CN117933069A - Inversion proxy model training method and device based on plasmas - Google Patents
Inversion proxy model training method and device based on plasmas Download PDFInfo
- Publication number
- CN117933069A CN117933069A CN202410021830.6A CN202410021830A CN117933069A CN 117933069 A CN117933069 A CN 117933069A CN 202410021830 A CN202410021830 A CN 202410021830A CN 117933069 A CN117933069 A CN 117933069A
- Authority
- CN
- China
- Prior art keywords
- sampling
- plasma
- distribution profile
- signal
- inversion
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 67
- 238000012549 training Methods 0.000 title claims abstract description 38
- 210000002381 plasma Anatomy 0.000 title claims description 114
- 238000005070 sampling Methods 0.000 claims abstract description 250
- 230000005855 radiation Effects 0.000 claims abstract description 107
- 238000000605 extraction Methods 0.000 claims abstract description 100
- 108010076504 Protein Sorting Signals Proteins 0.000 claims abstract description 91
- 238000009826 distribution Methods 0.000 claims abstract description 90
- 238000013528 artificial neural network Methods 0.000 claims description 28
- 238000003860 storage Methods 0.000 claims description 18
- 239000011553 magnetic fluid Substances 0.000 claims description 16
- 238000004590 computer program Methods 0.000 claims description 15
- 238000012545 processing Methods 0.000 claims description 14
- 238000012544 monitoring process Methods 0.000 claims description 7
- 238000007781 pre-processing Methods 0.000 claims description 7
- 230000002159 abnormal effect Effects 0.000 claims description 5
- 230000001502 supplementing effect Effects 0.000 claims description 5
- 244000309464 bull Species 0.000 claims description 2
- 238000010586 diagram Methods 0.000 description 15
- 230000006870 function Effects 0.000 description 13
- 238000003745 diagnosis Methods 0.000 description 11
- 230000008569 process Effects 0.000 description 9
- 230000006872 improvement Effects 0.000 description 8
- 238000005516 engineering process Methods 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 4
- 230000004927 fusion Effects 0.000 description 3
- 230000003287 optical effect Effects 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004587 chromatography analysis Methods 0.000 description 1
- 238000011217 control strategy Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000005055 memory storage Effects 0.000 description 1
- 229920001296 polysiloxane Polymers 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 239000010979 ruby Substances 0.000 description 1
- 229910001750 ruby Inorganic materials 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G21—NUCLEAR PHYSICS; NUCLEAR ENGINEERING
- G21B—FUSION REACTORS
- G21B1/00—Thermonuclear fusion reactors
- G21B1/25—Maintenance, e.g. repair or remote inspection
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01T—MEASUREMENT OF NUCLEAR OR X-RADIATION
- G01T1/00—Measuring X-radiation, gamma radiation, corpuscular radiation, or cosmic radiation
- G01T1/29—Measurement performed on radiation beams, e.g. position or section of the beam; Measurement of spatial distribution of radiation
- G01T1/2914—Measurement of spatial distribution of radiation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2113/00—Details relating to the application field
- G06F2113/08—Fluids
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Molecular Biology (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- High Energy & Nuclear Physics (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Plasma & Fusion (AREA)
- Measurement Of Radiation (AREA)
Abstract
The specification discloses a method and a device for training an inversion proxy model based on plasma, which acquire a sampling signal sequence, wherein the sampling signal sequence is obtained by sampling soft X-band radiation generated by the plasma at a plurality of sampling moments in advance, the sampling signal sequence corresponding to each sampling moment is obtained by sampling soft X-band radiation generated by the plasma at a plurality of sampling positions, and the radiation distribution profile label of the plasma at each sampling moment is acquired. And inputting the sampling signal sequence into a first feature extraction layer to obtain a first signal feature, and inputting the first signal feature into a second feature extraction layer to obtain a second signal feature. And inputting the second signal characteristics into a result prediction layer to obtain a radiation distribution profile prediction result of the second signal characteristics corresponding to each sampling moment output by the result prediction layer. And determining loss according to the radiation distribution profile prediction result and the radiation distribution profile annotation, and training the inversion proxy model according to the loss.
Description
Technical Field
The application relates to the field of nuclear fusion plasma diagnosis, in particular to an inversion proxy model training method and device based on plasmas.
Background
The realization of stable operation of the plasma is very critical for magnetic confinement controlled nuclear fusion, and magnetic fluid (MHD for short) in the plasma can cause interference to the stable operation of the plasma, so that the discharge of the plasma is unstable or fails. In general, the magnetic fluid can be monitored, and a corresponding control strategy is given based on the monitored magnetic fluid information, so that the plasma can stably run.
At present, in the field of plasma diagnosis, soft X-band radiation of a plasma polar section can be measured, then a chromatography reconstruction algorithm is developed, and the two-dimensional spatial distribution of the radiation is accurately inverted from line integral data with limited quantity and errors, so that the aim of measuring MHD activity by a core observation part is fulfilled. That is, in achieving monitoring of the magnetic fluid, it is necessary to rapidly analyze the radiation profile of the plasma. It is therefore an important issue how to rapidly resolve the radiation profile of a plasma based on soft X-band radiation generated by the plasma.
Based on the above, the specification provides a plasma-based inversion proxy model training method.
Disclosure of Invention
The present specification provides a method, apparatus, storage medium and electronic device for training inversion proxy model based on plasma, so as to at least partially solve the above-mentioned problems existing in the prior art.
