Chromatograph data processing device, data processing method and chromatograph

Abstract

The present invention provides a data processing device, a data processing method and a chromatograph, which are not easily affected by noise and can easily obtain peak characteristic points in a short time. The data processing device of the chromatograph comprises: a partial regression curve calculation unit, which obtains partial regression curves as partial regression curves for a plurality of data groups obtained by grouping a predetermined number of measured time series data; a partial characteristic point acquisition unit, which obtains partial characteristic points as partial characteristic points for each obtained partial regression curve; and an operation processing unit (163), which functions as a peak characteristic point acquisition unit for obtaining peak characteristic points in the measured time series data based on the partial characteristic points.

Description

Technical Field

[0001] The present invention relates to chromatography technology of a liquid chromatograph or the like, and in particular to a data processing device and a data processing method for a chromatograph and a chromatograph. Background Art

[0002] In the chromatograph, the type and amount of components contained in the analytical sample are obtained based on waveform data with time on the horizontal axis and signal intensity on the vertical axis. At this time, based on the data detected by the device, peak characteristic points such as the start point of the rise and the end point of the fall of the signal intensity are detected and waveform processing is performed.

[0003] As a technology for performing waveform processing as described above, for example, with the theme of "providing a chromatography device that can easily perform curve fitting using the nonlinear least squares method even for a chromatogram having multiple overlapping peaks", the following technology is known: "In a chromatography data processing device that processes data of a chromatogram obtained by separating a sample to be measured by a column and detecting the separated sample, for any time region in which there are multiple peaks of the chromatogram, fitting processing is performed on each peak compared to an earlier time region or a later time region, and the peak subjected to the fitting processing is subtracted from the chromatogram in the arbitrary time region, thereby decomposing the multiple peaks of the chromatogram."

[0004] In addition, with the theme of "making it easy to find appropriate feature points through the data processing device of the chromatograph", the following technology is also known: "The data processing device of the chromatograph that performs data processing based on the plotting data measured by the chromatograph comprises: a virtual curve calculation unit, which obtains a virtual curve based on the measured plotting data; a provisional feature point acquisition unit, which obtains provisional feature points based on the obtained virtual curve; and an operation processing unit, which functions as an actual plotting data feature point extraction unit, which extracts actual plotting data feature points corresponding to the above-mentioned provisional feature points in the above-mentioned measured plotting data" (for example, refer to patent document 2).

[0005] [Prior art literature]

[0006] [Patent Document]

[0007] [Patent Document 1] Japanese Patent Application Publication No. 2006-177980

[0008] [Patent Document 2] Japanese Patent Application Publication No. 2019-174399 Summary of the invention

[0009] [Problems to be Solved by the Invention]

[0010] In the technique like Patent Document 1, the calculated virtual curve and feature points are easily greatly affected by noise.

[0011] In addition, in the technology such as Patent Document 2, it takes a long time to study the type of appropriate regression curve after finding the feature points. That is, it is necessary to study which type of regression curve to select and which time axis range in the measurement data to perform regression and search for feature points.

[0012] The present invention has been made in view of the above-mentioned problems, and an object of the present invention is to easily obtain peak characteristic points in a short time without being easily affected by noise.

[0013] [Methods for solving the problem]

[0014] In order to achieve the above-mentioned purpose, the present invention is a data processing device for a chromatograph, which performs data processing based on plotted data measured by the chromatograph, and is characterized in that the data processing device for the chromatograph comprises: a partial regression curve calculation unit, which obtains partial regression curves as partial regression curves for a plurality of data groups obtained by grouping a specified number of measured time series data; a partial feature point acquisition unit, which obtains partial feature points as partial feature points for each obtained partial regression curve; and a peak feature point acquisition unit, which obtains peak feature points in the measured time series data based on the partial feature points.

[0015] Thus, by obtaining a partial regression curve for a predetermined number of time series plotted data, feature points can be obtained in a relatively short time. Furthermore, it is also possible to easily perform regression using a function of a relatively low degree, and thus it is also possible to easily suppress a regression curve that is offset due to the influence of noise.

[0016] [Effects of the invention]

[0017] In the present invention, the peak characteristic point can be easily obtained in a short time without being affected by noise. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a block diagram showing a schematic configuration of a chromatograph.

