Trace gas multiple filling repeatability error analysis and compensation method

Abstract

The present application relates to the technical field of error analysis, in particular to a trace gas multiple filling repeatability error analysis and compensation method, comprising the following steps: collecting gas pressure pulsation data and container inner wall adsorption layer thickness change data in the filling stage; inputting the pressure pulsation data into a dynamic coupling analysis module to calculate the correlation degree parameters of adsorption hysteresis effect and pressure fluctuation spectrum; establishing an adsorption-pressure dynamic coupling correction model through a phase space reconstruction algorithm to generate composite correction parameters including valve response advance and temperature compensation slope; inputting the composite correction parameters into the filling actuator to complete the reconstruction of electromagnetic valve control timing and dynamic correction of temperature control curve before the start of the next filling cycle. The present application can identify the deep coupling characteristics causing filling errors and improve the sensitivity and pertinence of error analysis.

Description

Technical Field

[0001] This invention relates to the field of error analysis technology, and in particular to a method for error analysis and compensation of repeated filling of trace gases. Background Technology

[0002] During the filling of high-purity or trace gases, due to the adsorption effect on the inner wall of the container, the gas will be unevenly adsorbed onto the inner surface for a period of time after entering the container, resulting in a deviation between the actual filling amount and the target set amount. Especially in multiple repeated filling tasks, due to the superimposed disturbances of the adsorption layer state, gas pressure fluctuations and temperature control response, problems such as unpredictable filling deviation and difficulty in converging repeatability errors often occur, posing a challenge to high-precision process control.

[0003] In existing technologies, filling control systems typically rely solely on a single-channel pressure sensor to obtain gas flow status and make control corrections through simple threshold judgments or fixed compensation coefficients. This lacks the ability to perceive and respond to the dynamic behavior of adsorption. Furthermore, the formation rate of the adsorption film is affected by temperature, pressure, and container material, exhibiting strong nonlinearity and hysteresis. Traditional methods struggle to effectively quantify the coupling relationship between the adsorption hysteresis process and fluid disturbances, resulting in a lack of adaptability in the control strategy and an inability to fundamentally compensate for errors during the filling process.

[0004] Especially in gas metering scenarios that require high repeatability, such as trace standard gas configuration, trace reactive gas filling, or experimental ultra-clean gas control systems, it is often difficult to achieve cross-cycle stable control through static or empirical compensation strategies alone, and there is also a lack of a complete system for dynamic coupling modeling. Summary of the Invention

[0005] This invention provides a method for analyzing and compensating for repeatability errors in multiple fillings of trace gases. It integrates multi-source real-time sensing, dynamic correlation analysis, and control command reconstruction into a full-process compensation method to improve the consistency and accuracy stability of multiple filling processes.

[0006] A method for analyzing and compensating for repeatability errors in multiple fillings of trace gases, including the following steps:

[0007] S1, deploy pressure sensors and adsorption film thickness detection devices (ultrasonic thickness probes) in the filling pipeline to simultaneously collect gas pressure pulsation data and adsorption layer thickness change data on the inner wall of the container during the filling stage.

[0008] S2, input the pressure pulsation data obtained in S1 into the dynamic coupling analysis module, and calculate the correlation parameter between the adsorption hysteresis effect and the pressure fluctuation spectrum by combining the adsorption layer thickness change data.

[0009] S3, based on the correlation parameters output by S2, establishes an adsorption-pressure dynamic coupling correction model through a phase space reconstruction algorithm, generating composite correction parameters including valve response advance and temperature compensation slope;

[0010] S4 inputs the composite correction parameters of S3 into the filling actuator to complete the reconstruction of the solenoid valve control timing and the dynamic correction of the temperature control curve before the start of the next filling cycle.

[0011] Optionally, S1 specifically includes:

[0012] S11, three pressure sensor arrays are set axially in the filling pipeline, located 5cm downstream of the intake control valve, in the middle section of the pipeline and 3cm near the container interface, respectively.

[0013] S12, the adsorption membrane thickness detection device uses an ultrasonic thickness probe. The ultrasonic thickness probe is embedded in the inner wall of the container at a 45° oblique incident angle, with multiple detection points evenly distributed along the circumference.

