Magnetic-inertial confinement fusion array signal acquisition and management system based on distributed architecture

Abstract

The invention discloses a magnetic-inertial confinement fusion array signal acquisition and management system based on a distributed architecture, and relates to the technical field of magnetic-inertial confinement fusion, the system comprises a signal acquisition front end, a data fusion and model building center, a real-time online model updating module, a data management module and a user interaction module; according to the invention, by adopting the signal acquisition nodes designed by a distributed architecture, comprehensive and efficient acquisition and management of multi-modal signals in a magnetic-inertial confinement fusion experiment are realized, the precision and the real-time performance of signal acquisition are improved, and on the basis of a traditional physical signal sensor array, the reliability and the reliability of the system are improved. According to the invention, an optical image acquisition device and an acoustic sensor are added, the dimensions of signal acquisition are enriched, data support is provided for subsequent data processing and analysis, and meanwhile, preprocessing and fusion of multi-modal data and construction and optimization of a learning model are realized through a data fusion and model construction center.

Description

Technical Field

[0001] The present invention relates to the technical field of magneto-inertial confinement fusion, and particularly to a magneto-inertial confinement fusion array signal acquisition and management system based on a distributed architecture. Background Art

[0002] With the continuous progress of science and technology, magneto-inertial confinement fusion, as an important field of energy research, has received extensive attention in recent years. Magneto-inertial confinement fusion experiments involve complex physical processes, including magnetic field confinement, particle acceleration and collision, and energy release. These processes are accompanied by the generation of various physical signals, such as magnetic field strength changes, particle flux fluctuations, temperature changes, pressure changes, plasma luminescence images, and acoustic wave characteristics. In order to deeply study and understand the magneto-inertial confinement fusion mechanism, it is particularly important to accurately acquire, fuse, and analyze these multi-modal signals.

[0003] Traditional signal acquisition and management systems have many deficiencies when facing the complex scenarios of magneto-inertial confinement fusion experiments. On the one hand, traditional systems often adopt a centralized architecture with limited acquisition nodes, making it difficult to comprehensively cover the key structures of the experimental device, resulting in limited integrity and accuracy of signal acquisition. On the other hand, when dealing with multi-modal signals, traditional systems lack an effective fusion mechanism, so that the correlation and complementarity between signals cannot be fully utilized, affecting subsequent data analysis and model construction. In addition, traditional systems also have many limitations in model updating and data management, such as the problems of untimely model updating and low data management efficiency, which are difficult to meet the actual needs of experimental research.

[0004] Therefore, developing a magneto-inertial confinement fusion array signal acquisition and management system based on a distributed architecture will help improve the accuracy and efficiency of experimental research and promote the further development of magneto-inertial confinement fusion technology. Summary of the Invention

[0005] The purpose of the present invention is to make up for the deficiencies of the existing technology and provide a magneto-inertial confinement fusion array signal acquisition and management system based on a distributed architecture. The system comprehensively and efficiently acquires physical signal data such as magnetic field strength, particle flux, temperature, and pressure, as well as optical image and acoustic feature data through signal acquisition nodes designed with a distributed architecture, and uses a data fusion and model construction center to perform data fusion and learning model construction to achieve real-time online model updating and data management.

[0006] To solve the above technical problems, the present invention provides the following technical solution: A magnetic-inertial confinement fusion array signal acquisition and management system based on a distributed architecture, the system comprising: a signal acquisition front end, a data fusion and model construction center, a real-time online model update module, a data management module, and a user interaction module;

[0007] The signal acquisition front end: Adopting a distributed architecture, the acquisition nodes are distributed on the key structures of the device to ensure that various physical signals can be comprehensively captured. Each acquisition node measures the physical signal data of magnetic field strength, particle flux, temperature, and pressure through a sensor array of traditional physical signals, and an optical image acquisition device captures the luminescence image data of the plasma at different stages. An acoustic sensor acquires the acoustic characteristics data of plasma oscillation, the sound waves generated by the operation of the device, and the energy release. At the same time, the acquired original signals are preprocessed and standardized, and feature extraction is performed according to the modal type;