[0019] Figure 2 This is a block diagram showing a schematic configuration of a data processing device for a chromatograph.

[0020] Figure 3 is a diagram showing an example of a chromatogram.

[0021] Figure 4 It is a diagram showing an example of schematic data for explaining an example of data processing.

[0022] Figure 5It is a diagram showing an example of a curve involved in data processing.

[0023] Figure 6 is a diagram showing an example of an envelope method in a scanning type PCF;

[0024] Description of Reference Numerals

[0025] 100 Liquid Chromatograph

[0026] 110 Mobile phase container

[0027] 120 Pumps

[0028] 130 Autosampler

[0029] 140 columns

[0030] 141 Column oven

[0031] 150 Detector

[0032] 160 Data processing device

[0033] 161 Control Processing Unit

[0034] 161a Control Unit

[0035] 161b Measurement condition setting unit

[0036] 161c Records Department

[0037] 162 Data retention unit

[0038] 163 Calculation Processing Unit

[0039] 163a Signal Processing Unit

[0040] 163b Operation unit

[0041] 163c Judgment Department

[0042] 170 Display unit DETAILED DESCRIPTION

[0043] Hereinafter, embodiments of the present invention will be described in detail based on the drawings.

[0044] (Overall description)

[0045] (Structure of Liquid Chromatograph 100)

[0046] Figure 1The schematic structure of the liquid chromatograph 100 is shown. The liquid chromatograph 100 includes: a mobile phase container 110 storing a liquid as a mobile phase; a pump 120 for conveying the mobile phase; an automatic sampler 130 for injecting a sample; a column 140 for separating the components in the sample by keeping the temperature constant through a column oven 141; a detector 150 for detecting the separated components; a data processing device 160 for processing the detection results; and a display unit 170 for displaying the processing results. In addition, the various components constituting the liquid chromatograph 100 can be configured in the same manner as a general device, except for the processing content of the data processing device 160, and therefore detailed description is omitted.

[0047] (Detailed Structure of Data Processing Device 160)

[0048] like Figure 2 As shown, the data processing device 160 includes a control processing unit 161 , a data holding unit 162 , and an arithmetic processing unit 163 .

[0049] The control processing unit 161 controls the overall operation of the liquid chromatograph 100 and includes a control unit 161a, a measurement condition setting unit 161b for setting measurement conditions based on an operation of an operation panel (not shown), and a recording unit 161c for recording measurement results.

[0050] The data storage unit 162 stores data processed based on the measurement results and the like.

[0051] The calculation processing unit 163 performs processing based on the measurement results and functions as a partial regression curve calculation unit, a partial feature point acquisition unit, and a feature point acquisition unit. Specifically, for example, it includes a signal processing unit 163a that performs D / A conversion on the analog signal output from the detector 150, a calculation unit 163b that extracts and analyzes feature points, and a determination unit 163c that determines the analysis results.

[0052] (Data processing action)

[0053] In the liquid chromatograph 100, the measurement operation obtains, for example, Figure 3 The waveform data shown in FIG. 1 is waveform data in which the horizontal axis is time and the vertical axis is signal strength.

[0054] Since the relationship between time and components is known in advance, the components contained in the analysis sample are determined by the retention time of the peak top of the waveform data on the horizontal axis (qualitative processing).

[0055] In addition, the amount of the component contained is measured based on the peak area of ​​the waveform data (quantitative calculation processing).

[0056] These processes are extracted by Figure 3The peak feature points such as the peak apex, start point, end point, valley point, shoulder point, etc. are shown as typical examples, or a baseline segment is set based on these peak feature points.

[0057] Through the processing of the calculation processing unit 163 , the peak feature points are extracted as shown below.

[0058] (Extraction of peak feature points)

[0059] The plot data P1 to P1000 of 1000 points, for example, for each predetermined sampling period, which are discrete data actually detected by the detector 150, are obtained. Figure 4 As shown, continuous plot data P_Start to P_End in a predetermined time interval from time t_Start to time t_End are set as processing targets. Here, time t_Start and time t_End may be determined in advance or may be designated by a user.

[0060] From the start point to the end point of the time interval, the continuous 7-point plot data P_n1 to P_n7 are selected while moving 1 point each time, for example Figure 5 As shown, a partial regression curve approximating the plotted data P_n1 to P_n7 at the seven points is obtained, that is, a quadratic curve C. The quadratic curve C is obtained by, for example, the least squares method. Alternatively, a curve of a cubic function or higher, a Gaussian curve, etc., or various approximation methods other than the least squares method may be used.

