Induction eddy current prediction model construction method, prediction method and electronic device

Abstract

The present application discloses a method for constructing an induced eddy current prediction model, an induced eddy current prediction method and an electronic device, which are suitable for predicting the induced eddy current of the vacuum wall in a magnetic confinement nuclear fusion device, wherein the model construction method includes: obtaining multiple groups of diagnostic signals in the historical discharge process to construct a training set; inputting the data samples in the training set into the initial prediction model to obtain predicted induced eddy current data; mapping the diagnostic signal and the predicted induced eddy current data to uniform variable parameters respectively, determining the error loss between the two groups of uniform variable parameters, adjusting the parameters of the initial prediction model based on the error loss, until the error loss reaches the preset expected value, and obtaining a trained induced eddy current prediction model. The training data are all derived from the real diagnostic data in the discharge process, have prediction accuracy, and when the eddy current induction prediction model trained by the training method is applied, the computing power requirement is low, and long-term coverage prediction can be achieved.

Description

Technical Field

[0001] The present application relates to the field of nuclear fusion technology, and in particular to a method for constructing an induced eddy current prediction model, a prediction method and an electronic device. Background Art

[0002] A magnetic confinement nuclear fusion device is a device that uses a magnetic field to confine and control high-temperature plasma to achieve controlled nuclear fusion. Taking the tokamak device as an example, the vacuum wall of the tokamak device is usually made of metal materials. During the operation of the tokamak device, the changing currents in various coils (for example, poloidal field coils, toroidal field coils, central solenoid coils, etc.) will generate induced eddy currents on the vacuum wall. The existence of induced eddy currents has many effects on the operation of the device, such as changing the shape of the breakdown area, changing the plasma configuration, and introducing errors into the acquisition results of various diagnostic measurement equipment. Therefore, quantitative analysis of the distribution of these induced eddy currents, especially the rapid acquisition of the distribution of induced eddy currents during the discharge operation of the device, is of great significance for the normal operation of the tokamak device and the study of many physical problems.

[0003] However, due to the continuous and closed structure of the vacuum wall, the distribution of the induced eddy currents is difficult to obtain through actual measurement. Therefore, in the related art, the induced eddy currents on the vacuum wall are often obtained through computer simulation.

[0004] In one of the related technologies, finite element simulation is used for three-dimensional simulation. For example, analysis software such as AnsysMaxwell and COMSOL is used. Based on the three-dimensional mechanical structure and boundary conditions of the vacuum wall, combined with the discharge configuration during device operation, the current waveform in each active coil is set to calculate the distribution of the induced eddy current. However, three-dimensional simulation requires huge computer computing power and is very time-consuming. For example, if a simulation is only performed for a certain instantaneous scene, it will take several days to achieve a high degree of accuracy. Therefore, it is difficult to conduct long-term coverage analysis of the eddy current distribution under different discharge configurations, and it is difficult to meet the needs of real-time analysis and real-time feedback control of the millisecond-level discharge window of the tokamak device.

[0005] In another related technology, the vacuum wall is constructed as a ring current wire model of a combination of many sufficiently thin circular rings, in which each equivalent circular ring is a ring current wire. In its main idea, it is necessary to ignore the annular asymmetry caused by the existence of each window on the vacuum chamber, so that the equivalent ring current wire has a common normal axis, and the shape and size of the cross section of each ring current wire does not change with the change of the annular coordinates, so that it can be approximately considered that the eddy current runs its own way in each ring current wire, and there is no current transmission between each ring current wire and other adjacent ring current wires. The size of the eddy current inside each ring current wire is annularly equal, thereby simplifying the three-dimensional problem to two dimensions. Based on this, the mutual inductance matrix between the active coil and the vacuum wall can be calculated, and the actual current waveform in the active coil can be obtained through experimental acquisition. By numerically solving the circuit equation of the system, the time-space distribution function of the eddy current can be obtained. However, when the induced eddy current obtained by the solution is multiplied by the mutual inductance matrix between the toroidal current wire and the magnetic diagnostic device, the magnetic flux share contributed by the induced eddy current to the magnetic diagnostic device is significantly different from the actual magnetic flux share contributed by the eddy current obtained by subtracting the magnetic flux share contributed by the active coil from the total magnetic flux. In addition, the actual time consumed in a typical solution is still in the order of minutes, which is too long compared to the discharge window of milliseconds. Therefore, the solution response speed cannot meet the needs of real-time analysis and real-time feedback control.

[0006] Therefore, how to determine the induced eddy current in the discharge process faster and more accurately becomes a technical problem that needs to be solved urgently. Summary of the invention

[0007] The present application provides an induced eddy current prediction model construction method, an induced eddy current prediction method and an electronic device, so as to at least solve the technical problem existing in the related art of how to determine the induced eddy current in the discharge process more quickly and accurately.

[0008] According to a first aspect of the present application, a method for constructing an induced eddy current prediction model for a magnetic confinement nuclear fusion device is provided, which is suitable for predicting induced eddy currents of a vacuum wall of a magnetic confinement nuclear fusion device, and comprises: acquiring multiple groups of diagnostic signals in a historical discharge process of the magnetic confinement nuclear fusion device to construct a training set, wherein the diagnostic signals comprise current data of multiple active coils in the magnetic confinement nuclear fusion device and total magnetic flux data of each magnetic diagnostic device in multiple magnetic diagnostic devices; inputting data samples in the training set into an initial prediction model to obtain predicted induced eddy current data; respectively mapping the diagnostic signals and the predicted induced eddy current data to uniform variable parameters, determining an error loss between two groups of the uniform variable parameters, and adjusting the parameters of the initial prediction model based on the error loss until the error loss reaches a preset expected value, thereby obtaining a trained induced eddy current prediction model.

