A diagnostic agent model training method, device, medium and electronic equipment

By training the diagnostic agent model, combined with the two-dimensional physical parameter profile chord integral information of the plasma, the problem of low accuracy in plasma diagnosis is solved, and high-precision measurement and complexity of plasma physical parameters are achieved. Adaptability in the environment.

CN118861824BActive Publication Date: 2024-11-22ZHEJIANG LAB
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411102196.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-12
Publication Date
2024-11-22
Estimated Expiration
2044-08-12

AI Technical Summary

Technical Problem

The existing optical string integral diagnostic technology is difficult to accurately measure physical parameters in plasma diagnosis, and deep learning models ignore the inherent connection between signals and physical phenomena, resulting in low diagnostic accuracy in complex environments and unable to meet high accuracy. need.

Method used

By obtaining the two-dimensional physical parameter profile and measurement signals of the plasma, using the embedded layer and the encoding layer to extract features, combining chord integral information and physical boundaries, the diagnostic agent model is trained to learn the intrinsic connection between the measurement signal and the physical information, and thus accurately Infer the physical parameters of the plasma.

Benefits of technology

The diagnostic agent model has improved the measurement accuracy of plasma physical parameters, enhanced the adaptability and migration ability to complex environments, and met the needs of high-precision diagnostics.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118861824B_ABST
    Figure CN118861824B_ABST
Patent Text Reader

Abstract

The present specification discloses a training method, device, medium and electronic device for a diagnostic proxy model, the method comprising: inputting a first measurement signal of a plasma into an embedding layer of a diagnostic proxy model to be trained to determine a first signal vector. Inputting the first chord integral information of a measurement optical path corresponding to the first measurement signal into the embedding layer to determine a first position vector. Inputting the first physical boundary of the plasma into the embedding layer to determine a first area vector. Inputting the first signal vector, the first position vector and the first area vector into the coding layer of the diagnostic proxy model to be trained to determine a first coding feature. Based on the first coding feature, a first predicted profile is determined. The diagnostic proxy model to be trained is trained based on the first predicted profile and the two-dimensional physical parameter profile of the plasma, so that the diagnostic proxy model can better learn the intrinsic connection and law between the measurement signal and the physical information, thereby accurately inferring the physical parameters of the plasma.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method, device, medium and electronic device for training a diagnostic agent model. Background Art

[0002] With the continuous development of science and technology, optical chord integration diagnostic technology has received widespread attention. This optical chord integration diagnostic technology can determine the internal state and properties of an object by capturing the spectral signal emitted or reflected by the object. In addition, this optical chord integration diagnostic technology has many application scenarios, especially in plasma diagnostic scenarios.

[0003] At present, in the plasma diagnosis scenario, the optical string integral diagnostic technology can be used to measure the physical parameters of the plasma, which can be the density distribution of the plasma or the soft X-ray radiation distribution. The density distribution of the plasma or the soft X-ray radiation distribution is the most important parameter in the magnetic confinement fusion device, which is not only directly related to the stable operation of the plasma, but also has a direct impact on the safe operation of the fusion reactor. Therefore, how to train the diagnostic agent model to accurately determine the physical parameters of the plasma is a very important issue.

[0004] Based on this, this specification provides a method for training a diagnostic agent model. Summary of the invention

[0005] This specification provides a diagnostic agent model training method, device, medium and electronic device to partially solve the above-mentioned problems existing in the prior art.

[0006] This manual adopts the following technical solutions:

[0007] This specification provides a method for training a diagnostic agent model, including:

[0008] Acquire a preset two-dimensional physical parameter profile of the plasma, and determine a first physical boundary corresponding to the plasma;

[0009] Determine a first measurement signal corresponding to the plasma collected by a preset diagnostic device, and determine first chord integral information of a measurement optical path corresponding to the first measurement signal;

[0010] Inputting the first measurement signal into the embedding layer of the diagnostic agent model to be trained to determine a first signal vector; inputting the first chord integral information into the embedding layer to determine a first position vector; inputting the first physical boundary into the embedding layer to determine a first region vector;

[0011] Inputting the first signal vector, the first position vector and the first region vector into the encoding layer of the diagnostic agent model to be trained to determine a first encoding feature;

[0012] Determining a first predicted profile corresponding to the plasma according to the first coding feature;

[0013] The diagnostic agent model to be trained is trained according to the first predicted profile and the two-dimensional physical parameter profile.

[0014] Optionally, the diagnostic device comprises a plurality of channels, and the dimension of the first measurement signal is the same as the number of the channels;

[0015] Inputting the first measurement signal into the embedding layer of the diagnostic agent model to be trained to determine a first signal vector specifically includes:

[0016] The first measurement signal is input into an embedding layer of a diagnostic agent model to be trained, so as to perform a matrix transformation on the first measurement signal through the embedding layer to obtain a first signal vector of a specified dimension.

[0017] Optionally, the diagnostic device includes a plurality of channels, and the first chord integral information is the length of the measurement light path corresponding to each channel passing through the plasma;

[0018] Inputting the first chord integral information into the embedding layer to determine a first position vector specifically includes:

[0019] The first chord integral information is input into the embedding layer so as to perform matrix transformation on the first chord integral information through the embedding layer to obtain a first position vector of a specified dimension.

[0020] Optionally, the first physical boundary is a region corresponding to the plasma in a preset measurement space;

[0021] Inputting the first physical boundary into the embedding layer to determine a first region vector specifically includes:

[0022] Gridding the preset measurement space to obtain a number of grid areas;

[0023] Marking each grid area according to the first physical boundary to obtain a first area matrix corresponding to the first physical boundary;

[0024] The first region matrix is ​​input into the embedding layer so as to perform matrix transformation on the first region matrix through the embedding layer to obtain a first region vector of a specified dimension.

[0025] Optionally, the encoding layer includes several encoding modules, each encoding module includes a normalization network, a multi-head attention network, a residual connection network and a feedforward neural network;

[0026] Inputting the first signal vector, the first position vector and the first region vector into the encoding layer of the diagnostic agent model to be trained to determine a first encoding feature specifically includes:

[0027] Determine a first embedding vector according to the first signal vector, the first position vector and the first region vector;

[0028] The first embedding vector is input into the encoding layer of the diagnostic agent model to be trained, so as to perform feature extraction on the first embedding vector through a plurality of encoding modules included in the encoding layer to obtain a first encoding feature.

[0029] Optionally, the method further comprises:

[0030] Determine a second measurement signal, second chord integral information, and second physical boundary corresponding to a first target task and use them as a first fine-tuning sample, and determine a two-dimensional density distribution profile corresponding to the first fine-tuning sample and use it as a first fine-tuning annotation; wherein the first target task is a task of measuring the density distribution of the plasma;

[0031] Inputting the second measurement signal in the first fine-tuning sample into the embedding layer of the trained diagnostic agent model to determine a second signal vector; inputting the second chord integral information in the first fine-tuning sample into the embedding layer to determine a second position vector; inputting the second physical boundary in the first fine-tuning sample into the embedding layer to determine a second region vector;

[0032] Inputting the second signal vector, the second position vector and the second region vector into the encoding layer of the diagnostic agent model to determine a second encoding feature;

[0033] Determining a second prediction profile corresponding to the first fine-tuning sample according to the second encoding feature;

[0034] The diagnostic proxy model is fine-tuned according to the second predicted profile and the first fine-tuning annotation to obtain a first target proxy model.