[0061] For example, when the starting point of the time interval is P1 and the end point is P15, the quadratic curves C_P1~P7 based on the plotting data P1~P7, the quadratic curves C_P2~P8 based on the plotting data P2~P8, ..., the quadratic curves C_P8~P14 based on the plotting data P9~P15, and the quadratic curves C_P9~P15 based on the plotting data P9~P15 are respectively calculated.

[0062] That is, when there are m points of plotted data within a time interval, and when n consecutive points of plotted data (where m>n) are selected, m-n+1 quadratic curves C are obtained.

[0063] For the group of quadratic curves C obtained in the entire area within the time interval, confirm whether the differential feature points (maximum points, minimum points) serving as partial feature points are included within the range of the selected data point group in each quadratic curve. If the differential feature points are included, the time corresponding to each differential feature point is used as a candidate for the time of the peak feature point (peak apex, peak start point, peak end point). Based on the data points at the time corresponding to these differential feature points, the peak feature points (peak apex, peak start point, peak end point) of the approximate waveform of the time series data (plotted data) of the detection signal are set, and the peak feature points are output.

[0064] Specifically, for the group of quadratic curves C obtained in the entire region within the time interval, the peak characteristic points are obtained according to the following procedure.

[0065] (1) That is, for example Figure 5 (a) plot data P2, Figure 5 The point on the quadratic curve C at the time of the plot data P13 of (i) is set as a partial feature point as a minimum point. Figure 5 The points on the quadratic curve C at the time of the plot data P8 of (d) to (f) are set as partial characteristic points as local maximum points.

[0066] (2) However, when the time of such a maximum point or minimum point is not within the time range for obtaining each partial regression curve, it can be ignored. In addition, not only within the time range for obtaining each partial regression curve, but also within the time range of ±2 sampling periods from the center of the time range, it can be set as a partial feature point.

[0067] (3) In addition, Figure 5 In the examples (d) to (f), all of them reach a maximum at the time of the plotted data P8, but when there are multiple quadratic curves with such maximum points in a specified time range and the times of reaching the maximum points are different from each other, these maximum points can also be summarized as follows.

[0068] For example, when the difference between the times of the maximum points is within two sampling periods (sense range), it can be assumed that there is a peak, and the time when the maximum value or the raw data in these maximum points becomes the maximum is set as the peak feature point. The above-mentioned sense range is set, for example, based on the assumed peak width, sampling period, etc., and is not limited to a fixed setting, and can also be set by the user.

[0069] Here, liquid chromatography is a measurement in which the measured value changes with time, and each measured value is a continuous value per unit time. In contrast, due to the separation performance of the column and the influence of noise, the data collected per unit time changes. Therefore, the data in a certain range can be considered as a block of a phenomenon that cannot be separated (represented by one data).

[0070] In addition, for example, even if there is a maximum point at a time that is more than 2 sampling cycles away, if all the plotted raw data at the time in between are within the specified noise amplitude, it can be determined that these plotted data are data of a peak affected by noise, for example, the side with the larger raw data can be used as the peak characteristic point.

[0071] Here, the noise amplitude specified above is, for example, as follows. That is, in the measurement data, a value of a certain amount of amplitude is added due to various factors accompanying the measurement. This value is defined only by the amplitude, and there are non-periodic components and periodic components, but the components whose period and acceptance correlation with the measurement peak are not fixed are considered to be noise components. That is, in the measurement value, the noise amplitude amount is considered to be an uncertain value. That is, even if the measurement value A>measurement value B, if the measurement value A-measurement value B<noise amplitude, the difference between the exact values ​​of A and B is uncertain. The noise amplitude as described above can be set, for example, according to the amplitude of the measurement value at a time sufficiently away from the peak point.

[0072] As described above, by obtaining a partial regression curve for a predetermined number of time series plotted data, feature points can be obtained in a relatively short time. Moreover, it is also possible to easily use a function of a relatively low degree for regression, etc., so it is also possible to easily suppress a regression curve that is offset due to the influence of noise, etc. In particular, it is also possible not to select the type of regression curve or the time axis range for obtaining the regression curve, so it is easy to quickly and objectively perform processing.