The technical scheme adopted in the specification is as follows:
The specification provides a inversion proxy model training method based on plasmas, the inversion proxy model comprises a first feature extraction layer, a second feature extraction layer and a result prediction layer, the second feature extraction layer comprises a multi-head attention network with a residual connection structure and a forward neural network with a residual connection structure, the first feature extraction layer and the result prediction layer all comprise full connection layers, and the method comprises the following steps:
Acquiring a sampling signal sequence, wherein the sampling signal sequence is obtained by sampling soft X-rays generated by plasma at a plurality of sampling moments in advance, and the sampling signal sequence corresponding to each sampling moment is obtained by sampling the soft X-rays generated by the plasma at a plurality of sampling positions; the radiation distribution profile label of the plasma at each sampling moment is obtained;
inputting the sampling signal sequence into the first feature extraction layer to obtain a first signal feature;
Inputting the first signal characteristics into the second characteristic extraction layer, so that the multi-head attention network with the residual connection structure carries out weighting processing on the first signal characteristics, and the forward neural network with the residual connection structure carries out characteristic extraction on the first signal characteristics at each sampling moment to obtain second signal characteristics output by the second characteristic extraction layer;
Inputting the second signal characteristics into the result prediction layer to obtain a radiation distribution profile prediction result of the second signal characteristics corresponding to each sampling moment output by the result prediction layer;
determining loss according to the radiation distribution profile prediction result and the radiation distribution profile label;
And training the inversion proxy model according to the loss.
Optionally, inputting the sampling signal sequence into the first feature extraction layer specifically includes:
Preprocessing the sampling signal sequence, and inputting the preprocessed sampling signal sequence into the first feature extraction layer; wherein the preprocessing comprises: removing abnormal signals in the sampling signal sequence and/or supplementing missing signals in the sampling signal sequence.
Optionally, the inversion proxy model further includes a third feature extraction layer, where the third feature extraction layer includes a multi-head attention network with a residual connection structure and a forward neural network with a residual connection structure;
Outputting the second signal characteristic to the result prediction layer, wherein the method specifically comprises the following steps:
inputting the second signal characteristics into the third characteristic extraction layer to obtain third signal characteristics;
And inputting the third signal characteristic into the result prediction layer.
Optionally, soft X-rays generated by the plasma are sampled at several sampling moments, specifically including:
at a number of sampling instants, soft X-rays generated by the plasma are sampled in accordance with a soft X-ray diagnostic system.
Optionally, determining the loss according to the radiation distribution profile prediction result and the radiation distribution profile annotation specifically includes:
For each sampling time, determining a difference value between a radiation distribution profile prediction result corresponding to the sampling time and a radiation distribution profile label corresponding to the sampling time;
judging whether the difference value is within a preset range or not;
if yes, determining loss corresponding to the sampling moment by adopting a first formula according to the radiation distribution profile prediction result corresponding to the sampling moment and the radiation distribution profile label corresponding to the sampling moment;
if not, determining the loss corresponding to the sampling moment by adopting a second formula according to the radiation distribution profile prediction result corresponding to the sampling moment and the radiation distribution profile label corresponding to the sampling moment.
Optionally, training the inversion proxy model according to the loss specifically includes:
and training the inversion proxy model according to the obtained loss corresponding to each sampling time.
Optionally, the method further comprises:
acquiring a signal sequence to be predicted, wherein the signal sequence to be predicted is obtained by sampling soft X-rays generated by plasma to be predicted at a plurality of sampling moments, and the sampling signal sequence corresponding to each sampling moment is obtained by sampling soft X-rays generated by the plasma to be predicted at a plurality of sampling positions;
Inputting the signal to be predicted into a trained inversion proxy model to obtain a radiation distribution profile of the plasma to be predicted, which is output by the trained inversion proxy model;
And monitoring the magnetic fluid state in the plasma to be predicted according to the obtained radiation distribution profile of the plasma to be predicted.
The utility model provides a inversion proxy model trainer based on plasma, inversion proxy model includes first characteristic extraction layer, second characteristic extraction layer and result prediction layer, the second characteristic extraction layer is including the bull attention network that possesses residual connection structure and possess the forward neural network of residual connection structure, first characteristic extraction layer with result prediction layer all includes full tie-in layer, the device specifically includes:
The data acquisition module is used for acquiring a sampling signal sequence, wherein the sampling signal sequence is obtained by sampling soft X-rays generated by plasma at a plurality of sampling moments in advance, and the sampling signal sequence corresponding to each sampling moment is obtained by sampling the soft X-rays generated by the plasma at a plurality of sampling positions; the radiation distribution profile label of the plasma at each sampling moment is obtained;
the first feature extraction module is used for inputting the sampling signal sequence into the first feature extraction layer to obtain a first signal feature;
the second feature extraction module is used for inputting the first signal features into the second feature extraction layer, so that the multi-head attention network with the residual connection structure performs weighting processing on the first signal features, and the forward neural network with the residual connection structure performs feature extraction on the first signal features at each sampling moment to obtain second signal features output by the second feature extraction layer;
The result prediction module is used for inputting the second signal characteristics into the result prediction layer to obtain a radiation distribution profile prediction result of the second signal characteristics corresponding to each sampling moment output by the result prediction layer;
the loss determination module is used for determining loss according to the radiation distribution profile prediction result and the radiation distribution profile annotation;
And the model training module is used for training the inversion proxy model according to the loss.
Optionally, the first feature extraction module is specifically configured to pre-process the sampled signal sequence, and input the pre-processed sampled signal sequence into the first feature extraction layer; wherein the preprocessing comprises: removing abnormal signals in the sampling signal sequence and/or supplementing missing signals in the sampling signal sequence.
Optionally, the inversion proxy model further includes a third feature extraction layer, where the third feature extraction layer includes a multi-head attention network with a residual connection structure and a forward neural network with a residual connection structure;
the result prediction module is specifically configured to input the second signal feature into the third feature extraction layer to obtain a third signal feature; and inputting the third signal characteristic into the result prediction layer.
Optionally, the data acquisition module is specifically configured to sample soft X-rays generated by the plasma according to a soft X-ray diagnostic system at a plurality of sampling moments.
Optionally, the loss determining module is specifically configured to determine, for each sampling time, a difference between a radiation distribution profile prediction result corresponding to the sampling time and a radiation distribution profile label corresponding to the sampling time; judging whether the difference value is within a preset range or not; if yes, determining loss corresponding to the sampling moment by adopting a first formula according to the radiation distribution profile prediction result corresponding to the sampling moment and the radiation distribution profile label corresponding to the sampling moment; if not, determining the loss corresponding to the sampling moment by adopting a second formula according to the radiation distribution profile prediction result corresponding to the sampling moment and the radiation distribution profile label corresponding to the sampling moment.
