matrix size of the neural network model is converted to 12×128;

[0020] Global pooling layer: Construct 1 global pooling layer to average the final neural network feature map extracted by the neural network model over the entire spectral range; At this time, the output matrix size of the neural network model is converted to 1×128;

[0021] Fully connected layer: Construct one fully connected layer each with 256, 64, and 1 convolutional kernels to perform dimensional transformation on the output of the neural network model to obtain the final ion temperature scalar output with a length of 1.

[0022] Furthermore, the training process of the neural network model is as follows:

[0023] The weight parameters of each convolutional neural network layer in the neural network model are obtained through training by the stochastic gradient descent method. The network parameters are trained using the established historical experimental dataset, that is, the relationship between the input data X and the label ion temperature Y in the historical experimental data is fitted to obtain a function F that satisfies Y = F(X), and a calculation model for inferring the ion temperature from the spectral data is obtained, and the calculation model of the ion temperature is used as the trained neural network model.

[0024] Furthermore, the specific steps for training the weight parameters are as follows:

[0025] Randomly obtain 128 spectral data and the corresponding ion temperature labels from the historical experimental dataset as a batch of data each time, and input them into the neural network model for calculation to obtain the temperature result Y';

[0026] Take the mean absolute error between the temperature result Y' and the ion temperature label Y as the loss function, and minimize the inference loss through the gradient descent method to update the weight parameters of the neural network model;

[0027] Iterate the above process repeatedly until the loss function no longer decreases. At this time, the weight parameters are the final parameters.

[0028] Furthermore, deploy the trained neural network model under the TFLite inference framework, and input the multi-channel real-time acquired spectral data into the trained neural network model for multi-channel real-time ion temperature parallel inference to obtain the real-time ion temperature, including:

[0029] Deploy the trained neural network model under the TFLite inference framework;

[0030] Based on the magnetic confinement fusion device, spectral data of multiple channels are obtained at each moment, and the spectral data of these multiple channels are used as a batch of samples and input into the neural network model trained under the TFLite inference framework;

[0031] Based on the trained neural network model, parallel inference of multi-channel real-time ion temperature is performed to calculate the real-time ion temperature.

[0032] Furthermore, it also includes: cleaning abnormal data from the spectral data obtained in real time for multiple channels.

[0033] In a second aspect, the present invention further provides a multi-channel ion temperature inference system for a magnetic confinement fusion device, and this system includes:

[0034] An acquisition unit, configured to obtain a charge exchange recombination spectral curve as input data X according to the historical experimental data of the magnetic confinement fusion device;

[0035] A label unit, configured to perform data analysis on the input data X to obtain ion temperature data as the ion temperature label Y;

[0036] A dataset construction unit, configured to construct a historical experimental dataset according to the input data X and the ion temperature label Y;

[0037] A model construction and training unit, configured to construct a neural network model and use the historical experimental dataset to train the neural network model to obtain a trained neural network model;

[0038] A real-time parallel inference unit, configured to deploy the trained neural network model under the TFLite inference framework, and input the spectral data obtained in real time for multiple channels into the trained neural network model to perform parallel inference of multi-channel real-time ion temperature to obtain the real-time ion temperature.

[0039] Furthermore, the execution process of the real-time parallel inference unit is as follows:

[0040] Deploy the trained neural network model under the TFLite inference framework;

[0041] Based on the magnetic confinement fusion device, spectral data of multiple channels are obtained at each moment, and the spectral data of these multiple channels are used as a batch of samples and input into the neural network model trained under the TFLite inference framework;

[0042] Based on the trained neural network model, parallel inference of multi-channel real-time ion temperature is performed to calculate the real-time ion temperature.

[0043] In a third aspect, the present invention further provides a computer-readable storage medium storing a computer program which, when executed by a processor, implements the above-mentioned multi-channel ion temperature inference method for a magnetic confinement fusion device.

