Spectroscopic analysis system and spectroscopic analysis method

Abstract

To provide a spectroscopic analysis system, etc. that makes it possible to appropriately conduct measurement even when measurement time, etc is limited.SOLUTION: A spectroscopic analysis system 100 comprises: an operation panel 21 that accepts the upper-limit value of measurement time of a spectroscopic spectrum and / or the lower-limit value of measurement accuracy as a user set condition regarding the spectroscopic spectrum measurement of a sample; and a control unit 32 that derives a prescribed recommended measurement condition that would satisfy the user set condition and causes a display unit 22 to show the recommended measurement condition. The recommended measurement condition is at least one of the wavelength range of light that is used for measuring spectroscopic spectra, the wavelength sampling interval of light, the slit width of diffraction gratings 2a, 8a of a spectroscope that spectrally separates light, and the wavelength sweep speed of light.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a spectroscopic analysis system and the like. [Background technology]

[0002] Spectroscopic analysis is known as a technique for determining the composition or quantifying the concentration of a substance by measuring the spectrum of light absorbed or emitted by the substance. Regarding fluorescence fingerprint analysis, which is one such spectroscopic analysis method, for example, Patent Document 1 describes that "fluorescence intensity is measured while the excitation wavelength for irradiation and the fluorescence wavelength for observation are changed stepwise for each of a plurality of extracted samples, and multiple pieces of fluorescence fingerprint information are obtained as fluorescence fingerprint continuum information."

[0003] Furthermore, Non-Patent Document 1 states that "In fluorescence fingerprint (or excitation-emission matrix) measurements, the intensity of fluorescence is measured while changing both the wavelength conditions of the excitation light and the wavelength conditions of the fluorescence being observed (solid arrow in Figure 2). In other words, a brute force investigation is conducted to determine whether electronic excitation occurs in the target sample and whether fluorescence is emitted." [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Publication No. 2020-76612 [Non-patent literature]

[0005] [Non-Patent Document 1] Mizuki Tsuta et al., "Food Quality Evaluation Technology Using Fluorescence Fingerprinting and Its Applications," Journal of the Agricultural Chemical Society of Japan, "Chemistry and Biology," Vol. 53, No. 5, 2015, pp. 285-292 Summary of the Invention [Problem to be solved by the invention]

[0006] For example, if you try to obtain high-resolution measurement results across the entire wavelength range using spectroscopic analysis, it takes a long time to measure. On the other hand, if you measure only specific wavelengths using filter spectroscopy, there is a risk that important spectral data will not be obtained. It is desirable to be able to perform appropriate measurements even when there are restrictions on the measurement time, etc., based on spectroscopic analysis.

[0007] Therefore, an object of the present invention is to provide a spectroscopic analysis system or the like that can perform appropriate measurements even when there are restrictions on measurement time or the like. [Means for solving the problem]

[0008] In order to solve such problems, the spectroscopic analysis system of the present invention includes an input unit that accepts input of at least one of an upper limit value of the measurement time for the spectroscopic analysis spectrum and a lower limit value of the measurement accuracy as user-defined conditions for measuring the spectroscopic analysis spectrum of a sample, and a control unit that derives predetermined recommended measurement conditions that satisfy the user-defined conditions and displays the recommended measurement conditions on a display device, wherein the recommended measurement conditions are at least one of the wavelength range of light used in measuring the spectroscopic analysis spectrum, the sampling interval of the wavelength of the light, the slit width of the diffraction grating of a spectrometer that disperses the light, and the sweep speed of the wavelength of the light. [Effects of the Invention]

[0009] According to the present invention, it is possible to provide a spectroscopic analysis system or the like that can perform appropriate measurements even when there are restrictions on measurement time or the like. [Brief explanation of the drawings]

[0010] [Figure 1] 1 is a configuration diagram of a spectroscopic analysis system according to a first embodiment. [Figure 2A] FIG. 2 is a diagram showing an example of a spectroscopic analysis spectrum obtained by the spectroscopic analysis system according to the first embodiment. [Figure 2B]FIG. 2 is a diagram showing a state in which a spectroscopic analysis spectrum obtained by the spectroscopic analysis system according to the first embodiment has been subjected to a predetermined preprocessing. [Figure 2C] FIG. 2 is an explanatory diagram of one-dimensional data of a spectroscopic analysis spectrum obtained by the spectroscopic analysis system according to the first embodiment. [Figure 3] FIG. 2 is a functional block diagram of a control unit included in the spectroscopic analysis system according to the first embodiment. [Figure 4] 4 is a flowchart showing the processing of a control unit of the spectroscopic analysis system according to the first embodiment. [Figure 5] 4 is a display example of a setting screen for measurement conditions and the like in the spectroscopic analysis system according to the first embodiment. [Figure 6] 4 is a display example of a screen showing measurement and analysis results in the spectroscopic analysis system according to the first embodiment. [Figure 7A] FIG. 2 is an explanatory diagram showing how the range of the target wavelength region is changed when the GAWLSPLS method is used in the spectroscopic analysis system according to the first embodiment. [Figure 7B] FIG. 2 is an explanatory diagram showing a measurement wavelength range set based on the GAWLSPLS method when the GAWLSPLS method is used in the spectroscopic analysis system according to the first embodiment. [Figure 8] FIG. 2 is an explanatory diagram showing an example in which the fluorescence wavelength is fixed and the excitation wavelength is swept to a predetermined value in the spectroscopic analysis system according to the first embodiment. [Figure 9] FIG. 4 is an explanatory diagram showing experimental results of a fluorescence fingerprint analysis performed by the spectroscopic analysis system according to the first embodiment. [Figure 10A] 10 is a flowchart showing the processing of a control unit of a spectroscopic analysis system according to a second embodiment. [Figure 10B] 10 is a flowchart showing the processing of a control unit of a spectroscopic analysis system according to a second embodiment. [Figure 11A] FIG. 11 is a diagram showing experimental results when RMSECV is used as an index of goodness of fit in the spectroscopic analysis system according to the fourth embodiment. [Figure 11B]FIG. 10 is a diagram showing experimental results when C1 is used as an index of goodness of fit in the spectroscopic analysis system according to the fourth embodiment. [Figure 12A] FIG. 13 is a diagram showing prediction results of a verification sample when RMSECV is used as an index of goodness of fit in the spectroscopic analysis system according to the fourth embodiment. [Figure 12B] FIG. 11 is a diagram showing the prediction results of the verification sample when C1 is used as an index of goodness of fit in the spectroscopic analysis system according to the fourth embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0011] First Embodiment FIG. 1 is a configuration diagram of a spectroscopic analysis system 100 according to the first embodiment. The spectroscopic analysis system 100 shown in FIG. 1 is a system that performs, for example, composition determination and concentration quantification of a sample based on spectroscopic analysis. Spectroscopic analysis is a method for determining the composition and concentration quantification of a sample based on the spectrum of light absorbed or emitted by the sample. Samples that can be subjected to such spectroscopic analysis include, but are not limited to, human or animal cells and blood, as well as culture solutions, foods, beverages, dyes, and minerals (e.g., gemstones). Spectroscopic analysis may also be used to measure airborne particles. Spectroscopic analysis methods include fluorescence fingerprint analysis and absorption spectroscopy, but the following describes the use of fluorescence fingerprint analysis as an example.

[0012] As shown in Fig. 1, the spectroscopic analysis system 100 includes a photometer unit 10, an operation unit 20, and a data processing unit 30. The photometer unit 10 is a device that irradiates a sample M1 with excitation light and measures the fluorescence emitted from the sample M1 in response to this irradiation. The photometer unit 10 is shielded by a housing (not shown) to prevent light from entering the photometer unit 10 from outside.

[0013] For example, when light is irradiated onto sample M1, the electrons contained in the molecules of sample M1 absorb the energy of the light, transitioning to a higher energy level and becoming excited. The light that causes this phenomenon is called "excitation light." Furthermore, the light emitted when the excited electrons return to their original ground state is called "fluorescence."

[0014] In fluorescence fingerprint analysis, the wavelength of the excitation light irradiated onto the sample M1 and the wavelength of the fluorescence to be measured are changed, and the fluorescence intensity is measured. This results in a spectroscopic spectrum (3D fluorescence spectrum, fluorescence fingerprint) consisting of three components: excitation wavelength, fluorescence wavelength, and fluorescence intensity. Because such a spectroscopic spectrum is unique to a substance, it is possible to determine the composition and quantify the concentration of the substance (sample). In fluorescence fingerprint analysis, the "light used to measure the spectroscopic spectrum" is the excitation light and fluorescence.