[0008] The data fusion and model construction center: Receives the multi-modal feature data uploaded by each acquisition node through a secure communication network. For the optical image and magnetic field strength data with spatial correlation, fusion is performed according to the optical-magnetic field spatial fusion formula. For the acoustic data and particle flux signals with time correlation, fusion is performed according to the acoustic-flux time fusion formula. Based on the fused multi-modal data feature set, a learning model for magnetic-inertial confinement fusion experiments is constructed through the multi-modal experimental data prediction model formula. This model comprehensively considers the weighted sum of the fusion features of the optical image and magnetic field strength and the fusion features of the acoustic data and particle flux signal in the time series, plus a constant term, aiming to predict the current state through historical data features, and optimizes the parameters of the model by the stochastic gradient descent method, and evaluates the trained model using real-time data;

[0009] The real-time online model update module: When new data is uploaded from the acquisition node, the new data is preprocessed, the data is fused and feature extracted, clustering analysis is performed on the new fused feature data, and representative points are selected. The similarity between the representative points and the historical data points is calculated to determine the local update area. Within the local update area, the correlation weights between the historical data points and the representative points in the area are calculated, and weighted prediction is performed using the existing model and weights. The predicted value is compared with the true value to obtain the prediction error. The gradients of the error with respect to each parameter of the model are calculated respectively, and the model parameters are updated according to the gradients and the learning rate;

[0010] The data management module: Classifies and stores the acquired original multi-modal data according to the time series, acquisition nodes, and modal types, stores the intermediate data and final model parameters during the model training process, and simultaneously establishes an indexing mechanism for fast retrieval and query of data;

[0011] The user interaction module: provides an operation interface for users to facilitate the configuration and management of the system, and displays experimental data and model results in an intuitive chart form.

[0012] Further, the acquisition devices used in the front-end signal acquisition are:

[0013] Sensor arrays for traditional physical signals: magnetic field strength sensors, particle flux sensors, temperature sensors;

[0014] Related to optical image acquisition: high-resolution CCD cameras, high-resolution CMOS cameras;

[0015] Related to acoustic sensors: piezoelectric acoustic sensors.

[0016] Furthermore, in the front-end signal acquisition, feature extraction is performed according to the modal type. For physical signals, the features F B (t) of the physical signals are extracted through the physical signal feature extraction formula. The formula is: where w is the time window length and B(t) is the magnetic field strength; for optical image data, the image features F I (t) are extracted through the optical image feature extraction formula. The formula is: I(x,y) is the image pixel value; for acoustic data, the acoustic features F A (t) are extracted through the acoustic feature extraction formula. The formula is: where A(t) represents the amplitude of the acoustic wave signal, f i is the discrete frequency point obtained by performing a fast Fourier transform FFT on the acoustic signal, and N is the number of points of the FFT.

[0017] Furthermore, in the data fusion and model construction center, for optical images and magnetic field strength data with spatial correlation, fusion is performed according to the optical-magnetic field spatial fusion formula. Let the preprocessed optical image feature matrix be I, where the element I ij represents the feature value of the image at the position (i,j), and the magnetic field strength matrix is B. The element B ij corresponds to the magnetic field strength value at the corresponding position, and both have the same dimension M×N. Normalization is performed on the two, and the formula is: The fusion feature matrix F is calculated, and the formula is: where α and β are weight coefficients, and 0 < α < 1, β ≥ 0, and r is a relatively small integer.

[0018] Furthermore, in the data fusion and model construction center, for the acoustic data and particle flux signals with time correlation, they are fused according to the acoustic-flux time fusion formula. Let the acoustic data be a(t) and the particle flux signal be p(t), where t represents time. Calculate the moving average of the acoustic data, and the formula is: where w is the width of the moving window, which is selected according to the fluctuation frequency of the data. Calculate the moving average of the particle flux signal, and the formula is: Fuse according to the acoustic-flux time fusion formula, and the formula is:

[0019] where k 1 , k 2 , k 3 are fusion coefficients, and k 1 + k 2 + k 3 = 1, and k 1 , k 2 , k 3 ≥ 0.

[0020] Furthermore, in the data fusion and model construction center, a learning model for magneto-inertial fusion experiments is constructed through the multi-modal experimental data prediction model formula. Let the fusion feature of the optical image and magnetic field strength after formula fusion be F img_mag (t), and the fusion feature of the acoustic data and particle flux signal be F acou_flux (t). The calculation formula is: where θ 1 , θ 2 , θ 3 are the parameters of the model, and w img_mag (i) and w acou_flux (j) are weight coefficients.