[0035] Optionally, the method further comprises:

[0036] determining a physical boundary of the plasma to be measured;

[0037] Determine a signal to be measured of the plasma to be measured collected by a preset diagnostic device, and determine chord integral information to be measured of a measurement optical path corresponding to the signal to be measured;

[0038] The signal to be measured, the chord integral information to be measured and the physical boundary to be measured are input into a trained diagnostic proxy model to determine a predicted profile of the plasma to be measured output by the diagnostic proxy model.

[0039] This specification provides a training device for a diagnostic agent model, including:

[0040] A first determination module, used to obtain a preset two-dimensional physical parameter profile of the plasma and determine a first physical boundary corresponding to the plasma;

[0041] A second determination module is used to determine a first measurement signal corresponding to the plasma collected by a preset diagnostic device, and to determine first chord integral information of a measurement optical path corresponding to the first measurement signal;

[0042] An embedding module, used for inputting the first measurement signal into an embedding layer of a diagnostic agent model to be trained to determine a first signal vector; inputting the first chord integral information into the embedding layer to determine a first position vector; inputting the first physical boundary into the embedding layer to determine a first area vector;

[0043] An encoding module, used for inputting the first signal vector, the first position vector and the first region vector into an encoding layer of the diagnostic agent model to be trained to determine a first encoding feature;

[0044] A prediction module, configured to determine a first predicted profile corresponding to the plasma according to the first coding feature;

[0045] A training module is used to train the diagnostic agent model to be trained according to the first predicted section and the two-dimensional physical parameter section.

[0046] This specification provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned training method of a diagnostic agent model.

[0047] This specification provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned training method for a diagnostic agent model when executing the program.

[0048] At least one of the above technical solutions adopted in this specification can achieve the following beneficial effects:

[0049] The training method of the diagnostic proxy model provided in this specification first obtains a preset two-dimensional physical parameter profile of the plasma and determines the first physical boundary corresponding to the plasma. Determine the first measurement signal corresponding to the plasma collected based on the preset diagnostic device, and determine the first chord integral information of the measurement optical path corresponding to the first measurement signal. Input the first measurement signal into the embedding layer of the diagnostic proxy model to be trained to determine the first signal vector. Input the first chord integral information into the embedding layer to determine the first position vector. Input the first physical boundary into the embedding layer to determine the first area vector. Afterwards, input the first signal vector, the first position vector and the first area vector into the encoding layer of the diagnostic proxy model to be trained to determine the first encoding feature. According to the first encoding feature, determine the first predicted profile corresponding to the plasma. According to the first predicted profile and the two-dimensional physical parameter profile, train the diagnostic proxy model to be trained.

[0050] It can be seen from the above method that when training the diagnostic proxy model, the present application first obtains the preset two-dimensional physical parameter profile of the plasma and determines the first physical boundary corresponding to the plasma. Determine the first measurement signal corresponding to the plasma collected by the preset diagnostic device, and determine the first chord integral information of the measurement optical path corresponding to the first measurement signal. Input the first measurement signal into the embedding layer of the diagnostic proxy model to be trained to determine the first signal vector. Input the first chord integral information into the embedding layer to determine the first position vector. Input the first physical boundary into the embedding layer to determine the first area vector. Afterwards, input the first signal vector, the first position vector and the first area vector into the encoding layer of the diagnostic proxy model to be trained to determine the first encoding feature. According to the first encoding feature, determine the first predicted profile corresponding to the plasma. According to the first predicted profile and the two-dimensional physical parameter profile, train the diagnostic proxy model to be trained. By training the diagnostic proxy model based on the first measurement signal, the first chord integral analysis and the first physical boundary, the diagnostic proxy model can better learn the intrinsic connection and law between the measurement signal and the physical information, thereby deeply analyzing the signal emitted or reflected from the plasma, so as to accurately infer the physical parameters of the plasma. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The drawings described herein are used to provide a further understanding of this specification and constitute a part of this specification. The illustrative embodiments and descriptions of this specification are used to explain this specification and do not constitute an improper limitation on this specification. In the drawings:

[0052] Figure 1 A flowchart of a method for training a diagnostic agent model provided in this specification;

[0053] Figure 2 A schematic diagram of a training process of a coding layer provided in this specification;

[0054] Figure 3 A schematic diagram of a process for determining a first coding feature provided in this specification;

[0055] Figure 4 A schematic diagram of a pre-training process of a diagnostic agent model provided in this specification;

[0056] Figure 5 A schematic diagram of a fine-tuning process of a diagnostic agent model provided in this specification;

[0057] Figure 6 A schematic diagram of a training device for a diagnostic agent model provided in this specification;

[0058] Figure 7 A method corresponding to the Figure 1 Schematic diagram of the structure of an electronic device. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solutions and advantages of this specification more clear, the technical solutions of this specification will be clearly and completely described below in combination with the specific embodiments of this specification and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of this specification, not all of them. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this specification.

[0060] The technical solutions provided by the embodiments of this specification are described in detail below in conjunction with the accompanying drawings.

[0061] Figure 1 A flowchart of a method for training a diagnostic agent model provided in this specification includes the following steps:

[0062] S100: Acquire a preset two-dimensional physical parameter profile of the plasma, and determine a first physical boundary corresponding to the plasma.

[0063] At present, the optical string integral diagnosis method mainly relies on the physical model based on the first principles in practical applications. When analyzing complex physical phenomena, this physical model needs to consider many factors, which may include the diversity of time and space scales, the interaction between different physical fields, and the complex coupling mechanism within the system. These factors lead to a significant increase in computational costs, and the construction and solution process of the physical model is time-consuming and labor-intensive. In addition, the optical string integral diagnosis method can also directly map the measurement signal to the diagnosis result through a deep learning model in practical applications, but ignores the complex intrinsic connection between the signal and the physical phenomenon. This direct mapping method lacks an in-depth understanding of the physical mechanism behind the measurement signal and fails to make full use of the rich information in the measurement signal. In addition, the existing deep learning model ignores the characteristics of the diagnostic device and the influence of the environment of signal acquisition. In addition, the migration ability and generalization of the deep learning model in new or changing application scenarios are severely limited. The deep learning model has low accuracy when processing unseen data, or its performance in complex environments is often unsatisfactory, and cannot meet the needs of high-precision diagnosis.

[0064] Based on this, in this specification, the device for training the diagnostic agent model can obtain a preset two-dimensional physical parameter profile of the plasma and determine the first physical boundary corresponding to the plasma. Among them, the device for training the diagnostic agent model can be a server, or an electronic device such as a desktop computer, a laptop computer, etc. For the convenience of description, the following only takes the server as the execution subject to illustrate the training method of the diagnostic agent model provided in this specification.