[0015] 1, photometer section 10 includes light source 1, excitation-side spectroscope 2 (spectroscope), beam splitter 3, monitor detector 4, excitation-side filter 5, sample mounting section 6, fluorescence-side filter 7, fluorescence-side spectroscope 8 (spectroscope), and detector 9. In addition to the components described above, photometer section 10 also includes excitation-side pulse motor 11, fluorescence-side pulse motor 12, excitation-side filter pulse motor 13, and fluorescence-side filter pulse motor 14.

[0016] The light source 1 emits a predetermined light. Examples of such light source 1 include a xenon lamp, a halogen lamp, and a medium-pressure mercury lamp. The excitation-side spectroscope 2 separates the light emitted from the light source 1 into predetermined wavelengths. The excitation-side spectroscope 2 includes a diffraction grating 2a for extracting light of a predetermined range of wavelengths from light containing various wavelengths. The diffraction grating 2a is configured, for example, by providing a plurality of parallel fine grooves at predetermined intervals on the surface of an optical material. The angle of incidence of light onto the diffraction grating 2a changes depending on the rotation angle of the diffraction grating 2a, and accordingly, the wavelength of the light extracted from the diffraction grating 2a changes. The excitation-side pulse motor 11 adjusts the rotation angle of the diffraction grating 2a of the excitation-side spectroscope 2 based on commands from the control unit 32.

[0017] The beam splitter 3 splits the light from the excitation-side spectroscope 2 into two beams (splits the light beam into two). The monitor detector 4 measures the intensity of one of the beams split by the beam splitter 3. The measurement result from the monitor detector 4 is output as a predetermined electrical signal to the A / D converter 31. The excitation-side filter 5 is a filter that transmits light of a predetermined wavelength range and blocks the remaining light, and is disposed between the beam splitter 3 and the sample mounting unit 6. The excitation-side filter 5 includes, for example, multiple cut filters (not shown). A predetermined cut filter selected from the multiple cut filters is positioned on the optical path by the excitation-side filter pulse motor 13. The excitation-side filter pulse motor 13 moves a predetermined cut filter included in the excitation-side filter 5 onto the optical path based on a command from the control unit 32.

[0018] The sample mounting section 6 is a holder for mounting a sample M1 to be subjected to fluorescence fingerprint analysis. If the sample M1 is a liquid or gas, a container or the like (not shown) containing the sample M1 is mounted on the sample mounting section 6. Then, light (excitation light) transmitted through the excitation-side filter 5 is incident on the sample M1.

[0019] The fluorescence side filter 7 is a filter that transmits light of a predetermined range of wavelengths and blocks the remaining light, and is disposed between the sample mounting unit 6 and the fluorescence side spectroscope 8. The fluorescence side filter 7 includes, for example, a plurality of cut filters (not shown). A predetermined cut filter selected from the plurality of cut filters is then positioned on the optical path by the fluorescence side filter pulse motor 14. The fluorescence side filter pulse motor 14 moves a predetermined cut filter included in the fluorescence side filter 7 onto the optical path based on a command from the control unit 32.

[0020] Fluorescence side spectroscope 8 disperses the light (fluorescence) emitted from sample M1 and is equipped with a diffraction grating 8a. Fluorescence side pulse motor 12 adjusts the rotation angle of diffraction grating 8a of fluorescence side spectroscope 8 based on commands from control unit 32. Detector 9 converts the light (fluorescence) from fluorescence side spectroscope 8 into a predetermined electrical signal. The electrical signal (analog signal) from detector 9 is output to A / D converter 31. Note that the configuration shown in FIG. 1 is an example, and the configuration of photometer unit 10 is not limited to this.

[0021] 1 accepts data input by the user and displays the processing results of the control unit 32. The operation unit 20 includes an operation panel 21 (input unit) and a display unit 22 (display device). The operation panel 21 accepts input of measurement conditions for fluorescence fingerprint analysis and the like based on user operation, and includes predetermined keys (not shown) and buttons (not shown). Note that a keyboard (not shown) or a mouse (not shown) may be used instead of the operation panel 21. The display unit 22 is, for example, a display, and displays a setting screen for measurement conditions and the like as well as predetermined processing results of the control unit 32 and the like.

[0022] The data processing unit 30 shown in FIG. 1 includes an A / D converter 31 and a control unit 32. The A / D converter 31 converts analog signals input from the monitor detector 4 and the detector 9 into digital signals. The control unit 32 performs predetermined processing based on the digital signals input from the A / D converter 31 and data input via the operation unit 20. For example, the control unit 32 comprehensively measures the intensity of fluorescence emitted by a combination of excitation wavelength and fluorescence wavelength by repeatedly sweeping the wavelength of the fluorescence emitted from the sample M1 at a predetermined sampling interval while fixing the wavelength of the excitation light irradiated on the sample M1 and then shifting to the next excitation wavelength. In this manner, a spectroscopic analysis spectrum (a three-dimensional fluorescence spectrum, a fluorescence fingerprint) consisting of three components: excitation wavelength, fluorescence wavelength, and fluorescence intensity, is obtained. As described above, a spectroscopic analysis spectrum is specific to the composition of a substance. In the following description, the reference numeral for the sample M1 will be omitted as appropriate.

[0023] FIG. 2A is a diagram showing an example of a spectroscopic analysis spectrum obtained by the spectroscopic analysis system. The vertical axis in Figure 2A represents the wavelength of the excitation light irradiated onto the sample (excitation wavelength). The horizontal axis in Figure 2A represents the wavelength of the fluorescence emitted from the sample (fluorescence wavelength). In Figure 2A, the spectroscopic analysis spectrum is represented by contour lines according to the intensity of the fluorescence emitted from the sample (fluorescence intensity). The non-fluorescent region R1 and the regions R2 and R3 related to scattered light shown in Figure 2A are not particularly related to fluorescence and contain unnecessary data, so they are excluded from the analysis as explained below.

[0024] The non-fluorescent region R1 shown in Figure 2A is a region where the fluorescent wavelength is shorter than the excitation wavelength. In reality, the energy of the fluorescent light is smaller than the energy of the excitation light, so the phenomenon of the fluorescent wavelength becoming shorter than the excitation wavelength (i.e., the energy becoming greater) does not occur. Therefore, the non-fluorescent region R1, which does not meet the definition of fluorescence, is excluded from the analysis.

[0025] Furthermore, excitation light reflected from the sample surface and detected as is is first-order scattered light, and is therefore excluded from the analysis. For example, a region R2 within ±30 [nm] of a line (not shown) where the excitation wavelength and fluorescence wavelength are equal is excluded from the analysis. A region R3 of higher-order (second-order or third-order) scattered light is also excluded from the analysis. The control unit 32 (see FIG. 1) is responsible for excluding the non-fluorescent region R1 and the regions R2 and R3 associated with scattered light from the analysis.

[0026] FIG. 2B shows the spectroscopic analysis spectrum after predetermined preprocessing. Note that Figure 2B shows the spectroscopic analysis spectrum after preprocessing, excluding the non-fluorescent region R1 and the scattered light-related regions R2 and R3 (see Figure 2A), which are not particularly necessary for analysis. The control unit 32 (see Figure 1) converts the preprocessed spectroscopic analysis spectrum into a one-dimensional spectrum, as shown in Figure 2C.

[0027] FIG. 2C is an explanatory diagram of one-dimensional data of the spectroscopic analysis spectrum. The horizontal axis in Figure 2C represents the fluorescence wavelength for each excitation wavelength. In other words, the horizontal axis in Figure 2C represents the fluorescence wavelength when focusing on one of the multiple excitation wavelengths in the spectroscopic analysis spectrum. The vertical axis in Figure 2C represents the fluorescence intensity. As shown in Figure 2C, for example, when the excitation wavelength is 250 [nm], a predetermined spectrum is obtained within the fluorescence wavelength range of 285 [nm] to 800 [nm]. Note that predetermined spectra can also be obtained in a similar manner for other excitation wavelengths. The control unit 32 (see Figure 1) aligns this data for each excitation wavelength, expands it into vectors, and then performs multivariate analysis. Details of multivariate analysis will be described later.

[0028] FIG. 3 is a functional block diagram of the control unit 32 included in the spectroscopic analysis system. 3 includes electronic circuits such as a CPU (Central Processing Unit), a ROM (Read Only Memory), a RAM (Random Access Memory), and various interfaces (not shown) as its hardware configuration. The CPU reads out a program stored in the ROM and loads it into the RAM, and executes various processes. Note that the control unit 32 does not necessarily have to be configured as a single device, but may be configured as multiple devices connected via signal lines, or may include a server (not shown) connected via a network (not shown).

[0029] 3, the control unit 32 includes a storage unit 321, a calculation unit 322, and a communication interface 323. The storage unit 321 stores data including measurement condition data 321a, analysis condition data 321b, measurement data 321c, analysis data 321d, and a regression model 321e.

