Online analyzer sampling accuracy self-calibration method and system based on electrical signal feedback

Abstract

The application provides an online analyzer sampling precision self-calibration method and system based on electric signal feedback, relates to the technical field of analyzer sampling calibration, and comprises the following steps: collecting electric signals in a sampling loop, constructing an electric signal original feature field, inputting an electric signal fingerprint tensor space into an electric signal structure evolution model, generating a forward-looking prediction tensor, constructing a hidden state inversion channel, reversely analyzing multi-dimensional features from electric signal structure distortion features, constructing a hidden state coupling parameter field, performing evolution constraint optimization of a multi-constraint coupling relationship surface, and generating a self-calibration control vector. The technical problems that sampling precision drift cannot be sensed in advance and self-calibrated in the sampling process of an online analyzer in the prior art are solved. The technical effects that real-time feature extraction, drift prediction and active calibration are performed by using electric signal feedback, and the sampling precision and reliability are improved are achieved.

Description

Technical Field

[0001] This invention relates to the field of analyzer sampling calibration technology, specifically to an online analyzer sampling accuracy self-calibration method and system based on electrical signal feedback. Background Technology

[0002] In fields such as industrial process analysis, environmental monitoring, and energy and chemical engineering, online analyzers have become crucial devices for continuously acquiring information on the composition and state of media. Their sampling accuracy directly determines the reliability of the analytical results. However, the sampling process of traditional online analyzers heavily relies on the coordinated operation of components such as mechanical valves, sampling pumps, and sampling channels. These actuators are susceptible to wear, contamination, pressure fluctuations, temperature changes, and variations in pipeline flexibility during long-term operation, leading to deviations in sampling volume, sampling sequence, and sampling flow pattern from the calibration conditions, thus causing sampling accuracy drift.

[0003] Current sampling accuracy calibration methods mainly rely on manual periodic inspections or fixed-cycle calibrations, making it difficult to detect subtle dynamic deviations during the sampling process in a timely manner. Although some methods use macroscopic parameters such as flow rate and pressure for monitoring, they cannot capture the coupling characteristics between microscale electrical signal disturbances and the transient response of the actuator, resulting in a lag in the prediction of accuracy drift and the inability to achieve early correction. In addition, the voltage, current, and other electrical signals generated during the analyzer sampling process contain rich transient and microstructural information, such as phase jumps, non-uniform energy accumulation, and high-frequency jitter. These microstructural features are closely related to the microhysteresis of the sampling valve, the elastic instability of the pump, and the channel structure response, but traditional methods have failed to effectively utilize these electrical signal features for accuracy discrimination or self-calibration.

[0004] Existing online analyzers lack real-time identification and prediction of electrical signal microstructure disturbances during the sampling process, resulting in a technical problem where sampling accuracy drift cannot be detected in advance or self-calibrated. Summary of the Invention

[0005] The purpose of this application is to provide a method and system for self-calibrating the sampling accuracy of an online analyzer based on electrical signal feedback, in order to solve the technical problem that existing online analyzers lack real-time identification and prediction of electrical signal microstructure disturbances during the sampling process, resulting in the inability to detect and self-calibrate sampling accuracy drift in advance.

[0006] In view of the above problems, this application provides a method and system for self-calibrating the sampling accuracy of an online analyzer based on electrical signal feedback.

[0007] The first aspect of this application provides a self-calibration method for sampling accuracy of an online analyzer based on electrical signal feedback. The method includes: during the sampling process of the online analyzer, acquiring electrical signals in the sampling loop with high temporal resolution, constructing an original feature field of the electrical signal containing transient electric field perturbation structures, current texture distribution structures, and phase topological jump structures; constructing an electrical signal fingerprint tensor space characterizing the stability of the electrical signal microstructure based on the original feature field; inputting the electrical signal fingerprint tensor space into an electrical signal structure evolution model, and generating a forward-looking prediction tensor of sampling accuracy drift using dynamic trajectory similarity mapping of the electrical signal fingerprint evolution trajectory; constructing a hidden state inversion channel for electrical signal-sampling behavior across physical domains based on the forward-looking prediction tensor, wherein the hidden state inversion channel is used to reverse-analyze the dynamic response characteristics of the sampling valve micro-hysteresis, the instantaneous elastic instability characteristics of the sampling pump, and the compliance perturbation characteristics of the sampling channel structure from the electrical signal structure distortion characteristics, constructing a hidden state coupling parameter field; and performing evolutionary constraint optimization of multi-constraint coupling relationship surfaces based on the hidden state coupling parameter field to generate a self-calibration control vector.

[0008] Optionally, the latent coupling parameter field is projected onto a multi-dimensional constraint space to construct a multi-constraint coupling relationship surface. The multi-dimensional constraint space includes an electrical signal structure stability constraint domain, a sampling action timing constraint domain, and a sampling channel structure response constraint domain. Based on the multi-constraint coupling relationship surface, the offset gradient field and constraint tension distribution field of the latent coupling parameter field in the multi-dimensional constraint space are calculated to form a sampling behavior instability driving weight map. Based on the sampling behavior instability driving weight map, an adaptive constraint weight reconstruction matrix is ​​constructed. The adaptive constraint weight reconstruction matrix is ​​used to dynamically reshape the multi-constraint coupling relationship surface to generate a constraint energy evolution path. The constraint energy evolution path is used to perform evolutionary optimization iterative calculations to generate a self-calibrating control vector.

[0009] Optionally, a constraint energy potential well distribution is constructed based on the constraint energy evolution path, and a multi-scale optimization search initial strategy set is generated based on the migration trajectory of the hidden coupling parameter field in the constraint energy potential well. Based on the multi-scale optimization search initial strategy set, a constraint fluctuation perturbation factor is introduced to perform non-stationary evolution modulation on the multi-constraint coupling relationship surface, forming a dynamic evolution search domain. Cross-scale collaborative convergence operations are performed using the dynamic evolution search domain to identify stable coupling attractor states between the electrical signal structure stability constraint domain, the sampling action timing constraint domain, and the sampling channel response constraint domain. The stable coupling attractor states are mapped to a self-calibration control vector containing timing correction weights, phase perturbation suppression weights, and energy injection modulation weights.

[0010] Optionally, the temporal distribution weights of the electrical signal structural distortion features are constrained and modulated based on the prospective prediction tensor to form a prospective time-constrained feature set; the prospective time-constrained feature set is used to construct a cross-physical domain mapping association structure between the electrical signal features and the physical domain of sampling behavior; in the cross-physical domain mapping association structure, the propagation weights of the hidden state inversion path are dynamically constrained and updated using the prospective prediction tensor to establish a hidden state inversion channel constrained by the prospective drift trend.

[0011] Optionally, the phase distortion topological density parameters, microscale dynamic jitter spectrum parameters, and energy cluster non-uniform aggregation parameters of the original feature field of the electrical signal are extracted, and an electrical signal fingerprint tensor space characterizing the stability of the electrical signal microstructure is constructed based on the extraction results.

[0012] Optionally, the original feature field of the electrical signal is subjected to scale-wise temporal decoupling processing to form a multi-time-resolution electrical signal structure subfield; in each electrical signal structure subfield, a phase topological jump neighborhood graph structure is constructed, the phase abnormal connectivity distribution is calculated, and the phase distortion topological density parameter is constructed; microscale spectral perturbation analysis is performed on the electrical signal structure subfield to extract the non-equilibrium distribution state of jitter energy in the local frequency band and generate microscale dynamic jitter spectral parameters; energy cluster spatial clustering mapping is performed on the electrical signal structure subfield to extract the non-smoothness characterization quantity of the cluster boundary and generate energy cluster non-uniform aggregation parameters.

[0013] Optionally, the electrical signal fingerprint tensor space is decomposed by a time sliding window to construct a multi-scale electrical signal fingerprint evolution sub-trajectory set; a fingerprint evolution reference trajectory library is constructed based on the multi-scale electrical signal fingerprint evolution sub-trajectory set, and a dynamic trajectory similarity matrix between the real-time electrical signal fingerprint evolution trajectory and the fingerprint evolution reference trajectory library is calculated; based on the dynamic trajectory similarity matrix, joint modeling of time extrapolation and structural morphology migration is performed on the electrical signal fingerprint evolution trajectory to generate a forward-looking prediction tensor representing the future drift trend of sampling accuracy.

[0014] Optionally, the self-calibration control vector is time-series recorded to construct a calibration control library; the calibration control library is then time-series encrypted and archived.

[0015] Optionally, the original feature field of the electrical signal is subjected to signal anomaly identification, monitoring and early warning information is configured, and anomaly reporting management of the monitoring and early warning information is implemented.

[0016] A second aspect of this application provides a self-calibration system for sampling accuracy of an online analyzer based on electrical signal feedback. The system includes: a signal acquisition module for acquiring electrical signals in the sampling loop with high time resolution during the online analyzer sampling process, constructing the acquired electrical signals into an original feature field containing transient electric field perturbation structures, current texture distribution structures, and phase topological transition structures; a tensor space construction module for constructing an electrical signal fingerprint tensor space characterizing the stability of the electrical signal microstructure based on the original electrical signal feature field; and a prediction tensor generation module for inputting the electrical signal fingerprint tensor space into the electrical signal structure. An evolutionary model is constructed, which uses the dynamic trajectory similarity mapping of the electrical signal fingerprint evolution trajectory to generate a forward-looking prediction tensor of sampling accuracy drift; a reverse analysis module is used to construct a hidden state inversion channel of electrical signal-sampling behavior across the physical domain based on the forward-looking prediction tensor. The hidden state inversion channel is used to reverse analyze the dynamic response characteristics of the sampling valve micro-hysteresis, the instantaneous elastic instability characteristics of the sampling pump, and the structural compliance perturbation characteristics of the sampling channel from the structural distortion characteristics of the electrical signal, and construct a hidden state coupling parameter field; a control vector generation module is used to perform evolutionary constraint optimization of multi-constraint coupling relationship surfaces based on the hidden state coupling parameter field to generate a self-calibrating control vector.

[0017] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0018] The method provided in this application involves acquiring electrical signals in the sampling loop with high temporal resolution during the sampling process of an online analyzer. The acquired electrical signals are then constructed into an original feature field containing transient electric field perturbation structures, current texture distribution structures, and phase topological jump structures. Based on this original feature field, an electrical signal fingerprint tensor space characterizing the stability of the electrical signal microstructure is constructed. This electrical signal fingerprint tensor space is input into an electrical signal structure evolution model, and a forward-looking prediction tensor of sampling accuracy drift is generated using the dynamic trajectory similarity mapping of the electrical signal fingerprint evolution trajectory. Based on this forward-looking prediction tensor, a hidden state inversion channel for electrical signal-sampling behavior across physical domains is constructed. This hidden state inversion channel is used to reverse-analyze the micro-hysteresis dynamic response characteristics of the sampling valve, the instantaneous elastic instability characteristics of the sampling pump, and the compliance perturbation characteristics of the sampling channel structure from the electrical signal structure distortion characteristics, constructing a hidden state coupling parameter field. Based on this hidden state coupling parameter field, evolutionary constraint optimization of multi-constraint coupling relationship surfaces is performed to generate a self-calibrating control vector. This technology achieves the technical effect of using electrical signal feedback for real-time feature extraction, drift prediction, and active calibration, enabling online adaptive adjustment of sampling accuracy and improving sampling accuracy and reliability.

[0019] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in this application 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 merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating the self-calibration method for sampling accuracy of an online analyzer based on electrical signal feedback provided in this application.

[0022] Figure 2 A schematic diagram of the sampling accuracy self-calibration system for an online analyzer based on electrical signal feedback provided in this application.

[0023] Figure labeling: Signal acquisition module 11, tensor space construction module 12, prediction tensor generation module 13, reverse analysis module 14, control vector generation module 15. Detailed Implementation

[0024] This application provides a method and system for self-calibrating the sampling accuracy of an online analyzer based on electrical signal feedback. It addresses the technical problem of existing online analyzers lacking real-time identification and prediction of electrical signal microstructure disturbances during sampling, leading to the inability to detect and self-calibrate sampling accuracy drift in advance. The method achieves the technical effect of utilizing electrical signal feedback for real-time feature extraction, drift prediction, and active calibration, enabling online adaptive adjustment of sampling accuracy and improving sampling accuracy and reliability.

[0025] The technical solutions of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. It should also be noted that, for ease of description, only the parts related to the present invention are shown in the accompanying drawings, not all of them.

[0026] Example 1, as Figure 1 As shown, this application provides a self-calibration method for the sampling accuracy of an online analyzer based on electrical signal feedback. The self-calibration method for the sampling accuracy of an online analyzer based on electrical signal feedback includes:

[0027] During the sampling process of the online analyzer, the electrical signals in the sampling loop are acquired with high time resolution. The acquired electrical signals are then constructed into an original feature field of electrical signals that includes transient electric field disturbance structure, current texture distribution structure and phase topological jump structure.

[0028] Specifically, during the sampling process of the online analyzer, the electrical signals in the sampling circuit are acquired with high time resolution through a high-speed data acquisition module or sampling card, and the transient fluctuation information of voltage, current and related signals is obtained in real time.

[0029] Based on the acquired electrical signals, an original feature field is constructed. This original feature field refers to the structured representation of the changes in the electrical signals in the temporal and spatial dimensions. The original feature field includes transient electric field perturbation structures, current texture distribution structures, and phase topological jump structures. The transient electric field perturbation structure refers to the brief and irregular changes in parameters such as electric field intensity and direction at the sampling instant. This reflects short-term electric field changes caused by rapid opening and closing of the sampling valve, sudden start-up of the sampling pump, or fluid disturbances during the sampling process. Signal processing algorithms, such as wavelet transform, can be used to perform multi-scale analysis of the electrical signals in the sampling loop, effectively extracting transient components and thus identifying the transient electric field perturbation structure.

[0030] The texture distribution structure of current reflects the distribution characteristics and variation law of current at different times and locations. When current flows in the sampling loop, due to the different parameters such as resistance and inductance of each component in the loop, as well as the influence of sampling action on the current path, the current signal exhibits a unique texture distribution. The Fourier transform algorithm can be used to convert the current signal in the time domain into a frequency domain signal, analyze the amplitude and phase information of different frequency components, and obtain the texture distribution structure of the current signal.

[0031] Phase topology jumps refer to sudden, discontinuous changes in the phase of an electrical signal during sampling. These phase changes are closely related to factors such as the signal's transmission path and load variations. Even minor hysteresis in the sampling valve or slight deformation of the sampling channel can cause phase topology jumps during sampling. By using the Hilbert-Huang transform algorithm, the electrical signal is adaptively decomposed to extract its instantaneous phase information, thereby accurately identifying phase topology jump structures.

[0032] By acquiring electrical signals in the sampling circuit and constructing the original characteristic field of the electrical signals, an accurate and reliable data foundation is provided for subsequent analysis and processing, thereby improving the reliability and accuracy of the online analyzer's sampling accuracy self-calibration.

[0033] Based on the original feature field of the electrical signal, an electrical signal fingerprint tensor space is constructed to characterize the stability of the electrical signal microstructure.

[0034] Furthermore, the phase distortion topological density parameters, microscale dynamic jitter spectrum parameters, and energy cluster non-uniform aggregation parameters of the original feature field of the electrical signal are extracted. Based on the extraction results, an electrical signal fingerprint tensor space characterizing the stability of the electrical signal microstructure is constructed.

[0035] Furthermore, the phase distortion topological density parameters, microscale dynamic jitter spectrum parameters, and energy cluster non-uniform aggregation parameters of the original feature field of the electrical signal are extracted, including: performing multi-scale temporal decoupling processing on the original feature field of the electrical signal to form multi-time resolution electrical signal structure subfields; constructing a phase topological jump neighborhood graph structure in each electrical signal structure subfield, calculating the phase abnormal connectivity distribution, and constructing phase distortion topological density parameters; performing microscale spectral perturbation analysis on the electrical signal structure subfields to extract the non-equilibrium distribution state of jitter energy in local frequency bands and generate microscale dynamic jitter spectrum parameters; and performing energy cluster spatial clustering mapping on the electrical signal structure subfields to extract the non-smoothness characterization of cluster boundaries and generate energy cluster non-uniform aggregation parameters.

[0036] Specifically, the original feature field of the electrical signal is first subjected to scale-wise time-series decoupling processing. Wavelet transform can be used for this process, as it offers multi-resolution analysis capabilities. By scaling and translating the wavelet basis functions, the original feature field is decomposed into sub-bands at different scales, i.e., different frequency ranges. Appropriate wavelet basis functions, such as Daubechies wavelets or Symlet wavelets, are selected to effectively capture different features of the electrical signal. The number of decomposition levels is determined; more levels result in finer time and frequency resolutions, but also increase computational complexity. Using wavelet decomposition algorithms, the original feature field is progressively decomposed into low-frequency approximations and high-frequency details. The low-frequency approximations reflect the long-term trend and low-frequency components of the signal, while the high-frequency details contain transient changes and high-frequency noise. The low-frequency approximations are further decomposed to obtain lower-frequency information at a coarser scale and higher-frequency information at a finer scale. This process is repeated until multiple electrical signal structure sub-fields with different time resolutions are formed.

[0037] For example, Symlet wavelets are used to perform scaled temporal decoupling on the original feature field of an electrical signal, performing a three-level decomposition. After the first level of decomposition, the Symlet wavelet basis functions, through scaling and translation operations, decompose the original feature field of the electrical signal into a low-frequency approximate subfield and a high-frequency detail subfield. The low-frequency approximate subfield mainly contains the fundamental frequency component, while the high-frequency detail subfield contains high-frequency components such as harmonics and noise. The low-frequency approximate subfield of the first level is then decomposed using Symlet wavelets for a second level, resulting in an even lower-frequency approximate subfield and a relatively high-frequency detail subfield. The lower-frequency approximate subfield more clearly reflects the long-term trend of the fundamental frequency, while the relatively high-frequency detail subfield contains some subtle changes in the fundamental frequency. A third level of decomposition is then performed on the low-frequency approximate subfield of the second level, further obtaining a more refined time-resolution subfield. Through this decomposition, the original feature field of the electrical signal is decomposed into multi-time-resolution electrical signal structure subfields.

[0038] For each electrical signal structure subfield in a multi-time-resolution electrical signal structure subfield, the electrical signal sampling points within it are used as nodes in a neighborhood graph. Node connection rules are determined based on phase changes. If the phase change between adjacent sampling points exceeds a preset jump threshold, a connection edge is established between these two nodes, constructing a phase topology jump neighborhood graph structure. This phase topology jump neighborhood graph structure reflects the topological relationship of electrical signal phase jumps. For each node in the phase topology jump neighborhood graph structure, the number of anomalous nodes connected by edges within a certain neighborhood range is counted, representing the number of anomalous phase changes. This number of anomalous nodes is used as the local anomalous connectivity of that node. The certain neighborhood range refers to the area encompassed by a radius of 5 sampling points centered on a node, or by taking 3 connected sampling points before and after the node. The local anomalous connectivity of each node is obtained by traversing all nodes. Statistical analysis methods, such as calculating the mean and variance, are used to obtain the phase anomalous connectivity distribution of the entire neighborhood graph. This phase anomalous connectivity distribution reflects the overall distribution of electrical signal phase jumps. The phase distortion topology density parameter is constructed by normalizing or weighted integration of the phase anomaly connectivity distribution. Normalization is usually achieved by scaling the connectivity values ​​of each node to an interval or standardizing the distribution. For example, min-max normalization is used to map the connectivity to the [0,1] interval, or z-score normalization is used to eliminate the magnitude difference between different subfields, making the connectivity comparable at different scales. Weighted integration introduces weight factors on the normalized connectivity to distinguish the importance of nodes. For example, weights are set according to indicators such as jump amplitude, phase gradient, temporal proximity, or betweenness centrality of nodes in the graph, and weighted summation or weighted averaging is performed on the connectivity. The phase distortion topology density parameter is used to quantify the degree of distortion of the electrical signal phase in the topology and can reflect the abnormal changes of the electrical signal phase in the local region.

[0039] Short-Time Fourier Transform (STFT) or Hilbert-Huang Transform (HHT) are used to perform time-frequency joint analysis of the structural subfields of electrical signals. STFT uses a sliding window function, such as a Hamming window, to truncate local segments of the signal and performs segment-by-segment Fourier transforms to obtain the time-frequency distribution matrix. HHT adaptively decomposes the electrical signal into multiple Intrinsic Mode Functions (IMFs) using Empirical Mode Decomposition (EMD), and then performs a Hilbert transform on each IMF to obtain the instantaneous frequency and amplitude, constructing a time-frequency spectrum. By analyzing the time-frequency spectrum, frequency regions with concentrated energy or specific characteristics are identified, determining the target local frequency band.

[0040] For each time point on the time spectrum, the sum of the energy of all frequency components within the target local frequency band is calculated. For the STFT results, this is obtained by summing the elements within the corresponding frequency band range in the time spectrum matrix. For the HHT results, the energy contribution of each IMF within the target frequency band is calculated based on the instantaneous amplitude and instantaneous frequency, and the energies of all relevant IMFs are summed. A curve showing the energy variation within the target frequency band over time is plotted with time as the x-axis and the calculated energy value as the y-axis. This curve reflects the dynamic change of the electrical signal energy within the target local frequency band. The energy variation curve within the target frequency band over time is treated as a time series data point. The time series data is normalized to have a mean of 0 and a standard deviation of 1 to eliminate the influence of data dimensions. Parameters such as skewness coefficient and kurtosis are statistically analyzed to quantify its non-equilibrium characteristics. For example, a skewness coefficient greater than zero indicates that the energy is shifted to the right, and a kurtosis greater than 3 indicates that the distribution exhibits a peaked state. The frequency range information of the target local frequency band, the energy variation curve over time, and the calculated skewness coefficient and kurtosis parameters are integrated together. The frequency range can be represented by the start frequency and the end frequency. The energy change curve over time can be stored in the form of an array or a data table. The skewness coefficient and kurtosis are used as scalar values ​​to generate microscale dynamic jitter spectrum parameters that include time, frequency, and energy imbalance. The microscale dynamic jitter spectrum parameters reflect the spectral perturbation characteristics of the electrical signal at the microscale.

[0041] Algorithms such as K-means clustering or density clustering are used to perform spatial clustering mapping on the structural subfields of the electrical signal. Energy points in the electrical signal are clustered according to their spatial distribution characteristics, such as features extracted from the time and frequency domains, forming different energy clusters. During the clustering process, the non-smoothness of the cluster boundaries is extracted. This non-smoothness reflects the complexity and irregularity of the energy cluster boundaries. By calculating the distance from points within a cluster to the cluster center, points with a distance greater than a certain threshold are considered boundary points. The local geometric information of the boundary points is used to calculate the non-smoothness of the cluster boundaries. For example, the curvature at the boundary point is calculated; a larger curvature indicates a higher degree of bending at that point, and thus a greater non-smoothness. Alternatively, the rate of change of the distribution density of points in the neighborhood around the boundary point is calculated; a more drastic change in density indicates a greater non-smoothness.

[0042] Energy cluster non-uniform aggregation parameters are generated based on the cluster boundary non-smoothness characterization. For example, based on the relationship between cluster boundary non-smoothness and energy distribution within the cluster, boundary non-smoothness is used as a weighting factor to weight the energy within the cluster, thus obtaining the energy cluster non-uniform aggregation parameters. Specifically, for each cluster, the total energy of all points within it is calculated, and the total energy is weighted according to the boundary non-smoothness characterization. If the boundary non-smoothness is large, it indicates that the cluster boundary is complex and the energy distribution may be more uneven, so a larger weight is assigned; conversely, a smaller weight is assigned. The weighted energy of all clusters is then combined to obtain the energy cluster non-uniform aggregation parameters. These parameters describe the non-uniform spatial distribution of electrical signal energy, reflecting the degree of concentration and dispersion of energy in the electrical signal.

[0043] The extracted phase distortion topology density parameters, microscale dynamic jitter spectrum parameters, and energy cluster non-uniform aggregation parameters reflect the stability characteristics of the electrical signal microstructure from different dimensions. The phase distortion topology density parameters, microscale dynamic jitter spectrum parameters, and energy cluster non-uniform aggregation parameters are used as the dimensions of a tensor, and the value of each dimension corresponds to the normalized value of the corresponding parameter. The electrical signal fingerprint tensor space is constructed, which can comprehensively characterize the stability of the electrical signal microstructure.

[0044] By analyzing the original feature field of electrical signals, an electrical signal fingerprint tensor space can be constructed, which can comprehensively and accurately characterize the microstructure stability of electrical signals. This helps to improve the reliability and accuracy of the online analyzer's sampling accuracy self-calibration scheme, and realize the precise monitoring and prediction of electrical signal states.

[0045] The electrical signal fingerprint tensor space is input into the electrical signal structure evolution model, and the dynamic trajectory similarity mapping of the electrical signal fingerprint evolution trajectory is used to generate a prospective prediction tensor of sampling accuracy drift.

[0046] Furthermore, the electrical signal fingerprint tensor space is input into the electrical signal structure evolution model, including: performing time-sliding window decomposition on the electrical signal fingerprint tensor space to construct a multi-scale electrical signal fingerprint evolution sub-trajectory set; constructing a fingerprint evolution reference trajectory library based on the multi-scale electrical signal fingerprint evolution sub-trajectory set, and calculating the dynamic trajectory similarity matrix between the real-time electrical signal fingerprint evolution trajectory and the fingerprint evolution reference trajectory library; performing joint modeling of time extrapolation and structural morphology migration on the electrical signal fingerprint evolution trajectory based on the dynamic trajectory similarity matrix to generate a forward-looking prediction tensor characterizing the future drift trend of sampling accuracy.

[0047] Specifically, the electrical signal structure evolution model is a composite prediction model that integrates tensor structure learning, temporal dynamics prediction, and trajectory similarity discrimination. The training process of the electrical signal structure evolution model includes three stages: feature trajectory extraction, reference trajectory library generation, and time-series dynamics learning. The specific steps are as follows: Based on a large amount of raw electrical signal data collected during historical operation phases, the model performs raw feature field construction processing to obtain a set of electrical signal fingerprint tensors containing phase distortion features, spectral jitter features, and energy cluster aggregation features. The fingerprint tensor set is then normalized, scale-aligned, and noise-suppressed preprocessing to obtain training samples, ensuring the stability and feature comparability of subsequent training processes. A time-sliding window decomposition is performed on the electrical signal fingerprint tensor set, dividing the electrical signal fingerprint tensor space into multiple time-segment sub-tensors to obtain a set of multi-scale electrical signal fingerprint evolution sub-trajectories used to describe the microstructure evolution state of the electrical signal.

[0048] Based on the fingerprint evolution sub-trajectories of all training samples, a fingerprint evolution reference trajectory library is constructed using trajectory pattern extraction methods based on clustering analysis, such as k-shape clustering, HDBSCAN, or trajectory clustering based on dynamic time warping. This reference trajectory library contains multiple representative evolution patterns and their statistical characteristics, used to characterize the behavioral features of electrical signals under different evolutionary states, such as stable, slow drift, and rapid transitions. A temporal dynamics prediction model is trained using these fingerprint evolution sub-trajectories. This model can employ temporal networks such as LSTM and ARIMA to model complex nonlinear morphological transitions, while a Koopman operator model based on linear observable space mapping is used to model near-linear structural evolution relationships. The temporal dynamics prediction parameter model is optimized by minimizing the loss function, enabling the temporal dynamics prediction model to accurately fit the changes in the fingerprint evolution sub-trajectories along the time axis.

[0049] By fusing the trained dynamic prediction model with the fingerprint evolution reference trajectory library, a complete electrical signal structure evolution model is constructed. During the operation phase, it can generate corresponding fingerprint evolution trajectories based on real-time fingerprint tensor input, and realize the prediction of the microstructure change trend and drift look-ahead determination of the electrical signal through dynamic similarity matching with the reference trajectory library.

[0050] By inputting the electrical signal fingerprint tensor space into the electrical signal structure evolution model, the model first performs a time-sliding window decomposition on the electrical signal fingerprint tensor space, dividing it into multiple sub-tensors of different time periods. These sub-tensors contain information on the stability of the signal microstructure at different time periods, resulting in a multi-scale set of electrical signal fingerprint evolution sub-trajectories. For the real-time acquired electrical signal fingerprint evolution trajectory, the dynamic trajectory similarity between it and each reference trajectory in the fingerprint evolution reference trajectory library is calculated. For example, a dynamic time warping algorithm is used to process time series of different lengths, and the similarity between two time series is measured by calculating the minimum distance between them. All calculated similarity values ​​are combined into a dynamic trajectory similarity matrix. The rows of the dynamic trajectory similarity matrix represent the real-time electrical signal fingerprint evolution trajectory, the columns represent the trajectories in the reference trajectory library, and the matrix parameters represent the similarity values ​​between the two.

[0051] Furthermore, a temporal dynamics prediction model is used to jointly model the evolution trajectory of the electrical signal fingerprint based on a dynamic trajectory similarity matrix, performing time extrapolation and structural morphology transfer. Time extrapolation refers to using time series prediction models, such as LSTM and ARIMA, to predict the future trend of the real-time fingerprint evolution trajectory based on the evolution patterns of historical similar trajectories. Structural morphology transfer modeling focuses on the structural changes of the fingerprint evolution trajectory under the similarity mapping. By analyzing the structural differences between similar trajectories, and based on the Koopman operator model of linear observable space mapping, the migration results of the real-time fingerprint evolution trajectory to possible future states are obtained. The time extrapolation and structural morphology transfer results are fused to generate a forward-looking prediction tensor characterizing the future drift trend of sampling accuracy. The dimensions of this forward-looking prediction tensor include a time dimension and a sampling accuracy-related parameter dimension. The time dimension reflects the predicted time range, while the sampling accuracy-related parameter dimension contains specific information about the sampling accuracy drift, such as the direction and magnitude of the drift.

[0052] By analyzing the forward-looking prediction tensor, the drift of electrical signal sampling accuracy can be predicted in advance, providing a basis for the subsequent adjustment and optimization of the online analyzer's sampling accuracy, and improving the accuracy and reliability of the online analyzer's sampling accuracy self-calibration.

[0053] Based on the aforementioned forward-looking prediction tensor, a hidden state inversion channel for electrical signal-sampling behavior across the physical domain is constructed. The hidden state inversion channel is used to reverse analyze the micro-hysteresis dynamic response characteristics of the sampling valve, the instantaneous elastic instability characteristics of the sampling pump, and the compliance perturbation characteristics of the sampling channel structure from the structural distortion characteristics of the electrical signal, and to construct a hidden state coupled parameter field.

[0054] Furthermore, constructing a hidden state inversion channel across physical domains for electrical signal-sampling behavior based on the prospective prediction tensor includes: constraining and modulating the temporal distribution weights of electrical signal structural distortion features based on the prospective prediction tensor to form a prospective time-constrained feature set; using the prospective time-constrained feature set to construct a cross-physical domain mapping association structure between electrical signal features and the physical domain of sampling behavior; and in the cross-physical domain mapping association structure, using the prospective prediction tensor to dynamically constrain and update the propagation weights of the hidden state inversion path to establish a hidden state inversion channel constrained by the prospective drift trend.

[0055] Specifically, based on the sampling frequency of the electrical signal and the actual analysis requirements, the time series of the electrical signal is divided into several consecutive time periods. For example, if the sampling frequency of the electrical signal is 1000Hz, that is, 1000 data points are collected per second, then every 100 data points are divided into a time period, and every 0.1 seconds constitutes a time segment. The length of the time segment can be adjusted according to specific circumstances. Within each time segment, the structural distortion features of the electrical signal are extracted. These structural distortion features include, but are not limited to, the degree of phase distortion, frequency offset, and amplitude change rate. For example, the degree of phase distortion is measured by calculating the difference between the signal phase and the ideal phase. For the frequency offset, the signal is converted to the frequency domain using Fourier transform, and the difference between the actual frequency and the reference frequency is compared.

[0056] The prospective prediction tensor is analyzed by calculating its statistical characteristics across different time dimensions, such as mean, variance, maximum, and minimum values, to determine the trend and extent of sampling accuracy drift within different time periods. For example, the mean of sampling accuracy drift within each time period is calculated; if the mean is large for a certain time period, it indicates that the sampling accuracy may have changed significantly within that time period.

[0057] Based on the analysis results of the prospective prediction tensor, a temporal distribution weight is calculated for the electrical signal structural distortion features within each time segment. The weight can be calculated using either a drift trend-based weighting or a similarity-based weighting. For example, in drift trend-based weighting, if the prospective prediction tensor shows a significant drift trend in sampling accuracy within a certain time period (e.g., a large drift), the weight of the electrical signal structural distortion features within that time period should be large; conversely, if the drift is small, the weight should be small. The weight of the i-th time period can be calculated by comparing the mean of the sampling accuracy drift within the i-th time period with the sum of the absolute values ​​of the mean of the sampling accuracy drift within all n time periods. In similarity-based weighting, the electrical signal structural distortion features of each time period are compared with the features of the corresponding time period in the prospective prediction tensor. Higher similarity results in a larger weight. Similarity calculations can use methods such as cosine similarity and Euclidean distance. The electrical signal structural distortion feature within each time segment is multiplied by its corresponding time distribution weight to obtain a weighted feature. The weighted features of all time segments are then combined to form a look-ahead time-constrained feature set. This look-ahead time-constrained feature set integrates the electrical signal structural distortion feature and its time distribution weights adjusted based on the future sampling accuracy drift trend. This set more accurately reflects the impact of the electrical signal on sampling behavior at different times, including electrical signal features such as phase distortion trend, spectral jitter energy migration, and energy cluster reconstruction trend. For example, let the electrical signal structural distortion feature vector for the i-th time segment be x. i The weight is w i Then the weighted eigenvector is x. i w i Combine all n weighted feature vectors to form the look-ahead time-constrained feature set X=[x1w1,x2w2……x n w n ].

[0058] The phase distortion trend, spectral jitter energy transfer, and energy cluster reconstruction trend from the look-ahead time constraint feature set are used as the basic inputs for the electrical signal features, and arranged in a time series manner to form a multimodal electrical signal time feature vector set with predictive guidance attributes. Simultaneously, the micro-hysteresis response features of the sampling valve, the transient elastic instability features of the sampling pump, and the compliance disturbance features of the sampling channel structure in the physical domain of sampling behavior are parameterized to form a physical domain state vector set characterizing the dynamic properties of sampling behavior. The micro-hysteresis response features of the sampling valve include opening and closing time delays and response speeds; the transient elastic instability features of the sampling pump include pressure fluctuation amplitude and flow rate change; and the compliance disturbance features of the sampling channel structure include the vibration frequency and damping coefficient of the fluid within the channel.

[0059] A cross-domain correlation basis is constructed between the time feature vector set of multimodal electrical signals and the physical domain state vector set. This correlation basis is implemented by introducing a cross-domain feature projection operator, projecting the multidimensional microstructure features of the electrical signal feature domain onto a response dimension corresponding to the physical domain characteristics, enabling a mappable relationship between the two within a unified feature coordinate system. The projection process extracts key indicators such as trend gradients, local mutation factors, and energy migration directionality from the temporal variations of the electrical signal features, generating cross-domain feature descriptors that correspond to physical domain parameters. Based on these cross-domain feature descriptors, a cross-physical domain mapping correlation matrix is ​​constructed. This matrix describes the response path induced by electrical signal micro-perturbations in the sampling behavior physical domain. By calculating the correlation distribution, response sensitivity distribution, and influence channel strength distribution between the electrical signal feature domain and the physical domain state, a cross-physical domain mapping correlation structure with structured, multidimensional coupling characteristics is formed. The cross-physical domain mapping and correlation structure can map phase anomalies, spectral disturbances, and energy accumulation changes in electrical signals to sampling valve hysteresis, sampling pump elastic response changes, and channel compliance disturbances, respectively, thus establishing a correspondence between the microstructural features of electrical signals and the physical processes of sampling behavior.

[0060] In the cross-physical domain mapping association structure, the propagation weights of the hidden state inversion path are dynamically updated using a forward-looking prediction tensor. The hidden state inversion path refers to the information transmission path that infers the physical domain characteristics of sampling behavior from electrical signal characteristics during the cross-physical domain mapping process. The dynamic constraint update of the propagation weights is based on the prediction of the sampling accuracy drift trend by the forward-looking prediction tensor. If the prediction tensor indicates that the sampling accuracy will change drastically at some point in the future, the propagation weights of the hidden state inversion path will be adjusted accordingly near that point to more accurately capture the impact of this change on the physical domain characteristics of sampling behavior. By establishing a hidden state inversion channel constrained by the forward-looking drift trend, the inversion channel can adjust the information transmission weights in real time based on the prediction information when analyzing hidden states such as sampling valve micro-hysteresis, sampling pump transient instability, and channel compliance disturbances, thereby improving the accuracy of the inversion.

[0061] Using the established latent state inversion channel, the micro-hysteresis dynamic response characteristics of the sampling valve, the instantaneous elastic instability characteristics of the sampling pump, and the compliance disturbance characteristics of the sampling channel structure are analyzed inversely from the structural distortion characteristics of the electrical signal. Specifically, the micro-hysteresis dynamic response characteristics of the sampling valve are extracted by analyzing the distortion characteristics in the electrical signal related to the opening and closing process of the sampling valve, combined with the latent state inversion channel, to obtain information such as the time delay and amplitude changes of the micro-hysteresis dynamic response. The instantaneous elastic instability characteristics of the sampling pump are analyzed by focusing on the distortion characteristics in the electrical signal related to the operating state of the sampling pump, such as pressure fluctuations and flow changes, and the latent state inversion channel is used to analyze the occurrence time and degree of instantaneous elastic instability of the sampling pump. The compliance disturbance characteristics of the sampling channel structure are obtained by starting from the distortion characteristics in the electrical signal reflecting the structural characteristics of the sampling channel, and using the latent state inversion channel, obtaining the frequency, amplitude, and other characteristics of the compliance disturbance of the sampling channel structure.

[0062] The micro-hysteresis dynamic response characteristics of the sampling valve, the instantaneous elastic instability characteristics of the sampling pump, and the compliance disturbance characteristics of the sampling channel structure obtained by analysis are integrated to construct a hidden coupling parameter field. The hidden coupling parameter field is a multi-dimensional parameter space, in which each dimension corresponds to a physical domain feature of the sampling behavior. The hidden coupling parameter field can comprehensively describe the cross-physical domain coupling relationship between the electrical signal and the sampling behavior.

[0063] By performing reverse analysis of the structural distortion characteristics of electrical signals based on the hidden state inversion channel, it is possible to accurately identify the hidden state of key physical behaviors such as the micro-hysteresis dynamic response of the sampling valve, the instantaneous elastic instability of the sampling pump, and the structural compliance disturbance of the sampling channel. This improves the comprehensiveness and observability of the micro-dynamic changes inside the online analyzer sampling, thereby effectively improving the pertinence, accuracy, and real-time performance of the self-calibration strategy, and enhancing the long-term stability and anti-drift capability of the online analyzer sampling accuracy.

[0064] Based on the hidden coupling parameter field, the evolution constraint optimization of the multi-constraint coupling relationship surface is performed to generate a self-calibrating control vector.

[0065] Furthermore, based on the latent coupling parameter field, evolutionary constraint optimization of the multi-constraint coupling relationship surface is performed to generate a self-calibrating control vector, including: projecting the latent coupling parameter field onto a multi-dimensional constraint space to construct a multi-constraint coupling relationship surface, wherein the multi-dimensional constraint space includes an electrical signal structure stability constraint domain, a sampling action timing constraint domain, and a sampling channel structure response constraint domain; calculating the offset gradient field and constraint tension distribution field of the latent coupling parameter field in the multi-dimensional constraint space based on the multi-constraint coupling relationship surface to form a sampling behavior instability driving weight map; constructing a constraint weight adaptive reconstruction matrix based on the sampling behavior instability driving weight map; dynamically reshaping the multi-constraint coupling relationship surface using the constraint weight adaptive reconstruction matrix to generate a constraint energy evolution path; and performing evolutionary optimization iterative calculations using the constraint energy evolution path to generate a self-calibrating control vector.

[0066] Specifically, the implicit coupling parameter field is projected onto a multi-dimensional constraint space, which includes an electrical signal structural stability constraint domain, a sampling action timing constraint domain, and a sampling channel structural response constraint domain. These constraints reflect the tolerance range of the online analyzer under different physical dimensions. The electrical signal structural stability constraint domain primarily focuses on the stability of the electrical signal during transmission and processing, ensuring that the signal is not excessively distorted due to external interference or internal factors, thus affecting the accurate judgment of the sampling equipment's status. The sampling action timing constraint domain focuses on the timing sequence and intervals of the actions of devices such as sampling valves and pumps, ensuring that the sampling process proceeds in an orderly manner according to the predetermined procedure, avoiding sampling failures or inaccurate data due to timing irregularities. The sampling channel structural response constraint domain focuses on the structural changes and response characteristics of the sampling channel during the sampling process, such as the channel's compliance and elasticity, ensuring that the sampling channel can adapt to different sampling requirements and environmental changes. By projecting the latent coupling parameter field onto the multidimensional constraint space, a multi-constraint coupling relationship surface is constructed, which reflects the offset state of the latent coupling parameter field under different constraint conditions.

[0067] For example, the electrical signal structure stability constraint domain is used to limit the allowable deviation range of fingerprint features such as phase distortion, jitter spectrum and energy clusters under normal operating conditions. The allowable deviation of phase distortion topological density can be set to ±0.05 to ±0.2 radian density units of the baseline value, or limited to 5% to 20% of the baseline level in the form of topological connectivity ratio. The microscale dynamic jitter spectrum energy is allowed to vary around the steady-state baseline in the range of ±1 to ±6 dB. The allowable deviation of the non-uniform aggregation degree of energy clusters can be set to fluctuate within the normalized range of 0.1 to 0.5 for the cluster boundary non-smoothness index. The sampling action timing constraint domain is used to limit the timing and consistency of actions of actuators such as sampling valves and sampling pumps during operation. The allowable tolerance for valve opening delay can be set to ±1 to ±10 milliseconds of the nominal action delay. The rise time during pump switching can be allowed to fluctuate within ±2% to ±15% of the nominal rise time. The minimum alignment interval for continuous action triggering can be limited to 0.9 to 1.1 times the baseline interval to ensure that the action chain does not experience clock instability. The channel equivalent impedance of the sampling channel structural response constraint domain is allowed to deviate within ±2% to ±15% of the nominal value, and the channel local compliance is allowed to vary within the normalized range of ±0.05 to ±0.3 of the baseline compliance. The above ranges are only examples; the actual tolerances should be adaptively determined based on the equipment's historical calibration data, long-term operating statistical averages, and their confidence intervals.

[0068] The offset gradient field of the latent coupled parameter field in the multidimensional constraint space is calculated based on the multi-constraint coupling relationship surface. This offset gradient field reflects the rate and trend of change of the latent coupled parameter field in various directions within the multidimensional constraint space. For each point on the multi-constraint coupling relationship surface, its partial derivative with respect to each parameter is calculated, yielding the gradient vector under each constraint condition. By calculating the gradient vectors under all constraint conditions, the changes in the parameter field under different constraints are comprehensively obtained, forming the offset gradient field. For example, in the sampling action timing constraint domain, the partial derivatives with respect to the sampling valve opening time parameter and the sampling pump start-up time parameter are calculated to obtain the gradient vector under that constraint condition, reflecting the sensitivity of the sampling action timing to the parameter field.

[0069] Simultaneously, the constraint tension distribution field of the latent coupling parameter field in the multi-dimensional constraint space is calculated based on the multi-constraint coupling relationship surface. This constraint tension distribution field reflects the constraint strength and range of each constraint condition on the latent coupling parameter field. For each constraint condition, its constraint tension is defined as the magnitude of the gradient vector. The interaction between multiple constraints is analyzed, and the comprehensive constraint tension is calculated. This comprehensive constraint tension is obtained by weighted summation, and the weight of each constraint condition can be set according to its importance and influence. The magnitude of the gradient vector reflects the drastic change at that point; using it as constraint tension can measure the constraint strength of the constraint condition on the parameter field. Weighted summation comprehensively considers the combined effect of multiple constraints, making the comprehensive constraint tension more accurately reflect the overall constraint situation of the parameter field, thus obtaining the constraint tension distribution field. For example, in the sampling channel structural response constraint domain, if the compliance of the channel has a significant impact on the sampling process, a larger weight is assigned to that constraint condition. Based on the offset gradient field and the constraint tension distribution field, the driving weight of each parameter on sampling behavior instability is calculated. For parameter features... Its driving weights , where a i Is it related to constraint C? i The relevant adjustment coefficients can be adjusted according to the actual situation. i To constrain tension, For parameter features Partial derivatives. The driving weights of all parameters in the hidden coupling parameter field are graphically displayed to form a sampling behavior instability driving weight map. The sampling behavior instability driving weight map can be in the form of heat map, bar chart, etc., which can clearly show the degree of driving effect of each parameter on sampling behavior instability.

[0070] The driving weight information corresponding to each constraint is extracted from the sampling behavior instability driving weight graph and used as the initial constraint weights. Based on the initial constraint weights, a dynamic update rule for the matrix elements is formulated according to the real-time feedback information of the sampling behavior and the optimization objective. For example, based on the gradient descent method, the constraint weight matrix elements are updated according to the gradient of the objective function, such as the stability index of the sampling behavior, to construct an adaptive constraint weight reconstruction matrix. This adaptive constraint weight reconstruction matrix is ​​a dynamically adjusted matrix whose elements are updated according to the real-time state and the optimization objective. By dynamically updating the elements of the adaptive constraint weight reconstruction matrix, the weight relationship between each constraint can be adjusted in real time, making the optimization process more flexible and effective.

[0071] A constraint-weighted adaptive reconstruction matrix is ​​used to dynamically reshape a multi-constraint coupling surface. For each point on the surface, the corresponding constraint value is weighted and combined using the constraint-weighted adaptive reconstruction matrix to adjust the influence of each constraint on the parameter field, generating a constraint energy evolution path. This path reflects the change process and energy variation of the hidden coupled parameter field from the initial state to the optimal state under different constraints. Appropriate path search algorithms, such as simulated annealing and genetic algorithms, can be used to find a path from the initial parameter point to the optimal parameter point on the reshaped multi-constraint coupling surface, gradually decreasing the energy function. The energy function is the sum of the squares of the differences between the current constraint and the objective value of each constraint, using the constraint weights as weighting coefficients. During the search process, the parameter value and corresponding energy value of each intermediate state are recorded, forming the constraint energy evolution path.

[0072] Then, the constrained energy evolution path is used to perform evolutionary optimization iterative calculations, continuously adjusting the parameter values ​​of the hidden coupling parameter field so that the hidden coupling parameter field gradually approaches the optimal state, generating a self-calibration control vector. The self-calibration control vector is a set of control parameters that integrates information from multiple aspects such as electrical signals, sampling actions, and sampling channels. It can automatically adjust the control parameters of the sampling equipment according to the actual operating status and needs of the online analyzer, realizing self-calibration and optimization control of the sampling process.

[0073] By dynamically reshaping the surface, constraints that significantly impact sampling stability are highlighted while those with less impact are weakened. This makes the reshaped surface more conducive to optimizing the self-calibration of the online analyzer's sampling accuracy, improving optimization efficiency and calibration specificity, and reducing ineffective control actions. Through evolutionary constraint optimization of multi-constraint coupling surfaces, the online analyzer's adaptability and control accuracy are enhanced, ensuring the stability and accuracy of the sampling process.

[0074] Furthermore, the evolutionary optimization iterative operation is performed using the constrained energy evolution path to generate a self-calibrating control vector, including: constructing a constrained energy potential well distribution based on the constrained energy evolution path; generating a multi-scale optimization search initial strategy set based on the migration trajectory of the hidden coupling parameter field in the constrained energy potential well; introducing a constrained fluctuation perturbation factor on the multi-constrained coupling relationship surface to perform non-stationary evolution modulation, forming a dynamic evolution search domain; performing cross-scale collaborative convergence operation using the dynamic evolution search domain to identify stable coupling attractor states between the electrical signal structure stability constraint domain, the sampling action timing constraint domain, and the sampling channel response constraint domain; and mapping the stable coupling attractor states into a self-calibrating control vector containing timing correction weights, phase perturbation suppression weights, and energy injection modulation weights.

[0075] Specifically, a constrained energy potential well distribution is constructed based on the constrained energy evolution path. This distribution reflects the potential well-like shape of the energy function under different constraints, with different positions corresponding to different energy levels. Regions with lower energy represent the bottom of the potential well, signifying a state closer to the optimization objective. By analyzing the energy values ​​and trends at each point along the constrained energy evolution path, the shape, location, and depth of the constrained energy potential well can be accurately obtained. The migration trajectory of the hidden coupled parameter field within the constrained energy potential well is also considered. This trajectory data includes the values ​​of various parameters in the parameter field at different times, such as the amplitude, frequency, and phase parameters of the signal in electrical signal processing.

[0076] The migration trajectory is analyzed for characteristics across different time periods, and indices such as the average rate of change and magnitude of parameter variation are calculated. For example, the average rate of change of parameters over short periods, such as several sampling periods, is calculated to capture rapid parameter changes, while the magnitude of change over longer periods, such as the entire migration process, is calculated to capture the overall trend of parameter variation. The characteristics of parameter features at different spatial locations are also analyzed. For instance, clustering algorithms are used to group points in the parameter field according to the similarity of parameter values, dividing the field into different regions. The parameter characteristics of each region are analyzed, and the average value and variance of parameters within each region are calculated. Finally, the energy value of each point on the migration trajectory is calculated using the energy function of the constraint energy potential well, and the trend of energy value variation is analyzed to identify regions with lower and higher energy. Regions with lower energy represent the bottom of the potential well, while regions with higher energy represent the edge of the potential well. Regions with lower energy typically correspond to states closer to the optimization objective, while regions with higher energy may have greater optimization potential.

[0077] Feature extraction at different scales provides a comprehensive understanding of the migration trajectory from multiple perspectives, offering a basis for generating targeted initial strategy sets. Temporal scale features reflect the dynamic changes in parameters, spatial scale features reveal the spatial distribution patterns of the parameter field, and energy scale features are directly related to the optimization objective. Based on the rapid parameter changes obtained from temporal scale feature analysis, rapid adjustment strategies are formulated. For example, when parameters change significantly within a short period, a larger search step size is used to quickly track the parameter change trend. By setting a threshold, when the parameter change rate exceeds this threshold, the search step size is increased to n times the original size (n > 1). For stages of slow parameter change, a smaller search step size is used for refined searching. When the parameter change rate is below a certain threshold, the search step size is reduced to m times the original size (m < 1). Dynamically adjusting the search step size according to the rate of parameter change improves search accuracy while maintaining search efficiency, making the search process more adaptable to dynamic parameter changes.

[0078] Simultaneously, based on the clustering results extracted from spatial scale features, different search strategies are formulated for different clustering regions. For regions with relatively concentrated parameter values, a dense search is conducted within that region; for regions with relatively dispersed parameter values, a dispersed search is adopted to expand the search range. For example, in dense search regions, the spacing between search points is set to a smaller value, while in dispersed search regions, the spacing is set to a larger value. Different spatial regions have different parameter characteristics, and using targeted search strategies can fully utilize spatial information and increase the probability of finding the optimal solution. Based on the potential well bottom region extracted from energy scale features, this region is designated as the key search area. A large number of search points are set near the bottom of the potential well, and a small search step size is used for fine-grained searching to find a state closer to the optimal solution. With the center point of the potential well bottom as the reference, search points are distributed at a certain density within a certain range. For the edge region of the potential well, an exploratory search strategy is adopted. For example, starting from a point on the edge of the potential well, a search is conducted in a certain direction and with a certain step size. If the energy is found to decrease, the search continues along that direction; if the energy increases, the search direction is adjusted. Energy is directly related to the optimization objective. Developing different search strategies based on energy levels can guide the search process towards a better direction, improving search efficiency. Integrating the initial strategies generated at the time, spatial, and energy scales forms a multi-scale optimization search initial strategy set. This set includes information such as multiple search starting points and search step sizes at different scales.

[0079] Based on the initial strategy set for multi-scale optimization search, a constraint fluctuation perturbation factor is introduced. This factor is designed to simulate the fluctuations and uncertainties that may exist in the constraints during actual online analyzer sampling, allowing the constraints to vary randomly within a certain range. Then, non-stationary evolution modulation is applied to the multi-constraint coupling surface. That is, the multi-constraint coupling surface is dynamically changed according to a set perturbation law, forming a dynamic evolution search domain. By introducing the perturbation factor for non-stationary evolution modulation, the surface of the multi-constraint coupling relationship is no longer fixed but dynamically adjusts with changes in the perturbation factor. This makes the dynamic evolution search domain closer to the complex environment of reality, allowing the search process to take place in a more realistic and broader space, increasing the probability of finding the optimal solution.

[0080] After obtaining the dynamic evolution search domain, a cross-scale collaborative convergence operation is performed. This operation is an optimization algorithm that comprehensively considers information from different scales, enabling simultaneous search and optimization at multiple scales. Through mutual collaboration and feedback, the search range is gradually narrowed, improving search efficiency. At the time scale, based on the periodic characteristics of the electrical signal and the timing logic of the sampling actions, a time window is set, and the search range is gradually narrowed. The correlation changes in the stability of the electrical signal structure, the timing of sampling actions, and the response of the sampling channel at each time node are analyzed. At the spatial scale, considering factors such as the transmission path of the electrical signal and the spatial layout of the sampling device, search sub-regions are divided, and the interaction of the three constraint domains within different regions is analyzed. At the energy scale, considering energy distribution and loss, the energy transfer and balance relationships between the constraint domains are evaluated. Through the interactive feedback and collaborative processing of multi-scale information, the search direction and parameters are continuously adjusted, and the iterative algorithm is used to gradually approach the stable state. When the coupling relationship indicators between the three constraint domains, such as correlation coefficient and energy transfer efficiency, tend to be stable and meet the preset threshold in multiple consecutive iterations, the stable coupling attractor state between the output electrical signal structure stability constraint domain, the sampling action timing constraint domain, and the sampling channel response constraint domain is obtained. The stable coupling attractor state represents the optimal or near-optimal state of the online analyzer under different sampling constraint conditions.

[0081] The time deviation distribution of key timing nodes in the stable coupled attractor state is analyzed to determine timing correction weights. Nodes with large deviations are assigned larger weights to enhance correction. Regarding phase, the phase perturbation pattern in the stable coupled attractor state is analyzed, and phase perturbation suppression weights are determined based on perturbation amplitude and frequency, with weights increasing in areas of severe perturbation. In the energy dimension, the energy transfer and distribution in the stable coupled attractor state are analyzed, and energy injection modulation weights are set, with weights increased in areas of insufficient energy. Finally, the timing correction weights, phase perturbation suppression weights, and energy injection modulation weights are integrated to obtain a self-calibration control vector. This vector precisely controls and adjusts various key parameters of the online analyzer, ensuring optimal sampling operation. Specifically, the timing correction weights adjust the timing sequence of electrical signals or sampling actions to ensure optimal timing; the phase perturbation suppression weights suppress potential phase fluctuations and improve the stability of the online analyzer's sampling process; and the energy injection modulation weights adjust the injection amount and method according to energy requirements.

[0082] By constructing a constrained energy potential well distribution, generating an initial strategy set, forming a dynamic evolution search domain, identifying stable coupled attractor states, and finally generating a self-calibration control vector, online, adaptive, and forward-looking control of sampling accuracy is achieved. This improves the stability and performance of the online analyzer's sampling accuracy self-calibration, thereby ensuring the reliability and accuracy of the entire sampling and analysis process.

[0083] Furthermore, generating the self-calibration control vector also includes: recording the self-calibration control vector in a time sequence to construct a calibration control library; and managing the calibration control library in a time-series encrypted archive.

[0084] Specifically, the self-calibration control vectors are recorded in time sequence to construct a calibration control library. For example, a linked list or array can be used. Each time a new self-calibration control vector is generated, the specific value of the control vector and the timestamp of its generation are recorded for each node or element, ensuring the accuracy and completeness of the records. Constructing the calibration control library involves organizing and storing the recorded time-series control vector data. Appropriate tables or sets are created based on the characteristics of the control vectors and storage requirements. The control vector data and its time sequence information are inserted into the database, and indexes are established to improve query efficiency and facilitate subsequent retrieval and use of the calibration control data. Encryption algorithms are used to encrypt and archive the data in the calibration control library. For example, a symmetric encryption algorithm (AES) or an asymmetric encryption algorithm (RSA) can be selected based on the data's security level and performance requirements. An encryption key is generated based on the encryption algorithm and used to encrypt the data in the database, ensuring data security during storage.

[0085] Furthermore, the method also includes: identifying signal anomalies in the original feature field of the electrical signal, configuring monitoring and early warning information, and managing the abnormal reporting of monitoring and early warning information.

[0086] Specifically, the original characteristic field of the electrical signal, including voltage, current, and phase characteristics, is compared with the characteristics of normal signals to locate abnormal signal features that deviate from the normal range. Based on preset anomaly level standards, corresponding monitoring and early warning information is configured for different identified anomalies, clarifying the anomaly type and severity. Anomaly reporting management is implemented, and when an anomaly is detected, the early warning information is promptly and accurately sent to relevant personnel through preset channels, such as SMS, email, and audible and visual alarms, ensuring timely response and the effectiveness and accuracy of the online analyzer's sampling process.

[0087] Example 2 is based on the same inventive concept as the online analyzer sampling accuracy self-calibration method based on electrical signal feedback in the previous examples, such as... Figure 2 As shown, this application provides an online analyzer sampling accuracy self-calibration system based on electrical signal feedback, wherein the online analyzer sampling accuracy self-calibration system based on electrical signal feedback includes:

[0088] The signal acquisition module 11 is used to acquire electrical signals in the sampling loop with high time resolution during the sampling process of the online analyzer, and to construct the acquired electrical signals into an original feature field of electrical signals containing transient electric field disturbance structure, current texture distribution structure and phase topological jump structure.

[0089] Tensor space construction module 12 is used to construct an electrical signal fingerprint tensor space characterizing the stability of the electrical signal microstructure based on the original feature field of the electrical signal.

[0090] The prediction tensor generation module 13 is used to input the electrical signal fingerprint tensor space into the electrical signal structure evolution model and generate a prospective prediction tensor for sampling accuracy drift by using the dynamic trajectory similarity mapping of the electrical signal fingerprint evolution trajectory.

[0091] The reverse analysis module 14 is used to construct a hidden state inversion channel for electrical signal-sampling behavior across the physical domain based on the prospective prediction tensor. The hidden state inversion channel is used to reverse analyze the micro-hysteresis dynamic response characteristics of the sampling valve, the instantaneous elastic instability characteristics of the sampling pump, and the compliance disturbance characteristics of the sampling channel structure from the electrical signal structural distortion characteristics, and to construct a hidden state coupled parameter field.

[0092] The control vector generation module 15 is used to perform evolution constraint optimization of the multi-constraint coupling relationship surface based on the hidden coupling parameter field, and generate a self-calibrating control vector.

[0093] Furthermore, the control vector generation module 15 is also used to: project the latent coupling parameter field onto a multi-dimensional constraint space to construct a multi-constraint coupling relationship surface, wherein the multi-dimensional constraint space includes an electrical signal structure stability constraint domain, a sampling action timing constraint domain, and a sampling channel structure response constraint domain; calculate the offset gradient field and constraint tension distribution field of the latent coupling parameter field in the multi-dimensional constraint space based on the multi-constraint coupling relationship surface to form a sampling behavior instability driving weight map; construct a constraint weight adaptive reconstruction matrix based on the sampling behavior instability driving weight map; dynamically reshape the multi-constraint coupling relationship surface using the constraint weight adaptive reconstruction matrix to generate a constraint energy evolution path; and perform evolutionary optimization iterative calculations using the constraint energy evolution path to generate a self-calibrating control vector.

[0094] Furthermore, the control vector generation module 15 is also used to: construct a constraint energy potential well distribution based on the constraint energy evolution path; generate a multi-scale optimization search initial strategy set based on the migration trajectory of the hidden coupling parameter field in the constraint energy potential well; introduce a constraint fluctuation perturbation factor based on the multi-constraint coupling relationship surface to perform non-stationary evolution modulation, forming a dynamic evolution search domain; perform cross-scale collaborative convergence operation using the dynamic evolution search domain to identify the stable coupling attractor state between the electrical signal structure stability constraint domain, the sampling action timing constraint domain, and the sampling channel response constraint domain; and map the stable coupling attractor state into a self-calibration control vector containing timing correction weights, phase perturbation suppression weights, and energy injection modulation weights.

[0095] Furthermore, the reverse analysis module 14 is also used to: constrain and modulate the temporal distribution weights of the electrical signal structural distortion features based on the prospective prediction tensor to form a prospective time-constrained feature set; construct a cross-physical domain mapping association structure between the electrical signal features and the physical domain of sampling behavior using the prospective time-constrained feature set; and dynamically constrain and update the propagation weights of the hidden state inversion path using the prospective prediction tensor in the cross-physical domain mapping association structure to establish a hidden state inversion channel constrained by the prospective drift trend.

[0096] Furthermore, the tensor space construction module 12 is also used to: extract the phase distortion topological density parameters, microscale dynamic jitter spectrum parameters and energy cluster non-uniform aggregation parameters of the original feature field of the electrical signal, and construct an electrical signal fingerprint tensor space that characterizes the stability of the electrical signal microstructure based on the extraction results.

[0097] Furthermore, the tensor space construction module 12 is also used to: perform scaled temporal decoupling processing on the original feature field of the electrical signal to form a multi-time resolution electrical signal structure subfield; in each electrical signal structure subfield, construct a phase topological jump neighborhood graph structure, calculate the phase abnormal connectivity distribution, and construct phase distortion topological density parameters; perform microscale spectral perturbation analysis on the electrical signal structure subfield, extract the non-equilibrium distribution state of jitter energy in the local frequency band, and generate microscale dynamic jitter spectral parameters; perform energy cluster spatial clustering mapping on the electrical signal structure subfield, extract the non-smoothness characterization quantity of the cluster boundary, and generate energy cluster non-uniform aggregation parameters.

[0098] Furthermore, the prediction tensor generation module 13 is also used to: perform time-sliding window decomposition on the electrical signal fingerprint tensor space to construct a multi-scale electrical signal fingerprint evolution sub-trajectory set; construct a fingerprint evolution reference trajectory library based on the multi-scale electrical signal fingerprint evolution sub-trajectory set, and calculate the dynamic trajectory similarity matrix between the real-time electrical signal fingerprint evolution trajectory and the fingerprint evolution reference trajectory library; perform joint modeling of time extrapolation and structural morphology migration on the electrical signal fingerprint evolution trajectory based on the dynamic trajectory similarity matrix to generate a forward-looking prediction tensor characterizing the future drift trend of sampling accuracy.

[0099] Furthermore, the control vector generation module 15 is also used to: record the self-calibration control vector in time sequence to construct a calibration control library; and perform time-series encrypted archiving management of the calibration control library.

[0100] Furthermore, the signal acquisition module 11 is also used to: identify signal anomalies in the original feature field of the electrical signal, configure monitoring and early warning information, and manage the abnormal reporting of monitoring and early warning information.

[0101] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0102] Obviously, those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A self-calibration method for sampling accuracy of an online analyzer based on electrical signal feedback, characterized in that, The method includes: During the sampling process of the online analyzer, the electrical signals in the sampling loop are acquired with high time resolution, and the acquired electrical signals are constructed into an original feature field of electrical signals containing transient electric field disturbance structure, current texture distribution structure and phase topological jump structure. Based on the original feature field of the electrical signal, an electrical signal fingerprint tensor space characterizing the stability of the electrical signal microstructure is constructed; The electrical signal fingerprint tensor space is input into the electrical signal structure evolution model, and a forward-looking prediction tensor of sampling accuracy drift is generated by using the dynamic trajectory similarity mapping of the electrical signal fingerprint evolution trajectory. Based on the aforementioned forward-looking prediction tensor, a hidden state inversion channel for electrical signal-sampling behavior across the physical domain is constructed. The hidden state inversion channel is used to reverse analyze the micro-hysteresis dynamic response characteristics of the sampling valve, the instantaneous elastic instability characteristics of the sampling pump, and the compliance disturbance characteristics of the sampling channel structure from the structural distortion characteristics of the electrical signal, and to construct a hidden state coupled parameter field. Based on the hidden coupling parameter field, the evolution constraint optimization of the multi-constraint coupling relationship surface is performed to generate a self-calibrating control vector; include: The hidden coupling parameter field is projected onto a multi-dimensional constraint space to construct a multi-constraint coupling relationship surface. The multi-dimensional constraint space includes an electrical signal structure stability constraint domain, a sampling action timing constraint domain, and a sampling channel structure response constraint domain. Based on the multi-constraint coupling relationship surface, the offset gradient field and constraint tension distribution field of the hidden coupling parameter field in the multi-dimensional constraint space are calculated to form a sampling behavior instability driving weight map. Based on the sampling behavior instability-driven weight graph, an adaptive reconstruction matrix of constraint weights is constructed. The adaptive reconstruction matrix of constraint weights is used to dynamically reshape the multi-constraint coupling surface to generate a constraint energy evolution path. The constrained energy evolution path is used to perform evolutionary optimization iterative calculations to generate a self-calibrating control vector. include: Construct a constrained energy potential well distribution based on the constrained energy evolution path, and generate a multi-scale optimization search initial strategy set based on the migration trajectory of the hidden coupling parameter field in the constrained energy potential well. Based on the initial strategy set for multi-scale optimization search, a constraint fluctuation perturbation factor is introduced to perform non-stationary evolution modulation on the multi-constraint coupling relationship surface, forming a dynamic evolution search domain. The dynamic evolution search domain is used to perform cross-scale collaborative convergence operations to identify stable coupling attractor states between the electrical signal structure stability constraint domain, the sampling action timing constraint domain, and the sampling channel response constraint domain. The stable coupling attractor state is mapped to a self-calibrating control vector that includes timing correction weights, phase perturbation suppression weights, and energy injection modulation weights.

2. The online analyzer sampling accuracy self-calibration method based on electrical signal feedback as described in claim 1, characterized in that, Based on the aforementioned forward-looking prediction tensor, a hidden state inversion channel for electrical signal-sampling behavior across the physical domain is constructed, including: Based on the aforementioned forward-looking prediction tensor, the temporal distribution weights of the electrical signal structural distortion features are constrained and modulated to form a forward-looking time-constrained feature set. The aforementioned look-ahead time-constrained feature set is used to construct a cross-physical domain mapping and correlation structure between electrical signal features and the physical domain of sampling behavior; In the cross-physical domain mapping association structure, the propagation weights of the hidden state inversion path are dynamically constrained and updated using the forward prediction tensor, thereby establishing a hidden state inversion channel constrained by the forward drift trend.

3. The online analyzer sampling accuracy self-calibration method based on electrical signal feedback as described in claim 1, characterized in that, The phase distortion topological density parameters, microscale dynamic jitter spectrum parameters, and energy cluster non-uniform aggregation parameters of the original feature field of the electrical signal are extracted. Based on the extraction results, an electrical signal fingerprint tensor space characterizing the stability of the electrical signal microstructure is constructed.

4. The online analyzer sampling accuracy self-calibration method based on electrical signal feedback as described in claim 3, characterized in that, The phase distortion topological density parameters, microscale dynamic jitter spectrum parameters, and energy cluster non-uniform aggregation parameters of the original feature field of the electrical signal are extracted, including: The original feature field of the electrical signal is subjected to scale-wise temporal decoupling processing to form a multi-time resolution electrical signal structure subfield; In each electrical signal structure subfield, a phase topological jump neighborhood graph structure is constructed, the phase anomaly connectivity distribution is calculated, and the phase distortion topological density parameters are constructed. Microscale spectral perturbation analysis is performed on the electrical signal structure subfield to extract the non-equilibrium distribution state of jitter energy in the local frequency band and generate microscale dynamic jitter spectrum parameters; Perform energy cluster spatial clustering mapping on the electrical signal structure subfield, extract the non-smoothness characterization of the cluster boundary, and generate the non-uniform aggregation parameters of the energy clusters.

5. The online analyzer sampling accuracy self-calibration method based on electrical signal feedback as described in claim 1, characterized in that, The electrical signal fingerprint tensor space is input into the electrical signal structure evolution model, including: The electrical signal fingerprint tensor space is decomposed by a time sliding window to construct a set of multi-scale electrical signal fingerprint evolution sub-trajectories; A fingerprint evolution reference trajectory library is constructed based on a set of multi-scale electrical signal fingerprint evolution sub-trajectories, and a dynamic trajectory similarity matrix between the real-time electrical signal fingerprint evolution trajectory and the fingerprint evolution reference trajectory library is calculated. Based on the dynamic trajectory similarity matrix, the evolution trajectory of the electrical signal fingerprint is jointly modeled by time extrapolation and structural morphology migration, generating a forward-looking prediction tensor that represents the future drift trend of sampling accuracy.

6. The online analyzer sampling accuracy self-calibration method based on electrical signal feedback as described in claim 1, characterized in that, Generating a self-calibrating control vector also includes: The self-calibration control vector is time-series recorded to construct a calibration control library; The calibration control library is archived and managed using time-series encryption.

7. The online analyzer sampling accuracy self-calibration method based on electrical signal feedback as described in claim 1, characterized in that, The original feature field of the electrical signal is used to identify signal anomalies, configure monitoring and early warning information, and manage the abnormal reporting of monitoring and early warning information.

8. A self-calibration system for sampling accuracy of an online analyzer based on electrical signal feedback, characterized in that, The steps for implementing the online analyzer sampling accuracy self-calibration method based on electrical signal feedback as described in any one of claims 1 to 7 include: The signal acquisition module is used to acquire electrical signals in the sampling loop with high time resolution during the sampling process of the online analyzer, and to construct the acquired electrical signals into an original feature field of electrical signals containing transient electric field disturbance structure, current texture distribution structure and phase topological jump structure. The tensor space construction module is used to construct an electrical signal fingerprint tensor space characterizing the stability of the electrical signal microstructure based on the original feature field of the electrical signal. The prediction tensor generation module is used to input the electrical signal fingerprint tensor space into the electrical signal structure evolution model, and generate a prospective prediction tensor of sampling accuracy drift by using the dynamic trajectory similarity mapping of the electrical signal fingerprint evolution trajectory. The reverse analysis module is used to construct a hidden state inversion channel for electrical signal-sampling behavior across the physical domain based on the prospective prediction tensor. The hidden state inversion channel is used to reverse analyze the micro-hysteresis dynamic response characteristics of the sampling valve, the instantaneous elastic instability characteristics of the sampling pump, and the compliance perturbation characteristics of the sampling channel structure from the electrical signal structural distortion characteristics, and to construct a hidden state coupled parameter field. The control vector generation module is used to perform evolution constraint optimization of the multi-constraint coupling relationship surface based on the hidden coupling parameter field, and generate a self-calibrating control vector.

Images