[0021] Furthermore, the steps for training the learning model of magneto-inertial fusion experiments in the data fusion and model construction center are as follows:

[0022] (1) Preprocess and fuse the optical images, magnetic field strengths, acoustic and particle flux data of magneto-inertial fusion experiments to obtain the fused feature sequences F img_mag (t) and F acou_flux (t), as well as the corresponding true values y true (t). Divide them into a training set and a test set according to 8:2. The training set is used for model training and parameter optimization, and the test set is used to evaluate the performance of the trained model;

[0023] (2) Randomly initialize the model parameters θ 1 , θ 2, θ 3 , based on the preliminary analysis of the data, set the weight decay parameter σ img_mag and σ acou_flux and the parameter of the learning rate η. The weight decay parameter is used to prevent the model from overfitting, and the learning rate controls the step size of the parameter update during the training process of the model;

[0024] (3) For each time step t in the training set, according to the current model parameters and the input fused feature sequence, calculate the predicted value y(t) of the model. The formula is: Calculate the error e(t) between the predicted value y(t) and the true value y true (t), e(t) = y(t) - y true (t). Calculate the gradient of the new data with respect to the model parameters according to the error. The formula is: Similarly, update θ 2 , θ 3 . Repeat the above steps to perform multiple iterative trainings on the training set until the mean squared error of the model on the training set no longer decreases significantly. During the iterative training process, the model continuously adjusts the parameters to reduce the error between the predicted value and the true value, thereby improving the accuracy and generalization ability of the model;

[0025] (4) After the training is completed, use the test set to evaluate the model. For each time step t in the test set, calculate the predicted value y(t) of the model and compare it with the true value y true (t), and calculate the mean squared error MSE on the test set. The formula is: where T test is the number of time steps in the test set. Judge the performance of the model according to the evaluation index. The smaller the mean squared error, the higher the prediction accuracy of the model and the better the performance.

[0026] Furthermore, in the real-time online model update module, the new fused feature data is clustered through the influence index. Let the new fused feature data set be Z = {z 1 , z 2 , …, z m}, where z i is composed of the fused feature of the optical image and the magnetic field strength and the fused feature of the acoustic data and the particle flux signal . Calculate the correlation degree d ij between data points. The formula is: d ij = α 0 · where α 0is the weight coefficient, M and N are the estimated maximum difference ranges of the two fusion features in the dataset respectively. The calculation of the correlation degree is based on the cosine similarity, considering the numerical differences between the fusion features of the optical image and the magnetic field strength, and the fusion features of the acoustic data and the particle flux signal. The differences are mapped to the interval (-1, 1) through the cosine function to reflect the similarity degree between data points, and the correlation matrix D is constructed, where D ij = d ij , calculate the influence index p of each data point i , and the formula is: p i = where k is a positive integer. When the influence index between two data points is greater than the threshold θ, they are classified into the same category, and thus the clustering result is obtained. Select the data point with the largest influence index in each cluster as the representative point R s . The representative point can represent the characteristics of the data within the cluster to a certain extent. By selecting the representative point, the amount of data for subsequent calculations can be reduced, while the key information of the data is retained, providing a basis for determining the local update area.

[0027] Furthermore, in the real-time online model update module, the similarity between the representative point and the historical data points is calculated to determine the local update area. Let the representative point be r, the historical data point be h, and the dimension of the fusion feature be d. Define the metric function M based on feature distance and trend similarity as follows: where f r,k and f h,k are the values of the representative point r and the historical data point h on the k-th dimensional fusion feature respectively, t r,k and t h,k are the characteristic trend values corresponding in the time dimension, and T is a normalization constant. This metric function comprehensively considers the distance between the representative point and the historical data points in the feature space and the trend similarity in the time dimension. By weighted summing the features of each dimension, a comprehensive similarity metric value is obtained. For each representative point r i , determine its local update area U, where μ ∈ (0, 1) is an adjustable threshold parameter. By determining the local update area, unnecessary computational amount is reduced, the efficiency of model update is improved, and at the same time, it is ensured that the updated model can better adapt to the feature changes of new data.

