Neutral beam power regulation model construction method and device, equipment and medium

Abstract

The invention discloses a neutral beam power regulation model construction method, apparatus and device, and a medium. The method comprises the steps of obtaining a historical discharge data set of a neutral beam injection experiment; and taking the power and the diversion coefficient in the historical discharge data set as input data, taking the set value of each path of power supply in the historical discharge data set as output data, and inputting the input data and the output data into a pre-constructed deep neural network model for training to obtain a neutral beam power regulation model. According to the scheme, the adjustment model for determining the target power and the setting parameters corresponding to the neutral beam can be provided, energy and time waste in the neutral beam injection process is avoided, and the working efficiency of the Tokamak device is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent control of large-scale electrophysical equipment, and in particular to a method, device, equipment and medium for constructing a neutral beam power regulation model. Background Art

[0002] The fully superconducting tokamak device is equipped with a neutral beam injection system. The main function of the neutral beam injection system is to inject a high-energy neutral beam into the tokamak device, causing it to collide with the background plasma to exchange energy, thereby achieving the purpose of heating the background plasma of the tokamak device. Neutral beam injection heating is the auxiliary heating method with the clearest physical mechanism and the most obvious heating effect among the tokamak heating methods.

[0003] Currently, focused long-pulse neutral beam injection (NBI) exhibits a strongly coupled, nonlinear relationship between the neutral beam setting parameters and the target power. For each NB injection, it's difficult to determine the corresponding neutral beam setting parameters for the target power. This can easily lead to the NB injection result not meeting the target power, resulting in wasted energy and time during the NB injection process, and impacting the efficiency of the tokamak. Summary of the Invention

[0004] In order to solve the above problems, the present invention proposes a method, device, equipment and medium for constructing a neutral beam power regulation model, and provides a regulation model for determining the set parameters corresponding to the target power and the neutral beam, thereby avoiding energy and time waste in the neutral beam injection process and improving the working efficiency of the tokamak device.

[0005] An embodiment of the present invention provides a method for constructing a neutral beam power adjustment model, the method comprising:

[0006] Obtain historical discharge datasets from neutral beam injection experiments;

[0007] The power and conductivity coefficient in the historical discharge data set are used as input data, and the set values ​​of each power supply in the historical discharge data set are used as output data. The input data and the output data are input into a pre-built deep neural network model for training to obtain a neutral beam power regulation model.

[0008] Preferably, each power supply includes a filament power supply, an arc power supply, a high voltage power supply and a suppressor power supply.

[0009] Preferably, the method further comprises:

[0010] Inputting the target power and the currently obtained conductivity coefficient into the neutral beam power regulation model, and outputting the set value of each power supply;

[0011] The neutral beam injection system is controlled according to the set values ​​of each power supply.

[0012] Preferably, after obtaining the historical discharge dataset of the neutral beam injection experiment, the method further comprises:

[0013] Performing data cleaning on the historical discharge data set to remove missing values ​​and outliers;

[0014] The cleaned data is normalized and the parameter dimensions are unified.

[0015] Preferably, the input data and the output data are input into a pre-built deep neural network model for training to obtain a neutral beam power regulation model, including:

[0016] Dividing the input data and the output data into a training set and a test set according to a preset ratio;

[0017] Performing dimensionless processing on the training set to obtain training samples;

[0018] Initialize the weight matrix and bias vector of each layer in the deep neural network model and set the initial model parameters;

[0019] Inputting the input data in the training sample into the deep neural network model for forward propagation, calculating through the hidden layer and output layer to obtain a predicted output; calculating the loss value between the predicted output and the corresponding output data in the training sample; performing backpropagation based on the loss value, recursively calculating the error of the output layer and hidden layer layer by layer, and correcting the corresponding weight matrix according to the error of each layer;

[0020] Performing dimensionless processing on the validation set to obtain a test sample, and calculating the validation error of the trained deep neural network model based on the test sample;

[0021] When the fluctuation of the verification error of N consecutive test samples is within a preset range, outputting the current deep neural network model as the neutral beam power adjustment model;

[0022] Otherwise, continue to use the training samples to train the deep neural network model.