[0065] In this specification, plasma is a collection of charged particles composed of ions (usually positive ions) and free electrons, and these charged particles are usually in a highly ionized state. The plasma can be used in many fields, such as energy, industry, astronomy, and medical fields. The plasma can be used in nuclear fusion experiments, that is, in magnetic confinement plasma devices such as tokamaks. Specifically, the plasma can be confined in a ring device by magnetic confinement, and power can be applied to heat the plasma in the ring device to promote nuclear fusion reactions between hydrogen isotopes (deuterium and tritium), thereby achieving nuclear fusion. Of course, the plasma can also be plasma in other fields, and this specification does not specifically limit it.

[0066] The above-mentioned physical parameter may be at least one of the density distribution or soft X-ray radiation distribution of the plasma, and the two-dimensional physical parameter profile may be at least one of the two-dimensional density distribution profile or the two-dimensional soft X-ray radiation distribution profile of the plasma. Of course, the above-mentioned physical parameter may also be other physical parameters of the plasma, and the above-mentioned two-dimensional physical parameter profile may also be other two-dimensional physical parameter profiles of the plasma. It should be noted that the above-mentioned two-dimensional physical parameter profile may be preset by the server, that is, the two-dimensional physical parameter profile may be a virtual two-dimensional parameter profile, the virtual two-dimensional parameter profile may be directly randomly generated by the server, or may be randomly generated by the server based on the two-dimensional parameter profile of the plasma obtained in history or any existing two-dimensional parameter profile of the plasma, or may be preset by the administrator and uploaded to the server. Of course, the above-mentioned two-dimensional physical parameter profile may also be a real two-dimensional parameter profile, the two-dimensional physical parameter profile may be a two-dimensional parameter profile of the plasma obtained in history, that is, it is determined by other machine learning models or any existing methods in history, such as the two-dimensional physical parameter profile may be determined based on the Fourier-Bessel analytical method, and this specification does not limit it. Of course, the above two-dimensional physical parameter profile can also be any existing two-dimensional parameter profile of plasma, which is not specifically limited in this specification.

[0067] In this specification, the first physical boundary can be the size of the two-dimensional physical parameter profile, that is, the size of the two-dimensional physical parameter profile. The first physical boundary is the area corresponding to the plasma in the pre-set measurement space. It should be noted that the measurement space can be a virtual space pre-set by the server, and the first physical boundary is also pre-set by the server, that is, the first physical boundary can be a virtual physical boundary. Both the measurement space and the first physical boundary can be pre-randomly set by the server. Of course, the measurement space can also be randomly generated by the server based on the space when the plasma is measured in history. The first physical boundary can also be randomly generated by the server based on the first physical boundary corresponding to the plasma obtained in history or the first physical boundary of any existing plasma. The measurement space or the first physical boundary can also be pre-set by the administrator and uploaded to the server.

[0068] In addition, the above-mentioned measurement space may also be a real measurement space, that is, the collection space when the pre-set diagnostic device collects the signal reflected or emitted by the plasma. The size of the collection space can be determined according to the position of the diagnostic device and the position of the plasma. Of course, the size of the collection space can also be pre-set by the server, and this specification does not specifically limit it. The above-mentioned measurement space may be a two-dimensional space. Based on this, the above-mentioned first physical boundary may also be a real physical boundary, and the first physical boundary may be the first physical boundary corresponding to the plasma obtained in history or the first physical boundary of any existing plasma. It should be noted that the above-mentioned diagnostic device may be a diagnostic device or a diagnostic system pre-built in the tokamak device for measuring the physical parameters of the plasma. The diagnostic device or the diagnostic system includes a number of sensors, which can be used to sample the signal reflected or emitted by the plasma. The sensors may have different numbers of channels, and the installation positions or orientations of the sensors may be different, that is, the sensors may be located at different positions for sampling the signal reflected or emitted by the plasma at different positions or orientations. For ease of explanation, the following takes the diagnostic device or the diagnostic system including a sensor, and the sensor including a number of channels as an example, that is, the diagnostic device or the diagnostic system including a number of channels as an example. The server can establish a communication connection with the diagnostic device or diagnostic system in the tokamak device, so as to obtain the measurement signal collected by the diagnostic device or diagnostic system. Of course, the above-mentioned diagnostic device can also be any existing device for collecting signals when measuring the two-dimensional physical parameter profile of plasma. The above-mentioned diagnostic device can be a laser interferometer diagnostic device, such as a far-infrared laser interferometer, and can also be a soft X-ray diagnostic device. The server can establish a communication connection with the diagnostic device, so as to obtain the measurement signal collected by the diagnostic device.

[0069] S102: Determine a first measurement signal corresponding to the plasma collected by a preset diagnostic device, and determine first chord integral information of a measurement optical path corresponding to the first measurement signal.

[0070] In this specification, the server may determine a first measurement signal corresponding to the plasma collected by a pre-set diagnostic device, and determine the first chord integral information of the measurement optical path corresponding to the first measurement signal. The above-mentioned diagnostic device may be a virtual diagnostic device, and the server may set up virtual diagnostic devices at different positions in the measurement space, and the virtual diagnostic device is provided with an optical path layout, that is, a layout of the measurement optical path, and the optical path layout may include a starting point and an azimuth (i.e., direction) of the measurement optical path. The optical path layout may be randomly set by the server, or may be randomly generated by the server based on the optical path layout of a real diagnostic device when measuring the physical parameters of plasma in history, or randomly generated based on the optical path layout of any existing diagnostic device (i.e., a virtual or real diagnostic device), and this specification does not make specific limitations.

[0071] The above-mentioned measuring optical path is also a virtual optical path, and the first chord-integrated information of the measuring optical path may also be virtual chord-integrated information. The first chord-integrated information is the chord length (i.e., length) of the measuring optical path through the plasma, that is, how long the measuring optical path passes through the plasma. The first chord-integrated information can be determined based on the optical path layout of the diagnostic device and the basic information of the plasma. The basic information may include the position, shape, and volume of the plasma. It should be noted that the position of the plasma is the position of the container containing the plasma, and the shape and volume of the plasma are the shape and volume of the container containing the plasma.

[0072] Based on this, the first measurement signal can also be a virtual measurement signal. According to the chord integral physics principle, the first chord integral information and the virtual two-dimensional physical parameter profile can be multiplied to obtain the first measurement signal. At this time, the first measurement signal and the two-dimensional physical parameter profile are both ideal values, and any diagnostic system error and accidental error are not considered. It should be noted that there is a one-to-one correspondence between the virtual two-dimensional physical parameter profile, the virtual diagnostic device, the virtual measurement optical path, the virtual first chord integral information and the virtual first measurement signal.

[0073] In addition, the above-mentioned diagnostic device may also be a real diagnostic device, and the details are as described in the above-mentioned step S100, which will not be repeated here. The first chord-integrated information may also be real chord-integrated information, and the first chord-integrated information may be the chord length of the real measurement light path through the plasma in history, or any existing measurement light path through the plasma. This specification does not make specific restrictions. The first measurement signal may also be a real measurement signal, and the first measurement signal may be a measurement signal obtained based on a real diagnostic device in history, or any existing real measurement signal. This specification does not make specific restrictions. It should be noted that there is a one-to-one correspondence between the real two-dimensional physical parameter profile, the real diagnostic device, the real measurement light path, the real first chord-integrated information, and the real first measurement signal.