[0030] The measurement condition data 321a is data indicating the measurement conditions (see FIG. 5) for the spectroscopic analysis spectrum of the sample. The analysis condition data 321b is data indicating the analysis conditions (see FIG. 5) for the spectroscopic analysis spectrum of the sample. The measurement data 321c is data such as the spectroscopic analysis spectrum obtained by measuring the sample. The analysis data 321d is data indicating the analysis results of the sample. The regression model 321e is a prediction model used in the analysis of the sample. In the following description, the symbol for the regression model 321e will be omitted as appropriate.

[0031] The calculation unit 322 shown in FIG. 3 includes a condition setting unit 322a, a measurement control unit 322b, a wavelength region generation unit 322c, a model generation unit 322d, a model evaluation unit 322e, and a display control unit 322f. When predetermined measurement conditions and analysis conditions are input via the operation panel 21 (see FIG. 1), the condition setting unit 322a stores them in the storage unit 321 as measurement condition data 321a and analysis condition data 321b. The measurement control unit 322b measures the spectroscopic analysis spectrum of the sample using the photometer unit 10 (see FIG. 1). After the spectroscopic analysis spectrum is measured, the wavelength region generation unit 322c generates a set of wavelength region candidates to be used in generating the regression model 321e. The wavelength region is specified by each range of excitation wavelength and fluorescence wavelength.

[0032] The model generating unit 322d generates a regression model 321e for obtaining a predetermined response variable (composition, concentration, etc. of the sample) based on the spectroscopic analysis spectrum. The model evaluation unit 322e evaluates the prediction performance and measurement time of the regression model 321e. The display control unit 322f causes the display unit 22 (see FIG. 1) to display the measurement data 321c, the analysis data 321d, and also the evaluation results of the model evaluation unit 322e. The communication interface 323 inputs and outputs data to and from the A / D converter 31 (see FIG. 1), the operation panel 21 (see FIG. 1), and the display unit 22 (see FIG. 1).

[0033] For example, when performing fluorescence fingerprint analysis of samples transported one after another on a conveyer belt (not shown) in a factory or facility (i.e., performing in-line measurement), if the measurement time is too long, the number of samples (products) processed per unit time will be reduced. Furthermore, if measurement is performed focusing only on a specific excitation wavelength or fluorescence wavelength, the measurement time will be shortened, but important spectra for determining the composition and quantifying the concentration of the sample may not be obtained. Therefore, in the first embodiment, the user sets an upper limit on the measurement time for performing fluorescence fingerprint analysis of a single sample, and the control unit 32 generates a regression model 321e that will obtain highly accurate analysis results within this measurement time.

[0034] FIG. 4 is a flowchart showing the processing of the control unit of the spectroscopic analysis system (also see FIG. 1 as appropriate). 4 is a process that is performed in advance to generate recommended measurement conditions and a regression model prior to inline measurement in a factory or facility, for example. The use of the spectroscopic analysis system 100 (see FIG. 1) is not limited to inline measurement, but may also be offline measurement performed outside a production line, and it can also be used for a variety of purposes, such as medical-related and food-related testing and research, as well as individual experiments.

[0035] In step S101, the control unit 32 sets an upper limit value for the measurement time using the condition setting unit 322a (see FIG. 3). That is, the control unit 32 accepts input of an upper limit value for the measurement time of the spectroscopic analysis spectrum as a user-set condition for measuring the spectroscopic analysis spectrum of the sample based on an operation by the user via the operation panel 21 (input unit) (input process). Note that the "upper limit value for the measurement time" is the upper limit value of the range that the user can tolerate with regard to the time required to measure one sample.

[0036] Next, in step S102, the control unit 32 sets measurement conditions using the condition setting unit 322a (see FIG. 3). That is, the control unit 32 sets measurement conditions in the spectroscopic analysis system 100 based on an operation by the user via the operation panel 21.

[0037] FIG. 5 shows an example of a screen for setting measurement conditions, etc. The setting screen of FIG. 5 is displayed on the display unit 22 (see FIG. 1) when the user inputs the upper limit of the measurement time based on the spectroscopic analysis method ("maximum measurement time" in FIG. 5) and predetermined measurement conditions. As described above, the "measurement conditions" shown in Fig. 5 are the conditions for measuring the spectroscopic analysis spectrum of a sample. In the example of Fig. 5, the excitation wavelength range, fluorescence wavelength range, excitation light sampling interval, fluorescence sampling interval, excitation light slit width, fluorescence slit width, and wavelength scan speed are set as such "measurement conditions" (S102 in Fig. 4).

[0038] The "excitation wavelength range" shown in Figure 5 is the range of excitation wavelengths used in fluorescence fingerprint analysis. The "fluorescence wavelength range" is the range of fluorescence wavelengths used in fluorescence fingerprint analysis. The "excitation light sampling interval" is the sampling interval used when the excitation wavelength is swept. The "fluorescence sampling interval" is the sampling interval used when the fluorescence wavelength is swept.

[0039] The "excitation light slit width" shown in Figure 5 is the slit width of the diffraction grating 2a of the excitation side spectroscope 2 (see Figure 1). The "fluorescence slit width" is the slit width of the diffraction grating 8a of the fluorescence side spectroscope 8 (see Figure 1). The "wavelength scan speed" is the sweep speed (scan speed) when one of the excitation wavelength and the fluorescence wavelength is fixed and the other is swept.

[0040] As these measurement conditions, for example, the excitation wavelength range and the fluorescence wavelength range may each be set to the range of 250 to 750 [nm] (or a part of that range). Furthermore, the sampling interval of the excitation wavelength may be set to 10 [nm], the sampling interval of the fluorescence wavelength to 5 [nm], the slit width of the excitation light and fluorescence to 5 [nm], and the wavelength scan speed to 60,000 [nm / min]. Note that the above-mentioned numerical values ​​are merely examples and do not particularly limit the measurement conditions. Furthermore, predetermined default values ​​may be displayed as the numerical values ​​of each item of the measurement conditions, allowing the user to change the value of each item from the default value as appropriate.

[0041] The "analysis conditions" shown in FIG. 5 are conditions under which a predetermined analysis is performed based on the spectroscopic analysis spectrum of a sample. Note that the "maximum measurement time" relates to both measurement and analysis, but in the example of FIG. 5, it is displayed under the "analysis conditions." When the "maximum measurement time" is input by the user through the operation panel 21 (see FIG. 1), the control unit 32 sets this "maximum measurement time" as the upper limit of the measurement time (S101 in FIG. 4). The "maximum measurement time" may be, for example, a value of 10 seconds or less, or may be a predetermined value longer than 10 seconds.

[0042] The "wavelength region selection method" shown in FIG. 5 is a method for selecting a predetermined measurement wavelength region from the "excitation wavelength range" and "fluorescence wavelength range" set by the user. That is, the control unit 32 (see FIG. 1) displays multiple candidates for the wavelength region selection method, which is a method for selecting a predetermined measurement wavelength region, on the display unit 22 (display device: see FIG. 1), and executes a predetermined wavelength region selection method that the user selects from the multiple candidates through operation by the user via the operation panel 21 (input unit: see FIG. 1). This increases the user's freedom in setting the wavelength region selection method. Note that a method using random numbers or other well-known methods may be used as the wavelength region selection method.

[0043] The "regression method" shown in Fig. 5 is a method used to generate a regression model based on multivariate analysis. That is, the control unit 32 (see Fig. 1) displays a plurality of candidate regression methods for multivariate analysis used in analyzing the spectroscopic spectrum on the display unit 22 (display device: see Fig. 1), and executes a predetermined regression method selected from the plurality of candidates by a user through an operation panel 21 (input unit: see Fig. 1). As will be described in detail later, candidate regression methods that may be used include, for example, partial least squares (PLS) and Lasso regression, as well as multiple regression analysis, principal component regression analysis, RF regression, and SVM regression, as appropriate. It should be noted that for the "wavelength region selection method" and "regression method," multiple candidates may be displayed in a pull-down menu, and one may be selected from the multiple candidates. Furthermore, it is not necessary for the user to set all of the measurement conditions and analysis conditions shown in Figure 5; only some of those shown in Figure 5 may be set.

[0044] The "exclusion region" shown in Figure 5 is a region that is not particularly necessary for data analysis, such as the non-fluorescent region R1 (see Figure 2A) and scattered light regions R2 and R3 (see Figure 2A). The "measurement results" shown in Figure 5 are the results of measurement and analysis based on spectroscopic analysis (e.g., fluorescent fingerprint analysis). After the results of measurement and analysis based on spectroscopic analysis are obtained, the specific details of the "measurement results" are displayed (see Figure 6).

[0045] Returning to FIG. 4 again, the explanation will be continued. After setting an upper limit for the measurement time (S101) and setting the measurement conditions (S102), in step S103, the control unit 32 measures fluorescence fingerprint data using the measurement control unit 322b. That is, the control unit 32 performs fluorescence fingerprint analysis on a predetermined sample based on the measurement conditions set in step S102. Note that, in order to distinguish the composition of the sample and quantify the concentration, multiple samples with different concentrations and compositions are prepared, and fluorescence fingerprint data is measured for each sample in turn. Then, a spectroscopic analysis spectrum (see FIG. 2A) is generated by measuring the fluorescence fingerprint data.