[0073] In addition, the group of plotting data used to obtain the regression curve of each part is not limited to the group in which the sampling period is shifted by 1, and even if the sampling period is shifted by 2 or more, as long as the peak characteristic point is obtained with the necessary accuracy, the plotting data of each group is not necessarily limited to continuous plotting data, and scattered data, aggregated data, etc. can also be used.

[0074] (Scanning CCF: Complementally Curve Fitting and Scanning PCF: Scanning Partial Curve-Fitting)

[0075] The Savitzky-Golay method is a type of moving average and a smoothing method, but the purpose of suppressing the influence of noise is the same as that of the present embodiment. In the moving average, a window of a certain time interval is cut out, and a representative value is assigned to the representative point at the central moment of the window. The window is, for example, 7 or 9 points, and the representative values ​​of the detection signals representing the moments are sequentially replaced with moving average values ​​by scanning in the reverse direction of the moments. Alternatively, it can also be scanned forward. In short, set the function form to y=f(t), summarize the signal data points within the time window, and make the value of y of the target variable correspond to the moment t of the explanatory variable. If the weighting coefficient of the Savitzky-Golay method is switched, the differential coefficient such as y=f'(t) can also be calculated. The smoothing method is just a method to offset noise, but the purpose of this embodiment is to find the true (assumed to be true) curve (true curve).

[0076] In this embodiment, there is also a time window, and the window is scanned, which is common to the Savitzky-Golay method, etc. However, the characteristic of this embodiment is that, instead of assigning a representative value to the center time, an imaginary curve is regressed in each window, and as described later, there are characteristics of a scanning type CCF or a scanning type PCF.

[0077] In the case of quadratic curves, sometimes a vertex or valley point is obtained. For example, in the case of analyzing the vicinity of a vertex, the t coordinate and y coordinate of the vertex are determined for each window. That is, the vertex time t and signal intensity y are the differential feature points of each regression curve. Even if the window is shifted little by little, as long as the vertex coordinates (t, y) are roughly the same, it can be determined that the place is the real vertex. The same is true for the valley point. The method of finding this differential feature point is the scope of scanning CCF.

[0078] This method is not limited to 1st and 2nd order, and can also be extended to higher order polynomials. The imaginary curve is also not limited to probability density functions such as normal distribution, exponential functions, logarithmic functions and other analytical mathematical functions. Furthermore, it can be extended to any function. Since the curve fitting process performed in addition to each time window is scanned in sequence, this method is called scanning CCF (Complementally Curve Fitting). In order to obtain differential feature points, the curve fitting process is used in addition, which is the origin of the name.

[0079] On the other hand, the method of finding the true curve means using the same time window for the entire area of ​​the true curve and sequentially scanning the curve fitting process partially, which is called scanning PCF (Scanning Partial Curve-Fitting). Scanning PCF is not a supplementary use of curve fitting processing, but is mainly performed to find the true curve. Scanning CCF is used in differential feature point search, and scanning PCF is used in true curve search.

[0080] Furthermore, similar signal analysis methods include spline interpolation such as cubic spline. Spline interpolation also uses polynomials, but it is a method of interpolating between actual data points, and does not aim to find the true curve buried in noise. In addition, the above-mentioned moving average does not have an interpolation function. The scanning PCF removes the influence of noise and also has an interpolation function.

[0081] (Examples of Scanning CCF and Scanning PCF)

[0082] The scanning CCF finds differential feature points such as vertices and valley points, but in the scanning PCF, the difference of the imaginary curve in the interval repeated between the two windows, that is, the residual, etc. is considered. For example, by evaluating the residual sum of squares of the two imaginary curves, it is possible to compare whether the imaginary curves are consistent or not so consistent. By considering the consistent group of imaginary curves as the true curve, the true curve in the entire area of ​​the time window interval can be found. In fact, when the y-coordinates of the longitudinal axis direction of each time window are consistent within a certain allowable width, the consistent imaginary curves are extracted, and the true curve is generated by connecting them adjacent to each other. The scanning PCF has a feature in finding the true curve, which is different from the smoothing process that offsets the noise component.