Optionally, the model training module is specifically configured to train the inversion proxy model according to the obtained loss corresponding to each sampling time.
Optionally, the apparatus further comprises a model application module;
The model application module is specifically used for acquiring a signal sequence to be predicted, wherein the signal sequence to be predicted is obtained by sampling soft X-rays generated by plasma to be predicted at a plurality of sampling moments, and the sampling signal sequence corresponding to each sampling moment is obtained by sampling soft X-rays generated by the plasma to be predicted at a plurality of sampling positions; inputting the signal to be predicted into a trained inversion proxy model to obtain a radiation distribution profile of the plasma to be predicted, which is output by the trained inversion proxy model; and monitoring the magnetic fluid state in the plasma to be predicted according to the obtained radiation distribution profile of the plasma to be predicted.
The present specification provides a computer readable storage medium storing a computer program which when executed by a processor implements the above-described plasma-based inversion proxy model training method.
The present specification provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the above-described plasma-based inversion proxy model training method when executing the program.
The above-mentioned at least one technical scheme that this specification adopted can reach following beneficial effect:
In the inversion proxy model training method based on the plasmas provided in the specification, a sampling signal sequence can be obtained first, the sampling signal sequence is obtained by sampling soft X-rays generated by the plasmas at a plurality of sampling moments in advance, the sampling signal sequence corresponding to each sampling moment is obtained by sampling soft X-rays generated by the plasmas at a plurality of sampling positions, and the radiation distribution profile marking of the plasmas at each sampling moment can be obtained. And then, inputting the sampled signal sequence into a first feature extraction layer to obtain first signal features, inputting the first signal features into a second feature extraction layer to enable the multi-head attention network with the residual connection structure to carry out weighting processing on the first signal features, and enabling the forward neural network with the residual connection structure to carry out feature extraction on the first signal features at each sampling moment to obtain second signal features output by the second feature extraction layer. Then, the second signal characteristics can be input into a result prediction layer to obtain a radiation distribution profile prediction result of the second signal characteristics corresponding to each sampling time output by the result prediction layer. Finally, the loss can be determined according to the radiation distribution profile prediction result and the radiation distribution profile annotation, and the inversion proxy model is trained according to the loss.
According to the method, the inversion proxy model is constructed by constructing the structure of the inversion proxy model, namely adopting the first feature extraction layer, the second feature extraction layer and the result prediction layer, the second feature extraction layer comprises a multi-head attention network with a residual connection structure and a forward neural network with a residual connection structure, so that the head attention network weights the features, the forward neural network with the residual connection structure can extract the time sequence information features of a soft X-ray signal sequence at each sampling moment, and the method is different from a least square inversion proxy method, a Bayesian inversion proxy method and the like which are adopted in the plasma diagnosis field at present, and improves the accuracy of a radiation distribution section of plasma predicted by the inversion proxy model while improving the prediction efficiency.
Drawings
The accompanying drawings, which are included to provide a further understanding of the specification, illustrate and explain the exemplary embodiments of the present specification and their description, are not intended to limit the specification unduly. In the drawings:
FIG. 1 is a schematic flow chart of an inversion proxy model training method based on plasma in the specification;
FIG. 2a is a schematic diagram of an inversion proxy model provided in the present specification;
FIG. 2b is a schematic diagram of an inversion proxy model provided in the present specification;
FIG. 3 is a schematic structural diagram of an inversion proxy model provided in the present specification;
FIG. 4 is a schematic diagram of an inversion proxy model training device based on plasma provided in the present specification;
fig. 5 is a schematic view of the electronic device corresponding to fig. 1 provided in the present specification.
Detailed Description
For the purposes of making the objects, technical solutions and advantages of the present specification more apparent, the technical solutions of the present specification will be clearly and completely described below with reference to specific embodiments of the present specification and corresponding drawings. It will be apparent that the described embodiments are only some, but not all, of the embodiments of the present specification. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are intended to be within the scope of the present disclosure.
The following describes in detail the technical solutions provided by the embodiments of the present specification with reference to the accompanying drawings.
Fig. 1 is a schematic flow chart of an inversion proxy model training method based on plasma provided in the present specification, which specifically includes the following steps:
S100: acquiring a sampling signal sequence, wherein the sampling signal sequence is obtained by sampling soft X-rays generated by plasma at a plurality of sampling moments in advance, and the sampling signal sequence corresponding to each sampling moment is obtained by sampling the soft X-rays generated by the plasma at a plurality of sampling positions; and acquiring a radiation distribution profile label of the plasma at each sampling moment.
In general, controlled nuclear fusion can be achieved by magnetically confining, i.e., confining the plasma in a toroidal device and energizing to heat the plasma in the toroidal device. The aforementioned ring device is referred to as a tokamak device. In practical applications, on the one hand, tokamak devices often suffer from various magnetic fluid instabilities, which makes the plasma discharge unstable or fails. On the other hand, the radiation distribution profile of the plasma can be analyzed to realize the real-time monitoring of the magnetic fluid interference and timely give out a control method, so that the stable discharge of the plasma is ensured. In addition, although the magnetic fluid in the plasma is very complex, the radiation distribution profile of the plasma can be analyzed by measuring the soft X-ray (i.e., soft X-band radiation) signal generated by the plasma. Therefore, the radiation distribution profile of the plasma can be obtained through inversion of soft X-rays, so that the magnetic fluid interference can be monitored based on the radiation distribution profile of the plasma, and stable discharge of the plasma can be ensured.