[0046] Next, in step S104, the control unit 32 generates a set of candidate wavelength regions using the wavelength region generation unit 322c (see FIG. 3). That is, the control unit 32 generates a set of candidate wavelength regions from the entire range specified by the respective ranges of excitation wavelength and fluorescence wavelength set in the measurement conditions (S102). Here, the "wavelength region" refers to the range specified by the respective ranges of excitation wavelength and fluorescence wavelength (see wavelength regions G3 and G4 in FIG. 7A). For example, the control unit 32 generates a set of candidate wavelength regions to be used in generating a regression model based on a predetermined random number. Note that it is desirable for the set of candidate wavelength regions to include multiple wavelength regions. This is because including multiple wavelength regions makes it possible to include multiple locations that significantly contribute to sample composition determination and concentration quantification without making each wavelength region particularly large.

[0047] In step S105 of Fig. 4, the control unit 32 generates a regression model for each wavelength region using the model generation unit 322d (see Fig. 3). That is, the control unit 32 generates a regression model (prediction model) corresponding to each candidate wavelength region based on the spectroscopic analysis spectra of multiple samples with different compositions and concentrations. For example, PLS regression analysis or Lasso regression is used as a multivariate analysis method for generating the regression model.

[0048] Next, in step S106, the control unit 32 determines whether any of the set of wavelength region candidates has a measurement time that is equal to or less than a predetermined upper limit. The method for calculating the measurement time depends on the specific measurement method used by the photometer unit 10 (see FIG. 1). For example, if the process of sweeping the fluorescence wavelength while fixing the excitation wavelength, and then, when the sweeping of the fluorescence wavelength is completed, changing the excitation wavelength and sweeping the fluorescence wavelength again is repeated, the control unit 32 calculates the measurement time as follows: That is, the control unit 32 calculates the measurement time based on the sum of the fluorescence wavelength sweep time, the fluorescence wavelength return time, and the excitation wavelength movement time.

[0049] In step S106, if there is no measurement time that is equal to or shorter than the upper limit value (S106: No), the process of the control unit 32 returns to step S102. In this case, a predetermined message prompting the user to change the measurement conditions is displayed on the display unit 22 (see FIG. 1). Then, the measurement conditions are changed appropriately based on the user's operation via the operation panel 21 (see FIG. 1).

[0050] Furthermore, in step S106, if there is any wavelength region among the plurality of wavelength regions for which the measurement time is equal to or less than the upper limit (S106: Yes), the processing of the control unit 32 proceeds to step S107. In step S107, the control unit 32 verifies the predictive performance of the regression model using the model evaluation unit 322e (see FIG. 3). Note that the target of step S107 is the regression model associated with any wavelength region among the plurality of wavelength regions for which the measurement time based on the fluorescence fingerprint analysis is equal to or less than a predetermined upper limit.

[0051] Cross-validation, for example, is used as a method for verifying the predictive performance of a regression model. When performing cross-validation, the control unit 32 divides the training data (the spectroscopic analysis spectra of multiple samples) into multiple groups. As a specific example, consider here a case where the training data is divided into five groups (5-fold cross-validation). For example, if there are a total of 20 samples whose concentrations and compositions are known, the control unit 32 divides the spectroscopic analysis spectrum data of the 20 samples into a total of five groups of four each.

[0052] In 5-fold cross-validation, the control unit 32 reserves one predetermined group for verifying predictive performance and generates a new regression model using the remaining four groups. The control unit 32 then generates the regression model a total of five times by sequentially changing the group used for verifying predictive performance. The control unit 32 then determines hyperparameters based on the predetermined cross-validation. Note that "hyperparameters" are predetermined parameters for setting the behavior of the machine learning algorithm.

[0053] For example, when PLS regression is used to generate a regression model, the number of PLS ​​components is a hyperparameter. For example, root-mean-square error (RMSE) or mean absolute error (MAE) is used as an evaluation index of prediction performance. To give a specific example, when RMSE is used to evaluate prediction performance, the control unit 32 sets the average value of RMSE in five repeated cross-validations as RMSECV, which is the evaluation index of prediction performance.

[0054] Then, the control unit 32 sets the hyperparameters so that the RMSECV, which is an evaluation index of the prediction performance, is minimized. In this way, in step S108, the control unit 32 calculates the RMSECV based on the hyperparameters (the number of PLS ​​components) optimized under the predetermined measurement conditions and wavelength region. Note that the evaluation index of the prediction performance (e.g., RMSECV) is calculated for each regression model that satisfies the conditions in step S106.

[0055] Next, in step S108, the control unit 32 selects the regression model with the highest prediction performance. For example, the control unit 32 selects the regression model with the smallest RMSECV value for the optimized number of PLS ​​components. In step S109, the control unit 32 determines, by the model evaluation unit 322e (see FIG. 3), whether or not the prediction performance satisfies a predetermined target value. For example, the control unit 32 determines whether or not the RMSECV of the regression model selected in step S108 is equal to or less than a predetermined value. The predetermined value is a target value for the prediction performance, and is set in advance.

[0056] In step S109, if the predicted performance does not satisfy the target value (S109: No), the process of the control unit 32 returns to step S102. In this case, a message prompting the user to change the measurement conditions is displayed on the display unit 22 (see FIG. 1). Also, in step S109, if the predicted performance satisfies the target value (S109: Yes), the process of the control unit 32 proceeds to step S110.

[0057] In step S110, the control unit 32 displays the results of the measurement and analysis using the display control unit 322f (see FIG. 3). That is, the control unit 32 derives predetermined recommended measurement conditions that satisfy the user-set conditions regarding the measurement time, and displays these recommended measurement conditions on the display unit 22 (display device: see FIG. 1) (display processing). This allows the user to understand the recommended measurement conditions for performing measurements with relatively high predictive performance while keeping the measurement time below a predetermined upper limit. These measurement and analysis results are used when the user performs in-line measurements of samples in factories or facilities. After performing the processing of step S110, the control unit 32 ends the series of processing (END).

[0058] Figure 6 shows an example of the measurement and analysis results displayed on the screen. In Figure 6, the recommended measurement conditions for inline measurement, etc., are displayed on the display unit 22 (see Figure 1), including the excitation light sampling interval and fluorescence sampling interval for measuring the spectroscopic analysis spectrum, as well as the excitation light slit width, fluorescence slit width, excitation wavelength range, wavelength scan speed, and measurement time. These recommended measurement conditions are the measurement conditions when using the regression model with the highest prediction performance (S106: Yes, S108 in Figure 4) among those for which the measurement time meets a specified upper limit. Note that although Figure 6 does not specifically display the numerical values ​​in each column of the recommended measurement conditions, in reality, specific numerical values ​​are displayed. Furthermore, the recommended measurement conditions may also include the fluorescence wavelength range.

[0059] In the example of FIG. 6, the control unit 32 displays the recommended measurement conditions on the display unit 22 (display device: see FIG. 1), and also displays the spectroscopic analysis spectrum based on the fluorescence fingerprint analysis on the display unit 22. The spectroscopic analysis spectrum displayed on the display unit 22 shows measurement wavelength regions G1 and G2, which are specified by the excitation wavelength range and the fluorescence wavelength range as the wavelength range of the recommended measurement conditions. For example, the control unit 32 may highlight the measurement wavelength regions G1 and G2 by displaying them in a color or pattern different from other regions, or by displaying their boundaries with a bold frame line. This allows the user to understand at a glance which region should be used to measure the spectroscopic analysis spectrum.

[0060] Furthermore, as shown in Fig. 6, it is preferable that the display unit 22 (display device: see Fig. 1) displays both the recommended measurement conditions and the spectroscopic analysis spectrum on one screen. This allows the user to check the values ​​of each item of the recommended measurement conditions while also checking the measurement wavelength ranges G1 and G2 of the spectroscopic analysis spectrum on the same screen. Therefore, visibility is improved when the user checks the recommended measurement conditions and the measurement wavelength ranges G1 and G2.

[0061] Furthermore, the number of measurement wavelength regions G1, G2 (two in the example of FIG. 6) may be set based on a user's operation via the operation panel 21 (input unit: see FIG. 1). This increases the user's flexibility when causing the control unit 32 to generate the measurement wavelength regions G1, G2. The range of the number of measurement wavelength regions may also be set based on a user's operation via the operation panel 21. For example, when the user operates the operation panel 21 to set the number of measurement wavelength regions to two or more and five or less, the wavelength region generation unit 322c (see FIG. 3) generates a set of wavelength region candidates for each of the cases where the number of measurement wavelength regions is two, three, four, and five.