[0074] S104: Input the first measurement signal into the embedding layer of the diagnostic agent model to be trained to determine a first signal vector; input the first chord integral information into the embedding layer to determine a first position vector; input the first physical boundary into the embedding layer to determine a first area vector.

[0075] In this specification, the server may input the first measurement signal into the embedding layer of the diagnostic proxy model to be trained to determine the first signal vector. The first chord integral information is input into the embedding layer to determine the first position vector. At the same time, the first physical boundary is input into the embedding layer to determine the first area vector. Among them, the diagnostic proxy model to be trained includes at least an embedding layer and a coding layer, and the embedding layer is used to perform a matrix transformation on the input data to obtain a vector of a specified dimension. The diagnostic proxy model to be trained is used to learn the chord integral physical principle, that is, to determine the physical parameters of the plasma at least by measuring the signal and the chord integral information. The specified dimension can be pre-set by the server, and the specified dimension is n×m, wherein n is the number of channels of the diagnostic device, m=r×z, and r×z is the resolution of the above-mentioned measurement space.

[0076] In this specification, since the diagnostic device may include several channels, and each channel has a corresponding measurement optical path, and each measurement optical path has a corresponding measurement signal, the first measurement signal includes several dimensions, and the dimension of the first measurement signal is the same as the number of channels (that is, the number of channels included in the diagnostic device). The first measurement signal can be represented by In_s, and the first measurement signal can be an n×1 array. The first signal vector is obtained by matrix transformation of the first measurement signal through the embedding layer, and the dimension of the first signal vector is a specified dimension, that is, n×m, then the first signal vector is a matrix of the specified dimension, that is, an n×m matrix. The above-mentioned embedding layer can be a convolution layer or a linear layer with a 1D convolution kernel. Based on this, when the first measurement signal is input into the embedding layer of the diagnostic agent model to be trained to determine the first signal vector, the server can input the first measurement signal into the embedding layer of the diagnostic agent model to be trained, so as to perform a matrix transformation on the first measurement signal through the embedding layer to obtain a first signal vector of the specified dimension.

[0077] In the present specification, since the diagnostic device may include several channels, and each channel has a corresponding measurement signal, each measurement signal has corresponding chord integral information, the first chord integral information is the length of the measurement light path corresponding to each channel through the plasma, and there are n first measurement signals, then there are n sub-chord integral information, the first chord integral information includes several (i.e., n) sub-chord integral information, each sub-chord integral information can be represented by a matrix of the size of the measurement space, i.e., a sub-chord integral information matrix of r×z, then the first chord integral information can be a matrix of n×r×z. The first position vector is obtained by performing a matrix transformation on the first chord integral information through the embedding layer, and the dimension of the first position vector is a specified dimension, i.e., n×m, then the first position vector is a matrix of the specified dimension, i.e., a matrix of n×m. Specifically, when the first chord integral information is input into the embedding layer to determine the first position vector, the server may input the first chord integral information into the embedding layer, so as to perform a matrix transformation on the first chord integral information through the embedding layer to obtain the first position vector of the specified dimension.

[0078] In this specification, the above-mentioned first physical boundary is the area corresponding to the plasma in the pre-set measurement space. For the same application scenario (i.e., measuring a physical parameter of the same plasma), there is generally only one physical boundary. The first physical boundary can be represented by a matrix of the size of the measurement space, i.e., the first physical boundary matrix of r×z. In order to align the dimensions of the first signal vector, the first position vector, and the first area vector, the server can flatten the first physical boundary into a 1×m array through the embedding layer, and then stack it into an n×m matrix to obtain the first area vector of the specified dimension. Specifically, when the first physical boundary is input into the embedding layer to determine the first area vector, the server can input the first physical boundary into the embedding layer to perform a matrix transformation on the first physical boundary through the embedding layer to obtain the first area vector of the specified dimension.

[0079] In addition, the server may first grid the measurement space, and then mark the grid areas after division according to the first physical boundary, that is, mark the grid areas after division within the first physical boundary to determine the area corresponding to the plasma (that is, the marked grid areas), and then determine the first area matrix corresponding to the first physical boundary according to the marked grid areas, that is, the above-mentioned r×z first physical boundary matrix. Specifically, when the first physical boundary is input into the embedding layer to determine the first area vector, the server may grid the preset measurement space to obtain several grid areas. Then, mark each grid area according to the first physical boundary to obtain the first area matrix corresponding to the first physical boundary. Input the first area matrix into the embedding layer to perform matrix transformation on the first area matrix through the embedding layer to obtain the first area vector of the specified dimension. Among them, the size of each grid area is the same and can be a preset size. The first area matrix is ​​a first physical boundary matrix of r×z, which is a matrix composed of 0 and 1, 1 indicates that the grid area is within the first physical boundary and the grid area is marked, and 0 indicates that the grid area is not within the first physical boundary and the grid area is not marked.

[0080] S106: Input the first signal vector, the first position vector and the first region vector into the encoding layer of the diagnostic agent model to be trained to determine a first encoding feature.

[0081] S108: Determine a first predicted profile corresponding to the plasma according to the first coding feature.

[0082] In this specification, the server may input the first signal vector, the first position vector and the first area vector into the coding layer of the diagnostic agent model to be trained to determine the first coding feature. Then, based on the first coding feature, determine the first predicted profile corresponding to the plasma. Among them, in the above step S106, the server may determine the first embedding vector based on the first signal vector, the first position vector and the first area vector, and then input the first embedding vector into the coding layer of the diagnostic agent model to be trained to determine the first coding feature, and the dimension of the first coding feature is the specified dimension. When determining the first embedding vector, the first signal vector, the first position vector and the first area vector may be operated or stacked to generate the first embedding vector. By generating the final first embedding vector through operation or stacking, the effective fusion of physical information and measurement signals is achieved, so that the predicted profile generated by the diagnostic agent model is more accurate.

[0083] In addition, in order to ensure that the diagnostic proxy model can encode the embedding vector of the specified dimension in each application scenario, ensure that the diagnostic proxy model can handle inputs of different scales (i.e., inputs of different dimensions), and in order to make the diagnostic proxy model have better mobility, the server may fill the above-mentioned first embedding vector to obtain an embedding vector of a preset dimension, and the preset dimension may be n_max×m_max, where n_max is the maximum number of measurement signals (i.e., the maximum number of channels) acceptable to the diagnostic proxy model, and m_max is the maximum resolution of the measurement space. Therefore, in the above-mentioned step S106, when the first embedding vector is input into the encoding layer of the diagnostic proxy model to be trained and the first encoding feature is determined, the server may fill the first embedding vector to obtain the maximum embedding vector of the preset dimension, and input the maximum embedding vector into the encoding layer of the diagnostic proxy model to be trained to determine the first encoding feature. Among them, when filling the first embedding vector, the server may fill the first embedding vector with a null value to obtain the maximum embedding vector of the preset dimension.