[0044] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0045] The multi-channel ion temperature inference method and system for a magnetic confinement fusion device of the present invention fit the mapping relationship from spectral data obtained from historical experiments to plasma ion temperature based on a neural network model, and use multi-channel parallel technology to meet the requirement of obtaining real-time ion temperature in control feedback, and can infer the ion temperature more efficiently and quickly. The present invention meets the requirement of real-time ion temperature measurement in a feedback control scenario. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, form a part of this application, and do not limit the embodiments of the present invention. In the drawings:

[0047] Figure 1 is a flowchart of the multi-channel ion temperature inference method for a magnetic confinement fusion device of the present invention;

[0048] Figure 2 is a detailed flowchart of the multi-channel ion temperature inference method for a magnetic confinement fusion device of the present invention;

[0049] Figure 3 is a comparison diagram of the ion temperature output by the neural network model of the present invention and the label value;

[0050] Figure 4 is a structural block diagram of the multi-channel ion temperature inference system for a magnetic confinement fusion device of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0051] To make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the embodiments and the drawings. The illustrative embodiments and descriptions thereof of the present invention are only used to explain the present invention and do not limit the present invention.

[0052] The spectral decomposition method in the existing temperature inference for a magnetic confinement fusion device cannot meet the problem of real-time inference of ion temperature. The existing spectral decomposition method assumes the composition structure of the spectrum, writes the spectral intensity as a form of superposition of multiple Gaussian peaks, and iteratively solves to obtain the ion temperature by the method of least squares fitting. This method can obtain relatively accurate plasma ion temperature information, but this method needs to iteratively solve the diagnostic data, cannot perform parallel computing, has a slow computing speed, and is difficult to meet the requirement of real-time measurement in a feedback control scenario.

[0053] Therefore, the present invention designs a multi-channel ion temperature inference method and system for a magnetic confinement fusion device. By fitting the mapping relationship from spectral data obtained from historical experiments to plasma ion temperature based on a neural network model and using multi-channel parallel technology, the requirement of obtaining real-time ion temperature in control feedback is realized, and the ion temperature can be inferred more efficiently and quickly. Among them, the spectral data obtained from historical experiments also needs to be analyzed by existing methods first to obtain the ion temperature label Y to prepare for constructing the historical experiment dataset.

[0054] Example 1

[0055] As Figure 1 and Figure 2 shown, the multi-channel ion temperature inference method for a magnetic confinement fusion device of the present invention includes:

[0056] Step 1, obtain the charge exchange recombination spectrum curve as the input data X; and perform data analysis on the input data X to obtain the ion temperature data as the ion temperature label Y;

[0057] Step 1 specifically includes:

[0058] Step 11, obtain the charge exchange recombination spectrum curve as the input data X according to the historical experiment data of the magnetic confinement fusion device; the input data X refers to a spectrum with a dimension of 1×512, that is, the input data X is a spectrum with a length of 512, which characterizes the spectral intensity count (counts) of the radiation emitted by the charge exchange recombination reaction between the ions in the plasma and the neutral beam ions in different wavelength bands, and its order of magnitude is roughly in the range of 0 to 4000 counts.

[0059] Step 12, through an offline data analysis method, perform manual data analysis on the input data X to obtain the ion temperature data as the ion temperature label Y. Usually, Y is a scalar number characterizing the ion temperature at a specific position in the plasma, and its order of magnitude is roughly in the range of 0 to 3000 electron volts (eV).

[0060] Specifically, the offline data analysis method refers to writing the spectral intensity as a form of superposition of multiple Gaussian peaks by assuming the composition structure of the spectrum, and using the least squares fitting method to iteratively solve to obtain the ion temperature data as the ion temperature label Y; in addition, the obtained ion temperature data is also subjected to abnormal data cleaning.