[0023] Preferably, the hidden layer is 3 layers, and the number of nodes in the output layer is 4;

[0024] The predicted output of the input data of the pth training sample is

[0025] Among them, W (4) is the weight matrix of the output layer, b (4) is the bias vector of the output layer, a (l) is the output value of the lth hidden layer, l=1,2,3, a (3)=ReLU(z (3) ), z (l) is the linear output of the lth hidden layer, z (3) =W (3) a (2) +b (3) , W (l) is the weight matrix of the lth hidden layer, b (l) is the bias vector of the lth hidden layer, z (1) =W (1) x (p) +b (1) ,a (1) =ReLU(z (1) )=max(0,z (1) ), x (p) is the input data of the pth training sample.

[0026] Preferably, the error of the output layer is

[0027] The error δ of the lth hidden layer (l) =(W (l+1)T δ (l+1) )*ReLU′(z (l) );

[0028] Where B is the number of training samples, represents the prediction result of the input data in the training sample, Y′; represents the output data in the training sample; W (l+1) is the weight matrix of the next layer of the lth hidden layer, δ (l+1) is the error of the next layer of the lth hidden layer; define the function x is the input value for the defined function.

[0029] Yet another embodiment of the present invention provides a device for constructing a neutral beam power adjustment model, the device comprising:

[0030] Acquisition module, used to obtain historical discharge datasets of neutral beam injection experiments;

[0031] A training module is used to take the power and conductivity in the historical discharge data set as input data, take the set value of each power supply in the historical discharge data set as output data, input the input data and the output data into a pre-built deep neural network model for training, and obtain a neutral beam power regulation model.

[0032] Another embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the neutral beam power adjustment model construction method described in any one of the above embodiments is implemented.

[0033] Another embodiment of the present invention provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the neutral beam power adjustment model construction method described in any one of the above embodiments.

[0034] The present invention provides a method, apparatus, device, and medium for constructing a neutral beam power regulation model. The method obtains a historical discharge dataset from a neutral beam injection experiment; uses the power and conductivity in the historical discharge dataset as input data, uses the set values ​​of each power source in the historical discharge dataset as output data, and inputs the input and output data into a pre-constructed deep neural network model for training, thereby obtaining a neutral beam power regulation model. This application solution can provide a regulation model that determines the set parameters corresponding to the target power and the neutral beam, thereby avoiding energy and time waste during the neutral beam injection process and improving the operating efficiency of the tokamak device. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 1 is a flow chart of a method for constructing a neutral beam power adjustment model provided by an embodiment of the present invention;

[0036] Figure 2 is another flow chart of a method for constructing a neutral beam power adjustment model provided by an embodiment of the present invention;

[0037] Figure 3 is a schematic diagram of the structure of a deep neural network model provided by an embodiment of the present invention;

[0038] Figure 4 is a graph showing the mean square error of training data of a neutral beam power adjustment model provided by an embodiment of the present invention;

[0039] Figure 5 1 is a schematic diagram comparing the predicted and actual values ​​of the power setting values ​​of various power lines provided by an embodiment of the present invention;

[0040] Figure 6 1 is a schematic structural diagram of a neutral beam power regulation model construction device provided by an embodiment of the present invention;

[0041] Figure 7 It is a structural diagram of a terminal device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0043] This application provides a method for constructing a neutral beam power regulation model, see Figure 1 , is a flow chart of a method for constructing a neutral beam power adjustment model provided by an embodiment of the present invention, the method comprising steps S1 to S2:

[0044] Step S1, obtaining a historical discharge data set of a neutral beam injection experiment;

[0045] Step S2: using the power and conductivity in the historical discharge data set as input data, and the set value of each power supply in the historical discharge data set as output data, inputting the input data and the output data into a pre-built deep neural network model for training to obtain a neutral beam power regulation model.