[0009] According to a second aspect, an embodiment of the present application provides an induced eddy current prediction method, characterized in that:

[0010] Acquire the diagnostic data of the active coil within a preset time window corresponding to the time to be predicted during the discharge process to construct input features;

[0011] The input features are input into an induced eddy current prediction model constructed according to the induced eddy current prediction model construction method described in any one of the first aspects above, to obtain predicted induced eddy currents of the vacuum wall of the magnetic confinement nuclear fusion device during the discharge process.

[0012] According to a third aspect, an embodiment of the present application provides an electronic device, comprising: a processor and a memory; the memory is used to store a computer program; the processor is used to execute the computer program and implement the induced eddy current prediction model construction method described in any one of the first aspects and / or execute the induced eddy current prediction method described in the second aspect when executing the computer program.

[0013] Compared with the prior art, this application has at least the following effective effects:

[0014] The model training method of the present application obtains multiple groups of diagnostic signals in the historical discharge process of the magnetic confinement nuclear fusion device to construct a training set, wherein the diagnostic signals include current data of multiple active coils in the magnetic confinement nuclear fusion device and total magnetic flux data of multiple magnetic diagnostic equipment; inputs the data samples in the training set into the initial prediction model to obtain predicted induced eddy current data; maps the diagnostic signals and the predicted induced eddy current data to uniform variable parameters respectively, determines the error loss between the two groups of the uniform variable parameters, adjusts the parameters of the initial prediction model based on the error loss until the error loss reaches a preset expected value, and obtains a trained induced eddy current prediction model. The training data of the induced eddy current prediction model all come from the real diagnostic data during the discharge process of the magnetic confinement nuclear fusion device. During the training process, the diagnostic signal and the predicted induced eddy current data are respectively mapped to unified variable parameters to judge the accuracy of the predicted value, so that the model can better capture the relationship between the induced eddy current and the diagnostic data. To a certain extent, it avoids the solution error caused by the neglect or approximate processing of some entity details in the process of establishing the physical model in the prior art, and shows a higher prediction accuracy. At the same time, when the eddy current induction prediction model trained by the training method is used, the computing power requirement is low. Therefore, the diagnostic data in the discharge process can be processed in real time, and the induced eddy current can be predicted at any time in the millisecond discharge window, so it can have a long-term coverage prediction.

[0015] Furthermore, the induced eddy current prediction model adopts at least two layers of stacked recurrent neural networks to capture the temporal distribution relationship of the induced eddy currents in the training samples, and uses tanh as the activation function to cope with the possible drastic changes of the induced eddy currents, which is conducive to the convergence of model training and improves the accuracy of the final prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0017] Figure 1 A schematic diagram of a process flow of a method for constructing an induced eddy current prediction model for a vacuum wall of a magnetic confinement nuclear fusion device provided in this application;

[0018] Figure 2 A schematic diagram comparing the activation function in this implementation provided by this application with other activation functions;

[0019] Figure 3 A flow chart of a method for determining an error loss between a true value and a predicted value provided in the present application;

[0020] Figure 4 A flow chart of another method for determining the error loss between a true value and a predicted value provided in the present application;

[0021] Figure 5 A schematic diagram of the process flow of the induced eddy current prediction method provided in this application;

[0022] Figure 6 A schematic diagram of the prediction effect of the eddy current induction prediction method provided in this application;

[0023] Figure 7 Schematic diagram of an electronic device provided for this application. DETAILED DESCRIPTION

[0024] In order to more clearly illustrate the overall concept of the present application, a detailed description is given below in an illustrative manner in conjunction with the accompanying drawings.

[0025] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0026] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0027] The present application proposes a method for constructing an induced eddy current prediction model, which is used to predict the induced eddy currents induced on the vacuum wall during the discharge process of a magnetic confinement nuclear fusion device. Figure 1 As shown, the method for constructing an induced eddy current prediction model may include the following steps:

[0028] S10, acquiring multiple groups of diagnostic signals in the historical discharge process of the magnetic confinement nuclear fusion device to construct a training set. The magnetic confinement nuclear fusion device may include a tokamak, a stellarator and a magnetic mirror. In this embodiment, the tokamak is used as an example for description.

[0029] In this embodiment, the diagnostic signal is a working parameter used by the diagnostic device to detect the discharge process, and may include a coil current signal of the active coil and a magnetic flux of the magnetic diagnostic device.

[0030] It should be noted that the Tokamak contains various types of coils, such as the Poloidal Field (PF) coil, the Toroidal Field (TF) coil, the Center Solenoid (CS) coil, and the Merging-Compression (MC) coil. Since the current of the TF coil can be almost considered as a constant value during the discharge of the Tokamak device, the induced eddy current it generates is relatively weak. In addition, the arrangement direction of the TF coil is perpendicular to other coils, and the induced eddy currents it generates are also distributed in other directions. Therefore, the induced eddy current prediction model involved in this application does not need to consider the impact of the TF coil. In this case, the active coils in this application mainly include the center solenoid coil, the poloidal field coil and the MC coil. The magnetic diagnostic equipment may include a magnetic probe and a flux ring.