In general, there is a soft X-ray diagnosis system for measuring magnetic fluid information built in advance in a tokamak device, and the soft X-ray diagnosis system includes a plurality of sensors, the plurality of sensors may be used for sampling soft X-rays generated by plasma, the plurality of sensors may have different channel numbers, and installation positions (orientations) of the plurality of sensors may be different, that is, the plurality of detectors may be located at different positions for sampling soft X-rays generated by plasma at different positions (orientations). In the present specification, the execution body for training the inversion proxy model based on the plasma may be any computing device having computing capability, such as a server, a terminal, etc., where a soft X-ray diagnosis system may be built in advance, a function for measuring magnetic fluid activity similar to that of the soft X-ray diagnosis system may be provided, and a function for communicating with the soft X-ray diagnosis system in the tokamak device may be provided, so that the soft X-ray diagnosis system may establish communication, and obtain soft X-ray signals acquired by the soft X-ray diagnosis system, and then the computing device may acquire a sampling signal sequence.
In one or more embodiments of the present specification, the sampling signal sequence may be obtained by sampling soft X-rays generated by the plasma at a plurality of sampling times in advance by the computing device using the soft X-ray diagnostic system, and the sampling signal sequence corresponding to each sampling time is obtained by sampling soft X-rays generated by the plasma at a plurality of sampling positions. That is, the computing device may obtain a sampling signal sequence obtained by sampling soft X-rays generated by the plasma at a plurality of sampling moments, and the sampling signal sequence corresponding to each sampling moment is obtained by sampling soft X-rays generated by the plasma at a plurality of sampling positions. For example: assuming that there are N sampling instants and M sampling positions, the sampled signal sequence is an nxm signal matrix.
It should be noted that, the above-mentioned sampling position is a sampling position, that is, a sensor mounting position (position) in the soft X-ray diagnostic system, and the sensor mounting position in the soft X-ray system as described above may be different and may be used to sample soft X-rays generated by plasma in different directions, so that the computing device may obtain a sampling signal sequence for sampling soft X-rays at different sampling positions at each sampling time.
Moreover, the computing device may also obtain radiation profile annotations for the plasma at each sampling instant. I.e. the computing device may obtain radiation profile annotations for the sequence of sampled signals at each sampling instant. The radiation distribution profile label may be an accurate radiation distribution profile of the plasma calculated in advance based on other modes, for example, the radiation distribution profile of the sampling signal sequence at each sampling time may be determined based on a non-gaussian smoothing process, the radiation distribution profile of the sampling signal sequence at each sampling time may be determined based on a fourier-bessel analysis method, and the radiation distribution profile of the sampling signal sequence at each sampling time may be determined based on a least square method.
S102: and inputting the sampling signal sequence into the first feature extraction layer to obtain a first signal feature.
S104: and inputting the first signal characteristics into the second characteristic extraction layer, so that the multi-head attention network with the residual connection structure performs weighting processing on the first signal characteristics, and the forward neural network with the residual connection structure performs characteristic extraction on the first signal characteristics at each sampling moment to obtain second signal characteristics output by the second characteristic extraction layer.
S106: and inputting the second signal characteristics into the result prediction layer to obtain a radiation distribution profile prediction result of the second signal characteristics corresponding to each sampling moment output by the result prediction layer.
Fig. 2a and fig. 2b are schematic structural diagrams of an inversion proxy model provided in the present disclosure. It can be seen that the inversion proxy model provided in the present specification includes a first feature extraction layer, which may be a fully connected layer, and a second feature extraction layer, which may include a multi-head attention network with a residual connection structure and a forward neural network with a residual connection structure, and a result prediction layer, which may be a fully connected layer.
The computing device may input the sequence of sampled signals into a first feature extraction layer resulting in a first signal feature. The first feature extraction layer may adjust, for each sampling instant, a feature dimension of a sampled signal sequence corresponding to the sampling instant. Along the above example, if the sampling signal sequence is an nxm matrix, the first feature extraction layer may adjust the dimension of the sampling signal sequence corresponding to each sampling time, that is, adjust the dimension of the matrix by 1×m, and if the dimension M may be adjusted to the dimension Y, the first signal feature of the first feature extraction layer may be an nxy feature matrix.
The computing device may then input the first signal features into a second feature extraction layer, the multi-headed attention network with the residual connection structure in the second feature extraction layer may perform a weighting process on the input first signal features, and the forward neural network with the residual connection structure in the second feature extraction layer may perform feature extraction on the first signal features corresponding to each sampling time to obtain second signal features output by the second feature extraction layer. Specifically, the multi-head attention network with the residual connection structure can weight the first signal characteristics corresponding to the sampling signal sequence to obtain the weighted first signal characteristics. And then, according to the time sequence information reflected by the sampling moments, carrying out feature extraction on the weighted first signal features corresponding to each sampling moment based on the forward neural network with the residual error connection structure so as to extract the features of single sampling moment and obtain the second signal features corresponding to each sampling moment, thereby obtaining the second signal features corresponding to the sampling moments.
Further, in one or more embodiments of the present disclosure, the multi-headed attention network with residual connection structure may include three full-connection sub-networks, the forward neural network with residual connection structure may include two full-connection sub-networks, and a GeLU activation function may be further included between the two full-connection sub-networks. The computing device may perform feature extraction on the first signal features through the three fully-connected subnetworks, respectively, to obtain intermediate signal features that are output by the three fully-connected subnetworks, respectively. A weight for weighting the first signal feature may then be determined based on the three intermediate signal features obtained to weight the first signal feature based on the weight.
The computing device may then input the second signal characteristic into the result prediction layer to obtain a radiation profile prediction result of the second signal characteristic corresponding to each sampling instant output by the result prediction layer.
S108: and determining loss according to the radiation distribution profile prediction result and the radiation distribution profile label.
S110: and training the inversion proxy model according to the loss.
Finally, the computing device may determine the loss based on the radiation profile prediction and the radiation profile annotation, and train the inversion proxy model based on the loss. Specifically, for each sampling instant, the computing device may determine a difference between the radiation profile prediction result corresponding to the sampling instant and the radiation profile annotation corresponding to the sampling instant. And judging whether the difference is in a preset range, if so, determining the loss corresponding to the sampling moment by adopting a first formula according to the radiation distribution profile prediction result corresponding to the sampling moment and the radiation distribution profile label corresponding to the sampling moment, and if not, determining the loss corresponding to the sampling moment by adopting a second formula according to the radiation distribution profile prediction result corresponding to the sampling moment and the radiation distribution profile label corresponding to the sampling moment.