[0014] Optionally, the pressure sensor records gas pressure pulsation data at a sampling rate of 1 kHz. The ultrasonic probe scans the thickness of the adsorption layer at a frequency of 100Hz to generate data on the changes in the adsorption layer thickness. ,in:

[0015] This represents the instantaneous gas pressure value collected by the pressure sensor at time t and axial position x in the pipeline. This indicates the thickness of the adsorbed film at time t and the angular position x of the inner wall of the container, relative to the initial reference thickness. The change in quantity.

[0016] Optionally, S2 specifically includes:

[0017] S21, for pressure pulsation data Time-frequency joint analysis is performed, and the energy distribution matrix of the preset frequency band is extracted by wavelet packet decomposition. ;

[0018] S22, Data on the variation of adsorption layer thickness Differential processing was performed to obtain adsorption rate time-series data. ;

[0019] S23, construct the pressure-adsorption co-phase space, and calculate the energy distribution matrix using the window sliding mutual information algorithm. With adsorption rate Time-varying correlation coefficient ;

[0020] S24, perform spectral analysis on the time-varying correlation coefficient and extract the correlation parameter K within the preset characteristic frequency band.

[0021] Optionally, the time-varying correlation coefficient Represented as:

[0022] ;

[0023] in, It is the adsorption-pressure coupling correlation coefficient at time t. It is a local energy matrix standard deviation adsorption rate standard deviation This indicates summation over a sliding window within a specified frequency and position range.

[0024] Optionally, S24 relates to the time-varying correlation coefficient. Perform spectral analysis, calculate its power spectral density, and extract the correlation parameter K within the characteristic frequency band [0.5, 5] Hz, satisfying:

[0025] ;

[0026] in, The Fourier transform of the correlation coefficient is represented by K, which is the energy ratio index of the adsorption-pressure coupling characteristic frequency band and serves as the input correlation parameter for the subsequent dynamic correction model. df is the frequency integral variable.

[0027] Optionally, S3 specifically includes:

[0028] S31, based on Takens' theorem, an adsorption-pressure joint phase space is constructed, and the correlation parameters, pressure fluctuation data, and adsorption layer thickness change rate are combined to form a multi-dimensional joint vector sequence. :

[0029] S32, determine the optimal embedding dimension m, and reconstruct the joint vector sequence into a phase space trajectory matrix M by a time sliding window;

[0030] S33, train the radial basis function neural network to establish the mapping relationship between the phase space trajectory and the compensation parameters:

[0031] ,in, This is the lead time for the solenoid valve response. Let M be the slope of the temperature compensation curve, M be the phase space trajectory matrix, and RBFNN(·) be the radial basis function neural network function mapper. The input is the principal component vector, and the output is the two-dimensional compensation parameters.

[0032] Optionally, the radial basis function neural network function mapper includes an input layer, a hidden layer, and an output layer, wherein:

[0033] In the input layer structure design: the phase space trajectory matrix collected in the historical filling task is used as input data. The principal components covering the main feature energy are extracted by the singular value decomposition method and normalized to form a standardized feature vector for network modeling.

[0034] In the hidden layer structure design, radial basis functions in the form of Gaussian kernels are selected. The center of the basis function is determined by performing cluster analysis on the feature vectors of the training data. At the same time, the expansion scale parameter is adaptively set according to the statistical distribution of samples within the cluster. The number of hidden layer nodes is not fixed, and the node size is optimized by the Akaike information content criterion.

[0035] In the output layer, a dual-output structure is constructed, corresponding to the valve response advance and the target parameters of temperature compensation slope, respectively. The output node receives weighted inputs from all hidden layer nodes, and a linear mapping method is used to generate continuously adjustable compensation quantities.

[0036] Optionally, S4 specifically includes:

[0037] S41, Solenoid valve timing reconfiguration: Based on the valve response advance amount generated in the previous stage. The solenoid valve control pulse sequence is reconstructed, including adjusting the original opening time forward; a high-frequency PWM modulation signal is superimposed during the filling acceleration stage, and the final solenoid valve control instruction set is a discrete pulse sequence.

[0038] S42, Dynamic Correction of Temperature Control Curve: Constructing Temperature Compensation Slope The segmented temperature control function driven sends a ramp command to the temperature control unit in advance during the pre-correction stage before filling, and adopts feedforward-feedback composite control.