[0031] Since the induced eddy current on the vacuum wall is caused by the current change in the active coil during the discharge process, and the current in the active coil and the induced eddy current will both contribute to the magnetic flux share of the magnetic diagnostic device, the total magnetic flux of the magnetic diagnostic device minus the magnetic flux share contributed by the active coil can obtain the magnetic flux share contributed by the induced eddy current, and based on the relationship between the magnetic flux contributed by the induced eddy current and the induced eddy current, the induced eddy current can be obtained. Therefore, in this embodiment, when constructing the training sample, the current data of the active coil and the total magnetic flux of the magnetic diagnostic device are used for construction. After obtaining the current data of the active coil and the original signal of the magnetic diagnostic device, they are preprocessed. Specifically, the current signals of multiple active coils are subjected to noise reduction processing using a sliding window average filter, and the original signals of multiple magnetic diagnostic devices are integrated to obtain the total magnetic flux of each magnetic diagnostic device jointly acted by the active coil and the vacuum wall.

[0032] In this embodiment, multiple diagnostic data samples can be obtained from the MDSplus database of the tokamak discharge operation, and the diagnostic data samples can be the current data of the active coil at different times under the discharge configuration and the total magnetic flux data of the magnetic diagnostic equipment. In this embodiment, the current data of 9 active coils including 3 central solenoid coils, 5 poloidal field coils and 1 MC coil can be obtained, and the total magnetic flux data of 64 magnetic diagnostic equipment including 49 magnetic probes and 15 flux rings can be obtained.

[0033] For a diagnostic data sample, a target time for training the model to predict the induced eddy current is selected, and a preset time length is traced back before the selected target time as a preset time window, the target time is within the discharge time window, and the preset time length is less than the discharge time window. In this embodiment, n milliseconds can be traced back based on the target time, and n milliseconds can be used as the preset time window. The preset window time can be equally spaced into m segments for sampling, together with the two moments at the beginning and end of the time window, for a total of m+1 sampling moments.

[0034] The interpolation functions corresponding to the current data and the total magnetic flux data are used to interpolate the m+1 sampling moments, and (m+1)×9 coil current interpolation results and (m+1)×64 total magnetic flux interpolation results of the magnetic diagnostic equipment are obtained. In this embodiment, the coil current interpolation results and the total magnetic flux interpolation results can be arranged in the form of a matrix or in the form of a sequence. For the convenience of calculation, the matrix arrangement is used as an example for explanation in this embodiment.

[0035] The coil current interpolation result and the total magnetic flux interpolation result are combined to obtain a matrix sample corresponding to a diagnostic data sample. All diagnostic data samples are traversed according to the method for constructing matrix samples described in the above embodiment to obtain matrix samples corresponding to all diagnostic data samples to construct a training set. Correspondingly, the data samples in the training set can be arranged in the form of a matrix or in the form of a sequence. For the convenience of calculation, the matrix arrangement is used as an example for explanation in this embodiment.

[0036] S20, input the data samples in the training set into the initial prediction model to obtain the predicted induced eddy current data. The initial prediction model can adopt a recurrent neural network model. In this embodiment, the recurrent neural network has at least two stacked recursive layers. Exemplarily, two recursive layers are used as an example for explanation, and more stacked recursive layers are also applicable in this embodiment. Among them, the second recurrent neural network receives the output of the first recurrent neural network and calculates the value of the predicted induced eddy current at m+1 sampling moments in the window.

[0037] The nonlinear activation function uses tanh, and its expression is: Image as Figure 2 As shown in the figure. Compared with the sigmoid function, the tanh activation function is centered at 0, can achieve positive and negative balance, and has a wider output range, and is not easy to fall into the saturation area where the gradient is very small; compared with the ReLU function, it avoids the instability of the gradient and will not make the output less than 0 become a gradient value of 0, causing the weight to be unable to update normally. Considering the characteristics of the induced eddy current that may change dramatically, the characteristics of the tanh function are conducive to the convergence of model training and improve the accuracy of the final prediction.

[0038] During the training process, in each iterative training process, the data samples of the training set can be input into the initial model at one time, the model parameters can be updated, and the iterative training can be repeated multiple times until the initial model converges.

[0039] In order to improve training efficiency and reduce memory requirements, in one embodiment, the data samples in the training set can also be divided into multiple batches, each batch includes multiple data samples. During training, the model parameters are updated once for each batch, all batches are trained, the first iteration training is completed, and the iteration training is repeated multiple times until the initial model converges. In this embodiment, for the division of batches, a preset number of data samples can be randomly selected as a batch, and the data samples can be divided according to the time of collection, and can also be divided according to the source of the data samples from the fusion device.

[0040] S30, mapping the diagnostic signal and the predicted induced eddy current data to uniform variable parameters respectively, determining the error loss between the two sets of uniform variable parameters, adjusting the parameters of the initial prediction model based on the error loss until the error loss reaches a preset expected value, and obtaining a trained induced eddy current prediction model. In this embodiment, due to the continuous and closed structure of the vacuum wall, the distribution of its induced eddy currents is difficult to obtain through actual measurement. In the model training process, the error between the true value and the predicted value cannot be used to adjust the model. Therefore, in this embodiment, the predicted value and the training data sample can be mapped to the intermediate uniform variable parameters, and the uniform variable parameters corresponding to the two can be used to determine the error loss, and then the model training can be performed.

