CN117933070A - Training method, device and storage medium of radiation distribution prediction model - Google Patents

Training method, device and storage medium of radiation distribution prediction model Download PDF

Info

Publication number
CN117933070A
CN117933070A CN202410022577.6A CN202410022577A CN117933070A CN 117933070 A CN117933070 A CN 117933070A CN 202410022577 A CN202410022577 A CN 202410022577A CN 117933070 A CN117933070 A CN 117933070A
Authority
CN
China
Prior art keywords
radiation distribution
weight
loss
feature
prediction model
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
Application number
CN202410022577.6A
Other languages
Chinese (zh)
Other versions
CN117933070B (en
Inventor
杨仁杰
王智君
魏一雄
张泽宇
王聪
王海平
魏宁
高超霖
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang Lab
Original Assignee
Zhejiang Lab
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Zhejiang Lab filed Critical Zhejiang Lab
Priority to CN202410022577.6A priority Critical patent/CN117933070B/en
Publication of CN117933070A publication Critical patent/CN117933070A/en
Application granted granted Critical
Publication of CN117933070B publication Critical patent/CN117933070B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01T—MEASUREMENT OF NUCLEAR OR X-RADIATION
    • G01T1/00—Measuring X-radiation, gamma radiation, corpuscular radiation, or cosmic radiation
    • G01T1/16—Measuring radiation intensity
    • 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
    • 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
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • 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
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • 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
    • G06F2123/00—Data types
    • G06F2123/02—Data types in the time domain, e.g. time-series data
    • 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)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Computing Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Medical Informatics (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • High Energy & Nuclear Physics (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Analysing Materials By The Use Of Radiation (AREA)

Abstract

The specification discloses a training method, a training device and a storage medium of a radiation distribution prediction model, wherein the radiation distribution prediction model comprises a first feature extraction layer, a first weight adjustment layer and a radiation distribution output layer, features of a sampling signal sequence are extracted through the first feature extraction layer, and the features extracted by the first feature extraction layer contain time information. In addition, the first weight adjustment layer can be used for adjusting the weight of the extracted features so as to improve the accuracy of the output predicted radiation distribution. Finally, the time sequence features are extracted through the radiation distribution output layer, and the time sequence features are extracted because the sampling signal sequence comprises time information, so that the time sequence evolution rule of the radiation distribution can be obtained. Based on the time series characteristics, a predicted radiation distribution image of the plasma corresponding to each predicted time is obtained. The radiation distribution image is constructed through the radiation distribution prediction model, so that the efficiency and the accuracy of constructing the radiation distribution image are improved.