[0046] During the specific implementation of this embodiment, the data preparation stage obtains a historical discharge data set from the neutral beam injection experiment database; collects historical discharge data of the neutral beam injection experiment, filters out complete samples including power, conductivity and setting values ​​of each power supply, and constructs a historical discharge data set.

[0047] It should be noted that after obtaining the data, normalization processing can be performed, that is, the input data and output data are normalized and mapped to [0, 1] or standardized to outputs with mean 0 and variance 1, to avoid the influence of dimensional differences on training and also facilitate the stable calculation of activation functions;

[0048] A deep neural network model is constructed. The deep neural network model includes an input layer, a hidden layer, and an output layer. The weights and parameters of the deep neural network model are randomly initialized.

[0049] During model training, the power and conductance coefficient in the historical discharge data set are used as input data, and the set values ​​of each power supply in the historical discharge data set are used as output data. The training sample is input, and after linear transformation and ReLU activation of the hidden layer, the output layer calculates the power supply set value prediction.

[0050] The hidden layer error is calculated layer by layer, and the weights and biases are updated based on the error to iteratively optimize the model.

[0051] It should be noted that when this case is implemented, the model is also evaluated using the validation set. If the validation set loss no longer decreases, for example, if overfitting occurs, such as when the training set loss continues to decrease and the validation set loss increases, the learning rate is adjusted and the optimal model is saved.

[0052] According to the neutral beam power regulation model, by inputting the target power and conductance coefficient, the set values ​​of each output power supply can be calculated to assist in adjusting the experimental parameters.

[0053] By learning from historical data patterns, the model can accurately predict power supply setpoints corresponding to power and conductivity, reducing manual debugging and adapting to the complex multi-physics coupling scenarios of neutral beam injection experiments. This improves the precision of neutral beam injection parameter control and reduces experimental trial-and-error costs. This avoids energy and time waste during the neutral beam injection process, improving the efficiency of the tokamak.

[0054] In another embodiment provided by the present invention, each power supply includes a filament power supply, an arc power supply, a high-voltage power supply, and a suppressor power supply.

[0055] During the specific implementation of this embodiment, during neutral beam injection, the outputs of the filament power supply, arc power supply, high-voltage power supply, and suppressor power supply directly affect the physical processes of beam generation, acceleration, and transmission: filament power supply: heating the filament to generate thermal electrons, providing initial particles for arc discharge; arc power supply: maintaining arc chamber plasma discharge, generating high-density ions; high-voltage power supply: accelerating ions to form a high-energy beam; suppressor power supply: suppressing secondary electron emission, thereby improving beam transmission efficiency.

[0056] The set value of the power supply has a strong nonlinear coupling with the neutral beam output power target and conductivity coefficient input into the experiment, and is affected by complex factors such as plasma collision, electric field distribution, and equipment characteristics.

[0057] By exploring the complex relationship between the power conduction coefficient and the four power supplies, data-driven coordinated control of multiple power supplies can be achieved, which not only meets the physical requirements of neutral beam injection, but also greatly improves experimental efficiency and control accuracy.

[0058] In another embodiment of the present invention, the method further includes:

[0059] Inputting the target power and the currently obtained conductivity coefficient into the neutral beam power regulation model, and outputting the set value of each power supply;

[0060] The neutral beam injection system is controlled according to the set values ​​of each power supply.

[0061] During the specific implementation of this embodiment, historical discharge data is used to train a deep neural network to construct a neutral beam power regulation model, and then precise control of the neutral beam injection system is achieved through model prediction.

[0062] During model prediction, the set value of each power supply is determined based on the target power and conductance coefficient.

[0063] The target power, representing the energy level expected to be output by the neutral beam injection system, is the core indicator of operational requirements, that is, the power target for plasma heating in the tokamak device.

[0064] The conductivity coefficient reflects the transmission characteristics of the plasma beam during the generation and transmission of the neutral beam. It is strongly correlated with physical processes such as particle density and collision frequency, and is a key state quantity that affects the adaptability of power supply parameters.

