A method and system for processing data analysis in a chromatography system

Abstract

The present application relates to the technical field of data analysis, in particular to a data analysis processing method and system for processing a chromatographic system, comprising the following steps: collecting a chromatographic matrix, calculating curvature variation, constructing a segmented function, mapping to a manifold space, screening a drift retention trunk, calculating a difference and a variation rate, cutting a high variation area, constructing a three-dimensional tensor, performing sparse decomposition to obtain an independent peak component, locally fitting to extract a peak shape moment feature, reconstructing a peak shape according to a fitting weight to generate an identification identifier, in the present application, a manifold space is constructed based on curvature mutation point division, a signal trunk is identified using a gradient variation rate and drift noise is screened out, a high gradient section of a signal response surface is constructed in combination with a retention time difference and a variation rate, potential overlapping peak areas in a complex structure are effectively located, a local subsection is fitted by a least square method and a peak shape moment feature is extracted, weighted reconstruction is performed in combination with global information, and the refinement expression of peak shape information and the enhancement of recognition accuracy are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data analysis, in particular to a data analysis processing method and system for processing in a chromatographic system. BACKGROUND

[0002] The technical field of data analysis involves various methods and processes for collecting, organizing, modeling, analyzing, and interpreting raw data. Its core tasks include data acquisition, data cleaning, data modeling, statistical analysis, pattern recognition, and visualization. This technology is widely used in biomedicine, chemical engineering, financial statistics, marketing, and other industries, aiming to extract useful information from massive data and assist decision-making. In experimental analysis scenarios, data analysis often combines professional instruments such as mass spectrometers and chromatographs to process collected continuous signals or detect peak shapes, and by analyzing their time series or frequency distribution, qualitative or quantitative information about specific substances can be obtained.

[0003] Among them, the traditional data analysis processing method for processing in a chromatographic system refers to a processing method based on the original detection signal output by the chromatograph. This method usually includes specific operation steps such as baseline calibration, peak identification, peak area calculation, and retention time comparison in the chromatogram. In the traditional method, baseline calibration mainly uses fixed threshold method or moving average method to suppress background noise, peak identification generally uses first-order derivative change and window sliding algorithm to mark the starting and ending points of the peak, peak area calculation usually uses truncation integral method or fitting curve integral method, and retention time comparison relies on fixed standard substances or existing databases for comparison and identification. These methods can achieve data analysis to some extent, but their processing flow mainly relies on static rules, and their adaptability is limited under conditions such as complex baseline fluctuation and multiple peak overlap. SUMMARY

[0004] To solve the technical problems existing in the prior art, the present application provides a data analysis processing method for processing in a chromatographic system, comprising the following steps:

[0005] S1: Collecting the original chromatographic data matrix, calculating the first-order and second-order curvature gradient change rate to construct a segmented function, mapping the chromatographic signal to the manifold space sub-domain defined by the curvature mutation point, calculating the slope direction stability index in the manifold space sub-domain, filtering the drift term and retaining the signal trunk, and generating a baseline corrected chromatographic signal;

[0006] S2: Calculating the first-order difference value and the second-order variation rate for the baseline corrected chromatographic signal, selecting the signal segment in the high threshold interval according to the second-order variation rate, cutting the original chromatographic data matrix according to the signal segment, and constructing a three-dimensional tensor structure section;

[0007] S3: performing sparse decomposition operation based on the three-dimensional tensor structure section, resolving tensor element separation independent response vector, deconstructing single filamentous spectrum response, and obtaining independent chromatographic peak component;

[0008] S4: dividing the independent chromatographic peak component into local sub-sections, performing least square fitting on the local sub-sections, constructing local fitting curve, extracting peak shape moment feature from the local fitting curve, and composing local peak shape feature set;

[0009] S5: calculating weight coefficient according to fitting quality of the local sub-sections, adjusting local feature combined with global peak shape information, reconstructing peak shape feature, and generating chromatographic peak recognition identifier.

[0010] As a further scheme of the present application, the baseline corrected chromatographic signal includes pure retention time sequence, corrected response intensity vector and residual background noise term, the three-dimensional tensor structure section includes sample sequence dimension index, clipping time dimension coordinate and section response numerical block, the independent chromatographic peak component includes single filamentous spectrum, independent response abundance value and component resolution mixing weight coefficient, the local peak shape feature set includes segmented fitting curve parameter, local morphological skewness value and section residual statistics, and the chromatographic peak recognition identifier includes comprehensive peak shape fingerprint code, reconstructed confidence score and category attribution label.

[0011] As a further scheme of the present application, the specific steps of S1 are:

[0012] S101: collecting original chromatographic data matrix, performing discrete differential operation on retention time sequence in the original chromatographic data matrix, obtaining first-order and second-order curvature gradient change rate, constructing segmented function according to mutation point position of sign transition of the curvature gradient change rate value, mapping chromatographic signal sequence to corresponding data paragraph in manifold space according to interval definition rule of the segmented function, and generating manifold space mapping sub-domain;

[0013] S102: performing linear regression analysis on each data paragraph in the manifold space mapping sub-domain, extracting slope value of baseline variation trend term, counting sign flip frequency and amplitude fluctuation variance of slope value between adjacent sampling points, quantifying direction consistency degree of baseline trend in time dimension in the sub-domain by calculating statistical distribution characteristics of slope direction in continuous time window, and generating slope direction stability index;

[0014] S103: performing numerical comparison between the slope direction stability index and preset baseline drift judgment threshold, identifying and separating linear increasing and decreasing trend drift term with convergent index value, eliminating the drift term from the manifold space mapping sub-domain, reserving signal main part with peak shape local extreme value characteristics, performing time sequence splicing and smooth reconstruction on the signal main part processed in each sub-domain, and generating baseline corrected chromatographic signal.

[0015] As a further scheme of the present application, the specific steps of S2 are:

[0016] S201: Based on the baseline-corrected chromatographic signal, a first-order difference operation is performed on the response intensity values along the retention time dimension to obtain a first-order gradient sequence representing the signal instantaneous slope within a single sample, the first-order gradient sequences corresponding to adjacent sample rows are extracted and the change rate ratio of the two at the same retention time point is calculated, the gradient evolution intensity of the signal response surface in the time and sample double dimensions is quantified according to the ratio result, and each ratio is matrix-encapsulated according to the original time and sample coordinate position to generate a second-order gradient variation rate matrix;

[0017] S202: Numerical comparison is performed on the gradient variation values in the second-order gradient variation rate matrix and the lower limit of the preset high variation rate threshold interval, the mutation points falling into the high variation rate threshold interval are identified, the mutation point set that is continuously distributed in the retention time dimension and has an accumulated length exceeding the minimum peak width reference value is searched, the mutation point set is combined and the boundary is smoothed according to its adjacent relationship on the time axis, the effective distribution range of the densely overlapping signal cluster is circled, and the high-activity gradient signal segment is generated.

[0018] S203: According to the start and end time coordinates and sample index range defined by the high-activity gradient signal segment, directional data extraction is performed on the original chromatographic data matrix, background data outside the segment boundary is excluded, invalid information is separated, and the extracted multiple two-dimensional data sub-blocks are spatially stacked and structurally reorganized according to the original arrangement order of the sample sequence to construct a three-dimensional data structure of retention time, sample sequence and response intensity, and generate a three-dimensional tensor structure section.