[0062] In order to create a regression model with high prediction performance, it is important to verify many wavelength regions. However, if wavelength regions are generated using a brute force method, it may take a long time to improve prediction performance. Therefore, for example, when optimizing the wavelength regions, the control unit 32 may perform GAWLSPLS (Genetic Algorithm-based Wavelength Selection Partial Least Squares) based on a genetic algorithm. The following description of GAWLSPLS corresponds to the processing of steps S104 to S108 in FIG. 4.

[0063] FIG. 7A is an explanatory diagram showing how the range of the target wavelength region is changed when the GAWLSPLS method is used. In the GAWLSPLS method, the control unit 32 (see Figure 1) first specifies the number of wavelength regions to be selected. In the example of Figure 7A, two wavelength regions G3 and G4 are specified. These wavelength regions G3 and G4 are defined by the starting excitation wavelength, starting fluorescence wavelength, and the size of each region (length of the excitation wavelength range, length of the fluorescence wavelength range) when sweeping the excitation wavelength and fluorescence wavelength. In the genetic algorithm, the starting excitation wavelength and region size of each wavelength region G3 and G4 are assigned to a specified chromosome.

[0064] Then, the wavelength range to be used for analysis is derived from each chromosome. Based on these wavelength ranges, a prediction model (calibration model) for, for example, the concentration (objective variable) of a predetermined substance contained in a sample is constructed. In GAWLSPLS, for example, the above-mentioned RMSECV is used as an index of fitness for the genetic algorithm. Then, an analysis wavelength range (wavelength range to be analyzed) is determined from the chromosome, and the fitness for this analysis wavelength range is calculated.

[0065] In the GAWLSPLS method, suitable chromosomes are selected from the viewpoint of minimizing RMSECV, an index of fitness, based on a genetic algorithm. A constraint is set that the time required to measure a wavelength region must be equal to or less than a predetermined upper limit (maximum measurement time). The control unit 32 (see FIG. 1) generates chromosomes (initial population) for a predetermined number of wavelength regions, and calculates and evaluates the fitness. The control unit 32 then selects from these chromosomes those whose measurement time is equal to or less than the predetermined upper limit.

[0066] If the predetermined convergence condition specified by the user is satisfied, the chromosome with the lowest RMSECV, an index of fitness, in the population becomes the solution. If the predetermined convergence condition is not satisfied, the control unit 32 generates a next-generation population by selecting, crossing over, and mutating chromosomes from the population, and evaluates this population. The control unit 32 repeats this series of processes until the predetermined convergence condition is satisfied, thereby deriving a suitable analysis wavelength region. Alternatively, the control unit 32 may obtain multiple analysis wavelength regions by appropriately changing the number of wavelength regions and the method of generating random numbers.

[0067] FIG. 7B is an explanatory diagram showing measurement wavelength regions G5 and G6 set based on the GAWLSPLS method. In the example of Figure 7B, two trapezoidal measurement wavelength ranges G5 and G6 are set as wavelength ranges used for fluorescence fingerprint analysis such as in-line measurement. Note that the wavelength range specified by the range of excitation wavelength and fluorescence wavelength does not necessarily have to be rectangular, and may be trapezoidal as shown in Figure 7B. By using the GAWLSPLS method described above, it is possible to efficiently specify wavelength ranges that satisfy specified conditions based on a genetic algorithm.

[0068] Furthermore, when measurements are performed under multiple conditions, it may take a long time to measure the spectroscopic analysis spectrum. Therefore, as will be explained below, the control unit 32 may use a spectroscopic analysis spectrum acquired under a specific measurement condition to generate a pseudo spectroscopic analysis spectrum under a different measurement condition. An example of an experimental result using this method is shown below.

[0069] In the experiment, the vitamin E concentration in edible oil (15 samples) was used as the objective variable. Measurements were performed under the following conditions: an excitation wavelength range of 250–450 nm, an excitation wavelength sampling interval of 1 nm, an emission wavelength range of 250–450 nm, and an emission wavelength sampling interval of 2 nm. Measurements were performed under three different conditions: an excitation wavelength sampling interval of 1 nm, as described above, as well as 2 nm and 3 nm. The GAWLSPLS method was used to optimize the excitation and emission wavelength ranges. The upper limit (maximum measurement time) of the measurement time was set to 50 seconds. When the excitation wavelength sampling interval was increased, the fluorescence intensity was integrated (i.e., the sum of the fluorescence intensities) to generate a spectroscopic analysis spectrum.

[0070] FIG. 8 is an explanatory diagram showing an example in which the fluorescence wavelength is fixed and the excitation wavelength is swept to a predetermined value. For example, consider a case where data is acquired at sampling intervals of 1 nm during measurement, such as 250 nm, 251 nm, 252 nm, etc., as shown in Figure 8. By utilizing data obtained by sampling the excitation wavelength every 1 nm in this way, it is possible to generate a pseudo-spectroscopic analysis spectrum when the excitation wavelength is sampled every 3 nm.

[0071] One method for generating a pseudo-spectroscopic spectrum is to not use data with excitation wavelengths of 251 [nm] or 252 [nm]. In other words, data with an excitation wavelength of 250 [nm] + 3k (k is an integer) is used, while data with wavelengths of 251 [nm] + 3k and 252 [nm] + 3k are not used in the analysis. In this way, the measurement time can be shortened by thinning out the data, so to speak.

[0072] Another method is to integrate the fluorescence intensities at excitation wavelengths of 250 [nm], 251 [nm], and 252 [nm]. That is, the sum of the fluorescence intensities at excitation wavelengths of 250 [nm] + 3k (k is an integer), 251 [nm] + 3k, and 252 [nm] + 3k may be calculated and associated with, for example, 251 [nm] + 3k. In this way, the measurement time can be shortened by associating the sum of the fluorescence intensities at three adjacent excitation wavelengths at a predetermined sampling interval with one of the three excitation wavelengths (or the average value of the three excitation wavelengths). Another method is to integrate the fluorescence intensities when the excitation wavelength is 250 [nm] and 252 [nm]. That is, the excitation wavelength is 250 [nm] + 3k (k is an integer), and 252 The sum of the fluorescence intensities at [nm]+3k may be calculated and associated with, for example, 250[nm]+3k. In this manner, it is also possible to generate a pseudo-spectroscopic spectrum. The pseudo-spectroscopic spectrum is treated as separate data from the original spectroscopic spectrum measured at a sampling interval of 1 nm. The same applies when sweeping the fluorescence wavelength instead of the excitation wavelength.

[0073] In this way, when measuring a spectroscopic spectrum based on fluorescence fingerprint analysis, the control unit 32 fixes one of the excitation wavelength and the fluorescence wavelength and sweeps the other wavelength at a predetermined sampling interval. In this case, for each of n other wavelengths (n is a natural number) swept at the predetermined sampling interval, the control unit 32 generates data in which the sum of some or all of the n fluorescence intensities corresponding one-to-one to the n other wavelengths corresponds to one of the n other wavelengths. Note that the control unit 32 may also generate data in which the sum corresponds to the average value of the n other wavelengths. Then, the control unit 32 generates a new pseudo spectroscopic spectrum based on the generated data, and generates a prediction model for analyzing the spectroscopic spectrum based on this pseudo spectroscopic spectrum.

[0074] In the experiment, we adopted a method of integrating the fluorescence intensities at excitation wavelengths of 250 nm, 251 nm, and 252 nm (i.e., taking the sum of the fluorescence intensities). This improved the S / N ratio in fluorescence fingerprint analysis. This is because integrating the fluorescence intensities produced an effect similar to that of actually obtaining measurement results when the amount of fluorescent light is increased.

[0075] When integrating (i.e., adding up) the fluorescence intensities at two excitation wavelengths with different values, the control unit 32 may double the slit width of the excitation light. When integrating the fluorescence intensities at three excitation wavelengths, the control unit 32 may triple the slit width of the excitation light. This can improve the S / N ratio. Additionally, the control unit 32 may adjust the scanning speed of the fluorescence wavelengths as appropriate.

[0076] FIG. 9 is an explanatory diagram showing the experimental results of the fluorescent fingerprint analysis. In the example shown in Figure 9, the RMSECV, an index of prediction performance, was smallest when the excitation wavelength sampling interval was 2 nm, and the measurement time was 49.5 seconds. In this way, by setting the excitation wavelength sampling interval to 2 nm, we were able to identify recommended measurement conditions with relatively high prediction accuracy within a measurement time below the specified upper limit (maximum measurement time). Furthermore, when the excitation wavelength sampling interval was increased to 3 nm, the RMSECV, an index of prediction performance, was slightly lower, but the measurement time was 15.3 seconds, and a similar level of prediction accuracy was achieved in a shorter time.