[0028] Compared with the prior art, the magnetic-inertial confinement fusion array signal acquisition and management system based on a distributed architecture has the following beneficial effects:

[0029] I. By adopting signal acquisition nodes designed with a distributed architecture, the present invention realizes the acquisition and management of multi-modal signals in magneto-inertial confinement fusion experiments, improves the accuracy and real-time performance of signal acquisition. By adding optical image acquisition devices and acoustic sensors on the basis of traditional physical signal sensor arrays, the dimension of signal acquisition is enriched, providing data support for subsequent data processing and analysis. At the same time, the present invention also realizes the preprocessing, fusion of multi-modal data, and the construction and optimization of learning models through the data fusion and model construction center, improving the efficiency of data processing.

[0030] II. The present invention conducts clustering analysis on the new fusion feature data by introducing an influence index and selects representative points for local update, realizing the real-time online optimization of the learning model. This not only reduces the cost of model update but also improves the adaptability of the model to new data, enabling the model to better cope with the complex and changeable environmental conditions in magneto-inertial confinement fusion experiments. At the same time, the present invention also defines a metric function based on feature distance and trend similarity to determine the local update area, thus ensuring the accuracy and effectiveness of model update, not only improving the stability and reliability of the model but also providing reliable technical support for the research and application of magneto-inertial confinement fusion experiments.

[0031] Other advantages, objectives, and features of the present invention will, to some extent, be elaborated in the subsequent specification, and to some extent, will be obvious to those skilled in the art based on the study of the following text, or can be learned from the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or in the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings without creative efforts based on these drawings.

[0033] Figure 1 It is a flowchart of a magneto-inertial confinement fusion array signal acquisition and management system based on a distributed architecture;

[0034] Figure 2 It is a module connection architecture diagram of a magneto-inertial confinement fusion array signal acquisition and management system based on a distributed architecture. DETAILED DESCRIPTION OF THE INVENTION

[0035] To further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention objective, the following, in combination with the drawings and preferred embodiments, details the specific implementation manner, structure, features, and their effects of the present invention as follows.

[0036] Example 1

[0037] Research on Plasma Behavior in Magneto-Inertial Confinement Fusion Experiments

[0038] Deploy multiple acquisition nodes on the key structures of the magneto-inertial confinement fusion experimental device. For example, evenly distribute magnetic field strength sensors and particle flux sensors around the plasma generation area to monitor the changes in magnetic field strength and particle flux in real time; install temperature sensors near the outer shell of the device to measure the temperature changes during device operation; at the same time, set up a high-resolution CCD camera at a position where the plasma emission image can be clearly captured, and reasonably arrange piezoelectric acoustic sensors around the device to collect the acoustic signals of plasma oscillations and the sound waves generated by device operation.

[0039] The acquisition nodes continuously collect data. For example, the magnetic field strength sensors collect magnetic field strength data at a certain frequency, the CCD camera captures plasma emission images at a set frame rate, and the acoustic sensors record acoustic signals in real time. Synchronously preprocess and standardize the collected original signals. For physical signals, extract the physical signal feature F B (t) with the formula: For optical image data, extract the image feature F I (t) with the formula: For acoustic data, extract the acoustic feature F A (t) with the formula:

[0040] Transmit the multi-modal feature data uploaded by each acquisition node to the data fusion and model construction center through a secure communication network. For the optical image and magnetic field strength data with spatial correlation, perform fusion according to the optical-magnetic spatial fusion formula. Let the preprocessed optical image feature matrix be I, where the element I ij represents the feature value of the image at the position (i,j), and the magnetic field strength matrix is B, and the element B ij corresponds to the magnetic field strength value at the corresponding position. The two have the same dimension M×N. Normalize the two with the formula: Calculate the fusion feature matrix F with the formula: For the acoustic data and particle flux signal with time correlation, perform fusion according to the acoustic-flux time fusion formula. Let the acoustic data be a(t) and the particle flux signal be p(t), where t represents time. Calculate the moving average of the acoustic data with the formula: Among them, w is the width of the sliding window, which is selected according to the fluctuation frequency of the data. Calculate the sliding average of the particle flux signal. The formula is: Fuse according to the acoustic-flux time fusion formula. The formula is:

[0041] Based on the fused multi-modal data feature set, use the multi-modal experimental data prediction model formula to construct a learning model. Let the fused feature of the optical image and magnetic field strength be F img_mag (t), and the fused feature of the acoustic data and particle flux signal be F acou_flux (t). The calculation formula is:

[0042] F acou_flux (t - j) + θ 3 , based on the fused feature sequence F img_mag (t) and F acou_flux (t), and the corresponding true value y true (t), randomly initialize the model parameters θ 1 、θ 2 、θ 3 . According to the preliminary analysis of the data, set the weight decay parameters σ img_mag and σ acou_flux and the parameter of the learning rate η. For each time step t of the training set, calculate the predicted value y(t) of the model according to the current model parameters and the input fused feature sequence, and calculate the error e(t) between the predicted value y(t) and the true value y true (t) = y(t) - y true (t). Calculate the gradient of the new data on the model parameters according to the error. The formula is: Similarly, update θ 2 、θ 3 . Repeat the above steps to perform multiple iterative trainings on the training set until the mean square error of the model on the training set no longer decreases significantly. After the training is completed, use the test set to evaluate the model. For each time step t in the test set, calculate the predicted value y9t) of the model and compare it with the true value y true (t), and calculate the mean square error MSE on the test set. The formula is: Among them, T test is the number of time steps of the test set. Judge the performance of the model according to the evaluation index.

[0043] When new data is uploaded from the acquisition nodes, cluster analysis is performed on the new fused feature data. For example, a new set of data containing the fused features of optical images and magnetic field intensity, as well as the fused features of acoustic data and particle flux signals, is newly acquired, and the correlation degree d between data points is calculated ij , and the formula is: where α 0 is the weight coefficient, M and N are the estimated maximum difference ranges of the two fused features in the dataset respectively, and a correlation matrix D is constructed, where D ij =d ij , and the influence index p of each data point is calculated i , and the formula is: where k is a positive integer. When the influence index between two data points is greater than the threshold θ, they are classified into the same class, and thus the clustering result is obtained. The data point with the largest influence index in each cluster is selected as the representative point R s .

[0044] Calculate the similarity between the representative point and the historical data points to determine the local update area. For each representative point, according to the metric function M based on feature distance and trend similarity, the formula is: For each representative point r i , its local update area is determined Calculate the correlation weight between the historical data points and the representative point within the local update area, perform weighted prediction using the existing model and weights, compare the predicted value with the true value to obtain the prediction error, and update the model parameters.

[0045] The data management module classifies and stores the collected original multi-modal data according to time series, acquisition nodes, and modal types, backs up the intermediate data and final model parameters during the model training process, and establishes an efficient indexing mechanism. Users can conveniently configure and manage the system through the operation interface of the user interaction module. For example, they can view the status of data acquisition nodes, adjust the data acquisition frequency, and at the same time can intuitively view the experimental data and model results in the form of charts, such as viewing the magnetic field intensity change curve and particle flux change trend graph of the plasma at different stages, as well as the result display of the model's prediction of plasma behavior.

[0046] In summary, in the magneto-inertial confinement fusion experiment, the system of the present invention provides comprehensive support for plasma behavior research through multi-modal data acquisition, fusion and model construction, as well as real-time update functions, effectively collects and integrates data, constructs accurate models, updates in a timely manner to adapt to changes, and assists scientific researchers in deeply understanding plasma characteristics and promoting related research progress.

[0047] Example Two:

[0048] Magnetic - Inertial Confinement Fusion Device Operation Status Monitoring and Fault Warning

[0049] Deploy acquisition nodes at each key part of the magnetic - inertial confinement fusion device, including installing magnetic field intensity sensors near the magnetic field generating device, arranging particle flux sensors and temperature sensors in the plasma confinement area, installing acoustic sensors near the cooling system pipes of the device, and setting high - resolution CMOS cameras at appropriate positions around the device to collect image data.

[0050] The acquisition nodes continuously collect various physical signal data and image data. For example, the magnetic field intensity sensors continuously monitor the magnetic field intensity, the particle flux sensors record the particle flux, the temperature sensors detect the temperature, the acoustic sensors capture the operating sound of the device, and the CMOS cameras capture images related to the appearance of the device and the internal plasma.