In this specification, the loss may be an L1-smooth loss, and the following formula may be adopted:
wherein x n represents a radiation distribution profile label of a sampling signal sequence corresponding to each sampling time, y n represents a radiation distribution profile prediction result of the sampling signal sequence corresponding to each sampling time predicted based on the sampling signal sequence, beta is a super parameter, and the super parameter can be preset according to specific requirements. Determining whether the difference is within a preset range, i.e. determining whether the absolute value of the difference is smaller than a preset value (the super parameter beta), a first formula is The second formula is los2= |x n-yn | -0.5×beta, then the computing device may determine a loss corresponding to the sampling time based on the first formula and the second formula, and train the inversion proxy model according to the obtained loss corresponding to each sampling time.
In the inversion proxy model training method based on the plasmas provided in the present specification and shown in fig. 1, the inversion proxy model is constructed by constructing the structure of the inversion proxy model, that is, by adopting the first feature extraction layer, the second feature extraction layer and the result prediction layer, and the second feature extraction layer comprises a multi-head attention network with a residual connection structure and a forward neural network with a residual connection structure, so that the head attention network weights the features, and the forward neural network with the residual connection structure can extract the time sequence information features of the soft X-ray signal sequence at each sampling moment, which is different from the least square inversion method, the bayesian inversion method and the like currently adopted in the plasma diagnosis field, so that the speed of inverting and analyzing the radiation distribution profile of plasmas is greatly improved, that is, the radiation distribution profile of plasmas can be quickly inverted and analyzed based on soft X-wave band radiation generated by the plasmas. And the accuracy of the radiation distribution profile of the plasma predicted by the inversion proxy model is improved while the prediction efficiency is improved.
In one or more embodiments of the present disclosure, in the step S102, that is, when the calculation inputs the sampled signal sequence into the first feature extraction layer, the sampled signal sequence may be preprocessed to determine an outlier and a missing value in the sampled signal sequence, and then the preprocessed sampled signal sequence is input into the first feature extraction layer. Wherein the preprocessing comprises: removing abnormal signals in the sampled signal sequence, and/or supplementing missing signals in the sampled signal sequence, and/or correcting error measurement channel information.
In addition, as shown in fig. 3, a schematic structural diagram of an inversion proxy model is provided in the present disclosure. It can be seen that in one or more embodiments of the present description, the inversion proxy model may further include a third feature extraction layer, which may be identical to the second feature extraction layer, that is, including a multi-headed attention network with a residual connection structure and a forward neural network with a residual connection structure, and the multi-headed attention network with a residual connection structure may include three full connection sub-networks, and the forward neural network with a residual connection structure may include two full connection sub-networks, and a GeLU activation function may be further included in between the two full connection sub-networks.
Then in step S106, when the second signal feature is input into the result prediction layer, the computing device may first input the second signal feature into the third feature extraction layer to obtain a third signal feature, and then input the third signal feature into the result prediction layer.
Two feature extraction layers comprising a multi-head attention network with a residual connection structure and a forward neural network with a residual connection structure are constructed in the inversion proxy model, so that the accuracy of the prediction result of the inversion proxy model is further improved while the reasoning speed and the model size of the inversion proxy model are ensured to be within a reasonable range.
Furthermore, the specification also provides an application method of the inversion proxy model after training.
Specifically, the computing device may obtain a signal sequence to be predicted, where the signal sequence to be predicted is obtained by sampling soft X-rays generated by plasma to be predicted at a plurality of sampling moments, and the signal sequence to be sampled corresponding to each sampling moment is obtained by sampling soft X-rays generated by plasma to be predicted at a plurality of sampling positions. And then inputting the signal to be predicted into the trained inversion proxy model to obtain the radiation distribution profile of the plasma to be predicted, which is output by the trained inversion proxy model. And finally, monitoring the magnetic fluid state in the plasma to be predicted according to the obtained radiation distribution profile of the plasma to be predicted.
Further, in applying the trained inversion proxy model, a technical method for accelerating the reasoning speed of the model (such as using TensorRT reasoning frames) can be used to accelerate the reasoning speed of the inversion proxy model in practical application.
Based on the method, the inversion proxy model is obtained through training, the radiation distribution profile of the plasma can be obtained rapidly, the existing software framework optimized for the neural network can be utilized for operation, and the acceleration can be carried out by utilizing hardware such as a GPU (graphics processing unit) and the like, so that the reasoning efficiency of the inversion proxy model is improved. In addition, because the input sampling signal sequence is a sampling signal sequence of a plurality of sampling moments, the inversion proxy model can utilize time sequence information, so that the inversion proxy model can learn characteristics in the middle of the sampling information sequence, time sequence information and the like according to the time sequence relation among the sampling moments, the accuracy of a prediction result output by the inversion proxy model is improved, and the problem that the predicted radiation distribution profile of adjacent sampling moments fluctuates too much due to inaccurate output results of the inversion proxy model is avoided, so that the state of magnetic fluid in plasma is difficult to monitor.
Based on the above description, the embodiment of the present disclosure further correspondingly provides a schematic diagram of an inversion proxy model training device based on plasma, as shown in fig. 4.
Fig. 4 is a schematic diagram of an inversion proxy model training device based on plasma according to an embodiment of the present disclosure, where the inversion proxy model includes a first feature extraction layer, a second feature extraction layer, and a result prediction layer, the second feature extraction layer includes a multi-head attention network with a residual connection structure and a forward neural network with a residual connection structure, and the first feature extraction layer and the result prediction layer each include a full connection layer, and the device includes:
The data acquisition module 400 is configured to acquire a sampling signal sequence, where the sampling signal sequence is obtained by sampling soft X-rays generated by plasma at a plurality of sampling moments in advance, and the sampling signal sequence corresponding to each sampling moment is obtained by sampling soft X-rays generated by plasma at a plurality of sampling positions; the radiation distribution profile label of the plasma at each sampling moment is obtained;
A first feature extraction module 402, configured to input the sampled signal sequence into the first feature extraction layer to obtain a first signal feature;
A second feature extraction module 404, configured to input the first signal feature into the second feature extraction layer, so that the multi-head attention network with a residual connection structure performs weighting processing on the first signal feature, and the forward neural network with a residual connection structure performs feature extraction on the first signal feature at each sampling time to obtain a second signal feature output by the second feature extraction layer;
The result prediction module 406 is configured to input the second signal feature into the result prediction layer, and obtain a radiation distribution profile prediction result of the second signal feature corresponding to each sampling time output by the result prediction layer;
a loss determination module 408, configured to determine a loss according to the radiation distribution profile prediction result and the radiation distribution profile annotation;
a model training module 410, configured to train the inversion proxy model according to the loss.