Description

Training method, device and storage medium of radiation distribution prediction model
Technical Field
The present disclosure relates to the field of nuclear fusion, and in particular, to a method and apparatus for training a radiation distribution prediction model, and a storage medium.
Background
The energy generated by the nuclear fusion reaction is clean energy, and more energy is released by the nuclear fusion reaction, so that more available energy is obtained. Despite the great potential of nuclear fusion, the current implementation of controlled nuclear fusion still faces technical challenges.
In fusion plasma research, the plasma can be stably confined in a limited volume by a magnetic field, and enough temperature and pressure conditions are created so that a nuclear fusion reaction occurs. While maintaining the stability of the plasma is critical in producing nuclear fusion reactions, the plasma is susceptible to the interference of various magnetic fluid (Magnetohydrodynamics, MHD) activities, resulting in plasma instability. The energy released by the fusion reaction is expressed in the form of plasma discharge, and the unstable plasma may cause unstable plasma discharge or failure of discharge, i.e., failure of the fusion reaction. In the field of plasma diagnostics, soft X-band radiation of a polar cross-section of a plasma can be measured by a soft X-ray diagnostic system, which can be observed by constructing a radiation distribution image of the cross-section of the plasma to monitor the stability of the plasma. At present, a radiation distribution image of a section of the plasma can be constructed by a Bayesian-based Non-stationary Gaussian process tomography (Non-Stationary Gaussian Process Tomography, NSGPT) method, but the time is long, and the stability of the plasma cannot be monitored in real time.
Based on this, the present specification provides a training method of a radiation distribution prediction model.
Disclosure of Invention
The present disclosure provides a method and apparatus for training a radiation distribution prediction model, a storage medium, and an electronic device, so as to partially solve the foregoing problems in the prior art.
The technical scheme adopted in the specification is as follows:
the specification provides a training method of a radiation distribution prediction model, wherein the radiation distribution prediction model comprises a first feature extraction layer, a first weight adjustment layer and a radiation distribution output layer, and the method comprises the following steps:
Acquiring a sampling signal sequence, wherein the sampling signal sequence comprises soft X-ray signals at a plurality of moments, and the soft X-ray signals are obtained by sampling soft X-rays generated by plasma at preset moments; acquiring a radiation distribution image of the plasma at each moment, and taking the radiation distribution image as a first tag of the sampling signal sequence;
inputting the sampling signal sequence into the first feature extraction layer to obtain a first initial feature;
Inputting the first initial feature into the first weight adjustment layer so that the first weight adjustment layer adjusts the weight of the first initial feature to obtain an adjusted first feature output by the first weight adjustment layer as a first adjustment feature;
Inputting the first adjustment feature into the radiation distribution output layer so that the radiation distribution output layer extracts time sequence features in the first adjustment feature, and obtaining a predicted radiation distribution image of the plasma corresponding to each moment output by the radiation distribution output layer according to the time sequence features;
And training the radiation distribution prediction model according to the predicted radiation distribution image and the first label.
Optionally, the radiation distribution prediction model further comprises a second feature extraction layer; the method further comprises the steps of:
inputting the sampling signal sequence into the second feature extraction layer to obtain a second initial feature;
Inputting the first initial feature into the first weight adjustment layer, so that the first weight adjustment layer adjusts the weight of the first initial feature to obtain an adjusted first feature output by the first weight adjustment layer, wherein the adjusted first feature is used as a first adjustment feature and specifically comprises the following steps:
Inputting the first initial feature and the second initial feature into the first weight adjustment layer, so that the first weight adjustment layer adjusts the weight of the first initial feature and adjusts the weight of the second initial feature to obtain an adjusted first initial feature and an adjusted second initial feature which are output by the first weight adjustment layer;
And determining a first adjustment feature according to the adjusted first initial feature and the adjusted second initial feature output by the first weight adjustment layer.
Optionally, the radiation distribution prediction model further comprises a second weight adjustment layer; the method further comprises the steps of:
inputting the first initial feature and the second initial feature into the second weight adjustment layer, so that the second weight adjustment layer adjusts the weight of the first initial feature and adjusts the weight of the second initial feature to obtain an adjusted first initial feature and an adjusted second initial feature which are output by the second weight adjustment layer;
And determining a second adjustment feature according to the adjusted first initial feature and the adjusted second initial feature output by the second weight adjustment layer.
Optionally, the radiation distribution prediction model further comprises a magnetic fluid state prediction output layer; the method further comprises the steps of:
Acquiring the state of magnetic fluid in the plasma at each moment, wherein the state of the magnetic fluid comprises stable and unstable as a second label of the sampling signal sequence;
Inputting the second adjustment characteristic into a magnetic fluid state prediction output layer so that the magnetic fluid state prediction output layer outputs a predicted magnetic fluid state corresponding to each moment;
Training the radiation distribution prediction model according to the predicted radiation distribution image and the first label, specifically including:
and training the radiation distribution prediction model according to the predicted radiation distribution image, the first label, the predicted magnetic fluid state and the second label.
Optionally, training the radiation distribution prediction model according to the predicted radiation distribution image, the first tag, the predicted magnetic fluid state and the second tag specifically includes:
Determining a first loss at each moment of each iteration according to the predicted radiation distribution image and the first label;
determining a first loss change rate according to the first loss of the current iteration and the first loss of the last iteration of the current iteration;
And training the radiation distribution prediction model according to the first loss change rate, the predicted magnetic fluid state and the second label.
Optionally, training the radiation distribution prediction model according to the first loss change rate, the predicted magnetic fluid state and the second label specifically includes:
Determining a second loss at each moment of each iteration according to the predicted magnetic fluid state and the second label;
Determining a second loss change rate according to the second loss of the current iteration and the second loss of the last iteration of the current iteration;
And training the radiation distribution prediction model according to the first loss change rate and the second loss change rate.
Optionally, training the radiation distribution prediction model according to the first loss change rate and the second loss change rate specifically includes:
When the first loss change rate is larger than the second loss change rate, determining the weight of the first loss as a first weight, and determining the weight of the second loss as a second weight, wherein the first weight is smaller than the second weight;
determining a final loss according to the first weight, the second weight, the first loss and the second loss;
And training the radiation distribution prediction model according to the final loss.
Optionally, the method further comprises:
Obtaining soft X-ray signals to be predicted at a plurality of moments to obtain a sampling signal sequence to be predicted, wherein the soft X-ray signals to be predicted are obtained by sampling soft X-rays generated by plasma at preset moments;
Inputting the sampling signal sequence to be predicted into a radiation distribution prediction model after training, and obtaining a radiation distribution image corresponding to each moment of the plasma output by the radiation distribution prediction model after training;
And determining the state of the plasma according to the radiation distribution image corresponding to each moment of the plasma output by the trained radiation distribution prediction model, wherein the state of the plasma comprises stable and unstable.
The present specification provides a training device of a radiation distribution prediction model, the radiation distribution prediction model including a first feature extraction layer, a first weight adjustment layer, and a radiation distribution output layer, the device including:
the sampling signal sequence acquisition module is used for acquiring a sampling signal sequence, wherein the sampling signal sequence comprises soft X-ray signals at a plurality of moments, and the soft X-ray signals are obtained by sampling soft X-rays generated by plasma at preset moments; acquiring a radiation distribution image of the plasma at each moment, and taking the radiation distribution image as a first tag of the sampling signal sequence;