[0077] <Effects> According to the first embodiment, the control unit 32 derives recommended measurement conditions and a regression model for performing highly accurate substance discrimination and concentration quantification within a time that is equal to or less than the upper limit of the measurement time according to the user's needs. This enables highly accurate substance discrimination and concentration quantification even in use cases where measurement time constraints are important, such as in-line measurements for industrial processes. It is also possible to appropriately set the upper limit of the measurement time according to the specific conditions required in the industrial process. Thus, according to the first embodiment, a spectroscopic analysis system 100 can be provided that enables appropriate measurement even when there are constraints on the measurement time, etc.

[0078] Second Embodiment The second embodiment differs from the first embodiment in that, when the measurement conditions are optimized based on the spectroscopic analysis spectrum, learning data is measured again under the optimized measurement conditions, and a regression model is generated using the remeasured learning data. Note that other aspects (such as the configuration of the spectroscopic analysis system 100: see FIG. 1) are the same as those of the first embodiment. Therefore, only the differences from the first embodiment will be described, and a description of overlapping aspects will be omitted.

[0079] 10A and 10B are flowcharts showing the processing of the control unit 32 of the spectroscopic analysis system according to the second embodiment (see also FIG. 1 as appropriate). 10A are the same as those in the first embodiment (see FIG. 4), and therefore will not be described further. In step S103a shown in FIG. 10A, control unit 32 not only actually measures the fluorescence fingerprint data (spectroscopic analysis spectrum) of a predetermined sample, but also calculates pseudo fluorescence fingerprint data based on this fluorescence fingerprint data.

[0080] For example, the control unit 32 calculates the sum of the fluorescence intensities at excitation wavelengths of 250[nm]+3k (k is an integer), 251[nm]+3k, and 252[nm]+3k, and associates the sum of the fluorescence intensities with the excitation wavelength of 251[nm]+3k. Note that the method for calculating the pseudo-spectroscopic analysis spectrum is the same as that described in the first embodiment, and therefore will not be described here. Furthermore, if the predicted performance satisfies the predetermined target value in step S109 (S109: Yes), the process of the control unit 32 proceeds to step S120 in FIG. 10B.

[0081] 10B, the control unit 32 measures the fluorescence fingerprint data (spectroscopic analysis spectrum) again under the optimized measurement conditions. Note that in the process of step S121, the spectroscopic analysis spectrum of the sample is actually measured. In step S122, the control unit 32 generates a regression model for the wavelength region. This "wavelength region" is a predetermined wavelength region associated with a predetermined regression model whose prediction performance satisfies the target value (S109: Yes in FIG. 10A). In this manner, the control unit 32 actually measures the spectroscopic analysis spectrum under recommended measurement conditions based on the pseudo spectroscopic analysis spectrum (S121), and based on the spectroscopic analysis spectrum obtained by this measurement, generates again a regression model (prediction model) for analyzing the spectroscopic analysis spectrum (S122). By the control unit 32 generating again the regression model, it is possible to reduce the influence of errors associated with the use of the pseudo spectroscopic analysis spectrum in step S103a (see FIG. 10A), and obtain a regression model with high prediction performance.

[0082] Next, in step S123, the control unit 32 verifies the prediction performance of the regression model. Note that the method for verifying the prediction performance of the regression model is the same as that in step S107 (see FIG. 10A), and therefore description thereof will be omitted. In step S124, the control unit 32 determines whether the prediction performance of the regression model satisfies a predetermined target value. If the prediction performance does not satisfy the target value in step S124 (S124: No), the process of the control unit 32 returns to step S102 (see FIG. 10A). In this case, a message prompting the user to change the measurement conditions is displayed on the display unit 22 (see FIG. 1).

[0083] Furthermore, in step S124, if the predicted performance satisfies the predetermined target value (S124: Yes), the process of the control unit 32 proceeds to step S125. In step S125, the control unit 32 causes the measurement results and analysis results to be displayed on the display unit 22. After performing the process of step S125, the control unit 32 ends the series of processes (END).

[0084] <Effects> According to the second embodiment, a pseudo-spectroscopic analysis spectrum is generated based on a predetermined mathematical process. This eliminates the need for inspectors to measure spectroscopic analysis spectra under various measurement conditions, thereby reducing the workload on the inspectors and shortening the time required to identify a regression model with high predictive performance. Furthermore, the control unit 32 measures the spectroscopic analysis spectrum again under optimized measurement conditions and generates a regression model based on the measurement results. This ensures sufficient predictive accuracy of the regression model even when a pseudo-spectroscopic analysis spectrum is used.

[0085] Third Embodiment The third embodiment differs from the first embodiment in that the sampling intervals and the like under the recommended measurement conditions are different for multiple wavelength regions. Note that other aspects (such as the configuration of the spectroscopic analysis system 100: see FIG. 1) are the same as those of the first embodiment. Therefore, only the differences from the first embodiment will be described, and a description of overlapping aspects will be omitted.

[0086] The third embodiment will be described with reference to FIG. 6. For example, it is desirable to set the excitation wavelength sampling interval and the fluorescence wavelength sampling interval for each of the multiple measurement wavelength ranges G1 and G2. Specifically, by setting a high resolution in a predetermined measurement wavelength range G1 of the spectroscopic analysis spectrum, detailed shape data of the spectroscopic analysis spectrum can be obtained. This improves the accuracy when the control unit 32 determines the composition and quantifies the concentration of the sample. Furthermore, for another measurement wavelength range G2 of the spectroscopic analysis spectrum, information obtained by integrating peak intensities may be important. In such cases, since integration requires time, the control unit 32 can reduce the resolution to obtain high prediction performance in a relatively short time. In this way, the control unit 32 may change the measurement conditions according to the characteristics of each of the multiple regions included in the spectroscopic analysis spectrum.

[0087] In this way, it is preferable that the plurality of measurement wavelength ranges contain excitation wavelengths and / or fluorescence wavelengths with different sampling intervals included in the predetermined recommended measurement conditions. Furthermore, the sampling intervals of at least one of the excitation wavelengths and fluorescence wavelengths in the plurality of measurement wavelength ranges are set, for example, based on a genetic algorithm. Alternatively, an inspector may set the sampling intervals of the excitation wavelengths and fluorescence wavelengths in the measurement wavelength ranges G1 and G2 based on past experimental data.

[0088] <Effects> According to the third embodiment, the control unit 32 sets different measurement conditions based on the characteristics of multiple measurement wavelength ranges G1 and G2 specified in the ranges of excitation wavelength and fluorescence wavelength, thereby enabling highly accurate determination of sample composition and concentration quantification.

[0089] Fourth Embodiment The fourth embodiment differs from the first embodiment in that a predetermined evaluation index different from RMSECV is used to avoid so-called overfitting (overlearning). The rest is the same as the first embodiment. Therefore, only the parts that are different from the first embodiment will be described, and a description of the overlapping parts will be omitted.

[0090] The fourth embodiment will be described with reference to Fig. 1. In the first embodiment described above, the case where RMSECV is used as an index of prediction performance has been described. However, when RMSECV is minimized in regression analysis such as PLS regression, overfitting (overlearning) may occur. Therefore, in order to suppress excessive fitting to the spectroscopic analysis spectrum data used when creating the prediction model and to improve generalization performance, the control unit 32 may perform the following processing.

[0091] For example, in the GAWLSPLS method, as a way to deal with overfitting when selecting a suitable wavelength range from multiple analysis wavelength ranges, a prediction model may be created using an index instead of RMSECV. Specifically, overfitting can be suppressed by using the following evaluation index. In the following equation (1), j is the number of PLS ​​components, B2 is the Euclidean norm of the regression coefficient vector, and b is the regression coefficient vector. In addition, DW in equation (2) is the first derivative of the normalized regression coefficient vector. In addition, J in equation (3) is the Euclidean norm of the change in the regression coefficient.

[0092]

number

[0093]

number

[0094]

number

[0095] For example, if the regression coefficients contain noise components, the sum of the absolute values ​​of the regression coefficients increases, resulting in larger values ​​for each index (B2, DW, J). Therefore, each of these indexes (B2, DW, J) may be appropriately used as an evaluation index indicating the complexity of the regression coefficient vector. For example, even if the RMSECV is relatively small, it is desirable to penalize it if the complexity of the prediction model is high. Specifically, to improve the prediction performance for an unknown model, it is desirable for each index (B2, DW, J) to have a small value. The smaller the RMSECV and the smaller the indexes (B2, DW, J), the higher the prediction performance for an unknown sample.

[0096] Since RMSECV and each index (B2, DW, J) have different units, the generalization performance of the regression model can be improved by determining the number of PLS ​​components C1, which will be explained next. Note that j in equation (4) is the number of PLS ​​components, and RMSECV min is the minimum value of RMSECV, and RMSECV max is the maximum value of RMSECV. In addition, I included in equation (4) is one of B2, DW, and J, and Ij is the value when the number of PLS ​​components is j. min is the minimum value of I, and I max is the maximum value of I. The first term on the right side of equation (4) is the value of RMSECV normalized by the maximum minus the minimum value. The second term on the right side of equation (4) is the value of I normalized by the maximum minus the minimum value.