[0041] A training set is constructed by acquiring multiple groups of diagnostic signals in the historical discharge process of a magnetic confinement nuclear fusion device, wherein the diagnostic signals include current data of multiple active coils in the magnetic confinement nuclear fusion device and total magnetic flux data of multiple magnetic diagnostic equipment; the data samples in the training set are input into an initial prediction model to obtain predicted induced eddy current data; the diagnostic signals and the predicted induced eddy current data are respectively mapped to uniform variable parameters, the error loss between the two groups of uniform variable parameters is determined, and the parameters of the initial prediction model are adjusted based on the error loss until the error loss reaches a preset expected value, thereby obtaining a trained induced eddy current prediction model. The training data of the induced eddy current prediction model are all derived from the real diagnostic data during the discharge process of the magnetic confinement nuclear fusion device. During the training process, the diagnostic signal and the predicted induced eddy current data are respectively mapped to the unified variable parameter to judge the accuracy of the predicted value, so that the model can better capture the relationship between the induced eddy current and the diagnostic data, and to a certain extent avoid the solution error caused by ignoring or approximating some entity details in the process of establishing the physical model in the prior art, and show a higher prediction accuracy. At the same time, when the eddy current induction prediction model trained by the training method is applied, the computing power requirement is low. Therefore, the diagnostic data during the discharge process can be processed in real time, and the induced eddy current can be predicted at any time in the millisecond discharge window, so that a long-term coverage prediction can be provided. As an embodiment, the unified variable parameter can use the actual induced eddy current contribution to the magnetic diagnostic device obtained by calculation as the real value in the unified variable parameter, and the predicted induced eddy current contribution to the magnetic diagnostic device is used as the predicted value in the same variable parameter to determine the error loss, and then train the model.

[0042] In one embodiment, if Figure 3 As shown, the steps of mapping the diagnostic signal and the predicted induced eddy current data to uniform variable parameters respectively and determining the error loss between the two sets of uniform variable parameters include the following steps:

[0043] S301, determining the first magnetic flux share that the active coil contributes to the magnetic diagnostic device based on the current data. As an exemplary embodiment, the magnetic flux share that the active coil contributes to the magnetic diagnostic device can be calculated by the current data and the mutual inductance data between the active coil and the magnetic diagnostic device. Among them, the mutual inductance data group is associated with the relative position relationship between the active coil and the magnetic diagnostic device. Therefore, in this embodiment, the coil coordinate information of the active coil and the installation coordinate information of the magnetic diagnostic device are obtained; the first mutual inductance data between the active coil and the magnetic diagnostic device is determined based on the coil coordinate information and the installation coordinate information; the first magnetic flux share is determined based on the current data and the first mutual inductance data. Exemplarily, taking 9 active coils, 64 magnetic diagnostic devices and m+1 sampling moments as an example, the coordinates of the 9 active coils and the installation coordinates of the 64 magnetic diagnostic devices are recorded in the MDSplus database, and the mutual inductance matrix (size is 9×64) between the 9 coils and the 64 magnetic diagnostic devices is calculated; the current data of the above-mentioned active coils can be a current matrix of size (m+1)×9, and the current matrix of the active coils is multiplied by the number of turns of each coil, and then multiplied by the mutual inductance matrix between the active coils and the magnetic diagnostic devices to obtain a first magnetic flux share matrix of size (m+1)×64, in which the 9 coils act together on the 64 magnetic diagnostic devices at the m+1 sampling moments within the preset window.

[0044] S302. Determine the second magnetic flux share that the vacuum wall contributes to the magnetic diagnostic device based on the total magnetic flux data and the first magnetic flux share. As an exemplary embodiment, the total magnetic flux of the magnetic diagnostic device is contributed by the current of the active coil and the induced eddy current of the vacuum wall. Therefore, after obtaining the first magnetic flux share of the active coil for the magnetic diagnostic device, the magnetic flux share obtained by subtracting the first magnetic flux share from the total magnetic flux share of the magnetic diagnostic device is the second magnetic flux share actually contributed by the induced eddy current.

[0045] S303. Determine the predicted magnetic flux share that the vacuum wall contributes to the magnetic diagnostic device based on the predicted induced eddy current data. After obtaining the second magnetic flux share that the actual induced eddy current contributes, the unified variable parameter corresponding to the real value in the second magnetic flux share data sample can be used to determine the predicted magnetic flux share that the vacuum wall contributes to the magnetic diagnostic device through the predicted induced eddy current data as the unified parameter variable corresponding to the predicted value.

[0046] In this embodiment, the vacuum wall can be simulated as a plurality of annular current wires, and the coordinate information of each annular current wire is recorded, and the second mutual inductance data between the vacuum wall and the magnetic diagnostic device is determined based on the coordinate information of the annular current wire and the installation coordinate information of the magnetic diagnostic device. In this embodiment, the above data samples are all from actual experiments, so the effects of entity details and annular asymmetry can be more fully considered. In this embodiment, the vacuum wall can be divided into 209 annular current wires, and more or less numbers are also applicable in this embodiment.

[0047] After obtaining the coordinate information of the annular current filament and the installation coordinate information of the magnetic diagnostic device, the second mutual inductance data between the annular current filament and the magnetic diagnostic device is calculated. In this embodiment, taking 64 magnetic diagnostic devices as an example, a mutual inductance matrix of size 209×64 is obtained as the second mutual inductance matrix between the vacuum wall and the magnetic diagnostic device.

[0048] The predicted induced eddy current matrix corresponding to the current matrix at the m+1 sampling moments output by the initial prediction model is multiplied by the second mutual inductance matrix to obtain a predicted flux share matrix of size (m+1)×64, in which the predicted induced eddy current output by the initial prediction model at the m+1 sampling moments within a preset time window contributes to the magnetic diagnostic equipment.