[0019] As a further scheme of the present application, the setting process of the high variation rate threshold interval is specifically: by statistically analyzing the numerical frequency distribution of the global elements in the second-order gradient variation rate matrix, a frequency distribution histogram of the variation rate values is constructed and a distribution envelope curve is fitted, the curvature extreme point or inflection point position of the envelope curve when transitioning from the high-frequency low-amplitude area to the low-frequency high-amplitude area is calculated, the variation rate value corresponding to the inflection point is identified as the demarcation threshold between background noise fluctuation and signal mutation, and the demarcation threshold is set as the lower limit of the high variation rate threshold interval, while the maximum variation rate value in the matrix is selected as the upper limit of the high variation rate threshold interval.

[0020] As a further scheme of the present application, the specific steps of S3 are:

[0021] S301: For the three-dimensional tensor structure section, a tensor decomposition objective function containing an L1 norm sparse constraint term is constructed, an alternating least squares iterative strategy is adopted to minimize the reconstruction error, coupled information in a multidimensional data array is demixed into characteristic factor matrices in spectral modal, chromatographic modal and sample modal, low-rank sparse components representing the internal structure of the signal are extracted, and a sparse factor loading matrix is generated;

[0022] S302: A unimodality index of a chromatographic dimension load vector is calculated for the sparse factor loading matrix, a peak shape integrity is evaluated, a degree of collinearity within an overlapping cluster is determined in combination with a spectral similarity coefficient, an effective component vector distinguished from a noise background is screened according to an energy contribution rate, a redundant residual term is removed, and a characteristic vector group representing an independent chemical component is separated, and a decoupled signal vector cluster is generated.

[0023] S303: Based on the decoupled signal vector cluster, a rank-one tensor reconstruction operation is performed on each group of independent characteristic vectors, a single response surface of each component in the original retention time dimension is recovered, a spectral response curve and a chromatographic outflow curve in a monofilament distribution are analyzed from the reconstruction result, and an independent chromatographic peak component is obtained.

[0024] As a further scheme of the application, the process of screening the effective component vector distinguished from the noise background according to the energy contribution rate is specifically: calculating the ratio of the square value of the core singular value corresponding to each characteristic factor matrix to the sum of squares of all singular values as the energy contribution rate, arranging the energy contribution rates in descending order, and intercepting the vector whose cumulative contribution rate reaches a preset interpretation threshold as the effective component vector.

[0025] As a further scheme of the application, the specific steps of S4 are:

[0026] S401: A morphological discretization segmentation process is performed on the independent chromatographic peak component along the retention time axis, the turning boundaries of waveform changes are locked according to the zero-crossing points of the first-order derivative and the curvature extreme points, the continuous chromatographic peak response curve is cut into a plurality of independent data units with single monotonicity or convexity and concavity geometric characteristics, the discrete point set including the retention time coordinates and signal response amplitudes in the unit is extracted, and a peak shape local sampling sequence is generated.

[0027] S402: A polynomial regression equation is constructed according to the spatial distribution characteristics of the discrete data points in the peak shape local sampling sequence, the sum of squares of Euclidean distances of the observation data points to the theoretical regression curve is calculated, the polynomial coefficients are iteratively adjusted by solving a normal equation set, the sum of squares of residuals is minimized, a smooth mathematical trajectory continuously describing the evolution trend of the signal response with time in each local sampling interval is established, and a segmented morphological regression curve is generated.

[0028] S403: Perform geometric moment feature extraction operation based on the segmented morphological regression curve, calculate the first-order origin moment defining the relative position of the waveform centroid and the second-order central moment quantifying the amplitude distribution width of the signal, combine the skewness coefficient describing the left-right asymmetry of the curve and the kurtosis value representing the sharpness of the peak, vectorize the multi-dimensional morphological statistics extracted from each segment according to the original time sequence logic, and obtain the local peak shape feature set.

[0029] As a further scheme of the present application, the specific steps of S5 are:

[0030] S501: Perform inverse transformation and maximum-minimum normalization operation on the segment residual statistics recorded in the local peak shape feature set, quantify the deviation degree of the segmented fitting curve relative to the original data points, calculate the numerical weight representing the data reliability of each sub-segment according to the distribution of the deviation value, construct a gradient sequence for distinguishing the importance of multiple local morphological features in the overall waveform expression, and generate a fitting quality weight coefficient vector;

[0031] S502: Call the fitting quality weight coefficient vector to perform weighted aggregation operation on the first-order origin moment and second-order central moment features in the local peak shape feature set, smooth the abnormal fluctuations of low-weight segments using the morphological parameters of high-weight segments, splice the adjusted local feature parameters into a high-dimensional numerical matrix describing the complete waveform topology structure according to the time sequence logic, and generate reconstructed global peak shape description information;

[0032] S503: Perform feature hash encoding processing on the reconstructed global peak shape description information, construct a unique digital fingerprint, calculate a confidence score representing the reliability of the reconstruction result according to the cumulative weight proportion in the aggregation operation process, project the description sub into a preset standard feature space to perform similarity matching, determine the component attribution category, and combine the fingerprint, score and category label to generate a chromatographic peak recognition identifier.

[0033] A data analysis processing system for processing in a chromatographic system, the system comprising:

[0034] A baseline correction processing module acquires an original chromatographic data matrix, calculates first-order and second-order curvature gradient change rates to construct a segmented function, maps the chromatographic signal to a sub-domain of the manifold space defined by the curvature mutation point, calculates the slope direction stability index in the sub-domain of the manifold space, filters out the drift term and retains the signal trunk, and generates a baseline corrected chromatographic signal.

[0035] A three-dimensional tensor analysis module calculates first-order difference values and second-order variation rates for the baseline corrected chromatographic signal, selects signal segments in a high threshold interval according to the second-order variation rates, trims the original chromatographic data matrix according to the signal segments, and constructs a three-dimensional tensor structure section.

[0036] The chromatographic component processing module performs a sparse decomposition operation based on the three-dimensional tensor structure section, analyzes tensor element separation independent response vectors, deconstructs a single filament spectrum response, and obtains independent chromatographic peak components;

[0037] The peak shape feature analysis module divides the independent chromatographic peak components into local sub-sections, performs a least square fitting on the local sub-sections, constructs a local fitting curve, extracts peak shape moment features from the local fitting curve, and forms a local peak shape feature set;

[0038] The recognition identifier implementation module calculates a weight coefficient according to the fitting quality of the local sub-sections, adjusts the local features in combination with global peak shape information, reconstructs the peak shape features, and generates a chromatographic peak recognition identifier.

[0039] Compared with the prior art, the application has the following advantages and positive effects:

[0040] In the application, by constructing a manifold space divided based on curvature mutation points, using gradient change rate to identify signal stems and filter out drift noise, stable correction of chromatographic signals in a baseline fluctuation background is realized, high gradient sections of signal response surfaces are constructed in combination with retention time difference and variation rate, potential overlapping peak regions in complex structures are effectively located, single spectrum components are deconstructed from cluster structures by means of tensor sparse decomposition algorithm, the overlapping peak analysis capability is improved, local sub-sections are fitted by least square method and peak shape moment features are extracted, weighted reconstruction is performed in combination with global information, the peak shape information is refined and the recognition accuracy is enhanced, and the problem of insufficient analysis capability of traditional methods under baseline drift, multi-peak coupling and peak recognition misalignment is solved. BRIEF DESCRIPTION OF DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0042] Figure 1 It is a step flowchart of the present application.

[0043] Figure 2 It is a S1 refinement diagram of the present application.

[0044] Figure 3 It is a S2 refinement diagram of the present application.

[0045] Figure 4 It is a S3 refinement diagram of the present application.

[0046] Figure 5S4 refinement schematic diagram of the present application;

[0047] Figure 6 S5 refinement schematic diagram of the present application;

[0048] Figure 7 System module diagram of the present application. DETAILED DESCRIPTION

[0049] The technical solutions in the present application will be described below with reference to the drawings.

[0050] In the embodiments of the present application, the words such as "example", "for example" are used to represent as an example, illustration or description. Any embodiment or design scheme described as "example" in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the word "example" is intended to present the concept in a specific manner. In addition, in the embodiments of the present application, the meaning expressed by "and / or" can be both, or can be one of the two.

[0051] In the embodiments of the present application, "image" and "picture" can be used interchangeably at times. It should be pointed out that when the distinction is not emphasized, the meanings expressed are consistent. "Of", "corresponding" and "corresponding" can be used interchangeably at times. It should be pointed out that when the distinction is not emphasized, the meanings expressed are consistent.

[0052] In the embodiments of the present application, sometimes the subscript such as W1 can be written in the form of non-subscript such as W1. When the distinction is not emphasized, the meanings expressed are consistent.

[0053] To make the technical problems, technical solutions and advantages to be solved by the present application clearer, specific embodiments will be described in detail below with reference to the drawings.

[0054] Please refer to Figure 1 The embodiments of the present application provide a data analysis processing method for processing a chromatographic system, comprising the following steps:

[0055] S1: acquiring an original chromatographic data matrix containing retention time and signal intensity through a chromatographic detection device, calculating the first and second curvature gradient change rates of the retention time sequence to construct a segmented function, mapping the chromatographic signal to a sub-domain in the manifold space defined by the curvature mutation point, calculating the slope direction stability index of the baseline change trend item in each manifold space sub-domain, screening the drift item with linear increasing and decreasing trend and retaining the signal stem with peak-shaped local extremum characteristics, and generating a baseline corrected chromatographic signal;

[0056] S2: Calculate the first-order difference value in the retention time dimension and the second-order variation rate between adjacent rows for the baseline-corrected chromatogram signal, determine the region where there is a gradient mutation on the signal response surface according to the second-order variation rate, select a continuous signal segment with the second-order variation rate in the preset high threshold interval, and cut the original chromatogram data matrix according to the signal segment to construct a three-dimensional tensor structure section;

[0057] S3: Perform sparse decomposition operation based on the three-dimensional tensor structure section, analyze the tensor elements in the peak overlapping region to separate the independent response vectors in the overlapping cluster, and deconstruct the single filament spectrum response from the overlapping cluster signal to obtain an independent chromatographic peak component;

[0058] S4: Divide the independent chromatographic peak component into a plurality of local sub-sections, call the least square method to perform segmented fitting processing for the data point position and morphological feature of each local sub-section, construct a local fitting curve that can describe the change trend of the sub-section, extract the peak shape moment feature from the local fitting curve, and form a local peak shape feature set;

[0059] S5: Calculate the weight coefficient according to the fitting quality of each local sub-section in the local peak shape feature set, call the weighted reconstruction algorithm to adjust and merge the local features in combination with the global peak shape information, reconstruct the peak shape feature, and generate a chromatographic peak recognition identifier;

[0060] The baseline-corrected chromatogram signal includes a pure retention time sequence, a corrected response intensity vector, and a residual background noise term. The three-dimensional tensor structure section includes a sample sequence dimension index, a cutting time dimension coordinate, and a section response numerical block. The independent chromatographic peak component includes a single filament spectrum, an independent response abundance value, and a component de-mixing weight coefficient. The local peak shape feature set includes a segmented fitting curve parameter, a local morphological skewness value, and a section residual statistic. The chromatographic peak recognition identifier includes a comprehensive peak shape fingerprint code, a reconstruction confidence score, and a category attribution label.

[0061] Please refer to Figure 2 , the specific steps of S1 are as follows:

[0062] S101: Collect an original chromatogram data matrix, perform discrete differential operation on the retention time sequence in the original chromatogram data matrix to obtain the first-order and second-order curvature gradient variation rate, construct a segmented function according to the mutation point position where the curvature gradient variation rate value changes in sign, map the chromatogram signal sequence to the corresponding data paragraph in the manifold space according to the interval definition rule of the segmented function, and generate a manifold space mapping sub-domain;

[0063] The processor extracts the response intensity values of the current sampling point and the previous adjacent sampling point in sequence, calculates the difference between the two and divides by the sampling time interval to obtain the first derivative value representing the signal change rate; then, the first derivative sequence obtained is subjected to adjacent point difference operation again to obtain the second derivative value representing the signal curvature, i.e. the second curvature gradient change rate. The piecewise function is constructed according to the position of the abrupt point where the curvature gradient change rate value undergoes a sign transition, and the specific operation is to detect the zero point of the second curvature gradient sequence, identify the time when the value changes from positive to negative or from negative to positive, and these times correspond to the inflection points or start and end boundaries of the chromatographic waveform; the processor sets these sign transition points as the interval nodes of the piecewise function, and divides the continuous time sequence into several independent time intervals. The chromatographic signal sequence is mapped to the corresponding data paragraphs in the manifold space according to the interval definition rule of the piecewise function, and this process involves regarding the response intensity sequence in each time interval as a local tangent space or sub-manifold piece in the high-dimensional manifold space, keeping the internal topological structure unchanged, and converting its coordinates from one-dimensional time axis to characteristic coordinates of the manifold space, thereby generating a manifold space mapping sub-domain containing local geometric features. For example, in a sequence containing baseline drift and weak signal, three intervals are identified by the second derivative sign change, respectively corresponding to the flat baseline, signal rising edge and signal falling edge, and the system maps the three data segments into three independent sub-domains in the manifold space, respectively, for subsequent processing.

[0064] S102: Perform linear regression analysis on each data paragraph in the manifold space mapping sub-domain, extract the slope value of the baseline trend item, count the sign flip frequency and amplitude fluctuation variance of the slope value between adjacent sampling points, and quantify the direction consistency degree of the baseline trend in the time dimension in the sub-domain by calculating the statistical distribution characteristics of the slope direction in the continuous time window, to generate a slope direction stability index;

[0065] Within each sub-domain, the instantaneous slope is calculated point by point, the number of times of the sign of the slope switching between positive and negative is counted, and the number is divided by the total number of sampling points in the sub-domain to obtain the sign flipping frequency; at the same time, the dispersion degree of all instantaneous slope values relative to the average slope of the sub-domain is calculated to obtain the amplitude fluctuation variance. By calculating the statistical distribution characteristics of the slope direction in the continuous time window, the direction consistency degree of the baseline trend in the time dimension in the sub-domain is quantified, and the slope direction stability index is generated. The calculation logic of the index is: the reciprocal of the sign flipping frequency and the reciprocal of the amplitude fluctuation variance are normalized and weighted to sum. If a certain sub-domain is a stable baseline drift, the slope direction basically remains constant (such as continuously positive), the sign flipping frequency is very low and the variance is small, and the calculated stability index value is high; if a certain sub-domain is high-frequency noise, the slope direction frequently jumps, the flipping frequency is high and the variance is large, and the stability index value is very low. For example, set the time window length to fifty sampling points, for a smooth rising baseline drift, the calculated sign flipping frequency is close to zero, and the slope direction stability index value tends to one; while for a section of severe oscillation noise, the stability index value tends to zero.

[0066] S103: Based on the numerical comparison between the slope direction stability index and the preset baseline drift judgment threshold, the linear increase and decrease trend drift item with the index value tending to converge is identified and separated, the drift item is removed from the manifold space mapping sub-domain, the signal main part with the peak-shaped local extreme value characteristics is retained, the processed signal main part of each sub-domain is time series spliced and smoothed, and a baseline corrected chromatogram signal is generated;

[0067] The setting of baseline drift determination threshold is based on statistical analysis of chromatograms of a large number of blank solvent samples, calculation of probability density function of stability index of pure noise area, and selection of index value when cumulative probability distribution reaches 95% as baseline drift determination threshold. Linear increasing and decreasing trend drift items are identified and separated, and the specific determination logic is that when the slope direction stability index of a subdomain is greater than the threshold, it is determined that the subdomain is mainly composed of linear baseline drift, and the processor identifies the linear regression fitting straight line in the subdomain as a drift item; when the index is less than the threshold, it is determined that the subdomain contains nonlinear chromatographic peak signals or complex fluctuations. The drift items are removed from the manifold space mapping subdomain, that is, the linear component identified as the drift item is subtracted from the original signal, and the remaining residual signal is taken as the reserved part. The signal main part with peak shape local extremum characteristics is reserved, and the signal main part after processing of each subdomain is time series spliced and smoothed to generate a baseline corrected chromatographic signal. The splicing process uses a weighted average method to process the breakpoints at the boundaries of adjacent subdomains to eliminate step mutations; the smoothing reconstruction uses a five-point cubic smoothing algorithm to denoise the spliced sequence. For example, if the calculated baseline drift determination threshold is 0.8, and the stability index of a subdomain is 0.9, the subdomain is determined to be baseline drift and is deducted; the index of another subdomain is 0.3, and the subdomain data is reserved. The final output baseline corrected chromatographic signal is a pure data sequence that removes low-frequency background drift and maintains the integrity of the chromatographic peak.

[0068] Referring to Figure 3 , the specific steps of S2 are:

[0069] S201: Based on the baseline corrected chromatographic signal, a first-order difference operation is performed on the response intensity value along the retention time dimension to obtain a first-order gradient sequence representing the instantaneous slope of the signal in a single sample, the first-order gradient sequences corresponding to adjacent sample rows are extracted, and the change rate ratio of the two at the same retention time point is calculated. According to the ratio result, the gradient evolution degree of the signal response surface in the time and sample double dimensions is quantified, each ratio is matrix encapsulated according to the original time and sample coordinate position, and a second-order gradient variation rate matrix is generated;

[0070] The difference between the current time point response value and the last time point response value is calculated for each row sample in the matrix to obtain a first-order gradient sequence representing the instantaneous change rate of the signal in a single sample. The first-order gradient sequences corresponding to adjacent sample rows are extracted and the change rate ratio at the same retention time point is calculated. This process aims to capture the differences in signal evolution of different samples (such as different batches of continuous samples or repeated needles in the same experiment) at the same retention time position. Specifically, the processor takes the first-order gradient value of the Nth row sample at time T as the numerator and the first-order gradient value of the N-1th row sample at time T as the denominator (the denominator needs to add a small anti-zero constant), and calculates the gradient ratio. According to the ratio result, the gradient evolution intensity of the signal response surface in the time and sample double dimensions is quantified. If the ratio deviates significantly from one, it indicates that the signal at this position has a dramatic morphological change or retention time drift between samples. Each ratio is matrix-encapsulated according to the original time and sample coordinate position to generate a second-order gradient variation rate matrix. In this matrix, the row index corresponds to the retention time point, the column index corresponds to the sample number, and the matrix element value is the gradient ratio calculated above. For example, if the signal rise rate of the previous sample is ten and the rise rate of the current sample suddenly changes to twenty at a certain retention time point, the variation rate at this position is two, indicating that there is a strong gradient evolution between samples at this position, which may correspond to the separation degree change of overlapping peaks or the emergence of new impurity interference.

[0071] S202: Perform numerical comparison on the gradient variation value in the second-order gradient variation rate matrix and the lower limit of the preset high variation rate threshold interval, identify the mutation point where the variation value falls within the high variation rate threshold interval, retrieve a set of mutation points that are continuously distributed in the retention time dimension and have an accumulated length exceeding a minimum peak width reference value, and perform merging and boundary smoothing processing on the mutation point set according to its adjacency relationship on the time axis to delineate the effective distribution range of the dense overlapping signal cluster and generate a high-activity gradient signal segment.

[0072] The setting process of the high variation rate threshold interval is as follows: by statistically analyzing the numerical frequency distribution of the global elements in the second-order gradient variation rate matrix, a frequency distribution histogram of the variation rate value is constructed and a distribution envelope curve is fitted, the curvature extreme point or inflection point position of the envelope curve when transitioning from the high-frequency low-amplitude region to the low-frequency high-amplitude region is calculated, the variation rate value corresponding to the inflection point is identified as the threshold value between background noise fluctuation and signal mutation, and the threshold value is set as the lower limit of the high variation rate threshold interval. At the same time, the maximum variation rate value in the matrix is selected as the upper limit of the high variation rate threshold interval.

[0073] The setting process of the high variability threshold interval is specifically as follows: the frequency distribution of the numerical values of the global elements in the second-order gradient variability rate matrix is counted, and a frequency distribution histogram of the variability rate numerical values is constructed. The envelope curve of the distribution of the histogram is fitted by using the kernel density estimation method, and the curvature extreme point or inflection point position of the envelope curve when it transitions from the high-frequency low-amplitude area (mainly composed of background noise) to the low-frequency high-amplitude area (mainly composed of signal mutations) is calculated. The variability rate value corresponding to the inflection point is identified as the threshold value between background noise fluctuations and signal mutations, and the threshold value is set as the lower limit of the high variability threshold interval. At the same time, the maximum variability rate value in the matrix is selected as the upper limit of the high variability threshold interval. For example, after calculation, the variability rate of the background noise is mainly concentrated between 0.9 and 1.1, and the inflection point appears at 1.5. Therefore, 1.5 is set as the lower limit of the threshold value. The set of mutation points that are continuously distributed in the retention time dimension and have an accumulated length exceeding the minimum peak width reference value is searched, and the minimum peak width reference value is set according to the theoretical plate number and dead time of the chromatographic column, for example, 5 sampling points. The mutation point set is merged and the boundary is smoothed according to its adjacent relationship on the time axis, that is, the discontinuous sections that are too close are connected, and a certain number of sampling points are expanded outward to include the slow-changing area at the beginning and end of the signal, the effective distribution range of the dense overlapping signal cluster is circled, and the high-activity gradient signal section is generated.