[0097]

number

[0098] In the fourth embodiment, the aforementioned B2 (Euclidean norm of the regression coefficient vector) is used as I in equation (4), and the GAWLSPLS method is applied using C1 as an index of prediction accuracy. In this way, the control unit 32 (see FIG. 1) searches for the measurement wavelength region based on a genetic algorithm, and in the genetic algorithm, assigns, for example, C1 as a predetermined penalty function indicating the degree of overlearning to RMSECV, which is an index of the goodness of fit of a prediction model for analyzing a spectroscopic spectrum. The control unit 32 then evaluates the prediction model based on the penalty function. As described above, RMSECV is the average value of the root mean square error (RMSE) in cross-validation.

[0099] <Experimental Results> As an example, we will explain the verification of the quantification of glucose concentration in a culture medium using near-infrared spectroscopy. CHO cells were used as CRL-12445 (ATCC), and the medium was DMEM-low glucose (Sigma-Aldrich). After spreading CHO cells in the medium, the cell count was measured using an automated fluorescent cell counter LUNA-FL (Logos Biosystems). The cell count was 1 x 10 5 ~3×10 5 This culture solution was seeded into a spinner flask, and the culture solution was stirred with a stirrer and stored in an incubator (temperature: 37°C, CO2 concentration: 5%, air concentration: 95%).

[0100] The incubator used was a Personal CO2 Multigas Incubator APM50DR (manufactured by Astex). Four cultures were performed to prepare samples for building the calibration model (22 samples). Six cultures were also performed to prepare samples for verifying the predictive performance of the created prediction model (calibration model), resulting in a total of 23 samples.

[0101] In addition to the actual culture samples, the culture medium before and after cultivation, and glucose were mixed to create mock culture samples, which were used for measurements. For the culture medium after cultivation, the culture medium was used to remove cells and other substances using a 0.2 μm filter 7 days after the start of cultivation. A total of 102 samples were prepared so that the glucose concentration of these mixtures was in the range of 0 to 8 g / L, with the glucose concentration varying from 0.5 to 0.6 g / L.

[0102] In addition, the simulated culture medium was processed as data for building a calibration model in bulk, and transfer learning was performed using the simulated culture medium data. The transfer learning method used was the Frustratingly Easy Domain Adaptation method. As a preprocessing method, the number of wavelength points when fitting the Savitzky-Golay method was set to 21, the degree of the polynomial used for fitting was set to 2, and the degree of differentiation afterwards was set to 1.

[0103] A prediction model was constructed based on the spectral data (spectroscopic analysis spectrum) obtained in this way and the glucose concentration measured by the enzyme electrode method. The GAWLSPLS method was used to select the wavelength range. The maximum measurement time was 600 seconds. Two fitness indices for the genetic algorithm were used: RMSECV and C1. The number of wavelength ranges ranged from 1 to 10, and 10 iterations were performed for each range, yielding a total of 100 wavelength range data. The glucose concentrations of the validation samples (23 samples) were predicted using the prediction model based on the data from these 100 wavelength ranges, and the root-mean-square error prediction (RMSEP) was calculated from the prediction results. The results are shown in Figures 11A and 11B.

[0104] FIG. 11A shows the experimental results when RMSECV is used as the index of goodness of fit. 11A, the horizontal axis represents RMSECV and the vertical axis represents RMSEP. When RMSECV is used as the index of goodness of fit, as shown in FIG. 11A, there is no particular tendency for RMSEP to decrease as the RMSECV value decreases.

[0105] FIG. 11B shows the experimental results when C1 is used as the index of goodness of fit. The horizontal and vertical axes in FIG. 11B are the same as those in FIG. 11A. When C1 was used as the goodness of fit index, as shown in Figure 11B, it was confirmed that as the RMSECV decreased, the RMSEP also decreased. In other words, by using C1 as the goodness of fit index, overfitting was suppressed and the prediction performance for the validation sample improved.

[0106] Next, predictions were made for the validation sample using the analytical wavelength range that minimized RMSECV under each condition (fit index: RMSECV or C1). In FIG. 11A, the data with the smallest RMSECV is indicated by a circle Q1. Similarly, in FIG. 11B, the data with the smallest RMSECV is indicated by a circle Q2. Thus, the prediction results for the analytical wavelength range that minimized RMSECV are shown in FIGS. 12A and 12B.

[0107] FIG. 12A is a diagram showing prediction results of validation samples when RMSECV is used as an index of goodness of fit. The vertical axis of FIG. 12A represents the predicted glucose concentration of the verification sample. The horizontal axis of FIG. 12A represents the actual glucose concentration measured by the enzyme electrode method. FIG. 12A also shows a line L1 where the predicted and actual measurements are equal. The concentration of data points on this line indicates high prediction performance for the verification sample.

[0108] FIG. 12B is a diagram showing prediction results for validation samples when C1 is used as the index of goodness of fit. The vertical and horizontal axes in FIG. 12B are the same as those in FIG. 12A. FIG. 12B also shows a line L2 where the predicted value and the actual measured value are equal. As shown in FIG. 12B, when C1 is used as the fitness index, the results are the same as when RMSECV is used as the fitness index and transfer learning is not performed (see FIG. 12A The prediction performance is improved compared to the previous example (see reference). The RMSEP improves from 0.46 [g / L] to 0.33 [g / L] when C1 is used as the fitness index. Furthermore, when C1 is used as the fitness index and transfer learning is performed, the RMSEP becomes 0.26 [g / L]. Indicators By combining the modifications with the application of transfer learning, prediction accuracy can be further improved.

[0109] <Effects> According to the fourth embodiment, overlearning can be suppressed by using the C1 value or the like as an index of goodness of fit, thereby improving the prediction performance of the prediction model. Therefore, sample composition determination and concentration quantification can be performed with high accuracy.

[0110] <<Variations>> Although the spectroscopic analysis system 100 according to the present invention has been described in each embodiment, the present invention is not limited to these descriptions and various modifications can be made. For example, in the first embodiment, the case where the upper limit of the measurement time of the spectroscopic analysis spectrum is set by the user has been described, but this is not limiting. That is, the lower limit of the measurement accuracy of the spectroscopic analysis spectrum may be set by the user. Regarding the processing flow of the control unit 32, as an alternative to step S101 in FIG. 4, the control unit 32 sets a lower limit of the measurement accuracy in response to an input operation by the user. Furthermore, as an alternative to step S106, the control unit 32 determines whether there is any measurement accuracy that is equal to or greater than the lower limit. By performing such processing, the user can set the measurement accuracy when measuring the spectroscopic analysis spectrum, thereby improving usability for the user. Note that when RMSECV, in which the smaller the value, the higher the measurement accuracy, is used as the measurement accuracy, the maximum value within the allowable range of RMSECV is used as the "lower limit of the measurement accuracy." Furthermore, both the upper limit of the measurement time and the lower limit of the measurement accuracy may be set by the user. That is, as user-set conditions for measuring the spectroscopic spectrum of a sample, input of at least one of the upper limit of the measurement time of the spectroscopic spectrum and the lower limit of the measurement accuracy may be accepted by operating the operation panel 21 (input unit: see FIG. 1). This allows the user to search for recommended measurement conditions desired by the user in terms of either or both of the measurement time and the measurement accuracy. Note that the setting of the lower limit of the measurement accuracy can also be applied to the second to fourth embodiments.

[0111] Furthermore, in each embodiment, when the measurement conditions are set in step S102 (see FIG. 4), a case has been described in which one predetermined value is input for each of the excitation light sampling interval, fluorescence sampling interval, excitation light slit width, fluorescence slit width, and wavelength scan speed, as shown in FIG. 5, but this is not limiting. That is, the user may input multiple candidate values ​​or predetermined numerical ranges for some or all of the measurement conditions. Then, the control unit 32 may set an optimal value for each item based on a genetic algorithm or the like.

[0112] Furthermore, in each embodiment, a case has been described in which fluorescence fingerprint analysis is used as an example of a spectroscopic analysis method, but this is not limiting. For example, absorptiometry (also called absorption spectroscopy), which is one of the spectroscopic analysis methods, may be used. When absorptiometry is used, the concentration of a predetermined substance is quantified by measuring the spectrum of light absorbed by the sample out of the light irradiated onto the sample. Note that in absorptiometry, the "light used to measure the spectroscopic analysis spectrum" is light absorbed by the sample.