[0065] Through training, the deep neural network learns the nonlinear mapping relationship between target power, permeability, and the set values ​​for the filament power supply, arc power supply, high-voltage power supply, and suppressor power supply. (For example, when power demand is high, the high-voltage power supply needs to increase the accelerating voltage, while the filament power supply needs to adapt the heating current to maintain the arc discharge plasma density.) After inputting the target power and current permeability, the model outputs precise set values ​​for the four power supplies based on the learned patterns, achieving precise control of the control parameters.

[0066] Based on the four power supply set values ​​output by the model, the power supply hardware of the neutral beam injection system is directly regulated, such as by issuing instructions through control systems such as PLC and DCS, so that the power supply output matches the current, voltage and other parameters predicted by the model, thereby allowing the neutral beam injection system to operate stably at the target power.

[0067] In another embodiment provided by the present invention, after obtaining a historical discharge dataset of a neutral beam injection experiment, the method further includes:

[0068] Performing data cleaning on the historical discharge data set to remove missing values ​​and outliers;

[0069] The cleaned data is normalized and the parameter dimensions are unified.

[0070] In the specific implementation of this embodiment, after obtaining the historical discharge data set, the historical discharge data set is cleaned and standardized;

[0071] The data cleaning is used to remove missing values ​​and outliers, and the standardization process uses Min-Max normalization to eliminate dimensional differences in various parameters.

[0072] Dimensionlessization makes the impact of different features (such as power and conductivity) on the model more balanced, avoiding the situation where features with large values ​​suppress features with small values.

[0073] In another embodiment provided by the present invention, the input data and the output data are input into a pre-built deep neural network model for training to obtain a neutral beam power adjustment model, including:

[0074] Dividing the input data and the output data into a training set and a test set according to a preset ratio;

[0075] Performing dimensionless processing on the training set to obtain training samples;

[0076] Initialize the weight matrix and bias vector of each layer in the deep neural network model and set the initial model parameters;

[0077] Inputting the input data in the training sample into the deep neural network model for forward propagation, calculating through the hidden layer and output layer to obtain a predicted output; calculating the loss value between the predicted output and the corresponding output data in the training sample; performing backpropagation based on the loss value, recursively calculating the error of the output layer and hidden layer layer by layer, and correcting the corresponding weight matrix according to the error of each layer;

[0078] Performing dimensionless processing on the validation set to obtain a test sample, and calculating the validation error of the trained deep neural network model based on the test sample;

[0079] When the fluctuation of the verification error of N consecutive test samples is within a preset range, outputting the current deep neural network model as the neutral beam power adjustment model;

[0080] Otherwise, continue to use the training samples to train the deep neural network model.

[0081] When implementing this embodiment, see Figure 2 , is another flow chart of the method for constructing a neutral beam power adjustment model provided by an embodiment of the present invention.

[0082] The deep neural network model is a five-layer fully connected network, including an input layer, a hidden layer, and an output layer. The input layer has 2 nodes, the output layer has 4 nodes, and the hidden layers have 64, 48, and 32 nodes respectively.

[0083] The model input parameters are set as power and conductance, and the output is the set values ​​of the four power supplies: arc, filament, high voltage, and suppressor. The network structure, learning rate, training rounds and other hyperparameters are automatically adjusted according to the experimental data to ensure the convergence and generalization ability of the model on the validation set.

[0084] Split the training set and test set from the historical discharge dataset according to a preset ratio (such as 7:3, 8:2, etc.);

[0085] The training set and test set are dimensionless. This eliminates dimension differences (such as power unit kW and conductivity coefficient dimensionless) to make model training more stable. The training set input X and output Y are respectively:

[0086] Perform dimensionless processing, where μ and σ are the mean and standard deviation of the corresponding dimensions respectively.

[0087] Splitting the dataset avoids model overfitting, and dimensionless transformation makes the impact of different features on model updates more balanced, thereby improving training efficiency.

[0088] Randomly initialize the weight matrix W of each layer of the deep neural network (l) With the bias vector b (l) , and set training parameters such as learning rate η, number of iterations, early stopping threshold N, etc.