[0074] S203: According to the start and end time coordinates and sample index range defined by the high-activity gradient signal section, perform directional data extraction on the original chromatographic data matrix, eliminate background data outside the section boundary, separate invalid information, and perform spatial stacking and structured reorganization on the extracted multiple two-dimensional data sub-blocks according to the original arrangement order of the sample sequence, to construct a three-dimensional data structure of retention time, sample sequence and response intensity, and generate a three-dimensional tensor structure section;

[0075] The time start point T_start, the time end point T_end, the sample start number S_start, and the sample end number S_end of the high activity gradient signal section record are read, and these index parameters are used as the slicing basis to accurately cut out a rectangular data block containing the target signal from the large original data matrix. The background data outside the section boundary is removed, the invalid information is separated, that is, the smooth baseline or pure noise data not falling within the high activity section is discarded, thereby significantly reducing the data amount for subsequent processing. The extracted multiple two-dimensional data sub-blocks are spatially stacked and structured according to the original arrangement order of the sample sequence. The specific operation is as follows: if M two-dimensional data blocks of different samples within the same retention time period are extracted, the processor aligns and stacks the M data blocks along the third dimension (the sample dimension), constructs a three-dimensional data structure of the retention time (the first dimension), the sample sequence (the second dimension), and the response intensity (the third dimension), generates a three-dimensional tensor structure section. For example, the extracted data block covers the retention time of the tenth to the twentieth minute and involves the first to the fiftieth sample. These data are reorganized into a three-dimensional tensor with a dimension of “time length multiplied by sample number multiplied by response intensity”, and the tensor centrally contains the complex overlapping peak signal information to be analyzed.

[0076] Please refer to Figure 4 , the specific steps of S3 are as follows:

[0077] S301: For the three-dimensional tensor structure section, a tensor decomposition target function containing an L1 norm sparse constraint term is constructed, an alternating least squares iterative strategy is adopted to minimize the reconstruction error, the coupled information in the multi-dimensional data array is demixed into characteristic factor matrices in the spectral mode, the chromatographic mode, and the sample mode, the low-rank sparse components representing the internal structure of the signal are extracted, and a sparse factor loading matrix is generated;

[0078] The objective function aims to find the best combination of factor matrices so that the Euclidean distance between the reconstructed tensor and the original tensor is minimized, while adding an L1 regularization term in the objective function to punish the number of non-zero elements in the factor matrix, so as to promote the sparsity of the decomposition result, thereby suppressing noise and meaningless background signals. An alternating least squares iterative strategy is adopted to minimize the reconstruction error. The specific implementation process is as follows: the processor initializes the factor matrix, then fixes the factor matrix of two modes, and solves the factor matrix of the third mode that minimizes the objective function. In this way, the feature factor matrices under the spectral mode, the chromatographic mode and the sample mode are updated in turn until the objective function converges or the maximum number of iterations is reached. The coupled information in the multi-dimensional data array is unmixed into the feature factor matrices under the spectral mode (representing spectral characteristics), the chromatographic mode (representing elution curves) and the sample mode (representing concentration distribution), the low-rank sparse components representing the internal structure of the signal are extracted, and the sparse factor loading matrix is generated. For example, when processing liquid chromatography-diode array detector data, the three compound signals overlapped together can be separated into three independent spectral vectors, three independent chromatographic elution curve vectors and corresponding concentration distribution vectors through the decomposition process.

[0079] S302: Calculate the unimodality index of the chromatographic dimension loading vector for the sparse factor loading matrix, evaluate the peak shape integrity, determine the collinearity degree within the overlapping cluster combined with the spectral similarity coefficient, sort and select the effective component vector distinguished from the noise background according to the energy contribution rate, eliminate redundant residual terms and separate the feature vector group representing independent chemical components, and generate the decoupled signal vector cluster;

[0080] The process of sorting and selecting the effective component vector distinguished from the noise background according to the energy contribution rate is specifically as follows: calculating the ratio of the square value of the core singular value corresponding to each feature factor matrix to the sum of the squares of all singular values as the energy contribution rate, arranging the energy contribution rates in descending order, and taking the vectors whose cumulative contribution rate reaches the preset explanation threshold as the effective component vectors;

[0081] The calculation method of the unimodality index is to count the number of local maximum points in the loading vector. If there is only one significant maximum value and the two sides are monotonically decreasing, the unimodality index is one. If there are multiple maximum values, the index is reduced according to the ratio of the secondary maximum value to the primary maximum value. The degree of collinearity within the overlapping cluster is determined by combining the spectral similarity coefficient, that is, the cosine value of the angle between the decomposed spectral vector and the vector in the standard material spectrum library is calculated. According to the energy contribution rate, the effective component vector that is different from the noise background is screened. The specific process is as follows: the square value of the core singular value corresponding to each characteristic factor matrix is calculated as the energy value of the component; the energy value is divided by the sum of the squares of all decomposed singular values to obtain the relative energy contribution rate. The energy contribution rate is arranged in descending order, and the vector with a cumulative contribution rate reaching a preset interpretation threshold (for example, 99.99%) is taken as the effective component vector, and the remaining energy contribution is very low and has no clear chemical meaning. The redundant residual term is removed and the feature vector group representing independent chemical components is separated to generate the decoupled signal vector cluster. For example, if five components are decomposed, the cumulative energy contribution rate of the first three components has reached 99.95%, and all have good unimodality characteristics, then the three are retained as decoupled signals, and the last two are discarded as fitting residuals.

[0082] S303: Based on the decoupled signal vector cluster, rank-one tensor reconstruction operation is performed on each group of independent feature vectors to recover the single response surface of each component in the original retention time dimension, and the single filament distribution spectral response curve and chromatographic elution curve are analyzed from the reconstruction result to obtain independent chromatographic peak components;

[0083] The spectral vector, chromatographic vector and sample concentration vector corresponding to each effective component are extracted, and the three vectors are recombined using tensor outer product operation to recover the single response surface of the component in the original retention time dimension. This operation restores the one-dimensional feature vector to a three-dimensional response distribution with the same dimension as the original data, but at this time it only contains the contribution of a single chemical component. The single filament distribution spectral response curve and chromatographic elution curve are analyzed from the reconstruction result, that is, from the reconstructed single response surface, the maximum response value corresponding to the section is extracted as a pure spectral curve along the retention time axis, and the section at the maximum absorption wavelength is extracted as a pure chromatographic elution curve along the spectral wavelength axis, to obtain independent chromatographic peak components. This step ensures that each separated chromatographic peak corresponds to a pure chemical substance, eliminating the mutual interference between overlapping peaks. For example, for a binary mixture that is not completely separated, after this step, the system outputs two independent data sets, the first set only contains the complete chromatographic peak and spectrum of component A, and the second set only contains the complete data of component B, which are completely decoupled in value.