[0039] S43, Control Command Fusion: Completed before the start of the next filling cycle via the time-triggered scheduler of the real-time operating system.

[0040] A waveform generator that programs the valve control instruction set into the FPGA;

[0041] The temperature control function parameters are written into the temperature control module of the PLC.

[0042] Optionally, the feedforward-feedback composite control includes:

[0043] Feedforward term: based on The predicted temperature compensation amount;

[0044] Feedback item: Adjust PID parameters based on real-time temperature difference ΔT.

[0045] The beneficial effects of this invention are:

[0046] This invention constructs a dual-channel data stream of adsorption film thickness and pressure pulsation by deploying a pressure sensor and an ultrasonic thickness probe in the filling pipeline. Based on wavelet packet decomposition, time-varying mutual information, and spectral energy ratio, a dynamic coupling analysis model is built, and a quantifiable correlation parameter K is proposed. This effectively captures the microscopic nonlinear relationship between adsorption hysteresis and fluid disturbance. Compared with the traditional single-variable model that only relies on pressure fluctuations, this invention can accurately identify the deep coupling characteristics that cause filling errors, and improve the sensitivity and pertinence of error analysis.

[0047] This invention utilizes Takens' theorem to construct a jointly embedded adsorption-pressure phase space and uses a radial basis function neural network to realize the mapping reasoning from historical trajectories to compensation parameters. The output includes two-dimensional control parameters including the solenoid valve response advance and temperature compensation slope. Through multi-level mechanisms such as dimensionality reduction and dynamic node optimization and screening, the generalization ability of the model is improved. It has the ability to adaptively identify the source of control error and generate accurate correction parameters in real time for different filling environments, breaking through the limitation of traditional static correction mechanisms within the cycle that cannot cope with fluctuation disturbances.

[0048] This invention designs a hard real-time compensation execution architecture, based on FPGA waveform generator and PLC temperature control module respectively loading compensation instructions, integrating Δt-driven pre-adjustment pulse optimization and α-driven piecewise integral temperature control curve, and adopting a feedforward-feedback composite control strategy to improve the system's synergistic compensation effect for hysteresis adsorption, boundary disturbance and thermal inertia response, effectively avoid the accumulation of repetitive errors, and improve the long-term consistency and control accuracy of the filling system. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention;

[0051] Figure 2 This is a schematic diagram illustrating the filling process according to an embodiment of the present invention. Detailed Implementation

[0052] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should also be noted that, to make the embodiments more comprehensive, the following embodiments are the best and preferred embodiments, and those skilled in the art can use other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0053] like Figures 1-2 As shown, the method for analyzing and compensating for repeatability errors in multiple fillings of trace gases includes the following steps:

[0054] S1, deploy pressure sensors and adsorption film thickness detection devices (ultrasonic thickness probes) in the filling pipeline to simultaneously collect gas pressure pulsation data and adsorption layer thickness change data on the inner wall of the container during the filling stage.

[0055] S2, input the pressure pulsation data obtained in S1 into the dynamic coupling analysis module, and calculate the correlation parameter between the adsorption hysteresis effect and the pressure fluctuation spectrum by combining the adsorption layer thickness change data.

[0056] S3, based on the correlation parameters output by S2, establishes an adsorption-pressure dynamic coupling correction model through a phase space reconstruction algorithm, generating composite correction parameters including valve response advance and temperature compensation slope;

[0057] S4 inputs the composite correction parameters of S3 into the filling actuator to complete the reconstruction of the solenoid valve control timing and the dynamic correction of the temperature control curve before the start of the next filling cycle.

[0058] S1 specifically includes:

[0059] S11, three pressure sensor arrays are set axially in the filling pipeline, located 5cm downstream of the intake control valve, in the middle section of the pipeline and 3cm near the container interface, respectively.

[0060] S12, the adsorption membrane thickness detection device uses an ultrasonic thickness probe. The ultrasonic thickness probe is embedded in the inner wall of the container at a 45° oblique incident angle, with multiple detection points evenly distributed along the circumference.