[0089] For example, set the learning rate α = 5×10-4, Adam hyperparameters β1 = 0.9, β2 = 0.999, ε = 10 -8 , batch size B=32, maximum round Emax=200.

[0090] The model is given an initial state through random initialization, and then gradually optimized through backpropagation. The parameter configuration determines the training rhythm.

[0091] The input data of the training set (power and conductivity after dimensionless conversion) is input into the model. After linear transformation of the hidden layer and ReLU activation, the predicted power setting value is obtained at the output layer.

[0092] Use a loss function, such as mean square error loss, to quantify the difference between the predicted value and the actual power setting value;

[0093] Starting from the output layer, reversely calculate the error δ of each layer, and then use the error to update the weights and biases.

[0094] The dimensionless test set is input into the model being trained, and the validation error is calculated using the same loss calculation logic as the training set. The validation error is continuously monitored. When the validation error fluctuations for N consecutive test samples are very small (e.g., the change is <0.001), the model is considered to have converged and training is stopped. Otherwise, the iteration is continued.

[0095] The validation set monitors the model for overfitting, and an early stopping mechanism prevents excessive model iterations that can lead to poor generalization, allowing the optimal model to be saved promptly. When the early stopping condition is met (i.e., when the error is stable after N consecutive test samples), the currently trained deep neural network is saved as the neutral beam power regulation model for actual control.

[0096] Output a usable model to prepare for subsequent control of the neutral beam injection system.

[0097] By splitting the training and test sets and combining them with early stopping training, the model's generalization capability is improved, ensuring accurate prediction of power supply setpoints even in actual neutral beam experiments. Backpropagation uses mathematical derivation to precisely adjust weights, significantly reducing model training time.

[0098] In another embodiment provided by the present invention, the number of hidden layers is 3, and the number of nodes in the output layer is 4;

[0099] The predicted output of the input data of the pth training sample is

[0100] Among them, W (4) is the weight matrix of the output layer, b (4) is the bias vector of the output layer, a (l) is the output value of the lth hidden layer, l=1,2,3, a (3) =ReLU(z (3) ), z (l) is the linear output of the lth hidden layer, z (3) =W (3) a (2) +b (3) , W (l) is the weight matrix of the lth hidden layer, b (l) is the bias vector of the lth hidden layer, z (1) =W (1) x (p) +b (1) ,a (1) =ReLU(z (1) )=max(0,z (1) ), x (p) is the input data of the pth training sample.

[0101] When this embodiment is implemented, Figure 3 , is a schematic diagram of the structure of a deep neural network model provided by an embodiment of the present invention. The deep neural network model is a five-layer model. The network input is normalized power and conductance. There are three hidden layers and four output layer nodes, corresponding to the set values ​​of the arc power supply, filament power supply, high voltage power supply, and suppressor power supply.

[0102] The mean square error curve of the training data of the neutral beam power regulation model designed according to the above structure is as follows: Figure 4 The final test data prediction results are shown as follows Figure 5 shown.

[0103] During forward propagation, for the p-th training sample (x (p) ,y (p) ):

[0104] Hidden layer 1 output: z (1) =W (1) x (p) +b (1) ,a (1) =ReLU(z(1) )=max(0,z (1) );

[0105] Hidden layer 2 output: z (2) =W (2) a (1) +b (2) ,a (2) =ReLU(z (2) );

[0106] Hidden layer 3 output: z (3) =W (3) a (2) +b (3) ,a (3) =ReLU(z (3) );

[0107] Output layer prediction:

[0108] Among them, W (4) is the weight matrix of the output layer, b (4) is the bias vector of the output layer, a (l) is the output value of the lth hidden layer, l=1,2,3,z (l) is the linear output of the lth hidden layer, W (l) is the weight matrix of the lth hidden layer, b (l) is the bias vector of the lth hidden layer, x (p) is the input data of the pth training sample.

