Multi-parameter collaborative control method and system for vacuum filtration of liquid hydrogen spherical tank

Abstract

The present application relates to liquid hydrogen storage and transportation equipment vacuum heat insulation manufacturing technical field, disclose a kind of liquid hydrogen spherical tank vacuum filtering multi-parameter collaborative control method and system, including obtaining the temperature, vacuum degree and vibration monitoring data in liquid hydrogen spherical tank and its space coordinates, based on temperature field space distribution division physical field subarea, calculate the mean value of vacuum degree and vibration intensity mean value of each subarea, generate multi-physical field heterogeneous subarea topology structure;Tensor decomposition algorithm is used to extract physical field coupling factor matrix, identify the dominant coupling relationship between physical field;According to the subarea period quality label matrix scheduling multi-parameter monitoring data transmission in shallow fading period, scheduling collaborative control instruction is completed before the arrival of depth fading, control instruction is output to actuating mechanism.The present application is in the extreme environment of temperature field, vacuum degree field, vibration field multi-physical field coupling influence, improves the overall performance and safety of liquid hydrogen spherical tank vacuum filtering multi-parameter collaborative control system.

Description

Technical Field

[0001] This invention relates to the field of vacuum insulation manufacturing technology for liquid hydrogen storage and transportation equipment, and more specifically, to a method and system for multi-parameter coordinated control of vacuum filtration in liquid hydrogen spherical tanks. Background Technology

[0002] As a cryogenic liquid hydrogen storage device, the liquid hydrogen spherical tank requires a vacuum insulation layer to reduce heat transfer through a vacuum process, while a filtration system removes impurities from the liquid hydrogen to ensure its purity. During the simultaneous rapid cooling and vacuum filtration phase, the interior of the spherical tank exhibits a complex multiphysics distribution: the temperature gradients from -100°C at the liquid hydrogen inlet to -253°C in the core region, and the vacuum level varies from... Pa gradually decreased Pa, while the vibrations generated by the refrigeration equipment, vacuum pump and filter pump are transmitted through the pipes and tank to form a vibration field.

[0003] Multi-parameter collaborative control systems rely on wireless sensor networks to collect distributed monitoring data such as temperature, pressure, flow rate, and filtration efficiency, and then issue control commands. However, wireless communication channels are simultaneously affected by the coupling of three physical fields: temperature, vacuum, and vibration, exhibiting significant spatiotemporal heterogeneity. This restricts the reliability of multi-parameter data transmission and the real-time performance of the control system: Spatially, channel attenuation varies greatly across different temperature regions, leading to resource waste in high-temperature areas and transmission failure in extremely low-temperature areas with a single transmission parameter configuration; temporally, the rapid multipath fading caused by vibration and the slow time-varying fading caused by temperature and vacuum levels combine to form complex fading modes across multiple time scales, making it impossible to avoid deep fading periods with fixed-time transmission; traditional methods model the channel based on a single physical field, failing to capture the nonlinear coupling effects and spatial correlation characteristics of multiple physical fields; and the lack of coordinated optimization of spatial partitioning and time scheduling results in low efficiency of multi-parameter data transmission and insufficient reliability of collaborative control decisions. Summary of the Invention

[0004] This invention provides a multi-parameter collaborative control method and system for vacuum filtration in liquid hydrogen spherical tanks, addressing the shortcomings of traditional methods in related technologies, such as the one-sidedness of modeling based on a single physical field and the neglect of coupling effects by simple linear superposition; resource waste in high-temperature zones and transmission failure in extremely low-temperature zones due to single transmission parameter configuration; inability of traditional methods to distinguish and capture fading at multiple time scales; inability of fixed-time transmission to avoid data loss and command delays caused by deep fading; and suboptimal overall performance resulting from independent optimization of spatial and temporal dimensions in traditional methods.

[0005] This invention provides a multi-parameter coordinated control method for vacuum filtration in a liquid hydrogen spherical tank, comprising:

[0006] Acquire temperature, vacuum, and vibration monitoring data and their spatial coordinates inside the liquid hydrogen spherical tank. Use spatial interpolation algorithms to generate three-dimensional spatial distribution maps of temperature field, vacuum field, and vibration field. Divide the physical field into partitions based on the spatial distribution of temperature field, calculate the mean vacuum degree and mean vibration intensity of each partition, and generate a multi-physics heterogeneous partition topology.

[0007] The spatiotemporal data of temperature field, vacuum field and vibration field are organized into third-order tensors. The physical field coupling factor matrix is ​​extracted using tensor decomposition algorithm. The weight distribution of each physical field in the coupling factor is analyzed to identify the dominant coupling relationship between physical fields.

[0008] The channel quality data sequence of wireless links in each partition is obtained. The channel fading sequence is separated into slow time-varying components and fast time-varying components using a time series decomposition algorithm. A mapping function between the physical field coupling factor and the slow time-varying component of the channel is established. A correlation model between vibration spectrum characteristics and the fast time-varying component of the channel is established. The channel state evolution sequence in the future time window of each partition is predicted.

[0009] Based on the predicted channel fast time-varying component sequence, the deep fading period and shallow fading period of each partition are identified, and a partition period quality label matrix is ​​generated.

[0010] For each physical field partition, a capacity optimization algorithm is used to generate a partition-specific transmission parameter configuration based on the predicted slow time-varying state of the channel, and the time slot ratio of each partition is allocated according to the partition dynamic index and the number of nodes.

[0011] Based on the quality label matrix of the time period, multi-parameter monitoring data is scheduled to be transmitted during the shallow fading period. The scheduling and coordination control command is issued before the deep fading arrives and the control command is output to the actuator.

[0012] Furthermore, the specific method for generating three-dimensional spatial distribution maps of the temperature field, vacuum field, and vibration field using spatial interpolation algorithms is as follows:

[0013] The Kriging spatial interpolation algorithm is used for the monitoring data of each physical field. The spatial coordinates and monitoring values ​​of the known monitoring points are input, and the estimated value of the location to be interpolated is calculated by weighted linear combination.

[0014] Kriging interpolation weights are obtained by solving the Kriging equations. The equations are in the form that the linear combination of the weights of each monitoring point and the semivariogram values ​​is equal to the semivariogram values ​​from the point to be interpolated to each monitoring point. The constraint is that the sum of all weights is equal to 1.

[0015] The semivariogram is obtained by fitting the distance between monitoring points and the difference in monitoring values. It is represented by an exponential model or a spherical model. The model parameters include nugget value, partial sill value and range. The spatial variation characteristics of historical monitoring data are determined by fitting the data using the least squares method.

[0016] Furthermore, the specific method for dividing the physical field into zones based on the spatial distribution of the temperature field is as follows:

[0017] Based on the three-dimensional spatial distribution map of the temperature field, extract the set of isotherms, calculate the temperature gradient between adjacent isotherms, and identify the locations where the temperature gradient is greater than the threshold as candidates for partition boundaries.

[0018] The region growing clustering algorithm is used to aggregate spatial regions with temperature gradients less than a threshold into a single temperature partition. Spatial grid points of unassigned partitions are selected as seed points. The temperature gradient between adjacent grid points and seed points is checked. If it is less than the threshold, the adjacent points are added to the current partition and used as new seed points. This expansion is repeated until it cannot continue. Then, a new seed point is selected to start a new partition until all grid points are assigned.

[0019] The temperature difference threshold within a temperature zone is set to 2 to 5 times the measurement accuracy of the temperature sensor, and the temperature gradient threshold is set to the mean of the global temperature gradient plus one standard deviation.

[0020] Furthermore, the specific method for extracting the physical field coupling factor matrix using the tensor decomposition algorithm is as follows:

[0021] The spatiotemporal data of the three physical fields are organized into a third-order tensor, with the first dimension being the number of spatial grid points, the second dimension being the number of time series sampling points, and the third dimension being the number of physical field types.

[0022] Each physical field data is normalized based on its range, scaling it to the range of 0 to 1 to eliminate dimensional differences.

[0023] The third-order tensor is decomposed into the sum of the outer products of several rank components using the CP tensor decomposition algorithm. Each rank component contains a spatial mode factor, a temporal mode factor, and a physical field coupling factor.

[0024] The factor matrix is ​​solved iteratively using the alternating least squares method. The third factor matrix is ​​updated while the other two factor matrices are fixed. The process is continued until convergence by minimizing the mean square norm of the reconstruction error. The convergence condition is that the relative change in the objective function value between two adjacent iterations is less than a set threshold.

[0025] Furthermore, the specific method for identifying the dominant coupling relationship between physical fields is as follows:

[0026] Analyze each column vector of the physical field coupling factor matrix. The three elements of the coupling factor vector correspond to the weighting coefficients of the temperature field, vacuum field, and vibration field, respectively.