[0084] Please refer to Figure 5 , the specific steps of S4 are:

[0085] S401: Perform morphological discretization segmentation processing along the retention time axis for independent chromatographic peak components, lock the turning boundary of waveform change according to the zero-crossing point of the first-order derivative and the extreme point of the curvature, cut the continuous chromatographic peak response curve into multiple independent data units with single monotonicity or convexity and concavity geometric characteristics, extract the discrete point set including retention time coordinates and signal response amplitude in the unit, and generate a local sampling sequence of peak shape;

[0086] Lock the turning boundary of waveform change according to the zero-crossing point of the first-order derivative, that is, identify the point where the slope changes from positive to negative as the peak vertex; lock the point where the curvature changes most sharply as the inflection point according to the extreme point of the second-order derivative. Use these key points as cutting boundaries to cut the continuous chromatographic peak response curve into multiple independent data units with single monotonicity or convexity and concavity geometric characteristics. For example, a standard chromatographic peak is divided into four independent units: left peak to left inflection point (concave rising), left inflection point to peak top (convex rising), peak top to right inflection point (convex falling), and right inflection point to falling peak (concave falling). Extract the discrete point set including retention time coordinates and signal response amplitude in the unit, and generate a local sampling sequence of peak shape. This process decomposes the complex nonlinear waveform into several simple geometric elements, facilitating subsequent high-precision local fitting.

[0087] S402: Construct a polynomial regression equation according to the spatial distribution characteristics of the discrete data points in the local sampling sequence of peak shape, calculate the Euclidean distance square sum of the observed data points to the theoretical regression curve, adjust the polynomial coefficients by solving the normal equation system, minimize the residual square sum, establish a smooth mathematical trajectory that continuously describes the signal response evolution trend over time in each local sampling interval, and generate a segmented morphological regression curve;

[0088] Set a low-order polynomial model (such as a cubic polynomial), which has the mathematical form of response value equal to the linear combination of time to multiple powers. Calculate the Euclidean distance square sum of the observed data points to the theoretical regression curve as the target loss function. Adjust the polynomial coefficients by solving the normal equation system, the processor constructs a linear equation system containing the Vandermonde matrix, and solves the coefficient vector using Gaussian elimination method or matrix decomposition method, so that the residual square sum of all data points to the fitted curve is minimized, thereby establishing a smooth mathematical trajectory that continuously describes the signal response evolution trend over time in each local sampling interval, and generating a segmented morphological regression curve. For example, for the "left peak to left inflection point" segment, the system calculates a specific set of polynomial coefficients, so that the polynomial curve can closely fit the rising trend of the original sampling points, while filtering out high-frequency random noise, obtaining a smooth analytical curve segment.

[0089] S403: Perform geometric moment feature extraction operation based on the segmented morphological regression curve, calculate the first-order origin moment defining the relative position of the waveform centroid and the second-order central moment quantifying the amplitude distribution width of the signal, combine the skewness coefficient describing the left-right asymmetry of the curve and the kurtosis value representing the sharpness of the peak, vectorize and aggregate the multi-dimensional morphological statistics extracted from each segment according to the original time sequence logic, and obtain the local peak shape feature set;

[0090] Use integral algorithm to perform numerical integration on the fitted segmented curve equation, calculate the first-order origin moment defining the relative position of the waveform centroid, that is, calculate the integral value of the retention time multiplied by the response intensity divided by the integral value of the response intensity. This index quantifies the center of gravity of signal energy on the time axis; Calculate the second-order central moment quantifying the amplitude distribution width of the signal, that is, calculate the integral value of the response intensity multiplied by the square of the retention time minus the centroid time. This index reflects the diffusion degree or frequency band width of the chromatographic peak. Combine the skewness coefficient (third-order normalized central moment) describing the left-right asymmetry of the curve and the kurtosis value (fourth-order normalized central moment) representing the sharpness of the peak, vectorize and aggregate the multi-dimensional morphological statistics extracted from each segment according to the original time sequence logic, and obtain the local peak shape feature set. For example, for a severely tailing chromatographic peak, the first-order origin moment calculated will be biased towards the direction of larger retention time, and the skewness coefficient will be positive. For a sharp needle-shaped peak, the second-order central moment value is very small and the kurtosis value is very large. These feature values are packaged into a feature vector as the digital fingerprint of the chromatographic peak shape.

[0091] Please refer to Figure 6 The specific steps of S5 are as follows:

[0092] S501: Perform inverse transformation and maximum-minimum normalization operation on the segment residual statistics recorded in the local peak shape feature set, quantify the deviation degree of the segmented fitting curve relative to the original data points, calculate the numerical weight representing the data reliability of each subsegment according to the distribution of the deviation value, construct a gradient sequence for distinguishing the importance of multiple local morphological features in the overall waveform expression, and generate a fitting quality weight coefficient vector;

[0093] The residual sum of squares of each segment is obtained from the regression analysis step, and the smaller the value, the more accurate the fitting. The inverse transformation is performed on SSE, so that the segment with high fitting accuracy (small residual) obtains a larger value, and vice versa. Then, the sequence of inverse transformed values is subjected to maximum minimum value normalization, mapping all values to the closed interval of zero to one, quantifying the deviation degree of the segmented fitting curve relative to the original data points. According to the distribution of the deviation values, the numerical weight representing the data reliability of each sub-segment is calculated, a gradient sequence is constructed to distinguish the importance of multiple local morphological features in the overall waveform expression, and a fitting quality weight coefficient vector is generated. For example, if the peak top part has high signal-to-noise ratio and the fitting residual is extremely small, the weight coefficient after calculation is close to one; while the baseline noise at the peak is large and the fitting residual is high, the weight coefficient may be only 0.2. In this way, the system automatically gives higher importance to high-quality data segments.

[0094] S502: calling the fitting quality weight coefficient vector to perform weighted aggregation operation on the first-order origin moment and second-order central moment features in the local peak shape feature set, smoothing and correcting the abnormal fluctuations of low-weight segments by using the morphological parameters of high-weight segments, splicing the adjusted local feature parameters into a high-dimensional numerical matrix describing the complete waveform topological structure according to the time sequence logic, and generating reconstructed global peak shape description information;

[0095] The morphological feature value of each segment is multiplied by its corresponding weight coefficient, then the weighted feature values are summed and divided by the sum of the weight coefficients, thereby obtaining the global feature parameters after weighted averaging. The abnormal fluctuations of low-weight segments are smoothed and corrected by using the morphological parameters of high-weight segments, that is, when the feature value of a low-weight segment deviates from the weighted average value by more than a certain threshold, the trend value derived from the high-weight segment is used to replace or strongly constrain it. The adjusted local feature parameters are spliced into a high-dimensional numerical matrix describing the complete waveform topological structure according to the time sequence logic, and the reconstructed global peak shape description information is generated. This matrix not only contains the fine geometric features of each segment, but also integrates the confidence information based on data quality. For example, the final generated description information shows that the centroid position of the chromatographic peak is determined by the high-confidence data in the peak top region, while the noise interference in the edge region is effectively suppressed through low-weight processing.