[0109] In another embodiment provided by the present invention, the error of the output layer is

[0110] The error δ of the lth hidden layer (l) =(W (l+1)T δ (l+1) )*ReLU′(z (l) );

[0111] Where B is the number of training samples, represents the prediction result of the input data in the training sample, Y′ represents the output data in the training sample; W (l+1) is the weight matrix of the next layer of the lth hidden layer, δ (l+1) is the error of the next layer of the lth hidden layer; define the function x is the input value for the defined function.

[0112] In the specific implementation of this embodiment, when calculating the loss, the mean square error loss is calculated for the current batch of samples:

[0113] Through back propagation, the output layer error

[0114] The hidden layer error is recursively calculated layer by layer: the error δ of the lth hidden layer (l) =(W (l+1)T δ (l+1) )*ReLU′(z (l) );

[0115] Where B is the number of training samples, represents the prediction result of the input data in the training sample, Y′ represents the output data in the training sample; W (l+1) is the weight matrix of the next layer of the lth hidden layer, δ (l+1) is the error of the next layer of the lth hidden layer; define the function x is the input value for the defined function.

[0116] The neutral beam power regulation model construction scheme provided in this application can, during the experimental operation, enable the host computer to call the neutral beam power regulation model according to the current target power and conductance coefficient, and the FPGA to instantly output the corresponding control signal to drive each power module to complete a fast and stable power adjustment process, thereby ensuring the stable operation of the neutral beam injection system under complex working conditions.

[0117] Another embodiment of the present invention provides a device for constructing a neutral beam power adjustment model, see Figure 6 , is a schematic structural diagram of a device for constructing a neutral beam power adjustment model provided by an embodiment of the present invention, the device comprising:

[0118] Acquisition module, used to obtain historical discharge datasets of neutral beam injection experiments;

[0119] A training module is used to take the power and conductivity in the historical discharge data set as input data, take the set value of each power supply in the historical discharge data set as output data, input the input data and the output data into a pre-built deep neural network model for training, and obtain a neutral beam power regulation model.

[0120] The neutral beam power regulation model construction device provided in this embodiment can execute all the steps and functions of the neutral beam power regulation model construction method provided in any of the above embodiments, and the specific functions of the device are not described in detail here.

[0121] See also Figure 7, is a schematic diagram of the structure of a terminal device provided by an embodiment of the present invention. The terminal device includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a neutral beam power adjustment model construction program. When the processor executes the computer program, the steps of each of the above-mentioned neutral beam power adjustment model construction method embodiments are implemented, such as Figure 1 Alternatively, the processor implements the functions of the modules in the above-mentioned device embodiments when executing the computer program.

[0122] Exemplarily, the computer program can be divided into one or more modules, and the one or more modules are stored in the memory and executed by the processor to complete the present invention. The one or more modules can be a series of computer program instruction segments that can perform specific functions, and the instruction segments are used to describe the execution process of the computer program in the neutral beam power regulation model construction device. For example, the computer program can be divided into a detection module, an output power control module, and a window control module. The specific functions of each module have been described in detail in the neutral beam power regulation model construction method provided in any of the above embodiments, and the specific functions of the device will not be repeated here.

[0123] The neutral beam power regulation model construction device can be a computing device such as a desktop computer, laptop, PDA, or cloud server. The neutral beam power regulation model construction device can include, but is not limited to, a processor and a memory. Those skilled in the art will appreciate that the schematic diagram is merely an example of a neutral beam power regulation model construction device and does not constitute a limitation on the neutral beam power regulation model construction device. The device can include more or fewer components than shown, or a combination of certain components, or different components. For example, the neutral beam power regulation model construction device can also include input and output devices, network access devices, buses, etc.

[0124] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the neutral beam power regulation model construction device, and utilizes various interfaces and lines to connect various parts of the entire neutral beam power regulation model construction device.