[0027] Calculate the absolute value percentage of the weight coefficients of each physical field, and determine the coupling mode type of each rank component based on the weight percentage. If the weight percentage of a certain physical field exceeds 50%, it is determined to be a coupling mode dominated by a single physical field. If the sum of the weights of two physical fields exceeds 80%, it is determined to be a coupling mode dominated by dual physical field coupling. If the weights of the three physical fields are relatively balanced, it is determined to be a coupling mode of three physical field synergistic coupling.

[0028] The frequency of each type of coupling mode in all rank components is statistically analyzed, and the coupling mode type with the highest frequency is identified as the overall dominant coupling relationship.

[0029] Furthermore, the specific method for establishing the mapping function between the physical field coupling factor and the slow time-varying components of the channel is as follows:

[0030] Obtain the coupling factor vector, mean temperature, and mean vacuum level for each partition, and perform Z-score standardization on the mean temperature and mean vacuum level.

[0031] Establish a linear regression mapping function, with the inputs being the elements of the physical field coupling factor vector, the mean temperature, the mean vacuum degree, and the interaction term between temperature and vacuum degree, and the output being the slow time-varying component of the channel.

[0032] A training sample set is constructed using historical monitoring data, and the mean squared error is used as the loss function. The regression coefficient vector is solved by the normal equation to minimize the loss function.

[0033] Furthermore, the specific method for establishing the correlation model between vibration spectrum characteristics and fast time-varying components of the channel is as follows:

[0034] The vibration data were subjected to Fast Fourier Transform to extract the vibration spectrum features, including the amplitude and frequency of each frequency component, and the main frequency components with a cumulative energy ratio of more than 90% were selected.

[0035] The time correlation of fast time-varying components is established using an autoregressive model. The model input is the channel fast time-varying component sequence at historical time and the sinusoidal function representation of the vibration spectrum characteristics at the current time. The output is the predicted value of the channel fast time-varying component at the current time.

[0036] The autoregression order is determined based on the autocorrelation function of the fast time-varying components of the channel, and the minimum time lag that makes the autocorrelation coefficient drop below 0.1 is selected.

[0037] The autoregressive coefficients and vibration influence coefficients are solved by the least squares method, and the phase parameters are determined by fitting a sine function to each vibration frequency component individually.

[0038] Furthermore, the specific method for generating partition-specific transmission parameter configurations using capacity optimization algorithms is as follows:

[0039] The goal is to maximize channel capacity, which is solved under the conditions of satisfying transmit power constraints and bit error rate constraints. The channel capacity is determined by the channel bandwidth, transmit power, antenna gain, noise power spectral density and path loss coefficient. The path loss coefficient is obtained by converting the slow time-varying state of the channel from decibel units to a linear ratio.

[0040] For each time period within the time window, the path loss coefficient corresponding to each time period is calculated and a weighted average is performed. The weights are determined based on the quality label matrix of the time period partition. The shallow fading time period is assigned a larger weight, and the deep fading time period is assigned a smaller weight.

[0041] The optimal scheme is selected from the set of candidate modulation schemes using a discrete search method. For each modulation scheme, the minimum signal-to-noise ratio required to meet the target bit error rate is calculated in reverse according to its theoretical bit error rate formula. Then, the required transmit power is calculated, and the scheme with the largest channel capacity is selected from all feasible modulation schemes.

[0042] For zones with temperatures below -200°C, a temperature correction factor is introduced to improve the transmit power margin. The temperature correction factor increases as the temperature decreases.

[0043] Furthermore, it also includes online update steps:

[0044] Collect actual transmission performance data for each partition and time period, including success rate, average latency, and total system power consumption;

[0045] The channel prediction error is calculated and defined as the root mean square error between the predicted channel state and the actual channel state.

[0046] The regression coefficients of the physical field coupling-channel mapping function are updated using a recursive least squares algorithm. The regression coefficients are updated by multiplying the difference between the actual observed values ​​and the predicted values ​​by the gain matrix.

[0047] The autoregressive coefficients and vibration influence coefficients of the vibration-fading correlation model are updated using the gradient descent algorithm, and the parameters are adjusted along the negative direction of the error gradient based on the prediction error.

[0048] A forgetting factor is introduced to assign an exponentially decaying weight to historical data. A smaller forgetting factor is set for partitions with a faster time-varying rate, and a larger forgetting factor is set for partitions with a slower time-varying rate.

[0049] This invention provides a multi-parameter coordinated control system for vacuum filtration in a liquid hydrogen spherical tank, comprising:

[0050] The physical field monitoring module is used to acquire temperature, vacuum, and vibration monitoring data inside the liquid hydrogen spherical tank.

[0051] The spatial partitioning module is used to generate three-dimensional spatial distribution maps of temperature field, vacuum field and vibration field and divide the physical field into partitions;

[0052] The coupling analysis module is used to perform tensor decomposition on multiphysics spatiotemporal data and extract the physical field coupling factor matrix.

[0053] The channel prediction module is used to separate the multi-timescale components of channel fading and predict the channel state evolution sequence within the future time window of each partition.

[0054] The quality label generation module is used to generate a quality label matrix for different time periods.

[0055] The parameter configuration module is used to generate the transmission parameter configuration and time slot allocation scheme for each partition;

[0056] The transmission scheduling module is used to schedule the transmission times of multi-parameter monitoring data and collaborative control commands;

[0057] The online learning module is used to collect actual transmission performance data and update the physical field coupling-channel mapping function.

[0058] The beneficial effects of this invention are as follows:

[0059] According to the present invention, by organizing the spatiotemporal data of the three physical fields—temperature field, vacuum field, and vibration field—during the vacuum filtration process of a liquid hydrogen spherical tank into third-order tensors and performing CP decomposition, the coupling mode factor matrix between the physical fields is extracted, capturing the nonlinear coupling relationship and co-evolution mode between temperature-vacuum degree, temperature-vibration, and vacuum degree-vibration. This overcomes the shortcomings of traditional methods that are based on single physical field modeling and ignore coupling effects by simple linear superposition, thus achieving an accurate description of the wireless channel characteristics under the influence of multi-physical field coupling.

[0060] The physical field is partitioned based on the spatial distribution of the temperature field, and the average temperature, average vacuum degree and average vibration intensity of each partition are integrated to establish a spatial mapping function between the physical field coupling factor and the slow time-varying components of the wireless channel. For the channel heterogeneity of different physical field partitions, partition-specific transmission parameter configurations (transmit power, modulation method and coding rate) are generated. This overcomes the problems of resource waste in high temperature areas and transmission failure in extremely low temperature areas caused by single transmission parameter configuration, thus realizing differentiated adaptive transmission in spatially heterogeneous channels.

[0061] By using a time series decomposition algorithm, the channel fading sequence is separated into slow time-varying components and fast time-varying components. The slow time-varying components are associated with slow changes in temperature and vacuum, while the fast time-varying components are associated with rapid multipath phase fluctuations caused by vibration. A vibration-fading correlation model is established to predict the channel state evolution within the future time window and identify deep fading periods and shallow fading periods. This overcomes the shortcomings of traditional methods in distinguishing and capturing fading at multiple time scales, thus achieving accurate prediction of channel evolution in the time dimension.

[0062] Based on the predicted channel state time evolution sequence, multi-parameter monitoring data is scheduled to be transmitted to the control center first during shallow fading periods, while non-delay-sensitive data is delayed or coding redundancy is increased during deep fading periods. Delay-sensitive collaborative control commands are scheduled to be issued before deep fading occurs, realizing optimized scheduling of transmission time based on the channel spatiotemporal quality map. This overcomes the problem that fixed-time transmission cannot avoid data loss and command delay caused by deep fading. Therefore, it ensures the reliable real-time transmission of multi-parameter monitoring data such as temperature, pressure, flow, and filtering efficiency, and the accurate and timely issuance of collaborative control commands such as cooling control, vacuum regulation, and filtering flow control.

[0063] By generating differentiated transmission parameter configurations for each physical field partition to adapt to spatial heterogeneity, and scheduling data transmission times according to the predicted time period quality label matrix to adapt to temporal heterogeneity, the coordinated optimization of spatial partition configuration and time scheduling is achieved. This overcomes the problem of suboptimal overall performance caused by independent optimization of spatial and temporal dimensions in traditional methods. Therefore, in extreme environments with the coupled influence of multiple physical fields such as temperature field, vacuum field, and vibration field, the overall performance and safety of the multi-parameter coordinated control system for vacuum filtration of liquid hydrogen spherical tanks are improved. Attached Figure Description

[0064] Figure 1 This is a flowchart of a multi-parameter coordinated control method for vacuum filtration in a liquid hydrogen spherical tank according to the present invention;

[0065] Figure 2 This is a hybrid bar graph of channel state temporal evolution and multi-timescale component prediction according to the present invention. Detailed Implementation

[0066] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0067] At least one embodiment of the present invention discloses a multi-parameter coordinated control method for vacuum filtration in a liquid hydrogen spherical tank, such as... Figure 1 As shown, it includes the following steps:

[0068] Step 100: Obtain the spatiotemporal distribution data of the physical field inside the spherical tank and generate a multi-physics heterogeneous partition topology.