[0083] In a scanning PCF, it is preferred to use a quadratic or cubic polynomial for the imaginary curve. However, the special case of the curve is a linear straight line, so the regression line is not unusable for the scanning PCF. As mentioned above, it is the duty of this embodiment to study the degree of difference between adjacent regression curves. In the case of a linear straight line, consider the consistency and inconsistency of the first-order differential coefficient (i.e., the slope) in each time window. If they are roughly consistent, it can be determined that the slope of the consistent interval is roughly constant. In the case of finding the rate of change of the first-order differential coefficient, i.e., the regression line, it is sufficient to regress to a first-order polynomial, i.e., a straight line. This embodiment is a method for finding a true curve containing a true straight line from deviated, i.e., noise-overlapping detection signal data. At the same time, if the regression function is a differentiable function, the differential coefficient of the function can also be calculated.

[0084] The shoulder point, i.e., the inflection point, which is a characteristic of the shoulder peak, can also be found by expanding this logic. The inflection point is located at the moment when the second-order partial differential coefficient becomes zero. If each time window is regressed to a quadratic function, the first-order differential coefficient can of course be obtained, and the inflection point can also be found by observing its rate of change per unit time. The first-order differential coefficient is not a representative value at each moment like the smoothing process, and it is also a feature that it is found as a first-order differential curve in the scanning PCF.

[0085] The smoothing process mainly moves the time window to the side, but this embodiment does not necessarily need to move it frame by frame. Scanning can be performed regardless of skipping one or two. In addition, the number of points in the time window can be selected as 7 points, 9 points, 11 points, etc. Furthermore, the expected peak width can be used as input to automatically calculate the sampling period and the number of points in the time window. In the case of skipping one or skipping two, a bunching process of adding data points can also be performed. In fact, both scanning PCF and scanning CCF preferably have 20 to 30 data points per peak, or at most 50 points. It is important for this embodiment to use the representative peak width as an input value and automatically calculate the sampling period, the number of bunching points, etc. In other words, the representative sampling period of HPLC (high performance liquid chromatography) is hundreds of milliseconds, and the representative sampling period of UHPLC (ultra-high performance liquid chromatography) is tens of milliseconds.

[0086] In the case of scanning CCF, the comparison method between several time windows is also considered. For example, if the valley point time t of each time window is within the positive and negative sampling period, it can be judged as stable. Similarly, the vertical axis y value of the valley point is within a certain allowable range, which is also a criterion for judgment. The allowable range of the vertical and horizontal rectangles can also be set at the coordinates (t, y).

[0087] In the case of scanning PCF, there is also an allowable width for finding the true curve. An envelope of a certain width such as positive and negative mV is set for one of the two virtual curves to be compared in the vertical direction, and it is set to be suitable when the other virtual curve exists within the envelope. It is not suitable from the position outside the envelope. Through the comparison of the residual sum of squares mentioned above, the envelope is sometimes more convenient. Use Figure 6 This will be described in detail later.

[0088] As a more advanced technique, the contribution rate method is considered. A series of time windows are numbered T1, T2, T3, etc., and the result of the regression operation is represented by the contribution rate R1. 2 , R2 2 、R3 2Determine whether the regression process is good or bad. The more successful the regression process is, the better the explanation is given by the regression curve, and the closer the contribution rate is to 1. The regression curve that performs the regression well is reliably weighted to find the index of the true curve. In addition, the method of using the contribution rate as an index can also be used when finding the differential feature point.

[0089] Finally, if the first-order differential coefficient is obtained from the true curve through scanning PCF, the start or end point of the peak can also be found. This can also be understood as finding the differential feature point. In addition, the true curve obtained from the scanning PCF is the true peak waveform itself, so the accurate retention time and peak area can be calculated from it.

[0090] (Simple scanning PCF)

[0091] Simply put, for example, in the case of a time window that aggregates 5 data points, there is a weighted regression method as follows: the data point in the center is assigned a weight of 1, the data points next to it are assigned a weight of 2 / 3, the data points next to it are assigned a weight of 1 / 3, etc., and the least squares method is used for calculation. Then, the coefficient group of the regression curve such as the quadratic polynomial is assigned at the time of the center point. By scanning the center moment, each coefficient group is assigned to each center moment in turn. In the case of interpolation, these adjacent regression curves are mixed point by point. That is, at this moment, the ratio of the regression curve at this moment is set to 1.0, and the ratio of the regression curve at the adjacent moment is set to 0.0 and added. At the internal dividing point of the moment adjacent to the moment, according to the moment difference, each regression curve is mixed according to the ratio of the internal dividing point. At the midpoint, they are mixed at a ratio of 0.5 respectively. At adjacent moments, the ratio of the regression function at adjacent moments is 1.0. In this way, the resulting curve is a true curve.

