CN105259136A - Measuring-point-free temperature correction method of near-infrared correction model - Google Patents

Abstract

The invention relates to a measuring-point-free temperature correction modeling method used in near-infrared spectrum measurement. The measuring-point-free temperature correction modeling method comprises the steps that representative samples are collected, and it is ensured that the range of measurement requirements can be covered by the physical parameters to be measured of the samples; near-infrared spectrums of the samples are acquired at different temperature levels; preprocessing and principal component analysis are conducted on the acquired spectrums respectively according to the temperatures and the physical parameters to be measured; the two different spectrums obtained through preprocessing are combined, and combined spectral data are produced; finally, principal component analysis is conducted on the combined spectrums, and a physical parameter correction model is established through partial least squares. The measuring-point-free temperature correction modeling method uses the temperatures as separated implied factor variables to be involved in the near-infrared modeling process, thus when the near-infrared measurement is performed, measurement of physical properties at different temperatures can be completed by relying on the temperature adaptability of the model itself without direct temperature measurement information and relevant calculation, and the established model can have better universality and temperature adaptability.

Description

Near infrared correction without measuring point temperature correction

Technical field

The present invention relates in near-infrared spectral measurement without measuring point temperature adjustmemt modeling method, be applicable to the quick detection of substance viscosity easily influenced by ambient temperature, sweat alanine concentration, food quality, quality of agricultural product, medicine quality, oil product of gasoline etc., also can be used for the measurement of human body Woundless blood sugar concentration, soil constituent and mineralogical composition etc.

Background technology

In recent years, near-infrared spectral analysis technology with its fast detection, Non-Destructive Testing, without chemical contamination, the advantage such as easy and simple to handle, sample preparation is simple, be widely used in the industries such as petrochemical complex, food, agricultural, medicine, become one of qualitative and quantitative analysis technology with fastest developing speed.Absorption near infrared spectrum district mainly comes from the state change that molecular vibration or rotation cause, the vibration of its each group is easily subject to the impact of the external conditions such as temperature, especially when measuring fluid sample, the rising of temperature can cause the hydroxy number of stretching vibration to reduce and the increase of the number of free vibration, thus produce the skew of vibrational spectrum, the near-infrared spectroscopy set up under making specified temp can only be applicable to the sample quality analysis at this temperature, and it is undesirable for the sample quality analytical effect of different temperatures, this shortcoming strongly limit the application of near-infrared spectrometers modeling technique.

In order to obtain good accuracy of analysis, measure the impact that effectively can reduce temperature variation at a constant temperature, but temperature cannot accurately control in actual applications, therefore solve temperature to be suggested successively the certain methods that near infrared spectrum affects, as preprocessing procedures rejects the impact of temperature in spectrum; Choosing affects insensitive wave band to temperature and sets up analytical model; The method corrected in adopting, is included in temperature information in mathematical model; Measure the temperature of sample while gathering spectrum, set up Temperature correction model etc.These methods may be used for overcoming the interference of testing sample temperature variation to Quantitative Analysis Model, but, also do not have general rule to use which kind of method at present under judging which kind of situation, and will select according to particular problem.Therefore, under temperature variation, set up more general, that thermal adaptability is stronger near infrared detection calibration model, can effectively apply very crucial near infrared technology.

Summary of the invention

The method that the present invention proposes, sample spectra is obtained under different temperatures level, temperature is participated near infrared modeling process as the implicit variable factors be separated, thus when using near-infrared measuring, can rely on the adaptability of model to temperature itself complete different temperatures under physical measurement, do not need direct temperature metrical information and correlation computations, make set up model have better versatility and thermal adaptability.

The present invention for achieving the above object, adopts following technical scheme:

Step of the present invention is divided into two parts.Part I, the experimental design of modeling data and spectral collection; Part II, the pre-service of near infrared spectrum and the foundation of calibration model.

The experimental facilities of modeling data comprises, and the sample cell (2) that (1) can regulate sample temperature temperature meter (3) the near infrared spectrum collection instrument (4) of displays temperature change can not produce the optic probe of obviously impact on sample temperature.(5) the computer recording device be connected with near infrared spectrum collection instrument.

Experiment of the present invention and data collection step as follows:

Experimental procedure one: the minimum and maximum temperature value of confirmatory sample.Temperature range is divided into multiple level value.Each temperature levels is generally greater than thermometric instruments resolution 5 times, effectively distinguishes precision to reach.

Experimental procedure two: the highest with within the scope of minimum temperature, data are analyzed to the primary standard that all samples physical parameter obtains at defined temperature.

Experimental procedure three: the spectroscopic data collecting sample under different temperatures level.Record corresponding sample temperature value simultaneously.This temperature value participates in modeling as an implicit factor, and the exact numerical of record temperature to this method not necessarily.