[0069] Real-time temperature monitoring data and their spatial coordinates from a distributed temperature sensor array deployed within a liquid hydrogen spherical tank are acquired. Real-time vacuum monitoring data and their spatial coordinates from a pressure sensor array are also acquired. Vibration amplitude and frequency data and their spatial coordinates from a vibration sensor are obtained. Kriging spatial interpolation is used to process the temperature, vacuum, and vibration data respectively, generating three-dimensional spatial distribution maps of the temperature, vacuum, and vibration fields inside the spherical tank. Based on the temperature field spatial distribution maps, isotherm clustering is used to divide the tank space into several temperature zones, with the temperature difference within each zone less than a set threshold. The average vacuum level and average vibration intensity within each zone are obtained. The temperature zone identifier, average vacuum level, and average vibration intensity are correlated to generate a heterogeneous physical field partition topology, and a zone identifier is assigned to each multi-parameter sensor node.

[0070] Furthermore, the calculation methods for the average vacuum level and average vibration intensity within each of the aforementioned zones are as follows: Based on the spatial range of the temperature zone, extract the vacuum level values ​​of all spatial grid points within the temperature zone from the three-dimensional spatial distribution map of the vacuum field, and calculate the arithmetic mean of these vacuum level values ​​as the average vacuum level of the zone; extract the vibration amplitude values ​​of all spatial grid points within the temperature zone from the three-dimensional spatial distribution map of the vibration field, and calculate the root mean square value of the vibration amplitude as the average vibration intensity of the zone. The calculation formula is as follows: ,in For grid point indexing, The number of grid points within the partition. For the first Vibration amplitude at each grid point.

[0071] It should be noted that the input to the Kriging spatial interpolation algorithm is the known spatial coordinates and monitoring values ​​of the monitoring points, and the output is the estimated value of the location to be interpolated, using a weighted linear combination. Calculate the interpolated values, where For monitoring point indexing, For the number of monitoring points, the weight Determined by minimizing the estimated variance.

[0072] Furthermore, the aforementioned Kriging interpolation weights The Kriging equations are obtained by solving the system of equations, which has the following form: ,

[0073] in For monitoring point indexing, For monitoring points and The semivariogram values ​​between For Lagrange multipliers, the constraint is: The aforementioned semivariogram is based on the distance between monitoring points. The index model is commonly used and is obtained by fitting the difference between the monitored values ​​and the actual values. or spherical model It means that among them Value of a nugget. For the partial sill value, For variable ranges, these parameters were determined by fitting the spatial variability characteristics of historical monitoring data using the least squares method.

[0074] It should be noted that the isotherm clustering method includes the following steps: extracting a set of isotherms based on the three-dimensional spatial distribution map of the temperature field. An isotherm is a curve formed by connecting spatial points with equal temperature values; calculating the temperature gradient between adjacent isotherms and identifying locations with temperature gradients greater than a threshold as candidate partition boundaries; and using a region growing clustering algorithm to aggregate spatial regions with temperature gradients less than the threshold into single temperature partitions, generating partition identifiers and partition spatial ranges.

[0075] Furthermore, the specific implementation steps of the aforementioned region growing clustering algorithm are as follows: select a spatial grid point of an unassigned partition as a seed point and mark it as the current partition; check the neighboring grid points of the seed point, and if the temperature gradient between the neighboring point and the seed point is less than a threshold, add the neighboring point to the current partition and use it as a new seed point; repeat the expansion process until the current partition can no longer be expanded; select the grid point of the next unassigned partition as the seed point of the new partition, and repeat the above process until all grid points are assigned to a certain partition.

[0076] Furthermore, the aforementioned temperature difference threshold within the temperature zones is determined based on the temperature monitoring accuracy of the collaborative control system, specifically set to 2 to 5 times the measurement accuracy of the temperature sensors, ensuring the relative uniformity of the temperature field within the zones. The aforementioned temperature gradient threshold is determined based on the statistical characteristics of the temperature field within the spherical tank space, specifically calculated as the global temperature gradient mean plus one standard deviation, ensuring that the identified zone boundaries are located in areas of rapid temperature change.

[0077] In a practical application of a liquid hydrogen storage facility, the spherical tank has a volume of 100 cubic meters and is equipped with 24 temperature sensors, 12 pressure sensors, and 6 vibration sensors. Physical field data were collected 30 minutes after the vacuum filtration operation was started. The collected raw data are shown in Table 1.

[0078] Table 1 shows the raw monitoring data of some sensor nodes:

[0079]

[0080] The monitoring data was spatially interpolated using the Kriging interpolation algorithm to generate a three-dimensional spatial distribution map. Then, the temperature zones were divided using isotherm clustering, with a temperature difference threshold of 15°C and a temperature gradient threshold of 18°C / m. The zoning results are shown in Table 2.

[0081] Table 2. Topological structure of heterogeneous partitioning of physical fields:

[0082]

[0083] Step 200: Extract the coupling mode factor of the multiphysics field and identify the dominant coupling relationship between the physics fields.

[0084] The spatiotemporal data of the temperature field, vacuum field, and vibration field generated in step 100 are organized into a third-order tensor. The first dimension The number of spatial grid points, the second dimension The third dimension represents the number of time series sampling points. The number of physical field types is specified (temperature field, vacuum field, and vibration field, a total of 3). Since the temperature field, vacuum field, and vibration field have different dimensions, before organizing the third-order tensor, the data for each physical field are normalized based on their range to scale them to the range of 0 to 1, eliminating the influence of dimensional differences on tensor decomposition. The CP tensor decomposition algorithm is then used to decompose the third-order tensor into rank tensors. The sum of the factor matrices:

[0085]

[0086] in For rank component index, Let be the rank of the tensor decomposition. For spatial pattern factor, For time pattern factor, The physical field coupling factor. This represents the outer product operation. It extracts the physical field coupling factor matrix. ,in The physical field coupling factor of the first rank component. The physical field coupling factor of the second rank component. For the first The physical field coupling factors of each rank component are analyzed to determine the weight distribution of the three physical fields of temperature, vacuum degree, and vibration in each coupling factor, and to identify the dominant coupling relationship and spatial cooperative mode of temperature-vacuum degree-vibration.

[0087] Furthermore, the specific method for identifying the dominant coupling relationship of the physical field mentioned above is as follows: [The text abruptly ends here, likely due to an incomplete sentence or a formatting error.] Each column vector Analysis is performed on the coupling factor vector. The three elements correspond to the weight coefficients of the temperature field, vacuum field, and vibration field, respectively. The absolute value percentage of the weight coefficient of each physical field is calculated. If the weight percentage of a certain physical field exceeds 50%, the coupling mode is dominated by a single physical field. If the sum of the weights of two physical fields exceeds 80%, the coupling mode is dominated by two physical fields. If the weights of the three physical fields are relatively balanced, the coupling mode is a three-physical-field synergistic coupling. The overall dominant coupling relationship is determined by statistically analyzing the frequency of occurrence of each coupling mode in all rank components.

[0088] It should be noted that the input to the CP tensor decomposition algorithm is a third-order tensor. The output is a spatial mode factor matrix. Time pattern factor matrix Coupling factor matrix of physical fields The objective function is minimized by iteratively solving the factor matrix using the alternating least squares method. .

[0089] Furthermore, the specific implementation steps of the aforementioned alternating least squares method are as follows: Initialize the factor matrix. , and It is a random matrix; fixed and Expand the third-order tensor into a matrix along the first dimension. By solving the least squares problem renew ,in Represents the Khatri-Rao product. Indicates pseudo-inverse; fixed and Expand the tensor along the second dimension, through renew ;fixed and Expand the tensor along the third dimension, through renew Repeat the above update process until the objective function converges or the maximum number of iterations is reached. The aforementioned convergence condition is that the relative change in the objective function value between two adjacent iterations is less than a set threshold, i.e. ,in For the first The objective function value of the next iteration. The convergence threshold is usually set to 1. to .

[0090] Furthermore, the aforementioned tensor decomposition rank Based on the balance between reconstruction error and model complexity, the kernel consistency diagnostic method is specifically adopted. The minimum rank value that makes the kernel consistency index close to 100% is selected from the candidate rank values. The value is usually in the range of 3 to 8, which ensures that the main coupling modes between physical fields can be fully captured while avoiding overfitting.