the first initial feature determining module is used for inputting the sampling signal sequence into the first feature extracting layer to obtain a first initial feature;
The first adjustment feature determining module is used for inputting the first initial feature into the first weight adjustment layer so that the first weight adjustment layer adjusts the weight of the first initial feature to obtain an adjusted first feature output by the first weight adjustment layer as a first adjustment feature;
The output module is used for inputting the first adjustment feature into the radiation distribution output layer so that the radiation distribution output layer extracts the time sequence feature in the first adjustment feature, and a predicted radiation distribution image of the plasma corresponding to each moment output by the radiation distribution output layer is obtained according to the time sequence feature;
And the training module is used for training the radiation distribution prediction model according to the predicted radiation distribution image and the first label.
Optionally, the radiation distribution prediction model further comprises a second feature extraction layer; the first adjustment feature determining module is specifically configured to input the sampling signal sequence into the second feature extraction layer to obtain a second initial feature; inputting the first initial feature and the second initial feature into the first weight adjustment layer, so that the first weight adjustment layer adjusts the weight of the first initial feature and adjusts the weight of the second initial feature to obtain an adjusted first initial feature and an adjusted second initial feature which are output by the first weight adjustment layer; and determining a first adjustment feature according to the adjusted first initial feature and the adjusted second initial feature output by the first weight adjustment layer.
Optionally, the radiation distribution prediction model further comprises a second weight adjustment layer; the apparatus further comprises:
The second adjustment feature determining module is used for inputting the first initial feature and the second initial feature into the second weight adjustment layer so that the second weight adjustment layer adjusts the weight of the first initial feature and adjusts the weight of the second initial feature to obtain an adjusted first initial feature and an adjusted second initial feature which are output by the second weight adjustment layer; and determining a second adjustment feature according to the adjusted first initial feature and the adjusted second initial feature output by the second weight adjustment layer.
Optionally, the radiation distribution prediction model further comprises a magnetic fluid state prediction output layer; the training module is specifically configured to obtain a state of magnetic fluid in the plasma at each moment, where the state of the magnetic fluid includes stable and unstable, as a second tag of the sampling signal sequence; inputting the second adjustment characteristic into a magnetic fluid state prediction output layer so that the magnetic fluid state prediction output layer outputs a predicted magnetic fluid state corresponding to each moment; and training the radiation distribution prediction model according to the predicted radiation distribution image, the first label, the predicted magnetic fluid state and the second label.
Optionally, the training module is specifically configured to determine, according to the predicted radiation distribution image and the first label, a first loss at each moment of each iteration; determining a first loss change rate according to the first loss of the current iteration and the first loss of the last iteration of the current iteration; and training the radiation distribution prediction model according to the first loss change rate, the predicted magnetic fluid state and the second label.
Optionally, the training module is specifically configured to determine, according to the predicted magnetic fluid state and the second tag, a second loss at each moment of each iteration; determining a second loss change rate according to the second loss of the current iteration and the second loss of the last iteration of the current iteration; and training the radiation distribution prediction model according to the first loss change rate and the second loss change rate.
Optionally, the training module is specifically configured to determine, when the first loss change rate is greater than the second loss change rate, that the weight of the first loss is a first weight, and determine that the weight of the second loss is a second weight, where the first weight is less than the second weight; determining a final loss according to the first weight, the second weight, the first loss and the second loss; and training the radiation distribution prediction model according to the final loss.
Optionally, the apparatus further comprises:
The prediction module is used for obtaining soft X-ray signals to be predicted at a plurality of moments to obtain a sampling signal sequence to be predicted, wherein the soft X-ray signals to be predicted are obtained by sampling soft X-rays generated by plasma at preset moments; inputting the sampling signal sequence to be predicted into a radiation distribution prediction model after training, and obtaining a radiation distribution image corresponding to each moment of the plasma output by the radiation distribution prediction model after training; and determining the state of the plasma according to the radiation distribution image corresponding to each moment of the plasma output by the trained radiation distribution prediction model, wherein the state of the plasma comprises stable and unstable.
The present specification provides a computer readable storage medium storing a computer program which when executed by a processor implements the above-described method of training a radiation distribution prediction model.
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 a training method of the radiation distribution prediction model described above when executing the program.
The above-mentioned at least one technical scheme that this specification adopted can reach following beneficial effect:
As can be seen from the training method of the radiation distribution prediction model provided in the present specification, the radiation distribution prediction model includes a first feature extraction layer, a first weight adjustment layer, and a radiation distribution output layer, features of the sampling signal sequence are extracted by the first feature extraction layer, and the features extracted by the first feature extraction layer include time information. In addition, the first weight adjustment layer can be used for adjusting the weight of the extracted features so as to improve the accuracy of the output predicted radiation distribution. Finally, the time sequence features are extracted through the radiation distribution output layer, and because the sampling signal sequence comprises time information, the time sequence evolution rule of the radiation distribution can be obtained, the time sequence features are extracted, and the predicted radiation distribution image of the plasma corresponding to each predicted moment is obtained according to the time sequence features. The radiation distribution image is constructed through the radiation distribution prediction model, so that the efficiency and the accuracy of constructing the radiation distribution image are improved.
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 a training method of a radiation distribution prediction model provided in the present specification;
FIG. 2 is a schematic structural diagram of a radiation distribution prediction model provided in the present specification;
FIG. 3 is a schematic structural diagram of a first feature extraction layer provided in the present specification;
fig. 4 is a schematic structural diagram of a first weight adjustment layer provided in the present disclosure;
FIG. 5 is a schematic view of the structure of the radiation distribution output layer provided in the present specification;
FIG. 6 is a schematic diagram of a time module structure provided in the present specification;
FIG. 7 is a schematic diagram of another radiation distribution prediction model provided in the present specification;
FIG. 8 is a schematic diagram of the attention module provided in the present specification;
FIG. 9 is a schematic diagram of a training apparatus for a radiation distribution prediction model provided herein;
fig. 10 is a schematic structural diagram of an 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 a training method of a radiation distribution prediction model provided in the present specification, including the following steps:
S100: acquiring a sampling signal sequence, wherein the sampling signal sequence comprises soft X-ray signals at a plurality of moments, and the soft X-ray signals are obtained by sampling soft X-rays generated by plasma at preset moments; and acquiring a radiation distribution image of the plasma at each moment as a first tag of the sampling signal sequence.
Nuclear energy is available through nuclear fusion reactions, in which the nuclei of light elements combine to form heavier elements, accompanied by the release of a large amount of energy, the source of nuclear energy. To cause nuclear fusion to occur, i.e., to cause a stable discharge of the plasma, it is necessary to cause the plasma to be in a stable operating state. Today, it is possible to determine whether the plasma is in a steady state by the state of the magnetic fluid (MHD) within the plasma. While the state of the MHD can be determined by the radiation distribution image of the profile of the plasma. Therefore, the state of the plasma can be judged in an assisted manner by the radiation distribution image of the cross section of the plasma.