[0061] Step 2, construct a historical experiment dataset according to the input data X and the ion temperature label Y;

[0062] Based on the principle of charge exchange recombination spectroscopy diagnosis, the input data X should be collected during the period of neutral beam heating. It can include spectral data from different spatial channels. The dimensions, units, and other information of the input data X and the ion temperature label Y are shown in Table 1. By collecting and analyzing X in the historical experimental data generated during the operation of the fusion facility, each spectrum X with a length of 512 and the inferred ion temperature label Y with a length of 1 are used as a set of data in the dataset (that is, each set of data in the historical experimental dataset is represented as a spectrum with a dimension of 1×512 and the inferred ion temperature label with a dimension of 1×1). A large amount of historical experimental datasets can be gradually accumulated. To ensure the reliability of the neural network model, this dataset should contain no less than 100,000 sets of data. For the data whose ion temperature cannot be inferred by the least squares fitting method and the data with obvious errors in the inferred ion temperature, they need to be excluded from the dataset.

[0063] Table 1 Key information table of the input data X and the ion temperature label Y of the model

[0064]

[0065] Step 3: Construct a neural network model and use the historical experimental dataset to train the neural network model to obtain a trained neural network model;

[0066] Step 3 is mainly to establish a network model based on a one-dimensional convolutional neural network. This neural network model is composed of a one-dimensional convolutional neural network and a fully connected layer, etc. The main construction idea is as follows:

[0067] First, the input layer: The size of the input layer is 1×512;

[0068] Secondly, the first convolutional neural network layer: Construct 4 first convolutional neural network layers with 32 convolutional kernels. Based on the first convolutional neural network layer, preliminary feature extraction is performed on the input data X, and the data is converted into the form of a neural network feature map; at this time, the output matrix size of the neural network model is converted to 256×32;

[0069] Subsequently, the second convolutional neural network layer and the third convolutional neural network layer: Construct 3 second convolutional neural network layers with 64 convolutional kernels and 3 third convolutional neural network layers with 128 convolutional kernels; every 3 neural network layers are combined in the form of cross-layer residual connections, that is, the first convolutional neural network layer, the second convolutional neural network layer, and the third convolutional neural network layer are combined in the form of cross-layer residual connections to obtain the final neural network feature map; this architecture comprehensively ensures the deep feature representation ability of the neural network and the tolerance ability to the gradient vanishing problem. At this time, the output matrix size of the neural network model is converted to 12×128;

[0070] Next, the global pooling layer: Construct 1 global pooling layer to average the final neural network feature map extracted by the neural network model across the entire spectral range, so as to add a strong translation invariance constraint to the fitting of the neural network and improve its performance. At this time, the output matrix size of the neural network model is converted to 1×128;

[0071] Finally, the fully connected layer: Construct one fully connected layer each with 256, 64, and 1 convolutional kernels to perform dimensional transformation on the output of the neural network model, and obtain the final ion temperature scalar output with a length of 1.

[0072] The input of the above neural network model is the spectrum with a length of 512 for each channel of each frame, and the output is the temperature value with a length of 1. The weight parameters of each convolutional neural network layer in the neural network model are obtained through training by the stochastic gradient descent method. The network parameters are trained using the established historical experimental dataset, that is, the relationship between the input data X and the labeled ion temperature Y in the historical experimental data is fitted to obtain the function F that satisfies Y = F(X), and a calculation model for inferring the ion temperature from the spectral data is obtained. During the training process, 128 spectral data and the corresponding ion temperature labels are randomly grabbed from the historical experimental dataset as a batch of data each time, and they are input into the neural network model for calculation to obtain the temperature result Y'. The mean absolute error between the temperature result Y' and the ion temperature label Y is used as the loss function, and the inference loss is minimized through the gradient descent method to update the weight parameters of the neural network model. This process iterates the above process repeatedly until the loss function no longer decreases. The weight parameters at this time are the final parameters.

[0073] Step 4: Deploy the trained neural network model under the TFLite inference framework, and input the multi-channel spectral data obtained in real time into the trained neural network model to perform parallel inference of the multi-channel real-time ion temperature and obtain the real-time ion temperature.