[0091] The physics data in Table 1, along with the time series data from the past 10 minutes, are organized into a third-order tensor. The dataset contains 256 spatial grid points, 120 temporal sampling points (sampled every 5 seconds), and 3 physics types. After range-based normalization of each physics data, it is decomposed using the CP tensor decomposition algorithm, with a set rank. The method converged after 85 iterations using alternating least squares. The extracted physical field coupling factor matrix... Some of the results are shown in Table 3.

[0092] Table 3 Physical field coupling factor matrix:

[0093]

[0094] Analysis of the coupling factor matrix revealed that the temperature-vacuum dual-field coupling mode occurred most frequently (twice), accounting for 40%, indicating that the co-evolution of the temperature field and the vacuum field during the vacuum filtration process of the liquid hydrogen spherical tank has a dominant influence on the wireless channel characteristics.

[0095] Step 300: Separate the multi-timescale components of channel fading and establish the mapping relationship between the physical field and the channel.

[0096] For each physical field partition generated in step 100, obtain the historical channel quality data sequence of the wireless links within the partition, including the signal-to-noise ratio sequence. Bit error rate sequence and fading depth sequence .

[0097] Furthermore, the aforementioned historical channel quality data is obtained through the channel monitoring function of the wireless sensor nodes. Specifically, the sensor nodes measure the received signal strength and noise power in real time during data transmission, calculate the signal-to-noise ratio (SNR) and record it; detect data packet transmission errors through checksums, and calculate the bit error rate by counting the number of error bits per unit time; calculate the fading depth based on the attenuation of the measured SNR relative to ideal channel conditions, expressed in decibels. All channel quality data are stored in the database of the control center in time series for modeling purposes.

[0098] Channel fading sequences are decomposed using time series decomposition algorithms. Separation into slow time-varying components and fast time-varying components ,in The slow time-varying component characterizes the long-term effects of slow changes in temperature and vacuum on the channel, while the fast time-varying component characterizes the rapid multipath phase fluctuations caused by vibration.

[0099] Establish a spatial mapping function between the physical field coupling factor and the slow time-varying components of the channel. Obtain the physical field coupling factor matrix generated in step 200. Coupling factor vectors corresponding to each partition Obtain the average temperature of the physical field partition. and the average vacuum degree The mean temperature and mean vacuum level were standardized using Z-scores to eliminate dimensional differences, and a mapping function was established using regression analysis.

[0100]

[0101] in For physics field type index, The number of physical field types is equal to 3. For the first The coupling factor of a physical field The regression coefficients are determined by fitting historical data using the least squares method.

[0102] The aforementioned physical field coupling-channel mapping function is a linear regression model, whose input is the physical field coupling factor vector. Average temperature and the average vacuum degree The output is the channel slow time-varying component. The training of the physical field coupling-channel mapping function employs supervised learning, utilizing historical monitoring data to construct a training sample set. Each training sample contains an input feature vector. and the corresponding true slow time-varying component observations ,

[0103] in The coupling factor of the first physical field. The coupling factor for the second physical field. This indicates transpose, and the loss function uses mean squared error. ,in For training sample index, The number of training samples.

[0104] Solving the regression coefficient vector using the least squares method ,in The regression coefficient of the first physical field coupling factor. For the first The regression coefficients of each physical field coupling factor are used to minimize the loss function.

[0105] Furthermore, the specific calculation method for solving the regression coefficients using the aforementioned least squares method is as follows:

[0106] Organize the feature vectors of all training samples into a design matrix Organize the corresponding real observations into vectors Through the normal equation Solving for the regression coefficient vector, the normal equation is obtained by applying the loss function with respect to... Find the partial derivative and set it to zero to obtain the result.

[0107] A vibration-fading correlation model was established. Vibration spectral features were extracted from the vibration data using a Fast Fourier Transform. ,in For the first The amplitude of each frequency component, For the first The frequencies of each frequency component. The time correlation of the rapidly time-varying components is established using an autoregressive model:

[0108]

[0109] in For autoregressive time lag indexes, Let the order be the autoregressive order. These are the autoregressive coefficients. Index of vibration frequency components, The number of vibration frequency components. For the first The frequency of each vibration frequency component and These represent the influence coefficients and phases of the vibration frequency component on channel fading, respectively. Use white noise. Predict future time windows for each partition. Channel state time evolution sequence ,in The time window length is determined based on the decision cycle of the collaborative control system, and is usually set to 2 to 5 times the control cycle, with a value range of 10 to 60 seconds.

[0110] Furthermore, the aforementioned phase The range of values ​​is constrained by Within the interval, the physical meaning of the phase parameters is ensured by applying periodic boundary conditions during model training. This relates to the aforementioned white noise. It is zero-mean Gaussian white noise, and its standard deviation is determined based on the measured noise level of the fast time-varying components of the channel. It is usually 10% to 20% of the standard deviation of the channel fading depth measurement. In model training and prediction, the noise standard deviation is estimated by statistical residuals of historical data.

[0111] The aforementioned vibration-fading correlation model is an autoregressive model, whose input is the fast time-varying component sequence of the channel at historical moments. and the vibration spectrum characteristics at the current moment The output is the predicted value of the fast time-varying components of the channel at the current moment. The vibration-fading correlation model is trained using supervised learning. A training sample set is constructed using historical channel fading data and vibration monitoring data. Each training sample contains an input vector:

[0112]

[0113] and the corresponding real fast time-varying component observations ,in The amplitude of the first vibration frequency component. The frequency of the first vibration frequency component. For the first The amplitude of each vibration frequency component, For the first The frequency of each vibration frequency component is used, and the loss function employs the mean square error:

[0114] ,in For training time indexing, For the first At each training time step, the model parameters are solved using the least squares method. , Minimize the loss function.

[0115] Furthermore, the specific process of solving the autoregressive model parameters using the aforementioned least squares method is as follows: organize the input vectors at all training times into a matrix. Organize the corresponding real observations into vectors Model parameter vector By solving the normal equation Obtain, among which These are the first-order autoregressive coefficients. For the first Autoregressive coefficient of order, The influence coefficient of the first vibration frequency component. For the first The influence coefficient of each vibration frequency component, for the phase parameter This is determined by individually fitting a sine function to each vibration frequency component, thus maximizing the correlation between the vibration term and the residual signal.

[0116] Furthermore, the aforementioned autoregressive order The autoregression order is determined based on the autocorrelation function of the fast time-varying components of the channel. Specifically, the minimum time lag that reduces the autocorrelation coefficient to below 0.1 is chosen as the autoregression order, typically ranging from 3 to 10. The aforementioned number of vibration frequency components... Based on the energy distribution of the vibration spectrum, the main frequency component with a cumulative energy ratio of over 90% is selected, typically ranging from 5 to 20.

[0117] For the wireless links within Zone 2, acquire a 5-minute historical channel quality data sequence containing 300 sampling points. Use the Empirical Mode Decomposition (EMD) algorithm to analyze the channel fading sequence. The system is separated into slow time-varying components and fast time-varying components. Seven eigenmode function components are iteratively extracted. The sum of the first five high-frequency components is taken as the fast time-varying component, and the sum of the last two low-frequency components and the residual is taken as the slow time-varying component. The physical field coupling factor vector of Zone2 partition is extracted as follows: The average temperature was -182.5°C, and the average vacuum degree was [missing information]. Pa, after Z-score standardization, is -0.35 and 0.52, respectively. A mapping function is established and regression coefficients are obtained through least squares fitting, and the predicted values ​​of the slow time-varying components are calculated. dB. A Fast Fourier Transform was performed on the vibration data, extracting eight main frequency components, with a cumulative energy percentage of 92.3%, ranging from 15Hz to 85Hz. A vibration-fading correlation model was established, and the autoregressive order was set. The autoregressive coefficients and vibration influence coefficients were obtained by solving the least squares method. The channel state evolution sequence within the next 30-second time window was predicted, and some prediction results are shown in Table 4.

[0118] Table 4. Future Channel State Prediction Results for Zone 2 (Partial Moments):

[0119]

[0120] It should be noted that the time series decomposition algorithm uses the empirical mode decomposition method, and its input is the channel fading sequence. The output is a slow time-varying component. and fast time-varying components The intrinsic mode function components are extracted iteratively, and the sum of the high-frequency components is taken as the fast time-varying component, while the sum of the low-frequency components and the residual is taken as the slow time-varying component.