[0084] In this specification, in order to improve the migration ability and information processing ability of the diagnostic agent model, the diagnostic agent model can introduce deep learning and attention mechanisms to better process complex physical signals, learn the intrinsic connection between optical string integral signals and physical information, and the attention mechanism can enable the diagnostic agent model to pay special attention to the key features in the measurement signal and the relationship between the features, similar to the context understanding in natural language processing, so that the model is superior to traditional methods and existing deep learning applications. Therefore, the above-mentioned encoding layer may include a normalized network, a multi-head attention network, a residual connection network and a feedforward neural network, and the encoding layer may include two normalized networks (i.e., a first normalized network and a second normalized network) and two residual connection networks (i.e., a first residual connection network and a second residual connection network), so in the above-mentioned step S106, the server may determine the first embedding vector according to the first signal vector, the first position vector and the first area vector. Then the first embedding vector is input into the first normalized network in the encoding layer of the diagnostic agent model to be trained to determine the first normalized feature. Then the first normalized feature is input into the multi-head attention network in the encoding layer of the diagnostic agent model to be trained to determine the weighted feature. The weighted feature and the first embedded feature are then input into the first residual connection network in the encoding layer of the diagnostic proxy model to be trained to determine the first feature, and the first feature is then input into the second normalized network in the encoding layer of the diagnostic proxy model to be trained to determine the second normalized feature. The second normalized feature is then input into the feedforward neural network in the encoding layer of the diagnostic proxy model to be trained to determine the second feature. The first feature and the second feature are then input into the second residual connection network in the encoding layer of the diagnostic proxy model to be trained to determine the first encoded feature. Specifically, Figure 2 As shown, Figure 2 A schematic diagram of a training process of a coding layer provided in this specification.

[0085] In addition, the above encoding layer includes several encoding modules, each encoding module includes a normalized network, a multi-head attention network, a residual connection network and a feedforward neural network, and each encoding module may include two normalized networks (i.e., a first normalized network and a second normalized network) and two residual connection networks (i.e., a first residual connection network and a second residual connection network). Based on this, in the above step S106, the server may determine the first embedding vector according to the first signal vector, the first position vector and the first region vector. The first embedding vector is input into the encoding layer of the diagnostic agent model to be trained, so as to extract features of the embedding vector through several encoding modules included in the encoding layer to obtain a first encoding feature. Wherein, when the first embedding vector is extracted through several encoding modules included in the encoding layer to obtain the first encoding feature, the server may sequentially input the first embedding vector into each encoding module included in the encoding layer, obtain the output feature corresponding to the encoding module, and re-use the output feature as the first embedding vector, and re-input the next encoding module of the encoding module until the last encoding module is input, and the output feature corresponding to the last encoding module is determined and used as the first encoding feature. The process of inputting the first embedding vector into the encoding module to obtain the output feature corresponding to the encoding module is similar to the process of determining the first encoding feature according to the first embedding vector in the above step S106, and will not be repeated here. The process of determining the first encoding feature according to the first embedding vector can be as follows: Figure 3 As shown, Figure 3 is a schematic diagram of a process for determining a first coding feature provided in this specification, Figure 3 The encoding layer in the process shown may include N encoding modules, and each encoding module includes two normalization networks, a multi-head attention network, two residual connection networks, and a feedforward neural network.

[0086] In this specification, the above-mentioned first predicted profile may be at least one of the two-dimensional density distribution profile or the two-dimensional soft X-ray radiation distribution profile of the plasma. Of course, the above-mentioned first predicted profile may also be a predicted profile of other physical parameters of the plasma. It should be noted that the first predicted profile is a two-dimensional profile. In the above step S108, the server may input the first coding feature into the decoding layer to determine the first predicted profile corresponding to the plasma. Among them, the decoding layer may be a network layer pre-trained by the server, or any existing network layer for determining the two-dimensional physical parameter profile of the plasma. Of course, the decoding layer may also be an untrained network layer, and may be trained together with the diagnostic agent model to be trained, that is, the diagnostic agent model to be trained may also include a decoding layer, which is used to determine the first predicted profile according to the first coding feature. In addition, the server may also use any existing generation algorithm or machine learning model to generate the first predicted profile corresponding to the plasma based on the first coding feature.

[0087] S110: Training the diagnostic agent model to be trained according to the first predicted profile and the two-dimensional physical parameter profile.

[0088] In this specification, the server can train the diagnostic proxy model to be trained based on the first predicted profile and the two-dimensional physical parameter profile. Specifically, the server can train the diagnostic proxy model to be trained with the goal of minimizing the difference between the first predicted profile and the two-dimensional physical parameter profile. In addition, the server can also determine the first loss based on the first predicted profile and the two-dimensional physical parameter profile, and then train the diagnostic proxy model to be trained based on the first loss. The trained diagnostic proxy model can determine the two-dimensional physical parameter profile of the plasma based on the measurement signal, physical boundary and chord integral information corresponding to the plasma.

[0089] In this specification, in order to enable the diagnostic agent model to be trained to better learn the chord integral physics principle, the server can also determine the first prediction signal according to the first prediction profile and the first chord integral information, and then determine the second loss according to the first prediction signal and the first measurement signal. Afterwards, the diagnostic agent model to be trained is trained according to the first loss and the second loss. The first prediction signal is the product of the first prediction profile and the first chord integral information.

[0090] It can be seen from the above method that when training the diagnostic proxy model, the present application first obtains the preset two-dimensional physical parameter profile of the plasma and determines the first physical boundary corresponding to the plasma. Determine the first measurement signal corresponding to the plasma collected by the preset diagnostic device, and determine the first chord integral information of the measurement optical path corresponding to the first measurement signal. Input the first measurement signal into the embedding layer of the diagnostic proxy model to be trained to determine the first signal vector. Input the first chord integral information into the embedding layer to determine the first position vector. Input the first physical boundary into the embedding layer to determine the first area vector. Afterwards, input the first signal vector, the first position vector and the first area vector into the encoding layer of the diagnostic proxy model to be trained to determine the first encoding feature. According to the first encoding feature, determine the first predicted profile corresponding to the plasma. According to the first predicted profile and the two-dimensional physical parameter profile, train the diagnostic proxy model to be trained. By training the diagnostic proxy model based on the first measurement signal, the first chord integral analysis and the first physical boundary, the diagnostic proxy model can better learn the intrinsic connection and law between the measurement signal and the physical information, thereby deeply analyzing the signal emitted or reflected from the plasma, so as to accurately infer the physical parameters of the plasma.

[0091] In this specification, after obtaining the trained diagnostic proxy model, the server can determine the physical boundary to be measured of the plasma to be measured, determine the signal to be measured of the plasma to be measured collected by the preset diagnostic device, and determine the chord integral information to be measured of the measurement optical path corresponding to the signal to be measured. Afterwards, the signal to be measured, the chord integral information to be measured and the physical boundary to be measured are input into the trained diagnostic proxy model to determine the predicted profile of the plasma to be measured output by the diagnostic proxy model.