Optionally, the first feature extraction module 402 is specifically configured to pre-process the sampled signal sequence, and input the pre-processed sampled signal sequence into the first feature extraction layer; wherein the preprocessing comprises: removing abnormal signals in the sampling signal sequence and/or supplementing missing signals in the sampling signal sequence.
Optionally, the inversion proxy model further includes a third feature extraction layer, where the third feature extraction layer includes a multi-head attention network with a residual connection structure and a forward neural network with a residual connection structure;
the result prediction module 406 is specifically configured to input the second signal feature into the third feature extraction layer to obtain a third signal feature; and inputting the third signal characteristic into the result prediction layer.
Optionally, the data acquisition module 400 is specifically configured to sample soft X-rays generated by the plasma according to a soft X-ray diagnostic system at a plurality of sampling moments.
Optionally, the loss determining module 408 is specifically configured to determine, for each sampling time, a difference between a radiation distribution profile prediction result corresponding to the sampling time and a radiation distribution profile label corresponding to the sampling time; judging whether the difference value is within a preset range or not; if yes, determining loss corresponding to the sampling moment by adopting a first formula according to the radiation distribution profile prediction result corresponding to the sampling moment and the radiation distribution profile label corresponding to the sampling moment; if not, determining the loss corresponding to the sampling moment by adopting a second formula according to the radiation distribution profile prediction result corresponding to the sampling moment and the radiation distribution profile label corresponding to the sampling moment.
Optionally, the model training module 410 is specifically configured to train the inversion proxy model according to the obtained loss corresponding to each sampling time.
Optionally, the apparatus further comprises a model application module 412;
The model application module 412 is specifically configured to obtain a signal sequence to be predicted, where the signal sequence to be predicted is obtained by sampling soft X-rays generated by plasma to be predicted at a plurality of sampling moments, and the signal sequence to be sampled corresponding to each sampling moment is obtained by sampling soft X-rays generated by plasma to be predicted at a plurality of sampling positions; inputting the signal to be predicted into a trained inversion proxy model to obtain a radiation distribution profile of the plasma to be predicted, which is output by the trained inversion proxy model; and monitoring the magnetic fluid state in the plasma to be predicted according to the obtained radiation distribution profile of the plasma to be predicted.
The embodiments of the present specification also provide a computer readable storage medium storing a computer program operable to perform the plasma-based inversion proxy model training method described above.
Based on the above-mentioned method for training the inversion proxy model based on the plasma, the embodiment of the present specification further provides a schematic structural diagram of the electronic device shown in fig. 5. At the hardware level, as in fig. 5, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile storage, although it may include hardware required for other services. The processor reads the corresponding computer program from the nonvolatile memory into the memory and then runs to realize the inversion proxy model training method based on the plasmas.
Of course, other implementations, such as logic devices or combinations of hardware and software, are not excluded from the present description, that is, the execution subject of the following processing flows is not limited to each logic unit, but may be hardware or logic devices.
In the 90 s of the 20 th century, improvements to one technology could clearly be distinguished as improvements in hardware (e.g., improvements to circuit structures such as diodes, transistors, switches, etc.) or software (improvements to the process flow). However, with the development of technology, many improvements of the current method flows can be regarded as direct improvements of hardware circuit structures. Designers almost always obtain corresponding hardware circuit structures by programming improved method flows into hardware circuits. Therefore, an improvement of a method flow cannot be said to be realized by a hardware entity module. For example, a programmable logic device (Programmable Logic Device, PLD) (e.g., field programmable gate array (Field Programmable GATE ARRAY, FPGA)) is an integrated circuit whose logic functions are determined by user programming of the device. A designer programs to "integrate" a digital system onto a PLD without requiring the chip manufacturer to design and fabricate application-specific integrated circuit chips. Moreover, nowadays, instead of manually manufacturing integrated circuit chips, such programming is mostly implemented with "logic compiler (logic compiler)" software, which is similar to the software compiler used in program development and writing, and the original code before being compiled is also written in a specific programming language, which is called hardware description language (Hardware Description Language, HDL), but HDL is not just one, but a plurality of kinds, such as ABEL(Advanced Boolean Expression Language)、AHDL(Altera Hardware Description Language)、Confluence、CUPL(Cornell University Programming Language)、HDCal、JHDL(Java Hardware Description Language)、Lava、Lola、MyHDL、PALASM、RHDL(Ruby Hardware Description Language), and VHDL (Very-High-SPEED INTEGRATED Circuit Hardware Description Language) and Verilog are currently most commonly used. It will also be apparent to those skilled in the art that a hardware circuit implementing the logic method flow can be readily obtained by merely slightly programming the method flow into an integrated circuit using several of the hardware description languages described above.
The controller may be implemented in any suitable manner, for example, the controller may take the form of, for example, a microprocessor or processor and a computer readable medium storing computer readable program code (e.g., software or firmware) executable by the (micro) processor, logic gates, switches, application SPECIFIC INTEGRATED Circuits (ASICs), programmable logic controllers, and embedded microcontrollers, examples of controllers include, but are not limited to, the following microcontrollers: ARC625D, atmel AT91SAM, microchip PIC18F26K20, and Silicone Labs C8051F320, the memory controller may also be implemented as part of the control logic of the memory. Those skilled in the art will also appreciate that, in addition to implementing the controller in a pure computer readable program code, it is well possible to implement the same functionality by logically programming the method steps such that the controller is in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers, etc. Such a controller may thus be regarded as a kind of hardware component, and means for performing various functions included therein may also be regarded as structures within the hardware component. Or even means for achieving the various functions may be regarded as either software modules implementing the methods or structures within hardware components.