[0121] Furthermore, the specific steps for extracting intrinsic mode function (EMF) components using the aforementioned empirical mode decomposition are as follows: Identify all local maxima and local minima of the signal; perform cubic spline interpolation on all maxima to generate an upper envelope, and perform cubic spline interpolation on all minima to generate a lower envelope; calculate the mean of the upper and lower envelopes as the local mean curve; subtract the local mean curve from the original signal to obtain candidate EMFs; check whether the candidate function meets the EMF condition, i.e., the difference between the number of extrema and the number of zero-crossing points does not exceed 1 and the local mean is close to zero. If not, repeat the above process using the candidate function as a new input signal; extract the obtained EMF as a component, subtract this component from the original signal to obtain the residual signal, and repeat the above decomposition process on the residual signal until the residual signal becomes a monotonic function or has fewer than two extrema. The aforementioned criterion for judging that the local mean is close to zero is that the root mean square value of the local mean is less than 5% to 10% of the root mean square value of the candidate function.

[0122] Figure 2 Displays the channel state prediction results for the Zone2 partition within a 30-second time window.

[0123] Step 400: Generate a time-segment quality label matrix to provide a spatiotemporal quality map for data transmission.

[0124] Based on the channel fast time-varying component sequence predicted in step 300 It identifies the deep fading periods and shallow fading periods of each physical field partition within a future time window.

[0125] Define fading depth threshold ,when Time-marked time period During the deep fading period, when The time stamps are designated as periods of shallow fading. A quality label matrix for each time period is generated. ,in The number of physical field partitions. The number of time periods within the time window. For partitioned indexes, For time period index, matrix elements Indicates the first The partition in the _th ... The period is characterized by shallow fading (good channel quality). This indicates deep fading (poor channel quality). The time-segmented quality label matrix provides a spatiotemporal quality map for multi-parameter data transmission and control command issuance in collaborative control systems.

[0126] Furthermore, the aforementioned number of time periods Based on the length of the time window The granularity of time period division is determined, and the calculation method is as follows: ,in The time interval is determined based on the rate of change of the fast time-varying components of the channel, and is typically set to 1 to 5 seconds to ensure that rapid changes in channel quality can be captured. This indicates the rounding up operation.

[0127] Furthermore, the aforementioned fading depth threshold The threshold is determined based on the statistical distribution of historical channel fading data, specifically calculated as the median of the historical fading depth sequence, ensuring a basic balance between the time proportions of deep fading and shallow fading periods. Alternatively, the threshold can be adjusted to a value between the 60th and 80th percentiles of the historical fading depth, depending on the system's requirements for transmission reliability.

[0128] Based on the predicted channel fast time-varying component sequences in Table 4, historical channel fading data is analyzed to determine the fading depth threshold. dB. Identify deep fading and shallow fading periods in Zone2 within a 30-second time window, with the time interval set to [value missing]. The time window is divided into 10 time segments, with each segment defined as a second. Channel quality labels are applied to all time segments across the four partitions to generate a partition-time segment quality label matrix. Some of the results are shown in Table 5.

[0129] Table 5 Quality Label Matrix for Different Time Periods (Partial Time Periods):

[0130]

[0131] In the table, a value of 1 indicates shallow fading (good channel quality), and a value of 0 indicates deep fading (poor channel quality). This matrix provides a spatiotemporal quality map for the cooperative control system, guiding the timing of multi-parameter data transmission and control command issuance.

[0132] Step 500: Generate differentiated transmission parameter configurations and time slot allocation schemes for physical field partitions.

[0133] For each physical field partition generated in step 100, based on the channel slow time-varying state predicted in step 300... A capacity optimization algorithm is used to generate partition-specific transmission parameter configurations. The input to the capacity optimization algorithm is the channel's slow time-varying state. Given the channel parameters, the output is the optimal transmit power. Modulation method and coding rate To maximize channel capacity With the goal of satisfying the transmit power constraint, and bit error rate constraints Solve under the given conditions.

[0134] in For channel bandwidth, For antenna gain, For noise power spectral density, The path loss coefficient function maps the slow time-varying state of the channel to the path loss coefficient, and is calculated as follows: This indicates that the fading depth is converted from decibel units to a linear scale.

[0135] Furthermore, the specific implementation of the aforementioned capacity optimization algorithm in the time dimension is as follows: for future time windows... Each time period within Based on the channel slow time-varying state sequence predicted in step 300 Calculate the path loss coefficient for each time period. The path loss coefficients for all time periods are weighted and averaged, with the weights proportional to the expected data transmission volume in each time period, to calculate the average path loss coefficient. ,in For time period index, For the first The weight of each time period and satisfying ; using average path loss coefficient Replacement instantaneous path loss coefficient Capacity optimization is performed to ensure that the generated transmission parameter configuration can adapt to changes in channel conditions throughout the time window, guaranteeing continuous and stable transmission of multi-parameter data within the time window. The aforementioned time period weighting... Based on the partitioned time period quality label matrix generated in step 400 Determined, if the first The period is the period of shallow decay, i.e. Then a larger weight is assigned, if it is a period of deep fading. Then set a smaller weight, specifically calculated as follows: ,in Using time-period indexing allows capacity optimization to focus more on channel conditions during periods of shallow fading.

[0136] Furthermore, the aforementioned channel bandwidth Based on the communication protocol used by the wireless sensor network, for the IEEE 802.15.4 protocol, MHz, for the Bluetooth Low Energy protocol, MHz. The aforementioned antenna gain. Determined based on the sensor node antenna type; for omnidirectional antennas, Up to 2, for directional antennas, Up to 10. The aforementioned noise power spectral density Determined based on the thermal noise level of the working environment, under room temperature conditions. The noise level is reduced by dBm / Hz in the cryogenic environment of the liquid hydrogen tank. Up to -175dBm / Hz.

[0137] Furthermore, the aforementioned maximum transmission power The power consumption is determined based on the battery capacity and energy consumption budget of the wireless sensor node, while also considering the electromagnetic compatibility requirements and human safety radiation limits in the liquid hydrogen tank environment. It is usually set to 10 to 100 milliwatts to ensure that communication needs are met without affecting the safe operation of the equipment.

[0138] Modulation mode output by capacity optimization algorithm The modulation order is a numerical representation of the modulation order. Based on the channel quality, a candidate modulation scheme is selected and the output modulation order is mapped to a specific modulation scheme. The modulation order 2 corresponds to BPSK modulation, the modulation order 4 corresponds to QPSK modulation, the modulation order 16 corresponds to 16QAM modulation, and the modulation order 64 corresponds to 64QAM modulation.

[0139] Furthermore, the aforementioned target bit error rate Based on the data transmission reliability requirements of the liquid hydrogen spherical tank collaborative control system, the following settings are determined for monitoring data of safety-critical parameters such as temperature and pressure: For non-safety-critical auxiliary parameter monitoring data, set The aforementioned capacity optimization algorithm employs a discrete search method. The specific process is as follows: First, modulation schemes are selected sequentially from the candidate modulation scheme set. For each modulation scheme, its theoretical bit error rate formula is used... Reverse calculation satisfies Minimum signal-to-noise ratio required Then, the required transmit power is calculated in reverse using the channel capacity formula. ,like If the modulation scheme is feasible, select the scheme with the largest channel capacity from all feasible modulation schemes as the optimal transmission parameter configuration.

[0140] Furthermore, the aforementioned theoretical bit error rate formula is calculated using the appropriate formula depending on the modulation scheme. For BPSK modulation, For QPSK modulation, For 16QAM modulation, For 64QAM modulation, ,in Given a Gaussian Q-function, the inverse function of the Q-function is obtained by looking up a table or through numerical calculation to get the desired signal-to-noise ratio. The minimum signal-to-noise ratio is then calculated. Then, according to the definition of signal-to-noise ratio Inverse calculation of the required transmit power .

[0141] The aforementioned coding rate Based on the selected modulation scheme and target bit error rate, the standard coding rate configuration using channel coding technology is determined. For security-critical data, a lower coding rate is selected to enhance error correction capability, usually 1 / 2 or 2 / 3. For non-security-critical data, a higher coding rate is selected to improve transmission efficiency, usually 3 / 4 or 5 / 6.

[0142] Obtain the rate of change of multi-parameter data monitored within each physical field zone. Calculate the rate of change of temperature data. Pressure data change rate Traffic data change rate and rate of change of filtering efficiency data .

[0143] Furthermore, the aforementioned rates of change for each parameter are calculated using the finite difference method. Specifically, for a continuously sampled monitoring data sequence, the rate of change is calculated as the difference between the values ​​at two adjacent sampling times divided by the time interval. ,in The sampling time interval is determined based on the variation characteristics of each parameter. For parameters that change slowly, such as temperature and vacuum level, Up to 10 seconds, for parameters that change rapidly, such as flow rate and filtering efficiency. The calculation method for the rate of change of other parameters is similar up to 3 seconds.