The radiation distribution in the radiation distribution image of the profile of the plasma can be derived from soft X-ray signals detected by a soft X-ray diagnostic system which can be built in a tokamak device for generating nuclear fusion reactions. The soft X-ray diagnostic system has a plurality of detector arrays, each array having a plurality of view lines for measuring soft X-band radiation of the plasma profile, i.e. plasma radiation of the core. Soft X-ray diagnostic systems typically operate at high sampling rates, from a few kilohertz to hundreds of kilohertz, typical tokamak devices discharge HL-2A pulses typically lasting about seconds to tens of seconds, and the radiation profile image of the profile of the plasma constructed by NSGPT takes a long time, and thus, currently, real-time detection of plasma stability cannot be achieved. The radiation distribution prediction model obtained by the training method builds a radiation distribution image, improves the efficiency and accuracy of building the radiation distribution image, and realizes the real-time detection of the state of plasma. The subject matter of the present specification can be an electronic device, such as a server, terminal, etc., that trains a radiation distribution prediction model, which can communicate with a soft X-ray diagnostic system. The present disclosure is not limited to computing devices with computing capabilities that deploy radiation distribution prediction models and soft X-ray diagnostic systems, and the present disclosure is directed to computing devices as the main execution body for ease of description.
Since MHD instability in the plasma is a high frequency periodic physical phenomenon, soft X-ray signals can be acquired over a period of time to improve the accuracy of the predicted radiation distribution. Then, the detector in the soft X-ray diagnostic system may acquire soft X-ray signals at several moments that constitute a sampling signal sequence, which soft X-ray signals are obtained by sampling soft X-rays generated by the plasma at preset moments. That is, the sampling signal sequence can be obtained by measuring a plurality of detectors at different positions in a soft X-ray diagnosis system at a plurality of time points, and the sampling signal sequence can provide important information about the state of magnetic fluid in the fusion process.
The soft X-ray diagnostic system may then send the sampled signal sequence to a computing device that receives the sampled signal sequence sent by the soft X-ray diagnostic system, i.e., acquires the sampled signal sequence.
The computing device may also acquire a radiation distribution image of the plasma at each instant as a first tag of the sampled signal sequence for subsequent training of the radiation distribution prediction model in accordance with the first tag.
S102: and inputting the sampling signal sequence into the first feature extraction layer to obtain a first initial feature.
Fig. 2 is a schematic structural diagram of a radiation distribution prediction model provided in the present specification, as shown in fig. 2.
The radiation distribution prediction model comprises a first feature extraction layer, a first weight adjustment layer and a radiation distribution output layer. After the computing device acquires the sampling signal sequence, the sampling signal sequence can be input into a first feature extraction layer in the radiation distribution prediction model, and the first feature extraction layer extracts and outputs features in the sampling signal sequence to obtain first initial features.
Fig. 3 is a schematic structural diagram of the first feature extraction layer provided in the present specification, as shown in fig. 3.
The first feature extraction layer may be an Expert network (Expert), and the first feature extraction layer includes an activation function, a linear layer, and a random inactivation layer (Dropout), where the activation function may be a Relu activation function, which is not limited in this specification. The random inactivation layer can prevent overfitting and improve the generalization performance of the radiation distribution prediction model.
S104: and inputting the first initial feature into the first weight adjustment layer so that the first weight adjustment layer adjusts the weight of the first initial feature to obtain an adjusted first feature output by the first weight adjustment layer as a first adjustment feature.
Fig. 4 is a schematic structural diagram of the first weight adjustment layer provided in the present disclosure, as shown in fig. 4.
The first weight adjustment layer includes an activation function and a linear layer, and the first weight adjustment layer may include a gating network, which is not limited in this specification. The first initial feature is weighted, so that features which are more important for improving the model precision are focused in the model training process, and a radiation distribution prediction model with higher precision is obtained.
S106: and inputting the first adjustment feature into the radiation distribution output layer so that the radiation distribution output layer extracts the time sequence feature in the first adjustment feature, and obtaining a predicted radiation distribution image of the plasma corresponding to each moment output by the radiation distribution output layer according to the time sequence feature.
Fig. 5 is a schematic structural view of the radiation distribution output layer provided in the present specification, as shown in fig. 5.
The radiation distribution output layer comprises a number of time modules (TimeModel) and a number of normalization layers (Layer Normalization, LN). And extracting time sequence characteristics in the first adjustment characteristics through the time module of the radiation output layer, and obtaining a predicted radiation distribution image of the plasma corresponding to each moment output by the radiation distribution output layer according to the time sequence characteristics.
Specifically, fig. 6 is a schematic view of a time module structure provided in the present specification, as shown in fig. 6.
The time module comprises a multi-layer perceptron and an expansion convolution encoder module (DilatedConvEncoder), wherein the expansion convolution encoder module comprises a plurality of convolution blocks (ConvBlock), and each convolution block comprises a plurality of cavity convolution modules (DilatedConv d), a plurality of projection layers (Glue) and a residual connection structure (Conv 1 d).
After the first adjustment feature is input into the radiation distribution output layer, the first adjustment feature is processed by the expansion convolution encoder module through the multi-layer perceptron to obtain a final predicted radiation distribution image. Wherein the dilated convolutional encoder module is capable of effectively capturing local features in the time series data to provide more detailed time series information
The dilated convolutional encoder module processes the first adjustment feature by a series of convolutional blocks with different dilation rates, and the last convolutional block can be mapped by a linear layer to generate a feature that is adapted to the final output dimension. The expansion rate is the interval between values when the convolution kernel processes data, and represents the expansion size.
In the convolution block, nonlinear transformation is carried out on the first adjustment feature through a plurality of cavity convolution modules and activation functions which are filled in the same way, then a residual error connection structure is used for adding the convolution result and the first adjustment feature so as to keep low-level features, and finally the number of channels is adjusted through a projection layer so that the input and output dimensions are consistent. The residual connection can avoid the problem of gradient disappearance or gradient explosion, enhance the transmission and utilization effects of the characteristics, and flexibly adjust the number of input and output channels according to the needs of the projection layer so as to adapt to the requirements of different tasks.
For each hole convolution module, the computing device may calculate an appropriate amount of padding based on the convolution kernel size, the expansion ratio, and the number of input channels to keep the input and output shapes the same. That is, the cavity convolution module can make the input and output shapes consistent, and stability and accuracy of processing time sequence data are improved. In addition, the cavity convolution module can increase the receptive field, so that each convolution output contains a larger range of information, and the accuracy of the output predicted radiation distribution image is improved.
S108: and training the radiation distribution prediction model according to the predicted radiation distribution image and the first label.
Specifically, the difference between the predicted radiation distribution image and the first label is determined, a loss is determined according to the difference, the loss is reduced as a training target, and the radiation distribution prediction model is trained.
The method for training the radiation distribution prediction model shown in fig. 1 comprises a first feature extraction layer, a first weight adjustment layer and a radiation distribution output layer, wherein features of the sampling signal sequence are extracted through the first feature extraction layer, and the features extracted by the first feature extraction layer comprise time information. In addition, the first weight adjustment layer can be used for adjusting the weight of the extracted features so as to improve the accuracy of the output predicted radiation distribution. Finally, the time sequence features are extracted through the radiation distribution output layer, and because the sampling signal sequence comprises time information, the time sequence evolution rule of the radiation distribution can be obtained, the time sequence features are extracted, and the predicted radiation distribution image of the plasma corresponding to each predicted moment is obtained according to the time sequence features. The radiation distribution image is constructed through the radiation distribution prediction model, so that the efficiency and the accuracy of constructing the radiation distribution image are improved.