[0096] S503: performing feature hash coding processing on the reconstructed global peak shape description information to construct a unique digital fingerprint, calculating a confidence score representing the reliability of the reconstruction result according to the cumulative weight proportion in the aggregation operation process, projecting the description sub into a preset standard feature space to perform similarity matching, determining the component attribution category, combining the fingerprint, score and category label to generate a chromatographic peak identification identifier;

[0097] The position-sensitive hashing algorithm or perceptual hashing algorithm is used to project the high-dimensional numerical matrix into a low-dimensional binary space to construct a unique digital fingerprint. The fingerprint is a string of binary codes that can tolerate slight numerical fluctuations but is highly sensitive to structural changes. The confidence score representing the reliability of the reconstruction result is calculated based on the cumulative weight proportion in the aggregation operation process. The score is the average of all segmented weight coefficients, reflecting the overall signal-to-noise ratio level of the original data. The descriptor is projected into the preset standard feature space to perform similarity matching, calculate the Hamming distance or cosine similarity between the generated digital fingerprint and the known substance fingerprint in the standard database, and determine the component attribution category. The fingerprint, score, and category label are combined and packaged to generate a chromatographic peak recognition identifier. For example, after matching, the fingerprint of a certain chromatographic peak has a similarity of 98% with the standard fingerprint of "benzene" in the database, and the confidence score is 0.95. The system generates an identification identifier containing "ID: benzene, similarity: 0.98, confidence: 0.95", completing the qualitative analysis.

[0098] Please refer to Figure 7 A data analysis processing system for processing in a chromatographic system, the system comprising:

[0099] A baseline correction processing module acquires the original chromatographic data matrix, calculates the first-order and second-order curvature gradient change rates to construct a segmented function, maps the chromatographic signal to the manifold space sub-domain defined by the curvature mutation point, calculates the slope direction stability index in the manifold space sub-domain, filters out the drift term and retains the signal trunk, and generates a baseline-corrected chromatographic signal;

[0100] A three-dimensional tensor analysis module calculates the first-order difference value and the second-order variation rate for the baseline-corrected chromatographic signal, selects the signal segment in the high threshold interval according to the second-order variation rate, trims the original chromatographic data matrix according to the signal segment, and constructs a three-dimensional tensor structure section;

[0101] A chromatographic component processing module performs sparse decomposition operation based on the three-dimensional tensor structure section, analyzes the tensor element to separate independent response vectors, deconstructs a single filament spectrum response, and obtains independent chromatographic peak components;

[0102] A peak shape feature analysis module divides the independent chromatographic peak components into local subsegments, performs least squares fitting on the local subsegments, constructs a local fitting curve, extracts peak shape moment features from the local fitting curve, and forms a local peak shape feature set;

[0103] An identification identifier implementation module calculates weight coefficients according to the fitting quality of the local subsegments, adjusts the local features in combination with the global peak shape information, reconstructs the peak shape features, and generates a chromatographic peak recognition identifier.

[0104] The above merely illustrates the specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A data analysis and processing method for processing data in a chromatographic system, characterized in that, Includes the following steps: S1: Collect the raw chromatographic data matrix, calculate the first and second order curvature gradient change rates to construct a piecewise function, map the chromatographic signal to the manifold space subdomain defined by the curvature abrupt change point, calculate the slope direction stability index in the manifold space subdomain, screen the drift terms and retain the signal backbone, and generate the baseline-corrected chromatographic signal. S2: Calculate the first-order difference value and the second-order variation rate for the baseline correction chromatographic signal, select the signal segment in the high threshold range according to the second-order variation rate, trim the original chromatographic data matrix according to the signal segment, and construct a three-dimensional tensor structure section. S3: Perform sparse decomposition operation based on the three-dimensional tensor structure section, analyze the tensor elements to separate independent response vectors, deconstruct the single filamentary spectral response, and obtain independent chromatographic peak components. S4: Divide the independent chromatographic peak components into local sub-segments, perform least squares fitting on the local sub-segments, construct local fitting curves, extract peak shape moment features from the local fitting curves, and form a set of local peak shape features. S5: Calculate the weighting coefficients based on the fitting quality of the local sub-segments, adjust the local features in combination with the global peak shape information, reconstruct the peak shape features, and generate chromatographic peak identification identifiers; The baseline-corrected chromatographic signal includes a pure retention time series, a corrected response intensity vector, and a residual background noise term. The three-dimensional tensor structure cross-section includes a sample sequence dimension index, a clipped time dimension coordinate, and a cross-sectional response value block. The independent chromatographic peak components include a single filamentary spectrum, an independent response abundance value, and a component decomposition weighting coefficient. The local peak shape feature set includes piecewise fitting curve parameters, local morphological skewness values, and segment residual statistics. The chromatographic peak identification identifier includes a comprehensive peak shape fingerprint code, a reconstructed confidence score, and a category attribution label.

2. The data analysis and processing method for processing a chromatographic system according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Collect the original chromatographic data matrix, perform discrete differential operation on the retained time series in the original chromatographic data matrix to obtain the first and second curvature gradient change rates, construct a piecewise function based on the abrupt change point where the curvature gradient change rate changes sign, map the chromatographic signal sequence to the corresponding data segment in the manifold space according to the interval definition rules of the piecewise function, and generate the manifold space mapping subdomain. S102: Perform linear regression analysis on each data segment within the manifold space mapping subdomain, extract the slope value of the baseline change trend term, count the sign flip frequency and amplitude fluctuation variance of the slope values ​​between adjacent sampling points, and quantify the degree of directional consistency of the baseline trend in the subdomain in the time dimension by calculating the statistical distribution characteristics of the slope direction within a continuous time window, and generate a slope direction stability index. S103: Based on the slope direction stability index and the preset baseline drift judgment threshold, the numerical values ​​are compared to identify and separate the linear increase and decrease trend drift terms that tend to converge. The drift terms are removed from the manifold space mapping subdomain, and the signal backbone with peak-shaped local extreme value characteristics is retained. The signal backbone after processing each subdomain is time-series spliced ​​and smoothed to reconstruct the signal and generate the baseline correction chromatographic signal.

3. The data analysis and processing method for processing a chromatographic system according to claim 1, characterized in that, The specific steps of S2 are as follows: S201: Based on the baseline-corrected chromatographic signal, perform a first-order difference operation on the response intensity value along the retention time dimension to obtain a first-order gradient sequence characterizing the instantaneous slope of the signal within a single sample. Extract the first-order gradient sequences corresponding to adjacent sample rows and calculate the ratio of their rate of change at the same retention time point. Quantify the intensity of gradient evolution of the signal response surface in both time and sample dimensions based on the ratio result. Encapsulate each ratio in a matrix according to the original time and sample coordinate positions to generate a second-order gradient variability matrix. S202: Perform a numerical comparison between the gradient variation values ​​in the second-order gradient variation rate matrix and the lower limit of the preset high variation rate threshold interval, identify the mutation points where the variation values ​​fall into the high variation rate threshold interval, retrieve the set of mutation points that are continuously distributed in the retention time dimension and whose cumulative length exceeds the minimum peak width reference value, merge and smooth the mutation point set according to its adjacency relationship on the time axis, delineate the effective distribution range of densely overlapping signal clusters, and generate a high-activity gradient signal segment; S203: Based on the start and end time coordinates and sample index range defined by the high-activity gradient signal segment, perform directional data extraction on the original chromatographic data matrix, remove background data outside the segment boundaries, separate invalid information, and spatially stack and structurally reorganize the extracted multiple two-dimensional data sub-blocks according to the original arrangement order of the sample sequence to construct a three-dimensional data structure with three dimensions: retention time, sample sequence, and response intensity, and generate a three-dimensional tensor structure cross section.