[0144] Since the change rates of the various parameters have different dimensions, each change rate is normalized based on its range to scale it to a range of 0 to 1. Then, the maximum value of the change rates of all normalized parameters within the partition is taken as the partition dynamics index. Based on the dynamic indicators of the partitions and the requirements for coordinated control, a proportional allocation algorithm is used to allocate a certain percentage of radio time slots to each partition:

[0145]

[0146] in For partitioned indexes, For partitioned indexes, The number of physical field partitions. For the first The percentage of time slots in each partition, For the first Dynamic indicators for each partition For the first Dynamic indicators for each partition For the first Number of sensor nodes in each partition For the first The number of sensor nodes within each partition. Generate the transmission parameter configuration for each partition. and time slot allocation scheme ,in For the first The optimal transmit power for each partition. For the first The optimal modulation scheme for each partition. For the first The optimal coding rate for each partition.

[0147] Furthermore, the aforementioned time slot allocation scheme satisfies the following constraint: the sum of the time slot proportions of all partitions equals 1, i.e. The time slot percentage for each partition is a non-negative number, i.e. The time slot ratio of each partition automatically meets the above constraints according to the proportional allocation algorithm. The normalization term in the denominator ensures that the sum of the ratios is 1, and the dynamic index and the number of nodes in the numerator are both non-negative to ensure that the ratios are non-negative.

[0148] In this embodiment, to adapt to severe channel attenuation in extremely low temperature regions, the transmit power margin is further increased for partitions with temperatures below -200°C during the transmission parameter configuration generation in step 500. Specifically, a temperature correction factor is introduced into the capacity optimization algorithm. ,in Let be the temperature sensitivity coefficient, whose dimension is the reciprocal of temperature, such that... Since it is a dimensionless number, the corrected transmit power is: This ensures reliable transmission in extremely low temperature regions. The aforementioned temperature sensitivity coefficient... The value is determined based on the additional channel attenuation in the extremely low temperature region, typically ranging from 0.01 to 0.05, in the reciprocal of °C, which increases the transmit power in the core region at -253°C by 10% to 30% compared to the region at -200°C.

[0149] For the four physical field partitions in Table 2, based on the slow time-varying channel state predicted in step 300, a capacity optimization algorithm is used to generate partition-specific transmission parameter configurations. System parameters are set as follows: channel bandwidth. MHz (using IEEE 802.15.4 protocol), antenna gain Noise power spectral density dBm / Hz (low temperature environment) Maximum transmit power mW. Target bit error rate is set for safety-critical parameters. For non-safety-critical parameter settings The optimal scheme is selected from the candidate modulation scheme set {BPSK, QPSK, 16QAM, 64QAM} using a discrete search method.

[0150] For Zone3 and Zone4 partitions, since the temperature is below -200°C, a temperature correction factor is introduced and set. °C The temperature correction factor for Zone3 was calculated to be: The temperature correction factor for Zone4 is [value missing]. The configuration results of the partition-specific transmission parameters are shown in Table 6.

[0151] Table 6. Transmission parameter configuration and time slot allocation scheme for each partition:

[0152]

[0153] As shown in Table 6, Zone 4, located in the extremely low temperature region, suffers from severe channel attenuation and requires the most conservative transmission scheme (maximum transmit power, lowest order modulation, and lowest coding rate) to ensure transmission reliability, but this significantly reduces channel capacity. Zones 3 and 2, while having lower temperatures, have better vacuum levels, allowing for a balance between reliability and efficiency using QPSK modulation. Zone 1, with its higher temperature and best channel quality, achieves the highest channel capacity using 16QAM modulation.

[0154] Step 600: Schedule the transmission time of multi-parameter monitoring data and collaborative control commands, and output control commands to the actuator.

[0155] Based on the partitioned time period quality label matrix generated in step 400 The scheduling of temperature monitoring data, pressure monitoring data, flow monitoring data, and filtration efficiency monitoring data during periods of shallow fading ( Priority transmission to the control center is given to periods of deep fading. ( ), delay the transmission of non-latency-sensitive monitoring data, or increase the coding redundancy before transmitting latency-sensitive monitoring data.

[0156] Furthermore, the aforementioned increase in coding redundancy refers to reducing the coding rate during deep fading periods, specifically reducing the coding rate from 3 / 4 or 5 / 6 during normal periods to 1 / 2 or 2 / 3, thereby enhancing forward error correction capabilities to counteract channel quality degradation and ensuring the reliability of delay-sensitive monitoring data transmission.

[0157] The control center executes collaborative control decisions based on the received multi-parameter monitoring data, generating cooling control commands, vacuum adjustment commands, and filter flow control commands.

[0158] Furthermore, the specific execution process of the aforementioned collaborative control decision is as follows: Based on temperature monitoring data, the deviation between the current cooling rate and the target cooling curve is calculated; a cooling control command is generated using a PID control algorithm to adjust the opening of the temperature control valve; based on vacuum monitoring data, it is determined whether the current vacuum level has reached the target vacuum level range. If not, a vacuum adjustment command is generated to increase the vacuum pump power; if it exceeds the range, the vacuum pump power is reduced; based on filtration efficiency monitoring data and flow monitoring data, the working status of the filtration system is calculated. If the filtration efficiency decreases, a filtration flow control command is generated to reduce the flow rate to extend the filtration time; if the filtration efficiency is normal, the flow rate is maintained or appropriately increased to accelerate the filtration process.

[0159] For time-delay-sensitive control commands, based on the partitioned time-segment quality label matrix It identifies the future shallow fading period of the partition where the target actuator is located and issues control commands before deep fading occurs. Specifically, if the current time is... The target partition was predicted to be in If the period is characterized by deep fading, then in to Control command transmission is completed during the shallow fading period, among which This is the early warning time for deep fading. This refers to the duration of deep fading. Control commands are transmitted wirelessly to the corresponding actuator nodes. After receiving the commands, the actuator nodes output them to the temperature control valve of the cooling system, the vacuum pump controller of the vacuum system, and the flow regulating valve of the filtration system, achieving multi-parameter coordinated control.

[0160] Furthermore, the aforementioned deep fading warning time Based on the quality label matrix of different time periods Current moment The time interval until the start of the next deep fading period is determined.

[0161] Right now ,in Index the target partition. The aforementioned deep fading duration. The length of consecutive deep fading periods in the quality label matrix of the partitioned time period is determined, that is, the continuous deep fading period is counted starting from the start time of deep fading. The number of time periods multiplied by the time interval between time periods. Get duration ,in This represents the number of periods of continuous deep fading.

[0162] In this embodiment, to further improve the transmission reliability of safety-critical control commands, a multi-path redundancy transmission strategy is adopted for the emergency shutdown command and leakage protection command with the highest safety level during the control command transmission scheduling in step 600. Specifically, multiple non-intersecting paths of nodes in the partition where the target actuator is located are identified, and the control commands are sent in parallel through all paths. The actuator executes immediately upon receiving the first arrived and correctly verified copy of the command, discarding subsequent duplicate copies, thus ensuring highly reliable real-time delivery of safety-critical commands.

[0163] Furthermore, the specific method for identifying multiple non-intersecting paths mentioned above is as follows: the wireless sensor network is modeled as a directed graph, with sensor nodes and actuator nodes as nodes, and edges as wireless links between nodes; taking the control center as the source node and the target actuator as the destination node, the maximum flow algorithm is used to calculate the maximum flow value from the source node to the destination node, and the maximum flow value is equal to the maximum number of non-intersecting paths; the maximum flow is decomposed into multiple unit flows with a flow of 1 using a path decomposition algorithm, and each unit flow corresponds to a non-intersecting path; from the set of decomposed paths, 2 to 3 paths with the fewest hops are selected as redundant transmission paths to ensure that there are no common relay nodes between paths.

[0164] Based on the time-segment quality label matrix in Table 5, multi-parameter monitoring data transmission for each zone is scheduled. During time segment 1 (0-3 seconds), the channel quality of Zone1, Zone2, and Zone4 is good. The temperature, pressure, flow, and filtration efficiency data of these three zones are transmitted to the control center. Zone 3 is in a period of deep fading ( Non-delay-sensitive data was delayed during transmission. In period 2 (3-6 seconds), Zones 2 and 4 transitioned to deep fading, while Zones 1 and 3 maintained good channel quality, and data transmission from Zones 1 and 3 was scheduled. In period 3 (6-9 seconds), Zone 1 transitioned to deep fading, but temperature monitoring data was delay-sensitive; therefore, the coding rate was reduced from 3 / 4 to 1 / 2 before transmission during this period to enhance error correction capabilities. Throughout the 30-second time window, each zone optimized transmission timing based on the quality label matrix, achieving an average successful transmission rate of 97.8% for multi-parameter monitoring data, a 12.3 percentage point improvement compared to the fixed-time transmission method.