[0113] In each embodiment, the recommended measurement conditions are set as shown in Fig. 6, including the excitation light sampling interval and fluorescence sampling interval, as well as the excitation light slit width, fluorescence slit width, excitation wavelength range, fluorescence wavelength range, and wavelength scan speed. However, the present invention is not limited to this. In other words, the recommended measurement conditions may be at least one of the wavelength range of light used in measuring the spectroscopic analysis spectrum, the sampling interval of the light wavelength, the slit width of the diffraction grating of the spectrometer that disperses the light (e.g., the excitation-side spectrometer 2 or the fluorescence-side spectrometer 8; see Fig. 1), and the wavelength sweep speed (scan speed) of the light. Even in such cases, the control unit 32 can search for the recommended measurement conditions so that the specified measurement time and measurement accuracy are satisfied.

[0114] Furthermore, for example, the control unit 32 may store multiple prediction models in the storage unit 321 (see FIG. 3) and use different prediction models depending on the state of the process, including in-line measurement. That is, when performing in-line measurement of a sample in an industrial plant, the control unit 32 may switch between the prediction models actually used to analyze the spectroscopic analysis spectrum of the sample from among the multiple prediction models depending on the state of the industrial plant. Here, the measurement times of the spectroscopic analysis spectrum in the measurement wavelength ranges that are one-to-one associated with the multiple prediction models are different. Furthermore, the "industrial plant" is not limited to a factory, but also includes various facilities such as research facilities. For example, when an industrial process including in-line measurement is operating steadily and stably, the control unit 32 sets the time required to measure one sample to a relatively long predetermined time. Furthermore, when the control unit 32 detects that the industrial process is in an abnormal state, the control unit 32 shortens the time required to measure one sample in order to transition the industrial process to a steady and stable state. In this way, the control unit 32 can appropriately control the industrial process including in-line measurement by using different prediction models depending on the state of the industrial process.

[0115] In the first embodiment, the control unit 32 determines whether or not there is a spectroscopic analysis spectrum whose measurement time is equal to or less than the upper limit (S106 in FIG. 4), and then selects the regression model with the highest prediction performance (S108). However, this is not limiting. For example, the control unit 32 may determine the measurement time in descending order of prediction performance among the regression models for each wavelength region. This type of processing also produces the same effect.

[0116] In the first embodiment, the photometer unit 10 (see FIG. 1) is described as including the excitation-side filter 5 (see FIG. 1) and the fluorescence-side filter 7 (see FIG. 1), but these may be omitted as appropriate. In the first embodiment, the processing result of the control unit 32 (see FIG. 1) is displayed on the display unit 22 (see FIG. 1), but this is not limiting. For example, the processing result of the control unit 32 may be transmitted to a mobile terminal (not shown) such as a mobile phone, smartphone, or tablet. In this case, the display of the mobile terminal functions as the display unit. In the first embodiment, the case where the measurement time and the like are input by the user operating the input unit (see FIG. 1) has been described, but this is not limiting. For example, the measurement time and the like may be input based on the user operating a mobile terminal (see FIG. 1). In this case, the buttons or touch panel of the mobile terminal function as the "input unit." The same can be said for the second to fourth embodiments.

[0117] In addition, the functions of the spectroscopic analysis system 100 etc. described in each embodiment ( spectroscopy All or part of the program for realizing the analysis method may be executed by one or more computers such as a server (not shown). The program may be provided via a communication line or may be written to a recording medium such as a CD-ROM and distributed.

[0118] Furthermore, each embodiment has been described in detail to clearly explain the present invention, and is not necessarily limited to having all of the described configurations. Furthermore, it is possible to add, delete, or replace some of the configurations of the embodiments with other configurations. Furthermore, the above-described mechanisms and configurations are those considered necessary for explanation, and do not necessarily represent all of the mechanisms and configurations of the product. [Explanation of symbols]

[0119] 1 light source 2. Excitation side spectroscope (spectroscope) 2a Diffraction grating 3 Beam Splitter 4 Monitor Detector 5. Excitation filter 6. Sample placement area 7 Fluorescence side filter 8 Fluorescence side spectroscope (spectroscope) 8a Diffraction grating 9 Detector 10 Photometer section 21 Operation panel (input section) 22 Display section (display device) 32 Control section 321e Regression model (prediction model) 100 Spectroscopic Analysis System G1,G2 measurement wavelength range M1 sample

Claims

1. an input unit that receives input of at least one of an upper limit value of a measurement time of the spectroscopic analysis spectrum and a lower limit value of a measurement accuracy as a user-set condition for measuring the spectroscopic analysis spectrum of the sample; a control unit that derives predetermined recommended measurement conditions that satisfy the user-set conditions and displays the recommended measurement conditions on a display device; the recommended measurement conditions are at least one of a wavelength range of light used in measuring the spectroscopic spectrum, a sampling interval of the wavelength of the light, a slit width of a diffraction grating of a spectrometer that disperses the light, and a sweep speed of the wavelength of the light.

2. the control unit causes the display device to display the recommended measurement conditions and also causes the display device to display the spectroscopic analysis spectrum based on the fluorescence fingerprint analysis; The spectroscopic analysis spectrum displayed on the display device shows a measurement wavelength region specified by a range of excitation wavelengths and a range of fluorescence wavelengths as the wavelength range of the recommended measurement conditions.

2. The spectroscopic analysis system according to claim 1,

3. The display device displays both the recommended measurement conditions and the spectroscopic analysis spectrum on one screen.

3. The spectroscopic analysis system according to claim 2, wherein:

4. the number of measurement wavelength regions is set based on an operation by a user via the input unit; Or, The range of the number of measurement wavelength regions is set based on an operation by a user via the input unit.

3. The spectroscopic analysis system according to claim 2, wherein:

5. The control unit displays a plurality of candidates on the display device as a wavelength region selection method that is a method for selecting the measurement wavelength region, and executes a predetermined wavelength region selection method that is selected from the plurality of candidates by a user's operation via the input unit.

3. The spectroscopic analysis system according to claim 2, wherein:

6. The control unit displays a plurality of candidates for regression methods of multivariate analysis to be used in analyzing the spectroscopic spectrum on the display device, and executes a predetermined regression method selected from the plurality of candidates by a user's operation via the input unit.

2. The spectroscopic analysis system according to claim 1,

7. When measuring the spectroscopic analysis spectrum based on fluorescence fingerprint analysis, if one of the excitation wavelength and the fluorescence wavelength is fixed and the other wavelength is swept at a predetermined sampling interval, the control unit generates a new pseudo spectroscopic analysis spectrum based on data in which a value obtained by summing some or all of n fluorescence intensities corresponding one-to-one to the n other wavelengths for each of n other wavelengths (n is a natural number) swept at the predetermined sampling interval is associated with any of the n other wavelengths, or data in which the value obtained by summing is associated with an average value of the n other wavelengths, and generates a prediction model for analyzing the spectroscopic analysis spectrum based on the pseudo spectroscopic analysis spectrum.

2. The spectroscopic analysis system according to claim 1,

8. The control unit actually measures a spectroscopic analysis spectrum under the recommended measurement conditions based on the pseudo spectroscopic analysis spectrum, and generates a prediction model for analysis again based on the analysis spectrum obtained by the measurement.

8. The spectroscopic analysis system according to claim 7,

9. The recommended measurement conditions include excitation wavelengths and fluorescence wavelengths with different sampling intervals, and these wavelengths are mixed in a plurality of measurement wavelength ranges.

3. The spectroscopic analysis system according to claim 2, wherein:

10. The sampling interval of at least one of the plurality of measurement wavelength regions is set based on a genetic algorithm. The spectroscopic analysis system according to claim 9 .

11. the control unit searches for the measurement wavelength region based on a genetic algorithm, and in the genetic algorithm, assigns a predetermined penalty function indicating a degree of overlearning to an RMSECV, which is an index of goodness of fit of a prediction model for analyzing the spectroscopic analysis spectrum, and evaluates the prediction model based on the penalty function; The RMSECV is the average value of the mean square error in cross-validation.

3. The spectroscopic analysis system according to claim 2, wherein:

12. The control unit switches a prediction model to be actually used for analyzing the spectroscopic analysis spectrum of the sample from among a plurality of prediction models depending on the state of the industrial plant when performing in-line measurement of the sample in the industrial plant.

2. The spectroscopic analysis system according to claim 1,

13. an input process for receiving input of at least one of an upper limit value of a measurement time of the spectroscopic analysis spectrum and a lower limit value of a measurement accuracy as a user setting condition for measuring the spectroscopic analysis spectrum of the sample; a display process for deriving predetermined recommended measurement conditions that satisfy the user-set conditions and displaying the recommended measurement conditions on a display device, the recommended measurement conditions are at least one of a wavelength range of light used in measuring the spectroscopic spectrum, a sampling interval of the wavelength of the light, a slit width of a diffraction grating of a spectrometer that disperses the light, and a sweep speed of the wavelength of the light.

Images