After the radiation distribution prediction model is trained, the computing equipment can firstly acquire soft X-ray signals to be predicted at a plurality of moments to obtain a sampling signal sequence to be predicted, wherein the soft X-ray signals to be predicted are obtained by sampling soft X-rays generated by plasma at preset moments. And inputting the sampling signal sequence to be predicted into a radiation distribution prediction model after training, and obtaining a radiation distribution image corresponding to each moment of the plasma output by the radiation distribution prediction model after training. And finally, determining the state of the plasma according to the radiation distribution image corresponding to each moment of the plasma output by the trained radiation distribution prediction model, wherein the state of the plasma comprises stable and unstable.
Fig. 7 is a schematic structural diagram of another radiation distribution prediction model provided in the present specification, as shown in fig. 7.
The radiation distribution prediction model can output a predicted radiation distribution image and a predicted magnetic fluid state, and in the radiation distribution prediction model, the two output results are related, so that two tasks are trained in the predicted radiation distribution image simultaneously, training of the two tasks can be mutually promoted, and the obtained predicted radiation distribution image and the predicted magnetic fluid state are high in accuracy. Then, the specification can train the model using a Multi-tasking framework (Multi-gate Mixture of Experts, MMoE) as shown in fig. 7.
As shown in fig. 7, the radiation distribution prediction model may further include a second feature extraction layer, a second weight adjustment layer, and a magnetic fluid state prediction output layer. The structure of the second feature extraction layer may be identical to that of the first feature extraction layer, and the structure of the second weight adjustment layer may be identical to that of the first weight adjustment layer, which is not limited in this specification. The magnetic fluid state prediction output layer comprises a time module, a plurality of linear layers, a plurality of attention modules and an activation function.
Then, for step S104, the computing device may input the sequence of sampled signals into the second feature extraction layer, resulting in a second initial feature. And inputting the first initial feature and the second initial feature into the first weight adjustment layer so that the first weight adjustment layer adjusts the weight of the first initial feature and adjusts the weight of the second initial feature to obtain an adjusted first initial feature and an adjusted second initial feature which are output by the first weight adjustment layer. Finally, determining a first adjustment feature according to the adjusted first initial feature and the adjusted second initial feature output by the first weight adjustment layer, so as to execute step S106.
It should be noted that, the computing device may perform feature stitching on the adjusted first initial feature and the adjusted second initial feature output by the first weight adjustment layer to obtain the first adjustment feature, or may use other manners to obtain the first adjustment feature through the adjusted first initial feature and the adjusted second initial feature output by the first weight adjustment layer. The adjusted first initial feature and the adjusted second initial feature output by the first weight adjustment layer can also be directly input into the radiation distribution output layer, which is not limited in this specification.
Further, in order to enable the radiation distribution prediction model to output a predicted magnetic fluid state, the computing device may input the first initial feature and the second initial feature into the second weight adjustment layer first, so that the second weight adjustment layer adjusts the weight of the first initial feature, adjusts the weight of the second initial feature, obtains an adjusted first initial feature and an adjusted second initial feature output by the second weight adjustment layer, and determines a second adjustment feature according to the adjusted first initial feature and the adjusted second initial feature output by the second weight adjustment layer. The present disclosure is not limited to how to determine the second adjustment feature according to the adjusted first initial feature and the adjusted second initial feature output by the second weight adjustment layer.
And then, acquiring the state of magnetic fluid in the plasma at each moment, wherein the state of the magnetic fluid comprises stable and unstable as a second label of the sampling signal sequence. And then, inputting the second adjustment characteristic into a magnetic fluid state prediction output layer so that the magnetic fluid state prediction output layer outputs a predicted magnetic fluid state corresponding to each moment.
After the computing equipment inputs the second adjustment characteristic into the magnetic fluid state prediction output layer, the computing equipment captures the local characteristic in the time sequence data through a time module in the magnetic fluid state prediction output layer and provides more detailed time sequence information. And then the final predicted magnetic fluid state is obtained through treatment of a normalization layer, a random inactivation layer, an attention module, a linear layer and an activation function, wherein the activation function can be GELU, and the specification is not limited by the specification. The normalization layer and GELU activation functions are beneficial to improving the robustness and generalization capability of the model, and the random inactivation layer can prevent overfitting and improve the generalization performance of the model.
Fig. 8 is a schematic structural view of the attention module provided in the present specification, as shown in fig. 8.
The attention module includes a matrix multiplication (Matmul) function, a scaling layer, an activation function, which may be SoftMax, which is not limited in this specification. Through the attention module, the model can automatically learn important parts in time sequence data, better distinguish different types of characteristics and improve the accuracy of predicting the state of the magnetic fluid.
For step S108, the computing device may train the radiation distribution prediction model according to the predicted radiation distribution image, the first tag, the predicted magnetic fluid state, and the second tag.
Specifically, the computing device may determine a first loss at each time of each iteration based on the predicted radiation distribution image and the first tag, and determine a second loss at each time of each iteration based on the predicted magnetic fluid state and the second tag. And determining a final loss according to the first loss and the second loss. The radiation distribution prediction model is trained with the reduction of the final loss as a training target. The first loss and the second loss may be weighted and averaged to obtain a final loss, or the first loss and the second loss may be directly added to obtain a final loss, which is not limited in this specification.
It should be noted that, since the radiation distribution prediction model may perform two tasks, the training effects of the two tasks may not be synchronized during the training process. In order to synchronize the training effects of the two tasks of the radiation distribution prediction model, i.e. to enable the two tasks of the radiation distribution prediction model to reach the training target at the same time, different weights may be assigned to the first loss and the second loss.
Then, the computing device may determine a first loss rate of change based on the first loss for the current iteration and the first loss for the last iteration of the current iteration. And determining a second loss change rate according to the second loss of the current iteration and the second loss of the last iteration of the current iteration. And determining convergence speeds of the two tasks by judging the first loss change rate and the second loss change rate so as to adjust weights of the two losses in the current iteration, so that the two tasks reach a training target.
When the first loss change rate is greater than the second loss change rate, determining that the weight of the first loss is a first weight, and determining that the weight of the second loss is a second weight, wherein the first weight is smaller than the second weight. The first weight is greater than the second weight when the first loss rate of change is less than the second loss rate of change, and the first weight is equal to the second weight when the first loss rate of change is equal to the second loss rate of change.
That is, the loss change rate is inversely related to the loss weight, and the greater the loss change rate, the smaller the weight is allocated to the loss of the task, so that the model can pay more attention to another task with larger weight, and thus, the two tasks can reach the training target simultaneously. The training target may be the accuracy of the results output by the model, which is not limited in this specification.