[0092] Among them, the physical boundary to be measured is the area corresponding to the plasma to be measured in the preset measurement space, the signal to be measured is the signal collected by the preset diagnostic device, and the chord integral information to be measured is the length of the measurement light path through the plasma to be measured. The chord integral information to be measured can be determined by the server according to the position of the diagnostic device and the basic information of the plasma to be measured, and can also be determined by other computing devices, which is not specifically limited in this specification. The above-mentioned predicted profile can be a density distribution profile of the plasma to be measured, and can also be a soft X-ray radiation distribution profile of the plasma to be measured. The specific two-dimensional physical parameter profile is not specifically limited in this specification, and is mainly determined according to the diagnostic device. When the diagnostic device is a laser interferometer, the predicted profile can be a density distribution profile of the plasma to be measured, and when the diagnostic device is a soft X-ray diagnostic device, the predicted profile can be a soft X-ray radiation distribution profile of the plasma to be measured. The process of inputting the measured signal, the measured chord integral information and the measured physical boundary into the trained diagnostic proxy model to determine the predicted profile of the measured plasma output by the diagnostic proxy model is similar to the process described in steps S104 to S108 above, and will not be repeated here.

[0093] In this manual, the above Figure 1The process shown may be the pre-training stage of the diagnostic proxy model. The server may pre-train the diagnostic proxy model to be trained based on a large number of virtual training samples (i.e., the first measurement signal, the first physical boundary, and the first chord integral information) and virtual annotations (i.e., the two-dimensional physical parameter profile). Of course, in addition to using virtual training samples and annotations, real training samples and annotations may also be used to pre-train the diagnostic proxy model to be trained, so as to improve the robustness and generalization ability of the diagnostic proxy model and avoid overfitting to ideal conditions. In addition, the diagnostic proxy model to be trained can focus on learning the intrinsic connection and laws between the measurement signal and the physical information. This pre-training method provides the diagnostic proxy model with an in-depth understanding of the "language" of the optical chord integral, learns the intrinsic laws of physical information, and lays the foundation for subsequent fine-tuning for specific diagnostic tasks. Subsequent fine-tuning is performed for specific application scenarios, and the annotation data is used to improve the prediction accuracy of the diagnostic proxy model for specific physical problems. The pre-training-fine-tuning training strategy can ensure that the diagnostic proxy model is both physically interpretable and performs well in specific scenarios, thereby improving diagnostic efficiency and accuracy, and has a wide range of cross-domain application potential.

[0094] In this specification, the pre-training stage can be specifically as follows Figure 4 As shown, Figure 4 is a schematic diagram of a pre-training process of a diagnostic agent model provided in this specification, Figure 4 The diagnostic proxy model to be trained shown includes an embedding layer, a coding layer and a decoding layer as an example. The server can input the first measurement signal, the first chord integral information and the first physical boundary into the embedding layer of the diagnostic proxy model to be trained respectively to obtain the first signal vector, the first position vector and the first area vector. Then the first signal vector, the first position vector and the first area vector are input into the coding layer of the diagnostic proxy model to be trained to obtain the first coding feature. Then the first coding feature is input into the decoding layer of the diagnostic proxy model to be trained to obtain the first predicted profile. Subsequently, the diagnostic proxy model to be trained is pre-trained according to the first predicted profile and the two-dimensional physical parameter profile.

[0095] Based on this, after obtaining the trained diagnostic proxy model, the server can fine-tune the diagnostic proxy model based on the sample data corresponding to the downstream task to obtain the target proxy model, and the target proxy model can be used to perform the downstream task, that is, the server can use the trained diagnostic proxy model as the base model, and fine-tune the base model based on the sample data corresponding to the downstream task to obtain the target proxy model. The downstream task is at least one of the task of measuring the density distribution of plasma and the task of measuring the radiation distribution of plasma.

[0096] Specifically, when the downstream task is the first target task, and the first target task is a task to measure the density distribution of plasma, the server can determine the second measurement signal, second chord integral information, and second physical boundary corresponding to the first target task, and use them as the first fine-tuning sample, as well as determine the two-dimensional density distribution profile corresponding to the first fine-tuning sample, and use it as the first fine-tuning annotation. The second measurement signal in the first fine-tuning sample is input into the embedding layer of the trained diagnostic proxy model to determine the second signal vector. The second chord integral information in the first fine-tuning sample is input into the embedding layer to determine the second position vector. The second physical boundary in the first fine-tuning sample is input into the embedding layer to determine the second area vector. Afterwards, the second signal vector, the second position vector, and the second area vector are input into the encoding layer of the diagnostic proxy model to determine the second encoding feature. Based on the second encoding feature, the second predicted profile corresponding to the first fine-tuning sample is determined. Based on the second predicted profile and the first fine-tuning annotation, the diagnostic proxy model is fine-tuned to obtain the first target proxy model. Specifically, it can be as follows Figure 5 As shown, Figure 5 A schematic diagram of a fine-tuning process of a diagnostic agent model provided in this specification.

[0097] Among them, the above-mentioned first target task is to measure the density distribution task of plasma. The above-mentioned second measurement signal, second chord integral information and second physical boundary are all real data, and are also data corresponding to the first target task. The server can directly obtain the second measurement signal, second chord integral information and second physical boundary determined when the first target task was executed in history. The first fine-tuning annotation is the two-dimensional density distribution profile of the plasma, that is, the two-dimensional electron density distribution profile of the plasma. It should be noted that the above-mentioned measurement signals and chord integral information are all data at a certain moment in history. The specific process of fine-tuning the diagnostic agent model based on the first fine-tuning sample and the first fine-tuning annotation is the same as the above-mentioned process. Figure 1 The training process shown is similar and will not be repeated here.

[0098] The trained diagnostic proxy model may include an embedding layer and an encoding layer, and of course, a decoding layer. Therefore, the first target proxy model may also include an embedding layer, an encoding layer and a decoding layer. When the second prediction profile corresponding to the first fine-tuning sample is determined according to the second encoding feature, the server may input the second encoding feature into the decoding layer of the diagnostic proxy model to determine the second prediction profile corresponding to the first fine-tuning sample. In addition, the first target proxy model may only include the embedding layer and the encoding layer of the diagnostic proxy model, and of course, may also include an untrained first prediction layer. The first prediction layer may be fine-tuned together with the diagnostic proxy model when the diagnostic proxy model is fine-tuned according to the second prediction profile and the first fine-tuning annotation, that is, the server may fine-tune the diagnostic proxy model and the first prediction layer according to the second prediction profile and the first fine-tuning annotation. At this time, the diagnostic proxy model only includes an embedding layer and an encoding layer. The model structure of the first prediction layer may be set by the server according to the first target task. The first prediction layer may be a decoding module of a Transformer, or a convolutional neural network, or a multi-layer perceptron. This specification does not specifically limit this. The first prediction layer can be in various forms, depending on the requirements and computing power of the task, which is a task of predicting the output of the profile, which can be a classification task output or a regression task output. Based on this, when determining the second prediction profile corresponding to the first fine-tuning sample according to the second coding feature, the server can input the second coding feature into the pre-set first prediction layer to determine the second prediction profile corresponding to the first fine-tuning sample.

[0099] The above-mentioned first target proxy model is used to execute the first target task. Specifically, the server can determine the first signal to be diagnosed, the first chord integral information to be diagnosed and the first physical boundary to be diagnosed corresponding to the first plasma to be diagnosed, and input the first signal to be diagnosed, the first chord integral information to be diagnosed and the first physical boundary to be diagnosed into the first target proxy model to determine the density distribution profile corresponding to the first plasma to be diagnosed. The specific process is similar to the process described in the above steps S104 to S108, and will not be repeated here.

[0100] In the present specification, the downstream task may also be a second target task, and the second target task may be a task of measuring the radiation distribution of plasma, that is, a task of measuring the soft X-ray radiation distribution of plasma. The second fine-tuning sample corresponding to the second target task may include a third measurement signal, third chord integral information and a third physical boundary, and the second fine-tuning annotation is a two-dimensional radiation distribution profile. The process of fine-tuning the trained diagnostic proxy model based on the second fine-tuning sample and the second fine-tuning annotation is the same as the process of fine-tuning the trained diagnostic proxy model based on the first fine-tuning sample and the first fine-tuning annotation (that is, Figure 5The model obtained after fine-tuning is the second target proxy model, which is used to perform the second target task. The process of performing the second target task is similar to the process of performing the first target task by the first target proxy model, which is not repeated here.

[0101] In this specification, after obtaining the first target proxy model or the second target proxy model, the server may test the first target proxy model or the second target proxy model according to the test data set of the first target task or the second target task to evaluate the performance of the first target proxy model or the second target proxy model. The specific evaluation index may be accuracy, and the specific process is similar to any existing machine learning evaluation process, which will not be repeated here.

[0102] In this specification, the Tokamak device is often disturbed by various magnetic fluid instabilities, which makes the plasma discharge unstable or fails. Therefore, after obtaining the density distribution profile or radiation distribution profile of the plasma, the magnetic fluid disturbance can be monitored based on the density distribution profile or the radiation distribution profile, so as to ensure the stable discharge of the plasma. Specifically, the density distribution profile or the radiation distribution profile can be displayed to the management personnel so that the management personnel can monitor the magnetic fluid disturbance.

[0103] The above are one or more implementation methods of this specification. Based on the same idea, this specification also provides a corresponding diagnostic agent model training device, such as Figure 6 shown.

[0104] Figure 6 A schematic diagram of a training device for a diagnostic agent model provided in this specification includes:

[0105] A first determination module 200 is used to obtain a preset two-dimensional physical parameter profile of the plasma and determine a first physical boundary corresponding to the plasma;

[0106] A second determination module 202 is used to determine a first measurement signal corresponding to the plasma collected by a preset diagnostic device, and to determine first chord integral information of a measurement optical path corresponding to the first measurement signal;

[0107] The embedding module 204 is used to input the first measurement signal into the embedding layer of the diagnostic agent model to be trained to determine a first signal vector; input the first chord integral information into the embedding layer to determine a first position vector; input the first physical boundary into the embedding layer to determine a first region vector;

[0108] An encoding module 206, configured to input the first signal vector, the first position vector and the first region vector into an encoding layer of the diagnostic agent model to be trained to determine a first encoding feature;

[0109] A prediction module 208, configured to determine a first predicted profile corresponding to the plasma according to the first coding feature;

[0110] The training module 210 is used to train the diagnostic agent model to be trained according to the first predicted section and the two-dimensional physical parameter section.

[0111] Optionally, the diagnostic device comprises a plurality of channels, and the dimension of the first measurement signal is the same as the number of the channels;

[0112] The embedding module 204 is specifically used to input the first measurement signal into the embedding layer of the diagnostic agent model to be trained, so as to perform matrix transformation on the first measurement signal through the embedding layer to obtain a first signal vector of a specified dimension.

[0113] Optionally, the diagnostic device includes a plurality of channels, and the first chord integral information is the length of the measurement light path corresponding to each channel passing through the plasma;

[0114] The embedding module 204 is specifically used to input the first chord integral information into the embedding layer, so as to perform matrix transformation on the first chord integral information through the embedding layer to obtain a first position vector of a specified dimension.

[0115] Optionally, the first physical boundary is a region corresponding to the plasma in a preset measurement space;

[0116] The embedding module 204 is specifically used to grid the preset measurement space to obtain a plurality of grid areas; mark each grid area according to the first physical boundary to obtain a first area matrix corresponding to the first physical boundary; input the first area matrix into the embedding layer to perform a matrix transformation on the first area matrix through the embedding layer to obtain a first area vector of a specified dimension.

[0117] Optionally, the encoding layer includes several encoding modules, each encoding module includes a normalization network, a multi-head attention network, a residual connection network and a feedforward neural network;

[0118] The encoding module 206 is specifically used to determine a first embedding vector based on the first signal vector, the first position vector and the first area vector; input the first embedding vector into the encoding layer of the diagnostic agent model to be trained, so as to extract features of the first embedding vector through several encoding modules included in the encoding layer to obtain a first encoding feature.

[0119] Optionally, the training module 210 is also used to determine a second measurement signal, a second chord integral information and a second physical boundary corresponding to a first target task, and use them as a first fine-tuning sample, and to determine a two-dimensional density distribution profile corresponding to the first fine-tuning sample, and use them as a first fine-tuning annotation; wherein the first target task is a task of measuring the density distribution of the plasma; inputting the second measurement signal in the first fine-tuning sample into an embedding layer of a trained diagnostic proxy model to determine a second signal vector; inputting the second chord integral information in the first fine-tuning sample into the embedding layer to determine a second position vector; inputting the second physical boundary in the first fine-tuning sample into the embedding layer to determine a second area vector; inputting the second signal vector, the second position vector and the second area vector into an encoding layer of the diagnostic proxy model to determine a second encoding feature; determining a second predicted profile corresponding to the first fine-tuning sample based on the second encoding feature; and fine-tuning the diagnostic proxy model based on the second predicted profile and the first fine-tuning annotation to obtain a first target proxy model.

[0120] Optionally, the device further comprises:

[0121] The application module 212 is used to determine the physical boundary to be measured of the plasma to be measured; determine the signal to be measured of the plasma to be measured collected by a preset diagnostic device, and determine the chord integral information to be measured of the measurement optical path corresponding to the signal to be measured; input the signal to be measured, the chord integral information to be measured and the physical boundary to be measured into the trained diagnostic agent model, and determine the predicted profile of the plasma to be measured output by the diagnostic agent model.

[0122] This specification also provides a computer-readable storage medium, which stores a computer program, which can be used to execute the above Figure 1 A training method for a diagnostic agent model is provided.

[0123] This manual also provides Figure 7 The one shown corresponds to Figure 1 A schematic diagram of the electronic device. Figure 7As shown, at the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and may also include other hardware required for the business. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 1 The training method of the diagnostic agent model.

[0124] Of course, in addition to software implementation, this specification does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc., that is to say, the executor of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0125] In the 1990s, it was very clear whether the improvement of a technology was hardware improvement (for example, improvement of the circuit structure of diodes, transistors, switches, etc.) or software improvement (improvement of the method flow). However, with the development of technology, many improvements of the method flow today can be regarded as direct improvements of the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that the improvement of a method flow cannot be implemented with hardware entity modules. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital system on a PLD by programming it themselves, without having to ask chip manufacturers to design and make dedicated integrated circuit chips. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs, and the original code before compilation must also be written in a specific programming language, which is called hardware description language (HDL). There is not only one kind of HDL, but many 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), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages ​​and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that implements the logic method flow.

[0126] The controller may be implemented in any suitable manner, for example, the controller may take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (e.g., software or firmware) executable by the (micro)processor, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller, examples of which include but are not limited to the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320, and the memory controller may also be implemented as part of the control logic of the memory. It is also known to those skilled in the art that, in addition to implementing the controller in a purely computer-readable program code manner, the controller may be implemented in the form of a logic gate, a switch, an application-specific integrated circuit, a programmable logic controller, and an embedded microcontroller by logically programming the method steps. Therefore, such a controller may be considered as a hardware component, and the devices for implementing various functions included therein may also be considered as structures within the hardware component. Or even, the devices for implementing various functions may be considered as both software modules for implementing the method and structures within the hardware component.

[0127] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, 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.

[0128] For the convenience of description, the above device is described in various units according to their functions. Of course, when implementing this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0129] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Therefore, this specification may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0130] This specification is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0131] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0132] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0133] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0134] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0135] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media 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 disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0136] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0137] It should be understood by those skilled in the art that the embodiments of this specification may be provided as methods, systems or computer program products. Therefore, this specification may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0138] This specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. This specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0139] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0140] The above description is only an embodiment of the present specification and is not intended to limit the present specification. For those skilled in the art, the present specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present specification shall be included in the scope of the claims of the present specification.

Claims

1. A method for training a diagnostic agent model, characterized in that: The method comprises: Acquire a preset two-dimensional physical parameter profile of the plasma, and determine a first physical boundary corresponding to the plasma; Determine a first measurement signal corresponding to the plasma collected by a preset diagnostic device, and determine first chord integral information of a measurement optical path corresponding to the first measurement signal; Inputting the first measurement signal into the embedding layer of the diagnostic agent model to be trained to determine a first signal vector; inputting the first chord integral information into the embedding layer to determine a first position vector; inputting the first physical boundary into the embedding layer to determine a first region vector; Inputting the first signal vector, the first position vector and the first region vector into the encoding layer of the diagnostic agent model to be trained to determine a first encoding feature; Determining a first predicted profile corresponding to the plasma according to the first coding feature; The diagnostic agent model to be trained is trained according to the first predicted profile and the two-dimensional physical parameter profile.

2. The method according to claim 1, characterized in that The diagnostic device comprises a plurality of channels, the dimension of the first measurement signal being the same as the number of the channels; Inputting the first measurement signal into the embedding layer of the diagnostic agent model to be trained to determine a first signal vector specifically includes: The first measurement signal is input into an embedding layer of a diagnostic agent model to be trained, so as to perform a matrix transformation on the first measurement signal through the embedding layer to obtain a first signal vector of a specified dimension.

3. The method according to claim 1, characterized in that The diagnostic device comprises a plurality of channels, and the first chord integral information is the length of the measuring light path corresponding to each channel passing through the plasma; Inputting the first chord integral information into the embedding layer to determine a first position vector specifically includes: The first chord integral information is input into the embedding layer so as to perform matrix transformation on the first chord integral information through the embedding layer to obtain a first position vector of a specified dimension.

4. The method according to claim 1, characterized in that The first physical boundary is a region corresponding to the plasma in a preset measurement space; Inputting the first physical boundary into the embedding layer to determine a first region vector specifically includes: Gridding the preset measurement space to obtain a number of grid areas; Marking each grid area according to the first physical boundary to obtain a first area matrix corresponding to the first physical boundary; The first region matrix is ​​input into the embedding layer so as to perform matrix transformation on the first region matrix through the embedding layer to obtain a first region vector of a specified dimension.

5. The method according to claim 1, characterized in that The encoding layer includes several encoding modules, each of which includes a normalization network, a multi-head attention network, a residual connection network and a feedforward neural network; Inputting the first signal vector, the first position vector and the first region vector into the encoding layer of the diagnostic agent model to be trained to determine a first encoding feature specifically includes: Determine a first embedding vector according to the first signal vector, the first position vector and the first region vector; The first embedding vector is input into the encoding layer of the diagnostic agent model to be trained, so as to perform feature extraction on the first embedding vector through a plurality of encoding modules included in the encoding layer to obtain a first encoding feature.

6. The method according to claim 1, characterized in that The method further comprises: Determine a second measurement signal, second chord integral information, and second physical boundary corresponding to a first target task and use them as a first fine-tuning sample, and determine a two-dimensional density distribution profile corresponding to the first fine-tuning sample and use it as a first fine-tuning annotation; wherein the first target task is a task of measuring the density distribution of the plasma; Inputting the second measurement signal in the first fine-tuning sample into the embedding layer of the trained diagnostic agent model to determine a second signal vector; inputting the second chord integral information in the first fine-tuning sample into the embedding layer to determine a second position vector; inputting the second physical boundary in the first fine-tuning sample into the embedding layer to determine a second region vector; Inputting the second signal vector, the second position vector and the second region vector into the encoding layer of the diagnostic agent model to determine a second encoding feature; Determining a second prediction profile corresponding to the first fine-tuning sample according to the second encoding feature; The diagnostic proxy model is fine-tuned according to the second predicted profile and the first fine-tuning annotation to obtain a first target proxy model.

7. The method according to claim 1, characterized in that The method further comprises: determining a physical boundary of the plasma to be measured; Determine a signal to be measured of the plasma to be measured collected by a preset diagnostic device, and determine chord integral information to be measured of a measurement optical path corresponding to the signal to be measured; The signal to be measured, the chord integral information to be measured and the physical boundary to be measured are input into a trained diagnostic proxy model to determine a predicted profile of the plasma to be measured output by the diagnostic proxy model.

8. A diagnostic agent model training device, characterized in that: include: A first determination module, used to obtain a preset two-dimensional physical parameter profile of the plasma and determine a first physical boundary corresponding to the plasma; A second determination module is used to determine a first measurement signal corresponding to the plasma collected by a preset diagnostic device, and to determine first chord integral information of a measurement optical path corresponding to the first measurement signal; An embedding module, configured to input the first measurement signal into an embedding layer of a diagnostic agent model to be trained to determine a first signal vector; Inputting the first chord integral information into the embedding layer to determine a first position vector; inputting the first physical boundary into the embedding layer to determine a first area vector; An encoding module, used for inputting the first signal vector, the first position vector and the first region vector into an encoding layer of the diagnostic agent model to be trained to determine a first encoding feature; A prediction module, configured to determine a first predicted profile corresponding to the plasma according to the first coding feature; A training module is used to train the diagnostic agent model to be trained according to the first predicted section and the two-dimensional physical parameter section.

9. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method described in any one of claims 1 to 7 is implemented.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method described in any one of claims 1 to 7 is implemented.