The system, apparatus, module or unit set forth in the above embodiments may be implemented in particular by a computer chip or entity, or by a product having a certain function. One typical implementation is a computer. In particular, the computer may be, for example, a personal computer, a laptop computer, a cellular telephone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.
For convenience of description, the above devices are described as being functionally divided into various units, respectively. Of course, the functions of each element may be implemented in one or more software and/or hardware elements when implemented in the present specification.
It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
In one typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
The memory may include volatile memory in a computer-readable medium, random Access Memory (RAM) and/or nonvolatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of computer-readable media.
Computer readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of storage media for a computer include, but are not limited to, phase change memory (PRAM), static Random Access Memory (SRAM), dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium, which can be used to store information that can be accessed by a computing device. Computer-readable media, as defined herein, does not include transitory computer-readable media (transmission media), such as modulated data signals and carrier waves.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article or apparatus that comprises the element.
It will be appreciated by those skilled in the art that embodiments of the present description may be provided as a method, system, or computer program product. Accordingly, the present specification may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present description can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
The description may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.
In this specification, each embodiment is described in a progressive manner, and identical and similar parts of each embodiment are all referred to each other, and each embodiment mainly describes differences from other embodiments. In particular, for system embodiments, since they are substantially similar to method embodiments, the description is relatively simple, as relevant to see a section of the description of method embodiments.
The foregoing is merely exemplary of the present disclosure and is not intended to limit the disclosure. Various modifications and alterations to this specification will become apparent to those skilled in the art. Any modifications, equivalent substitutions, improvements, or the like, which are within the spirit and principles of the present description, are intended to be included within the scope of the claims of the present application.
Claims (10)
1. The inversion proxy model training method based on the plasmas is characterized by comprising a first feature extraction layer, a second feature extraction layer and a result prediction layer, wherein the second feature extraction layer comprises a multi-head attention network with a residual connection structure and a forward neural network with a residual connection structure, and the first feature extraction layer and the result prediction layer comprise full connection layers; the method comprises the following steps:
Acquiring a sampling signal sequence, wherein the sampling signal sequence is obtained by sampling soft X-rays generated by plasma at a plurality of sampling moments in advance, and the sampling signal sequence corresponding to each sampling moment is obtained by sampling the soft X-rays generated by the plasma at a plurality of sampling positions; the radiation distribution profile label of the plasma at each sampling moment is obtained;
inputting the sampling signal sequence into the first feature extraction layer to obtain a first signal feature;
Inputting the first signal characteristics into the second characteristic extraction layer, so that the multi-head attention network with the residual connection structure carries out weighting processing on the first signal characteristics, and the forward neural network with the residual connection structure carries out characteristic extraction on the first signal characteristics at each sampling moment to obtain second signal characteristics output by the second characteristic extraction layer;
Inputting the second signal characteristics into the result prediction layer to obtain a radiation distribution profile prediction result of the second signal characteristics corresponding to each sampling moment output by the result prediction layer;
determining loss according to the radiation distribution profile prediction result and the radiation distribution profile label;
And training the inversion proxy model according to the loss.
2. The method of claim 1, wherein inputting the sequence of sampled signals into the first feature extraction layer comprises:
Preprocessing the sampling signal sequence, and inputting the preprocessed sampling signal sequence into the first feature extraction layer; wherein the preprocessing comprises: removing abnormal signals in the sampling signal sequence and/or supplementing missing signals in the sampling signal sequence.
3. The method of claim 1, wherein the inversion proxy model further comprises a third feature extraction layer comprising a multi-headed attention network with residual connection and a forward neural network with residual connection;
inputting the second signal characteristic into the result prediction layer, wherein the method specifically comprises the following steps:
inputting the second signal characteristics into the third characteristic extraction layer to obtain third signal characteristics;
And inputting the third signal characteristic into the result prediction layer.
4. The method according to claim 1, characterized in that the soft X-rays generated by the plasma are sampled at several sampling instants, in particular comprising:
at a number of sampling instants, soft X-rays generated by the plasma are sampled in accordance with a soft X-ray diagnostic system.
5. The method of claim 1, wherein determining the loss based on the radiation profile prediction and the radiation profile annotation comprises:
For each sampling time, determining a difference value between a radiation distribution profile prediction result corresponding to the sampling time and a radiation distribution profile label corresponding to the sampling time;
judging whether the difference value is within a preset range or not;
if yes, determining loss corresponding to the sampling moment by adopting a first formula according to the radiation distribution profile prediction result corresponding to the sampling moment and the radiation distribution profile label corresponding to the sampling moment;
if not, determining the loss corresponding to the sampling moment by adopting a second formula according to the radiation distribution profile prediction result corresponding to the sampling moment and the radiation distribution profile label corresponding to the sampling moment.
6. The method of claim 5, wherein training the inversion proxy model based on the loss comprises:
and training the inversion proxy model according to the obtained loss corresponding to each sampling time.
7. The method of claim 1, wherein the method further comprises:
acquiring a signal sequence to be predicted, wherein the signal sequence to be predicted is obtained by sampling soft X-rays generated by plasma to be predicted at a plurality of sampling moments, and the sampling signal sequence corresponding to each sampling moment is obtained by sampling soft X-rays generated by the plasma to be predicted at a plurality of sampling positions;
Inputting the signal to be predicted into a trained inversion proxy model to obtain a radiation distribution profile of the plasma to be predicted, which is output by the trained inversion proxy model;
And monitoring the magnetic fluid state in the plasma to be predicted according to the obtained radiation distribution profile of the plasma to be predicted.