[0051] Synchronously pre - process and standardize the collected original signals. For physical signals, calculate the energy change rate feature F B (t) over a period of time through the physical signal feature extraction formula. The formula is as follows: For optical image data, calculate the brightness contrast feature F I (t) through the optical image feature extraction formula. The formula is as follows: For acoustic data, calculate the frequency centroid feature F A (t) through the acoustic feature extraction formula. The formula is as follows:

[0052] After the processed data is transmitted to the data fusion and model construction center, for the optical images and magnetic field intensity data with spatial correlation, perform fusion according to the optical - magnetic spatial fusion formula. Let the pre - processed optical image feature matrix be I, where the element I ij represents the feature value of the image at the position (i, j), and the magnetic field intensity matrix is B, and the element B ij corresponds to the magnetic field intensity value at the corresponding position. Both have the same dimension M×N. Normalize the two, and the formula is: Calculate the fusion feature matrix F, and the formula is: For the acoustic data and particle flux signals with time correlation, perform fusion according to the acoustic - flux time fusion formula. Let the acoustic data be a(t) and the particle flux signal be p(t), where t represents time. Calculate the moving average of the acoustic data, and the formula is: where w is the width of the moving window, selected according to the fluctuation frequency of the data. Calculate the moving average of the particle flux signal, and the formula is: Fusion is performed according to the acoustic-flux time fusion formula, and the formula is:

[0053] Based on the fused multi-modal data feature set, a learning model is constructed using the multi-modal experimental data prediction model formula. Let the fused feature of the optical image and magnetic field strength be F img_mag (t), and the fused feature of the acoustic data and particle flux signal be F acou_flux (t). The calculation formula is: Then, training and optimization are carried out. The initial values of the model parameters are set, the weight decay parameter and learning rate are determined according to the data characteristics, iterative training is performed using the training set, and the model performance is evaluated through the test set to ensure that the model can accurately predict the operating state of the device.

[0055] When new data is uploaded from the acquisition node, preprocess the new data, fuse the optical image and magnetic field strength data and the acoustic data and particle flux signal to extract features, and cluster the new fused feature data through the influence index. Let the new fused feature data set be Z = {z 1 , z 2 , …, z m}. Calculate the correlation degree d ij between data points. The formula is: Construct the correlation matrix D, where D ij = d ij . Calculate the influence index p i of each data point. The formula is: When the influence index between two data points is greater than the threshold θ, they are classified into the same class. Thus, in each cluster, select the data point with the largest influence index as the representative point R s . Calculate the similarity between the representative point and the historical data points to determine the local update region. Let the representative point be r, the historical data point be h, and the fused feature dimension be d. Define the metric function M based on feature distance and trend similarity as: The formula is For each representative point r i , determine its local update region U, Within the local update region, calculate the correlation weight between the historical data points in the region and the representative point, perform weighted prediction using the existing model and weights, compare the predicted value with the true value to obtain the prediction error, calculate the gradients of the error with respect to each parameter of the model respectively, and update the model parameters according to the gradients and learning rate.

[0056] The data management module classifies and stores the collected raw data, backs up the model data, and establishes an index to facilitate data retrieval and query. Through the user interaction module, users can view the device operation status data, such as real-time temperature data, magnetic field intensity data, and the prediction results of the model on the device operation status. When the model predicts that there may be a fault risk, the system can send a warning message to the user through the user interaction module. The user can take timely measures according to the warning message, such as checking the equipment and adjusting the operation parameters, to ensure the safe and stable operation of the magneto-inertial confinement fusion device.

[0057] In summary, for the operation status monitoring and fault warning of the magneto-inertial confinement fusion device, the present invention utilizes multi-modal signal acquisition, fusion modeling, and real-time update mechanisms to achieve precise control of the device operation status, can detect abnormalities and give warnings in a timely manner, and ensure the safe and stable operation of the device, which has important practical value.