[0165] The control center executes coordinated control decisions based on the received multi-parameter monitoring data. Analysis of temperature monitoring data revealed that the actual cooling rate of Zone 4 was 0.85°C / min, deviating from the target cooling curve by 0.60°C / min. A cooling control command was generated using a PID control algorithm, adjusting the temperature control valve opening from 45% to 38%. Analysis of vacuum monitoring data revealed that the current vacuum level of Zone 2 was... Pa, the target vacuum level has not been reached. Pa generates a vacuum adjustment command, increasing the vacuum pump power from 75% to 90%. Analysis of the filtration efficiency data reveals a current filtration efficiency of 92.5%, within the normal range. A filtration flow control command is generated to maintain the current flow rate of 2.8 cubic meters per hour. These control commands need to be sent to the corresponding actuator nodes. The target actuator for the cooling control command is located in Zone 4, the target actuator for the vacuum adjustment command is located in Zone 2, and the target actuator for the filtration flow control command is located in Zone 1. Referring to Table 5, at the current time (start of time period 1, 0 seconds), Zone 4 is identified as experiencing shallow fading during time period 1 (0-3 seconds) and continuous deep fading during time periods 2 and 3 (3-9 seconds). The deep fading warning time is calculated. seconds, duration of deep fading Therefore, the cooling control command is sent to the Zone4 actuator within the shallow fading period of Time 1. For the vacuum regulation command, Zone2 is identified as having shallow fading in Time 1, deep fading in Time 2, and shallow fading in Time 3. The vacuum regulation command is sent to the Zone2 actuator within Time 1. For the filter flow control command, Zone1 is identified as having shallow fading in Time 1. The command is sent to the Zone1 actuator within Time 1. For the emergency shutdown command (assuming a leakage risk trigger), the target actuator is located in Zone 3. A multi-path redundancy transmission strategy is adopted. The three non-intersecting paths of the target actuator are identified. The command is sent in parallel through the three paths and completed within the shallow fading period of Time 2, ensuring highly reliable real-time transmission. The transmission scheduling results are shown in Table 7.

[0166] Table 7. Results of Cooperative Control Command Transmission Scheduling:

[0167]

[0168] By optimizing the transmission time scheduling based on the channel spatiotemporal quality map, the reliability and real-time performance of multi-parameter monitoring data and collaborative control commands are significantly improved, ensuring the safe and stable operation of the multi-parameter collaborative control system for the vacuum filtration process of the liquid hydrogen spherical tank.

[0169] Step 700: Collect actual transmission performance data and update the physical field coupling-channel mapping function online.

[0170] Collect actual transmission performance data for each physical field partition at different times, including the success rate of multi-parameter monitoring data. Average delay of control commands Total system energy consumption .

[0171] Furthermore, the aforementioned success rate Defined as the ratio of the number of successfully received data packets to the total number of data packets sent, calculated as follows: ,in The number of data packets that the control center successfully received and verified. The total number of data packets sent by the sensor node. The aforementioned average latency. Defined as the average time from when a control command is issued by the control center to when it is received and acknowledged by the actuator node, it is calculated by taking the arithmetic mean of the delays of all control commands within the evaluation time window.

[0172] The channel prediction error is calculated and defined as the predicted channel state. With actual channel state Root mean square error:

[0173]

[0174] in For sampling time index, For the first Each sampling time, The number of sampling points is determined based on the model evaluation time window length and sampling interval, and is calculated as follows: ,in The time window for model evaluation is typically set to be 10 to 30 minutes. The sampling interval for channel quality data is typically 1 to 5 seconds. The transmission efficiency index is calculated and defined as the amount of data successfully transmitted per unit of energy consumption.

[0175]

[0176] in The total number of data packets sent within the evaluation time window. The internal count is the sum of the number of data packets sent by all sensor nodes across all partitions. The total system energy consumption is expressed in joules, including the transmit and receive energy consumption of all sensor nodes. The calculation method is as follows: ,in For sensor node indexing, This represents the total number of sensor nodes. and The first The transmit and receive power of each node. and These represent the transmit duration and receive duration of the sensor node within the evaluation time window, respectively.

[0177] The physical field coupling-channel mapping function and vibration-fading correlation model established in step 300 are updated using an online learning algorithm. The regression coefficients of the mapping function are updated using a recursive least squares algorithm.

[0178]

[0179] in For the regression coefficient vector, The input feature vector contains coupling factors, temperature, vacuum level, etc. For the actual observed slow time-varying components of the channel, This is the gain matrix. An adaptive filtering algorithm is used to update the autoregressive coefficients of the vibration-fading correlation model. and vibration influence coefficient Minimize the weighted sum of squared prediction errors.

[0180] The aforementioned adaptive filtering algorithm is used to update the parameters of the autoregressive model online, and its input is the current observation value of the fast time-varying component of the channel. and model predictions The output is the updated autoregressive coefficients. and vibration influence coefficient The gradient descent algorithm is used to determine the prediction error. Update the model parameters.

[0181] Furthermore, the specific calculation formula for updating the autoregressive model parameters using the aforementioned gradient descent algorithm is as follows: For the autoregressive coefficients, ,in The learning rate is typically between 0.001 and 0.01.

[0182] For the vibration influence coefficient The parameter update direction is the negative direction of the error gradient. The prediction error is reduced by gradually adjusting the parameters. The parameter update stops when the moving average of the prediction error stabilizes below the set threshold. The aforementioned stop threshold is determined according to the channel quality measurement accuracy and is usually set to 2 to 3 times the measurement standard deviation. For fading depth prediction error, the stop threshold is usually 1 to 3 dB.

[0183] In this embodiment of the application, in order to accelerate model convergence and improve prediction accuracy, a forgetting factor is introduced during the online learning process in step 700. By assigning exponentially decaying weights to historical data, the model focuses more on recent data.

[0184] Furthermore, the aforementioned forgetting factor The forgetting factor is determined based on the time-varying rate of the physical field distribution. For partitions with a faster time-varying rate, a smaller forgetting factor is set to quickly track channel changes, typically ranging from 0.90 to 0.95. For partitions with a slower time-varying rate, a larger forgetting factor is set to maintain model stability, typically ranging from 0.98 to 0.995.

[0185] The gain matrix of the recursive least squares algorithm is updated as follows:

[0186]

[0187]

[0188] in Let be the covariance matrix. By introducing a forgetting factor, the model can quickly adapt to the time-varying characteristics of the physical field distribution and channel properties.

[0189] Furthermore, the aforementioned covariance matrix Initialize to at the start of the recursive least squares algorithm ,in It is the identity matrix. The covariance matrix is ​​a relatively large positive number, typically ranging from 1000 to 10000, which makes the algorithm highly sensitive to new data in the early stages. As the recursive iteration proceeds, the covariance matrix gradually converges to a stable value that reflects the statistical characteristics of the data.

Claims

1. A multi-parameter coordinated control method for vacuum filtration in a liquid hydrogen spherical tank, characterized in that, Includes the following steps: Acquire temperature, vacuum, and vibration monitoring data and their spatial coordinates inside the liquid hydrogen spherical tank. Use spatial interpolation algorithms to generate three-dimensional spatial distribution maps of temperature field, vacuum field, and vibration field. Divide the physical field into partitions based on the spatial distribution of temperature field, calculate the mean vacuum degree and mean vibration intensity of each partition, and generate a multi-physics heterogeneous partition topology. The spatiotemporal data of temperature field, vacuum field and vibration field are organized into third-order tensors. The physical field coupling factor matrix is ​​extracted using tensor decomposition algorithm. The weight distribution of each physical field in the coupling factor is analyzed to identify the dominant coupling relationship between physical fields. The channel quality data sequence of wireless links in each partition is obtained. The channel fading sequence is separated into slow time-varying components and fast time-varying components using a time series decomposition algorithm. A mapping function between the physical field coupling factor and the slow time-varying component of the channel is established. A correlation model between vibration spectrum characteristics and the fast time-varying component of the channel is established. The channel state evolution sequence in the future time window of each partition is predicted. Based on the predicted channel fast time-varying component sequence, the deep fading period and shallow fading period of each partition are identified, and a partition period quality label matrix is ​​generated. For each physical field partition, a capacity optimization algorithm is used to generate a partition-specific transmission parameter configuration based on the predicted slow time-varying state of the channel, and the time slot ratio of each partition is allocated according to the partition dynamic index and the number of nodes. Based on the quality label matrix of the time period, multi-parameter monitoring data is scheduled to be transmitted during the shallow fading period. The scheduling and coordination control command is issued before the deep fading arrives and the control command is output to the actuator.