[0092] (Envelope method)

[0093] One of the two imaginary curves may or may not coincide with the other imaginary curve within a certain allowable width. Figure 6 The envelope method is shown in FIG. First, the five measured data points representing the time before and after t0 are used to regress the imaginary curve 0. -2 to -1 , t0, t1, t2 are five points. These subscripts are time marks of the imaginary curve 0.

[0094] Next, one data point is shifted backward and regressed to virtual curve 1. The degree to which virtual curve 1 is consistent with virtual curve 0 is evaluated. Therefore, an envelope is set above and below virtual curve 0 with the input width. Virtual curve 1 is repeated at 4 points from t-1 to t2 according to the above-mentioned time mark relative to virtual curve 0. Figure 6In the example, at these four points, the virtual curve 1 at t0 and t1 exists within the envelope. Therefore, the virtual curve 0 is considered to be consistent with the virtual curve 1 only at these two points. In this case, the true curve is calculated as shown below (Formula 1) and is only applied to the consistent moments. y0 represents the virtual curve 0, y1 represents the virtual curve 1, etc., and y represents the true (assumed to be true) curve. The coefficient of the true curve y is the average value of the coefficients of the virtual curves.

[0095] [Formula 1]

[0096] y0=a0+b0t+c0t 2 ...

[0097] y1=a1+b1t+c1t 2 ...

[0098]

[0099] exist Figure 6 In this case, regarding the true curve y, according to n=1 (n is the number of imaginary curves used to derive the true curve y - 1), which means imaginary curve 0 and imaginary curve 1, the moment only applies to the time intervals t0 and t1 when the imaginary curve 0 is consistent with the imaginary curve 1. Since the true curve y is obtained, its first-order differential curve y' and second-order differential curve y" are calculated based on the differential coefficients of the polynomial.

[0100] Here, the relationship between the upper and lower allowable widths of the envelope and the possibility of the true curve, that is, the reliability, is described. If the allowable width is narrow, the reliability is improved, but the possibility that the adjacent imaginary curve does not enter the envelope increases. In extreme cases, only one true curve is allocated at the center moment. On the contrary, if the allowable width is too wide, the number of object imaginary curves for calculating the average value increases. Although the average value can be obtained, the deviation becomes larger and the reliability decreases. A reference for an appropriate allowable width is preferably about a percentage of the peak height. Of course, the allowable width of the envelope can also be automatically input based on this percentage.

[0101] (Method of time-variable regression interval)

[0102] In the above embodiment, the time interval of the 5-point data is shifted backward one by one. Assuming that the imaginary curve obtained from the previous 5 points is well suited to the actual data, there is also a case where the next 6th data point can be described with a certain allowable width. It can be imagined by the aforementioned envelope method, or it can be grasped by the residual of the imaginary curve and the actual data. For example, the average value of the residual sum of squares in the case of 5 points is compared with the average value of the residual sum of squares in the case of 6 points. If the former is smaller, the regression process is stopped at 5 points. On the contrary, if the average value of the residual sum of squares obtained from 6 points in the latter is not large, it is judged that the regression curve can be extended, and the regression process is performed again in the time interval of 6 points. As a result, a new imaginary curve of the 6-point interval is obtained. By repeating this process, the regression time interval can be made variable. The method of the time-variable regression interval is different from the aforementioned envelope method. It is not a method of comparing two imaginary curves, but a method of comparing imaginary curves with actual data point sequences. In addition, for example, the method can be extended to compare two virtual curves by taking the possible curve obtained in the 7-point interval as the reference and moving one to the next 5-point time interval with the 8th data point as the end point. Only in the interval where the two virtual curves are consistent, the two virtual curves are connected to generate a true curve.