[0058] The above are only the preferred embodiments of the present invention, and do not impose any form of limitation on the present invention. Although the present invention has been disclosed above with preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications to equivalent embodiments by using the disclosed technical content within the scope of the technical solution of the present invention. However, as long as it does not depart from the content of the technical solution of the present invention, any brief modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. A magnetic-inertial confinement fusion array signal acquisition and management system based on a distributed architecture, characterized in that: The system includes: signal acquisition front end, data fusion and model building center, real-time online model update module, data management module and user interaction module; The signal acquisition front end: adopts a distributed architecture, and the acquisition nodes are distributed on the key structures of the device. Each acquisition node measures the physical signal data of magnetic field intensity, particle flux, temperature, and pressure through a sensor array of traditional physical signals, and an optical image acquisition device captures the luminous image data of plasma at different stages. The acoustic sensor collects the acoustic characteristic data of plasma oscillation, sound waves generated by equipment operation, and energy release. The collected original signals are preprocessed and standardized simultaneously, and feature extraction is performed according to the modal type. The data fusion and model building center: receives the multimodal feature data uploaded by each acquisition node through a secure communication network, fuses the optical image and magnetic field intensity data with spatial correlation according to the optical-magnetic field space fusion formula, fuses the acoustic data and particle flux signal with time correlation according to the acoustic-flux time fusion formula, builds a learning model for the magnetic-inertial confinement fusion experiment based on the fused multimodal data feature set and the multimodal experimental data prediction model formula, optimizes the parameters of the model by the stochastic gradient descent method, and evaluates the trained model using real-time data; The real-time online model update module: when new data is uploaded from the acquisition node, the new data is preprocessed, the data is fused and feature extracted, cluster analysis is performed on the new fused feature data, and representative points are selected, the similarity between the representative points and the historical data points is calculated to determine the local update area, and within the local update area, the correlation weights between the historical data points and the representative points in the area are calculated, and weighted prediction is performed using the existing model and weights, and the predicted value is compared with the true value to obtain the prediction error, and the gradient of the error to each model parameter is calculated respectively, and the model parameters are updated according to the gradient and the learning rate; The data management module: classifies and stores the collected original multimodal data according to time series, collection nodes and modality types, stores the intermediate data and final model parameters in the model training process, and establishes an index mechanism to quickly retrieve and query data; The user interaction module provides an operation interface for users to configure and manage the system conveniently, and displays experimental data and model results in an intuitive graphical form.

2. According to claim 1, a magnetic-inertial confinement fusion array signal acquisition and management system based on a distributed architecture is characterized in that: The acquisition device used in the signal acquisition front end is: Sensor arrays for traditional physical signals: magnetic field intensity sensors, particle flux sensors, temperature sensors; Optical image acquisition related: high-resolution CCD camera, high-resolution CMOS camera; Acoustic sensor related: piezoelectric acoustic sensor.

3. According to claim 1, a magnetic-inertial confinement fusion array signal acquisition and management system based on a distributed architecture is characterized in that: The signal acquisition front end performs feature extraction according to the modal type, and extracts the feature F of the physical signal through the physical signal feature extraction formula. B( t ) , the formula is: Among them, w is the time window length, B ( t ) is the magnetic field strength; for optical image data, the image feature F is extracted by the optical image feature extraction formula I( t ) , the formula is: I ( x,y ) is the image pixel value; for acoustic data, the acoustic feature F is extracted by the acoustic feature extraction formula A( t ) , the formula is: Among them A ( t ) Represents the amplitude of the sound wave signal, f i It is the discrete frequency point obtained by fast Fourier transforming the acoustic signal, and N is the number of FFT points.

4. The magnetic-inertial confinement fusion array signal acquisition and management system based on a distributed architecture according to claim 1, characterized in that: In the data fusion and model building center, the optical image and magnetic field intensity data with spatial correlation are fused according to the optical-magnetic field spatial fusion formula. Let the feature matrix of the preprocessed optical image be I, where the element I ij Indicates that the image is at position ( i,j ) The eigenvalue at the magnetic field strength matrix is ​​B, and the element B ij The magnetic field strength values ​​at the corresponding positions, and both have the same dimension M×N, the two are normalized, the formula is: Calculate the fusion feature matrix F, the formula is: Among them, α and β are weight coefficients, and 0<α<1, β≥0, and r is a small integer.