4. The data analysis and processing method for processing a chromatographic system according to claim 3, characterized in that, The process of setting the high variability threshold interval is as follows: by statistically analyzing the frequency distribution of global elements in the second-order gradient variability matrix, a frequency distribution histogram of variability values ​​is constructed and a distribution envelope curve is fitted. The curvature extreme point or inflection point of the envelope curve when it transitions from the high-frequency low-amplitude region to the low-frequency high-amplitude region is calculated. The variability value corresponding to the inflection point is identified as the boundary threshold between background noise fluctuations and signal mutations. The boundary threshold is set as the lower limit of the high variability threshold interval. At the same time, the maximum variability value inside the matrix is ​​selected as the upper limit of the high variability threshold interval.

5. The data analysis and processing method for processing a chromatographic system according to claim 1, characterized in that, The specific steps for S3 are as follows: S301: For the three-dimensional tensor structure section, construct a tensor decomposition objective function containing L1 norm sparse constraint terms, use an alternating least squares iterative strategy to minimize the reconstruction error, demix the coupled information in the multidimensional data array into feature factor matrices under spectral mode, chromatographic mode and sample mode, extract low-rank sparse components that characterize the intrinsic structure of the signal, and generate a sparse factor loading matrix. S302: Calculate the unimodality index of the chromatographic dimension loading vector for the sparse factor loading matrix, evaluate the peak shape integrity, determine the degree of collinearity within the overlapping cluster by combining the spectral similarity coefficient, sort and screen the effective component vectors that are different from the noise background according to the energy contribution rate, remove redundant residual terms and separate the feature vector groups representing independent chemical components to generate decoupled signal vector clusters. S303: Based on the decoupled signal vector cluster, perform rank-tensor reconstruction operation on each group of independent feature vectors to recover the single response surface of each component in the original retention time dimension, and analyze the spectral response curve and chromatographic elution curve with monofilament distribution from the reconstruction result to obtain independent chromatographic peak components.

6. The data analysis and processing method for processing a chromatographic system according to claim 5, characterized in that, The process of sorting and selecting effective component vectors that are different from the noise background based on energy contribution rate is as follows: the ratio of the square of the core singular value corresponding to each feature factor matrix to the sum of the squares of all singular values ​​is calculated as the energy contribution rate. The energy contribution rates are arranged in descending order, and vectors whose cumulative contribution rate reaches the preset explanatory power threshold are selected as effective component vectors.

7. The data analysis and processing method for processing a chromatographic system according to claim 1, wherein step S4 is as follows: S401: Perform morphological discretization segmentation processing on the independent chromatographic peak components along the retention time axis. Based on the zero-crossing point of the first derivative of the curve and the extreme point of curvature, lock the turning boundary of the waveform change. Divide the continuous chromatographic peak response curve into multiple independent data units with single monotonicity or convex-concave geometric features. Extract the discrete point set including retention time coordinates and signal response amplitude inside the unit to generate a peak shape local sampling sequence. S402: Based on the spatial distribution characteristics of discrete data points in the peak-shaped local sampling sequence, construct a polynomial regression equation, calculate the sum of squared Euclidean distances from the observed data points to the theoretical regression curve, iteratively adjust the polynomial coefficients by solving the normal equation system, minimize the sum of squared residuals, establish a smooth mathematical trajectory that continuously describes the trend of signal response with time evolution in each local sampling interval, and generate a piecewise morphological regression curve. S403: Based on the segmented morphological regression curve, perform geometric moment feature extraction operation, calculate the first-order origin moment that defines the relative position of the waveform centroid and the second-order central moment that defines the amplitude distribution width of the quantized signal, combine the skewness coefficient that describes the left and right asymmetry of the curve and the kurtosis value that characterizes the sharpness of the peak, and vectorize and aggregate the multidimensional morphological statistics extracted from each segment according to the original time series logic to obtain a set of local peak features.

8. The data analysis and processing method for processing a chromatographic system according to claim 1, characterized in that, The specific steps of S5 are as follows: S501: Perform reciprocal transformation and maximum / minimum normalization operations on the segment residual statistics recorded in the local peak feature set to quantify the deviation of the segmented fitting curve relative to the original data points. Calculate the numerical weight representing the reliability of each sub-segment data based on the distribution of the deviation values, construct a gradient sequence to distinguish the importance of multiple local morphological features in the overall waveform expression, and generate a fitting quality weight coefficient vector. S502: Call the fitted quality weight coefficient vector to perform weighted aggregation operation on the first-order origin moment and second-order central moment features in the local peak shape feature set, use the morphological parameters of the high-weight segment to smooth the abnormal fluctuations of the low-weight segment, and according to the time series logic, splice the adjusted local feature parameters into a high-dimensional numerical matrix describing the complete waveform topology to generate reconstructed global peak shape description information. S503: Perform feature hash encoding processing on the reconstructed global peak shape description information to construct a unique digital fingerprint. Calculate the confidence score that characterizes the reliability of the reconstruction result based on the cumulative weight ratio in the aggregation operation process. Project the descriptor to a preset standard feature space to perform similarity matching, determine the component's category, and encapsulate the fingerprint, score, and category label to generate a chromatographic peak identification identifier.

9. A data analysis and processing system for processing data in a chromatographic system, characterized in that, The system is used to implement the data analysis and processing method for processing a chromatographic system according to any one of claims 1-8, the system comprising: The baseline correction processing module acquires the raw chromatographic data matrix, calculates the first and second order curvature gradient change rates to construct a piecewise function, maps the chromatographic signal to the manifold space subdomain defined by the curvature abrupt change point, calculates the slope direction stability index within the manifold space subdomain, filters drift terms and retains the signal backbone, and generates a baseline-corrected chromatographic signal. The three-dimensional tensor analysis module calculates the first-order difference value and the second-order variability rate for the baseline-corrected chromatographic signal, selects the signal segment in the high threshold range based on the second-order variability rate, trims the original chromatographic data matrix based on the signal segment, and constructs a three-dimensional tensor structure section. The chromatographic component processing module performs sparse decomposition operation based on the three-dimensional tensor structure section, analyzes the tensor element to separate independent response vectors, deconstructs the single filamentary spectral response, and obtains independent chromatographic peak components. The peak shape feature analysis module divides the independent chromatographic peak components into local sub-segments, performs least squares fitting on the local sub-segments, constructs local fitting curves, extracts peak shape moment features from the local fitting curves, and forms a set of local peak shape features. The identification module calculates weight coefficients based on the fitting quality of the local sub-segments, adjusts local features in conjunction with global peak shape information, reconstructs peak shape features, and generates chromatographic peak identification identifiers.

Images