8. Inversion proxy model trainer based on plasma, its characterized in that, inversion proxy model includes first characteristic extraction layer, second characteristic extraction layer and result prediction layer, the second characteristic extraction layer is including the bull attention network that possesses residual connection structure and possess the forward neural network of residual connection structure, first characteristic extraction layer and result prediction layer all include full tie-in layer, the device specifically includes:
The data acquisition module is used for acquiring a sampling signal sequence, wherein the sampling signal sequence is obtained by sampling soft X-rays generated by plasma at a plurality of sampling moments in advance, and the sampling signal sequence corresponding to each sampling moment is obtained by sampling the soft X-rays generated by the plasma at a plurality of sampling positions; the radiation distribution profile label of the plasma at each sampling moment is obtained;
the first feature extraction module is used for inputting the sampling signal sequence into the first feature extraction layer to obtain a first signal feature;
the second feature extraction module is used for inputting the first signal features into the second feature extraction layer, so that the multi-head attention network with the residual connection structure performs weighting processing on the first signal features, and the forward neural network with the residual connection structure performs feature extraction on the first signal features at each sampling moment to obtain second signal features output by the second feature extraction layer;
The result prediction module is used for inputting the second signal characteristics into the result prediction layer to obtain a radiation distribution profile prediction result of the second signal characteristics corresponding to each sampling moment output by the result prediction layer;
the loss determination module is used for determining loss according to the radiation distribution profile prediction result and the radiation distribution profile annotation;
And the model training module is used for training the inversion proxy model according to the loss.
9. A computer readable storage medium, characterized in that the storage medium stores a computer program which, when executed by a processor, implements the method of any of the preceding claims 1-7.
10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the method of any of the preceding claims 1-7 when the program is executed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410021830.6A CN117933069B (en) | 2024-01-05 | 2024-01-05 | Inversion proxy model training method and device based on plasmas |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410021830.6A CN117933069B (en) | 2024-01-05 | 2024-01-05 | Inversion proxy model training method and device based on plasmas |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117933069A true CN117933069A (en) | 2024-04-26 |
| CN117933069B CN117933069B (en) | 2024-07-12 |
Family
ID=90765617
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410021830.6A Active CN117933069B (en) | 2024-01-05 | 2024-01-05 | Inversion proxy model training method and device based on plasmas |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117933069B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112365513A (en) * | 2020-11-23 | 2021-02-12 | 北京三快在线科技有限公司 | Model training method and device |
| CN114822541A (en) * | 2022-04-25 | 2022-07-29 | 中国人民解放军军事科学院国防科技创新研究院 | Method and system for recognizing silent voice based on back translation |
| WO2022213846A1 (en) * | 2021-04-07 | 2022-10-13 | 北京三快在线科技有限公司 | Training of recognition model |
| WO2022257840A1 (en) * | 2021-06-07 | 2022-12-15 | 维沃移动通信有限公司 | Information display method and apparatus, electronic device, and readable storage medium |
-
2024
- 2024-01-05 CN CN202410021830.6A patent/CN117933069B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112365513A (en) * | 2020-11-23 | 2021-02-12 | 北京三快在线科技有限公司 | Model training method and device |
| WO2022213846A1 (en) * | 2021-04-07 | 2022-10-13 | 北京三快在线科技有限公司 | Training of recognition model |
| WO2022257840A1 (en) * | 2021-06-07 | 2022-12-15 | 维沃移动通信有限公司 | Information display method and apparatus, electronic device, and readable storage medium |
| CN114822541A (en) * | 2022-04-25 | 2022-07-29 | 中国人民解放军军事科学院国防科技创新研究院 | Method and system for recognizing silent voice based on back translation |
Non-Patent Citations (2)
| Title |
|---|
| 余国刚;潘宇东;刘仪;: "AXUV光电二极管探测器在聚变装置HL-2A热辐射测量系统上的应用", 核电子学与探测技术, no. 06, 20 November 2006 (2006-11-20) * |
| 马瑞;王爱科;王灏;: "用人工神经网络预测HL-2A等离子体放电破裂", 核聚变与等离子体物理, no. 01, 15 March 2010 (2010-03-15) * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117933069B (en) | 2024-07-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110245047B (en) | Time sequence abnormality detection method, device and equipment | |
| Ferreira et al. | Deep learning for plasma tomography and disruption prediction from bolometer data | |
| CN115600157B (en) | Data processing method and device, storage medium and electronic equipment | |
| CN111935892B (en) | Method and apparatus for measuring plasma state | |
| CN115880685A (en) | Three-dimensional target detection method and system based on votenet model | |
| CN116230108A (en) | Intelligent medication decision method and system based on TDM and AI technologies | |
| CN117933069B (en) | Inversion proxy model training method and device based on plasmas | |
| CN116167431B (en) | Service processing method and device based on hybrid precision model acceleration | |
| CN117828360A (en) | Model training method, model training device, model code generating device, storage medium and storage medium | |
| CN117409466A (en) | Three-dimensional dynamic expression generation method and device based on multi-label control | |
| CN117933070B (en) | Training method, device and storage medium of radiation distribution prediction model | |
| CN117357132B (en) | Task execution method and device based on multi-layer brain network node participation coefficient | |
| CN117932406A (en) | Training method and device for classification model applied to brain-computer interface task | |
| CN117556266B (en) | Signal identification model training method, signal identification method and device | |
| CN116309924B (en) | Model training method, image display method and device | |
| CN118131159A (en) | Asynchronous track association method, device and system based on pseudo nearest neighbor distance | |
| CN118378726B (en) | Model training system, method, storage medium and electronic equipment | |
| CN116579414B (en) | Model training method, MRI thin layer data reconstruction method, device and equipment | |
| CN117975202B (en) | Model training method, service execution method, device, medium and equipment | |
| CN117455015B (en) | Model optimization method and device, storage medium and electronic equipment | |
| CN116224279B (en) | Target detection method and device, storage medium and electronic equipment | |
| CN117131633B (en) | Robot structural member optimization method and device based on heat conduction constitutive model | |
| CN117952182B (en) | Mixed precision model training method and device based on data quality | |
| CN117875789B (en) | Data processing method and device for sponge city drainage effect evaluation | |
| CN117726760A (en) | Training method and device for three-dimensional human body reconstruction model of video |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |