Particle turbulence data identification method and system based on artificial intelligence

Abstract

The embodiment of the invention discloses a particle turbulence data identification method and system based on artificial intelligence, and belongs to the technical field of data processing. The embodiment of the invention relates to a particle turbulence analysis technology of a Tokamak device. When a Tokamak device runs, particle turbulence data are collected by using multi-source sensors such as a magnetic probe array, a laser scatterometer and a Thomson scattering diagnosis system; performing integration processing on the acquired data to obtain an integrated data set comprising a time-aligned particle density-magnetic field coupling data segment and a space-calibrated temperature fluctuation-density fluctuation associated data block; inputting the integrated data set into a turbulence characteristic correlation model, performing multi-dimensional characteristic correlation analysis, and generating a correlation characteristic set of particle motion mode correlation characteristics and turbulence evolution rule correlation characteristics; and finally, accurately generating a particle motion pattern recognition result and a turbulence evolution rule recognition result of the Tokamak device according to the associated feature set.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a method and system for identifying particle turbulence data based on artificial intelligence. Background Technology

[0002] In tokamak device research, accurately acquiring and analyzing particle turbulence data within the device is crucial for understanding its operational mechanisms and optimizing its performance. Current technologies typically employ only a single type of sensor to collect particle turbulence-related data. This approach yields data with limited dimensions, failing to comprehensively reflect the complex characteristics of particle turbulence. Furthermore, existing technologies lack effective data integration and processing, resulting in a lack of temporal and spatial correlation between different data types and consequently, low data quality. In terms of data analysis, existing methods often only perform simple feature analysis, failing to uncover the deep correlation between particle motion patterns and turbulence evolution, and thus unable to accurately identify particle motion patterns and turbulence evolution laws. Therefore, accurately identifying particle turbulence data from tokamak devices is a key technical challenge that needs to be overcome. Summary of the Invention

[0003] This invention provides a method and system for identifying particle turbulence data based on artificial intelligence, which can solve or partially solve the technical problems involved in the background art.

[0004] This invention provides an artificial intelligence-based particle turbulence data identification method, applied to particle turbulence data identification technology. The method includes: collecting particle turbulence data through multi-source sensors during the operation of a tokamak device, including a magnetic probe array, a laser scattering instrument, and a Thomson scattering diagnostic system; integrating the particle turbulence data to obtain an integrated dataset, which includes time-aligned particle density-magnetic field coupling data segments and spatially calibrated temperature fluctuation-density fluctuation correlation data blocks; inputting the integrated dataset into a turbulence feature correlation model to perform multi-dimensional feature correlation analysis, generating a correlation feature set of particle motion pattern correlation features and turbulence evolution law correlation features; and generating particle motion pattern identification results and turbulence evolution law identification results for the tokamak device based on the correlation feature set. The particle motion pattern identification results include particle cyclotron motion features, random walk features, and constraint boundary escape features, while the turbulence evolution law identification results include turbulent coherent structure growth features, energy cascade transfer features, and dissipation decay features.

[0005] This invention provides a particle turbulence data identification technology, including at least one processor and a memory; the memory stores computer execution instructions; the at least one processor executes the computer execution instructions stored in the memory, causing the at least one processor to perform the above-described method.

[0006] This invention provides a readable storage medium on which a program or instruction is stored, and when the program or instruction is executed by a processor, it implements the steps of the above method.

[0007] This invention collects particle turbulence data during the operation of a tokamak device using multi-source sensors, enabling the acquisition of relevant information from different dimensions and ensuring the comprehensiveness and diversity of the data. The collected particle turbulence data is integrated to obtain an integrated dataset containing time-aligned particle density-magnetic field coupling data segments and spatially calibrated temperature fluctuation-density fluctuation correlation data blocks, improving data quality and usability and making the data more correlated and accurate in both time and space dimensions. Inputting the integrated dataset into a turbulence feature correlation model for multi-dimensional feature correlation analysis reveals the intrinsic relationship between particle motion patterns and turbulence evolution laws, generating a comprehensive and targeted set of correlation features. Based on this set of correlation features, accurate particle motion pattern recognition results and turbulence evolution law recognition results can be generated for the tokamak device, thereby accurately and comprehensively revealing particle behavior and turbulence phenomena within the tokamak device and guiding the operation and performance optimization of the tokamak device. Attached Figure Description

[0008] Figure 1 This is a flowchart of an artificial intelligence-based particle turbulence data identification method provided in an embodiment of the present invention.

[0009] Figure 2 This is a schematic diagram of a particle turbulence data identification technology provided in an embodiment of the present invention. Detailed Implementation

[0010] Figure 1 An artificial intelligence-based particle turbulence data identification method is shown, which is applied to particle turbulence data identification technology. The method includes the following steps 110-140.

[0011] Step 110: During the operation of the tokamak device, particle turbulence data are collected by a multi-source sensor, which includes a magnetic probe array, a laser scatterer, and a Thomson scattering diagnostic system.

[0012] After the tokamak device is started up and enters a stable operating state, multi-source sensors begin to collect particle turbulence data. Among them, the magnetic probe array consists of multiple magnetic probes arranged in a specified layout. These magnetic probes are distributed at different locations within the tokamak device and can sense changes in the magnetic field within the device in real time. Since particle turbulence causes disturbances in the magnetic field, the magnetic probes convert these changes into electrical signals. When collecting data, the magnetic probes sample at regular time intervals, continuously recording information such as the strength and direction of the magnetic field at different moments, thus obtaining a magnetic field disturbance signal with time-series labeling.

[0013] A laser scatterer operates by emitting a laser beam into the plasma within a tokamak device. When the laser interacts with particles in the plasma, scattering occurs. The laser scatterer detects various characteristics of the scattered light, such as intensity and frequency. By analyzing these characteristics, the density fluctuations of the particles in the plasma can be inferred. The laser scatterer also collects data sequentially over time, obtaining a set of values ​​reflecting the particle density fluctuations at each moment, ultimately forming a particle density fluctuation signal with a time-series label.

[0014] The Thomson scattering diagnostic system utilizes the interaction between a high-energy electron beam and particles in a plasma to generate Thomson scattering. Through detailed analysis of the scattered light, information on plasma temperature fluctuations can be obtained. During operation, the system measures and records the temperature changes of the plasma at specified time points, thus obtaining temperature fluctuation signals with time-series labels.

[0015] The aforementioned multi-source sensors monitor the plasma from different physical perspectives. For example, the magnetic field disturbance signal collected by the magnetic probe array can reflect the influence of particle motion on the magnetic field; the particle density fluctuation signal collected by the laser scattering instrument can reflect the changes in particle distribution in space; and the temperature fluctuation signal collected by the Thomson scattering diagnostic system helps to understand the changes in the energy state of the plasma.

[0016] Step 120: Perform data integration processing on the particle turbulence data to obtain an integrated dataset, which includes time-aligned particle density-magnetic field coupling data segments and spatially calibrated temperature fluctuation-density fluctuation correlation data blocks.

[0017] After collecting particle turbulence data, inconsistencies arise in time and space due to differences in the operating characteristics of different sensors and acquisition methods. To more effectively analyze and utilize the data, data integration processing is necessary.

[0018] In terms of time, particle density fluctuation signals, magnetic field disturbance signals, and temperature fluctuation signals may be acquired at different points in time. For example, the sampling frequency of a magnetic probe array may differ from that of a laser scattering instrument and a Thomson scattering diagnostic system, causing their acquired data to not directly correspond in time. To solve this problem, time offset error correction is needed for these signals. This can be achieved by finding characteristic points or reference points in each signal and aligning them on the time axis. For example, using the moment of occurrence of a specified physical event as a reference point, the time series of different signals can be adjusted to synchronize them in time.

[0019] Spatially, the measurement positions of different sensors also differ. The magnetic probe array, laser scatterer, and Thomson scattering diagnostic system perform measurements at different spatial locations within the tokamak device. The magnetic probe array may be distributed circumferentially within the device, the laser scatterer primarily measures radial particle density, and the Thomson scattering diagnostic system focuses on polar temperature measurement. To ensure spatial consistency of the data, spatial position calibration based on the tokamak device's geometric coordinate system is necessary. This involves mapping the circumferential coordinates of the magnetic probe array, the radial measurement coordinates of the laser scatterer, and the polar observation coordinates of the Thomson scattering diagnostic system onto this geometric coordinate system.

[0020] After time alignment and spatial calibration, we obtain a time-aligned particle density-magnetic field coupling data segment and a spatially calibrated temperature fluctuation-density fluctuation correlation data block. The particle density-magnetic field coupling data segment synchronously correlates the particle density fluctuation signal and the magnetic field perturbation signal in time, reflecting the relationship between particle density changes and magnetic field changes. The temperature fluctuation-density fluctuation correlation data block calibrates and correlates the temperature fluctuation signal and the density fluctuation signal in space, which helps to analyze the spatial correlation characteristics between temperature changes and particle density fluctuations. These data blocks and segments together constitute the integrated dataset.

[0021] In one example, the particle turbulence data includes time-series labeled particle density fluctuation signals, magnetic field disturbance signals, and temperature fluctuation signals. The data integration processing of the particle turbulence data yields an integrated dataset, including: Step 121: Perform time offset error correction on the particle density fluctuation signal, the magnetic field disturbance signal, and the temperature fluctuation signal respectively to generate the original synchronization data sequence.

[0022] To generate the original synchronized data sequence, time offset error correction needs to be performed on the particle density fluctuation signal, magnetic field disturbance signal, and temperature fluctuation signal. Because different sensors may have different sampling mechanisms and clock accuracies, these signals may be time-shifted. For example, the sampling clock of the magnetic probe array may deviate slightly from that of the laser scatterer, causing the data they collect to not correspond perfectly in time.

[0023] This step requires establishing a unified time reference standard. A specific physical event during the operation of the tokamak device, such as the start of a plasma discharge, can be chosen as the time reference point. Then, for the particle density fluctuation signal, by analyzing characteristic points in the data, such as signal peaks, troughs, or specified trend points, these characteristic points are compared with the time reference point to calculate the time offset of the particle density fluctuation signal relative to the time reference point. Based on this offset, the time series of the particle density fluctuation signal is adjusted to align with the time reference point.

[0024] For magnetic field disturbance signals, a similar method is used to analyze the feature points in the magnetic field disturbance signal, match them with the time reference point, determine the time offset of the magnetic field disturbance signal, and correct its time series.

[0025] The same steps are followed for temperature fluctuation signals: find the feature points in the temperature fluctuation signals, calculate the time offset, and then adjust the time series.

[0026] After correcting the time offset error of each of the three signals, they were synchronized in time, generating an original synchronized data sequence. This original synchronized data sequence provides a temporally consistent data basis for subsequent spatial calibration and further analysis.

[0027] Step 122: Based on the geometric coordinate system of the tokamak device, the original synchronous data sequence is spatially calibrated. The circumferential position coordinates of the magnetic probe array, the radial measurement position coordinates of the laser scatterer, and the poloidal observation position coordinates of the Thomson scattering diagnostic system are uniformly mapped to the geometric coordinate system to generate a spatially calibrated multi-physical quantity synchronous data set.

[0028] After obtaining the original synchronization data sequence, since the measurement positions of different sensors are distributed in different spatial directions of the tokamak device, the magnetic probe array mainly measures the circumferential magnetic field information, the laser scatterer focuses on the radial particle density measurement, and the Thomson scattering diagnostic system focuses on the polar temperature. In order to comprehensively analyze the relationship between these different physical quantities, it is necessary to perform spatial position calibration of the original synchronization data sequence based on the geometric coordinate system of the tokamak device.

[0029] The geometric coordinate system of a tokamak device is a unified coordinate system used to describe spatial positions within the device, with clearly defined coordinates and orientations. First, it is necessary to determine the actual coordinates of the measurement positions of the magnetic probe array, laser scatterer, and Thomson scattering diagnostic system within the device. The circumferential position coordinates of the magnetic probe array represent its position along the circumference of the tokamak device; the radial measurement position coordinates of the laser scatterer reflect its distance from the center to the edge of the device; and the poloidal observation position coordinates of the Thomson scattering diagnostic system reflect its position in the vertical direction of the device.

[0030] Then, using coordinate transformation, these coordinates in different directions are uniformly mapped to the geometric coordinate system of the tokamak device. For example, the circumferential position coordinates of the magnetic probe array are converted into corresponding coordinate values ​​in the geometric coordinate system according to its definition. Corresponding transformation rules are also applied to the radial measurement position coordinates of the laser scatterer and the poloidal observation position coordinates of the Thomson scattering diagnostic system.

[0031] After coordinate mapping is completed, the particle density fluctuation signal, magnetic field disturbance signal, and temperature fluctuation signal in the original synchronous data sequence are associated with the transformed coordinates. Thus, each data point corresponds to a specific location in the geometric coordinate system, forming a spatially calibrated multi-physical quantity synchronous data group. The data in this data group are synchronized in time and calibrated in space.

[0032] Step 123: Perform data segmentation processing on the multi-physical quantity synchronization data group, and extract the associated data units within the continuous time window according to the triggering characteristics of particle turbulence events. The associated data units include the temporal overlap interval of particle density fluctuation and magnetic field disturbance, and the spatial overlap region of temperature fluctuation and density fluctuation.

[0033] After obtaining a space-calibrated multi-physics synchronous data set, it is necessary to segment the data set into data segments to more effectively analyze particle turbulence events. Particle turbulence events typically have certain triggering characteristics, which can be identified through data analysis.

[0034] First, a continuous time window is determined. The size of this window can be determined based on the actual research needs and the characteristics of the particle turbulence event. Within this time window, a detailed analysis of the multi-physical quantity synchronous data set is performed. For data on particle density fluctuations and magnetic field disturbances, their overlapping intervals in the time domain are identified. Since particle density fluctuations and magnetic field disturbances may influence each other, their trends may show correlation within certain time periods. By analyzing the temporal changes of these two physical quantities, the time periods in which they change significantly simultaneously are identified; these time periods are the overlapping intervals in the time domain between particle density fluctuations and magnetic field disturbances. For data on temperature fluctuations and density fluctuations, their spatial overlap is considered. Since temperature fluctuations and density fluctuations may be spatially correlated, their changes may influence each other at certain spatial locations. By analyzing the spatially calibrated coordinates and data values ​​of these two physical quantities, the spatial regions in which they change simultaneously are identified; these regions are the spatial overlap regions between temperature fluctuations and density fluctuations.

[0035] Data from the aforementioned overlapping temporal and spatial regions are extracted to form correlated data units. These correlated data units contain information closely related to particle turbulence events and can more accurately reflect the temporal and spatial characteristics of particle turbulence phenomena.

[0036] Step 124: Generate the integrated dataset based on the associated data units.

[0037] Based on the previously extracted correlated data units, an integrated dataset can be generated. The correlated data units already contain data on the temporal overlap between particle density fluctuations and magnetic field disturbances, as well as the spatial overlap between temperature fluctuations and density fluctuations. These data are correlated in both time and space and are key information for analyzing particle turbulence phenomena.

[0038] These correlated data units were organized and combined to form a complete dataset. In this dataset, the particle density-magnetic field coupling data segment is time-aligned because it was extracted from overlapping temporal regions, ensuring temporal consistency; the temperature fluctuation-density fluctuation correlated data block is spatially calibrated because it was extracted from spatially overlapping regions, ensuring spatial consistency. This integrated dataset contains important time-aligned and spatially calibrated data, providing a high-quality data foundation for subsequent particle turbulence analysis. Further processing and analysis of this integrated dataset can provide a deeper understanding of particle motion patterns and turbulence evolution within the tokamak device.

[0039] Step 130: Input the integrated dataset into the turbulence feature association model, perform multi-dimensional feature association analysis, and generate a set of associated features of particle motion mode association features and turbulence evolution law association features.

[0040] After obtaining the integrated dataset, it is input into the turbulence feature correlation model for multi-dimensional feature correlation analysis. The turbulence feature correlation model is specifically designed for analyzing particle turbulence data. It can mine and analyze data from multiple dimensions to find the correlations between different physical quantities. This model performs a comprehensive analysis of the particle density-magnetic field coupling data segment and the temperature fluctuation-density fluctuation correlation data block in the integrated dataset. For the particle density-magnetic field coupling data segment, the model analyzes the interaction between particle density fluctuations and magnetic field disturbances, identifying their correlation characteristics at different time scales and spatial locations. For example, a sudden increase in particle density over a certain period may lead to a corresponding change in the magnetic field, and the model captures this correlation information.

[0041] For data blocks related to temperature fluctuations and density fluctuations, the model studies the causal relationship and evolution of these two factors. Within certain spatial regions, temperature changes may cause fluctuations in particle density; the model analyzes the manifestation of this correlation at different time points.

[0042] Through multi-dimensional feature correlation analysis, the model generates particle motion pattern correlation features and turbulence evolution law correlation features. Particle motion pattern correlation features reflect different particle motion modes within the tokamak device, such as cyclotron motion, random walk, and escape from constraint boundaries. Turbulence evolution law correlation features demonstrate the development and changes of turbulence at different stages, such as the growth of turbulent coherent structures, energy cascade transfer, and dissipation decay.

[0043] These particle motion pattern correlation features and turbulence evolution law correlation features are combined to form a correlation feature set. This correlation feature set contains rich information and provides an important basis for subsequent identification of particle motion patterns and turbulence evolution laws in the tokamak device.

[0044] As an example, the step of performing multi-dimensional feature correlation analysis to generate a set of correlation features for particle motion patterns and turbulence evolution laws includes: Step 131: Call the multi-scale feature extraction module of the turbulence feature association model to perform feature extraction processing on the particle density-magnetic field coupling data segment in the integrated dataset at different time resolutions, and generate multi-scale density magnetic field association features including high-frequency fluctuation features, mid-frequency coherence features and low-frequency trend features.

[0045] In the turbulence feature correlation model, the multi-scale feature extraction module is specifically used to extract features from the particle density-magnetic field coupling data segment in the integrated dataset. Since particle density fluctuations and magnetic field disturbances exhibit different characteristics at different time scales, analysis from multiple time resolutions is required.

[0046] The multi-scale feature extraction module first divides the particle density-magnetic field coupling data segment into different time windows. By setting time windows of different sizes, the data segment is divided into multiple sub-segments. Different time windows correspond to different time resolutions, enabling the capture of data variation characteristics at different time scales.

[0047] For high-frequency fluctuation characteristics, the module focuses on changes in data over shorter time scales. By analyzing the changes in particle density-magnetic field coupled data segments within a small time window, it extracts features such as the instantaneous rate of change and phase difference of particle density and magnetic field disturbances. These features can reflect the rapid changes of particles and magnetic fields in a short period of time, demonstrating the characteristics of high-frequency fluctuations.

[0048] For mid-frequency coherence features, the module searches for periodic changes in the data over a medium time scale. Within a medium-sized time window, it identifies periodic density-magnetic field synchronous oscillation modes in the particle density-magnetic field coupled data segment. These synchronous oscillation modes reflect the interaction patterns between particles and magnetic fields within a certain time range, forming mid-frequency coherence features.

[0049] For low-frequency trend features, the module analyzes the changing trends of the data over a longer time scale. Within a large time window, it extracts features such as the slope of the long-term changing trends of particle density and magnetic field disturbances. These features can reflect the overall changing trends of particles and magnetic fields over a longer period of time, forming low-frequency trend features.

[0050] Finally, the extracted high-frequency fluctuation features, mid-frequency coherence features, and low-frequency trend features are combined to form multi-scale density magnetic field correlation features. This feature set contains feature information of particle density-magnetic field coupled data segments at different time scales.

[0051] In a preferred embodiment, the step of calling the multi-scale feature extraction module of the turbulence feature association model to perform feature extraction processing on the particle density-magnetic field coupling data segment in the integrated dataset at different time resolutions, generating multi-scale density-magnetic field association features containing high-frequency fluctuation features, mid-frequency coherence features, and low-frequency trend features, including: Step 1311: Input the particle density-magnetic field coupling data segment into the sliding window processing layer of the multi-scale feature extraction module, and use the first time window, the second time window and the third time window to window and truncate the data segment to generate three sets of data sub-segments with different time spans.

[0052] In the multi-scale feature extraction module, the sliding window processing layer is a key part for windowing and truncating the particle density-magnetic field coupled data segment. This layer processes the data segment according to the size of the preset first time window, second time window and third time window.

[0053] The sizes of the first, second, and third time windows are set according to different time resolution requirements. The first time window is usually small and used to capture rapid changes in data over a short period of time; the second time window is of moderate size and can analyze the periodic characteristics of data at a medium time scale; the third time window is large and used to study the overall trend of data changes over a long period of time.

[0054] The sliding window processing layer starts from the beginning of the particle density-magnetic field coupling data segment and truncates it according to the size of the first time window to obtain the first data segment. Then, the window slides forward by a fixed step and truncates the second data segment, and so on, until the entire data segment is covered, generating a set of data segments based on the first time window.

[0055] The same sliding window truncation method is used for the second and third time windows. The particle density-magnetic field coupling data segment is truncated according to its respective window size and step size to generate two other sets of data segments with different time spans.

[0056] By using this windowing method, the particle density-magnetic field coupling data segment was divided into three data segments with different time spans. These data segments contain information about the data at different time scales, providing a foundation for subsequent extraction of high-frequency fluctuation features, mid-frequency coherence features, and low-frequency trend features.

[0057] Step 1312: Perform local feature extraction processing on each data segment: For the data segment in the first time window, calculate the instantaneous rate of change and phase difference of particle density and magnetic field disturbance as high-frequency fluctuation features; for the data segment in the second time window, identify the periodically occurring density-magnetic field synchronous oscillation mode as mid-frequency coherent features; for the data segment in the third time window, extract the slope of the long-term trend of particle density and magnetic field disturbance as low-frequency trend features.

[0058] After obtaining three sets of data segments with different time spans, local feature extraction processing needs to be performed on each set of data segments. Different time windows correspond to different feature extraction methods to capture the characteristics of the data at different time scales.

[0059] For the first time window data segment, due to its short time span, the focus is mainly on the rapid changes in particle density and magnetic field disturbances within a short period. By calculating the instantaneous rates of change of particle density and magnetic field disturbances, we can understand their rate of change at each instant. Simultaneously, calculating the phase difference between particle density and magnetic field disturbances allows us to analyze their relative relationship over time. These instantaneous rates of change and phase differences together constitute the high-frequency fluctuation characteristics, reflecting the high-frequency changes of particles and the magnetic field within a short time.

[0060] The second time window data segment has a moderate time span, making it suitable for identifying periodic changes in the data. Within this time window, the changes in the particle density-magnetic field coupled data segment are analyzed, and periodic density-magnetic field synchronous oscillation modes are identified. These synchronous oscillation modes indicate that particle density and magnetic field disturbances exhibit synchronous periodic changes within a certain time range, forming mid-frequency coherent characteristics and reflecting the interaction law between particles and magnetic fields on a medium time scale.

[0061] For the third time window data segment, due to its long time span, the focus is mainly on the long-term variation trends of particle density and magnetic field perturbation. By performing trend analysis on this data segment, the slope of the long-term variation trend of particle density and magnetic field perturbation is extracted. This slope can reflect the overall direction and rate of change of particles and magnetic field over a long period of time, forming a low-frequency trend feature, which reflects the macroscopic variation characteristics of particles and magnetic field on a long time scale.

[0062] Step 1313: The high-frequency fluctuation feature, the mid-frequency coherence feature and the low-frequency trend feature are spliced ​​along the feature dimension to generate a multi-scale density magnetic field correlation feature with multiple time scale representations.

[0063] After extracting high-frequency fluctuation features, mid-frequency coherence features, and low-frequency trend features, these features from different time scales need to be concatenated to integrate them. Concatenating along the feature dimension means arranging and combining these features in a certain order to form a feature set containing information from multiple time scales.

[0064] First, determine the feature dimensions of high-frequency fluctuation characteristics, mid-frequency coherence characteristics, and low-frequency trend characteristics. Each feature has corresponding dimensions, representing different characteristic attributes. For example, high-frequency fluctuation characteristics may include dimensions such as the instantaneous rate of change and phase difference of particle density and magnetic field disturbance; mid-frequency coherence characteristics may include the relevant parameter dimensions of the density-magnetic field synchronous oscillation mode; and low-frequency trend characteristics may include dimensions such as the slope of the long-term trend of particle density and magnetic field disturbance.

[0065] Then, these features are concatenated according to feature dimensions. High-frequency fluctuation features can be placed first, followed by mid-frequency coherent features, and finally low-frequency trend features. This arrangement can comprehensively show the hierarchical relationship of features at different time scales.

[0066] Through splicing, a multi-scale density magnetic field correlation feature with multiple time scales was generated. This feature set contains feature information of particle density-magnetic field coupling data segments at short, medium, and long time scales.

[0067] Step 132: Using the time-series dependency modeling module of the turbulence feature association model, perform time-series evolution analysis on the temperature fluctuation-density fluctuation association data block, extract the causal relationship between temperature fluctuation and density fluctuation within a continuous time step, and generate time-series association features including precursor signal features, synchronous coupling features and hysteresis response features.

[0068] In this embodiment of the invention, the time-series dependency modeling module of the turbulence feature correlation model is mainly used to perform time-series evolution analysis on the temperature fluctuation-density fluctuation correlation data block. Since temperature fluctuations and density fluctuations have a certain temporal correlation, the causal relationship can be extracted by analyzing their changes over continuous time steps.

[0069] The time-series dependency modeling module first divides the temperature fluctuation-density fluctuation correlation data block into time steps. The data block is divided into consecutive time step units according to the same time interval, and each time step unit contains the temperature fluctuation value and density fluctuation value at that moment.

[0070] For precursor signal characteristics, the module analyzes the relationship between temperature fluctuations at the current time step and density fluctuations at the previous time step. It measures the precursory impact of temperature fluctuations on density fluctuations by calculating a correlation index, such as the Pearson correlation coefficient. A large Pearson correlation coefficient indicates that density fluctuations at the previous time step may have influenced temperature fluctuations at the current time step, thus forming precursor signal characteristics.

[0071] For the synchronous coupling feature, the module focuses on the relationship between the temperature fluctuation and density fluctuation at the current time step. By calculating their covariance, it reflects the degree of synchronous change between temperature fluctuation and density fluctuation at the same time point. A larger covariance indicates a strong synchronous coupling relationship between temperature fluctuation and density fluctuation at the current time step, forming the synchronous coupling feature.

[0072] For hysteresis response characteristics, the module studies the relationship between density fluctuations at the current time step and temperature fluctuations at the previous time step. It measures the hysteresis response of density fluctuations to temperature fluctuations by calculating their mutual information entropy. A larger mutual information entropy indicates that the density fluctuation at the current time step may be a hysteresis response to the temperature fluctuation at the previous time step, thus forming a hysteresis response characteristic.

[0073] Finally, the precursor signal features, synchronization coupling features, and hysteresis response features are integrated to generate time-series correlation features. This feature set reflects the causal relationship between temperature fluctuations and density fluctuations over time.

[0074] In another preferred embodiment, the time-series dependency modeling module of the turbulence feature association model performs time-series evolution analysis on the temperature fluctuation-density fluctuation association data block, extracts the causal relationship between temperature fluctuation and density fluctuation within a continuous time step, and generates time-series association features including precursor signal features, synchronous coupling features, and hysteresis response features, including: Step 1321: Perform time step alignment processing on the temperature fluctuation-density fluctuation correlation data block, and divide the temperature fluctuation data and density fluctuation data into continuous time step units according to the same time interval.

[0075] To accurately analyze the temporal causal relationship between temperature fluctuations and density fluctuations, it is necessary to align the time steps of the temperature fluctuation-density fluctuation correlation data blocks. Since temperature fluctuation data and density fluctuation data may be collected at different time points or with different sampling frequencies, they need to be uniformly divided in time.

[0076] First, a suitable time interval is determined, which should be chosen based on the characteristics of the data and the needs of the analysis. Then, the temperature fluctuation data and density fluctuation data are divided into continuous time step units according to this time interval. Each time step unit contains the temperature fluctuation values ​​and density fluctuation values ​​within that time period. During the division process, data interpolation or sampling adjustments are required to ensure that each time step unit has corresponding temperature fluctuation values ​​and density fluctuation values. If there is no corresponding data at a certain time point, it can be estimated using interpolation methods based on data from adjacent time points. Through time step alignment, the temperature fluctuation data and density fluctuation data achieve temporal consistency, with each time step unit corresponding to the same time range.

[0077] Step 1322: For each time step unit, calculate the Pearson correlation coefficient between the temperature fluctuation value of the current time step and the density fluctuation value of the previous time step, as a precursor signal feature of temperature fluctuation on density fluctuation.

[0078] After completing the time step alignment process, for each time step unit, it is necessary to calculate the Pearson correlation coefficient between the temperature fluctuation value of the current time step and the density fluctuation value of the previous time step in order to obtain the precursor signal characteristics of temperature fluctuation on density fluctuation.

[0079] The Pearson correlation coefficient is an indicator used to measure the linear correlation between two variables. For each time step, the temperature fluctuation value of the current time step and the density fluctuation value of the previous time step are used as two variables for calculation.

[0080] First, calculate the mean of the temperature fluctuation value at the current time step and the mean of the density fluctuation value at the previous time step. Then, calculate the difference between each value and its respective mean. Next, multiply these differences and sum them to obtain the numerator of the covariance. Simultaneously, calculate the standard deviation of the temperature fluctuation value at the current time step and the density fluctuation value at the previous time step. Divide the numerator of the covariance by the product of the two standard deviations to obtain the Pearson correlation coefficient, which reflects the degree of linear correlation between the temperature fluctuation value at the current time step and the density fluctuation value at the previous time step. A large correlation coefficient indicates that the density fluctuation at the previous time step may have influenced the temperature fluctuation at the current time step, forming a precursor signal characteristic of temperature fluctuation on density fluctuation. By calculating the Pearson correlation coefficient for each time step unit, a set of data reflecting the precursor signal characteristics is obtained.

[0081] Step 1323: Calculate the covariance between the temperature fluctuation value and the density fluctuation value at the current time step, as the synchronous coupling feature between temperature fluctuation and density fluctuation.

[0082] After calculating the characteristics of the precursor signal, in order to analyze the synchronous changes of temperature fluctuations and density fluctuations at the same time point, it is necessary to calculate the covariance between the temperature fluctuation value and the density fluctuation value at the current time step, as the synchronous coupling characteristic of temperature fluctuations and density fluctuations.

[0083] Covariance is an indicator that measures the degree of coordinated change between two variables. For each time step, the temperature fluctuation and density fluctuation of the current time step are calculated as two variables.

[0084] First, calculate the mean of the temperature fluctuation and density fluctuation values ​​at the current time step. Then, calculate the difference between each value and its respective mean. Next, multiply these differences and sum them over all time step cells to obtain the covariance value. The magnitude and sign of the covariance reflect the relationship between temperature fluctuations and density fluctuations at the same time point. If the covariance is positive, it indicates that temperature fluctuations and density fluctuations tend to change in the same direction at the same time point; if the covariance is negative, it indicates that they tend to change in opposite directions; the larger the absolute value of the covariance, the stronger the degree of their synchronous change.

[0085] By calculating the covariance of each time step unit, a set of data reflecting the synchronous coupling characteristics of temperature fluctuations and density fluctuations is obtained. This data can reflect the interaction between temperature and density at the same time point and can be used to analyze the synchronous change characteristics of turbulence.

[0086] Step 1324: Calculate the mutual information entropy between the density fluctuation value at the current time step and the temperature fluctuation value at the previous time step, as a characteristic of the hysteresis response of density fluctuation to temperature fluctuation.

[0087] To analyze the hysteretic response of density fluctuations to temperature fluctuations, it is necessary to calculate the mutual information entropy between the density fluctuation value at the current time step and the temperature fluctuation value at the previous time step, as a characteristic of the hysteretic response of density fluctuations to temperature fluctuations.

[0088] Mutual information entropy is an indicator used to measure the degree of information sharing between two variables. For each time step, the density fluctuation value of the current time step and the temperature fluctuation value of the previous time step are used as two variables for calculation.

[0089] First, determine the probability distributions of the density fluctuation value at the current time step and the temperature fluctuation value at the previous time step. The probability distribution can be estimated by the frequency of different values ​​in the statistical data. Then, calculate the joint probability distribution of the two variables.

[0090] Next, based on the definition of mutual information entropy, the mutual information entropy is calculated using the joint probability distribution and the individual probability distributions. Mutual information entropy reflects the degree of information correlation between the density fluctuation value at the current time step and the temperature fluctuation value at the previous time step.

[0091] If the mutual information entropy is large, it indicates that the density fluctuation in the current time step may be affected by the temperature fluctuation in the previous time step, forming a hysteresis response characteristic of density fluctuation to temperature fluctuation. By calculating the mutual information entropy of each time step unit, a set of data reflecting the hysteresis response characteristics is obtained, which can reflect the time-lag correlation between density and temperature.

[0092] Step 1325: After normalizing the precursor signal features, synchronization coupling features and hysteresis response features, the time-series correlation features are generated.

[0093] After calculating the characteristics of the precursor signal, the synchronous coupling, and the hysteresis response, since the range and magnitude of these characteristics may differ, they need to be normalized to unify them onto a comparable scale.

[0094] The purpose of normalization is to map the value of each feature to a specified range, typically the interval [0, 1]. First, the value ranges for the precursor signal feature, synchronization coupling feature, and hysteresis response feature are determined. For the precursor signal feature, its value may fall within different numerical ranges; the synchronization coupling feature and the hysteresis response feature also have their own value ranges.

[0095] Then, a normalization method is applied to each feature. Common normalization methods, such as min-max normalization, can be used. For each feature, find its maximum and minimum values, subtract the minimum value from each feature value, and then divide by the difference between the maximum and minimum values ​​to obtain the normalized feature value.

[0096] By normalizing the precursor signal characteristics, synchronization coupling characteristics, and hysteresis response characteristics, they can be compared and analyzed at the same scale. The normalized precursor signal characteristics, synchronization coupling characteristics, and hysteresis response characteristics are then combined to generate a time-series correlation feature. This time-series correlation feature contains causal correlation information between temperature fluctuations and density fluctuations over time and exhibits consistency across scales.

[0097] Step 133: Using the feature interaction module of the turbulence feature association model, perform cross-correlation analysis on the multi-scale density magnetic field association features and the time-series association features to identify the coupling mode of particle motion and turbulence evolution, and generate a set of association features of the particle motion mode association features and the turbulence evolution law association features.

[0098] After obtaining the multi-scale density magnetic field correlation features and temporal correlation features, in order to gain a deeper understanding of the coupling mode between particle motion and turbulent evolution, it is necessary to use the feature interaction module of the turbulence feature correlation model to perform cross-correlation analysis on these two features.

[0099] The feature interaction module first inputs multi-scale density-magnetic field correlation features and temporal correlation features. These two features reflect the relationships between particles and magnetic fields, as well as between temperature and density, from different perspectives. The multi-scale density-magnetic field correlation features contain feature information of particle density-magnetic field coupling data at different time scales; the temporal correlation features reflect the causal relationship between temperature fluctuations and density fluctuations over time.

[0100] During the cross-correlation analysis, the module analyzes the mutual influence between multi-scale density magnetic field correlation features and temporal correlation features. For example, it investigates whether there is a correlation between high-frequency fluctuation features in multi-scale density magnetic field correlation features and precursor signal features in temporal correlation features. By calculating the correlation between different feature dimensions, it identifies the potential connections between them.

[0101] Based on the results of cross-correlation analysis, the coupling modes between particle motion and turbulent evolution are identified. For example, when the low-frequency trend characteristics in the multi-scale density magnetic field correlation features and the synchronous coupling characteristics in the time-series correlation features show a certain specific correlation, it may mean that there is a coupling relationship between the long-term motion trend of particles and the synchronous changes in temperature and density.

[0102] Based on the identified coupling patterns, feature subsets related to particle motion patterns and turbulence evolution patterns are extracted. The feature subset related to particle motion patterns includes information on features such as particle spinning motion, random walk, and escape from constraint boundaries; the feature subset related to turbulence evolution patterns includes information on features such as the growth of turbulent coherent structures, energy cascade transfer, and dissipation attenuation.

[0103] Finally, the correlation features of particle motion patterns and the correlation features of turbulence evolution are combined to form a correlation feature set, which integrates information from multi-scale density magnetic field correlation features and temporal correlation features.

[0104] In another preferred embodiment, the feature interaction module of the turbulence feature correlation model performs cross-correlation analysis on the multi-scale density magnetic field correlation features and the temporal correlation features to identify the coupling mode between particle motion and turbulence evolution, and generates a correlation feature set of the particle motion mode correlation features and the turbulence evolution law correlation features, including: Step 1331: Input the multi-scale density magnetic field correlation features and the temporal correlation features into the attention mechanism layer of the feature interaction module, and calculate the contribution weight of each feature dimension to the particle motion pattern through self-attention to generate a weighted feature set guided by particle motion.

[0105] In the feature interaction module, the attention mechanism layer is used to process multi-scale density magnetic field correlation features and temporal correlation features to determine the contribution weight of each feature dimension to the particle motion pattern. The self-attention mechanism allows the model to automatically focus on the parts of the features that are more important to the particle motion pattern.

[0106] First, the multi-scale density magnetic field correlation features and temporal correlation features are input into the attention mechanism layer. The attention mechanism layer performs a linear transformation on the input features, mapping them to different vector spaces. Through this transformation, the query vector, key vector, and value vector are obtained.

[0107] Then, the similarity between the query vector and the key vector is calculated. Similarity can be calculated using methods such as dot product. Based on the similarity, a weight is assigned to each feature dimension. The higher the similarity, the larger the weight, indicating a greater contribution of that feature dimension to the particle motion pattern.

[0108] The weights are then applied to the value vectors, and the value vectors are summed in a weighted manner. Thus, each feature dimension is weighted according to its contribution to the particle motion pattern.

[0109] Through self-attention computation, a weighted feature set guided by particle motion is generated. This weighted feature set highlights the feature dimensions that are more important to particle motion patterns and can be used to extract features associated with particle motion patterns.

[0110] Step 1332: Input the multi-scale density magnetic field correlation features and the time-series correlation features into the cross-correlation analysis layer of the feature interaction module, and calculate the cross-correlation matrix between different feature dimensions.

[0111] After inputting the multi-scale density magnetic field correlation features and temporal correlation features into the cross-correlation analysis layer of the feature interaction module, this layer calculates the cross-correlation matrix between different feature dimensions. The cross-correlation matrix reflects the degree of correlation between different feature dimensions.

[0112] The cross-correlation analysis layer first organizes the input multi-scale density magnetic field correlation features and temporal correlation features. It then determines the feature dimension of each feature, with each feature dimension representing a different feature attribute.

[0113] Then, for any two feature dimensions, calculate the cross-correlation coefficient between them. The cross-correlation coefficient can be obtained by calculating the covariance between the data sequences of the two feature dimensions and dividing it by the product of their standard deviations.

[0114] Combine all feature dimensions pairwise and calculate their cross-correlation coefficients to form a matrix. Each element in the matrix represents the cross-correlation coefficient between the corresponding two feature dimensions. The cross-correlation coefficient ranges from -1 to 1. The closer the value is to 1, the stronger the positive correlation between the two feature dimensions; the closer the value is to -1, the stronger the negative correlation; and the closer the value is to 0, the weaker the correlation.

[0115] By calculating the cross-correlation matrix between different feature dimensions, we can gain a comprehensive understanding of the interrelationships between various feature dimensions in the multi-scale density magnetic field correlation features and temporal correlation features. This can be used to extract particle motion mode correlation features and turbulence evolution law correlation features.

[0116] Step 1333: Based on the weighted feature set guided by particle motion and the cross-correlation matrix, extract the feature subsets related to particle spinning motion, random walk and constraint boundary escape as particle motion mode association features.

[0117] After obtaining the weighted feature set and cross-correlation matrix of particle motion, it is necessary to extract a subset of features related to particle spinning motion, random walk and constraint boundary escape based on this information, as the particle motion pattern association features.

[0118] First, based on the weighted feature set guided by particle motion, we identify the feature dimensions that contribute significantly to the particle motion pattern. These feature dimensions are given higher weights during the self-attention calculation process, indicating that they are more important in describing particle motion.

[0119] Then, by combining the cross-correlation matrix, the correlation between these important feature dimensions and particle cyclotron motion, random walk, and constraint boundary escape is analyzed. For particle cyclotron motion, the cross-correlation matrix is ​​used to find the high correlation between feature dimensions related to cyclotron motion. For example, some feature dimensions may be related to parameters such as the cyclotron radius and cyclotron frequency of the particle in the magnetic field. These related feature dimensions are extracted to form a feature subset related to particle cyclotron motion.

[0120] For random walks, we analyze the correlation between feature dimensions and the characteristic behavior of random walks. Random walks are usually characterized by the random movement of particles. We identify feature dimensions related to the randomness and diffusion of particle movement, and form a feature subset related to random walks.

[0121] For boundary escape, the feature dimensions related to the particle breaking through the boundary are determined based on the cross-correlation matrix. For example, some feature dimensions may be related to parameters such as the particle's energy and velocity. When these parameters reach certain conditions, the particle may escape the boundary. These related feature dimensions are extracted to form a feature subset related to boundary escape.

[0122] These feature subsets related to particle spinning motion, random walk, and constraint boundary escape are combined to form particle motion pattern association features. This feature set contains key feature information describing different particle motion patterns and can be used to identify particle motion patterns.

[0123] Step 1334: Based on the feature combinations that satisfy the correlation conditions in the cross-correlation matrix, extract the feature subsets that are related to the growth of turbulent coherent structures, energy cascade transfer and dissipation attenuation as the correlation features of turbulent evolution law.

[0124] After obtaining the cross-correlation matrix, in order to extract the feature subsets related to the evolution of turbulence, it is necessary to filter the feature combinations in the matrix that satisfy the correlation conditions.

[0125] First, identify the characteristic dimensions related to the growth of turbulent coherent structures, energy cascade transfer, and dissipation decay. For example, the growth of turbulent coherent structures may be related to features such as particle density aggregation and vortex formation; energy cascade transfer may be related to features such as energy transfer at different scales and transfer efficiency; and dissipation decay may be related to features such as energy loss and weakening of turbulence intensity.

[0126] Then, in the cross-correlation matrix, identify feature combinations that satisfy certain correlation conditions among these related feature dimensions. These correlation conditions can be set based on the needs and experience of the actual research. For example, a correlation threshold can be set; when the absolute value of the cross-correlation coefficient between two feature dimensions is greater than this threshold, they are considered to have a strong correlation.

[0127] Feature dimensions corresponding to feature combinations that satisfy the correlation condition are extracted to form feature subsets related to turbulent coherent structure growth, energy cascade transfer, and dissipation decay. For turbulent coherent structure growth, feature dimensions related to this process are combined to form corresponding feature subsets; the same method is used to extract corresponding feature subsets for energy cascade transfer and dissipation decay.

[0128] These feature subsets related to the growth of coherent structures in turbulence, energy cascade transfer, and dissipation decay are combined to form turbulence evolution law correlation features. This feature set contains key feature information describing turbulence at different evolution stages and can be used to identify turbulence evolution laws.

[0129] Step 1335: Combine the particle motion pattern correlation features with the turbulence evolution law correlation features to generate the correlation feature set.

[0130] After extracting the particle motion pattern correlation features and the turbulence evolution law correlation features respectively, they need to be combined in order to integrate these two parts of feature information.

[0131] First, determine the feature dimensions of the particle motion pattern correlation features and the turbulence evolution law correlation features. Each feature has a corresponding dimension, representing different feature attributes. Then, arrange and combine the particle motion pattern correlation features and the turbulence evolution law correlation features in a certain order. The particle motion pattern correlation features can be placed first, followed by the turbulence evolution law correlation features; this arrangement comprehensively demonstrates the hierarchical relationship between different types of features.

[0132] Through combined processing, a set of associated features is generated, which contains relevant feature information on particle motion patterns and turbulence evolution laws. Based on the set of associated features, the particle motion pattern recognition results and turbulence evolution law recognition results of the tokamak device can be generated.

[0133] Step 140: Generate particle motion pattern recognition results and turbulence evolution law recognition results of the tokamak device based on the associated feature set. The particle motion pattern recognition results include particle cyclonic motion characteristics, random walk characteristics, and constraint boundary escape characteristics. The turbulence evolution law recognition results include turbulent coherent structure growth characteristics, energy cascade transfer characteristics, and dissipation attenuation characteristics.

[0134] After obtaining the set of associated features, the particle motion pattern recognition results and turbulence evolution law recognition results of the tokamak device are generated based on the particle motion pattern association features and turbulence evolution law association features in the set.

[0135] For particle motion pattern recognition results, different particle motion patterns are identified by analyzing the particle motion pattern association features in the associated feature set. These particle motion pattern association features include feature information related to particle spinning motion, random walk, and constraint boundary escape. Further processing and judgment of this feature information determines the particle's motion pattern within the tokamak device. For example, based on the feature subset related to particle spinning motion, it is determined whether the particle exhibits spinning motion and its characteristic behavior; based on the feature subset related to random walk, the random walk characteristics of the particle are identified; based on the feature subset related to constraint boundary escape, it is determined whether the particle has experienced constraint boundary escape and its escape characteristics. These recognition results are combined to form the particle motion pattern recognition result, which includes particle spinning motion features, random walk features, and constraint boundary escape features.

[0136] For the identification results of turbulence evolution laws, the correlation features of turbulence evolution laws in the associated feature set are analyzed. These correlation features include characteristic information related to the growth of turbulent coherent structures, energy cascade transfer, and dissipation decay. By analyzing this characteristic information, the evolution laws of turbulence at different stages are identified. For example, based on the feature subset related to the growth of turbulent coherent structures, the growth characteristics of turbulent coherent structures are determined; based on the feature subset related to energy cascade transfer, the energy transfer law in turbulence is analyzed; and based on the feature subset related to dissipation decay, the dissipation decay characteristics of turbulence are understood. These identification results are combined to form the turbulence evolution law identification results, which include the characteristics of turbulent coherent structure growth, energy cascade transfer, and dissipation decay.

[0137] In one implementation, generating particle motion pattern recognition results and turbulence evolution law recognition results for the tokamak device based on the associated feature set includes: Step 141: Perform classification decision processing on the particle motion pattern association features in the association feature set to generate particle motion pattern recognition results containing the confidence of each motion pattern.

[0138] To generate particle motion pattern recognition results that include the confidence scores of each motion pattern, it is necessary to perform classification decision processing on the particle motion pattern association features in the associated feature set. This classification decision processing can determine the probability of a particle under different motion patterns.

[0139] First, the associated features of particle motion patterns are organized and analyzed. A subset of features related to particle spinning motion, random walk, and constraint boundary escape is identified. Then, for each motion pattern, an appropriate classification decision method is adopted. For particle spinning motion, the similarity between the feature values ​​of the relevant feature subset and a preset spinning motion feature template is calculated. The higher the similarity, the greater the probability that the particle is undergoing spinning motion; this similarity is then converted into a confidence level for spinning motion.

[0140] For random walks, we can analyze the probability distribution of eigenvalues ​​in relevant feature subsets. We then compare this distribution with a Gaussian distribution and calculate the degree of difference between them. The smaller the difference, the greater the probability that the particle will perform a random walk; this degree of difference is then converted into a confidence level for the random walk.

[0141] For constraint boundary escape, the eigenvalues ​​of the relevant feature subset can be compared with the critical eigenvalue for boundary escape. The ratio of the eigenvalue to the critical eigenvalue is calculated, and the probability of the particle escaping the constraint boundary is determined based on the magnitude of the ratio, which is then converted into a confidence level for constraint boundary escape.

[0142] Finally, the calculated confidence scores for cyclotron motion, random walk, and constraint boundary escape are normalized to ensure their values ​​fall within the range of [0, 1]. These normalized confidence scores are then combined to generate particle motion pattern recognition results that include the confidence scores for each motion pattern.

[0143] As one implementation, the step of performing classification decision processing on the particle motion pattern association features in the associated feature set to generate particle motion pattern recognition results containing the confidence levels of each motion pattern includes: Step 1411: Extract the first feature subset related to particle cyclotron motion from the particle motion pattern association features, calculate the Euclidean distance between the feature value of the first feature subset and the preset cyclotron motion feature template, and convert the Euclidean distance into cyclotron motion confidence.

[0144] When performing classification decision processing on the features associated with particle motion patterns, the first feature subset related to particle spinning motion is extracted. This feature subset includes various feature dimensions related to particle spinning motion, such as the particle's spinning radius and spinning frequency.

[0145] Then, the Euclidean distance between the eigenvalues ​​of the first feature subset and the preset cyclotron motion feature template is calculated. The preset cyclotron motion feature template is a pre-defined set of standard features that represents typical particle cyclotron motion characteristics. Euclidean distance is a method to measure the distance between two vectors. By calculating the Euclidean distance between the eigenvalue vector of the first feature subset and the vector of the preset cyclotron motion feature template, the degree of difference between them can be obtained.

[0146] The smaller the Euclidean distance, the more similar the feature values ​​of the first feature subset are to the preset cyclotron motion feature template, and the greater the probability that the particle will undergo cyclotron motion. The Euclidean distance can be converted into cyclotron motion confidence using a conversion function. For example, an inverse proportional function can be used, taking the Euclidean distance as input and obtaining a value between [0, 1] as the cyclotron motion confidence. The smaller the Euclidean distance, the higher the converted confidence, reflecting a greater probability that the particle will undergo cyclotron motion. In this way, the confidence of the particle undergoing cyclotron motion is obtained.

[0147] Step 1412: Extract the second feature subset related to random walk from the particle motion pattern associated features, calculate the probability distribution of the feature values ​​of the second feature subset and the KL divergence of the Gaussian distribution, and convert the KL divergence into random walk confidence.

[0148] When processing the features associated with particle motion patterns, a second feature subset related to random walks is then extracted. This feature subset includes feature dimensions related to particle random walks, such as the randomness and diffusion of particle trajectories.

[0149] Calculate the KL divergence between the probability distribution of the eigenvalues ​​of the second eigenset and the Gaussian distribution. KL divergence is a measure of the degree of difference between two probability distributions. The Gaussian distribution is a common probability distribution, often used to describe random phenomena. By calculating the KL divergence between the probability distribution of the eigenvalues ​​of the second eigenset and the Gaussian distribution, we can determine the degree of difference between them.

[0150] The smaller the KL divergence, the closer the probability distribution of the eigenvalues ​​in the second feature subset is to a Gaussian distribution, and the greater the probability that the particle will perform a random walk. The KL divergence can be converted into a random walk confidence score using a transformation function. For example, an inverse proportional function can be used, taking the KL divergence as input and obtaining a value between [0, 1] as the random walk confidence score. The smaller the KL divergence, the higher the converted confidence score, reflecting a greater probability that the particle will perform a random walk. In this way, the confidence score of the particle performing a random walk is obtained, further refining the particle motion pattern recognition results.

[0151] Step 1413: Extract the third feature subset related to constraint boundary escape from the particle motion pattern associated features, calculate the ratio of the feature value of the third feature subset to the critical feature value of boundary escape, and convert the ratio into constraint boundary escape confidence.

[0152] When processing the features associated with particle motion patterns, a third feature subset related to escape from the constraint boundary is extracted. This feature subset includes feature dimensions related to particle escape from the constraint boundary, such as particle energy and velocity. These features are closely related to whether the particle can break through the constraint boundary.

[0153] Calculate the ratio of the eigenvalues ​​of the third feature subset to the critical eigenvalue for boundary escape. The critical eigenvalue for boundary escape is a pre-defined standard value representing the critical condition for a particle to escape from the bounded boundary. By calculating the ratio of the eigenvalues ​​to the critical eigenvalue, we can understand the relationship between the particle's current state and the critical state for boundary escape.

[0154] If the ratio is greater than 1, it indicates that the particle's eigenvalues ​​exceed the boundary escape critical eigenvalues, and the particle is more likely to escape the constraint boundary. If the ratio is less than 1, it indicates that the particle's eigenvalues ​​have not reached the boundary escape critical eigenvalues, and the particle is less likely to escape the constraint boundary. This ratio can be converted into a constraint boundary escape confidence level using a mapping function. For example, the ratio can be directly mapped to the interval [0, 1]; the larger the ratio, the higher the constraint boundary escape confidence level. In this way, the confidence level of particle constraint boundary escape is obtained, providing important information about constraint boundary escape for particle motion pattern recognition results.

[0155] Step 1414: Normalize the confidence scores of the cyclotron motion, random walk, and constraint boundary escape to generate the particle motion pattern recognition result.

[0156] After calculating the confidence levels of the cyclotron motion, random walk, and constraint boundary escape, they need to be normalized to ensure comparability and consistency.

[0157] First, determine the range of values ​​for the confidence scores of the rotational motion, random walk, and constraint boundary escape. Typically, the original values ​​for these confidence scores may fall within different ranges.

[0158] Then, normalization methods are used to process them. Common normalization methods, such as min-max normalization, can be used. Find the maximum and minimum values ​​among the three confidence scores, subtract the minimum value from each confidence score, and then divide by the difference between the maximum and minimum values ​​to obtain the normalized confidence scores.

[0159] The normalized confidence scores range from [0, 1], allowing for a comprehensive comparison of the likelihood of particles in different motion modes. Combining the normalized confidence scores for cyclotron motion, random walk, and constraint boundary escape generates a particle motion pattern recognition result that includes the confidence scores for each motion mode. This result accurately reflects the probability of particles undergoing different motion modes within the tokamak device.

[0160] Step 142: Perform evolution stage division processing on the turbulence evolution law association features in the association feature set to generate turbulence evolution law identification results containing the duration of each evolution stage.

[0161] To generate turbulence evolution law identification results that include the duration of each evolution stage, it is necessary to perform evolution stage segmentation processing on the turbulence evolution law correlation features in the correlation feature set. Evolution stage segmentation processing can determine the development process and duration of turbulence in different stages.

[0162] First, the correlation characteristics of turbulent evolution are analyzed. A subset of features related to the growth of turbulent coherent structures, energy cascade transfer, and dissipation decay are identified. Then, appropriate analytical methods are applied for each evolution stage. For the growth stage of turbulent coherent structures, time series analysis of the relevant feature subsets is used to identify the time interval from the baseline value to the peak value. This time interval represents the growth stage of the turbulent coherent structures. The start and end times of this stage are recorded, and the duration is calculated.

[0163] For the energy cascade transfer stage, energy flow analysis is performed on the relevant feature subset. The time interval from the peak value to the stable value of the characteristic energy transfer efficiency is identified; this time interval is the energy cascade transfer stage. The start and end times of this stage are recorded, and the duration is calculated.

[0164] For the dissipation decay stage, amplitude decay analysis is performed on the relevant feature subset. The time interval from the peak value to the reference value is identified, which is the dissipation decay stage. The start and end times of this stage are recorded, and the duration is calculated.

[0165] Finally, the durations of each evolution stage are combined to generate a turbulence evolution law identification result that includes the duration of each evolution stage. This result can comprehensively show the development process and duration of turbulence at different stages.

[0166] As another implementation, the step of dividing the turbulence evolution law associated features in the associated feature set into evolutionary stages to generate turbulence evolution law identification results containing the duration of each evolutionary stage includes: Step 1421: Perform time series analysis on the growth characteristics of turbulent coherent structures in the correlation features of the turbulent evolution law, identify the first time interval from the start of the feature amplitude to the peak value as the growth stage, and record the first start time and the first end time of the first time interval.

[0167] To determine the growth stages of turbulent coherent structures, time-series analysis is needed on the growth characteristics of turbulent coherent structures within the correlated features of turbulent evolution. Time-series analysis can reveal the temporal variation patterns of these characteristics.

[0168] First, we need to define the baseline and peak values ​​of the growth characteristics of turbulent coherent structures. The baseline value is the value of the characteristic in its initial or relatively stable state, while the peak value is the maximum value reached by the characteristic during the growth process.

[0169] Then, the time series data of the growth characteristics of the turbulent coherent structure are traversed. Starting from the beginning of the time series, the time point when the characteristic amplitude begins to rise from the baseline value is found, and this time point is recorded as the first start time. The data is traversed again until the time point when the characteristic amplitude reaches its peak is found, and this time point is recorded as the first end time.

[0170] The time interval between the first start time and the first end time constitutes the growth stage of the turbulent coherent structure. This stage reflects the process from the initial formation of the turbulent coherent structure to its maximum scale. Recording the first start time and the first end time of this first time interval provides important temporal information about the growth stage of the turbulent coherent structure for subsequent generation of turbulence evolution law identification results.

[0171] Step 1422: Perform energy flow analysis on the energy cascade transfer characteristics in the correlation characteristics of the turbulent evolution law, identify the second time interval from the peak value to the stable value of the characteristic energy transfer efficiency as the transfer stage, and record the second start time and the second end time of the second time interval.

[0172] After determining the growth stages of the coherent turbulent structure, energy flow analysis is needed to identify the energy cascade transfer stages within the correlation characteristics of turbulent evolution. Energy flow analysis can reveal the energy transfer process in turbulence.

[0173] First, we need to define the peak and stable values ​​of energy transfer efficiency in energy cascade transfer characteristics. The peak value is the maximum energy transfer efficiency reached at a certain moment, while the stable value is the value at which the energy transfer efficiency tends to stabilize after a period of time.

[0174] Then, the time series data of energy cascade transfer characteristics are analyzed. The time point when the energy transfer efficiency reaches its peak is found from the time series and recorded as the second start time. Next, the data is observed again, and when the energy transfer efficiency decreases and tends to a stable value, this time point is recorded as the second end time.

[0175] The time interval between the second start time and the second end time constitutes the energy cascade transfer stage, which reflects the process of energy transfer from higher to lower energy levels in turbulence. Recording the second start time and the second end time of this second time interval provides crucial temporal information about the energy cascade transfer stage for generating identification results of turbulence evolution patterns.

[0176] Step 1423: Perform amplitude decay analysis on the dissipation decay feature in the correlation feature of the turbulence evolution law, identify the third time interval from the peak value to the reference value as the decay stage, and record the third start time and the third end time of the third time interval.

[0177] After determining the energy cascade transfer stage, in order to determine the dissipation decay stage, it is necessary to perform amplitude decay analysis on the dissipation decay characteristics in the correlation features of turbulence evolution. Amplitude decay analysis can reveal the decay law of the characteristics over time.

[0178] First, define the baseline and peak values ​​of the dissipation decay characteristic. The baseline value is the value of the characteristic in its initial or relatively stable state, while the peak value is the maximum value reached by the characteristic at a certain moment.

[0179] Then, the time series data of the dissipation decay characteristic is traversed. The time point when the characteristic amplitude reaches its peak is found from the time series and recorded as the third start time. Next, the data is observed again, and when the characteristic amplitude drops to the baseline value, this time point is recorded as the third end time.

[0180] The time interval between the third start time and the third end time is the dissipation decay stage, which reflects the gradual dissipation of energy and the gradual weakening of turbulence intensity. Recording the third start time and the third end time of this third time interval provides important temporal information about the dissipation decay stage for generating turbulence evolution law identification results.

[0181] Step 1424: Based on the time intervals of the growth stage, the transmission stage, and the decay stage, as well as the duration of each stage, generate the turbulence evolution law identification result.

[0182] After determining the time intervals and durations of the growth, propagation, and decay stages of turbulence, this information is integrated to generate turbulence evolution law identification results.

[0183] First, calculate the duration of each stage. For the growth stage, subtract the start time from the end time of the growth stage to obtain the duration of the growth stage; for the transmission stage, subtract the start time from the end time of the transmission stage to obtain the duration of the transmission stage; for the decay stage, subtract the start time from the end time of the decay stage to obtain the duration of the decay stage.

[0184] Next, the time interval information for the growth phase, transmission phase, and decay phase, as well as the calculated duration of each phase, are organized. They can be listed in chronological order: first, the start time, end time, and duration of the growth phase; then the corresponding information for the transmission phase; and finally, the information for the decay phase.

[0185] During the data processing, it is essential to ensure the accuracy and consistency of information at each stage. Time intervals must be clearly and comprehensively defined, and durations must be calculated precisely.

[0186] Combining this information generates a turbulence evolution identification result that includes the duration of each evolution stage. This result visually demonstrates the entire evolution process of turbulence from growth to transmission to decay, as well as the duration of each stage. For example, this result shows that the turbulence growth stage is relatively short, while the energy cascade transmission stage may last longer, and the duration of the dissipation decay stage varies depending on the specific circumstances. This information can be used to conduct in-depth research on the evolution of turbulence within a tokamak device and guide the implementation of corresponding control measures.

[0187] In an optional embodiment, the step of generating particle motion pattern recognition results and turbulence evolution law recognition results of the tokamak device based on the associated feature set further includes: Step 1431: Perform uncertainty assessment processing on the set of associated features, calculate the feature value fluctuation range of the particle motion mode associated features and the turbulence evolution law associated features using the Monte Carlo sampling method, and generate uncertainty assessment results containing feature confidence intervals.

[0188] To assess the uncertainties in the correlation features of particle motion patterns and turbulence evolution laws within the correlation feature set, a Monte Carlo sampling method was employed. Monte Carlo sampling is a statistical method based on random sampling that can simulate data uncertainty using a large number of random samples.

[0189] First, a feature space is determined for the correlation between particle motion patterns and turbulence evolution. This feature space encompasses the value range and dimensions of all relevant features. Then, random sampling is performed within this feature space. Each sampling yields a set of feature values, representing a possible state of particle motion patterns and turbulence evolution. Through multiple sampling operations, a large number of feature value samples are obtained.

[0190] For each feature dimension, analyze the distribution of feature values ​​for these samples. Calculate the mean and standard deviation of the feature values. Based on statistical principles, determine the range of fluctuation for the feature values ​​using the mean as the center and the standard deviation as the reference. For example, the range of fluctuation can be determined by adding or subtracting a certain multiple of the standard deviation from the mean; this range of fluctuation is the confidence interval for the feature.

[0191] By combining the confidence intervals of all feature dimensions, an uncertainty assessment result containing feature confidence intervals is generated. This result can reflect the possible range of changes in the particle motion mode correlation characteristics and the turbulence evolution law correlation characteristics under different conditions.

[0192] Step 1432: The particle motion pattern recognition result, the turbulence evolution law recognition result, and the uncertainty assessment result are fused together to generate a target recognition result containing a feature confidence interval. The target recognition result is used for plasma confinement performance analysis and turbulence control strategy optimization of the tokamak device.

[0193] After obtaining the particle motion pattern recognition results, turbulence evolution law recognition results, and uncertainty assessment results, they need to be fused to generate target recognition results containing feature confidence intervals.

[0194] First, let's clarify the key information in these three results. The particle motion pattern recognition result includes the confidence level of particles under different motion patterns; the turbulence evolution law recognition result includes the time interval and duration of each evolution stage of turbulence; and the uncertainty assessment result includes the feature confidence intervals of the correlation features of particle motion patterns and the correlation features of turbulence evolution laws.

[0195] Then, this information is integrated. The particle motion pattern recognition results and turbulence evolution law recognition results can be used as the main recognition information, and the feature confidence intervals in the uncertainty assessment results can be correlated with this recognition information. For example, for each motion pattern confidence level in the particle motion pattern recognition results, its corresponding feature confidence interval can be used to indicate the reliability of that confidence level. Similarly, for the time information of each evolution stage in the turbulence evolution law recognition results, corresponding feature confidence intervals can also be correlated to reflect the uncertainty of this time information.

[0196] This fusion process generates target identification results containing feature confidence intervals. These results not only include identification information on particle motion patterns and turbulence evolution, but also provide the uncertainty range of this information. For plasma confinement performance analysis of tokamak devices, these results can help determine the extent to which particle motion patterns and turbulence evolution affect plasma confinement, and the reliability of this influence. In terms of turbulence control strategy optimization, based on the uncertainty information in the target identification results, more flexible and reliable control strategies can be formulated to address particle motion and turbulence evolution under different conditions.

[0197] As a non-limiting embodiment, the method further includes: based on the particle motion pattern recognition results and turbulence evolution law recognition results, performing key feature screening processing on the associated feature set to identify a subset of key associated features whose contribution to particle motion pattern recognition and turbulence evolution law recognition exceeds a preset threshold; feeding the subset of key associated features back to the integrated dataset, performing feature enhancement processing on the particle density-magnetic field coupling data segment and temperature fluctuation-density fluctuation associated data block in the integrated dataset corresponding to the subset of key associated features to generate an enhanced associated feature set; and regenerating the optimized particle motion pattern recognition results and turbulence evolution law recognition results of the tokamak device based on the enhanced associated feature set.

[0198] Based on the particle motion pattern recognition results and turbulence evolution law recognition results, key feature screening is performed on the associated feature set. First, the contribution of each feature to particle motion pattern recognition and turbulence evolution law recognition is determined. This contribution can be determined by analyzing the correlation between the feature and the recognition results. For example, for particle motion pattern recognition, if a feature has a high correlation with the confidence level of particle cyclonic motion, it indicates that the feature contributes significantly to the recognition of particle cyclonic motion patterns.

[0199] A preset threshold is set, and features whose contribution exceeds the threshold are filtered out to form a key correlation feature subset. This subset contains key features that have a significant impact on the recognition results.

[0200] The key correlation feature subset is fed back into the integrated dataset. In the integrated dataset, the particle density-magnetic field coupling data segment and the temperature fluctuation-density fluctuation correlation data block corresponding to the key correlation feature subset are found.

[0201] Feature enhancement processing is then performed on these corresponding data segments and blocks. Various methods can be used for feature enhancement, such as smoothing the data to reduce noise interference, or weighting the data to highlight the influence of key features. Through these processes, key features become more apparent and prominent, generating an enhanced set of related features.

[0202] Based on the enhanced correlation feature set, particle motion pattern recognition and turbulence evolution law recognition are performed again. Because the key features in the enhanced correlation feature set are strengthened, the regenerated particle motion pattern recognition results and turbulence evolution law recognition results are more accurate and reliable, thus optimizing the recognition results.

[0203] As a non-limiting embodiment, the method further includes: performing uncertainty assessment processing on the particle motion pattern recognition results and turbulence evolution law recognition results to generate an uncertainty assessment report of the recognition results including the variance of feature contribution; extracting high uncertainty correlation features from the correlation feature set based on the uncertainty assessment report of the recognition results; performing supplementary correlation analysis on the corresponding data segments and data blocks in the integrated dataset for the high uncertainty correlation features to generate supplementary correlation features; fusing the supplementary correlation features with the correlation feature set to generate a corrected correlation feature set; and regenerating the optimized particle motion pattern recognition results and turbulence evolution law recognition results of the tokamak device based on the corrected correlation feature set.

[0204] Uncertainty assessment processing is performed on the particle motion pattern recognition results and turbulence evolution law recognition results. First, the contribution of each feature to the recognition results is analyzed. The contribution can be determined by calculating the correlation coefficient between the feature and the recognition result. Then, the variance of these contributions is calculated. Variance reflects the degree of fluctuation in the contribution; the larger the variance, the more unstable the influence of the feature on the recognition result, and the higher the uncertainty. This information on the variance of feature contributions is compiled into an uncertainty assessment report for the recognition results, which includes the variance of the contribution of each feature.

[0205] Based on the uncertainty assessment report of the identification results, high uncertainty correlation features are extracted from the correlation feature set. These features have a large contribution variance and have a significant impact on the uncertainty of the identification results.

[0206] For high-uncertainty correlation features, corresponding particle density-magnetic field coupling data segments and temperature fluctuation-density fluctuation correlation data blocks were identified in the integrated dataset. Supplementary correlation analysis was performed on these data segments and blocks. This supplementary correlation analysis could include more in-depth feature mining and joint analysis with other relevant features to obtain more information about the high-uncertainty correlation features.

[0207] By supplementing correlation analysis, supplementary correlation features are generated, which contain further understanding and information about highly uncertain correlation features. Merging these supplementary correlation features with the existing correlation feature set allows for a more complete and accurate correlation feature set, resulting in a revised correlation feature set.

[0208] Based on the revised set of associated features, particle motion pattern recognition and turbulence evolution law recognition are performed again. Because the revised set of associated features supplements and corrects highly uncertain associated features, the regenerated particle motion pattern recognition results and turbulence evolution law recognition results are more accurate and reliable, thus optimizing the recognition results.

[0209] As a non-limiting embodiment, the method further includes: decomposing the associated feature set into a particle motion-dominated associated feature subspace and a turbulence evolution-dominated associated feature subspace; generating corresponding subspace particle motion pattern recognition results and subspace turbulence evolution law recognition results based on the particle motion-dominated associated feature subspace and the turbulence evolution-dominated associated feature subspace, respectively; performing consensus analysis on the subspace particle motion pattern recognition results and the original particle motion pattern recognition results, and the subspace turbulence evolution law recognition results and the original turbulence evolution law recognition results, to generate a multi-subspace recognition result consensus report; and performing weighted fusion processing on the multi-subspace recognition result consensus report to generate fused and optimized particle motion pattern recognition results and turbulence evolution law recognition results.

[0210] The associated feature set is decomposed into a particle motion-dominated associated feature subspace and a turbulence evolution-dominated associated feature subspace. Based on the correlation between features and particle motion and turbulence evolution, the features in the associated feature set are classified. Features strongly correlated with particle motion form the particle motion-dominated associated feature subspace, and features strongly correlated with turbulence evolution form the turbulence evolution-dominated associated feature subspace.

[0211] The two subspaces are analyzed separately. For the particle motion-dominated correlation feature subspace, a similar method as before is used for particle motion pattern recognition, generating subspace particle motion pattern recognition results that reflect the particle motion patterns under the particle motion-dominated feature. For the turbulence evolution-dominated correlation feature subspace, turbulence evolution law recognition is also performed, generating subspace turbulence evolution law recognition results that reflect the turbulence evolution law under the turbulence evolution-dominated feature.

[0212] A consensus analysis is performed between the generated subspace identification results and the original identification results. For the subspace particle motion pattern identification results and the original particle motion pattern identification results, the differences in their confidence levels for each motion pattern are compared. The consensus can be measured by calculating similarity indices, such as cosine similarity. Similarly, for the subspace turbulence evolution law identification results and the original turbulence evolution law identification results, the differences in their time intervals and durations at each evolution stage are compared, and similarity indices can also be used to measure the consensus.

[0213] These consensus information are compiled into a multi-subspace identification result consensus report, which comprehensively demonstrates the consistency and differences between the subspace identification results and the original identification results. Based on the multi-subspace identification result consensus report, the identification results of each subspace are weighted and fused. Different weights are assigned to the subspace identification results according to their consensus level. Subspace identification results with high consensus level are given higher weights, and those with low consensus level are given lower weights. Through weighted fusion, the subspace identification results and the original identification results are integrated to generate fused and optimized particle motion pattern identification results and turbulence evolution law identification results. This optimized result integrates information from different subspaces, is more accurate and reliable, and can better reflect the real situation of particle motion and turbulence evolution within the tokamak device.

[0214] This invention collects particle turbulence data during the operation of a tokamak device using multi-source sensors, enabling the acquisition of relevant information from different dimensions and ensuring the comprehensiveness and diversity of the data. The collected particle turbulence data is integrated to obtain an integrated dataset containing time-aligned particle density-magnetic field coupling data segments and spatially calibrated temperature fluctuation-density fluctuation correlation data blocks, improving data quality and usability and making the data more correlated and accurate in both time and space dimensions. Inputting the integrated dataset into a turbulence feature correlation model for multi-dimensional feature correlation analysis reveals the intrinsic relationship between particle motion patterns and turbulence evolution laws, generating a comprehensive and targeted set of correlation features. Based on this set of correlation features, accurate particle motion pattern recognition results and turbulence evolution law recognition results can be generated for the tokamak device, thereby accurately and comprehensively revealing particle behavior and turbulence phenomena within the tokamak device and guiding the operation and performance optimization of the tokamak device.

[0215] Furthermore, Figure 2 This is a schematic diagram of the structure of a particle turbulence data identification technology 200 provided in an embodiment of the present invention. Figure 2 The particle turbulence data identification technology 200 shown includes a processor 210, which can call and run computer programs from memory to implement the methods in the embodiments of the present invention.

[0216] Optionally, such as Figure 2 As shown, the particle turbulence data identification technology 200 may further include a memory 230. The processor 210 can call and run computer programs from the memory 230 to implement the methods in the embodiments of the present invention. The memory 230 may be a separate device independent of the processor 210, or it may be integrated into the processor 210. Optionally, as... Figure 2 As shown, the particle turbulence data identification technology 200 may further include a transceiver 220. The processor 210 can control the transceiver 220 to interact with other devices. Specifically, it can send information or data to other devices or receive information or data sent by other devices. Optionally, the particle turbulence data identification technology 200 can implement the corresponding processes of the storage engine or components (such as processing modules) in the storage engine or devices with deployed storage engines in the various methods of the embodiments of the present invention. For the sake of brevity, these will not be elaborated further here. It should be understood that the processor in the embodiments of the present invention may be an integrated circuit chip with signal processing capabilities. It is understood that the memory in the embodiments of the present invention may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. It should be noted that the memory of the systems and methods described herein is intended to include, but is not limited to, suitable types of memory.

[0217] Based on the above, a readable storage medium is provided, on which a program or instructions are stored, and when the program or instructions are executed by a processor, the steps of the above method are implemented.

[0218] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the embodiments of the present invention are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the embodiments of the present invention without departing from the spirit and scope of protection of the embodiments of the present invention, and all of these forms are within the protection scope of the embodiments of the present invention.

Claims

1. A particle turbulence data identification method based on artificial intelligence, characterized in that, The method includes: During the operation of the tokamak device, particle turbulence data are collected by a multi-source sensor, which includes a magnetic probe array, a laser scatterer, and a Thomson scattering diagnostic system. The particle turbulence data is integrated to obtain an integrated dataset, which includes time-aligned particle density-magnetic field coupling data segments and spatially calibrated temperature fluctuation-density fluctuation correlation data blocks. The integrated dataset is input into the turbulence feature association model, and multi-dimensional feature association analysis is performed to generate a set of associated features of particle motion mode association features and turbulence evolution law association features; The particle motion pattern recognition results and turbulence evolution law recognition results of the tokamak device are generated based on the associated feature set. The particle motion pattern recognition results include particle cyclonic motion characteristics, random walk characteristics, and constraint boundary escape characteristics. The turbulence evolution law recognition results include turbulent coherent structure growth characteristics, energy cascade transfer characteristics, and dissipation attenuation characteristics.

2. The particle turbulence data identification method based on artificial intelligence according to claim 1, characterized in that, The particle turbulence data includes time-series labeled particle density fluctuation signals, magnetic field disturbance signals, and temperature fluctuation signals. The process of integrating the particle turbulence data to obtain an integrated dataset includes: Time offset error corrections are performed on the particle density fluctuation signal, the magnetic field disturbance signal, and the temperature fluctuation signal respectively to generate the original synchronization data sequence; The original synchronous data sequence is spatially calibrated using the geometric coordinate system of the tokamak device. The circumferential position coordinates of the magnetic probe array, the radial measurement position coordinates of the laser scatterer, and the poloidal observation position coordinates of the Thomson scattering diagnostic system are uniformly mapped to the geometric coordinate system to generate a spatially calibrated multi-physical quantity synchronous data set. The multi-physical quantity synchronous data group is segmented into data segments. Based on the triggering characteristics of particle turbulence events, associated data units within a continuous time window are extracted. The associated data units include the temporal overlap interval of particle density fluctuations and magnetic field disturbances, and the spatial overlap region of temperature fluctuations and density fluctuations. The integrated dataset is generated based on the associated data units.

3. The particle turbulence data identification method based on artificial intelligence according to claim 1, characterized in that, The multi-dimensional feature correlation analysis is performed to generate a set of correlation features for particle motion pattern correlation features and turbulence evolution law correlation features, including: The multi-scale feature extraction module of the turbulence feature association model is invoked to perform feature extraction processing on the particle density-magnetic field coupling data segment in the integrated dataset at different time resolutions, generating multi-scale density-magnetic field association features that include high-frequency fluctuation features, mid-frequency coherence features and low-frequency trend features. The time-series dependency modeling module of the turbulence feature association model is used to perform time-series evolution analysis on the temperature fluctuation-density fluctuation association data block, extract the causal relationship between temperature fluctuation and density fluctuation within a continuous time step, and generate time-series association features including precursor signal features, synchronous coupling features and hysteresis response features. Using the feature interaction module of the turbulence feature association model, cross-correlation analysis is performed on the multi-scale density magnetic field association features and the temporal association features to identify the coupling mode between particle motion and turbulence evolution, and to generate a set of association features of the particle motion mode association features and the turbulence evolution law association features.

4. The particle turbulence data identification method based on artificial intelligence according to claim 3, characterized in that, The multi-scale feature extraction module of the turbulence feature association model is invoked to perform feature extraction processing on the particle density-magnetic field coupling data segment in the integrated dataset at different time resolutions, generating multi-scale density-magnetic field association features containing high-frequency fluctuation features, mid-frequency coherence features, and low-frequency trend features, including: The particle density-magnetic field coupled data segment is input into the sliding window processing layer of the multi-scale feature extraction module, and the data segment is windowed and truncated using the first time window, the second time window and the third time window respectively, generating three sets of data sub-segments with different time spans; Local feature extraction processing is performed on each data segment: for the data segment in the first time window, the instantaneous rate of change and phase difference of particle density and magnetic field disturbance are calculated as high-frequency fluctuation features; for the data segment in the second time window, the periodically occurring density-magnetic field synchronous oscillation mode is identified as mid-frequency coherent features; for the data segment in the third time window, the slope of the long-term trend of particle density and magnetic field disturbance is extracted as low-frequency trend features. The high-frequency fluctuation features, the mid-frequency coherence features, and the low-frequency trend features are spliced ​​along the feature dimension to generate multi-scale density magnetic field correlation features with multiple time scale representations.

5. The particle turbulence data identification method based on artificial intelligence according to claim 3, characterized in that, The time-series dependency modeling module of the turbulence feature association model performs time-series evolution analysis on the temperature fluctuation-density fluctuation association data block, extracts the causal relationship between temperature fluctuation and density fluctuation within a continuous time step, and generates time-series association features including precursor signal features, synchronous coupling features, and hysteresis response features, including: The temperature fluctuation-density fluctuation correlation data block is aligned with the time step size, and the temperature fluctuation data and density fluctuation data are divided into continuous time step units with the same time interval. For each time step, the Pearson correlation coefficient between the temperature fluctuation value of the current time step and the density fluctuation value of the previous time step is calculated as a precursor signal feature of the temperature fluctuation on the density fluctuation. Calculate the covariance between the temperature fluctuation value and the density fluctuation value at the current time step, as a synchronous coupling feature between temperature fluctuation and density fluctuation. Calculate the mutual information entropy between the density fluctuation value at the current time step and the temperature fluctuation value at the previous time step, as a characteristic of the hysteresis response of density fluctuation to temperature fluctuation. The time-series correlation features are generated by normalizing the precursor signal features, synchronization coupling features, and hysteresis response features.

6. The particle turbulence data identification method based on artificial intelligence according to claim 3, characterized in that, The feature interaction module of the turbulence feature correlation model performs cross-correlation analysis on the multi-scale density magnetic field correlation features and the temporal correlation features to identify the coupling mode between particle motion and turbulence evolution, and generates a set of correlation features for the particle motion mode correlation features and the turbulence evolution law correlation features, including: The multi-scale density magnetic field correlation features and the temporal correlation features are input into the attention mechanism layer of the feature interaction module. The contribution weight of each feature dimension to the particle motion pattern is calculated through self-attention, and a weighted feature set guided by particle motion is generated. The multi-scale density magnetic field correlation features and the time-series correlation features are input into the cross-correlation analysis layer of the feature interaction module to calculate the cross-correlation matrix between different feature dimensions; Based on the weighted feature set guided by particle motion and the cross-correlation matrix, feature subsets related to particle spinning motion, random walk and constraint boundary escape are extracted as particle motion mode associated features. Based on the feature combinations that satisfy the correlation conditions in the cross-correlation matrix, a subset of features related to the growth of turbulent coherent structures, energy cascade transfer, and dissipation decay is extracted as the correlation features of turbulent evolution law. The particle motion pattern correlation features are combined with the turbulence evolution law correlation features to generate the correlation feature set.

7. The particle turbulence data identification method based on artificial intelligence according to claim 1, characterized in that, The generation of particle motion pattern recognition results and turbulence evolution law recognition results for the tokamak device based on the associated feature set includes: The particle motion pattern association features in the associated feature set are subjected to classification decision processing to generate particle motion pattern recognition results containing the confidence of each motion pattern; The turbulence evolution law associated features in the associated feature set are divided into evolution stages to generate turbulence evolution law identification results that include the duration of each evolution stage.

8. The particle turbulence data identification method based on artificial intelligence according to claim 7, characterized in that, The step of performing classification decision processing on the particle motion pattern association features in the associated feature set to generate particle motion pattern recognition results containing the confidence scores of each motion pattern includes: Extract the first feature subset related to particle spinning motion from the particle motion pattern association features, calculate the Euclidean distance between the feature values ​​of the first feature subset and the preset spinning motion feature template, and convert the Euclidean distance into spinning motion confidence. Extract the second feature subset related to random walk from the particle motion pattern association features, calculate the probability distribution of the feature values ​​of the second feature subset and the KL divergence of the Gaussian distribution, and convert the KL divergence into random walk confidence. Extract the third feature subset related to constraint boundary escape from the particle motion pattern association features, calculate the ratio of the feature value of the third feature subset to the critical feature value of boundary escape, and convert the ratio into constraint boundary escape confidence. The confidence scores of the cyclotron motion, random walk, and constraint boundary escape are normalized to generate the particle motion pattern recognition results. The process of dividing the turbulence evolution law correlation features in the correlation feature set into evolution stages to generate turbulence evolution law identification results containing the duration of each evolution stage includes: Time series analysis was performed on the growth characteristics of turbulent coherent structures in the correlation features of the turbulent evolution law. The first time interval from the start of the characteristic amplitude to the peak value was identified as the growth stage, and the first start time and the first end time of the first time interval were recorded. Energy flow analysis is performed on the energy cascade transfer characteristics in the correlation characteristics of the turbulent evolution law. The second time interval from the peak value to the stable value of the characteristic energy transfer efficiency is identified as the transfer stage, and the second start time and the second end time of the second time interval are recorded. An amplitude decay analysis was performed on the dissipation decay characteristics in the correlation characteristics of the turbulence evolution law. The third time interval from the peak value to the reference value was identified as the decay stage, and the third start time and the third end time of the third time interval were recorded. Based on the time intervals of the growth stage, the transmission stage, and the decay stage, as well as the duration of each stage, the turbulence evolution law identification result is generated.

9. The particle turbulence data identification method based on artificial intelligence according to claim 1, characterized in that, The process of generating particle motion pattern recognition results and turbulence evolution law recognition results for the tokamak device based on the associated feature set also includes: Uncertainty assessment processing is performed on the aforementioned set of associated features. The eigenvalue fluctuation range of the particle motion mode associated features and the turbulence evolution law associated features is calculated using the Monte Carlo sampling method, and uncertainty assessment results containing feature confidence intervals are generated. The particle motion pattern recognition results, the turbulence evolution law recognition results, and the uncertainty assessment results are fused together to generate target recognition results containing feature confidence intervals. These target recognition results are used for plasma confinement performance analysis and turbulence control strategy optimization of the tokamak device.

10. A particle turbulence data identification technology, characterized in that, The method includes at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, causing the at least one processor to perform the method according to any one of claims 1-9.

Images