Temperature is as follows as the implicit variable factors modeling method step be separated:

Modeling procedure one: carrying out spectrum take temperature model as the pre-service of target.Original spectrum is done first order derivative or second derivative operator, produce first derivative spectrum or second derivative spectra.The determination of derivative order is herein different with the characteristic of sample and physical parameter.Such as, to macromolecule high viscosity samples, be better with second derivative.Be better to low viscosity sample with first order derivative.

Modeling procedure two: do pivot analysis (PCA) to the derivative spectrum produced above, rejects statistics exceptional value, makes the pivot pattern of whole derivative spectrum data all within a statistical certainty.

Modeling procedure three: to the original spectrum pre-service that to carry out with physical property parameter mode to be measured be target.These pre-service comprise the superposition of one or more following algorithms: first order derivative, second derivative, maximum-minimum sandards, and basic bottom line corrects, scatter correction, constant bias correction, etc.The determination of Preprocessing Algorithm is different with physical parameter to be measured herein.

Modeling procedure four: do pivot analysis (PCA) to spectrum after the pre-service produced above, rejects statistics exceptional value, makes whole pretreated spectroscopic data pivot pattern all within a statistical certainty.

Modeling procedure five: take temperature as the derivative spectrum of target by what formed above and merge with spectrum after the pre-service that is target of physical property parameter to be measured.

Modeling procedure six: do pivot analysis (PCA) to the merging spectrum produced above, rejects statistics exceptional value, makes the pivot pattern of whole merging spectroscopic data all within a statistical certainty.

Using physical parameter to be measured in the original analysis value of a set point of temperature as predictive variable, to merge spectrum wave number as independent variable.Geophysical parameter prediction model is set up with partial least squares algorithm (PLS):

P＝B ₁y ₁+B ₂y ₂+…B _ny _n+A ₁x ₁+A ₂x ₂+…A _nx _n

Herein, P is the measured value under physical property variable set point of temperature, B _i, A _i, i=1,2 ... n is regression coefficient, y _i, x _ibe respectively after pre-service spectrum and derivative spectrum at wave number i=1,2 ... the numerical value at n place.

The present invention participates in temperature near infrared modeling process as the implicit variable factors be separated, thus when using near-infrared measuring, can rely on the adaptability of model to temperature itself complete different temperatures under physical measurement, do not need direct temperature metrical information and correlation computations, make set up model have preferably compensation effect to temperature, thus there is better versatility.

Accompanying drawing explanation

Fig. 1 is without measuring point temperature compensation experimental provision

The second derivative pre-processed spectrum of a kind of macromolecular compound of Fig. 2

Fig. 3 is based on the PCA mode chart of second derivative pre-processed spectrum

Fig. 4 first order derivative pre-processed spectrum

Fig. 5 first derivative spectrum principal element pattern

Fig. 6 implementation step block diagram

The merging spectrum of Fig. 7 different temperatures

Fig. 8 merges the PCA mode chart of spectrum

Fig. 9 merges the Viscosity Model that spectrum produces.

The wave number that Figure 10 Viscosity Model uses.

Figure 11 the inventive method result is to the adaptability of temperature

Embodiment

Below for a kind of viscosity measurement of macromolecular compound, specific implementation method is described.This example does not form and limits the scope of the inventive method.

Near infrared correction without measuring point temperature correction implementation step block diagram as shown in Figure 6, specifically comprise the following steps:

Step one: gather representative sample, ensure that the physical parameter to be measured of sample can cover the scope measured and require.Total number of samples is at 40-60.

Step 2: utilize the laboratory equipment shown in Fig. 1, gathers the near infrared spectrum of each sample respectively under 24 DEG C, 35 DEG C, 50 DEG C, 60 DEG C, 70 DEG C five different temperatures levels, records experiment condition as temperature etc. simultaneously.

Step 3: doing gathered spectrum with temperature is pre-service and the pivot analysis of target, produces derivative spectrum data.In example, second derivative process and pivot analysis are carried out to macromolecule high viscosity sample.As shown in Figure 2, principal element pattern as shown in Figure 3 for treatment effect.Second derivative pre-service is on the basis of first order derivative, extracts the spectrum of temperature information sensitivity again, effectively reduces temperature and the physical parameter overlap in modeling wave number.In the PCA mode chart shown in Fig. 3, have two points and other to have a little very large distance, this point is singular point, is rejected, make whole pretreated spectroscopic data pivot pattern all within a statistical certainty when modeling.