[0074] Step 4 specifically includes:

[0075] Use the TFLite inference framework to deploy the trained neural network model under the TFLite inference framework, make full use of the GPU memory width, simultaneously obtain the spectral data of multiple channels at each moment based on the magnetic confinement fusion device, and perform abnormal data cleaning on the multi-channel spectral data obtained in real time; use the multi-channel spectral data as a batch of samples and input them into the neural network model trained under the TFLite inference framework; based on the trained neural network model, perform parallel inference of the multi-channel real-time ion temperature and calculate the real-time ion temperature.

[0076] In the traditional method, data analysis needs to be performed on these 32 channels separately, and the total time consumption is 32 channels × the time consumption of a single channel. Since the neural network model is trained with a batch size of 128 in the calculation stage, in the inference stage, based on the same hardware architecture, parallel inference of 1 to 128 samples is also supported, and the time consumption of multi-channel parallel calculation will not increase significantly compared with that of a single channel. In the present invention, since the preposed spectral diagnosis system supports simultaneous measurement and output of 32 spectral channels, the neural network model performs parallel inference of 32 spectra in the application stage.

[0077] In specific implementation, this embodiment takes the ion temperature inference method of charge exchange recombination spectroscopy diagnosis of the HL-2A device in China as an example, and the specific steps are as follows:

[0078] (1) Collect the spectral data of 81 effective signals of charge exchange recombination spectroscopy diagnosis of the HL-2A device, use the least squares fitting method to fit each spectrum to obtain the corresponding ion temperature result, and check each ion temperature one by one to ensure the rationality of the result. The selected effective spectral data and the ion temperature data inferred therefrom are combined into a data set. Each discharge contains 20 - 60 frames of effective signals, each frame contains spectral data of 32 channels. After removing abnormal signals, there are 121,995 spectral signals in total, where the length of each spectral signal is 512, and the corresponding temperature forms a historical experimental data set. Collect the input diagnostic data X and the ion temperature label Y.

[0079] (2) Model construction and training. Construct a neural network model based on a one-dimensional convolutional neural network and a fully connected layer, etc. The length of the input node is 512, and the output node is 1 in length. Randomly divide the historical experimental data set constructed in the previous step into a training set and a test set according to a ratio of 4:1, train the weight parameter matrix in the neural network model on the training set, and perform test verification on the test set. During the training process, randomly grab 128 groups of data from the training set as a batch of data for training each time, and repeatedly iterate to obtain the final weight parameters. The trained model is tested on the test set to obtain the model performance evaluation. Use the coefficient of determination as the evaluation coefficient, where y i represents a certain label Y, represents the output of the corresponding data model, represents the average value of the label Y in the test set. The average value obtained from 15 tests is R2 = 0.950, and the model's inferred ion temperature result has a high degree of coincidence with the offline least squares fitting algorithm. Figure 3 Shows the comparison between the output of the neural network model and the label. It can be seen that at different channels and different times, the two are in good agreement.

[0080] (3) Real-time testing during the inference phase. After the training of the neural network model for ion temperature inference is completed, using the TFLite inference framework, on a general desktop computer configuration (CPU is i7-14700KF, GPU is NVIDIA RTX 4070TiS), ion temperature inference is performed simultaneously on 32 channels in the test set. The total time consumption for 32 channels is on average 1.22 milliseconds, meeting the multi-channel real-time diagnostic requirements for feedback control.

[0081] Compared with the existing technology, the present invention can obtain multi-channel ultrafast ion temperature inference and meet the real-time ion temperature measurement requirements in the feedback control scenario.

[0082] Embodiment 2

[0083] As Figure 4 shown, the difference between this embodiment and Embodiment 1 is that this embodiment provides a multi-channel ion temperature inference system for a magnetic confinement fusion device, and the functions of this system correspond one-to-one with those of the multi-channel ion temperature inference method for a magnetic confinement fusion device in Embodiment 1; this system includes:

[0084] An acquisition unit, configured to obtain a charge exchange recombination spectrum curve as input data X according to the historical experimental data of the magnetic confinement fusion device;

[0085] A labeling unit, configured to perform data analysis on the input data X to obtain ion temperature data as the ion temperature label Y;

[0086] A data set construction unit, configured to construct a historical experimental data set according to the input data X and the ion temperature label Y;

[0087] A model construction and training unit, configured to construct a neural network model and train the neural network model using the historical experimental data set to obtain a trained neural network model;

[0088] A real-time parallel inference unit, configured to deploy the trained neural network model under the TFLite inference framework, and input the multi-channel real-time acquired spectral data into the trained neural network model to perform multi-channel real-time ion temperature parallel inference to obtain the real-time ion temperature.