[0061] The pressure sensor records gas pressure pulsation data at a sampling rate of 1 kHz. The ultrasonic probe scans the thickness of the adsorption layer at a frequency of 100Hz to generate data on the changes in the adsorption layer thickness. ,in:

[0062] This represents the instantaneous gas pressure value collected by the pressure sensor at time t and axial position x in the pipeline. This indicates the thickness of the adsorbed film at time t and the angular position x of the inner wall of the container, relative to the initial reference thickness. The change in quantity.

[0063] S2 specifically includes:

[0064] S21, for pressure pulsation data Joint time-frequency analysis was performed, and the energy distribution matrix of the 0.1-100Hz frequency band was extracted using wavelet packet decomposition, yielding:

[0065] ;

[0066] in, This represents the local pressure energy density at frequency f and location x. This represents the wavelet packet decomposition operator based on the db6 wavelet basis. This represents the original pressure signal at time t and position x.

[0067] S22, Data on the variation of adsorption layer thickness By performing time differentiation processing, the adsorption rate time series data are obtained:

[0068] ;

[0069] in, This represents the instantaneous rate of change in the thickness of the adsorption layer. This represents the change in the thickness of the adsorption layer.

[0070] S23, construct the pressure-adsorption combined phase space, and use the windowed sliding mutual information algorithm to calculate the energy distribution matrix within the sliding time window. With adsorption rate Time-varying correlation coefficient ,satisfy:

[0071] ;

[0072] in, It is the adsorption-pressure coupling correlation coefficient at time t. It is a local energy matrix standard deviation The rate of change of adsorption layer thickness standard deviation This represents the summation over a sliding window within a specified frequency and location range, with the window width set to 1 / 5 of the characteristic time constant of the adsorption process, and the value range being... -200ms.

[0073] S24, regarding the correlation coefficient Perform spectral analysis, calculate its power spectral density, and extract the correlation parameter K within the characteristic frequency band [0.5, 5] Hz, satisfying:

[0074] ;

[0075] in, The Fourier transform of the correlation coefficient is represented by K, which is the energy ratio index of the adsorption-pressure coupling characteristic frequency band and serves as the input correlation parameter for the subsequent dynamic correction model. df is the frequency integral variable.

[0076] The above expression for the time-varying correlation coefficient is used to calculate the frequency domain energy distribution of the pressure signal at a certain time t. With the rate of change of adsorption layer thickness Standardized correlation coefficient between Specifically, it performs the following operations:

[0077] 1. At a given time t, calculate the weighted product of the signals at all spatial locations x and frequency components f, where, This represents the energy of pressure fluctuations at that location and frequency. This represents the rate of change of the adsorption layer thickness at position x.

[0078] 2. Divide the above weighted result by two standard deviations. and , express The overall degree of fluctuation, This indicates the overall fluctuation in the rate of change of the adsorption layer thickness.

[0079] 3. Obtain a standardized coefficient. Its value ranges between positive and negative. The larger the value, the more synchronous and coupled the pressure spectrum characteristics and adsorption change trend are at that moment.

[0080] The spectral characteristics of pressure fluctuations directly affect the adsorption / desorption process (such as boundary layer disturbances and changes in molecular collision frequency). The two are correlated, but this correlation is not a simple linear response. Therefore, a time-varying correlation coefficient is used.

[0081] The degree of "coupling in the same direction" is reflected in the molecular part through weighted inner products;

[0082] The denominator is normalized by standard deviation to eliminate problems such as inconsistency of dimensions and inaccurate data range;

[0083] Essentially, it can be viewed as a standardized cross-correlation indicator, which facilitates cross-time sliding window comparisons.

[0084] This coefficient forms the basis for subsequent spectral integral analysis and dynamic modeling. It preserves the time-domain variation trend while introducing frequency-domain energy factors, demonstrating the advantages of time-frequency coupling analysis.

[0085] S3 specifically includes:

[0086] S31, based on Takens' embedding theorem, constructs an adsorption-pressure joint phase space, and integrates the correlation parameter, pressure fluctuation data, and adsorption layer thickness change rate according to the delay time. Embeddings form a joint vector sequence :

[0087] ;

[0088] in, It is the delay time, determined by the first minimum value of the mutual information function (value...). =8-15ms), where n is the time series index. It is the adsorption-pressure coupling correlation parameter for time t. It is the pressure pulsation data over time t. The rate of change of the adsorption layer thickness is denoted as .