5. The magnetic-inertial confinement fusion array signal acquisition and management system based on a distributed architecture according to claim 1, characterized in that: In the data fusion and model building center, the acoustic data and particle flux signals with time correlation are fused according to the acoustic-flux time fusion formula. Assume that the acoustic data is a ( t ) , the particle flux signal is p ( t ) , where t represents time, the sliding average of the acoustic data is calculated as follows: Where w is the sliding window width, which is selected according to the fluctuation frequency of the data, and the sliding average of the particle flux signal is calculated. The formula is: Fusion is performed according to the acoustic-flux time fusion formula, which is: Among them, k1, k2, k3 are fusion coefficients, and k1+k2+k3=1, k1,k2,k3≥0.

6. The magnetic-inertial confinement fusion array signal acquisition and management system based on a distributed architecture according to claim 1, characterized in that: The data fusion and model building center constructs a learning model for magnetic-inertial confinement fusion experiments through a multimodal experimental data prediction model formula. The optical image and magnetic field intensity fusion feature after fusion through the formula is assumed to be F img_mag ( t ) , the fusion characteristics of acoustic data and particle flux signal are F acou_flux ( t ) , the calculation formula is: Among them, θ1, θ2, θ3 are the parameters of the model, w img_mag ( i ) and w acou_flux ( j ) is the weight coefficient.

7. The magnetic-inertial confinement fusion array signal acquisition and management system based on a distributed architecture according to claim 1, characterized in that: The steps of training the learning model for the magnetic-inertial confinement fusion experiment in the data fusion and model building center are as follows: (1) Preprocess and fuse the optical images, magnetic field intensity, acoustic and particle flux data of the magnetic-inertial confinement fusion experiment to obtain the fused characteristic sequence F img_mag ( t ) and F acou_flux ( t ) , and the corresponding true value y true( t ) , divided into training set and test set according to 8:2; (2) Randomly initialize the model parameters θ1, θ2, θ3, and set the weight decay parameter σ based on the preliminary analysis of the data. img_mag and σ acou_flux and the parameters of the learning rate η; (3) For each time step t in the training set, the model prediction value y is calculated based on the current model parameters and the input fusion feature sequence ( t ) , the formula is: Calculate the predicted value y ( t ) and the true value y true( t ) The error e ( t ) =y ( t ) -y true( t ) , calculate the gradient of the new data to the model parameters based on the error, the formula is: Similarly, update θ2 and θ3, repeat the above steps, and iterate the training set for multiple times until the mean square error of the model on the training set no longer decreases significantly; (4) After training, the model is evaluated using the test set. For each time step t in the test set, the model's predicted value y is calculated ( t ) , and the true value y true( t ) For comparison, the mean square error MSE on the test set is calculated as follows: Among them, T test The number of time steps for the test set is used to judge the performance of the model based on the evaluation metric.

8. The magnetic-inertial confinement fusion array signal acquisition and management system based on a distributed architecture according to claim 1, characterized in that: In the real-time online model updating module, the new fusion feature data is clustered by the influence index. Suppose the new fusion feature data set is Z = {z1, z2, ..., z m }, where z i Fusion of optical image and magnetic field intensity features And the fusion characteristics of acoustic data and particle flux signals Composition, calculate the correlation d between data points ij , the formula is: Where α0 is the weight coefficient, M and N are the estimated values ​​of the maximum difference range of the two fusion features in the data set, and the correlation matrix D is constructed, where D ij =d ij , calculate the influence index p of each data point i , the formula is: Where k is a positive integer. When the influence index between two data points is greater than the threshold θ, they are classified into the same category to obtain the clustering result. In each cluster, the data point with the largest influence index is selected as the representative point R. s .

9. The magnetic-inertial confinement fusion array signal acquisition and management system based on a distributed architecture according to claim 1, characterized in that: In the real-time online model update module, the similarity between the representative point and the historical data point is calculated to determine the local update area. Let the representative point be r, the historical data point be h, and the fusion feature dimension be d. The metric function M based on feature distance and trend similarity is defined as: where f r,k and f h,k are the values ​​of the representative point r and the historical data point h on the kth dimension fusion feature, t r,k and t h,k is the characteristic trend value corresponding to the time dimension, T is a normalized constant, for each representative point r i , determine its local update area U, where μ∈(0,1) is an adjustable threshold parameter.

Images