Finally, determining a final loss according to the first weight, the second weight, the first loss and the second loss, and training the radiation distribution prediction model according to the final loss. Of course, for each iteration, the computing device may perform weight distribution by the foregoing loss weight distribution method.
Different training strategies can be adopted at different stages of model training, and the stages can be divided according to the iteration times of the model. For example, the first 500 iterations are initial phases, the next 200 iterations are intermediate phases, and the last 50 iterations are final phases. Of course, the division may be performed according to the accuracy of the model output result, for example, the accuracy of the model output result reaches 50% as an initial stage, the accuracy of the model output result reaches 80% as an intermediate stage, and the accuracy of the model output result reaches 90% as a final stage.
When the radiation distribution prediction model is trained, the computing equipment can adopt a cosine annealing training method, and different training strategies can be adopted in different training stages of the radiation distribution prediction model through the method. I.e. a larger learning rate is used for rapid convergence in the initial stage, and the learning rate is gradually reduced in each training stage according to the form of a cosine curve. In other words, the learning rate is gradually reduced as training proceeds to improve the stability and generalization ability of the model. In the intermediate stage, the learning rate is adjusted using a cosine function to smoothly change the magnitude of the learning rate in different stages.
The learning rate can also be changed according to the final loss in the training process, namely, a larger learning rate is used in the first two tasks to quickly converge to a better model state, and when the loss is reduced to a certain degree, the learning rate is halved so as to more finely adjust model parameters in the later stage of training.
The foregoing is a schematic flow chart of a training method of the radiation distribution prediction model shown in fig. 1, and the present disclosure further provides a training device of the corresponding radiation distribution prediction model, as shown in fig. 9.
Fig. 9 is a schematic diagram of a training device for a radiation distribution prediction model provided in the present specification, including:
the sampling signal sequence acquisition module 900 is configured to acquire a sampling signal sequence, where the sampling signal sequence includes soft X-ray signals at a plurality of moments, where the soft X-ray signals are obtained by sampling soft X-rays generated by plasma at preset moments; acquiring a radiation distribution image of the plasma at each moment, and taking the radiation distribution image as a first tag of the sampling signal sequence;
a first initial feature determining module 902, configured to input the sampled signal sequence into the first feature extraction layer to obtain a first initial feature;
A first adjustment feature determining module 904, configured to input the first initial feature into the first weight adjustment layer, so that the first weight adjustment layer adjusts the weight of the first initial feature, to obtain an adjusted first feature output by the first weight adjustment layer, as a first adjustment feature;
An output module 906, configured to input the first adjustment feature into the radiation distribution output layer, so that the radiation distribution output layer extracts a time sequence feature in the first adjustment feature, and obtain a predicted radiation distribution image of the plasma corresponding to each moment output by the radiation distribution output layer according to the time sequence feature;
The training module 908 is configured to train the radiation distribution prediction model according to the predicted radiation distribution image and the first label.
Optionally, the radiation distribution prediction model further comprises a second feature extraction layer; the first adjustment feature determining module 904 is specifically configured to input the sampling signal sequence into the second feature extraction layer to obtain a second initial feature; inputting the first initial feature and the second initial feature into the first weight adjustment layer, so that the first weight adjustment layer adjusts the weight of the first initial feature and adjusts the weight of the second initial feature to obtain an adjusted first initial feature and an adjusted second initial feature which are output by the first weight adjustment layer; and determining a first adjustment feature according to the adjusted first initial feature and the adjusted second initial feature output by the first weight adjustment layer.
Optionally, the radiation distribution prediction model further comprises a second weight adjustment layer; the apparatus further comprises:
A second adjustment feature determining module 910, configured to input the first initial feature and the second initial feature into the second weight adjustment layer, so that the second weight adjustment layer adjusts the weight of the first initial feature, and adjusts the weight of the second initial feature, to obtain an adjusted first initial feature and an adjusted second initial feature output by the second weight adjustment layer; and determining a second adjustment feature according to the adjusted first initial feature and the adjusted second initial feature output by the second weight adjustment layer.
Optionally, the radiation distribution prediction model further comprises a magnetic fluid state prediction output layer; the training module 908 is specifically configured to obtain, as the second tag of the sampling signal sequence, a state of a magnetic fluid in the plasma at each moment, where the state of the magnetic fluid includes stable and unstable; inputting the second adjustment characteristic into a magnetic fluid state prediction output layer so that the magnetic fluid state prediction output layer outputs a predicted magnetic fluid state corresponding to each moment; and training the radiation distribution prediction model according to the predicted radiation distribution image, the first label, the predicted magnetic fluid state and the second label.
Optionally, the training module 908 is specifically configured to determine, according to the predicted radiation distribution image and the first label, a first loss at each moment in each iteration; determining a first loss change rate according to the first loss of the current iteration and the first loss of the last iteration of the current iteration; and training the radiation distribution prediction model according to the first loss change rate, the predicted magnetic fluid state and the second label.
Optionally, the training module 908 is specifically configured to determine, according to the predicted magnetic fluid state and the second tag, a second loss at each moment of each iteration; determining a second loss change rate according to the second loss of the current iteration and the second loss of the last iteration of the current iteration; and training the radiation distribution prediction model according to the first loss change rate and the second loss change rate.
Optionally, the training module 908 is specifically configured to determine, when the first loss change rate is greater than the second loss change rate, that the weight of the first loss is a first weight, and determine that the weight of the second loss is a second weight, where the first weight is less than the second weight; determining a final loss according to the first weight, the second weight, the first loss and the second loss; and training the radiation distribution prediction model according to the final loss.
Optionally, the apparatus further comprises:
The prediction module 912 is configured to obtain soft X-ray signals to be predicted at a plurality of times, to obtain a sampling signal sequence to be predicted, where the soft X-ray signals to be predicted are obtained by sampling soft X-rays generated by plasma at a preset time; inputting the sampling signal sequence to be predicted into a radiation distribution prediction model after training, and obtaining a radiation distribution image corresponding to each moment of the plasma output by the radiation distribution prediction model after training; and determining the state of the plasma according to the radiation distribution image corresponding to each moment of the plasma output by the trained radiation distribution prediction model, wherein the state of the plasma comprises stable and unstable.
The present specification provides a computer readable storage medium storing a computer program which when executed by a processor implements the above-described method of training a radiation distribution prediction model.
The present specification also provides a computer readable storage medium having stored thereon a computer program operable to perform a method of training a radiation distribution prediction model as provided in fig. 1 above.
The present specification also provides a schematic structural diagram of the electronic device shown in fig. 10, which corresponds to fig. 1. At the hardware level, as shown in fig. 10, the electronic device includes a processor, an internal bus, a network interface, a memory, and a nonvolatile storage, and may of course include hardware required by other services. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs to implement the training method of the radiation distribution prediction model described above with respect to fig. 1.
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: ARC 625D, 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 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 present description is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the specification. 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 (first die), 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 description.