[0089] S32, According to Takens' theorem, if the true dimension of a dynamical system is d, then its state can be reconstructed using an embedding dimension of at least m > 2d by constructing an embedding dimension of at least m > 2d from a one-dimensional time series, ensuring that the dynamical trajectories do not overlap (preserving topological equivalence). Therefore, the embedding dimension m must be large enough to unfold all degrees of freedom of the system state, but not too large to avoid introducing noise dimensions. The optimal embedding dimension is determined to be m = 6. The joint vector... Reconstructing the phase space trajectory matrix using a time sliding window:

[0090] ;

[0091] Where M is the reconstructed phase space trajectory matrix, and N is the total number of samples in the time series.

[0092] S33, d. Train a radial basis function neural network (RBFNN) to establish a nonlinear mapping relationship from the phase space trajectory to the compensation parameters: ,in, This is the lead time for the solenoid valve response. The slope of the temperature compensation curve is represented by RBFNN(·), which denotes a radial basis function neural network function mapper. The input is a principal component vector, and the output is two-dimensional compensation parameters. The specific scheme for the radial basis function neural network function mapper is as follows:

[0093] 1. Input layer design:

[0094] Input data preprocessing: The phase space trajectory matrix M is processed according to the following steps:

[0095] Singular Value Decomposition (SVD) Dimensionality Reduction: Perform SVD decomposition on M: The first three principal components (covering over 95% of the energy) are retained to generate feature vectors. .

[0096] Normalization: Max-min normalization is performed on the feature vectors to unify the value range of each dimension to [0,1].

[0097] 2. Hidden Layer Architecture: Radial Basis Function Configuration:

[0098] Basis function form: using a Gaussian kernel function: ,in For the j-th cluster center, For the corresponding width parameter;

[0099] Center point selection:

[0100] a. Perform K-means clustering (dynamic cluster number optimization) on the feature vectors of the training set, using the cluster centers as... ;

[0101] b. Width of basis functions , This represents the number of samples within the cluster.

[0102] Dynamic node optimization: Dynamically adjusting the number of hidden layer nodes J based on the AIC criterion. , where p is the number of samples, and the optimal J is determined by minimizing the AIC value (typical value 12-18);

[0103] 3. Output layer mapping: Two-parameter output mechanism:

[0104] Weighted Connections: A fully connected layer is used from the hidden layer to the output layer. There are two output nodes, corresponding to:

[0105] ;

[0106] ;

[0107] in, These are the weighting coefficients. As a bias term, the output layer uses a linear activation function.

[0108] 4. Training and Optimization: Supervised learning strategy is adopted.

[0109] Training data: The input is the phase space trajectory matrix M of historical filling tasks, and the output is the measured optimal compensation parameters. Loss function: ,in, This represents the weighting coefficient for the temperature compensation term; The loss function value is used to balance the two objectives of time and temperature, and the parameters are updated based on the optimizer. Regularization coefficient To prevent overfitting, the validation set loss decreases for 5 consecutive iterations. Training will be terminated at that time.

[0110] S4 specifically includes:

[0111] S41, Solenoid valve timing reconfiguration: Based on the valve response advance amount generated in the previous stage. The solenoid valve control pulse sequence is reconstructed, including adjusting the original opening time forward; a high-frequency PWM modulation signal is superimposed during the filling acceleration stage, and the final solenoid valve control instruction set is a discrete pulse sequence.

[0112] S42, Dynamic Correction of Temperature Control Curve: Constructing Temperature Compensation Slope The segmented temperature control function driven sends a ramp command to the temperature control unit in advance during the pre-correction stage before filling, and adopts feedforward-feedback composite control.

[0113] S43, Control Command Fusion: Completed before the start of the next filling cycle via the time-triggered scheduler of the real-time operating system.

[0114] A waveform generator that programs the valve control instruction set into the FPGA;

[0115] The temperature control function parameters are written into the temperature control module of the PLC.

[0116] Feedforward-feedback composite control includes:

[0117] Feedforward term: based on The predicted temperature compensation amount;

[0118] Feedback item: Adjust PID parameters based on real-time temperature difference ΔT.

[0119] S4 specifically includes:

[0120] S41, Solenoid valve timing reconfiguration: Based on the valve response advance amount generated in the previous stage. The original solenoid valve control timing is reconstructed, specifically including:

[0121] 1. Set the original solenoid valve opening time. Adjust forward to: ;

[0122] 2. During the filling acceleration phase, i.e. from Within a continuous 50ms period, a high-frequency PWM (Pulse Width Modulation) signal is superimposed, and its duty cycle increases according to the following pattern: ,in, The duty cycle of the PWM signal. The characteristic response time for material adsorption;

[0123] 3. The final generated solenoid valve control instruction set is a discrete pulse sequence, in the form of:

[0124] .

[0125] S42, Dynamic correction of temperature control curve:

[0126] Construct temperature-compensated slope Driven dynamic temperature control function :

[0127] ,when ;

[0128] in, The target temperature for the temperature control unit. To set the initial temperature, For real-time pressure signals, The pressure threshold for activating integral compensation (value 0.3 MPa).

[0129] During the pre-filling correction phase, a temperature ramp command is sent to the temperature control unit 10-15ms in advance, with the control ramp set as follows: ,in, This is the temperature calibration coefficient, obtained through thermodynamic calibration experiments, with a default value of 0.6-1.0.

[0130] A feedforward-feedback composite control mechanism is adopted:

[0131] Feedforward: Use Predicting temperature shift trends;

[0132] Feedback item: Based on real-time temperature difference Adaptively adjust PID parameters.

[0133] S43, 5ms before the start of the next filling cycle, completes the following operations through the time-triggered scheduler of the real-time operating system (RTOS):

[0134] The valve control instruction set is programmed into the waveform generator in the FPGA controller;

[0135] Write the temperature control function parameters into the temperature regulation module of the PLC controller.

[0136] The valve control instruction set is a set of pre-generated control commands that include the opening degree that the solenoid valve should be at each moment. The waveform generator refers to the PWM / pulse control module inside the FPGA, which is responsible for generating the required electrical signal waveform according to the precise time axis to drive the solenoid valve. The preset instructions or parameters are written into the registers or storage units of the control hardware so that they can be called at runtime.

[0137] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0138] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for analyzing and compensating for repeatability errors in multiple fillings of trace gases, characterized in that, Includes the following steps: S1, a pressure sensor and an adsorption film thickness detection device are deployed in the filling pipeline to simultaneously collect gas pressure pulsation data and adsorption layer thickness change data on the inner wall of the container during the filling stage. The adsorption film thickness detection device uses an ultrasonic thickness probe, and the pressure sensor records gas pressure pulsation data at a sampling rate of 1 kHz. The ultrasonic probe scans the thickness of the adsorption layer at a frequency of 100Hz to generate data on the changes in the adsorption layer thickness. ,in, Indicates time and pipeline axial position The instantaneous gas pressure value collected by the pressure sensor at that location. Indicates time Angular position relative to the inner wall of the container The corresponding adsorption film thickness relative to the reference initial thickness The change in; S2, input the pressure pulsation data obtained in S1 into the dynamic coupling analysis module, and combine it with the adsorption layer thickness change data to calculate the correlation parameter between the adsorption hysteresis effect and the pressure fluctuation spectrum; specifically including: S21, for pressure pulsation data Time-frequency joint analysis is performed, and the energy distribution matrix of the preset frequency band is extracted by wavelet packet decomposition. ; S22, Data on the variation of adsorption layer thickness Differential processing was performed to obtain adsorption rate time-series data. ; S23, construct the pressure-adsorption co-phase space and calculate the energy distribution matrix using the window sliding mutual information algorithm. With adsorption rate Time-varying correlation coefficient ; S24, Perform spectral analysis on the time-varying correlation coefficient and extract the correlation parameter K within the preset characteristic frequency band; S3, based on the correlation parameters output by S2, establishes an adsorption-pressure dynamic coupling correction model through a phase space reconstruction algorithm, generating composite correction parameters including valve response advance and temperature compensation slope; specifically including: S31, based on Takens' theorem, an adsorption-pressure joint phase space is constructed, and the correlation parameters, pressure fluctuation data, and adsorption layer thickness change rate are combined to form a multi-dimensional joint vector sequence. : S32, determine the optimal embedding dimension m, and reconstruct the joint vector sequence into a phase space trajectory matrix using a time sliding window. ; S33, train the radial basis function neural network to establish the mapping relationship between the phase space trajectory and the compensation parameters: ,in, This is the response lead time for the solenoid valve. The slope of the temperature compensation curve. Let be the phase space trajectory matrix, and RBFNN(·) be the radial basis function neural network function mapper. The input is the principal component vector, and the output is the two-dimensional compensation parameters. S4 inputs the composite correction parameters from S3 into the filling actuator to reconstruct the solenoid valve control timing and dynamically correct the temperature control curve before the start of the next filling cycle; specifically including: S41, Solenoid valve timing reconfiguration: Based on the valve response advance amount generated in the previous stage. The solenoid valve control pulse sequence is reconstructed, including adjusting the original opening time forward; a high-frequency PWM modulation signal is superimposed during the filling acceleration stage, and the final solenoid valve control instruction set is a discrete pulse sequence. S42, Dynamic Correction of Temperature Control Curve: Constructing Temperature Compensation Slope The segmented temperature control function driven sends a ramp command to the temperature control unit in advance during the pre-correction stage before filling, and adopts feedforward-feedback composite control. S43, Control Command Fusion: Completed before the start of the next filling cycle via the time-triggered scheduler of the real-time operating system. A waveform generator that programs the valve control instruction set into the FPGA; The temperature control function parameters are written into the temperature control module of the PLC.