[0049] S304. Calculate the first deviation between the second magnetic flux share and the predicted magnetic flux share to determine the error loss. In this embodiment, the second magnetic flux share is used as a unified variable parameter corresponding to the true value in the data sample, and the predicted magnetic flux share that the vacuum wall contributes to the magnetic diagnostic device determined by predicting the induced eddy current data is used as a unified variable parameter corresponding to the predicted value. Calculate the error between the two unified variable parameters as the error loss between the predicted value and the true value. In this embodiment, the mean square error between the second magnetic flux share and the predicted magnetic flux share can be calculated as the error loss between the predicted value and the true value.

[0050] Furthermore, the unified variable parameter can also be an induced eddy current. Referring to the description of the above embodiment, the induced eddy current of the real vacuum wall is calculated through the current data of the active coil. Specifically, the first magnetic flux share contributed by the active coil to the magnetic diagnostic device can be calculated through the current data of the active coil and the mutual inductance matrix between the active coil and the magnetic diagnostic device. For details, see the description of the calculation method of the first magnetic flux share in the above step S301; the second magnetic flux share contributed by the vacuum wall to the magnetic diagnostic device is obtained by the difference between the total magnetic flux and the first magnetic flux share, and the real induced eddy current of the vacuum wall is calculated based on the inversion of the mutual inductance matrix between the vacuum wall and the magnetic induction device, and the model is trained by using the determined error loss of the predicted induced eddy current data and the real induced eddy current data.

[0051] In order to improve the reliability of the inversion result, that is, the reliability of the real induced eddy current, the mutual inductance matrix can be subjected to singular value decomposition when the mutual inductance matrix is ​​inverted, so as to avoid the influence on the reliability of the inversion result caused by the strong singularity of the mutual inductance matrix.

[0052] Furthermore, since the current data in the active coil may change with time, a corresponding induced electromotive force will be generated on the vacuum wall. In this embodiment, the induced electromotive force exerted on the vacuum wall by the active coil and the correspondence between the induced electromotive force on the vacuum wall and the induced eddy current can be used to unify the true value and the predicted value into a unified variable parameter of the induced electromotive force.

[0053] Specifically, the circuit equation of eddy current dissipation can be used to multiply the time rate of change of the active coil current by the mutual inductance matrix between the active coil and the toroidal current wire as the actual total voltage drop; the sum of the self-inductance dissipation term contributed by the predicted induced eddy current plus the resistance dissipation term is taken as the predicted total voltage drop, and the difference between the two is calculated.

[0054] See also Figure 4 As shown, mapping the diagnostic signal and the predicted induced eddy current data to uniform variable parameters respectively, and determining the error loss between the two sets of uniform variable parameters also includes:

[0055] S401. Determine the induced electromotive force data of the active coil acting on the vacuum wall based on the current change rate. After obtaining the current data of the active coil, the current data of the active coil are respectively derived to obtain the current change rate of each active coil over time. Referring to the resampling of current data in the above embodiment, in this embodiment, the current data before resampling can be first processed to obtain the current change rate, and then interpolation sampling is performed according to the resampling method of the current data. The current data after resampling can also be derived to obtain the corresponding current change rate matrix.

[0056] Similarly, when calculating the induced electromotive force of the current data on the vacuum wall, the vacuum wall can be simulated as a plurality of annular current wires. After determining the coordinate position of the active coil and the coordinate position of each annular current wire, the mutual inductance matrix between the active coil and the annular current wire is calculated. The induced electromotive force matrix of the coil acting on each annular current wire of the vacuum wall at m+1 sampling moments within the preset time window is calculated using the current change rate matrix and the mutual inductance matrix as the unified variable parameter corresponding to the true value.

[0057] S402. Determine the predicted induced eddy current change rate of the predicted induced eddy current over time based on the predicted induced eddy current data. After obtaining the predicted induced eddy current matrix, derive the predicted induced eddy current matrix along the time dimension to obtain the predicted induced eddy current change rate matrix of the predicted induced eddy current over time.

[0058] S403. Determine the predicted annular pressure drop data of the vacuum wall based on the predicted induced eddy current change rate. In this embodiment, the induced electromotive force of the vacuum wall may come from the annular pressure drop caused by the self-inductance between the annular current wires, may come from the annular pressure drop caused by the resistance of the annular current wires, or may come from the annular pressure drop of both. In this embodiment, after obtaining the predicted induced eddy current change rate, the predicted annular pressure drop matrix in the vacuum wall can be calculated based on the source of the annular pressure drop by predicting the induced eddy current change rate as a unified variable parameter corresponding to the predicted value.

[0059] In one embodiment, the annular pressure drop caused by self-inductance is used as an example to calculate the predicted annular pressure drop, and the vacuum wall self-inductance data in the MDSplus database is obtained; in this embodiment, 209 annular current wires are also used as an example for explanation, and the self-inductance matrix of the vacuum wall is a matrix of size 209×209. The predicted annular pressure drop data contributed by self-inductance is determined based on the predicted induced eddy current change rate and the vacuum wall self-inductance data. Exemplarily, the predicted induced eddy current change rate matrix is ​​multiplied by the annular current wire self-inductance matrix to obtain the predicted annular pressure drop matrix contributed by the self-inductance of the predicted induced eddy current.

[0060] In another embodiment, the predicted annular voltage drop is calculated by taking the annular voltage drop caused by resistance as an example, and the vacuum wall resistance data is obtained, and the resistance data may be a resistance matrix of the annular current filament. The predicted annular voltage drop matrix contributed by the resistance dissipation is determined based on the predicted induced eddy current data and the resistance data.