Claims (10)

1. A method of training a radiation distribution prediction model, the radiation distribution prediction model comprising a first feature extraction layer, a first weight adjustment layer, and a radiation distribution output layer, the method comprising:
Acquiring a sampling signal sequence, wherein the sampling signal sequence comprises soft X-ray signals at a plurality of moments, and the soft X-ray signals are obtained by sampling soft X-rays generated by plasma at preset moments; acquiring a radiation distribution image of the plasma at each moment, and taking the radiation distribution image as a first tag of the sampling signal sequence;
inputting the sampling signal sequence into the first feature extraction layer to obtain a first initial feature;
Inputting the first initial feature into the first weight adjustment layer so that the first weight adjustment layer adjusts the weight of the first initial feature to obtain an adjusted first feature output by the first weight adjustment layer as a first adjustment feature;
Inputting the first adjustment feature into the radiation distribution output layer so that the radiation distribution output layer extracts time sequence features in the first adjustment feature, and obtaining a predicted radiation distribution image of the plasma corresponding to each moment output by the radiation distribution output layer according to the time sequence features;
And training the radiation distribution prediction model according to the predicted radiation distribution image and the first label.
2. The method of claim 1, wherein the radiation distribution prediction model further comprises a second feature extraction layer; the method further comprises the steps of:
inputting the sampling signal sequence into the second feature extraction layer to obtain a second initial feature;
Inputting the first initial feature into the first weight adjustment layer, so that the first weight adjustment layer adjusts the weight of the first initial feature to obtain an adjusted first feature output by the first weight adjustment layer, wherein the adjusted first feature is used as a first adjustment feature and specifically comprises the following steps:
Inputting the first initial feature and the second initial feature into the first weight adjustment layer, so that the first weight adjustment layer adjusts the weight of the first initial feature and adjusts the weight of the second initial feature to obtain an adjusted first initial feature and an adjusted second initial feature which are output by the first weight adjustment layer;
And determining a first adjustment feature according to the adjusted first initial feature and the adjusted second initial feature output by the first weight adjustment layer.
3. The method of claim 2, wherein the radiation distribution prediction model further comprises a second weight adjustment layer; the method further comprises the steps of:
inputting the first initial feature and the second initial feature into the second weight adjustment layer, so that the second weight adjustment layer adjusts the weight of the first initial feature and adjusts the weight of the second initial feature to obtain an adjusted first initial feature and an adjusted second initial feature which are output by the second weight adjustment layer;
And determining a second adjustment feature according to the adjusted first initial feature and the adjusted second initial feature output by the second weight adjustment layer.
4. The method of claim 3, wherein the radiation distribution prediction model further comprises a magnetic fluid state prediction output layer; the method further comprises the steps of:
Acquiring the state of magnetic fluid in the plasma at each moment, wherein the state of the magnetic fluid comprises stable and unstable as a second label of the sampling signal sequence;
Inputting the second adjustment characteristic into a magnetic fluid state prediction output layer so that the magnetic fluid state prediction output layer outputs a predicted magnetic fluid state corresponding to each moment;
Training the radiation distribution prediction model according to the predicted radiation distribution image and the first label, specifically including:
and training the radiation distribution prediction model according to the predicted radiation distribution image, the first label, the predicted magnetic fluid state and the second label.
5. The method of claim 4, wherein training the radiation distribution prediction model based on the predicted radiation distribution image, the first tag, the predicted magnetic fluid state, and the second tag, specifically comprises:
Determining a first loss at each moment of each iteration according to the predicted radiation distribution image and the first label;
determining a first loss change rate according to the first loss of the current iteration and the first loss of the last iteration of the current iteration;
And training the radiation distribution prediction model according to the first loss change rate, the predicted magnetic fluid state and the second label.
6. The method of claim 5, wherein training the radiation distribution prediction model based on the first loss rate of change, the predicted magnetic fluid state, and the second tag, comprises:
Determining a second loss at each moment of each iteration according to the predicted magnetic fluid state and the second label;
Determining a second loss change rate according to the second loss of the current iteration and the second loss of the last iteration of the current iteration;
And training the radiation distribution prediction model according to the first loss change rate and the second loss change rate.
7. The method of claim 6, wherein training the radiation distribution prediction model based on the first loss rate of change and the second loss rate of change, comprises:
When the first loss change rate is larger than the second loss change rate, determining the weight of the first loss as a first weight, and determining the weight of the second loss as a second weight, wherein the first weight is smaller than the second weight;
determining a final loss according to the first weight, the second weight, the first loss and the second loss;
And training the radiation distribution prediction model according to the final loss.
8. The method of claim 1, wherein the method further comprises:
Obtaining soft X-ray signals to be predicted at a plurality of moments to obtain a sampling signal sequence to be predicted, wherein the soft X-ray signals to be predicted are obtained by sampling soft X-rays generated by plasma at preset moments;
Inputting the sampling signal sequence to be predicted into a radiation distribution prediction model after training, and obtaining a radiation distribution image corresponding to each moment of the plasma output by the radiation distribution prediction model after training;
And determining the state of the plasma according to the radiation distribution image corresponding to each moment of the plasma output by the trained radiation distribution prediction model, wherein the state of the plasma comprises stable and unstable.
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 1-8.
10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the method of any of the preceding claims 1-8 when executing the program.
CN202410022577.6A 2024-01-05 2024-01-05 Training method, device and storage medium of radiation distribution prediction model Active CN117933070B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202410022577.6A CN117933070B (en) 2024-01-05 2024-01-05 Training method, device and storage medium of radiation distribution prediction model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202410022577.6A CN117933070B (en) 2024-01-05 2024-01-05 Training method, device and storage medium of radiation distribution prediction model