2. The method for repeatability error analysis and compensation of trace gas multiple fillings according to claim 1, characterized in that, S1 specifically includes: S11, three pressure sensor arrays are axially arranged in the filling pipeline, located downstream of the intake control valve, in the middle section of the pipeline and near the container interface, respectively. S12, the ultrasonic thickness probe is embedded in the inner wall of the container at a 45° oblique incident angle, with multiple detection points evenly distributed along the circumference.

3. The method for repeatability error analysis and compensation of trace gas multiple fillings according to claim 1, characterized in that, The time-varying correlation coefficient Represented as: ； in, It is a moment The adsorption-pressure coupling correlation coefficient, It is a local energy matrix standard deviation adsorption rate standard deviation This indicates summation over a sliding window within a specified frequency and position range.

4. The method for repeatability error analysis and compensation of trace gas multiple fillings according to claim 3, characterized in that, S24 relates to the time-varying correlation coefficient Spectral analysis was performed to calculate the power spectral density and extract the correlation parameters within the characteristic frequency band [0.5, 5] Hz. ,satisfy: ； in, The Fourier transform of the correlation coefficient The energy proportion of the adsorption-pressure coupling characteristic frequency band is used as an input correlation parameter for the subsequent dynamic correction model. For frequency integral variables.

5. The method for repeatability error analysis and compensation of trace gas multiple fillings according to claim 1, characterized in that, The radial basis function neural network function mapper includes an input layer, a hidden layer, and an output layer, wherein: In the input layer structure design: the phase space trajectory matrix collected in the historical filling task is used as input data. The principal components covering the main feature energy are extracted by the singular value decomposition method and normalized to form a standardized feature vector for network modeling. In the hidden layer structure design, radial basis functions in the form of Gaussian kernels are selected. The center of the basis function is determined by performing cluster analysis on the feature vectors of the training data. At the same time, the expansion scale parameter is adaptively set according to the statistical distribution of samples within the cluster. The number of hidden layer nodes is not fixed, and the node size is optimized by the Akaike information content criterion. In the output layer, a dual-output structure is constructed, corresponding to the valve response advance and the target parameters of temperature compensation slope, respectively. The output node receives weighted inputs from all hidden layer nodes, and a linear mapping method is used to generate continuously adjustable compensation quantities.

6. The method for repeatability error analysis and compensation of trace gas multiple fillings according to claim 1, characterized in that, The feedforward-feedback composite control includes: Feedforward term: based on The predicted temperature compensation amount; Feedback item: Adjust PID parameters based on real-time temperature difference ΔT.

Images