Step 4: original spectrum is carried out with physical property parameter mode to be measured be target pre-service and pivot analysis (PCA).Produce pre-processed spectrum data.In example, first order derivative pre-service and pivot analysis are carried out to macromolecule sample.Spectrum after treatment eliminates due to light source ages, and what probe vibrations and probe and the factor such as sample contacts degree were brought drift about spectrally down, while remain again the effective information that temperature affects spectrum peak and shape.Pre-processed spectrum as shown in Figure 4.Fig. 5 is the pivot mode chart of pre-processed spectrum.There are two singular points in Fig. 5, should give rejecting, no longer participate in modeling.

Step 5: the derivative spectrum produced above and pre-processed spectrum are merged, produces and merge spectroscopic data.Fig. 7 is the merging spectrum in different temperatures.In merging spectrum in the figure 7, left-half and first order derivative part, provide effective physical property modeling spectral information; Right half part and second derivative part, provide the spectral information of temperature compensation function.

Step 6: do pivot analysis (PCA) to the merging spectrum produced above, rejects statistics exceptional value, makes the pivot pattern of whole merging spectroscopic data all within a statistical certainty.Fig. 8 is the PCA mode chart merging spectrum.There are three singular points in Fig. 8, should give rejecting, no longer participate in modeling.

Step 7: using physical parameter original analysis value to be measured as predictive variable, to merge spectrum wave number as independent variable, set up geophysical parameter prediction model with partial least squares algorithm (PLS):

P＝B ₁y ₁+B ₂y ₂+…B _ny _n+A ₁x ₁+A ₂x ₂+…A _nx _n

Herein, P is the measured value under physical property variable set point of temperature, B _i, A _i, i=1,2 ... n is regression coefficient, y _i, x _ibe respectively after pre-service spectrum and derivative spectrum at wave number i=1,2 ... the numerical value at n place.

Fig. 9 is the Viscosity Model merging spectrum generation, and Figure 10 is the wave number that Viscosity Model uses, and Figure 11 is the comparison of result to thermal adaptability.The correlativity of Fig. 9 complex spectrum model predication value and measured value is 0.98, model accuracy R ²be 0.97.Wavelength band shown in Figure 10, first derivative spectrum section is 9056-4765cm ^-1, second derivative spectra section is chosen as 6024-4528cm ^-1.Figure 11 be the present invention propose without measuring point temperature compensation algorithm be fixed on comparing of modeling algorithm at 50 degree of temperature, as can be seen from the figure the measurement result of fixed temperature model has larger susceptibility to temperature variation, and the model that the inventive method is set up, there is preferably compensation effect to temperature.

Claims (5)

1. in near-infrared spectral measurement without a measuring point temperature adjustmemt modeling method, it is characterized in that the method comprises the steps:

Step one: obtain multiple sample physical parameter primary standard data at the specified temperature, and gather near infrared spectrum under different temperatures level;

Step 2: carrying out the near infrared spectrum gathered in step one take temperature as pre-service and the pivot analysis of target, produces derivative spectrum data; Carry out, with physical property parameter mode to be measured be target pre-service and pivot analysis, producing pre-processed spectrum to the original near infrared spectrum gathered in step one;

Step 3: the pre-processed spectrum that the derivative spectrum produced in step 2 and step 4 produce is carried out spectrum merging, produces and merge spectroscopic data;

Step 4: do pivot analysis to the merging spectroscopic data described in step 3, rejects statistics exceptional value, makes the pivot pattern of whole merging spectroscopic data all within a statistical certainty;

Step 5: using physical parameter original analysis value to be measured as predictive variable, to merge spectrum wave number as independent variable, set up geophysical parameter prediction model.

2. near infrared correction according to claim 1 without measuring point temperature correction, it is characterized in that: in described step one, the temperature range of collected specimens near infrared spectrum covers real-time physical parameter measuring tempeature scope; Multiple sample physical property parameters distribution to be measured is measured the scope of requirement and has enough intervals distinguished.

3. near infrared correction according to claim 1 without measuring point temperature correction, it is characterized in that: the Pretreated spectra being target for temperature in described step 2 is second derivative.

4. near infrared correction according to claim 1 without measuring point temperature correction, it is characterized in that: the Pretreated spectra being target for physical parameter in described step 2 comprises the superposition of one or more following algorithms: first order derivative, second derivative, maximum-minimum sandards, basis bottom line corrects, scatter correction, constant bias correction.

5. near infrared correction according to claim 1 without measuring point temperature correction, it is characterized in that: in described step 5 geophysical parameter prediction model foundation adopt partial least squares algorithm carry out the linear regression that temperature is implicit variable:

P＝B ₁y ₁+B ₂y ₂+…B _ny _n+A ₁x ₁+A ₂x ₂+…A _nx _n

Herein, P is the measured value under physical property variable set point of temperature, B _i, A _i, i=1,2 ... n is regression coefficient, y _i, x _ibe respectively after pre-service spectrum and derivative spectrum at wave number i=1,2 ... the numerical value at n place.