Publications (2)

Publication Number Publication Date
CN117933070A true CN117933070A (en) 2024-04-26
CN117933070B CN117933070B (en) 2024-07-05

Family

ID=90753029

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202410022577.6A Active CN117933070B (en) 2024-01-05 2024-01-05 Training method, device and storage medium of radiation distribution prediction model

Country Status (1)

Country Link
CN (1) CN117933070B (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112101551A (en) * 2020-09-25 2020-12-18 北京百度网讯科技有限公司 Method, apparatus, device and storage medium for training a model
CN112580800A (en) * 2020-12-11 2021-03-30 京东数字科技控股股份有限公司 Training method of data trend prediction model and prediction method of data trend
GB202112104D0 (en) * 2020-08-24 2021-10-06 Nvidia Corp Machine-learning techniques for oxygen therapy prediction using medical imaging data and clinical metadata
CN116402108A (en) * 2023-03-10 2023-07-07 支付宝(杭州)信息技术有限公司 Model training and graph data processing method, device, medium and equipment

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
GB202112104D0 (en) * 2020-08-24 2021-10-06 Nvidia Corp Machine-learning techniques for oxygen therapy prediction using medical imaging data and clinical metadata
CN112101551A (en) * 2020-09-25 2020-12-18 北京百度网讯科技有限公司 Method, apparatus, device and storage medium for training a model
CN112580800A (en) * 2020-12-11 2021-03-30 京东数字科技控股股份有限公司 Training method of data trend prediction model and prediction method of data trend
CN116402108A (en) * 2023-03-10 2023-07-07 支付宝(杭州)信息技术有限公司 Model training and graph data processing method, device, medium and equipment

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
朱元硕;李颖;卢渊;田野;: "基于向量空间模型的岩屑LIBS光谱分类识别方法", 光谱学与光谱分析, no. 09, 15 September 2017 (2017-09-15) *

Also Published As

Publication number Publication date
CN117933070B (en) 2024-07-05

Similar Documents

Publication Publication Date Title
CN117635822A (en) Model training method and device, storage medium and electronic equipment
CN117933070B (en) Training method, device and storage medium of radiation distribution prediction model
CN116842715B (en) Simulation data structuring processing system
CN117392485A (en) Image generation model training method, service execution method, device and medium
CN115841335B (en) Data processing method, device and equipment
CN117079777A (en) Medical image complement method and device, storage medium and electronic equipment
CN116403097A (en) Target detection method and device, storage medium and electronic equipment
CN117933069B (en) Inversion proxy model training method and device based on plasmas
CN117911630B (en) Three-dimensional human modeling method and device, storage medium and electronic equipment
CN114268965B (en) Mobile communication network coverage rate calculation method and device based on deep learning
CN117743824A (en) Model training and service execution method and device, storage medium and equipment
CN116501852B (en) Controllable dialogue model training method and device, storage medium and electronic equipment
CN117423445B (en) Intelligent finger ring control method and device based on user cluster perception
CN116822138B (en) Method and device for calculating conductivity of gunpowder gas and storage medium
CN117975202B (en) Model training method, service execution method, device, medium and equipment
CN118053518B (en) Chemical process time series data causal relationship graph construction method, device and medium
CN118131159A (en) Asynchronous track association method, device and system based on pseudo nearest neighbor distance
CN117808976B (en) Three-dimensional model construction method and device, storage medium and electronic equipment
CN117880444B (en) Human body rehabilitation exercise video data generation method guided by long-short time features
CN117393140B (en) Intelligent finger ring control method and device based on historical data
CN117893696B (en) Three-dimensional human body data generation method and device, storage medium and electronic equipment
CN114528923B (en) Video target detection method, device, equipment and medium based on time domain context
CN114205829B (en) Mobile communication network coverage rate calculating method based on MR data
CN117787418A (en) Risk identification method and device, storage medium and electronic equipment
CN118035784A (en) Abnormal event identification method, equipment and medium based on blood glucose monitoring

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