[0061] Furthermore, when determining the predicted annular voltage drop matrix, the predicted annular voltage drop matrix of self-inductance contribution can be used alone, or the predicted annular voltage drop matrix of resistance dissipation contribution can be used alone, or the predicted annular voltage drop matrix of self-inductance contribution and the predicted annular voltage drop matrix of resistance dissipation contribution can be used at the same time, and the sum of the predicted annular voltage drop matrix of self-inductance contribution and the predicted annular voltage drop matrix of resistance dissipation contribution is used as the predicted annular voltage drop matrix.

[0062] S404. Calculate the second deviation between the induced electromotive force data and the predicted annular voltage drop data to determine the error loss. As an exemplary embodiment, the induced electromotive force is used as a unified variable parameter corresponding to the true value in the data sample, and the predicted annular voltage drop is used as a unified variable parameter corresponding to the predicted value. Calculate the error between the two unified variable parameters as the error loss between the predicted value and the true value. In this embodiment, the mean square error between the induced electromotive force data and the predicted annular voltage drop data can be calculated as the error loss between the predicted value and the true value.

[0063] In one embodiment, the first deviation can be returned as a loss function of error loss, and the second deviation can be returned as a loss function of error loss. In order to improve the prediction accuracy of the model, in another embodiment, the first deviation and the second deviation can be combined and used together as the error loss between the true value and the prediction. In this embodiment, since the matrix corresponding to the second deviation is a toroidal current wire, its matrix size is (m+1)×209, and the matrix corresponding to the first deviation is a magnetic diagnostic device, its matrix size is (m+1)×64, in order to avoid the large difference in magnitude between the two, resulting in one of the weights being too small and being ignored by the model, in this embodiment, the second deviation is scaled at a fixed ratio, and the scaled value is the same or similar magnitude as the first deviation, and the sum of the first deviation and the second deviation is returned as the loss function.

[0064] When training the model, perform the forward pass, the optimizer can be Adam, and the weight_decay setting value is 3×10 -8 , update the weight parameters in the initial model. Input each batch of data samples in turn, and repeatedly execute the method of adjusting the model parameters in the above embodiment until the loss function remains basically stable and no longer decreases, then stop training. The embodiment of the present application also provides an induced eddy current prediction method, which is a preset method for predicting the induced eddy currents on the vacuum wall of a magnetic confinement nuclear fusion device at a certain moment during the discharge process. The induced eddy current prediction model trained in the above embodiment is deployed on the MDSplus server, and diagnostic data is obtained in real time during the discharge process of the magnetic confinement nuclear fusion device. Input features are constructed based on the diagnostic data in the preset time window corresponding to the moment to be predicted, and the input features are input into the induced eddy current prediction model to obtain the induced eddy currents on the vacuum wall at the moment to be predicted.

[0065] Specifically, see Figure 5 As shown, the following steps may be included:

[0066] S501, obtaining the diagnostic data of the active coils in the preset time window corresponding to the predicted moment during the discharge process of the magnetic confinement nuclear fusion device to construct input features. As an exemplary embodiment, the diagnostic data includes the current data of multiple active coils and the total magnetic flux data of multiple magnetic diagnostic devices, wherein the current data is subjected to noise reduction processing using a sliding window average filter; the original signals of the magnetic diagnostic devices are respectively integrated to obtain the total magnetic flux contributed by the active coils to the magnetic diagnostic device. In this embodiment, the active coils are taken as nine coils, such as 3 CS coils, 5 PF coils and 1 MC coil, and the magnetic diagnostic devices are taken as 64 magnetic diagnostic devices, such as 49 magnetic probes and 15 flux rings. The current data of the 9 active coils are sampled multiple times to obtain multiple current data sampling results, and the total magnetic flux of the 64 magnetic diagnostic devices is sampled in the same way to obtain multiple total magnetic flux sampling results, and the current data sampling matrix and the total magnetic flux sampling matrix are respectively constructed, and the current data sampling matrix and the total magnetic flux sampling matrix are merged to obtain the input feature matrix.

[0067] S502. Input the input features into the induced eddy current prediction model constructed by the induced eddy current prediction model construction method to obtain the predicted induced eddy current of the vacuum wall of the magnetic confinement nuclear fusion device during the discharge process.

[0068] In this embodiment, see Figure 6 As shown, taking Tokamak as an example, the test results of discharge are: Figure 6 Curve 1 is the total magnetic flux recorded by the actual magnetic diagnostic device minus the magnetic flux contribution of the active coil, and the magnetic flux contribution of the vacuum wall induced eddy current to the magnetic diagnostic device is obtained. Curve 2 is the magnetic flux contribution of the induced eddy current predicted by the induced eddy current prediction model. Curve 3 is the magnetic flux contribution of the induced eddy current obtained by solving the circuit differential equation using the traditional method. Among them, the horizontal axis is time, the vertical axis is magnetic flux, and the units are all in the international system of units. It can be seen that compared with the traditional method, the recurrent neural network model used in the present invention achieves significantly higher prediction accuracy and is more consistent with the actual experimental results.

[0069] According to another aspect of an embodiment of the present application, an electronic device for implementing the above-mentioned induced eddy current prediction model construction method and / or induced eddy current prediction method is also provided. The electronic device may be a server, a terminal, or a combination thereof.