[0089] As a further implementation, the execution process of the real-time parallel inference unit is as follows:

[0090] Deploy the trained neural network model under the TFLite inference framework;

[0091] Based on the magnetic confinement fusion device, simultaneously obtain spectral data of multiple channels at each moment, and use the spectral data of these multi-channels as a batch of samples and input them into the trained neural network model under the TFLite inference framework;

[0092] Based on the trained neural network model, multi-channel real-time parallel inference of ion temperature is performed to calculate the real-time ion temperature.

[0093] Among them, the execution process of each unit can be carried out according to the process steps of the multi-channel ion temperature inference method for the magnetic confinement fusion device in Embodiment 1, and will not be elaborated one by one in this embodiment.

[0094] Meanwhile, the present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned multi-channel ion temperature inference method for the magnetic confinement fusion device.

[0095] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0096] The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of the flows and / or blocks in the flowchart and / or block diagram can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the specified functions in Figure 1 one or more processes and / or blocks Figure 1 one or more blocks.

[0097] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device implements the specified functions in Figure 1 one or more processes and / or blocks Figure 1 one or more blocks.

[0098] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, so that a series of operation steps are performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions for implementing the steps of the process Figure 1 one process or a plurality of processes and / or blocks Figure 1 steps for implementing the functions specified in one block or a plurality of blocks.

[0099] The specific embodiments described above further elaborate the objectives, technical solutions and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A multi-channel ion temperature inference method for a magnetic confinement nuclear fusion device, characterized in that: The method includes: Acquiring a charge exchange composite spectrum curve as input data; and performing data analysis on the input data to obtain ion temperature data as an ion temperature label; Constructing a historical experimental data set according to the input data and the ion temperature label; Constructing a neural network model, and using a historical experimental data set to perform model training on the neural network model to obtain a trained neural network model; The trained neural network model is deployed under the TFLite inference framework, and the spectral data acquired in real time from multiple channels are input into the trained neural network model to perform parallel inference of the real-time ion temperature of multiple channels to obtain the real-time ion temperature.

2. The multi-channel ion temperature inference method for a magnetic confinement nuclear fusion device according to claim 1, characterized in that: Acquiring a charge exchange composite spectral curve as input data; and performing data analysis on the input data to obtain ion temperature data as an ion temperature label, including: According to historical experimental data of the magnetic confinement nuclear fusion device, a charge exchange composite spectrum curve is obtained as input data; the input data refers to a spectrum with a dimension of 1×512; The input data is analyzed by the least square fitting method, and abnormal data is cleaned to obtain ion temperature data as an ion temperature label; the ion temperature label refers to a scalar number of the ion temperature at a set position in the plasma.

3. The multi-channel ion temperature inference method for a magnetic confinement nuclear fusion device according to claim 1, characterized in that: Each set of data in the historical experimental data set is represented by a spectrum with a dimension of 1×512 and an ion temperature label with a dimension of 1×1 inferred therefrom.