2. The multi-parameter coordinated control method for vacuum filtration in a liquid hydrogen spherical tank according to claim 1, characterized in that, The specific method for generating the three-dimensional spatial distribution maps of temperature field, vacuum field, and vibration field using the spatial interpolation algorithm is as follows: The Kriging spatial interpolation algorithm is used for the monitoring data of each physical field. The spatial coordinates and monitoring values ​​of the known monitoring points are input, and the estimated value of the location to be interpolated is calculated by weighted linear combination. Kriging interpolation weights are obtained by solving the Kriging equations. The equations are in the form that the linear combination of the weights of each monitoring point and the semivariogram values ​​is equal to the semivariogram values ​​from the point to be interpolated to each monitoring point. The constraint is that the sum of all weights is equal to 1. The semivariogram is obtained by fitting the distance between monitoring points and the difference in monitoring values. It is represented by an exponential model or a spherical model. The model parameters include nugget value, partial sill value and range. The spatial variation characteristics of historical monitoring data are determined by fitting the data using the least squares method.

3. The multi-parameter coordinated control method for vacuum filtration in a liquid hydrogen spherical tank according to claim 1, characterized in that, The specific method for dividing the physical field into partitions based on the spatial distribution of the temperature field is as follows: Extract the set of isotherms from the three-dimensional spatial distribution map of the temperature field, calculate the temperature gradient between adjacent isotherms, and identify the locations where the temperature gradient is greater than the threshold as candidates for partition boundaries. The region growing clustering algorithm is used to aggregate spatial regions with temperature gradients less than a threshold into a single temperature partition. Spatial grid points of unassigned partitions are selected as seed points. The temperature gradient between adjacent grid points and seed points is checked. If it is less than the threshold, the adjacent points are added to the current partition and used as new seed points. This expansion is repeated until it cannot continue. Then, a new seed point is selected to start a new partition until all grid points are assigned. The temperature difference threshold within a temperature zone is set to 2 to 5 times the measurement accuracy of the temperature sensor, and the temperature gradient threshold is set to the global temperature gradient mean plus 1 standard deviation.

4. The multi-parameter coordinated control method for vacuum filtration in a liquid hydrogen spherical tank according to claim 1, characterized in that, The specific method for extracting the physical field coupling factor matrix using the tensor decomposition algorithm is as follows: The spatiotemporal data of the three physical fields are organized into a third-order tensor, with the first dimension being the number of spatial grid points, the second dimension being the number of time series sampling points, and the third dimension being the number of physical field types. Each physical field data is normalized based on its range, scaling it to the range of 0 to 1 to eliminate dimensional differences. The third-order tensor is decomposed into the sum of the outer products of several rank components using the CP tensor decomposition algorithm. Each rank component contains a spatial mode factor, a temporal mode factor, and a physical field coupling factor. The factor matrix is ​​solved iteratively using the alternating least squares method. The third factor matrix is ​​updated while the other two factor matrices are fixed. The process is continued until convergence by minimizing the mean square norm of the reconstruction error. The convergence condition is that the relative change in the objective function value between two adjacent iterations is less than a set threshold.

5. The multi-parameter coordinated control method for vacuum filtration in a liquid hydrogen spherical tank according to claim 4, characterized in that, The specific method for identifying the dominant coupling relationship between physical fields is as follows: Analyze each column vector of the physical field coupling factor matrix. The three elements of the coupling factor vector correspond to the weighting coefficients of the temperature field, vacuum field, and vibration field, respectively. Calculate the absolute value percentage of the weight coefficients of each physical field, and determine the coupling mode type of each rank component based on the weight percentage. If the weight percentage of a certain physical field exceeds 50%, it is determined to be a coupling mode dominated by a single physical field. If the sum of the weights of two physical fields exceeds 80%, it is determined to be a coupling mode dominated by dual physical field coupling. If the weights of the three physical fields are relatively balanced, it is determined to be a coupling mode of three physical field synergistic coupling. The frequency of each type of coupling mode in all rank components is statistically analyzed, and the coupling mode type with the highest frequency is identified as the overall dominant coupling relationship.

6. The multi-parameter coordinated control method for vacuum filtration in a liquid hydrogen spherical tank according to claim 1, characterized in that, The specific method for establishing the mapping function between the physical field coupling factor and the slow time-varying component of the channel is as follows: Obtain the coupling factor vector, mean temperature, and mean vacuum level for each partition, and perform Z-score standardization on the mean temperature and mean vacuum level. Establish a linear regression mapping function, with the inputs being the elements of the physical field coupling factor vector, the mean temperature, the mean vacuum degree, and the interaction term between temperature and vacuum degree, and the output being the slow time-varying component of the channel. A training sample set is constructed using historical monitoring data, and the mean squared error is used as the loss function. The regression coefficient vector is solved by the normal equation to minimize the loss function.

7. The multi-parameter coordinated control method for vacuum filtration in a liquid hydrogen spherical tank according to claim 1, characterized in that, The specific method for establishing the correlation model between vibration spectrum characteristics and fast time-varying components of the channel is as follows: The vibration data were subjected to Fast Fourier Transform to extract the vibration spectrum features, including the amplitude and frequency of each frequency component, and the main frequency components with a cumulative energy ratio of more than 90% were selected. The time correlation of fast time-varying components is established using an autoregressive model. The model input is the channel fast time-varying component sequence at historical time and the sinusoidal function representation of the vibration spectrum characteristics at the current time. The output is the predicted value of the channel fast time-varying component at the current time. The autoregression order is determined based on the autocorrelation function of the fast time-varying components of the channel, and the minimum time lag that makes the autocorrelation coefficient drop below 0.1 is selected. The autoregressive coefficients and vibration influence coefficients are solved by the least squares method, and the phase parameters are determined by fitting a sine function to each vibration frequency component individually.

8. The multi-parameter coordinated control method for vacuum filtration in a liquid hydrogen spherical tank according to claim 1, characterized in that, The specific method for generating partition-specific transmission parameter configurations using the capacity optimization algorithm is as follows: The goal is to maximize channel capacity, which is solved under the conditions of satisfying transmit power constraints and bit error rate constraints. The channel capacity is determined by the channel bandwidth, transmit power, antenna gain, noise power spectral density and path loss coefficient. The path loss coefficient is obtained by converting the slow time-varying state of the channel from decibel units to a linear ratio. For each time period within the time window, the path loss coefficient corresponding to each time period is calculated and a weighted average is performed. The weights are determined based on the quality label matrix of the time period partition. The shallow fading time period is assigned a larger weight, and the deep fading time period is assigned a smaller weight. The optimal scheme is selected from the set of candidate modulation schemes using a discrete search method. For each modulation scheme, the minimum signal-to-noise ratio required to meet the target bit error rate is calculated in reverse according to its theoretical bit error rate formula. Then, the required transmit power is calculated, and the scheme with the largest channel capacity is selected from all feasible modulation schemes. For zones with temperatures below -200°C, a temperature correction factor is introduced to improve the transmit power margin. The temperature correction factor increases as the temperature decreases.

9. The multi-parameter coordinated control method for vacuum filtration in a liquid hydrogen spherical tank according to claim 1, characterized in that, It also includes online update steps: Collect actual transmission performance data for each partition and time period, including success rate, average latency, and total system power consumption; The channel prediction error is calculated and defined as the root mean square error between the predicted channel state and the actual channel state. The regression coefficients of the physical field coupling-channel mapping function are updated using a recursive least squares algorithm. The regression coefficients are updated by multiplying the difference between the actual observed values ​​and the predicted values ​​by the gain matrix. The autoregressive coefficients and vibration influence coefficients of the vibration-fading correlation model are updated using the gradient descent algorithm, and the parameters are adjusted along the negative direction of the error gradient based on the prediction error. A forgetting factor is introduced to assign an exponentially decaying weight to historical data. A smaller forgetting factor is set for partitions with a faster time-varying rate, and a larger forgetting factor is set for partitions with a slower time-varying rate.

10. A multi-parameter coordinated control system for vacuum filtration of a liquid hydrogen spherical tank, used to execute the multi-parameter coordinated control method for vacuum filtration of a liquid hydrogen spherical tank according to any one of claims 1-9, characterized in that, include: The physical field monitoring module is used to acquire temperature, vacuum, and vibration monitoring data inside the liquid hydrogen spherical tank. The spatial partitioning module is used to generate three-dimensional spatial distribution maps of temperature field, vacuum field and vibration field and divide the physical field into partitions; The coupling analysis module is used to perform tensor decomposition on multiphysics spatiotemporal data and extract the physical field coupling factor matrix. The channel prediction module is used to separate the multi-timescale components of channel fading and predict the channel state evolution sequence within the future time window of each partition. The quality label generation module is used to generate a quality label matrix for different time periods. The parameter configuration module is used to generate the transmission parameter configuration and time slot allocation scheme for each partition; The transmission scheduling module is used to schedule the transmission times of multi-parameter monitoring data and collaborative control commands; The online learning module is used to collect actual transmission performance data and update the physical field coupling-channel mapping function.

Images