[0103] (Gaussian join using machine learning)

[0104] It is believed that the ideal shape of the peak of the chromatogram is a normal distribution, i.e., a Gaussian. Furthermore, most of them are represented by a probability density function such as a trailing peak, i.e., a Gaussian EMG corrected by an exponential function. Referring to the idea of ​​the scanning PCF of the present embodiment, even if it is not necessarily represented by a function, a method of finally connecting to a true curve can be thought of on the basis of partially being suitable for Gauss. Simply put, the peak can be divided into three parts in the vertical direction, which are divided into the part near the vertex, the middle part near the inflection point, and the bottom part. In each segmented part, the vertex of the Gaussian, the middle part of the left and right, and the bottom part of the left and right are approximated with different Gaussians. This may not be a vertical 3-segmentation, but the number of segments is increased to 4-segmentation and 5-segmentation. The vertical 3-segmentation is 5 intervals in the time direction, the vertical 4-segmentation is 7 intervals in the time direction, and the vertical 5-segmentation is 9 time intervals.

[0105] Regarding the calculation method of the appropriate number of divisions, Gaussian or EMG, it is expected that machine learning will be applied to the following topics: what kind of function should polynomials, etc. be regressed to according to the situation. A large number of chromatographic peak examples are pre-input into the machine learning with teachers and applied to individual analysis. The phenomenon of overlapping peaks and shoulder peaks is added, which further complicates the problem, but the adaptability of this embodiment is also increased accordingly. The pre-input examples can be measured data or artificially generated chromatographic peaks.

[0106] The Gaussian connection method using this machine learning is also a method of comparing a virtual curve and an actual data time series, similar to the above-mentioned method of time-variable regression intervals.

Claims

1. A data processing device for a chromatograph, which performs data processing based on plotted data measured by the chromatograph, characterized in that: The data processing device of the chromatograph has: a partial regression curve calculation unit for obtaining partial regression curves as partial regression curves for each of a plurality of data groups formed by grouping a predetermined number of measured time series data; a partial feature point acquisition unit that obtains partial feature points as partial feature points for each obtained partial regression curve; as well as The peak feature point acquisition unit obtains the peak feature point in the measured time series data based on the partial feature points.

2. The data processing device for a chromatograph according to claim 1, characterized in that: The regression curve is any one of a curve of a function of degree 2 or higher and a Gaussian curve.

3. The data processing device of the chromatograph according to claim 2, characterized in that: The multiple data groups are data groups that are offset from each other on the time axis.

4. The data processing device for a chromatograph according to claim 3, characterized in that: The partial feature points are at least one of a maximum point and a minimum point.

5. The data processing device for a chromatograph according to claim 2, characterized in that: The peak feature point is any one of a peak apex, a peak start point, and a peak end point.

6. A chromatograph, characterized in that: The chromatograph has: a chromatography unit that separates and measures components contained in a sample; and The data processing device of the chromatograph according to claim 1.

7. A method for processing data of a chromatograph, which performs data processing based on plotted data measured by the chromatograph, characterized in that: The data processing method of the chromatograph has the following features: a partial regression curve calculation step of obtaining partial regression curves as partial regression curves for each of a plurality of data groups formed by grouping a predetermined number of measured time series data; A partial feature point acquisition step, for each partial regression curve obtained, obtaining a partial feature point as a partial feature point; as well as The peak feature point acquisition step is to obtain the peak feature point in the measured time series data based on the partial feature point.

8. A data processing device for a chromatograph, which processes data based on plotted data measured by the chromatograph, characterized in that: The data processing device of the chromatograph has: a partial regression curve calculation unit for obtaining partial regression curves as partial regression curves for each of a plurality of data groups formed by grouping a predetermined number of measured time series data; a partial curve comparison section that compares the respectively obtained partial regression curves with each other; as well as A comprehensive regression curve acquisition unit is configured to obtain, based on the comparison result, a regression curve in the measured time series data, the regression curve having a larger range than each of the partial regression curves.

9. A data processing device for a chromatograph, which processes data based on plotted data measured by the chromatograph, characterized in that: The data processing device of the chromatograph has: a partial regression curve calculation unit for obtaining partial regression curves as partial regression curves for each of a plurality of data groups formed by grouping a predetermined number of measured time series data; a partial curve comparison unit that compares the respectively obtained partial regression curves with the time series data; as well as A comprehensive regression curve acquisition unit is configured to obtain, based on the comparison result, a regression curve in the measured time series data, the regression curve having a larger range than each of the partial regression curves.

Images