4. The multi-channel ion temperature inference method for a magnetic confinement nuclear fusion device according to claim 1, characterized in that: Build a neural network model, including: Input layer: The input layer size is 1×512; First convolutional neural network layer: construct the first convolutional neural network layer with 4 layers and 32 convolution kernels, and perform preliminary feature extraction on the input data based on the first convolutional neural network layer; at this time, the output matrix size of the neural network model is converted to 256×32; The second convolutional neural network layer and the third convolutional neural network layer: construct a 3-layer second convolutional neural network layer with 64 convolution kernels and a 3-layer third convolutional neural network layer with 128 convolution kernels; the first convolutional neural network layer, the second convolutional neural network layer and the third convolutional neural network layer are combined in the form of cross-layer residual connections to obtain the final neural network feature map; at this time, the output matrix size of the neural network model is converted to 12×128; Global pooling layer: construct a global pooling layer to average the final neural network feature map extracted by the neural network model over the entire spectral range; at this time, the output matrix size of the neural network model is converted to 1×128; Fully connected layer: Construct a fully connected layer with 256, 64, and 1 convolution kernels, transform the dimension of the neural network model output, and obtain the final ion temperature scalar output with a length of 1.

5. The multi-channel ion temperature inference method for a magnetic confinement nuclear fusion device according to claim 4, characterized in that: The training process of the neural network model is: The weight parameters of each convolutional neural network layer in the neural network model are obtained through training with the stochastic gradient descent method, and the network parameters are trained using the established historical experimental data set, that is, the relationship between the input data X and the label ion temperature Y in the historical experimental data is fitted to obtain a function F that satisfies Y=F(X), and a calculation model for inferring the ion temperature from the spectral data is obtained, and the calculation model of the ion temperature is used as the trained neural network model.

6. The multi-channel ion temperature inference method for a magnetic confinement nuclear fusion device according to claim 5, characterized in that: The specific steps of training the weight parameters are: Randomly obtain 128 spectral data and corresponding ion temperature labels from the historical experimental data set each time as a batch of data, and input them into the neural network model for calculation to obtain temperature results; The mean absolute error between the temperature result and the ion temperature label is used as the loss function, and the inference loss is minimized by the gradient descent method to update the weight parameters of the neural network model. The above process is iterated repeatedly until the loss function no longer decreases, and the weight parameter at this time is the final parameter.

7. The multi-channel ion temperature inference method for a magnetic confinement nuclear fusion device according to claim 1, characterized in that: The trained neural network model is deployed under the TFLite inference framework, and the spectral data acquired in real time by multiple channels is input into the trained neural network model to perform parallel inference of the real-time ion temperature of multiple channels to obtain the real-time ion temperature, including: Deploy the trained neural network model in the TFLite reasoning framework; Based on the magnetic confinement nuclear fusion device, the spectral data of multiple channels at each moment are simultaneously obtained, and the spectral data of multiple channels are taken as a batch of samples and input into the neural network model trained under the TFLite inference framework; Based on the trained neural network model, multi-channel real-time ion temperature parallel inference is performed to calculate the real-time ion temperature.

8. A multi-channel ion temperature inference system for a magnetic confinement nuclear fusion device, characterized in that: The system includes: An acquisition unit, used for acquiring a charge exchange composite spectrum curve as input data according to historical experimental data of a magnetic confinement nuclear fusion device; A label unit, used for performing data analysis on the input data to obtain ion temperature data as an ion temperature label; A data set construction unit, used to construct a historical experiment data set according to the input data and the ion temperature label; A model building and training unit, used to build a neural network model, and use a historical experimental data set to train the neural network model to obtain a trained neural network model; The real-time parallel inference unit is used to deploy the trained neural network model under the TFLite inference framework, and input the spectral data acquired in real time by multiple channels into the trained neural network model, perform multi-channel real-time ion temperature parallel inference, and obtain the real-time ion temperature.

9. The multi-channel ion temperature inference system for a magnetic confinement nuclear fusion device according to claim 8, characterized in that: The execution process of the real-time parallel inference unit is: Deploy the trained neural network model in the TFLite reasoning framework; Based on the magnetic confinement nuclear fusion device, the spectral data of multiple channels at each moment are simultaneously obtained, and the spectral data of multiple channels are taken as a batch of samples and input into the neural network model trained under the TFLite inference framework; Based on the trained neural network model, multi-channel real-time ion temperature parallel inference is performed to calculate the real-time ion temperature.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the multi-channel ion temperature inference method for a magnetic confinement nuclear fusion device as described in any one of claims 1 to 7 is implemented.

Images