[0070] Figure 7 is a structural block diagram of an electronic device according to an embodiment of the present application, such as Figure 7 As shown, it includes a processor 701, a communication interface 702, a memory 703 and a communication bus 704, wherein the processor 701, the communication interface 702 and the memory 703 communicate with each other through the communication bus 704, wherein,

[0071] Memory 703, used for storing computer programs;

[0072] The processor 701 is used to execute the computer program stored in the memory 703 to implement the following steps:

[0073] Acquire multiple groups of diagnostic signals in the historical discharge process of the magnetic confinement nuclear fusion device to construct a training set, wherein the diagnostic signals include current data of multiple active coils in the magnetic confinement nuclear fusion device and total magnetic flux data of multiple magnetic diagnostic devices;

[0074] Inputting the data samples in the training set into the initial prediction model to obtain predicted induced eddy current data;

[0075] The diagnostic signal and the predicted induced eddy current data are respectively mapped to uniform variable parameters, the error loss between the two sets of the uniform variable parameters is determined, and the parameters of the initial prediction model are adjusted based on the error loss until the error loss reaches a preset expected value, thereby obtaining a trained induced eddy current prediction model.

[0076] In one embodiment, the processor 701, when used to execute the computer program stored in the memory 703, may further implement the following steps:

[0077] Acquire the diagnostic data of the active coil within a preset time window corresponding to the time to be predicted during the discharge process to construct input features;

[0078] The input features are input into the induced eddy current prediction model constructed according to the above induced eddy current prediction model construction method to obtain the predicted induced eddy current of the vacuum wall of the Tokamak device during the discharge process.

[0079] The induced eddy current prediction model construction method and the induced eddy current prediction method in the above embodiment can be executed in the same electronic device or in different electronic devices. In one embodiment, the above communication bus can be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 7 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0080] The communication interface is used for communication between the above electronic device and other devices.

[0081] The memory may include RAM, or may include non-volatile memory, such as at least one disk memory. In addition, the memory may also be at least one storage device located far away from the aforementioned processor.

[0082] The above-mentioned processor can be a general-purpose processor, which can include but not be limited to: CPU (Central Processing Unit), NP (Network Processor), etc.; it can also be DSP (Digital Signal Processing), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0083] The specific examples in this embodiment can refer to the examples described in the above embodiments, and this embodiment will not be described in detail here.

[0084] It can be understood by those skilled in the art that Figure 7 The structure shown is for illustration only, and the device implementing the above-mentioned induced eddy current prediction model construction method and / or induced eddy current prediction method may be a terminal device. Figure 7 It does not limit the structure of the above electronic device. For example, the terminal device may also include Figure 7 More or fewer components (such as network interfaces, display devices, etc.) shown in, or having Figure 7 Different configurations shown.

[0085] A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the hardware related to the terminal device through a program, and the program can be stored in a computer-readable storage medium, which can include: a flash drive, ROM, RAM, a magnetic disk or an optical disk, etc.

[0086] According to another aspect of the embodiments of the present application, a storage medium is also provided, which can be used to execute the program code of the induced eddy current prediction model construction method and / or the induced eddy current prediction method.

[0087] In one embodiment, the storage medium may be located on at least one network device among the plurality of network devices in the network mentioned above.

[0088] Further, the storage medium is configured to store program codes for executing the following steps:

[0089] Acquire multiple groups of diagnostic signals in the historical discharge process of the magnetic confinement nuclear fusion device to construct a training set, wherein the diagnostic signals include current data of multiple active coils in the magnetic confinement nuclear fusion device and total magnetic flux data of multiple magnetic diagnostic devices;

[0090] Inputting the data samples in the training set into the initial prediction model to obtain predicted induced eddy current data;

[0091] The diagnostic signal and the predicted induced eddy current data are respectively mapped to uniform variable parameters, the error loss between the two sets of the uniform variable parameters is determined, and the parameters of the initial prediction model are adjusted based on the error loss until the error loss reaches a preset expected value, thereby obtaining a trained induced eddy current prediction model.

[0092] In one embodiment, the storage medium may also be configured to store program codes for performing the following steps:

[0093] Acquire the diagnostic data of the active coil within a preset time window corresponding to the time to be predicted during the discharge process to construct input features;

[0094] The input features are input into the induced eddy current prediction model constructed according to the above induced eddy current prediction model construction method to obtain the predicted induced eddy current of the vacuum wall of the Tokamak device during the discharge process.

[0095] The program codes corresponding to the induced eddy current prediction model construction method and the induced eddy current prediction method in the above embodiments may be stored in the same storage medium or in different storage media.

[0096] The specific examples in this embodiment can refer to the examples described in the above embodiments, which will not be described in detail in this embodiment.

[0097] Furthermore, the above storage medium may include but is not limited to: a U disk, a ROM, a RAM, a mobile hard disk, a magnetic disk or an optical disk, and other media that can store program codes.

[0098] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0099] If the integrated units in the above embodiments are implemented in the form of software functional units and sold or used as independent products, they can be stored in the above computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling one or more computer devices (which may be personal computers, servers, or network devices, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application.

[0100] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0101] In the several embodiments provided in the present application, it should be understood that the disclosed client can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0102] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution provided in this embodiment.