[0125] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the neutral beam power adjustment model construction device by running or executing the computer program and / or module stored in the memory and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function (such as a sound playback function, an image playback function, etc.); the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.). In addition, the memory can include a high-speed random access memory and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0126] Wherein, if the module integrated in the neutral beam power regulation model construction device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, it can implement the steps of the above-mentioned various method embodiments. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device that can carry the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0127] It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for constructing a neutral beam power regulation model, characterized in that: The method comprises: Obtain historical discharge datasets from neutral beam injection experiments; The power and conductivity coefficient in the historical discharge data set are used as input data, and the set values ​​of each power supply in the historical discharge data set are used as output data. The input data and the output data are input into a pre-built deep neural network model for training to obtain a neutral beam power regulation model.

2. The method for constructing a neutral beam power regulation model according to claim 1, wherein: The power supplies include filament power supply, arc power supply, high voltage power supply and suppressor power supply.

3. The method for constructing a neutral beam power regulation model according to claim 1, wherein: The method further comprises: Inputting the target power and the currently obtained conductivity coefficient into the neutral beam power regulation model, and outputting the set value of each power supply; The neutral beam injection system is controlled according to the set values ​​of each power supply.

4. The method for constructing a neutral beam power regulation model according to claim 1, wherein: After obtaining a historical discharge dataset of a neutral beam injection experiment, the method further includes: Performing data cleaning on the historical discharge data set to remove missing values ​​and outliers; The cleaned data is normalized and the parameter dimensions are unified.

5. The method for constructing a neutral beam power regulation model according to claim 1, wherein: Inputting the input data and the output data into a pre-built deep neural network model for training to obtain a neutral beam power adjustment model, including: Dividing the input data and the output data into a training set and a test set according to a preset ratio; Performing dimensionless processing on the training set to obtain training samples; Initialize the weight matrix and bias vector of each layer in the deep neural network model and set the initial model parameters; Inputting the input data in the training sample into the deep neural network model for forward propagation, calculating through the hidden layer and output layer to obtain a predicted output; calculating the loss value between the predicted output and the corresponding output data in the training sample; performing backpropagation based on the loss value, recursively calculating the error of the output layer and hidden layer layer by layer, and correcting the corresponding weight matrix according to the error of each layer; Performing dimensionless processing on the validation set to obtain a test sample, and calculating the validation error of the trained deep neural network model based on the test sample; When the fluctuation of the verification error of N consecutive test samples is within a preset range, outputting the current deep neural network model as the neutral beam power adjustment model; Otherwise, continue to use the training samples to train the deep neural network model.

6. The method for constructing a neutral beam power regulation model according to claim 5, characterized in that: The number of hidden layers is 3, and the number of nodes in the output layer is 4; The predicted output of the input data of the pth training sample is Among them, W (4) is the weight matrix of the output layer, b (4) is the bias vector of the output layer, a (l) is the output value of the lth hidden layer, l=1,2,3, a (3) =ReLU(z (3) ), z (l) is the linear output of the lth hidden layer, z (3) =W (3) a (2) +b (3) , W (l) is the weight matrix of the lth hidden layer, b (l) is the bias vector of the lth hidden layer, z (1) =W (1) x (p) +b (1) ,a (1) =ReLU(z (1) )=max(0,z (1) ), x (p) is the input data of the pth training sample.

7. The method for constructing a neutral beam power regulation model according to claim 5, wherein: The error of the output layer is The error δ of the lth hidden layer (l) =(W (l+1)T δ (l+1) )*ReLU′(z (l) ); Where B is the number of training samples, represents the prediction result of the input data in the training sample, Y′; represents the output data in the training sample; W (l+1) is the weight matrix of the next layer of the lth hidden layer, δ (l+1) is the error of the next layer of the lth hidden layer; define the function x is the input value for the defined function.

8. A neutral beam power regulation model construction device, characterized in that: The device comprises: Acquisition module, used to obtain historical discharge datasets of neutral beam injection experiments; A training module is used to take the power and conductivity in the historical discharge data set as input data, take the set value of each power supply in the historical discharge data set as output data, input the input data and the output data into a pre-built deep neural network model for training, and obtain a neutral beam power regulation model.

9. A terminal device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the method for constructing a neutral beam power regulation model according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the neutral beam power adjustment model construction method according to any one of claims 1 to 7.

Images