[0103] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0104] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0105] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A method for constructing an induced eddy current prediction model, suitable for predicting the induced eddy current of the vacuum wall in a magnetic confinement nuclear fusion device, characterized in that: include: Acquire multiple groups of diagnostic signals in the historical discharge process of the magnetic confinement nuclear fusion device to construct a training set, wherein the diagnostic signals include current data of multiple active coils in the magnetic confinement nuclear fusion device and total magnetic flux data of each magnetic diagnostic device in multiple magnetic diagnostic devices; Inputting the data samples in the training set into the initial prediction model to obtain predicted induced eddy current data; The diagnostic signal and the predicted induced eddy current data are respectively mapped to unified variable parameters, the error loss between the two groups of the unified variable parameters is determined, and the parameters of the initial prediction model are adjusted based on the error loss until the error loss reaches a preset expected value, so as to obtain a trained induced eddy current prediction model, wherein the unified variable parameters include the magnetic flux share contributed by the induced eddy current on the magnetic diagnostic device, and the error loss includes the error loss between the true value and the predicted value of the magnetic flux share contributed by the induced eddy current on the magnetic diagnostic device; or, the unified variable parameters include the induced eddy current, and the error loss includes the error loss between the true value of the induced eddy current and the predicted value of the induced eddy current.

2. The prediction model construction method according to claim 1, characterized in that: Mapping the diagnostic signal and the predicted induced eddy current data to uniform variable parameters respectively and determining the error loss between the two sets of uniform variable parameters comprises: determining a first magnetic flux share that the active coil contributes to the magnetic diagnostic device based on the current data; determining a second magnetic flux share that the vacuum wall contributes to the magnetic diagnostic device based on the total magnetic flux data and the first magnetic flux share; determining a predicted magnetic flux share that the vacuum wall contributes to the magnetic diagnostic device based on the predicted induced eddy current data; A first deviation between the second magnetic flux share and the predicted magnetic flux share is calculated to determine the error loss.

3. The prediction model construction method according to claim 1 or 2, characterized in that: The diagnostic signal also includes a current change rate of the current over time; The step of mapping the diagnostic signal and the predicted induced eddy current data to uniform variable parameters respectively and determining the error loss between the two sets of uniform variable parameters comprises: Determining the induced electromotive force data of the active coil acting on the vacuum wall based on the current change rate; Determining a predicted induced eddy current change rate of the predicted induced eddy current over time based on the predicted induced eddy current data; Determining predicted annular pressure drop data of the vacuum wall based on the predicted induced eddy current change rate and / or predicted induced eddy current data; A second deviation between the induced electromotive force data and the predicted annular voltage drop data is calculated to determine the error loss.

4. The prediction model construction method according to claim 3, characterized in that: The predicted annular pressure drop data include predicted annular pressure drop data of self-inductance contribution of the vacuum wall and / or predicted annular pressure drop data of resistance dissipation contribution of the vacuum wall; The step of determining the predicted annular pressure drop data of the vacuum wall based on the predicted induced eddy current data comprises: Acquiring self-inductance data of the vacuum wall; Determine predicted annular pressure drop data of self-inductance contribution based on the predicted induced eddy current change rate and the self-inductance data of the vacuum wall; and / or Acquiring resistance data of the vacuum wall; Predicted hoop voltage drop data for the resistive dissipation contribution is determined based on the predicted induced eddy current data and the electrical resistance data.

5. The prediction model construction method according to claim 1, characterized in that: The initial prediction model includes a recurrent neural network model, wherein the recurrent neural network has at least two stacked recursive layers; The step of inputting the data samples in the training set into the initial prediction model to obtain the predicted induced eddy current data comprises: The sample data in the training set is input into the recurrent neural network model, and the predicted induced eddy current data is obtained through at least two recursive layers using a tanh activation function.

6. The prediction model construction method according to claim 1, characterized in that: The training set is constructed by obtaining multiple groups of diagnostic signals from the historical discharge process of the magnetic confinement nuclear fusion device, including: Acquire multiple sets of diagnostic data under different discharge configurations and / or different discharge times; Resampling the diagnostic data within a preset time window to obtain a diagnostic data sampling matrix; The training set is constructed based on multiple groups of diagnostic data sampling matrices.

7. The prediction model construction method according to claim 6, characterized in that: The resampling of the diagnostic data within a preset time window to obtain a diagnostic data sampling matrix includes: Based on a preset time period before the selected target moment as a preset time window, the target moment is within the discharge time window, and the preset time period is less than the discharge time window; The diagnostic data is difference-resampled according to a preset sampling frequency within the preset time window to obtain a sampling matrix corresponding to the diagnostic data.

8. An induced eddy current prediction method, characterized in that: include: Acquire diagnostic data of an active coil in a preset time window corresponding to a time to be predicted during a discharge process of a magnetic confinement nuclear fusion device to construct input features; The input features are input into the induced eddy current prediction model constructed according to the induced eddy current prediction model construction method according to any one of claims 1 to 7 to obtain the predicted induced eddy currents of the vacuum wall of the magnetic confinement nuclear fusion device during the discharge process.

9. The method for predicting induced eddy currents according to claim 8, characterized in that: The diagnostic data includes current data of a plurality of active coils and total magnetic flux data of a plurality of magnetic induction devices; The step of acquiring the diagnostic data of the active coil in the preset time window corresponding to the predicted moment in the discharge process of the magnetic confinement nuclear fusion device to construct the input features includes: Select the time to be predicted; Sampling the current data and the total magnetic flux data within a preset time window before the time to be predicted to obtain sampled data of the current data and the total magnetic flux data; A data matrix of the input features is constructed using the sampled data.

10. An electronic device, characterized in that: include: Processor, memory; The memory is used to store computer programs; The processor is used to execute the computer program and implement the induced eddy current prediction model construction method according to any one of claims 1 to 7 and / or execute the induced eddy current prediction method according to claim 8 or 9 when executing the computer program.

Images